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 WARN_ON(cache
->reserved_pinned
> 0);
125 kfree(cache
->free_space_ctl
);
131 * this adds the block group to the fs_info rb tree for the block group
134 static int btrfs_add_block_group_cache(struct btrfs_fs_info
*info
,
135 struct btrfs_block_group_cache
*block_group
)
138 struct rb_node
*parent
= NULL
;
139 struct btrfs_block_group_cache
*cache
;
141 spin_lock(&info
->block_group_cache_lock
);
142 p
= &info
->block_group_cache_tree
.rb_node
;
146 cache
= rb_entry(parent
, struct btrfs_block_group_cache
,
148 if (block_group
->key
.objectid
< cache
->key
.objectid
) {
150 } else if (block_group
->key
.objectid
> cache
->key
.objectid
) {
153 spin_unlock(&info
->block_group_cache_lock
);
158 rb_link_node(&block_group
->cache_node
, parent
, p
);
159 rb_insert_color(&block_group
->cache_node
,
160 &info
->block_group_cache_tree
);
161 spin_unlock(&info
->block_group_cache_lock
);
167 * This will return the block group at or after bytenr if contains is 0, else
168 * it will return the block group that contains the bytenr
170 static struct btrfs_block_group_cache
*
171 block_group_cache_tree_search(struct btrfs_fs_info
*info
, u64 bytenr
,
174 struct btrfs_block_group_cache
*cache
, *ret
= NULL
;
178 spin_lock(&info
->block_group_cache_lock
);
179 n
= info
->block_group_cache_tree
.rb_node
;
182 cache
= rb_entry(n
, struct btrfs_block_group_cache
,
184 end
= cache
->key
.objectid
+ cache
->key
.offset
- 1;
185 start
= cache
->key
.objectid
;
187 if (bytenr
< start
) {
188 if (!contains
&& (!ret
|| start
< ret
->key
.objectid
))
191 } else if (bytenr
> start
) {
192 if (contains
&& bytenr
<= end
) {
203 btrfs_get_block_group(ret
);
204 spin_unlock(&info
->block_group_cache_lock
);
209 static int add_excluded_extent(struct btrfs_root
*root
,
210 u64 start
, u64 num_bytes
)
212 u64 end
= start
+ num_bytes
- 1;
213 set_extent_bits(&root
->fs_info
->freed_extents
[0],
214 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
215 set_extent_bits(&root
->fs_info
->freed_extents
[1],
216 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
220 static void free_excluded_extents(struct btrfs_root
*root
,
221 struct btrfs_block_group_cache
*cache
)
225 start
= cache
->key
.objectid
;
226 end
= start
+ cache
->key
.offset
- 1;
228 clear_extent_bits(&root
->fs_info
->freed_extents
[0],
229 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
230 clear_extent_bits(&root
->fs_info
->freed_extents
[1],
231 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
234 static int exclude_super_stripes(struct btrfs_root
*root
,
235 struct btrfs_block_group_cache
*cache
)
242 if (cache
->key
.objectid
< BTRFS_SUPER_INFO_OFFSET
) {
243 stripe_len
= BTRFS_SUPER_INFO_OFFSET
- cache
->key
.objectid
;
244 cache
->bytes_super
+= stripe_len
;
245 ret
= add_excluded_extent(root
, cache
->key
.objectid
,
250 for (i
= 0; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
251 bytenr
= btrfs_sb_offset(i
);
252 ret
= btrfs_rmap_block(&root
->fs_info
->mapping_tree
,
253 cache
->key
.objectid
, bytenr
,
254 0, &logical
, &nr
, &stripe_len
);
258 cache
->bytes_super
+= stripe_len
;
259 ret
= add_excluded_extent(root
, logical
[nr
],
269 static struct btrfs_caching_control
*
270 get_caching_control(struct btrfs_block_group_cache
*cache
)
272 struct btrfs_caching_control
*ctl
;
274 spin_lock(&cache
->lock
);
275 if (cache
->cached
!= BTRFS_CACHE_STARTED
) {
276 spin_unlock(&cache
->lock
);
280 /* We're loading it the fast way, so we don't have a caching_ctl. */
281 if (!cache
->caching_ctl
) {
282 spin_unlock(&cache
->lock
);
286 ctl
= cache
->caching_ctl
;
287 atomic_inc(&ctl
->count
);
288 spin_unlock(&cache
->lock
);
292 static void put_caching_control(struct btrfs_caching_control
*ctl
)
294 if (atomic_dec_and_test(&ctl
->count
))
299 * this is only called by cache_block_group, since we could have freed extents
300 * we need to check the pinned_extents for any extents that can't be used yet
301 * since their free space will be released as soon as the transaction commits.
303 static u64
add_new_free_space(struct btrfs_block_group_cache
*block_group
,
304 struct btrfs_fs_info
*info
, u64 start
, u64 end
)
306 u64 extent_start
, extent_end
, size
, total_added
= 0;
309 while (start
< end
) {
310 ret
= find_first_extent_bit(info
->pinned_extents
, start
,
311 &extent_start
, &extent_end
,
312 EXTENT_DIRTY
| EXTENT_UPTODATE
);
316 if (extent_start
<= start
) {
317 start
= extent_end
+ 1;
318 } else if (extent_start
> start
&& extent_start
< end
) {
319 size
= extent_start
- start
;
321 ret
= btrfs_add_free_space(block_group
, start
,
324 start
= extent_end
+ 1;
333 ret
= btrfs_add_free_space(block_group
, start
, size
);
340 static noinline
void caching_thread(struct btrfs_work
*work
)
342 struct btrfs_block_group_cache
*block_group
;
343 struct btrfs_fs_info
*fs_info
;
344 struct btrfs_caching_control
*caching_ctl
;
345 struct btrfs_root
*extent_root
;
346 struct btrfs_path
*path
;
347 struct extent_buffer
*leaf
;
348 struct btrfs_key key
;
354 caching_ctl
= container_of(work
, struct btrfs_caching_control
, work
);
355 block_group
= caching_ctl
->block_group
;
356 fs_info
= block_group
->fs_info
;
357 extent_root
= fs_info
->extent_root
;
359 path
= btrfs_alloc_path();
363 last
= max_t(u64
, block_group
->key
.objectid
, BTRFS_SUPER_INFO_OFFSET
);
366 * We don't want to deadlock with somebody trying to allocate a new
367 * extent for the extent root while also trying to search the extent
368 * root to add free space. So we skip locking and search the commit
369 * root, since its read-only
371 path
->skip_locking
= 1;
372 path
->search_commit_root
= 1;
377 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
379 mutex_lock(&caching_ctl
->mutex
);
380 /* need to make sure the commit_root doesn't disappear */
381 down_read(&fs_info
->extent_commit_sem
);
383 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
387 leaf
= path
->nodes
[0];
388 nritems
= btrfs_header_nritems(leaf
);
391 if (btrfs_fs_closing(fs_info
) > 1) {
396 if (path
->slots
[0] < nritems
) {
397 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
399 ret
= find_next_key(path
, 0, &key
);
403 if (need_resched() ||
404 btrfs_next_leaf(extent_root
, path
)) {
405 caching_ctl
->progress
= last
;
406 btrfs_release_path(path
);
407 up_read(&fs_info
->extent_commit_sem
);
408 mutex_unlock(&caching_ctl
->mutex
);
412 leaf
= path
->nodes
[0];
413 nritems
= btrfs_header_nritems(leaf
);
417 if (key
.objectid
< block_group
->key
.objectid
) {
422 if (key
.objectid
>= block_group
->key
.objectid
+
423 block_group
->key
.offset
)
426 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
) {
427 total_found
+= add_new_free_space(block_group
,
430 last
= key
.objectid
+ key
.offset
;
432 if (total_found
> (1024 * 1024 * 2)) {
434 wake_up(&caching_ctl
->wait
);
441 total_found
+= add_new_free_space(block_group
, fs_info
, last
,
442 block_group
->key
.objectid
+
443 block_group
->key
.offset
);
444 caching_ctl
->progress
= (u64
)-1;
446 spin_lock(&block_group
->lock
);
447 block_group
->caching_ctl
= NULL
;
448 block_group
->cached
= BTRFS_CACHE_FINISHED
;
449 spin_unlock(&block_group
->lock
);
452 btrfs_free_path(path
);
453 up_read(&fs_info
->extent_commit_sem
);
455 free_excluded_extents(extent_root
, block_group
);
457 mutex_unlock(&caching_ctl
->mutex
);
459 wake_up(&caching_ctl
->wait
);
461 put_caching_control(caching_ctl
);
462 btrfs_put_block_group(block_group
);
465 static int cache_block_group(struct btrfs_block_group_cache
*cache
,
466 struct btrfs_trans_handle
*trans
,
467 struct btrfs_root
*root
,
470 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
471 struct btrfs_caching_control
*caching_ctl
;
475 if (cache
->cached
!= BTRFS_CACHE_NO
)
479 * We can't do the read from on-disk cache during a commit since we need
480 * to have the normal tree locking. Also if we are currently trying to
481 * allocate blocks for the tree root we can't do the fast caching since
482 * we likely hold important locks.
484 if (trans
&& (!trans
->transaction
->in_commit
) &&
485 (root
&& root
!= root
->fs_info
->tree_root
)) {
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
);
2721 * We want to set the generation to 0, that way if anything goes wrong
2722 * from here on out we know not to trust this cache when we load up next
2725 BTRFS_I(inode
)->generation
= 0;
2726 ret
= btrfs_update_inode(trans
, root
, inode
);
2729 if (i_size_read(inode
) > 0) {
2730 ret
= btrfs_truncate_free_space_cache(root
, trans
, path
,
2736 spin_lock(&block_group
->lock
);
2737 if (block_group
->cached
!= BTRFS_CACHE_FINISHED
) {
2738 /* We're not cached, don't bother trying to write stuff out */
2739 dcs
= BTRFS_DC_WRITTEN
;
2740 spin_unlock(&block_group
->lock
);
2743 spin_unlock(&block_group
->lock
);
2745 num_pages
= (int)div64_u64(block_group
->key
.offset
, 1024 * 1024 * 1024);
2750 * Just to make absolutely sure we have enough space, we're going to
2751 * preallocate 12 pages worth of space for each block group. In
2752 * practice we ought to use at most 8, but we need extra space so we can
2753 * add our header and have a terminator between the extents and the
2757 num_pages
*= PAGE_CACHE_SIZE
;
2759 ret
= btrfs_check_data_free_space(inode
, num_pages
);
2763 ret
= btrfs_prealloc_file_range_trans(inode
, trans
, 0, 0, num_pages
,
2764 num_pages
, num_pages
,
2767 dcs
= BTRFS_DC_SETUP
;
2768 btrfs_free_reserved_data_space(inode
, num_pages
);
2772 btrfs_release_path(path
);
2774 spin_lock(&block_group
->lock
);
2775 block_group
->disk_cache_state
= dcs
;
2776 spin_unlock(&block_group
->lock
);
2781 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
2782 struct btrfs_root
*root
)
2784 struct btrfs_block_group_cache
*cache
;
2786 struct btrfs_path
*path
;
2789 path
= btrfs_alloc_path();
2795 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
2797 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
)
2799 cache
= next_block_group(root
, cache
);
2807 err
= cache_save_setup(cache
, trans
, path
);
2808 last
= cache
->key
.objectid
+ cache
->key
.offset
;
2809 btrfs_put_block_group(cache
);
2814 err
= btrfs_run_delayed_refs(trans
, root
,
2819 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
2821 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
) {
2822 btrfs_put_block_group(cache
);
2828 cache
= next_block_group(root
, cache
);
2837 if (cache
->disk_cache_state
== BTRFS_DC_SETUP
)
2838 cache
->disk_cache_state
= BTRFS_DC_NEED_WRITE
;
2840 last
= cache
->key
.objectid
+ cache
->key
.offset
;
2842 err
= write_one_cache_group(trans
, root
, path
, cache
);
2844 btrfs_put_block_group(cache
);
2849 * I don't think this is needed since we're just marking our
2850 * preallocated extent as written, but just in case it can't
2854 err
= btrfs_run_delayed_refs(trans
, root
,
2859 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
2862 * Really this shouldn't happen, but it could if we
2863 * couldn't write the entire preallocated extent and
2864 * splitting the extent resulted in a new block.
2867 btrfs_put_block_group(cache
);
2870 if (cache
->disk_cache_state
== BTRFS_DC_NEED_WRITE
)
2872 cache
= next_block_group(root
, cache
);
2881 btrfs_write_out_cache(root
, trans
, cache
, path
);
2884 * If we didn't have an error then the cache state is still
2885 * NEED_WRITE, so we can set it to WRITTEN.
2887 if (cache
->disk_cache_state
== BTRFS_DC_NEED_WRITE
)
2888 cache
->disk_cache_state
= BTRFS_DC_WRITTEN
;
2889 last
= cache
->key
.objectid
+ cache
->key
.offset
;
2890 btrfs_put_block_group(cache
);
2893 btrfs_free_path(path
);
2897 int btrfs_extent_readonly(struct btrfs_root
*root
, u64 bytenr
)
2899 struct btrfs_block_group_cache
*block_group
;
2902 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
2903 if (!block_group
|| block_group
->ro
)
2906 btrfs_put_block_group(block_group
);
2910 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
2911 u64 total_bytes
, u64 bytes_used
,
2912 struct btrfs_space_info
**space_info
)
2914 struct btrfs_space_info
*found
;
2918 if (flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
2919 BTRFS_BLOCK_GROUP_RAID10
))
2924 found
= __find_space_info(info
, flags
);
2926 spin_lock(&found
->lock
);
2927 found
->total_bytes
+= total_bytes
;
2928 found
->disk_total
+= total_bytes
* factor
;
2929 found
->bytes_used
+= bytes_used
;
2930 found
->disk_used
+= bytes_used
* factor
;
2932 spin_unlock(&found
->lock
);
2933 *space_info
= found
;
2936 found
= kzalloc(sizeof(*found
), GFP_NOFS
);
2940 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
2941 INIT_LIST_HEAD(&found
->block_groups
[i
]);
2942 init_rwsem(&found
->groups_sem
);
2943 spin_lock_init(&found
->lock
);
2944 found
->flags
= flags
& (BTRFS_BLOCK_GROUP_DATA
|
2945 BTRFS_BLOCK_GROUP_SYSTEM
|
2946 BTRFS_BLOCK_GROUP_METADATA
);
2947 found
->total_bytes
= total_bytes
;
2948 found
->disk_total
= total_bytes
* factor
;
2949 found
->bytes_used
= bytes_used
;
2950 found
->disk_used
= bytes_used
* factor
;
2951 found
->bytes_pinned
= 0;
2952 found
->bytes_reserved
= 0;
2953 found
->bytes_readonly
= 0;
2954 found
->bytes_may_use
= 0;
2956 found
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
2957 found
->chunk_alloc
= 0;
2959 init_waitqueue_head(&found
->wait
);
2960 *space_info
= found
;
2961 list_add_rcu(&found
->list
, &info
->space_info
);
2965 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
2967 u64 extra_flags
= flags
& (BTRFS_BLOCK_GROUP_RAID0
|
2968 BTRFS_BLOCK_GROUP_RAID1
|
2969 BTRFS_BLOCK_GROUP_RAID10
|
2970 BTRFS_BLOCK_GROUP_DUP
);
2972 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
2973 fs_info
->avail_data_alloc_bits
|= extra_flags
;
2974 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
2975 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
2976 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
2977 fs_info
->avail_system_alloc_bits
|= extra_flags
;
2981 u64
btrfs_reduce_alloc_profile(struct btrfs_root
*root
, u64 flags
)
2984 * we add in the count of missing devices because we want
2985 * to make sure that any RAID levels on a degraded FS
2986 * continue to be honored.
2988 u64 num_devices
= root
->fs_info
->fs_devices
->rw_devices
+
2989 root
->fs_info
->fs_devices
->missing_devices
;
2991 if (num_devices
== 1)
2992 flags
&= ~(BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID0
);
2993 if (num_devices
< 4)
2994 flags
&= ~BTRFS_BLOCK_GROUP_RAID10
;
2996 if ((flags
& BTRFS_BLOCK_GROUP_DUP
) &&
2997 (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
2998 BTRFS_BLOCK_GROUP_RAID10
))) {
2999 flags
&= ~BTRFS_BLOCK_GROUP_DUP
;
3002 if ((flags
& BTRFS_BLOCK_GROUP_RAID1
) &&
3003 (flags
& BTRFS_BLOCK_GROUP_RAID10
)) {
3004 flags
&= ~BTRFS_BLOCK_GROUP_RAID1
;
3007 if ((flags
& BTRFS_BLOCK_GROUP_RAID0
) &&
3008 ((flags
& BTRFS_BLOCK_GROUP_RAID1
) |
3009 (flags
& BTRFS_BLOCK_GROUP_RAID10
) |
3010 (flags
& BTRFS_BLOCK_GROUP_DUP
)))
3011 flags
&= ~BTRFS_BLOCK_GROUP_RAID0
;
3015 static u64
get_alloc_profile(struct btrfs_root
*root
, u64 flags
)
3017 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3018 flags
|= root
->fs_info
->avail_data_alloc_bits
&
3019 root
->fs_info
->data_alloc_profile
;
3020 else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3021 flags
|= root
->fs_info
->avail_system_alloc_bits
&
3022 root
->fs_info
->system_alloc_profile
;
3023 else if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3024 flags
|= root
->fs_info
->avail_metadata_alloc_bits
&
3025 root
->fs_info
->metadata_alloc_profile
;
3026 return btrfs_reduce_alloc_profile(root
, flags
);
3029 u64
btrfs_get_alloc_profile(struct btrfs_root
*root
, int data
)
3034 flags
= BTRFS_BLOCK_GROUP_DATA
;
3035 else if (root
== root
->fs_info
->chunk_root
)
3036 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
3038 flags
= BTRFS_BLOCK_GROUP_METADATA
;
3040 return get_alloc_profile(root
, flags
);
3043 void btrfs_set_inode_space_info(struct btrfs_root
*root
, struct inode
*inode
)
3045 BTRFS_I(inode
)->space_info
= __find_space_info(root
->fs_info
,
3046 BTRFS_BLOCK_GROUP_DATA
);
3050 * This will check the space that the inode allocates from to make sure we have
3051 * enough space for bytes.
3053 int btrfs_check_data_free_space(struct inode
*inode
, u64 bytes
)
3055 struct btrfs_space_info
*data_sinfo
;
3056 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3058 int ret
= 0, committed
= 0, alloc_chunk
= 1;
3060 /* make sure bytes are sectorsize aligned */
3061 bytes
= (bytes
+ root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
3063 if (root
== root
->fs_info
->tree_root
||
3064 BTRFS_I(inode
)->location
.objectid
== BTRFS_FREE_INO_OBJECTID
) {
3069 data_sinfo
= BTRFS_I(inode
)->space_info
;
3074 /* make sure we have enough space to handle the data first */
3075 spin_lock(&data_sinfo
->lock
);
3076 used
= data_sinfo
->bytes_used
+ data_sinfo
->bytes_reserved
+
3077 data_sinfo
->bytes_pinned
+ data_sinfo
->bytes_readonly
+
3078 data_sinfo
->bytes_may_use
;
3080 if (used
+ bytes
> data_sinfo
->total_bytes
) {
3081 struct btrfs_trans_handle
*trans
;
3084 * if we don't have enough free bytes in this space then we need
3085 * to alloc a new chunk.
3087 if (!data_sinfo
->full
&& alloc_chunk
) {
3090 data_sinfo
->force_alloc
= CHUNK_ALLOC_FORCE
;
3091 spin_unlock(&data_sinfo
->lock
);
3093 alloc_target
= btrfs_get_alloc_profile(root
, 1);
3094 trans
= btrfs_join_transaction(root
);
3096 return PTR_ERR(trans
);
3098 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
3099 bytes
+ 2 * 1024 * 1024,
3101 CHUNK_ALLOC_NO_FORCE
);
3102 btrfs_end_transaction(trans
, root
);
3111 btrfs_set_inode_space_info(root
, inode
);
3112 data_sinfo
= BTRFS_I(inode
)->space_info
;
3118 * If we have less pinned bytes than we want to allocate then
3119 * don't bother committing the transaction, it won't help us.
3121 if (data_sinfo
->bytes_pinned
< bytes
)
3123 spin_unlock(&data_sinfo
->lock
);
3125 /* commit the current transaction and try again */
3128 !atomic_read(&root
->fs_info
->open_ioctl_trans
)) {
3130 trans
= btrfs_join_transaction(root
);
3132 return PTR_ERR(trans
);
3133 ret
= btrfs_commit_transaction(trans
, root
);
3141 data_sinfo
->bytes_may_use
+= bytes
;
3142 spin_unlock(&data_sinfo
->lock
);
3148 * Called if we need to clear a data reservation for this inode.
3150 void btrfs_free_reserved_data_space(struct inode
*inode
, u64 bytes
)
3152 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3153 struct btrfs_space_info
*data_sinfo
;
3155 /* make sure bytes are sectorsize aligned */
3156 bytes
= (bytes
+ root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
3158 data_sinfo
= BTRFS_I(inode
)->space_info
;
3159 spin_lock(&data_sinfo
->lock
);
3160 data_sinfo
->bytes_may_use
-= bytes
;
3161 spin_unlock(&data_sinfo
->lock
);
3164 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
3166 struct list_head
*head
= &info
->space_info
;
3167 struct btrfs_space_info
*found
;
3170 list_for_each_entry_rcu(found
, head
, list
) {
3171 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
3172 found
->force_alloc
= CHUNK_ALLOC_FORCE
;
3177 static int should_alloc_chunk(struct btrfs_root
*root
,
3178 struct btrfs_space_info
*sinfo
, u64 alloc_bytes
,
3181 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
3182 u64 num_bytes
= sinfo
->total_bytes
- sinfo
->bytes_readonly
;
3183 u64 num_allocated
= sinfo
->bytes_used
+ sinfo
->bytes_reserved
;
3186 if (force
== CHUNK_ALLOC_FORCE
)
3190 * We need to take into account the global rsv because for all intents
3191 * and purposes it's used space. Don't worry about locking the
3192 * global_rsv, it doesn't change except when the transaction commits.
3194 num_allocated
+= global_rsv
->size
;
3197 * in limited mode, we want to have some free space up to
3198 * about 1% of the FS size.
3200 if (force
== CHUNK_ALLOC_LIMITED
) {
3201 thresh
= btrfs_super_total_bytes(&root
->fs_info
->super_copy
);
3202 thresh
= max_t(u64
, 64 * 1024 * 1024,
3203 div_factor_fine(thresh
, 1));
3205 if (num_bytes
- num_allocated
< thresh
)
3210 * we have two similar checks here, one based on percentage
3211 * and once based on a hard number of 256MB. The idea
3212 * is that if we have a good amount of free
3213 * room, don't allocate a chunk. A good mount is
3214 * less than 80% utilized of the chunks we have allocated,
3215 * or more than 256MB free
3217 if (num_allocated
+ alloc_bytes
+ 256 * 1024 * 1024 < num_bytes
)
3220 if (num_allocated
+ alloc_bytes
< div_factor(num_bytes
, 8))
3223 thresh
= btrfs_super_total_bytes(&root
->fs_info
->super_copy
);
3225 /* 256MB or 5% of the FS */
3226 thresh
= max_t(u64
, 256 * 1024 * 1024, div_factor_fine(thresh
, 5));
3228 if (num_bytes
> thresh
&& sinfo
->bytes_used
< div_factor(num_bytes
, 3))
3233 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
3234 struct btrfs_root
*extent_root
, u64 alloc_bytes
,
3235 u64 flags
, int force
)
3237 struct btrfs_space_info
*space_info
;
3238 struct btrfs_fs_info
*fs_info
= extent_root
->fs_info
;
3239 int wait_for_alloc
= 0;
3242 flags
= btrfs_reduce_alloc_profile(extent_root
, flags
);
3244 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
3246 ret
= update_space_info(extent_root
->fs_info
, flags
,
3250 BUG_ON(!space_info
);
3253 spin_lock(&space_info
->lock
);
3254 if (space_info
->force_alloc
)
3255 force
= space_info
->force_alloc
;
3256 if (space_info
->full
) {
3257 spin_unlock(&space_info
->lock
);
3261 if (!should_alloc_chunk(extent_root
, space_info
, alloc_bytes
, force
)) {
3262 spin_unlock(&space_info
->lock
);
3264 } else if (space_info
->chunk_alloc
) {
3267 space_info
->chunk_alloc
= 1;
3270 spin_unlock(&space_info
->lock
);
3272 mutex_lock(&fs_info
->chunk_mutex
);
3275 * The chunk_mutex is held throughout the entirety of a chunk
3276 * allocation, so once we've acquired the chunk_mutex we know that the
3277 * other guy is done and we need to recheck and see if we should
3280 if (wait_for_alloc
) {
3281 mutex_unlock(&fs_info
->chunk_mutex
);
3287 * If we have mixed data/metadata chunks we want to make sure we keep
3288 * allocating mixed chunks instead of individual chunks.
3290 if (btrfs_mixed_space_info(space_info
))
3291 flags
|= (BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
);
3294 * if we're doing a data chunk, go ahead and make sure that
3295 * we keep a reasonable number of metadata chunks allocated in the
3298 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
3299 fs_info
->data_chunk_allocations
++;
3300 if (!(fs_info
->data_chunk_allocations
%
3301 fs_info
->metadata_ratio
))
3302 force_metadata_allocation(fs_info
);
3305 ret
= btrfs_alloc_chunk(trans
, extent_root
, flags
);
3306 if (ret
< 0 && ret
!= -ENOSPC
)
3309 spin_lock(&space_info
->lock
);
3311 space_info
->full
= 1;
3315 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3316 space_info
->chunk_alloc
= 0;
3317 spin_unlock(&space_info
->lock
);
3319 mutex_unlock(&extent_root
->fs_info
->chunk_mutex
);
3324 * shrink metadata reservation for delalloc
3326 static int shrink_delalloc(struct btrfs_trans_handle
*trans
,
3327 struct btrfs_root
*root
, u64 to_reclaim
, int sync
)
3329 struct btrfs_block_rsv
*block_rsv
;
3330 struct btrfs_space_info
*space_info
;
3335 int nr_pages
= (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT
;
3337 unsigned long progress
;
3339 block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
3340 space_info
= block_rsv
->space_info
;
3343 reserved
= space_info
->bytes_may_use
;
3344 progress
= space_info
->reservation_progress
;
3350 if (root
->fs_info
->delalloc_bytes
== 0) {
3353 btrfs_wait_ordered_extents(root
, 0, 0);
3357 max_reclaim
= min(reserved
, to_reclaim
);
3359 while (loops
< 1024) {
3360 /* have the flusher threads jump in and do some IO */
3362 nr_pages
= min_t(unsigned long, nr_pages
,
3363 root
->fs_info
->delalloc_bytes
>> PAGE_CACHE_SHIFT
);
3364 writeback_inodes_sb_nr_if_idle(root
->fs_info
->sb
, nr_pages
);
3366 spin_lock(&space_info
->lock
);
3367 if (reserved
> space_info
->bytes_may_use
)
3368 reclaimed
+= reserved
- space_info
->bytes_may_use
;
3369 reserved
= space_info
->bytes_may_use
;
3370 spin_unlock(&space_info
->lock
);
3374 if (reserved
== 0 || reclaimed
>= max_reclaim
)
3377 if (trans
&& trans
->transaction
->blocked
)
3380 time_left
= schedule_timeout_interruptible(1);
3382 /* We were interrupted, exit */
3386 /* we've kicked the IO a few times, if anything has been freed,
3387 * exit. There is no sense in looping here for a long time
3388 * when we really need to commit the transaction, or there are
3389 * just too many writers without enough free space
3394 if (progress
!= space_info
->reservation_progress
)
3399 if (reclaimed
>= to_reclaim
&& !trans
)
3400 btrfs_wait_ordered_extents(root
, 0, 0);
3401 return reclaimed
>= to_reclaim
;
3405 * Retries tells us how many times we've called reserve_metadata_bytes. The
3406 * idea is if this is the first call (retries == 0) then we will add to our
3407 * reserved count if we can't make the allocation in order to hold our place
3408 * while we go and try and free up space. That way for retries > 1 we don't try
3409 * and add space, we just check to see if the amount of unused space is >= the
3410 * total space, meaning that our reservation is valid.
3412 * However if we don't intend to retry this reservation, pass -1 as retries so
3413 * that it short circuits this logic.
3415 static int reserve_metadata_bytes(struct btrfs_trans_handle
*trans
,
3416 struct btrfs_root
*root
,
3417 struct btrfs_block_rsv
*block_rsv
,
3418 u64 orig_bytes
, int flush
)
3420 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
3422 u64 num_bytes
= orig_bytes
;
3425 bool committed
= false;
3426 bool flushing
= false;
3429 spin_lock(&space_info
->lock
);
3431 * We only want to wait if somebody other than us is flushing and we are
3432 * actually alloed to flush.
3434 while (flush
&& !flushing
&& space_info
->flush
) {
3435 spin_unlock(&space_info
->lock
);
3437 * If we have a trans handle we can't wait because the flusher
3438 * may have to commit the transaction, which would mean we would
3439 * deadlock since we are waiting for the flusher to finish, but
3440 * hold the current transaction open.
3444 ret
= wait_event_interruptible(space_info
->wait
,
3445 !space_info
->flush
);
3446 /* Must have been interrupted, return */
3450 spin_lock(&space_info
->lock
);
3454 unused
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
3455 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
3456 space_info
->bytes_may_use
;
3459 * The idea here is that we've not already over-reserved the block group
3460 * then we can go ahead and save our reservation first and then start
3461 * flushing if we need to. Otherwise if we've already overcommitted
3462 * lets start flushing stuff first and then come back and try to make
3465 if (unused
<= space_info
->total_bytes
) {
3466 unused
= space_info
->total_bytes
- unused
;
3467 if (unused
>= num_bytes
) {
3468 space_info
->bytes_may_use
+= orig_bytes
;
3472 * Ok set num_bytes to orig_bytes since we aren't
3473 * overocmmitted, this way we only try and reclaim what
3476 num_bytes
= orig_bytes
;
3480 * Ok we're over committed, set num_bytes to the overcommitted
3481 * amount plus the amount of bytes that we need for this
3484 num_bytes
= unused
- space_info
->total_bytes
+
3485 (orig_bytes
* (retries
+ 1));
3489 * Couldn't make our reservation, save our place so while we're trying
3490 * to reclaim space we can actually use it instead of somebody else
3491 * stealing it from us.
3495 space_info
->flush
= 1;
3498 spin_unlock(&space_info
->lock
);
3504 * We do synchronous shrinking since we don't actually unreserve
3505 * metadata until after the IO is completed.
3507 ret
= shrink_delalloc(trans
, root
, num_bytes
, 1);
3514 * So if we were overcommitted it's possible that somebody else flushed
3515 * out enough space and we simply didn't have enough space to reclaim,
3516 * so go back around and try again.
3524 * Not enough space to be reclaimed, don't bother committing the
3527 spin_lock(&space_info
->lock
);
3528 if (space_info
->bytes_pinned
< orig_bytes
)
3530 spin_unlock(&space_info
->lock
);
3542 trans
= btrfs_join_transaction(root
);
3545 ret
= btrfs_commit_transaction(trans
, root
);
3554 spin_lock(&space_info
->lock
);
3555 space_info
->flush
= 0;
3556 wake_up_all(&space_info
->wait
);
3557 spin_unlock(&space_info
->lock
);
3562 static struct btrfs_block_rsv
*get_block_rsv(struct btrfs_trans_handle
*trans
,
3563 struct btrfs_root
*root
)
3565 struct btrfs_block_rsv
*block_rsv
;
3567 block_rsv
= trans
->block_rsv
;
3569 block_rsv
= root
->block_rsv
;
3572 block_rsv
= &root
->fs_info
->empty_block_rsv
;
3577 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
3581 spin_lock(&block_rsv
->lock
);
3582 if (block_rsv
->reserved
>= num_bytes
) {
3583 block_rsv
->reserved
-= num_bytes
;
3584 if (block_rsv
->reserved
< block_rsv
->size
)
3585 block_rsv
->full
= 0;
3588 spin_unlock(&block_rsv
->lock
);
3592 static void block_rsv_add_bytes(struct btrfs_block_rsv
*block_rsv
,
3593 u64 num_bytes
, int update_size
)
3595 spin_lock(&block_rsv
->lock
);
3596 block_rsv
->reserved
+= num_bytes
;
3598 block_rsv
->size
+= num_bytes
;
3599 else if (block_rsv
->reserved
>= block_rsv
->size
)
3600 block_rsv
->full
= 1;
3601 spin_unlock(&block_rsv
->lock
);
3604 static void block_rsv_release_bytes(struct btrfs_block_rsv
*block_rsv
,
3605 struct btrfs_block_rsv
*dest
, u64 num_bytes
)
3607 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
3609 spin_lock(&block_rsv
->lock
);
3610 if (num_bytes
== (u64
)-1)
3611 num_bytes
= block_rsv
->size
;
3612 block_rsv
->size
-= num_bytes
;
3613 if (block_rsv
->reserved
>= block_rsv
->size
) {
3614 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
3615 block_rsv
->reserved
= block_rsv
->size
;
3616 block_rsv
->full
= 1;
3620 spin_unlock(&block_rsv
->lock
);
3622 if (num_bytes
> 0) {
3624 spin_lock(&dest
->lock
);
3628 bytes_to_add
= dest
->size
- dest
->reserved
;
3629 bytes_to_add
= min(num_bytes
, bytes_to_add
);
3630 dest
->reserved
+= bytes_to_add
;
3631 if (dest
->reserved
>= dest
->size
)
3633 num_bytes
-= bytes_to_add
;
3635 spin_unlock(&dest
->lock
);
3638 spin_lock(&space_info
->lock
);
3639 space_info
->bytes_may_use
-= num_bytes
;
3640 space_info
->reservation_progress
++;
3641 spin_unlock(&space_info
->lock
);
3646 static int block_rsv_migrate_bytes(struct btrfs_block_rsv
*src
,
3647 struct btrfs_block_rsv
*dst
, u64 num_bytes
)
3651 ret
= block_rsv_use_bytes(src
, num_bytes
);
3655 block_rsv_add_bytes(dst
, num_bytes
, 1);
3659 void btrfs_init_block_rsv(struct btrfs_block_rsv
*rsv
)
3661 memset(rsv
, 0, sizeof(*rsv
));
3662 spin_lock_init(&rsv
->lock
);
3663 atomic_set(&rsv
->usage
, 1);
3665 INIT_LIST_HEAD(&rsv
->list
);
3668 struct btrfs_block_rsv
*btrfs_alloc_block_rsv(struct btrfs_root
*root
)
3670 struct btrfs_block_rsv
*block_rsv
;
3671 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3673 block_rsv
= kmalloc(sizeof(*block_rsv
), GFP_NOFS
);
3677 btrfs_init_block_rsv(block_rsv
);
3678 block_rsv
->space_info
= __find_space_info(fs_info
,
3679 BTRFS_BLOCK_GROUP_METADATA
);
3683 void btrfs_free_block_rsv(struct btrfs_root
*root
,
3684 struct btrfs_block_rsv
*rsv
)
3686 if (rsv
&& atomic_dec_and_test(&rsv
->usage
)) {
3687 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
3694 * make the block_rsv struct be able to capture freed space.
3695 * the captured space will re-add to the the block_rsv struct
3696 * after transaction commit
3698 void btrfs_add_durable_block_rsv(struct btrfs_fs_info
*fs_info
,
3699 struct btrfs_block_rsv
*block_rsv
)
3701 block_rsv
->durable
= 1;
3702 mutex_lock(&fs_info
->durable_block_rsv_mutex
);
3703 list_add_tail(&block_rsv
->list
, &fs_info
->durable_block_rsv_list
);
3704 mutex_unlock(&fs_info
->durable_block_rsv_mutex
);
3707 int btrfs_block_rsv_add(struct btrfs_trans_handle
*trans
,
3708 struct btrfs_root
*root
,
3709 struct btrfs_block_rsv
*block_rsv
,
3717 ret
= reserve_metadata_bytes(trans
, root
, block_rsv
, num_bytes
, 1);
3719 block_rsv_add_bytes(block_rsv
, num_bytes
, 1);
3726 int btrfs_block_rsv_check(struct btrfs_trans_handle
*trans
,
3727 struct btrfs_root
*root
,
3728 struct btrfs_block_rsv
*block_rsv
,
3729 u64 min_reserved
, int min_factor
)
3732 int commit_trans
= 0;
3738 spin_lock(&block_rsv
->lock
);
3740 num_bytes
= div_factor(block_rsv
->size
, min_factor
);
3741 if (min_reserved
> num_bytes
)
3742 num_bytes
= min_reserved
;
3744 if (block_rsv
->reserved
>= num_bytes
) {
3747 num_bytes
-= block_rsv
->reserved
;
3748 if (block_rsv
->durable
&&
3749 block_rsv
->freed
[0] + block_rsv
->freed
[1] >= num_bytes
)
3752 spin_unlock(&block_rsv
->lock
);
3756 if (block_rsv
->refill_used
) {
3757 ret
= reserve_metadata_bytes(trans
, root
, block_rsv
,
3760 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
3768 trans
= btrfs_join_transaction(root
);
3769 BUG_ON(IS_ERR(trans
));
3770 ret
= btrfs_commit_transaction(trans
, root
);
3777 int btrfs_block_rsv_migrate(struct btrfs_block_rsv
*src_rsv
,
3778 struct btrfs_block_rsv
*dst_rsv
,
3781 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
3784 void btrfs_block_rsv_release(struct btrfs_root
*root
,
3785 struct btrfs_block_rsv
*block_rsv
,
3788 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
3789 if (global_rsv
->full
|| global_rsv
== block_rsv
||
3790 block_rsv
->space_info
!= global_rsv
->space_info
)
3792 block_rsv_release_bytes(block_rsv
, global_rsv
, num_bytes
);
3796 * helper to calculate size of global block reservation.
3797 * the desired value is sum of space used by extent tree,
3798 * checksum tree and root tree
3800 static u64
calc_global_metadata_size(struct btrfs_fs_info
*fs_info
)
3802 struct btrfs_space_info
*sinfo
;
3806 int csum_size
= btrfs_super_csum_size(&fs_info
->super_copy
);
3808 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_DATA
);
3809 spin_lock(&sinfo
->lock
);
3810 data_used
= sinfo
->bytes_used
;
3811 spin_unlock(&sinfo
->lock
);
3813 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
3814 spin_lock(&sinfo
->lock
);
3815 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_DATA
)
3817 meta_used
= sinfo
->bytes_used
;
3818 spin_unlock(&sinfo
->lock
);
3820 num_bytes
= (data_used
>> fs_info
->sb
->s_blocksize_bits
) *
3822 num_bytes
+= div64_u64(data_used
+ meta_used
, 50);
3824 if (num_bytes
* 3 > meta_used
)
3825 num_bytes
= div64_u64(meta_used
, 3);
3827 return ALIGN(num_bytes
, fs_info
->extent_root
->leafsize
<< 10);
3830 static void update_global_block_rsv(struct btrfs_fs_info
*fs_info
)
3832 struct btrfs_block_rsv
*block_rsv
= &fs_info
->global_block_rsv
;
3833 struct btrfs_space_info
*sinfo
= block_rsv
->space_info
;
3836 num_bytes
= calc_global_metadata_size(fs_info
);
3838 spin_lock(&block_rsv
->lock
);
3839 spin_lock(&sinfo
->lock
);
3841 block_rsv
->size
= num_bytes
;
3843 num_bytes
= sinfo
->bytes_used
+ sinfo
->bytes_pinned
+
3844 sinfo
->bytes_reserved
+ sinfo
->bytes_readonly
+
3845 sinfo
->bytes_may_use
;
3847 if (sinfo
->total_bytes
> num_bytes
) {
3848 num_bytes
= sinfo
->total_bytes
- num_bytes
;
3849 block_rsv
->reserved
+= num_bytes
;
3850 sinfo
->bytes_may_use
+= num_bytes
;
3853 if (block_rsv
->reserved
>= block_rsv
->size
) {
3854 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
3855 sinfo
->bytes_may_use
-= num_bytes
;
3856 sinfo
->reservation_progress
++;
3857 block_rsv
->reserved
= block_rsv
->size
;
3858 block_rsv
->full
= 1;
3861 spin_unlock(&sinfo
->lock
);
3862 spin_unlock(&block_rsv
->lock
);
3865 static void init_global_block_rsv(struct btrfs_fs_info
*fs_info
)
3867 struct btrfs_space_info
*space_info
;
3869 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
3870 fs_info
->chunk_block_rsv
.space_info
= space_info
;
3871 fs_info
->chunk_block_rsv
.priority
= 10;
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
->global_block_rsv
.priority
= 10;
3876 fs_info
->global_block_rsv
.refill_used
= 1;
3877 fs_info
->delalloc_block_rsv
.space_info
= space_info
;
3878 fs_info
->trans_block_rsv
.space_info
= space_info
;
3879 fs_info
->empty_block_rsv
.space_info
= space_info
;
3880 fs_info
->empty_block_rsv
.priority
= 10;
3882 fs_info
->extent_root
->block_rsv
= &fs_info
->global_block_rsv
;
3883 fs_info
->csum_root
->block_rsv
= &fs_info
->global_block_rsv
;
3884 fs_info
->dev_root
->block_rsv
= &fs_info
->global_block_rsv
;
3885 fs_info
->tree_root
->block_rsv
= &fs_info
->global_block_rsv
;
3886 fs_info
->chunk_root
->block_rsv
= &fs_info
->chunk_block_rsv
;
3888 btrfs_add_durable_block_rsv(fs_info
, &fs_info
->global_block_rsv
);
3890 btrfs_add_durable_block_rsv(fs_info
, &fs_info
->delalloc_block_rsv
);
3892 update_global_block_rsv(fs_info
);
3895 static void release_global_block_rsv(struct btrfs_fs_info
*fs_info
)
3897 block_rsv_release_bytes(&fs_info
->global_block_rsv
, NULL
, (u64
)-1);
3898 WARN_ON(fs_info
->delalloc_block_rsv
.size
> 0);
3899 WARN_ON(fs_info
->delalloc_block_rsv
.reserved
> 0);
3900 WARN_ON(fs_info
->trans_block_rsv
.size
> 0);
3901 WARN_ON(fs_info
->trans_block_rsv
.reserved
> 0);
3902 WARN_ON(fs_info
->chunk_block_rsv
.size
> 0);
3903 WARN_ON(fs_info
->chunk_block_rsv
.reserved
> 0);
3906 int btrfs_truncate_reserve_metadata(struct btrfs_trans_handle
*trans
,
3907 struct btrfs_root
*root
,
3908 struct btrfs_block_rsv
*rsv
)
3910 struct btrfs_block_rsv
*trans_rsv
= &root
->fs_info
->trans_block_rsv
;
3915 * Truncate should be freeing data, but give us 2 items just in case it
3916 * needs to use some space. We may want to be smarter about this in the
3919 num_bytes
= btrfs_calc_trans_metadata_size(root
, 2);
3921 /* We already have enough bytes, just return */
3922 if (rsv
->reserved
>= num_bytes
)
3925 num_bytes
-= rsv
->reserved
;
3928 * You should have reserved enough space before hand to do this, so this
3931 ret
= block_rsv_migrate_bytes(trans_rsv
, rsv
, num_bytes
);
3937 void btrfs_trans_release_metadata(struct btrfs_trans_handle
*trans
,
3938 struct btrfs_root
*root
)
3940 if (!trans
->bytes_reserved
)
3943 BUG_ON(trans
->block_rsv
!= &root
->fs_info
->trans_block_rsv
);
3944 btrfs_block_rsv_release(root
, trans
->block_rsv
,
3945 trans
->bytes_reserved
);
3946 trans
->bytes_reserved
= 0;
3949 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle
*trans
,
3950 struct inode
*inode
)
3952 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3953 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
3954 struct btrfs_block_rsv
*dst_rsv
= root
->orphan_block_rsv
;
3957 * We need to hold space in order to delete our orphan item once we've
3958 * added it, so this takes the reservation so we can release it later
3959 * when we are truly done with the orphan item.
3961 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
3962 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
3965 void btrfs_orphan_release_metadata(struct inode
*inode
)
3967 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3968 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
3969 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
, num_bytes
);
3972 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle
*trans
,
3973 struct btrfs_pending_snapshot
*pending
)
3975 struct btrfs_root
*root
= pending
->root
;
3976 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
3977 struct btrfs_block_rsv
*dst_rsv
= &pending
->block_rsv
;
3979 * two for root back/forward refs, two for directory entries
3980 * and one for root of the snapshot.
3982 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 5);
3983 dst_rsv
->space_info
= src_rsv
->space_info
;
3984 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
3987 static unsigned drop_outstanding_extent(struct inode
*inode
)
3989 unsigned dropped_extents
= 0;
3991 spin_lock(&BTRFS_I(inode
)->lock
);
3992 BUG_ON(!BTRFS_I(inode
)->outstanding_extents
);
3993 BTRFS_I(inode
)->outstanding_extents
--;
3996 * If we have more or the same amount of outsanding extents than we have
3997 * reserved then we need to leave the reserved extents count alone.
3999 if (BTRFS_I(inode
)->outstanding_extents
>=
4000 BTRFS_I(inode
)->reserved_extents
)
4003 dropped_extents
= BTRFS_I(inode
)->reserved_extents
-
4004 BTRFS_I(inode
)->outstanding_extents
;
4005 BTRFS_I(inode
)->reserved_extents
-= dropped_extents
;
4007 spin_unlock(&BTRFS_I(inode
)->lock
);
4008 return dropped_extents
;
4011 static u64
calc_csum_metadata_size(struct inode
*inode
, u64 num_bytes
)
4013 return num_bytes
>>= 3;
4016 int btrfs_delalloc_reserve_metadata(struct inode
*inode
, u64 num_bytes
)
4018 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4019 struct btrfs_block_rsv
*block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
4021 unsigned nr_extents
= 0;
4024 if (btrfs_transaction_in_commit(root
->fs_info
))
4025 schedule_timeout(1);
4027 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
4029 spin_lock(&BTRFS_I(inode
)->lock
);
4030 BTRFS_I(inode
)->outstanding_extents
++;
4032 if (BTRFS_I(inode
)->outstanding_extents
>
4033 BTRFS_I(inode
)->reserved_extents
) {
4034 nr_extents
= BTRFS_I(inode
)->outstanding_extents
-
4035 BTRFS_I(inode
)->reserved_extents
;
4036 BTRFS_I(inode
)->reserved_extents
+= nr_extents
;
4038 to_reserve
= btrfs_calc_trans_metadata_size(root
, nr_extents
);
4040 spin_unlock(&BTRFS_I(inode
)->lock
);
4042 to_reserve
+= calc_csum_metadata_size(inode
, num_bytes
);
4043 ret
= reserve_metadata_bytes(NULL
, root
, block_rsv
, to_reserve
, 1);
4047 * We don't need the return value since our reservation failed,
4048 * we just need to clean up our counter.
4050 dropped
= drop_outstanding_extent(inode
);
4051 WARN_ON(dropped
> 1);
4055 block_rsv_add_bytes(block_rsv
, to_reserve
, 1);
4060 void btrfs_delalloc_release_metadata(struct inode
*inode
, u64 num_bytes
)
4062 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4066 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
4067 dropped
= drop_outstanding_extent(inode
);
4069 to_free
= calc_csum_metadata_size(inode
, num_bytes
);
4071 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
4073 btrfs_block_rsv_release(root
, &root
->fs_info
->delalloc_block_rsv
,
4077 int btrfs_delalloc_reserve_space(struct inode
*inode
, u64 num_bytes
)
4081 ret
= btrfs_check_data_free_space(inode
, num_bytes
);
4085 ret
= btrfs_delalloc_reserve_metadata(inode
, num_bytes
);
4087 btrfs_free_reserved_data_space(inode
, num_bytes
);
4094 void btrfs_delalloc_release_space(struct inode
*inode
, u64 num_bytes
)
4096 btrfs_delalloc_release_metadata(inode
, num_bytes
);
4097 btrfs_free_reserved_data_space(inode
, num_bytes
);
4100 static int update_block_group(struct btrfs_trans_handle
*trans
,
4101 struct btrfs_root
*root
,
4102 u64 bytenr
, u64 num_bytes
, int alloc
)
4104 struct btrfs_block_group_cache
*cache
= NULL
;
4105 struct btrfs_fs_info
*info
= root
->fs_info
;
4106 u64 total
= num_bytes
;
4111 /* block accounting for super block */
4112 spin_lock(&info
->delalloc_lock
);
4113 old_val
= btrfs_super_bytes_used(&info
->super_copy
);
4115 old_val
+= num_bytes
;
4117 old_val
-= num_bytes
;
4118 btrfs_set_super_bytes_used(&info
->super_copy
, old_val
);
4119 spin_unlock(&info
->delalloc_lock
);
4122 cache
= btrfs_lookup_block_group(info
, bytenr
);
4125 if (cache
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
4126 BTRFS_BLOCK_GROUP_RAID1
|
4127 BTRFS_BLOCK_GROUP_RAID10
))
4132 * If this block group has free space cache written out, we
4133 * need to make sure to load it if we are removing space. This
4134 * is because we need the unpinning stage to actually add the
4135 * space back to the block group, otherwise we will leak space.
4137 if (!alloc
&& cache
->cached
== BTRFS_CACHE_NO
)
4138 cache_block_group(cache
, trans
, NULL
, 1);
4140 byte_in_group
= bytenr
- cache
->key
.objectid
;
4141 WARN_ON(byte_in_group
> cache
->key
.offset
);
4143 spin_lock(&cache
->space_info
->lock
);
4144 spin_lock(&cache
->lock
);
4146 if (btrfs_super_cache_generation(&info
->super_copy
) != 0 &&
4147 cache
->disk_cache_state
< BTRFS_DC_CLEAR
)
4148 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
4151 old_val
= btrfs_block_group_used(&cache
->item
);
4152 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
4154 old_val
+= num_bytes
;
4155 btrfs_set_block_group_used(&cache
->item
, old_val
);
4156 cache
->reserved
-= num_bytes
;
4157 cache
->space_info
->bytes_reserved
-= num_bytes
;
4158 cache
->space_info
->bytes_used
+= num_bytes
;
4159 cache
->space_info
->disk_used
+= num_bytes
* factor
;
4160 spin_unlock(&cache
->lock
);
4161 spin_unlock(&cache
->space_info
->lock
);
4163 old_val
-= num_bytes
;
4164 btrfs_set_block_group_used(&cache
->item
, old_val
);
4165 cache
->pinned
+= num_bytes
;
4166 cache
->space_info
->bytes_pinned
+= num_bytes
;
4167 cache
->space_info
->bytes_used
-= num_bytes
;
4168 cache
->space_info
->disk_used
-= num_bytes
* factor
;
4169 spin_unlock(&cache
->lock
);
4170 spin_unlock(&cache
->space_info
->lock
);
4172 set_extent_dirty(info
->pinned_extents
,
4173 bytenr
, bytenr
+ num_bytes
- 1,
4174 GFP_NOFS
| __GFP_NOFAIL
);
4176 btrfs_put_block_group(cache
);
4178 bytenr
+= num_bytes
;
4183 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
4185 struct btrfs_block_group_cache
*cache
;
4188 cache
= btrfs_lookup_first_block_group(root
->fs_info
, search_start
);
4192 bytenr
= cache
->key
.objectid
;
4193 btrfs_put_block_group(cache
);
4198 static int pin_down_extent(struct btrfs_root
*root
,
4199 struct btrfs_block_group_cache
*cache
,
4200 u64 bytenr
, u64 num_bytes
, int reserved
)
4202 spin_lock(&cache
->space_info
->lock
);
4203 spin_lock(&cache
->lock
);
4204 cache
->pinned
+= num_bytes
;
4205 cache
->space_info
->bytes_pinned
+= num_bytes
;
4207 cache
->reserved
-= num_bytes
;
4208 cache
->space_info
->bytes_reserved
-= num_bytes
;
4210 spin_unlock(&cache
->lock
);
4211 spin_unlock(&cache
->space_info
->lock
);
4213 set_extent_dirty(root
->fs_info
->pinned_extents
, bytenr
,
4214 bytenr
+ num_bytes
- 1, GFP_NOFS
| __GFP_NOFAIL
);
4219 * this function must be called within transaction
4221 int btrfs_pin_extent(struct btrfs_root
*root
,
4222 u64 bytenr
, u64 num_bytes
, int reserved
)
4224 struct btrfs_block_group_cache
*cache
;
4226 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
4229 pin_down_extent(root
, cache
, bytenr
, num_bytes
, reserved
);
4231 btrfs_put_block_group(cache
);
4236 * btrfs_update_reserved_bytes - update the block_group and space info counters
4237 * @cache: The cache we are manipulating
4238 * @num_bytes: The number of bytes in question
4239 * @reserve: One of the reservation enums
4241 * This is called by the allocator when it reserves space, or by somebody who is
4242 * freeing space that was never actually used on disk. For example if you
4243 * reserve some space for a new leaf in transaction A and before transaction A
4244 * commits you free that leaf, you call this with reserve set to 0 in order to
4245 * clear the reservation.
4247 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4248 * ENOSPC accounting. For data we handle the reservation through clearing the
4249 * delalloc bits in the io_tree. We have to do this since we could end up
4250 * allocating less disk space for the amount of data we have reserved in the
4251 * case of compression.
4253 * If this is a reservation and the block group has become read only we cannot
4254 * make the reservation and return -EAGAIN, otherwise this function always
4257 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
4258 u64 num_bytes
, int reserve
)
4260 struct btrfs_space_info
*space_info
= cache
->space_info
;
4262 spin_lock(&space_info
->lock
);
4263 spin_lock(&cache
->lock
);
4264 if (reserve
!= RESERVE_FREE
) {
4268 cache
->reserved
+= num_bytes
;
4269 space_info
->bytes_reserved
+= num_bytes
;
4270 if (reserve
== RESERVE_ALLOC
) {
4271 BUG_ON(space_info
->bytes_may_use
< num_bytes
);
4272 space_info
->bytes_may_use
-= num_bytes
;
4277 space_info
->bytes_readonly
+= num_bytes
;
4278 cache
->reserved
-= num_bytes
;
4279 space_info
->bytes_reserved
-= num_bytes
;
4280 space_info
->reservation_progress
++;
4282 spin_unlock(&cache
->lock
);
4283 spin_unlock(&space_info
->lock
);
4287 int btrfs_prepare_extent_commit(struct btrfs_trans_handle
*trans
,
4288 struct btrfs_root
*root
)
4290 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4291 struct btrfs_caching_control
*next
;
4292 struct btrfs_caching_control
*caching_ctl
;
4293 struct btrfs_block_group_cache
*cache
;
4295 down_write(&fs_info
->extent_commit_sem
);
4297 list_for_each_entry_safe(caching_ctl
, next
,
4298 &fs_info
->caching_block_groups
, list
) {
4299 cache
= caching_ctl
->block_group
;
4300 if (block_group_cache_done(cache
)) {
4301 cache
->last_byte_to_unpin
= (u64
)-1;
4302 list_del_init(&caching_ctl
->list
);
4303 put_caching_control(caching_ctl
);
4305 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
4309 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
4310 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
4312 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
4314 up_write(&fs_info
->extent_commit_sem
);
4316 update_global_block_rsv(fs_info
);
4320 static int unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
4322 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4323 struct btrfs_block_group_cache
*cache
= NULL
;
4326 while (start
<= end
) {
4328 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
4330 btrfs_put_block_group(cache
);
4331 cache
= btrfs_lookup_block_group(fs_info
, start
);
4335 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
4336 len
= min(len
, end
+ 1 - start
);
4338 if (start
< cache
->last_byte_to_unpin
) {
4339 len
= min(len
, cache
->last_byte_to_unpin
- start
);
4340 btrfs_add_free_space(cache
, start
, len
);
4345 spin_lock(&cache
->space_info
->lock
);
4346 spin_lock(&cache
->lock
);
4347 cache
->pinned
-= len
;
4348 cache
->space_info
->bytes_pinned
-= len
;
4350 cache
->space_info
->bytes_readonly
+= len
;
4351 } else if (cache
->reserved_pinned
> 0) {
4352 len
= min(len
, cache
->reserved_pinned
);
4353 cache
->reserved_pinned
-= len
;
4354 cache
->space_info
->bytes_may_use
+= len
;
4356 spin_unlock(&cache
->lock
);
4357 spin_unlock(&cache
->space_info
->lock
);
4361 btrfs_put_block_group(cache
);
4365 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
4366 struct btrfs_root
*root
)
4368 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4369 struct extent_io_tree
*unpin
;
4370 struct btrfs_block_rsv
*block_rsv
;
4371 struct btrfs_block_rsv
*next_rsv
;
4377 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
4378 unpin
= &fs_info
->freed_extents
[1];
4380 unpin
= &fs_info
->freed_extents
[0];
4383 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
4388 if (btrfs_test_opt(root
, DISCARD
))
4389 ret
= btrfs_discard_extent(root
, start
,
4390 end
+ 1 - start
, NULL
);
4392 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
4393 unpin_extent_range(root
, start
, end
);
4397 mutex_lock(&fs_info
->durable_block_rsv_mutex
);
4398 list_for_each_entry_safe(block_rsv
, next_rsv
,
4399 &fs_info
->durable_block_rsv_list
, list
) {
4401 idx
= trans
->transid
& 0x1;
4402 if (block_rsv
->freed
[idx
] > 0) {
4403 block_rsv_add_bytes(block_rsv
,
4404 block_rsv
->freed
[idx
], 0);
4405 block_rsv
->freed
[idx
] = 0;
4407 if (atomic_read(&block_rsv
->usage
) == 0) {
4408 btrfs_block_rsv_release(root
, block_rsv
, (u64
)-1);
4410 if (block_rsv
->freed
[0] == 0 &&
4411 block_rsv
->freed
[1] == 0) {
4412 list_del_init(&block_rsv
->list
);
4416 btrfs_block_rsv_release(root
, block_rsv
, 0);
4419 mutex_unlock(&fs_info
->durable_block_rsv_mutex
);
4424 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
4425 struct btrfs_root
*root
,
4426 u64 bytenr
, u64 num_bytes
, u64 parent
,
4427 u64 root_objectid
, u64 owner_objectid
,
4428 u64 owner_offset
, int refs_to_drop
,
4429 struct btrfs_delayed_extent_op
*extent_op
)
4431 struct btrfs_key key
;
4432 struct btrfs_path
*path
;
4433 struct btrfs_fs_info
*info
= root
->fs_info
;
4434 struct btrfs_root
*extent_root
= info
->extent_root
;
4435 struct extent_buffer
*leaf
;
4436 struct btrfs_extent_item
*ei
;
4437 struct btrfs_extent_inline_ref
*iref
;
4440 int extent_slot
= 0;
4441 int found_extent
= 0;
4446 path
= btrfs_alloc_path();
4451 path
->leave_spinning
= 1;
4453 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
4454 BUG_ON(!is_data
&& refs_to_drop
!= 1);
4456 ret
= lookup_extent_backref(trans
, extent_root
, path
, &iref
,
4457 bytenr
, num_bytes
, parent
,
4458 root_objectid
, owner_objectid
,
4461 extent_slot
= path
->slots
[0];
4462 while (extent_slot
>= 0) {
4463 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
4465 if (key
.objectid
!= bytenr
)
4467 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
4468 key
.offset
== num_bytes
) {
4472 if (path
->slots
[0] - extent_slot
> 5)
4476 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4477 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
4478 if (found_extent
&& item_size
< sizeof(*ei
))
4481 if (!found_extent
) {
4483 ret
= remove_extent_backref(trans
, extent_root
, path
,
4487 btrfs_release_path(path
);
4488 path
->leave_spinning
= 1;
4490 key
.objectid
= bytenr
;
4491 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
4492 key
.offset
= num_bytes
;
4494 ret
= btrfs_search_slot(trans
, extent_root
,
4497 printk(KERN_ERR
"umm, got %d back from search"
4498 ", was looking for %llu\n", ret
,
4499 (unsigned long long)bytenr
);
4501 btrfs_print_leaf(extent_root
,
4505 extent_slot
= path
->slots
[0];
4508 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
4510 printk(KERN_ERR
"btrfs unable to find ref byte nr %llu "
4511 "parent %llu root %llu owner %llu offset %llu\n",
4512 (unsigned long long)bytenr
,
4513 (unsigned long long)parent
,
4514 (unsigned long long)root_objectid
,
4515 (unsigned long long)owner_objectid
,
4516 (unsigned long long)owner_offset
);
4519 leaf
= path
->nodes
[0];
4520 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
4521 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4522 if (item_size
< sizeof(*ei
)) {
4523 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
4524 ret
= convert_extent_item_v0(trans
, extent_root
, path
,
4528 btrfs_release_path(path
);
4529 path
->leave_spinning
= 1;
4531 key
.objectid
= bytenr
;
4532 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
4533 key
.offset
= num_bytes
;
4535 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
4538 printk(KERN_ERR
"umm, got %d back from search"
4539 ", was looking for %llu\n", ret
,
4540 (unsigned long long)bytenr
);
4541 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
4544 extent_slot
= path
->slots
[0];
4545 leaf
= path
->nodes
[0];
4546 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
4549 BUG_ON(item_size
< sizeof(*ei
));
4550 ei
= btrfs_item_ptr(leaf
, extent_slot
,
4551 struct btrfs_extent_item
);
4552 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
) {
4553 struct btrfs_tree_block_info
*bi
;
4554 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
4555 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
4556 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
4559 refs
= btrfs_extent_refs(leaf
, ei
);
4560 BUG_ON(refs
< refs_to_drop
);
4561 refs
-= refs_to_drop
;
4565 __run_delayed_extent_op(extent_op
, leaf
, ei
);
4567 * In the case of inline back ref, reference count will
4568 * be updated by remove_extent_backref
4571 BUG_ON(!found_extent
);
4573 btrfs_set_extent_refs(leaf
, ei
, refs
);
4574 btrfs_mark_buffer_dirty(leaf
);
4577 ret
= remove_extent_backref(trans
, extent_root
, path
,
4584 BUG_ON(is_data
&& refs_to_drop
!=
4585 extent_data_ref_count(root
, path
, iref
));
4587 BUG_ON(path
->slots
[0] != extent_slot
);
4589 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
4590 path
->slots
[0] = extent_slot
;
4595 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
4598 btrfs_release_path(path
);
4601 ret
= btrfs_del_csums(trans
, root
, bytenr
, num_bytes
);
4604 invalidate_mapping_pages(info
->btree_inode
->i_mapping
,
4605 bytenr
>> PAGE_CACHE_SHIFT
,
4606 (bytenr
+ num_bytes
- 1) >> PAGE_CACHE_SHIFT
);
4609 ret
= update_block_group(trans
, root
, bytenr
, num_bytes
, 0);
4612 btrfs_free_path(path
);
4617 * when we free an block, it is possible (and likely) that we free the last
4618 * delayed ref for that extent as well. This searches the delayed ref tree for
4619 * a given extent, and if there are no other delayed refs to be processed, it
4620 * removes it from the tree.
4622 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
4623 struct btrfs_root
*root
, u64 bytenr
)
4625 struct btrfs_delayed_ref_head
*head
;
4626 struct btrfs_delayed_ref_root
*delayed_refs
;
4627 struct btrfs_delayed_ref_node
*ref
;
4628 struct rb_node
*node
;
4631 delayed_refs
= &trans
->transaction
->delayed_refs
;
4632 spin_lock(&delayed_refs
->lock
);
4633 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
4637 node
= rb_prev(&head
->node
.rb_node
);
4641 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
4643 /* there are still entries for this ref, we can't drop it */
4644 if (ref
->bytenr
== bytenr
)
4647 if (head
->extent_op
) {
4648 if (!head
->must_insert_reserved
)
4650 kfree(head
->extent_op
);
4651 head
->extent_op
= NULL
;
4655 * waiting for the lock here would deadlock. If someone else has it
4656 * locked they are already in the process of dropping it anyway
4658 if (!mutex_trylock(&head
->mutex
))
4662 * at this point we have a head with no other entries. Go
4663 * ahead and process it.
4665 head
->node
.in_tree
= 0;
4666 rb_erase(&head
->node
.rb_node
, &delayed_refs
->root
);
4668 delayed_refs
->num_entries
--;
4671 * we don't take a ref on the node because we're removing it from the
4672 * tree, so we just steal the ref the tree was holding.
4674 delayed_refs
->num_heads
--;
4675 if (list_empty(&head
->cluster
))
4676 delayed_refs
->num_heads_ready
--;
4678 list_del_init(&head
->cluster
);
4679 spin_unlock(&delayed_refs
->lock
);
4681 BUG_ON(head
->extent_op
);
4682 if (head
->must_insert_reserved
)
4685 mutex_unlock(&head
->mutex
);
4686 btrfs_put_delayed_ref(&head
->node
);
4689 spin_unlock(&delayed_refs
->lock
);
4693 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
4694 struct btrfs_root
*root
,
4695 struct extent_buffer
*buf
,
4696 u64 parent
, int last_ref
)
4698 struct btrfs_block_rsv
*block_rsv
;
4699 struct btrfs_block_group_cache
*cache
= NULL
;
4702 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
4703 ret
= btrfs_add_delayed_tree_ref(trans
, buf
->start
, buf
->len
,
4704 parent
, root
->root_key
.objectid
,
4705 btrfs_header_level(buf
),
4706 BTRFS_DROP_DELAYED_REF
, NULL
);
4713 block_rsv
= get_block_rsv(trans
, root
);
4714 cache
= btrfs_lookup_block_group(root
->fs_info
, buf
->start
);
4715 if (block_rsv
->space_info
!= cache
->space_info
)
4718 if (btrfs_header_generation(buf
) == trans
->transid
) {
4719 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
4720 ret
= check_ref_cleanup(trans
, root
, buf
->start
);
4725 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
4726 pin_down_extent(root
, cache
, buf
->start
, buf
->len
, 1);
4730 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
4732 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
4733 btrfs_update_reserved_bytes(cache
, buf
->len
, RESERVE_FREE
);
4738 if (block_rsv
->durable
&& !cache
->ro
) {
4740 spin_lock(&cache
->lock
);
4742 cache
->reserved_pinned
+= buf
->len
;
4745 spin_unlock(&cache
->lock
);
4748 spin_lock(&block_rsv
->lock
);
4749 block_rsv
->freed
[trans
->transid
& 0x1] += buf
->len
;
4750 spin_unlock(&block_rsv
->lock
);
4755 * Deleting the buffer, clear the corrupt flag since it doesn't matter
4758 clear_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
);
4759 btrfs_put_block_group(cache
);
4762 int btrfs_free_extent(struct btrfs_trans_handle
*trans
,
4763 struct btrfs_root
*root
,
4764 u64 bytenr
, u64 num_bytes
, u64 parent
,
4765 u64 root_objectid
, u64 owner
, u64 offset
)
4770 * tree log blocks never actually go into the extent allocation
4771 * tree, just update pinning info and exit early.
4773 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
4774 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
4775 /* unlocks the pinned mutex */
4776 btrfs_pin_extent(root
, bytenr
, num_bytes
, 1);
4778 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
4779 ret
= btrfs_add_delayed_tree_ref(trans
, bytenr
, num_bytes
,
4780 parent
, root_objectid
, (int)owner
,
4781 BTRFS_DROP_DELAYED_REF
, NULL
);
4784 ret
= btrfs_add_delayed_data_ref(trans
, bytenr
, num_bytes
,
4785 parent
, root_objectid
, owner
,
4786 offset
, BTRFS_DROP_DELAYED_REF
, NULL
);
4792 static u64
stripe_align(struct btrfs_root
*root
, u64 val
)
4794 u64 mask
= ((u64
)root
->stripesize
- 1);
4795 u64 ret
= (val
+ mask
) & ~mask
;
4800 * when we wait for progress in the block group caching, its because
4801 * our allocation attempt failed at least once. So, we must sleep
4802 * and let some progress happen before we try again.
4804 * This function will sleep at least once waiting for new free space to
4805 * show up, and then it will check the block group free space numbers
4806 * for our min num_bytes. Another option is to have it go ahead
4807 * and look in the rbtree for a free extent of a given size, but this
4811 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
4814 struct btrfs_caching_control
*caching_ctl
;
4817 caching_ctl
= get_caching_control(cache
);
4821 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
4822 (cache
->free_space_ctl
->free_space
>= num_bytes
));
4824 put_caching_control(caching_ctl
);
4829 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
4831 struct btrfs_caching_control
*caching_ctl
;
4834 caching_ctl
= get_caching_control(cache
);
4838 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
4840 put_caching_control(caching_ctl
);
4844 static int get_block_group_index(struct btrfs_block_group_cache
*cache
)
4847 if (cache
->flags
& BTRFS_BLOCK_GROUP_RAID10
)
4849 else if (cache
->flags
& BTRFS_BLOCK_GROUP_RAID1
)
4851 else if (cache
->flags
& BTRFS_BLOCK_GROUP_DUP
)
4853 else if (cache
->flags
& BTRFS_BLOCK_GROUP_RAID0
)
4860 enum btrfs_loop_type
{
4861 LOOP_FIND_IDEAL
= 0,
4862 LOOP_CACHING_NOWAIT
= 1,
4863 LOOP_CACHING_WAIT
= 2,
4864 LOOP_ALLOC_CHUNK
= 3,
4865 LOOP_NO_EMPTY_SIZE
= 4,
4869 * walks the btree of allocated extents and find a hole of a given size.
4870 * The key ins is changed to record the hole:
4871 * ins->objectid == block start
4872 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4873 * ins->offset == number of blocks
4874 * Any available blocks before search_start are skipped.
4876 static noinline
int find_free_extent(struct btrfs_trans_handle
*trans
,
4877 struct btrfs_root
*orig_root
,
4878 u64 num_bytes
, u64 empty_size
,
4879 u64 search_start
, u64 search_end
,
4880 u64 hint_byte
, struct btrfs_key
*ins
,
4884 struct btrfs_root
*root
= orig_root
->fs_info
->extent_root
;
4885 struct btrfs_free_cluster
*last_ptr
= NULL
;
4886 struct btrfs_block_group_cache
*block_group
= NULL
;
4887 int empty_cluster
= 2 * 1024 * 1024;
4888 int allowed_chunk_alloc
= 0;
4889 int done_chunk_alloc
= 0;
4890 struct btrfs_space_info
*space_info
;
4891 int last_ptr_loop
= 0;
4894 int alloc_type
= (data
& BTRFS_BLOCK_GROUP_DATA
) ?
4895 RESERVE_ALLOC_NO_ACCOUNT
: RESERVE_ALLOC
;
4896 bool found_uncached_bg
= false;
4897 bool failed_cluster_refill
= false;
4898 bool failed_alloc
= false;
4899 bool use_cluster
= true;
4900 u64 ideal_cache_percent
= 0;
4901 u64 ideal_cache_offset
= 0;
4903 WARN_ON(num_bytes
< root
->sectorsize
);
4904 btrfs_set_key_type(ins
, BTRFS_EXTENT_ITEM_KEY
);
4908 space_info
= __find_space_info(root
->fs_info
, data
);
4910 printk(KERN_ERR
"No space info for %llu\n", data
);
4915 * If the space info is for both data and metadata it means we have a
4916 * small filesystem and we can't use the clustering stuff.
4918 if (btrfs_mixed_space_info(space_info
))
4919 use_cluster
= false;
4921 if (orig_root
->ref_cows
|| empty_size
)
4922 allowed_chunk_alloc
= 1;
4924 if (data
& BTRFS_BLOCK_GROUP_METADATA
&& use_cluster
) {
4925 last_ptr
= &root
->fs_info
->meta_alloc_cluster
;
4926 if (!btrfs_test_opt(root
, SSD
))
4927 empty_cluster
= 64 * 1024;
4930 if ((data
& BTRFS_BLOCK_GROUP_DATA
) && use_cluster
&&
4931 btrfs_test_opt(root
, SSD
)) {
4932 last_ptr
= &root
->fs_info
->data_alloc_cluster
;
4936 spin_lock(&last_ptr
->lock
);
4937 if (last_ptr
->block_group
)
4938 hint_byte
= last_ptr
->window_start
;
4939 spin_unlock(&last_ptr
->lock
);
4942 search_start
= max(search_start
, first_logical_byte(root
, 0));
4943 search_start
= max(search_start
, hint_byte
);
4948 if (search_start
== hint_byte
) {
4950 block_group
= btrfs_lookup_block_group(root
->fs_info
,
4953 * we don't want to use the block group if it doesn't match our
4954 * allocation bits, or if its not cached.
4956 * However if we are re-searching with an ideal block group
4957 * picked out then we don't care that the block group is cached.
4959 if (block_group
&& block_group_bits(block_group
, data
) &&
4960 (block_group
->cached
!= BTRFS_CACHE_NO
||
4961 search_start
== ideal_cache_offset
)) {
4962 down_read(&space_info
->groups_sem
);
4963 if (list_empty(&block_group
->list
) ||
4966 * someone is removing this block group,
4967 * we can't jump into the have_block_group
4968 * target because our list pointers are not
4971 btrfs_put_block_group(block_group
);
4972 up_read(&space_info
->groups_sem
);
4974 index
= get_block_group_index(block_group
);
4975 goto have_block_group
;
4977 } else if (block_group
) {
4978 btrfs_put_block_group(block_group
);
4982 down_read(&space_info
->groups_sem
);
4983 list_for_each_entry(block_group
, &space_info
->block_groups
[index
],
4988 btrfs_get_block_group(block_group
);
4989 search_start
= block_group
->key
.objectid
;
4992 * this can happen if we end up cycling through all the
4993 * raid types, but we want to make sure we only allocate
4994 * for the proper type.
4996 if (!block_group_bits(block_group
, data
)) {
4997 u64 extra
= BTRFS_BLOCK_GROUP_DUP
|
4998 BTRFS_BLOCK_GROUP_RAID1
|
4999 BTRFS_BLOCK_GROUP_RAID10
;
5002 * if they asked for extra copies and this block group
5003 * doesn't provide them, bail. This does allow us to
5004 * fill raid0 from raid1.
5006 if ((data
& extra
) && !(block_group
->flags
& extra
))
5011 if (unlikely(block_group
->cached
== BTRFS_CACHE_NO
)) {
5014 ret
= cache_block_group(block_group
, trans
,
5016 if (block_group
->cached
== BTRFS_CACHE_FINISHED
)
5017 goto have_block_group
;
5019 free_percent
= btrfs_block_group_used(&block_group
->item
);
5020 free_percent
*= 100;
5021 free_percent
= div64_u64(free_percent
,
5022 block_group
->key
.offset
);
5023 free_percent
= 100 - free_percent
;
5024 if (free_percent
> ideal_cache_percent
&&
5025 likely(!block_group
->ro
)) {
5026 ideal_cache_offset
= block_group
->key
.objectid
;
5027 ideal_cache_percent
= free_percent
;
5031 * The caching workers are limited to 2 threads, so we
5032 * can queue as much work as we care to.
5034 if (loop
> LOOP_FIND_IDEAL
) {
5035 ret
= cache_block_group(block_group
, trans
,
5039 found_uncached_bg
= true;
5042 * If loop is set for cached only, try the next block
5045 if (loop
== LOOP_FIND_IDEAL
)
5049 cached
= block_group_cache_done(block_group
);
5050 if (unlikely(!cached
))
5051 found_uncached_bg
= true;
5053 if (unlikely(block_group
->ro
))
5056 spin_lock(&block_group
->free_space_ctl
->tree_lock
);
5058 block_group
->free_space_ctl
->free_space
<
5059 num_bytes
+ empty_size
) {
5060 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
5063 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
5066 * Ok we want to try and use the cluster allocator, so lets look
5067 * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
5068 * have tried the cluster allocator plenty of times at this
5069 * point and not have found anything, so we are likely way too
5070 * fragmented for the clustering stuff to find anything, so lets
5071 * just skip it and let the allocator find whatever block it can
5074 if (last_ptr
&& loop
< LOOP_NO_EMPTY_SIZE
) {
5076 * the refill lock keeps out other
5077 * people trying to start a new cluster
5079 spin_lock(&last_ptr
->refill_lock
);
5080 if (last_ptr
->block_group
&&
5081 (last_ptr
->block_group
->ro
||
5082 !block_group_bits(last_ptr
->block_group
, data
))) {
5084 goto refill_cluster
;
5087 offset
= btrfs_alloc_from_cluster(block_group
, last_ptr
,
5088 num_bytes
, search_start
);
5090 /* we have a block, we're done */
5091 spin_unlock(&last_ptr
->refill_lock
);
5095 spin_lock(&last_ptr
->lock
);
5097 * whoops, this cluster doesn't actually point to
5098 * this block group. Get a ref on the block
5099 * group is does point to and try again
5101 if (!last_ptr_loop
&& last_ptr
->block_group
&&
5102 last_ptr
->block_group
!= block_group
&&
5104 get_block_group_index(last_ptr
->block_group
)) {
5106 btrfs_put_block_group(block_group
);
5107 block_group
= last_ptr
->block_group
;
5108 btrfs_get_block_group(block_group
);
5109 spin_unlock(&last_ptr
->lock
);
5110 spin_unlock(&last_ptr
->refill_lock
);
5113 search_start
= block_group
->key
.objectid
;
5115 * we know this block group is properly
5116 * in the list because
5117 * btrfs_remove_block_group, drops the
5118 * cluster before it removes the block
5119 * group from the list
5121 goto have_block_group
;
5123 spin_unlock(&last_ptr
->lock
);
5126 * this cluster didn't work out, free it and
5129 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
5133 /* allocate a cluster in this block group */
5134 ret
= btrfs_find_space_cluster(trans
, root
,
5135 block_group
, last_ptr
,
5137 empty_cluster
+ empty_size
);
5140 * now pull our allocation out of this
5143 offset
= btrfs_alloc_from_cluster(block_group
,
5144 last_ptr
, num_bytes
,
5147 /* we found one, proceed */
5148 spin_unlock(&last_ptr
->refill_lock
);
5151 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
5152 && !failed_cluster_refill
) {
5153 spin_unlock(&last_ptr
->refill_lock
);
5155 failed_cluster_refill
= true;
5156 wait_block_group_cache_progress(block_group
,
5157 num_bytes
+ empty_cluster
+ empty_size
);
5158 goto have_block_group
;
5162 * at this point we either didn't find a cluster
5163 * or we weren't able to allocate a block from our
5164 * cluster. Free the cluster we've been trying
5165 * to use, and go to the next block group
5167 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
5168 spin_unlock(&last_ptr
->refill_lock
);
5172 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
5173 num_bytes
, empty_size
);
5175 * If we didn't find a chunk, and we haven't failed on this
5176 * block group before, and this block group is in the middle of
5177 * caching and we are ok with waiting, then go ahead and wait
5178 * for progress to be made, and set failed_alloc to true.
5180 * If failed_alloc is true then we've already waited on this
5181 * block group once and should move on to the next block group.
5183 if (!offset
&& !failed_alloc
&& !cached
&&
5184 loop
> LOOP_CACHING_NOWAIT
) {
5185 wait_block_group_cache_progress(block_group
,
5186 num_bytes
+ empty_size
);
5187 failed_alloc
= true;
5188 goto have_block_group
;
5189 } else if (!offset
) {
5193 search_start
= stripe_align(root
, offset
);
5194 /* move on to the next group */
5195 if (search_start
+ num_bytes
>= search_end
) {
5196 btrfs_add_free_space(block_group
, offset
, num_bytes
);
5200 /* move on to the next group */
5201 if (search_start
+ num_bytes
>
5202 block_group
->key
.objectid
+ block_group
->key
.offset
) {
5203 btrfs_add_free_space(block_group
, offset
, num_bytes
);
5207 ins
->objectid
= search_start
;
5208 ins
->offset
= num_bytes
;
5210 if (offset
< search_start
)
5211 btrfs_add_free_space(block_group
, offset
,
5212 search_start
- offset
);
5213 BUG_ON(offset
> search_start
);
5215 ret
= btrfs_update_reserved_bytes(block_group
, num_bytes
,
5217 if (ret
== -EAGAIN
) {
5218 btrfs_add_free_space(block_group
, offset
, num_bytes
);
5222 /* we are all good, lets return */
5223 ins
->objectid
= search_start
;
5224 ins
->offset
= num_bytes
;
5226 if (offset
< search_start
)
5227 btrfs_add_free_space(block_group
, offset
,
5228 search_start
- offset
);
5229 BUG_ON(offset
> search_start
);
5230 btrfs_put_block_group(block_group
);
5233 failed_cluster_refill
= false;
5234 failed_alloc
= false;
5235 BUG_ON(index
!= get_block_group_index(block_group
));
5236 btrfs_put_block_group(block_group
);
5238 up_read(&space_info
->groups_sem
);
5240 if (!ins
->objectid
&& ++index
< BTRFS_NR_RAID_TYPES
)
5243 /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5244 * for them to make caching progress. Also
5245 * determine the best possible bg to cache
5246 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5247 * caching kthreads as we move along
5248 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5249 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5250 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5253 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
) {
5255 if (loop
== LOOP_FIND_IDEAL
&& found_uncached_bg
) {
5256 found_uncached_bg
= false;
5258 if (!ideal_cache_percent
)
5262 * 1 of the following 2 things have happened so far
5264 * 1) We found an ideal block group for caching that
5265 * is mostly full and will cache quickly, so we might
5266 * as well wait for it.
5268 * 2) We searched for cached only and we didn't find
5269 * anything, and we didn't start any caching kthreads
5270 * either, so chances are we will loop through and
5271 * start a couple caching kthreads, and then come back
5272 * around and just wait for them. This will be slower
5273 * because we will have 2 caching kthreads reading at
5274 * the same time when we could have just started one
5275 * and waited for it to get far enough to give us an
5276 * allocation, so go ahead and go to the wait caching
5279 loop
= LOOP_CACHING_WAIT
;
5280 search_start
= ideal_cache_offset
;
5281 ideal_cache_percent
= 0;
5283 } else if (loop
== LOOP_FIND_IDEAL
) {
5285 * Didn't find a uncached bg, wait on anything we find
5288 loop
= LOOP_CACHING_WAIT
;
5294 if (loop
== LOOP_ALLOC_CHUNK
) {
5295 if (allowed_chunk_alloc
) {
5296 ret
= do_chunk_alloc(trans
, root
, num_bytes
+
5297 2 * 1024 * 1024, data
,
5298 CHUNK_ALLOC_LIMITED
);
5299 allowed_chunk_alloc
= 0;
5301 done_chunk_alloc
= 1;
5302 } else if (!done_chunk_alloc
&&
5303 space_info
->force_alloc
==
5304 CHUNK_ALLOC_NO_FORCE
) {
5305 space_info
->force_alloc
= CHUNK_ALLOC_LIMITED
;
5309 * We didn't allocate a chunk, go ahead and drop the
5310 * empty size and loop again.
5312 if (!done_chunk_alloc
)
5313 loop
= LOOP_NO_EMPTY_SIZE
;
5316 if (loop
== LOOP_NO_EMPTY_SIZE
) {
5322 } else if (!ins
->objectid
) {
5324 } else if (ins
->objectid
) {
5331 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
5332 int dump_block_groups
)
5334 struct btrfs_block_group_cache
*cache
;
5337 spin_lock(&info
->lock
);
5338 printk(KERN_INFO
"space_info %llu has %llu free, is %sfull\n",
5339 (unsigned long long)info
->flags
,
5340 (unsigned long long)(info
->total_bytes
- info
->bytes_used
-
5341 info
->bytes_pinned
- info
->bytes_reserved
-
5342 info
->bytes_readonly
),
5343 (info
->full
) ? "" : "not ");
5344 printk(KERN_INFO
"space_info total=%llu, used=%llu, pinned=%llu, "
5345 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5346 (unsigned long long)info
->total_bytes
,
5347 (unsigned long long)info
->bytes_used
,
5348 (unsigned long long)info
->bytes_pinned
,
5349 (unsigned long long)info
->bytes_reserved
,
5350 (unsigned long long)info
->bytes_may_use
,
5351 (unsigned long long)info
->bytes_readonly
);
5352 spin_unlock(&info
->lock
);
5354 if (!dump_block_groups
)
5357 down_read(&info
->groups_sem
);
5359 list_for_each_entry(cache
, &info
->block_groups
[index
], list
) {
5360 spin_lock(&cache
->lock
);
5361 printk(KERN_INFO
"block group %llu has %llu bytes, %llu used "
5362 "%llu pinned %llu reserved\n",
5363 (unsigned long long)cache
->key
.objectid
,
5364 (unsigned long long)cache
->key
.offset
,
5365 (unsigned long long)btrfs_block_group_used(&cache
->item
),
5366 (unsigned long long)cache
->pinned
,
5367 (unsigned long long)cache
->reserved
);
5368 btrfs_dump_free_space(cache
, bytes
);
5369 spin_unlock(&cache
->lock
);
5371 if (++index
< BTRFS_NR_RAID_TYPES
)
5373 up_read(&info
->groups_sem
);
5376 int btrfs_reserve_extent(struct btrfs_trans_handle
*trans
,
5377 struct btrfs_root
*root
,
5378 u64 num_bytes
, u64 min_alloc_size
,
5379 u64 empty_size
, u64 hint_byte
,
5380 u64 search_end
, struct btrfs_key
*ins
,
5384 u64 search_start
= 0;
5386 data
= btrfs_get_alloc_profile(root
, data
);
5389 * the only place that sets empty_size is btrfs_realloc_node, which
5390 * is not called recursively on allocations
5392 if (empty_size
|| root
->ref_cows
)
5393 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
5394 num_bytes
+ 2 * 1024 * 1024, data
,
5395 CHUNK_ALLOC_NO_FORCE
);
5397 WARN_ON(num_bytes
< root
->sectorsize
);
5398 ret
= find_free_extent(trans
, root
, num_bytes
, empty_size
,
5399 search_start
, search_end
, hint_byte
,
5402 if (ret
== -ENOSPC
&& num_bytes
> min_alloc_size
) {
5403 num_bytes
= num_bytes
>> 1;
5404 num_bytes
= num_bytes
& ~(root
->sectorsize
- 1);
5405 num_bytes
= max(num_bytes
, min_alloc_size
);
5406 do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
5407 num_bytes
, data
, CHUNK_ALLOC_FORCE
);
5410 if (ret
== -ENOSPC
&& btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
5411 struct btrfs_space_info
*sinfo
;
5413 sinfo
= __find_space_info(root
->fs_info
, data
);
5414 printk(KERN_ERR
"btrfs allocation failed flags %llu, "
5415 "wanted %llu\n", (unsigned long long)data
,
5416 (unsigned long long)num_bytes
);
5417 dump_space_info(sinfo
, num_bytes
, 1);
5420 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, ins
->offset
);
5425 int btrfs_free_reserved_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
5427 struct btrfs_block_group_cache
*cache
;
5430 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
5432 printk(KERN_ERR
"Unable to find block group for %llu\n",
5433 (unsigned long long)start
);
5437 if (btrfs_test_opt(root
, DISCARD
))
5438 ret
= btrfs_discard_extent(root
, start
, len
, NULL
);
5440 btrfs_add_free_space(cache
, start
, len
);
5441 btrfs_update_reserved_bytes(cache
, len
, RESERVE_FREE
);
5442 btrfs_put_block_group(cache
);
5444 trace_btrfs_reserved_extent_free(root
, start
, len
);
5449 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
5450 struct btrfs_root
*root
,
5451 u64 parent
, u64 root_objectid
,
5452 u64 flags
, u64 owner
, u64 offset
,
5453 struct btrfs_key
*ins
, int ref_mod
)
5456 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5457 struct btrfs_extent_item
*extent_item
;
5458 struct btrfs_extent_inline_ref
*iref
;
5459 struct btrfs_path
*path
;
5460 struct extent_buffer
*leaf
;
5465 type
= BTRFS_SHARED_DATA_REF_KEY
;
5467 type
= BTRFS_EXTENT_DATA_REF_KEY
;
5469 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
5471 path
= btrfs_alloc_path();
5475 path
->leave_spinning
= 1;
5476 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
5480 leaf
= path
->nodes
[0];
5481 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
5482 struct btrfs_extent_item
);
5483 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
5484 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
5485 btrfs_set_extent_flags(leaf
, extent_item
,
5486 flags
| BTRFS_EXTENT_FLAG_DATA
);
5488 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
5489 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
5491 struct btrfs_shared_data_ref
*ref
;
5492 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
5493 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
5494 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
5496 struct btrfs_extent_data_ref
*ref
;
5497 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
5498 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
5499 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
5500 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
5501 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
5504 btrfs_mark_buffer_dirty(path
->nodes
[0]);
5505 btrfs_free_path(path
);
5507 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1);
5509 printk(KERN_ERR
"btrfs update block group failed for %llu "
5510 "%llu\n", (unsigned long long)ins
->objectid
,
5511 (unsigned long long)ins
->offset
);
5517 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
5518 struct btrfs_root
*root
,
5519 u64 parent
, u64 root_objectid
,
5520 u64 flags
, struct btrfs_disk_key
*key
,
5521 int level
, struct btrfs_key
*ins
)
5524 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5525 struct btrfs_extent_item
*extent_item
;
5526 struct btrfs_tree_block_info
*block_info
;
5527 struct btrfs_extent_inline_ref
*iref
;
5528 struct btrfs_path
*path
;
5529 struct extent_buffer
*leaf
;
5530 u32 size
= sizeof(*extent_item
) + sizeof(*block_info
) + sizeof(*iref
);
5532 path
= btrfs_alloc_path();
5536 path
->leave_spinning
= 1;
5537 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
5541 leaf
= path
->nodes
[0];
5542 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
5543 struct btrfs_extent_item
);
5544 btrfs_set_extent_refs(leaf
, extent_item
, 1);
5545 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
5546 btrfs_set_extent_flags(leaf
, extent_item
,
5547 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
5548 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
5550 btrfs_set_tree_block_key(leaf
, block_info
, key
);
5551 btrfs_set_tree_block_level(leaf
, block_info
, level
);
5553 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
5555 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
5556 btrfs_set_extent_inline_ref_type(leaf
, iref
,
5557 BTRFS_SHARED_BLOCK_REF_KEY
);
5558 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
5560 btrfs_set_extent_inline_ref_type(leaf
, iref
,
5561 BTRFS_TREE_BLOCK_REF_KEY
);
5562 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
5565 btrfs_mark_buffer_dirty(leaf
);
5566 btrfs_free_path(path
);
5568 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1);
5570 printk(KERN_ERR
"btrfs update block group failed for %llu "
5571 "%llu\n", (unsigned long long)ins
->objectid
,
5572 (unsigned long long)ins
->offset
);
5578 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
5579 struct btrfs_root
*root
,
5580 u64 root_objectid
, u64 owner
,
5581 u64 offset
, struct btrfs_key
*ins
)
5585 BUG_ON(root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
5587 ret
= btrfs_add_delayed_data_ref(trans
, ins
->objectid
, ins
->offset
,
5588 0, root_objectid
, owner
, offset
,
5589 BTRFS_ADD_DELAYED_EXTENT
, NULL
);
5594 * this is used by the tree logging recovery code. It records that
5595 * an extent has been allocated and makes sure to clear the free
5596 * space cache bits as well
5598 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
5599 struct btrfs_root
*root
,
5600 u64 root_objectid
, u64 owner
, u64 offset
,
5601 struct btrfs_key
*ins
)
5604 struct btrfs_block_group_cache
*block_group
;
5605 struct btrfs_caching_control
*caching_ctl
;
5606 u64 start
= ins
->objectid
;
5607 u64 num_bytes
= ins
->offset
;
5609 block_group
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
5610 cache_block_group(block_group
, trans
, NULL
, 0);
5611 caching_ctl
= get_caching_control(block_group
);
5614 BUG_ON(!block_group_cache_done(block_group
));
5615 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
5618 mutex_lock(&caching_ctl
->mutex
);
5620 if (start
>= caching_ctl
->progress
) {
5621 ret
= add_excluded_extent(root
, start
, num_bytes
);
5623 } else if (start
+ num_bytes
<= caching_ctl
->progress
) {
5624 ret
= btrfs_remove_free_space(block_group
,
5628 num_bytes
= caching_ctl
->progress
- start
;
5629 ret
= btrfs_remove_free_space(block_group
,
5633 start
= caching_ctl
->progress
;
5634 num_bytes
= ins
->objectid
+ ins
->offset
-
5635 caching_ctl
->progress
;
5636 ret
= add_excluded_extent(root
, start
, num_bytes
);
5640 mutex_unlock(&caching_ctl
->mutex
);
5641 put_caching_control(caching_ctl
);
5644 ret
= btrfs_update_reserved_bytes(block_group
, ins
->offset
,
5645 RESERVE_ALLOC_NO_ACCOUNT
);
5647 btrfs_put_block_group(block_group
);
5648 ret
= alloc_reserved_file_extent(trans
, root
, 0, root_objectid
,
5649 0, owner
, offset
, ins
, 1);
5653 struct extent_buffer
*btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
,
5654 struct btrfs_root
*root
,
5655 u64 bytenr
, u32 blocksize
,
5658 struct extent_buffer
*buf
;
5660 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
5662 return ERR_PTR(-ENOMEM
);
5663 btrfs_set_header_generation(buf
, trans
->transid
);
5664 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, buf
, level
);
5665 btrfs_tree_lock(buf
);
5666 clean_tree_block(trans
, root
, buf
);
5668 btrfs_set_lock_blocking(buf
);
5669 btrfs_set_buffer_uptodate(buf
);
5671 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
5673 * we allow two log transactions at a time, use different
5674 * EXENT bit to differentiate dirty pages.
5676 if (root
->log_transid
% 2 == 0)
5677 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
5678 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
5680 set_extent_new(&root
->dirty_log_pages
, buf
->start
,
5681 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
5683 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
5684 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
5686 trans
->blocks_used
++;
5687 /* this returns a buffer locked for blocking */
5691 static struct btrfs_block_rsv
*
5692 use_block_rsv(struct btrfs_trans_handle
*trans
,
5693 struct btrfs_root
*root
, u32 blocksize
)
5695 struct btrfs_block_rsv
*block_rsv
;
5696 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
5699 block_rsv
= get_block_rsv(trans
, root
);
5701 if (block_rsv
->size
== 0) {
5702 ret
= reserve_metadata_bytes(trans
, root
, block_rsv
,
5705 * If we couldn't reserve metadata bytes try and use some from
5706 * the global reserve.
5708 if (ret
&& block_rsv
!= global_rsv
) {
5709 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
5712 return ERR_PTR(ret
);
5714 return ERR_PTR(ret
);
5719 ret
= block_rsv_use_bytes(block_rsv
, blocksize
);
5724 ret
= reserve_metadata_bytes(trans
, root
, block_rsv
, blocksize
,
5727 spin_lock(&block_rsv
->lock
);
5728 block_rsv
->size
+= blocksize
;
5729 spin_unlock(&block_rsv
->lock
);
5731 } else if (ret
&& block_rsv
!= global_rsv
) {
5732 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
5738 return ERR_PTR(-ENOSPC
);
5741 static void unuse_block_rsv(struct btrfs_block_rsv
*block_rsv
, u32 blocksize
)
5743 block_rsv_add_bytes(block_rsv
, blocksize
, 0);
5744 block_rsv_release_bytes(block_rsv
, NULL
, 0);
5748 * finds a free extent and does all the dirty work required for allocation
5749 * returns the key for the extent through ins, and a tree buffer for
5750 * the first block of the extent through buf.
5752 * returns the tree buffer or NULL.
5754 struct extent_buffer
*btrfs_alloc_free_block(struct btrfs_trans_handle
*trans
,
5755 struct btrfs_root
*root
, u32 blocksize
,
5756 u64 parent
, u64 root_objectid
,
5757 struct btrfs_disk_key
*key
, int level
,
5758 u64 hint
, u64 empty_size
)
5760 struct btrfs_key ins
;
5761 struct btrfs_block_rsv
*block_rsv
;
5762 struct extent_buffer
*buf
;
5767 block_rsv
= use_block_rsv(trans
, root
, blocksize
);
5768 if (IS_ERR(block_rsv
))
5769 return ERR_CAST(block_rsv
);
5771 ret
= btrfs_reserve_extent(trans
, root
, blocksize
, blocksize
,
5772 empty_size
, hint
, (u64
)-1, &ins
, 0);
5774 unuse_block_rsv(block_rsv
, blocksize
);
5775 return ERR_PTR(ret
);
5778 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
,
5780 BUG_ON(IS_ERR(buf
));
5782 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
5784 parent
= ins
.objectid
;
5785 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
5789 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
5790 struct btrfs_delayed_extent_op
*extent_op
;
5791 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
5794 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
5796 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
5797 extent_op
->flags_to_set
= flags
;
5798 extent_op
->update_key
= 1;
5799 extent_op
->update_flags
= 1;
5800 extent_op
->is_data
= 0;
5802 ret
= btrfs_add_delayed_tree_ref(trans
, ins
.objectid
,
5803 ins
.offset
, parent
, root_objectid
,
5804 level
, BTRFS_ADD_DELAYED_EXTENT
,
5811 struct walk_control
{
5812 u64 refs
[BTRFS_MAX_LEVEL
];
5813 u64 flags
[BTRFS_MAX_LEVEL
];
5814 struct btrfs_key update_progress
;
5824 #define DROP_REFERENCE 1
5825 #define UPDATE_BACKREF 2
5827 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
5828 struct btrfs_root
*root
,
5829 struct walk_control
*wc
,
5830 struct btrfs_path
*path
)
5838 struct btrfs_key key
;
5839 struct extent_buffer
*eb
;
5844 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
5845 wc
->reada_count
= wc
->reada_count
* 2 / 3;
5846 wc
->reada_count
= max(wc
->reada_count
, 2);
5848 wc
->reada_count
= wc
->reada_count
* 3 / 2;
5849 wc
->reada_count
= min_t(int, wc
->reada_count
,
5850 BTRFS_NODEPTRS_PER_BLOCK(root
));
5853 eb
= path
->nodes
[wc
->level
];
5854 nritems
= btrfs_header_nritems(eb
);
5855 blocksize
= btrfs_level_size(root
, wc
->level
- 1);
5857 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
5858 if (nread
>= wc
->reada_count
)
5862 bytenr
= btrfs_node_blockptr(eb
, slot
);
5863 generation
= btrfs_node_ptr_generation(eb
, slot
);
5865 if (slot
== path
->slots
[wc
->level
])
5868 if (wc
->stage
== UPDATE_BACKREF
&&
5869 generation
<= root
->root_key
.offset
)
5872 /* We don't lock the tree block, it's OK to be racy here */
5873 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, blocksize
,
5878 if (wc
->stage
== DROP_REFERENCE
) {
5882 if (wc
->level
== 1 &&
5883 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
5885 if (!wc
->update_ref
||
5886 generation
<= root
->root_key
.offset
)
5888 btrfs_node_key_to_cpu(eb
, &key
, slot
);
5889 ret
= btrfs_comp_cpu_keys(&key
,
5890 &wc
->update_progress
);
5894 if (wc
->level
== 1 &&
5895 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
5899 ret
= readahead_tree_block(root
, bytenr
, blocksize
,
5905 wc
->reada_slot
= slot
;
5909 * hepler to process tree block while walking down the tree.
5911 * when wc->stage == UPDATE_BACKREF, this function updates
5912 * back refs for pointers in the block.
5914 * NOTE: return value 1 means we should stop walking down.
5916 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
5917 struct btrfs_root
*root
,
5918 struct btrfs_path
*path
,
5919 struct walk_control
*wc
, int lookup_info
)
5921 int level
= wc
->level
;
5922 struct extent_buffer
*eb
= path
->nodes
[level
];
5923 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
5926 if (wc
->stage
== UPDATE_BACKREF
&&
5927 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
5931 * when reference count of tree block is 1, it won't increase
5932 * again. once full backref flag is set, we never clear it.
5935 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
5936 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
5937 BUG_ON(!path
->locks
[level
]);
5938 ret
= btrfs_lookup_extent_info(trans
, root
,
5943 BUG_ON(wc
->refs
[level
] == 0);
5946 if (wc
->stage
== DROP_REFERENCE
) {
5947 if (wc
->refs
[level
] > 1)
5950 if (path
->locks
[level
] && !wc
->keep_locks
) {
5951 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
5952 path
->locks
[level
] = 0;
5957 /* wc->stage == UPDATE_BACKREF */
5958 if (!(wc
->flags
[level
] & flag
)) {
5959 BUG_ON(!path
->locks
[level
]);
5960 ret
= btrfs_inc_ref(trans
, root
, eb
, 1);
5962 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
5964 ret
= btrfs_set_disk_extent_flags(trans
, root
, eb
->start
,
5967 wc
->flags
[level
] |= flag
;
5971 * the block is shared by multiple trees, so it's not good to
5972 * keep the tree lock
5974 if (path
->locks
[level
] && level
> 0) {
5975 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
5976 path
->locks
[level
] = 0;
5982 * hepler to process tree block pointer.
5984 * when wc->stage == DROP_REFERENCE, this function checks
5985 * reference count of the block pointed to. if the block
5986 * is shared and we need update back refs for the subtree
5987 * rooted at the block, this function changes wc->stage to
5988 * UPDATE_BACKREF. if the block is shared and there is no
5989 * need to update back, this function drops the reference
5992 * NOTE: return value 1 means we should stop walking down.
5994 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
5995 struct btrfs_root
*root
,
5996 struct btrfs_path
*path
,
5997 struct walk_control
*wc
, int *lookup_info
)
6003 struct btrfs_key key
;
6004 struct extent_buffer
*next
;
6005 int level
= wc
->level
;
6009 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
6010 path
->slots
[level
]);
6012 * if the lower level block was created before the snapshot
6013 * was created, we know there is no need to update back refs
6016 if (wc
->stage
== UPDATE_BACKREF
&&
6017 generation
<= root
->root_key
.offset
) {
6022 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
6023 blocksize
= btrfs_level_size(root
, level
- 1);
6025 next
= btrfs_find_tree_block(root
, bytenr
, blocksize
);
6027 next
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
6032 btrfs_tree_lock(next
);
6033 btrfs_set_lock_blocking(next
);
6035 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, blocksize
,
6036 &wc
->refs
[level
- 1],
6037 &wc
->flags
[level
- 1]);
6039 BUG_ON(wc
->refs
[level
- 1] == 0);
6042 if (wc
->stage
== DROP_REFERENCE
) {
6043 if (wc
->refs
[level
- 1] > 1) {
6045 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
6048 if (!wc
->update_ref
||
6049 generation
<= root
->root_key
.offset
)
6052 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
6053 path
->slots
[level
]);
6054 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
6058 wc
->stage
= UPDATE_BACKREF
;
6059 wc
->shared_level
= level
- 1;
6063 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
6067 if (!btrfs_buffer_uptodate(next
, generation
)) {
6068 btrfs_tree_unlock(next
);
6069 free_extent_buffer(next
);
6075 if (reada
&& level
== 1)
6076 reada_walk_down(trans
, root
, wc
, path
);
6077 next
= read_tree_block(root
, bytenr
, blocksize
, generation
);
6080 btrfs_tree_lock(next
);
6081 btrfs_set_lock_blocking(next
);
6085 BUG_ON(level
!= btrfs_header_level(next
));
6086 path
->nodes
[level
] = next
;
6087 path
->slots
[level
] = 0;
6088 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
6094 wc
->refs
[level
- 1] = 0;
6095 wc
->flags
[level
- 1] = 0;
6096 if (wc
->stage
== DROP_REFERENCE
) {
6097 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
6098 parent
= path
->nodes
[level
]->start
;
6100 BUG_ON(root
->root_key
.objectid
!=
6101 btrfs_header_owner(path
->nodes
[level
]));
6105 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
, parent
,
6106 root
->root_key
.objectid
, level
- 1, 0);
6109 btrfs_tree_unlock(next
);
6110 free_extent_buffer(next
);
6116 * hepler to process tree block while walking up the tree.
6118 * when wc->stage == DROP_REFERENCE, this function drops
6119 * reference count on the block.
6121 * when wc->stage == UPDATE_BACKREF, this function changes
6122 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6123 * to UPDATE_BACKREF previously while processing the block.
6125 * NOTE: return value 1 means we should stop walking up.
6127 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
6128 struct btrfs_root
*root
,
6129 struct btrfs_path
*path
,
6130 struct walk_control
*wc
)
6133 int level
= wc
->level
;
6134 struct extent_buffer
*eb
= path
->nodes
[level
];
6137 if (wc
->stage
== UPDATE_BACKREF
) {
6138 BUG_ON(wc
->shared_level
< level
);
6139 if (level
< wc
->shared_level
)
6142 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
6146 wc
->stage
= DROP_REFERENCE
;
6147 wc
->shared_level
= -1;
6148 path
->slots
[level
] = 0;
6151 * check reference count again if the block isn't locked.
6152 * we should start walking down the tree again if reference
6155 if (!path
->locks
[level
]) {
6157 btrfs_tree_lock(eb
);
6158 btrfs_set_lock_blocking(eb
);
6159 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
6161 ret
= btrfs_lookup_extent_info(trans
, root
,
6166 BUG_ON(wc
->refs
[level
] == 0);
6167 if (wc
->refs
[level
] == 1) {
6168 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
6174 /* wc->stage == DROP_REFERENCE */
6175 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
6177 if (wc
->refs
[level
] == 1) {
6179 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
6180 ret
= btrfs_dec_ref(trans
, root
, eb
, 1);
6182 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
6185 /* make block locked assertion in clean_tree_block happy */
6186 if (!path
->locks
[level
] &&
6187 btrfs_header_generation(eb
) == trans
->transid
) {
6188 btrfs_tree_lock(eb
);
6189 btrfs_set_lock_blocking(eb
);
6190 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
6192 clean_tree_block(trans
, root
, eb
);
6195 if (eb
== root
->node
) {
6196 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
6199 BUG_ON(root
->root_key
.objectid
!=
6200 btrfs_header_owner(eb
));
6202 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
6203 parent
= path
->nodes
[level
+ 1]->start
;
6205 BUG_ON(root
->root_key
.objectid
!=
6206 btrfs_header_owner(path
->nodes
[level
+ 1]));
6209 btrfs_free_tree_block(trans
, root
, eb
, parent
, wc
->refs
[level
] == 1);
6211 wc
->refs
[level
] = 0;
6212 wc
->flags
[level
] = 0;
6216 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
6217 struct btrfs_root
*root
,
6218 struct btrfs_path
*path
,
6219 struct walk_control
*wc
)
6221 int level
= wc
->level
;
6222 int lookup_info
= 1;
6225 while (level
>= 0) {
6226 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
6233 if (path
->slots
[level
] >=
6234 btrfs_header_nritems(path
->nodes
[level
]))
6237 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
6239 path
->slots
[level
]++;
6248 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
6249 struct btrfs_root
*root
,
6250 struct btrfs_path
*path
,
6251 struct walk_control
*wc
, int max_level
)
6253 int level
= wc
->level
;
6256 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
6257 while (level
< max_level
&& path
->nodes
[level
]) {
6259 if (path
->slots
[level
] + 1 <
6260 btrfs_header_nritems(path
->nodes
[level
])) {
6261 path
->slots
[level
]++;
6264 ret
= walk_up_proc(trans
, root
, path
, wc
);
6268 if (path
->locks
[level
]) {
6269 btrfs_tree_unlock_rw(path
->nodes
[level
],
6270 path
->locks
[level
]);
6271 path
->locks
[level
] = 0;
6273 free_extent_buffer(path
->nodes
[level
]);
6274 path
->nodes
[level
] = NULL
;
6282 * drop a subvolume tree.
6284 * this function traverses the tree freeing any blocks that only
6285 * referenced by the tree.
6287 * when a shared tree block is found. this function decreases its
6288 * reference count by one. if update_ref is true, this function
6289 * also make sure backrefs for the shared block and all lower level
6290 * blocks are properly updated.
6292 void btrfs_drop_snapshot(struct btrfs_root
*root
,
6293 struct btrfs_block_rsv
*block_rsv
, int update_ref
)
6295 struct btrfs_path
*path
;
6296 struct btrfs_trans_handle
*trans
;
6297 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
6298 struct btrfs_root_item
*root_item
= &root
->root_item
;
6299 struct walk_control
*wc
;
6300 struct btrfs_key key
;
6305 path
= btrfs_alloc_path();
6311 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
6313 btrfs_free_path(path
);
6318 trans
= btrfs_start_transaction(tree_root
, 0);
6319 BUG_ON(IS_ERR(trans
));
6322 trans
->block_rsv
= block_rsv
;
6324 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
6325 level
= btrfs_header_level(root
->node
);
6326 path
->nodes
[level
] = btrfs_lock_root_node(root
);
6327 btrfs_set_lock_blocking(path
->nodes
[level
]);
6328 path
->slots
[level
] = 0;
6329 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
6330 memset(&wc
->update_progress
, 0,
6331 sizeof(wc
->update_progress
));
6333 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
6334 memcpy(&wc
->update_progress
, &key
,
6335 sizeof(wc
->update_progress
));
6337 level
= root_item
->drop_level
;
6339 path
->lowest_level
= level
;
6340 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6341 path
->lowest_level
= 0;
6349 * unlock our path, this is safe because only this
6350 * function is allowed to delete this snapshot
6352 btrfs_unlock_up_safe(path
, 0);
6354 level
= btrfs_header_level(root
->node
);
6356 btrfs_tree_lock(path
->nodes
[level
]);
6357 btrfs_set_lock_blocking(path
->nodes
[level
]);
6359 ret
= btrfs_lookup_extent_info(trans
, root
,
6360 path
->nodes
[level
]->start
,
6361 path
->nodes
[level
]->len
,
6365 BUG_ON(wc
->refs
[level
] == 0);
6367 if (level
== root_item
->drop_level
)
6370 btrfs_tree_unlock(path
->nodes
[level
]);
6371 WARN_ON(wc
->refs
[level
] != 1);
6377 wc
->shared_level
= -1;
6378 wc
->stage
= DROP_REFERENCE
;
6379 wc
->update_ref
= update_ref
;
6381 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
6384 ret
= walk_down_tree(trans
, root
, path
, wc
);
6390 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
6397 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
6401 if (wc
->stage
== DROP_REFERENCE
) {
6403 btrfs_node_key(path
->nodes
[level
],
6404 &root_item
->drop_progress
,
6405 path
->slots
[level
]);
6406 root_item
->drop_level
= level
;
6409 BUG_ON(wc
->level
== 0);
6410 if (btrfs_should_end_transaction(trans
, tree_root
)) {
6411 ret
= btrfs_update_root(trans
, tree_root
,
6416 btrfs_end_transaction_throttle(trans
, tree_root
);
6417 trans
= btrfs_start_transaction(tree_root
, 0);
6418 BUG_ON(IS_ERR(trans
));
6420 trans
->block_rsv
= block_rsv
;
6423 btrfs_release_path(path
);
6426 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
6429 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
6430 ret
= btrfs_find_last_root(tree_root
, root
->root_key
.objectid
,
6434 /* if we fail to delete the orphan item this time
6435 * around, it'll get picked up the next time.
6437 * The most common failure here is just -ENOENT.
6439 btrfs_del_orphan_item(trans
, tree_root
,
6440 root
->root_key
.objectid
);
6444 if (root
->in_radix
) {
6445 btrfs_free_fs_root(tree_root
->fs_info
, root
);
6447 free_extent_buffer(root
->node
);
6448 free_extent_buffer(root
->commit_root
);
6452 btrfs_end_transaction_throttle(trans
, tree_root
);
6454 btrfs_free_path(path
);
6457 btrfs_std_error(root
->fs_info
, err
);
6462 * drop subtree rooted at tree block 'node'.
6464 * NOTE: this function will unlock and release tree block 'node'
6466 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
6467 struct btrfs_root
*root
,
6468 struct extent_buffer
*node
,
6469 struct extent_buffer
*parent
)
6471 struct btrfs_path
*path
;
6472 struct walk_control
*wc
;
6478 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
6480 path
= btrfs_alloc_path();
6484 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
6486 btrfs_free_path(path
);
6490 btrfs_assert_tree_locked(parent
);
6491 parent_level
= btrfs_header_level(parent
);
6492 extent_buffer_get(parent
);
6493 path
->nodes
[parent_level
] = parent
;
6494 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
6496 btrfs_assert_tree_locked(node
);
6497 level
= btrfs_header_level(node
);
6498 path
->nodes
[level
] = node
;
6499 path
->slots
[level
] = 0;
6500 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
6502 wc
->refs
[parent_level
] = 1;
6503 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
6505 wc
->shared_level
= -1;
6506 wc
->stage
= DROP_REFERENCE
;
6509 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
6512 wret
= walk_down_tree(trans
, root
, path
, wc
);
6518 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
6526 btrfs_free_path(path
);
6530 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
6533 u64 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
6534 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
6537 * we add in the count of missing devices because we want
6538 * to make sure that any RAID levels on a degraded FS
6539 * continue to be honored.
6541 num_devices
= root
->fs_info
->fs_devices
->rw_devices
+
6542 root
->fs_info
->fs_devices
->missing_devices
;
6544 if (num_devices
== 1) {
6545 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
6546 stripped
= flags
& ~stripped
;
6548 /* turn raid0 into single device chunks */
6549 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
6552 /* turn mirroring into duplication */
6553 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
6554 BTRFS_BLOCK_GROUP_RAID10
))
6555 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
6558 /* they already had raid on here, just return */
6559 if (flags
& stripped
)
6562 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
6563 stripped
= flags
& ~stripped
;
6565 /* switch duplicated blocks with raid1 */
6566 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
6567 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
6569 /* turn single device chunks into raid0 */
6570 return stripped
| BTRFS_BLOCK_GROUP_RAID0
;
6575 static int set_block_group_ro(struct btrfs_block_group_cache
*cache
, int force
)
6577 struct btrfs_space_info
*sinfo
= cache
->space_info
;
6579 u64 min_allocable_bytes
;
6584 * We need some metadata space and system metadata space for
6585 * allocating chunks in some corner cases until we force to set
6586 * it to be readonly.
6589 (BTRFS_BLOCK_GROUP_SYSTEM
| BTRFS_BLOCK_GROUP_METADATA
)) &&
6591 min_allocable_bytes
= 1 * 1024 * 1024;
6593 min_allocable_bytes
= 0;
6595 spin_lock(&sinfo
->lock
);
6596 spin_lock(&cache
->lock
);
6603 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
6604 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
6606 if (sinfo
->bytes_used
+ sinfo
->bytes_reserved
+ sinfo
->bytes_pinned
+
6607 sinfo
->bytes_may_use
+ sinfo
->bytes_readonly
+
6608 cache
->reserved_pinned
+ num_bytes
+ min_allocable_bytes
<=
6609 sinfo
->total_bytes
) {
6610 sinfo
->bytes_readonly
+= num_bytes
;
6611 sinfo
->bytes_may_use
+= cache
->reserved_pinned
;
6612 cache
->reserved_pinned
= 0;
6617 spin_unlock(&cache
->lock
);
6618 spin_unlock(&sinfo
->lock
);
6622 int btrfs_set_block_group_ro(struct btrfs_root
*root
,
6623 struct btrfs_block_group_cache
*cache
)
6626 struct btrfs_trans_handle
*trans
;
6632 trans
= btrfs_join_transaction(root
);
6633 BUG_ON(IS_ERR(trans
));
6635 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
6636 if (alloc_flags
!= cache
->flags
)
6637 do_chunk_alloc(trans
, root
, 2 * 1024 * 1024, alloc_flags
,
6640 ret
= set_block_group_ro(cache
, 0);
6643 alloc_flags
= get_alloc_profile(root
, cache
->space_info
->flags
);
6644 ret
= do_chunk_alloc(trans
, root
, 2 * 1024 * 1024, alloc_flags
,
6648 ret
= set_block_group_ro(cache
, 0);
6650 btrfs_end_transaction(trans
, root
);
6654 int btrfs_force_chunk_alloc(struct btrfs_trans_handle
*trans
,
6655 struct btrfs_root
*root
, u64 type
)
6657 u64 alloc_flags
= get_alloc_profile(root
, type
);
6658 return do_chunk_alloc(trans
, root
, 2 * 1024 * 1024, alloc_flags
,
6663 * helper to account the unused space of all the readonly block group in the
6664 * list. takes mirrors into account.
6666 static u64
__btrfs_get_ro_block_group_free_space(struct list_head
*groups_list
)
6668 struct btrfs_block_group_cache
*block_group
;
6672 list_for_each_entry(block_group
, groups_list
, list
) {
6673 spin_lock(&block_group
->lock
);
6675 if (!block_group
->ro
) {
6676 spin_unlock(&block_group
->lock
);
6680 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_RAID1
|
6681 BTRFS_BLOCK_GROUP_RAID10
|
6682 BTRFS_BLOCK_GROUP_DUP
))
6687 free_bytes
+= (block_group
->key
.offset
-
6688 btrfs_block_group_used(&block_group
->item
)) *
6691 spin_unlock(&block_group
->lock
);
6698 * helper to account the unused space of all the readonly block group in the
6699 * space_info. takes mirrors into account.
6701 u64
btrfs_account_ro_block_groups_free_space(struct btrfs_space_info
*sinfo
)
6706 spin_lock(&sinfo
->lock
);
6708 for(i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
6709 if (!list_empty(&sinfo
->block_groups
[i
]))
6710 free_bytes
+= __btrfs_get_ro_block_group_free_space(
6711 &sinfo
->block_groups
[i
]);
6713 spin_unlock(&sinfo
->lock
);
6718 int btrfs_set_block_group_rw(struct btrfs_root
*root
,
6719 struct btrfs_block_group_cache
*cache
)
6721 struct btrfs_space_info
*sinfo
= cache
->space_info
;
6726 spin_lock(&sinfo
->lock
);
6727 spin_lock(&cache
->lock
);
6728 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
6729 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
6730 sinfo
->bytes_readonly
-= num_bytes
;
6732 spin_unlock(&cache
->lock
);
6733 spin_unlock(&sinfo
->lock
);
6738 * checks to see if its even possible to relocate this block group.
6740 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
6741 * ok to go ahead and try.
6743 int btrfs_can_relocate(struct btrfs_root
*root
, u64 bytenr
)
6745 struct btrfs_block_group_cache
*block_group
;
6746 struct btrfs_space_info
*space_info
;
6747 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
6748 struct btrfs_device
*device
;
6756 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
6758 /* odd, couldn't find the block group, leave it alone */
6762 min_free
= btrfs_block_group_used(&block_group
->item
);
6764 /* no bytes used, we're good */
6768 space_info
= block_group
->space_info
;
6769 spin_lock(&space_info
->lock
);
6771 full
= space_info
->full
;
6774 * if this is the last block group we have in this space, we can't
6775 * relocate it unless we're able to allocate a new chunk below.
6777 * Otherwise, we need to make sure we have room in the space to handle
6778 * all of the extents from this block group. If we can, we're good
6780 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
6781 (space_info
->bytes_used
+ space_info
->bytes_reserved
+
6782 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
6783 min_free
< space_info
->total_bytes
)) {
6784 spin_unlock(&space_info
->lock
);
6787 spin_unlock(&space_info
->lock
);
6790 * ok we don't have enough space, but maybe we have free space on our
6791 * devices to allocate new chunks for relocation, so loop through our
6792 * alloc devices and guess if we have enough space. However, if we
6793 * were marked as full, then we know there aren't enough chunks, and we
6808 index
= get_block_group_index(block_group
);
6813 } else if (index
== 1) {
6815 } else if (index
== 2) {
6818 } else if (index
== 3) {
6819 dev_min
= fs_devices
->rw_devices
;
6820 do_div(min_free
, dev_min
);
6823 mutex_lock(&root
->fs_info
->chunk_mutex
);
6824 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
6828 * check to make sure we can actually find a chunk with enough
6829 * space to fit our block group in.
6831 if (device
->total_bytes
> device
->bytes_used
+ min_free
) {
6832 ret
= find_free_dev_extent(NULL
, device
, min_free
,
6837 if (dev_nr
>= dev_min
)
6843 mutex_unlock(&root
->fs_info
->chunk_mutex
);
6845 btrfs_put_block_group(block_group
);
6849 static int find_first_block_group(struct btrfs_root
*root
,
6850 struct btrfs_path
*path
, struct btrfs_key
*key
)
6853 struct btrfs_key found_key
;
6854 struct extent_buffer
*leaf
;
6857 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
6862 slot
= path
->slots
[0];
6863 leaf
= path
->nodes
[0];
6864 if (slot
>= btrfs_header_nritems(leaf
)) {
6865 ret
= btrfs_next_leaf(root
, path
);
6872 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
6874 if (found_key
.objectid
>= key
->objectid
&&
6875 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
6885 void btrfs_put_block_group_cache(struct btrfs_fs_info
*info
)
6887 struct btrfs_block_group_cache
*block_group
;
6891 struct inode
*inode
;
6893 block_group
= btrfs_lookup_first_block_group(info
, last
);
6894 while (block_group
) {
6895 spin_lock(&block_group
->lock
);
6896 if (block_group
->iref
)
6898 spin_unlock(&block_group
->lock
);
6899 block_group
= next_block_group(info
->tree_root
,
6909 inode
= block_group
->inode
;
6910 block_group
->iref
= 0;
6911 block_group
->inode
= NULL
;
6912 spin_unlock(&block_group
->lock
);
6914 last
= block_group
->key
.objectid
+ block_group
->key
.offset
;
6915 btrfs_put_block_group(block_group
);
6919 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
6921 struct btrfs_block_group_cache
*block_group
;
6922 struct btrfs_space_info
*space_info
;
6923 struct btrfs_caching_control
*caching_ctl
;
6926 down_write(&info
->extent_commit_sem
);
6927 while (!list_empty(&info
->caching_block_groups
)) {
6928 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
6929 struct btrfs_caching_control
, list
);
6930 list_del(&caching_ctl
->list
);
6931 put_caching_control(caching_ctl
);
6933 up_write(&info
->extent_commit_sem
);
6935 spin_lock(&info
->block_group_cache_lock
);
6936 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
6937 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
6939 rb_erase(&block_group
->cache_node
,
6940 &info
->block_group_cache_tree
);
6941 spin_unlock(&info
->block_group_cache_lock
);
6943 down_write(&block_group
->space_info
->groups_sem
);
6944 list_del(&block_group
->list
);
6945 up_write(&block_group
->space_info
->groups_sem
);
6947 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
6948 wait_block_group_cache_done(block_group
);
6951 * We haven't cached this block group, which means we could
6952 * possibly have excluded extents on this block group.
6954 if (block_group
->cached
== BTRFS_CACHE_NO
)
6955 free_excluded_extents(info
->extent_root
, block_group
);
6957 btrfs_remove_free_space_cache(block_group
);
6958 btrfs_put_block_group(block_group
);
6960 spin_lock(&info
->block_group_cache_lock
);
6962 spin_unlock(&info
->block_group_cache_lock
);
6964 /* now that all the block groups are freed, go through and
6965 * free all the space_info structs. This is only called during
6966 * the final stages of unmount, and so we know nobody is
6967 * using them. We call synchronize_rcu() once before we start,
6968 * just to be on the safe side.
6972 release_global_block_rsv(info
);
6974 while(!list_empty(&info
->space_info
)) {
6975 space_info
= list_entry(info
->space_info
.next
,
6976 struct btrfs_space_info
,
6978 if (space_info
->bytes_pinned
> 0 ||
6979 space_info
->bytes_reserved
> 0 ||
6980 space_info
->bytes_may_use
> 0) {
6982 dump_space_info(space_info
, 0, 0);
6984 list_del(&space_info
->list
);
6990 static void __link_block_group(struct btrfs_space_info
*space_info
,
6991 struct btrfs_block_group_cache
*cache
)
6993 int index
= get_block_group_index(cache
);
6995 down_write(&space_info
->groups_sem
);
6996 list_add_tail(&cache
->list
, &space_info
->block_groups
[index
]);
6997 up_write(&space_info
->groups_sem
);
7000 int btrfs_read_block_groups(struct btrfs_root
*root
)
7002 struct btrfs_path
*path
;
7004 struct btrfs_block_group_cache
*cache
;
7005 struct btrfs_fs_info
*info
= root
->fs_info
;
7006 struct btrfs_space_info
*space_info
;
7007 struct btrfs_key key
;
7008 struct btrfs_key found_key
;
7009 struct extent_buffer
*leaf
;
7013 root
= info
->extent_root
;
7016 btrfs_set_key_type(&key
, BTRFS_BLOCK_GROUP_ITEM_KEY
);
7017 path
= btrfs_alloc_path();
7022 cache_gen
= btrfs_super_cache_generation(&root
->fs_info
->super_copy
);
7023 if (cache_gen
!= 0 &&
7024 btrfs_super_generation(&root
->fs_info
->super_copy
) != cache_gen
)
7026 if (btrfs_test_opt(root
, CLEAR_CACHE
))
7028 if (!btrfs_test_opt(root
, SPACE_CACHE
) && cache_gen
)
7029 printk(KERN_INFO
"btrfs: disk space caching is enabled\n");
7032 ret
= find_first_block_group(root
, path
, &key
);
7037 leaf
= path
->nodes
[0];
7038 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
7039 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
7044 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
7046 if (!cache
->free_space_ctl
) {
7052 atomic_set(&cache
->count
, 1);
7053 spin_lock_init(&cache
->lock
);
7054 cache
->fs_info
= info
;
7055 INIT_LIST_HEAD(&cache
->list
);
7056 INIT_LIST_HEAD(&cache
->cluster_list
);
7059 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
7061 read_extent_buffer(leaf
, &cache
->item
,
7062 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
7063 sizeof(cache
->item
));
7064 memcpy(&cache
->key
, &found_key
, sizeof(found_key
));
7066 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
7067 btrfs_release_path(path
);
7068 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
7069 cache
->sectorsize
= root
->sectorsize
;
7071 btrfs_init_free_space_ctl(cache
);
7074 * We need to exclude the super stripes now so that the space
7075 * info has super bytes accounted for, otherwise we'll think
7076 * we have more space than we actually do.
7078 exclude_super_stripes(root
, cache
);
7081 * check for two cases, either we are full, and therefore
7082 * don't need to bother with the caching work since we won't
7083 * find any space, or we are empty, and we can just add all
7084 * the space in and be done with it. This saves us _alot_ of
7085 * time, particularly in the full case.
7087 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
7088 cache
->last_byte_to_unpin
= (u64
)-1;
7089 cache
->cached
= BTRFS_CACHE_FINISHED
;
7090 free_excluded_extents(root
, cache
);
7091 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
7092 cache
->last_byte_to_unpin
= (u64
)-1;
7093 cache
->cached
= BTRFS_CACHE_FINISHED
;
7094 add_new_free_space(cache
, root
->fs_info
,
7096 found_key
.objectid
+
7098 free_excluded_extents(root
, cache
);
7101 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
7102 btrfs_block_group_used(&cache
->item
),
7105 cache
->space_info
= space_info
;
7106 spin_lock(&cache
->space_info
->lock
);
7107 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
7108 spin_unlock(&cache
->space_info
->lock
);
7110 __link_block_group(space_info
, cache
);
7112 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
7115 set_avail_alloc_bits(root
->fs_info
, cache
->flags
);
7116 if (btrfs_chunk_readonly(root
, cache
->key
.objectid
))
7117 set_block_group_ro(cache
, 1);
7120 list_for_each_entry_rcu(space_info
, &root
->fs_info
->space_info
, list
) {
7121 if (!(get_alloc_profile(root
, space_info
->flags
) &
7122 (BTRFS_BLOCK_GROUP_RAID10
|
7123 BTRFS_BLOCK_GROUP_RAID1
|
7124 BTRFS_BLOCK_GROUP_DUP
)))
7127 * avoid allocating from un-mirrored block group if there are
7128 * mirrored block groups.
7130 list_for_each_entry(cache
, &space_info
->block_groups
[3], list
)
7131 set_block_group_ro(cache
, 1);
7132 list_for_each_entry(cache
, &space_info
->block_groups
[4], list
)
7133 set_block_group_ro(cache
, 1);
7136 init_global_block_rsv(info
);
7139 btrfs_free_path(path
);
7143 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
7144 struct btrfs_root
*root
, u64 bytes_used
,
7145 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
7149 struct btrfs_root
*extent_root
;
7150 struct btrfs_block_group_cache
*cache
;
7152 extent_root
= root
->fs_info
->extent_root
;
7154 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
7156 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
7159 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
7161 if (!cache
->free_space_ctl
) {
7166 cache
->key
.objectid
= chunk_offset
;
7167 cache
->key
.offset
= size
;
7168 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
7169 cache
->sectorsize
= root
->sectorsize
;
7170 cache
->fs_info
= root
->fs_info
;
7172 atomic_set(&cache
->count
, 1);
7173 spin_lock_init(&cache
->lock
);
7174 INIT_LIST_HEAD(&cache
->list
);
7175 INIT_LIST_HEAD(&cache
->cluster_list
);
7177 btrfs_init_free_space_ctl(cache
);
7179 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
7180 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
7181 cache
->flags
= type
;
7182 btrfs_set_block_group_flags(&cache
->item
, type
);
7184 cache
->last_byte_to_unpin
= (u64
)-1;
7185 cache
->cached
= BTRFS_CACHE_FINISHED
;
7186 exclude_super_stripes(root
, cache
);
7188 add_new_free_space(cache
, root
->fs_info
, chunk_offset
,
7189 chunk_offset
+ size
);
7191 free_excluded_extents(root
, cache
);
7193 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
7194 &cache
->space_info
);
7197 spin_lock(&cache
->space_info
->lock
);
7198 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
7199 spin_unlock(&cache
->space_info
->lock
);
7201 __link_block_group(cache
->space_info
, cache
);
7203 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
7206 ret
= btrfs_insert_item(trans
, extent_root
, &cache
->key
, &cache
->item
,
7207 sizeof(cache
->item
));
7210 set_avail_alloc_bits(extent_root
->fs_info
, type
);
7215 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
7216 struct btrfs_root
*root
, u64 group_start
)
7218 struct btrfs_path
*path
;
7219 struct btrfs_block_group_cache
*block_group
;
7220 struct btrfs_free_cluster
*cluster
;
7221 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
7222 struct btrfs_key key
;
7223 struct inode
*inode
;
7227 root
= root
->fs_info
->extent_root
;
7229 block_group
= btrfs_lookup_block_group(root
->fs_info
, group_start
);
7230 BUG_ON(!block_group
);
7231 BUG_ON(!block_group
->ro
);
7234 * Free the reserved super bytes from this block group before
7237 free_excluded_extents(root
, block_group
);
7239 memcpy(&key
, &block_group
->key
, sizeof(key
));
7240 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
7241 BTRFS_BLOCK_GROUP_RAID1
|
7242 BTRFS_BLOCK_GROUP_RAID10
))
7247 /* make sure this block group isn't part of an allocation cluster */
7248 cluster
= &root
->fs_info
->data_alloc_cluster
;
7249 spin_lock(&cluster
->refill_lock
);
7250 btrfs_return_cluster_to_free_space(block_group
, cluster
);
7251 spin_unlock(&cluster
->refill_lock
);
7254 * make sure this block group isn't part of a metadata
7255 * allocation cluster
7257 cluster
= &root
->fs_info
->meta_alloc_cluster
;
7258 spin_lock(&cluster
->refill_lock
);
7259 btrfs_return_cluster_to_free_space(block_group
, cluster
);
7260 spin_unlock(&cluster
->refill_lock
);
7262 path
= btrfs_alloc_path();
7268 inode
= lookup_free_space_inode(root
, block_group
, path
);
7269 if (!IS_ERR(inode
)) {
7270 ret
= btrfs_orphan_add(trans
, inode
);
7273 /* One for the block groups ref */
7274 spin_lock(&block_group
->lock
);
7275 if (block_group
->iref
) {
7276 block_group
->iref
= 0;
7277 block_group
->inode
= NULL
;
7278 spin_unlock(&block_group
->lock
);
7281 spin_unlock(&block_group
->lock
);
7283 /* One for our lookup ref */
7287 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
7288 key
.offset
= block_group
->key
.objectid
;
7291 ret
= btrfs_search_slot(trans
, tree_root
, &key
, path
, -1, 1);
7295 btrfs_release_path(path
);
7297 ret
= btrfs_del_item(trans
, tree_root
, path
);
7300 btrfs_release_path(path
);
7303 spin_lock(&root
->fs_info
->block_group_cache_lock
);
7304 rb_erase(&block_group
->cache_node
,
7305 &root
->fs_info
->block_group_cache_tree
);
7306 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
7308 down_write(&block_group
->space_info
->groups_sem
);
7310 * we must use list_del_init so people can check to see if they
7311 * are still on the list after taking the semaphore
7313 list_del_init(&block_group
->list
);
7314 up_write(&block_group
->space_info
->groups_sem
);
7316 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
7317 wait_block_group_cache_done(block_group
);
7319 btrfs_remove_free_space_cache(block_group
);
7321 spin_lock(&block_group
->space_info
->lock
);
7322 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
7323 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
7324 block_group
->space_info
->disk_total
-= block_group
->key
.offset
* factor
;
7325 spin_unlock(&block_group
->space_info
->lock
);
7327 memcpy(&key
, &block_group
->key
, sizeof(key
));
7329 btrfs_clear_space_info_full(root
->fs_info
);
7331 btrfs_put_block_group(block_group
);
7332 btrfs_put_block_group(block_group
);
7334 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
7340 ret
= btrfs_del_item(trans
, root
, path
);
7342 btrfs_free_path(path
);
7346 int btrfs_init_space_info(struct btrfs_fs_info
*fs_info
)
7348 struct btrfs_space_info
*space_info
;
7349 struct btrfs_super_block
*disk_super
;
7355 disk_super
= &fs_info
->super_copy
;
7356 if (!btrfs_super_root(disk_super
))
7359 features
= btrfs_super_incompat_flags(disk_super
);
7360 if (features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
7363 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
7364 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
7369 flags
= BTRFS_BLOCK_GROUP_METADATA
| BTRFS_BLOCK_GROUP_DATA
;
7370 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
7372 flags
= BTRFS_BLOCK_GROUP_METADATA
;
7373 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
7377 flags
= BTRFS_BLOCK_GROUP_DATA
;
7378 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
7384 int btrfs_error_unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
7386 return unpin_extent_range(root
, start
, end
);
7389 int btrfs_error_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
7390 u64 num_bytes
, u64
*actual_bytes
)
7392 return btrfs_discard_extent(root
, bytenr
, num_bytes
, actual_bytes
);
7395 int btrfs_trim_fs(struct btrfs_root
*root
, struct fstrim_range
*range
)
7397 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7398 struct btrfs_block_group_cache
*cache
= NULL
;
7405 cache
= btrfs_lookup_block_group(fs_info
, range
->start
);
7408 if (cache
->key
.objectid
>= (range
->start
+ range
->len
)) {
7409 btrfs_put_block_group(cache
);
7413 start
= max(range
->start
, cache
->key
.objectid
);
7414 end
= min(range
->start
+ range
->len
,
7415 cache
->key
.objectid
+ cache
->key
.offset
);
7417 if (end
- start
>= range
->minlen
) {
7418 if (!block_group_cache_done(cache
)) {
7419 ret
= cache_block_group(cache
, NULL
, root
, 0);
7421 wait_block_group_cache_done(cache
);
7423 ret
= btrfs_trim_block_group(cache
,
7429 trimmed
+= group_trimmed
;
7431 btrfs_put_block_group(cache
);
7436 cache
= next_block_group(fs_info
->tree_root
, cache
);
7439 range
->len
= trimmed
;