2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include <linux/slab.h>
26 #include <linux/ratelimit.h>
31 #include "print-tree.h"
32 #include "transaction.h"
35 #include "free-space-cache.h"
37 /* control flags for do_chunk_alloc's force field
38 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
39 * if we really need one.
41 * CHUNK_ALLOC_FORCE means it must try to allocate one
43 * CHUNK_ALLOC_LIMITED means to only try and allocate one
44 * if we have very few chunks already allocated. This is
45 * used as part of the clustering code to help make sure
46 * we have a good pool of storage to cluster in, without
47 * filling the FS with empty chunks
51 CHUNK_ALLOC_NO_FORCE
= 0,
52 CHUNK_ALLOC_FORCE
= 1,
53 CHUNK_ALLOC_LIMITED
= 2,
57 * Control how reservations are dealt with.
59 * RESERVE_FREE - freeing a reservation.
60 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
62 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
63 * bytes_may_use as the ENOSPC accounting is done elsewhere
68 RESERVE_ALLOC_NO_ACCOUNT
= 2,
71 static int update_block_group(struct btrfs_trans_handle
*trans
,
72 struct btrfs_root
*root
,
73 u64 bytenr
, u64 num_bytes
, int alloc
);
74 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
75 struct btrfs_root
*root
,
76 u64 bytenr
, u64 num_bytes
, u64 parent
,
77 u64 root_objectid
, u64 owner_objectid
,
78 u64 owner_offset
, int refs_to_drop
,
79 struct btrfs_delayed_extent_op
*extra_op
);
80 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
81 struct extent_buffer
*leaf
,
82 struct btrfs_extent_item
*ei
);
83 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
84 struct btrfs_root
*root
,
85 u64 parent
, u64 root_objectid
,
86 u64 flags
, u64 owner
, u64 offset
,
87 struct btrfs_key
*ins
, int ref_mod
);
88 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
89 struct btrfs_root
*root
,
90 u64 parent
, u64 root_objectid
,
91 u64 flags
, struct btrfs_disk_key
*key
,
92 int level
, struct btrfs_key
*ins
);
93 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
94 struct btrfs_root
*extent_root
, u64 alloc_bytes
,
95 u64 flags
, int force
);
96 static int find_next_key(struct btrfs_path
*path
, int level
,
97 struct btrfs_key
*key
);
98 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
99 int dump_block_groups
);
100 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
101 u64 num_bytes
, int reserve
);
104 block_group_cache_done(struct btrfs_block_group_cache
*cache
)
107 return cache
->cached
== BTRFS_CACHE_FINISHED
;
110 static int block_group_bits(struct btrfs_block_group_cache
*cache
, u64 bits
)
112 return (cache
->flags
& bits
) == bits
;
115 static void btrfs_get_block_group(struct btrfs_block_group_cache
*cache
)
117 atomic_inc(&cache
->count
);
120 void btrfs_put_block_group(struct btrfs_block_group_cache
*cache
)
122 if (atomic_dec_and_test(&cache
->count
)) {
123 WARN_ON(cache
->pinned
> 0);
124 WARN_ON(cache
->reserved
> 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
,
471 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
472 struct btrfs_caching_control
*caching_ctl
;
475 caching_ctl
= kzalloc(sizeof(*caching_ctl
), GFP_NOFS
);
476 BUG_ON(!caching_ctl
);
478 INIT_LIST_HEAD(&caching_ctl
->list
);
479 mutex_init(&caching_ctl
->mutex
);
480 init_waitqueue_head(&caching_ctl
->wait
);
481 caching_ctl
->block_group
= cache
;
482 caching_ctl
->progress
= cache
->key
.objectid
;
483 atomic_set(&caching_ctl
->count
, 1);
484 caching_ctl
->work
.func
= caching_thread
;
486 spin_lock(&cache
->lock
);
488 * This should be a rare occasion, but this could happen I think in the
489 * case where one thread starts to load the space cache info, and then
490 * some other thread starts a transaction commit which tries to do an
491 * allocation while the other thread is still loading the space cache
492 * info. The previous loop should have kept us from choosing this block
493 * group, but if we've moved to the state where we will wait on caching
494 * block groups we need to first check if we're doing a fast load here,
495 * so we can wait for it to finish, otherwise we could end up allocating
496 * from a block group who's cache gets evicted for one reason or
499 while (cache
->cached
== BTRFS_CACHE_FAST
) {
500 struct btrfs_caching_control
*ctl
;
502 ctl
= cache
->caching_ctl
;
503 atomic_inc(&ctl
->count
);
504 prepare_to_wait(&ctl
->wait
, &wait
, TASK_UNINTERRUPTIBLE
);
505 spin_unlock(&cache
->lock
);
509 finish_wait(&ctl
->wait
, &wait
);
510 put_caching_control(ctl
);
511 spin_lock(&cache
->lock
);
514 if (cache
->cached
!= BTRFS_CACHE_NO
) {
515 spin_unlock(&cache
->lock
);
519 WARN_ON(cache
->caching_ctl
);
520 cache
->caching_ctl
= caching_ctl
;
521 cache
->cached
= BTRFS_CACHE_FAST
;
522 spin_unlock(&cache
->lock
);
525 * We can't do the read from on-disk cache during a commit since we need
526 * to have the normal tree locking. Also if we are currently trying to
527 * allocate blocks for the tree root we can't do the fast caching since
528 * we likely hold important locks.
530 if (trans
&& (!trans
->transaction
->in_commit
) &&
531 (root
&& root
!= root
->fs_info
->tree_root
) &&
532 btrfs_test_opt(root
, SPACE_CACHE
)) {
533 ret
= load_free_space_cache(fs_info
, cache
);
535 spin_lock(&cache
->lock
);
537 cache
->caching_ctl
= NULL
;
538 cache
->cached
= BTRFS_CACHE_FINISHED
;
539 cache
->last_byte_to_unpin
= (u64
)-1;
541 if (load_cache_only
) {
542 cache
->caching_ctl
= NULL
;
543 cache
->cached
= BTRFS_CACHE_NO
;
545 cache
->cached
= BTRFS_CACHE_STARTED
;
548 spin_unlock(&cache
->lock
);
549 wake_up(&caching_ctl
->wait
);
551 put_caching_control(caching_ctl
);
552 free_excluded_extents(fs_info
->extent_root
, cache
);
557 * We are not going to do the fast caching, set cached to the
558 * appropriate value and wakeup any waiters.
560 spin_lock(&cache
->lock
);
561 if (load_cache_only
) {
562 cache
->caching_ctl
= NULL
;
563 cache
->cached
= BTRFS_CACHE_NO
;
565 cache
->cached
= BTRFS_CACHE_STARTED
;
567 spin_unlock(&cache
->lock
);
568 wake_up(&caching_ctl
->wait
);
571 if (load_cache_only
) {
572 put_caching_control(caching_ctl
);
576 down_write(&fs_info
->extent_commit_sem
);
577 atomic_inc(&caching_ctl
->count
);
578 list_add_tail(&caching_ctl
->list
, &fs_info
->caching_block_groups
);
579 up_write(&fs_info
->extent_commit_sem
);
581 btrfs_get_block_group(cache
);
583 btrfs_queue_worker(&fs_info
->caching_workers
, &caching_ctl
->work
);
589 * return the block group that starts at or after bytenr
591 static struct btrfs_block_group_cache
*
592 btrfs_lookup_first_block_group(struct btrfs_fs_info
*info
, u64 bytenr
)
594 struct btrfs_block_group_cache
*cache
;
596 cache
= block_group_cache_tree_search(info
, bytenr
, 0);
602 * return the block group that contains the given bytenr
604 struct btrfs_block_group_cache
*btrfs_lookup_block_group(
605 struct btrfs_fs_info
*info
,
608 struct btrfs_block_group_cache
*cache
;
610 cache
= block_group_cache_tree_search(info
, bytenr
, 1);
615 static struct btrfs_space_info
*__find_space_info(struct btrfs_fs_info
*info
,
618 struct list_head
*head
= &info
->space_info
;
619 struct btrfs_space_info
*found
;
621 flags
&= BTRFS_BLOCK_GROUP_TYPE_MASK
;
624 list_for_each_entry_rcu(found
, head
, list
) {
625 if (found
->flags
& flags
) {
635 * after adding space to the filesystem, we need to clear the full flags
636 * on all the space infos.
638 void btrfs_clear_space_info_full(struct btrfs_fs_info
*info
)
640 struct list_head
*head
= &info
->space_info
;
641 struct btrfs_space_info
*found
;
644 list_for_each_entry_rcu(found
, head
, list
)
649 static u64
div_factor(u64 num
, int factor
)
658 static u64
div_factor_fine(u64 num
, int factor
)
667 u64
btrfs_find_block_group(struct btrfs_root
*root
,
668 u64 search_start
, u64 search_hint
, int owner
)
670 struct btrfs_block_group_cache
*cache
;
672 u64 last
= max(search_hint
, search_start
);
679 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
683 spin_lock(&cache
->lock
);
684 last
= cache
->key
.objectid
+ cache
->key
.offset
;
685 used
= btrfs_block_group_used(&cache
->item
);
687 if ((full_search
|| !cache
->ro
) &&
688 block_group_bits(cache
, BTRFS_BLOCK_GROUP_METADATA
)) {
689 if (used
+ cache
->pinned
+ cache
->reserved
<
690 div_factor(cache
->key
.offset
, factor
)) {
691 group_start
= cache
->key
.objectid
;
692 spin_unlock(&cache
->lock
);
693 btrfs_put_block_group(cache
);
697 spin_unlock(&cache
->lock
);
698 btrfs_put_block_group(cache
);
706 if (!full_search
&& factor
< 10) {
716 /* simple helper to search for an existing extent at a given offset */
717 int btrfs_lookup_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
720 struct btrfs_key key
;
721 struct btrfs_path
*path
;
723 path
= btrfs_alloc_path();
727 key
.objectid
= start
;
729 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
730 ret
= btrfs_search_slot(NULL
, root
->fs_info
->extent_root
, &key
, path
,
732 btrfs_free_path(path
);
737 * helper function to lookup reference count and flags of extent.
739 * the head node for delayed ref is used to store the sum of all the
740 * reference count modifications queued up in the rbtree. the head
741 * node may also store the extent flags to set. This way you can check
742 * to see what the reference count and extent flags would be if all of
743 * the delayed refs are not processed.
745 int btrfs_lookup_extent_info(struct btrfs_trans_handle
*trans
,
746 struct btrfs_root
*root
, u64 bytenr
,
747 u64 num_bytes
, u64
*refs
, u64
*flags
)
749 struct btrfs_delayed_ref_head
*head
;
750 struct btrfs_delayed_ref_root
*delayed_refs
;
751 struct btrfs_path
*path
;
752 struct btrfs_extent_item
*ei
;
753 struct extent_buffer
*leaf
;
754 struct btrfs_key key
;
760 path
= btrfs_alloc_path();
764 key
.objectid
= bytenr
;
765 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
766 key
.offset
= num_bytes
;
768 path
->skip_locking
= 1;
769 path
->search_commit_root
= 1;
772 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
,
778 leaf
= path
->nodes
[0];
779 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
780 if (item_size
>= sizeof(*ei
)) {
781 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
782 struct btrfs_extent_item
);
783 num_refs
= btrfs_extent_refs(leaf
, ei
);
784 extent_flags
= btrfs_extent_flags(leaf
, ei
);
786 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
787 struct btrfs_extent_item_v0
*ei0
;
788 BUG_ON(item_size
!= sizeof(*ei0
));
789 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
790 struct btrfs_extent_item_v0
);
791 num_refs
= btrfs_extent_refs_v0(leaf
, ei0
);
792 /* FIXME: this isn't correct for data */
793 extent_flags
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
798 BUG_ON(num_refs
== 0);
808 delayed_refs
= &trans
->transaction
->delayed_refs
;
809 spin_lock(&delayed_refs
->lock
);
810 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
812 if (!mutex_trylock(&head
->mutex
)) {
813 atomic_inc(&head
->node
.refs
);
814 spin_unlock(&delayed_refs
->lock
);
816 btrfs_release_path(path
);
819 * Mutex was contended, block until it's released and try
822 mutex_lock(&head
->mutex
);
823 mutex_unlock(&head
->mutex
);
824 btrfs_put_delayed_ref(&head
->node
);
827 if (head
->extent_op
&& head
->extent_op
->update_flags
)
828 extent_flags
|= head
->extent_op
->flags_to_set
;
830 BUG_ON(num_refs
== 0);
832 num_refs
+= head
->node
.ref_mod
;
833 mutex_unlock(&head
->mutex
);
835 spin_unlock(&delayed_refs
->lock
);
837 WARN_ON(num_refs
== 0);
841 *flags
= extent_flags
;
843 btrfs_free_path(path
);
848 * Back reference rules. Back refs have three main goals:
850 * 1) differentiate between all holders of references to an extent so that
851 * when a reference is dropped we can make sure it was a valid reference
852 * before freeing the extent.
854 * 2) Provide enough information to quickly find the holders of an extent
855 * if we notice a given block is corrupted or bad.
857 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
858 * maintenance. This is actually the same as #2, but with a slightly
859 * different use case.
861 * There are two kinds of back refs. The implicit back refs is optimized
862 * for pointers in non-shared tree blocks. For a given pointer in a block,
863 * back refs of this kind provide information about the block's owner tree
864 * and the pointer's key. These information allow us to find the block by
865 * b-tree searching. The full back refs is for pointers in tree blocks not
866 * referenced by their owner trees. The location of tree block is recorded
867 * in the back refs. Actually the full back refs is generic, and can be
868 * used in all cases the implicit back refs is used. The major shortcoming
869 * of the full back refs is its overhead. Every time a tree block gets
870 * COWed, we have to update back refs entry for all pointers in it.
872 * For a newly allocated tree block, we use implicit back refs for
873 * pointers in it. This means most tree related operations only involve
874 * implicit back refs. For a tree block created in old transaction, the
875 * only way to drop a reference to it is COW it. So we can detect the
876 * event that tree block loses its owner tree's reference and do the
877 * back refs conversion.
879 * When a tree block is COW'd through a tree, there are four cases:
881 * The reference count of the block is one and the tree is the block's
882 * owner tree. Nothing to do in this case.
884 * The reference count of the block is one and the tree is not the
885 * block's owner tree. In this case, full back refs is used for pointers
886 * in the block. Remove these full back refs, add implicit back refs for
887 * every pointers in the new block.
889 * The reference count of the block is greater than one and the tree is
890 * the block's owner tree. In this case, implicit back refs is used for
891 * pointers in the block. Add full back refs for every pointers in the
892 * block, increase lower level extents' reference counts. The original
893 * implicit back refs are entailed to the new block.
895 * The reference count of the block is greater than one and the tree is
896 * not the block's owner tree. Add implicit back refs for every pointer in
897 * the new block, increase lower level extents' reference count.
899 * Back Reference Key composing:
901 * The key objectid corresponds to the first byte in the extent,
902 * The key type is used to differentiate between types of back refs.
903 * There are different meanings of the key offset for different types
906 * File extents can be referenced by:
908 * - multiple snapshots, subvolumes, or different generations in one subvol
909 * - different files inside a single subvolume
910 * - different offsets inside a file (bookend extents in file.c)
912 * The extent ref structure for the implicit back refs has fields for:
914 * - Objectid of the subvolume root
915 * - objectid of the file holding the reference
916 * - original offset in the file
917 * - how many bookend extents
919 * The key offset for the implicit back refs is hash of the first
922 * The extent ref structure for the full back refs has field for:
924 * - number of pointers in the tree leaf
926 * The key offset for the implicit back refs is the first byte of
929 * When a file extent is allocated, The implicit back refs is used.
930 * the fields are filled in:
932 * (root_key.objectid, inode objectid, offset in file, 1)
934 * When a file extent is removed file truncation, we find the
935 * corresponding implicit back refs and check the following fields:
937 * (btrfs_header_owner(leaf), inode objectid, offset in file)
939 * Btree extents can be referenced by:
941 * - Different subvolumes
943 * Both the implicit back refs and the full back refs for tree blocks
944 * only consist of key. The key offset for the implicit back refs is
945 * objectid of block's owner tree. The key offset for the full back refs
946 * is the first byte of parent block.
948 * When implicit back refs is used, information about the lowest key and
949 * level of the tree block are required. These information are stored in
950 * tree block info structure.
953 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
954 static int convert_extent_item_v0(struct btrfs_trans_handle
*trans
,
955 struct btrfs_root
*root
,
956 struct btrfs_path
*path
,
957 u64 owner
, u32 extra_size
)
959 struct btrfs_extent_item
*item
;
960 struct btrfs_extent_item_v0
*ei0
;
961 struct btrfs_extent_ref_v0
*ref0
;
962 struct btrfs_tree_block_info
*bi
;
963 struct extent_buffer
*leaf
;
964 struct btrfs_key key
;
965 struct btrfs_key found_key
;
966 u32 new_size
= sizeof(*item
);
970 leaf
= path
->nodes
[0];
971 BUG_ON(btrfs_item_size_nr(leaf
, path
->slots
[0]) != sizeof(*ei0
));
973 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
974 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
975 struct btrfs_extent_item_v0
);
976 refs
= btrfs_extent_refs_v0(leaf
, ei0
);
978 if (owner
== (u64
)-1) {
980 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
981 ret
= btrfs_next_leaf(root
, path
);
985 leaf
= path
->nodes
[0];
987 btrfs_item_key_to_cpu(leaf
, &found_key
,
989 BUG_ON(key
.objectid
!= found_key
.objectid
);
990 if (found_key
.type
!= BTRFS_EXTENT_REF_V0_KEY
) {
994 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
995 struct btrfs_extent_ref_v0
);
996 owner
= btrfs_ref_objectid_v0(leaf
, ref0
);
1000 btrfs_release_path(path
);
1002 if (owner
< BTRFS_FIRST_FREE_OBJECTID
)
1003 new_size
+= sizeof(*bi
);
1005 new_size
-= sizeof(*ei0
);
1006 ret
= btrfs_search_slot(trans
, root
, &key
, path
,
1007 new_size
+ extra_size
, 1);
1012 ret
= btrfs_extend_item(trans
, root
, path
, new_size
);
1014 leaf
= path
->nodes
[0];
1015 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1016 btrfs_set_extent_refs(leaf
, item
, refs
);
1017 /* FIXME: get real generation */
1018 btrfs_set_extent_generation(leaf
, item
, 0);
1019 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1020 btrfs_set_extent_flags(leaf
, item
,
1021 BTRFS_EXTENT_FLAG_TREE_BLOCK
|
1022 BTRFS_BLOCK_FLAG_FULL_BACKREF
);
1023 bi
= (struct btrfs_tree_block_info
*)(item
+ 1);
1024 /* FIXME: get first key of the block */
1025 memset_extent_buffer(leaf
, 0, (unsigned long)bi
, sizeof(*bi
));
1026 btrfs_set_tree_block_level(leaf
, bi
, (int)owner
);
1028 btrfs_set_extent_flags(leaf
, item
, BTRFS_EXTENT_FLAG_DATA
);
1030 btrfs_mark_buffer_dirty(leaf
);
1035 static u64
hash_extent_data_ref(u64 root_objectid
, u64 owner
, u64 offset
)
1037 u32 high_crc
= ~(u32
)0;
1038 u32 low_crc
= ~(u32
)0;
1041 lenum
= cpu_to_le64(root_objectid
);
1042 high_crc
= crc32c(high_crc
, &lenum
, sizeof(lenum
));
1043 lenum
= cpu_to_le64(owner
);
1044 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
1045 lenum
= cpu_to_le64(offset
);
1046 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
1048 return ((u64
)high_crc
<< 31) ^ (u64
)low_crc
;
1051 static u64
hash_extent_data_ref_item(struct extent_buffer
*leaf
,
1052 struct btrfs_extent_data_ref
*ref
)
1054 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf
, ref
),
1055 btrfs_extent_data_ref_objectid(leaf
, ref
),
1056 btrfs_extent_data_ref_offset(leaf
, ref
));
1059 static int match_extent_data_ref(struct extent_buffer
*leaf
,
1060 struct btrfs_extent_data_ref
*ref
,
1061 u64 root_objectid
, u64 owner
, u64 offset
)
1063 if (btrfs_extent_data_ref_root(leaf
, ref
) != root_objectid
||
1064 btrfs_extent_data_ref_objectid(leaf
, ref
) != owner
||
1065 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
1070 static noinline
int lookup_extent_data_ref(struct btrfs_trans_handle
*trans
,
1071 struct btrfs_root
*root
,
1072 struct btrfs_path
*path
,
1073 u64 bytenr
, u64 parent
,
1075 u64 owner
, u64 offset
)
1077 struct btrfs_key key
;
1078 struct btrfs_extent_data_ref
*ref
;
1079 struct extent_buffer
*leaf
;
1085 key
.objectid
= bytenr
;
1087 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1088 key
.offset
= parent
;
1090 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1091 key
.offset
= hash_extent_data_ref(root_objectid
,
1096 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1105 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1106 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1107 btrfs_release_path(path
);
1108 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1119 leaf
= path
->nodes
[0];
1120 nritems
= btrfs_header_nritems(leaf
);
1122 if (path
->slots
[0] >= nritems
) {
1123 ret
= btrfs_next_leaf(root
, path
);
1129 leaf
= path
->nodes
[0];
1130 nritems
= btrfs_header_nritems(leaf
);
1134 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1135 if (key
.objectid
!= bytenr
||
1136 key
.type
!= BTRFS_EXTENT_DATA_REF_KEY
)
1139 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1140 struct btrfs_extent_data_ref
);
1142 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1145 btrfs_release_path(path
);
1157 static noinline
int insert_extent_data_ref(struct btrfs_trans_handle
*trans
,
1158 struct btrfs_root
*root
,
1159 struct btrfs_path
*path
,
1160 u64 bytenr
, u64 parent
,
1161 u64 root_objectid
, u64 owner
,
1162 u64 offset
, int refs_to_add
)
1164 struct btrfs_key key
;
1165 struct extent_buffer
*leaf
;
1170 key
.objectid
= bytenr
;
1172 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1173 key
.offset
= parent
;
1174 size
= sizeof(struct btrfs_shared_data_ref
);
1176 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1177 key
.offset
= hash_extent_data_ref(root_objectid
,
1179 size
= sizeof(struct btrfs_extent_data_ref
);
1182 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, size
);
1183 if (ret
&& ret
!= -EEXIST
)
1186 leaf
= path
->nodes
[0];
1188 struct btrfs_shared_data_ref
*ref
;
1189 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1190 struct btrfs_shared_data_ref
);
1192 btrfs_set_shared_data_ref_count(leaf
, ref
, refs_to_add
);
1194 num_refs
= btrfs_shared_data_ref_count(leaf
, ref
);
1195 num_refs
+= refs_to_add
;
1196 btrfs_set_shared_data_ref_count(leaf
, ref
, num_refs
);
1199 struct btrfs_extent_data_ref
*ref
;
1200 while (ret
== -EEXIST
) {
1201 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1202 struct btrfs_extent_data_ref
);
1203 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1206 btrfs_release_path(path
);
1208 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1210 if (ret
&& ret
!= -EEXIST
)
1213 leaf
= path
->nodes
[0];
1215 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1216 struct btrfs_extent_data_ref
);
1218 btrfs_set_extent_data_ref_root(leaf
, ref
,
1220 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
1221 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
1222 btrfs_set_extent_data_ref_count(leaf
, ref
, refs_to_add
);
1224 num_refs
= btrfs_extent_data_ref_count(leaf
, ref
);
1225 num_refs
+= refs_to_add
;
1226 btrfs_set_extent_data_ref_count(leaf
, ref
, num_refs
);
1229 btrfs_mark_buffer_dirty(leaf
);
1232 btrfs_release_path(path
);
1236 static noinline
int remove_extent_data_ref(struct btrfs_trans_handle
*trans
,
1237 struct btrfs_root
*root
,
1238 struct btrfs_path
*path
,
1241 struct btrfs_key key
;
1242 struct btrfs_extent_data_ref
*ref1
= NULL
;
1243 struct btrfs_shared_data_ref
*ref2
= NULL
;
1244 struct extent_buffer
*leaf
;
1248 leaf
= path
->nodes
[0];
1249 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1251 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1252 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1253 struct btrfs_extent_data_ref
);
1254 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1255 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1256 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1257 struct btrfs_shared_data_ref
);
1258 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1259 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1260 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1261 struct btrfs_extent_ref_v0
*ref0
;
1262 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1263 struct btrfs_extent_ref_v0
);
1264 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1270 BUG_ON(num_refs
< refs_to_drop
);
1271 num_refs
-= refs_to_drop
;
1273 if (num_refs
== 0) {
1274 ret
= btrfs_del_item(trans
, root
, path
);
1276 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
)
1277 btrfs_set_extent_data_ref_count(leaf
, ref1
, num_refs
);
1278 else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
)
1279 btrfs_set_shared_data_ref_count(leaf
, ref2
, num_refs
);
1280 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1282 struct btrfs_extent_ref_v0
*ref0
;
1283 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1284 struct btrfs_extent_ref_v0
);
1285 btrfs_set_ref_count_v0(leaf
, ref0
, num_refs
);
1288 btrfs_mark_buffer_dirty(leaf
);
1293 static noinline u32
extent_data_ref_count(struct btrfs_root
*root
,
1294 struct btrfs_path
*path
,
1295 struct btrfs_extent_inline_ref
*iref
)
1297 struct btrfs_key key
;
1298 struct extent_buffer
*leaf
;
1299 struct btrfs_extent_data_ref
*ref1
;
1300 struct btrfs_shared_data_ref
*ref2
;
1303 leaf
= path
->nodes
[0];
1304 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1306 if (btrfs_extent_inline_ref_type(leaf
, iref
) ==
1307 BTRFS_EXTENT_DATA_REF_KEY
) {
1308 ref1
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1309 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1311 ref2
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1312 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1314 } else if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1315 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1316 struct btrfs_extent_data_ref
);
1317 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1318 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1319 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1320 struct btrfs_shared_data_ref
);
1321 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1322 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1323 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1324 struct btrfs_extent_ref_v0
*ref0
;
1325 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1326 struct btrfs_extent_ref_v0
);
1327 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1335 static noinline
int lookup_tree_block_ref(struct btrfs_trans_handle
*trans
,
1336 struct btrfs_root
*root
,
1337 struct btrfs_path
*path
,
1338 u64 bytenr
, u64 parent
,
1341 struct btrfs_key key
;
1344 key
.objectid
= bytenr
;
1346 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1347 key
.offset
= parent
;
1349 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1350 key
.offset
= root_objectid
;
1353 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1356 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1357 if (ret
== -ENOENT
&& parent
) {
1358 btrfs_release_path(path
);
1359 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1360 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1368 static noinline
int insert_tree_block_ref(struct btrfs_trans_handle
*trans
,
1369 struct btrfs_root
*root
,
1370 struct btrfs_path
*path
,
1371 u64 bytenr
, u64 parent
,
1374 struct btrfs_key key
;
1377 key
.objectid
= bytenr
;
1379 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1380 key
.offset
= parent
;
1382 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1383 key
.offset
= root_objectid
;
1386 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
1387 btrfs_release_path(path
);
1391 static inline int extent_ref_type(u64 parent
, u64 owner
)
1394 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1396 type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1398 type
= BTRFS_TREE_BLOCK_REF_KEY
;
1401 type
= BTRFS_SHARED_DATA_REF_KEY
;
1403 type
= BTRFS_EXTENT_DATA_REF_KEY
;
1408 static int find_next_key(struct btrfs_path
*path
, int level
,
1409 struct btrfs_key
*key
)
1412 for (; level
< BTRFS_MAX_LEVEL
; level
++) {
1413 if (!path
->nodes
[level
])
1415 if (path
->slots
[level
] + 1 >=
1416 btrfs_header_nritems(path
->nodes
[level
]))
1419 btrfs_item_key_to_cpu(path
->nodes
[level
], key
,
1420 path
->slots
[level
] + 1);
1422 btrfs_node_key_to_cpu(path
->nodes
[level
], key
,
1423 path
->slots
[level
] + 1);
1430 * look for inline back ref. if back ref is found, *ref_ret is set
1431 * to the address of inline back ref, and 0 is returned.
1433 * if back ref isn't found, *ref_ret is set to the address where it
1434 * should be inserted, and -ENOENT is returned.
1436 * if insert is true and there are too many inline back refs, the path
1437 * points to the extent item, and -EAGAIN is returned.
1439 * NOTE: inline back refs are ordered in the same way that back ref
1440 * items in the tree are ordered.
1442 static noinline_for_stack
1443 int lookup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1444 struct btrfs_root
*root
,
1445 struct btrfs_path
*path
,
1446 struct btrfs_extent_inline_ref
**ref_ret
,
1447 u64 bytenr
, u64 num_bytes
,
1448 u64 parent
, u64 root_objectid
,
1449 u64 owner
, u64 offset
, int insert
)
1451 struct btrfs_key key
;
1452 struct extent_buffer
*leaf
;
1453 struct btrfs_extent_item
*ei
;
1454 struct btrfs_extent_inline_ref
*iref
;
1465 key
.objectid
= bytenr
;
1466 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1467 key
.offset
= num_bytes
;
1469 want
= extent_ref_type(parent
, owner
);
1471 extra_size
= btrfs_extent_inline_ref_size(want
);
1472 path
->keep_locks
= 1;
1475 ret
= btrfs_search_slot(trans
, root
, &key
, path
, extra_size
, 1);
1482 leaf
= path
->nodes
[0];
1483 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1484 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1485 if (item_size
< sizeof(*ei
)) {
1490 ret
= convert_extent_item_v0(trans
, root
, path
, owner
,
1496 leaf
= path
->nodes
[0];
1497 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1500 BUG_ON(item_size
< sizeof(*ei
));
1502 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1503 flags
= btrfs_extent_flags(leaf
, ei
);
1505 ptr
= (unsigned long)(ei
+ 1);
1506 end
= (unsigned long)ei
+ item_size
;
1508 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
) {
1509 ptr
+= sizeof(struct btrfs_tree_block_info
);
1512 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_DATA
));
1521 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1522 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1526 ptr
+= btrfs_extent_inline_ref_size(type
);
1530 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1531 struct btrfs_extent_data_ref
*dref
;
1532 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1533 if (match_extent_data_ref(leaf
, dref
, root_objectid
,
1538 if (hash_extent_data_ref_item(leaf
, dref
) <
1539 hash_extent_data_ref(root_objectid
, owner
, offset
))
1543 ref_offset
= btrfs_extent_inline_ref_offset(leaf
, iref
);
1545 if (parent
== ref_offset
) {
1549 if (ref_offset
< parent
)
1552 if (root_objectid
== ref_offset
) {
1556 if (ref_offset
< root_objectid
)
1560 ptr
+= btrfs_extent_inline_ref_size(type
);
1562 if (err
== -ENOENT
&& insert
) {
1563 if (item_size
+ extra_size
>=
1564 BTRFS_MAX_EXTENT_ITEM_SIZE(root
)) {
1569 * To add new inline back ref, we have to make sure
1570 * there is no corresponding back ref item.
1571 * For simplicity, we just do not add new inline back
1572 * ref if there is any kind of item for this block
1574 if (find_next_key(path
, 0, &key
) == 0 &&
1575 key
.objectid
== bytenr
&&
1576 key
.type
< BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1581 *ref_ret
= (struct btrfs_extent_inline_ref
*)ptr
;
1584 path
->keep_locks
= 0;
1585 btrfs_unlock_up_safe(path
, 1);
1591 * helper to add new inline back ref
1593 static noinline_for_stack
1594 int setup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1595 struct btrfs_root
*root
,
1596 struct btrfs_path
*path
,
1597 struct btrfs_extent_inline_ref
*iref
,
1598 u64 parent
, u64 root_objectid
,
1599 u64 owner
, u64 offset
, int refs_to_add
,
1600 struct btrfs_delayed_extent_op
*extent_op
)
1602 struct extent_buffer
*leaf
;
1603 struct btrfs_extent_item
*ei
;
1606 unsigned long item_offset
;
1612 leaf
= path
->nodes
[0];
1613 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1614 item_offset
= (unsigned long)iref
- (unsigned long)ei
;
1616 type
= extent_ref_type(parent
, owner
);
1617 size
= btrfs_extent_inline_ref_size(type
);
1619 ret
= btrfs_extend_item(trans
, root
, path
, size
);
1621 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1622 refs
= btrfs_extent_refs(leaf
, ei
);
1623 refs
+= refs_to_add
;
1624 btrfs_set_extent_refs(leaf
, ei
, refs
);
1626 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1628 ptr
= (unsigned long)ei
+ item_offset
;
1629 end
= (unsigned long)ei
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
1630 if (ptr
< end
- size
)
1631 memmove_extent_buffer(leaf
, ptr
+ size
, ptr
,
1634 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1635 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
1636 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1637 struct btrfs_extent_data_ref
*dref
;
1638 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1639 btrfs_set_extent_data_ref_root(leaf
, dref
, root_objectid
);
1640 btrfs_set_extent_data_ref_objectid(leaf
, dref
, owner
);
1641 btrfs_set_extent_data_ref_offset(leaf
, dref
, offset
);
1642 btrfs_set_extent_data_ref_count(leaf
, dref
, refs_to_add
);
1643 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1644 struct btrfs_shared_data_ref
*sref
;
1645 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1646 btrfs_set_shared_data_ref_count(leaf
, sref
, refs_to_add
);
1647 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1648 } else if (type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
1649 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1651 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
1653 btrfs_mark_buffer_dirty(leaf
);
1657 static int lookup_extent_backref(struct btrfs_trans_handle
*trans
,
1658 struct btrfs_root
*root
,
1659 struct btrfs_path
*path
,
1660 struct btrfs_extent_inline_ref
**ref_ret
,
1661 u64 bytenr
, u64 num_bytes
, u64 parent
,
1662 u64 root_objectid
, u64 owner
, u64 offset
)
1666 ret
= lookup_inline_extent_backref(trans
, root
, path
, ref_ret
,
1667 bytenr
, num_bytes
, parent
,
1668 root_objectid
, owner
, offset
, 0);
1672 btrfs_release_path(path
);
1675 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1676 ret
= lookup_tree_block_ref(trans
, root
, path
, bytenr
, parent
,
1679 ret
= lookup_extent_data_ref(trans
, root
, path
, bytenr
, parent
,
1680 root_objectid
, owner
, offset
);
1686 * helper to update/remove inline back ref
1688 static noinline_for_stack
1689 int update_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1690 struct btrfs_root
*root
,
1691 struct btrfs_path
*path
,
1692 struct btrfs_extent_inline_ref
*iref
,
1694 struct btrfs_delayed_extent_op
*extent_op
)
1696 struct extent_buffer
*leaf
;
1697 struct btrfs_extent_item
*ei
;
1698 struct btrfs_extent_data_ref
*dref
= NULL
;
1699 struct btrfs_shared_data_ref
*sref
= NULL
;
1708 leaf
= path
->nodes
[0];
1709 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1710 refs
= btrfs_extent_refs(leaf
, ei
);
1711 WARN_ON(refs_to_mod
< 0 && refs
+ refs_to_mod
<= 0);
1712 refs
+= refs_to_mod
;
1713 btrfs_set_extent_refs(leaf
, ei
, refs
);
1715 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1717 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1719 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1720 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1721 refs
= btrfs_extent_data_ref_count(leaf
, dref
);
1722 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1723 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1724 refs
= btrfs_shared_data_ref_count(leaf
, sref
);
1727 BUG_ON(refs_to_mod
!= -1);
1730 BUG_ON(refs_to_mod
< 0 && refs
< -refs_to_mod
);
1731 refs
+= refs_to_mod
;
1734 if (type
== BTRFS_EXTENT_DATA_REF_KEY
)
1735 btrfs_set_extent_data_ref_count(leaf
, dref
, refs
);
1737 btrfs_set_shared_data_ref_count(leaf
, sref
, refs
);
1739 size
= btrfs_extent_inline_ref_size(type
);
1740 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1741 ptr
= (unsigned long)iref
;
1742 end
= (unsigned long)ei
+ item_size
;
1743 if (ptr
+ size
< end
)
1744 memmove_extent_buffer(leaf
, ptr
, ptr
+ size
,
1747 ret
= btrfs_truncate_item(trans
, root
, path
, item_size
, 1);
1749 btrfs_mark_buffer_dirty(leaf
);
1753 static noinline_for_stack
1754 int insert_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1755 struct btrfs_root
*root
,
1756 struct btrfs_path
*path
,
1757 u64 bytenr
, u64 num_bytes
, u64 parent
,
1758 u64 root_objectid
, u64 owner
,
1759 u64 offset
, int refs_to_add
,
1760 struct btrfs_delayed_extent_op
*extent_op
)
1762 struct btrfs_extent_inline_ref
*iref
;
1765 ret
= lookup_inline_extent_backref(trans
, root
, path
, &iref
,
1766 bytenr
, num_bytes
, parent
,
1767 root_objectid
, owner
, offset
, 1);
1769 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
);
1770 ret
= update_inline_extent_backref(trans
, root
, path
, iref
,
1771 refs_to_add
, extent_op
);
1772 } else if (ret
== -ENOENT
) {
1773 ret
= setup_inline_extent_backref(trans
, root
, path
, iref
,
1774 parent
, root_objectid
,
1775 owner
, offset
, refs_to_add
,
1781 static int insert_extent_backref(struct btrfs_trans_handle
*trans
,
1782 struct btrfs_root
*root
,
1783 struct btrfs_path
*path
,
1784 u64 bytenr
, u64 parent
, u64 root_objectid
,
1785 u64 owner
, u64 offset
, int refs_to_add
)
1788 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1789 BUG_ON(refs_to_add
!= 1);
1790 ret
= insert_tree_block_ref(trans
, root
, path
, bytenr
,
1791 parent
, root_objectid
);
1793 ret
= insert_extent_data_ref(trans
, root
, path
, bytenr
,
1794 parent
, root_objectid
,
1795 owner
, offset
, refs_to_add
);
1800 static int remove_extent_backref(struct btrfs_trans_handle
*trans
,
1801 struct btrfs_root
*root
,
1802 struct btrfs_path
*path
,
1803 struct btrfs_extent_inline_ref
*iref
,
1804 int refs_to_drop
, int is_data
)
1808 BUG_ON(!is_data
&& refs_to_drop
!= 1);
1810 ret
= update_inline_extent_backref(trans
, root
, path
, iref
,
1811 -refs_to_drop
, NULL
);
1812 } else if (is_data
) {
1813 ret
= remove_extent_data_ref(trans
, root
, path
, refs_to_drop
);
1815 ret
= btrfs_del_item(trans
, root
, path
);
1820 static int btrfs_issue_discard(struct block_device
*bdev
,
1823 return blkdev_issue_discard(bdev
, start
>> 9, len
>> 9, GFP_NOFS
, 0);
1826 static int btrfs_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
1827 u64 num_bytes
, u64
*actual_bytes
)
1830 u64 discarded_bytes
= 0;
1831 struct btrfs_bio
*bbio
= NULL
;
1834 /* Tell the block device(s) that the sectors can be discarded */
1835 ret
= btrfs_map_block(&root
->fs_info
->mapping_tree
, REQ_DISCARD
,
1836 bytenr
, &num_bytes
, &bbio
, 0);
1838 struct btrfs_bio_stripe
*stripe
= bbio
->stripes
;
1842 for (i
= 0; i
< bbio
->num_stripes
; i
++, stripe
++) {
1843 if (!stripe
->dev
->can_discard
)
1846 ret
= btrfs_issue_discard(stripe
->dev
->bdev
,
1850 discarded_bytes
+= stripe
->length
;
1851 else if (ret
!= -EOPNOTSUPP
)
1855 * Just in case we get back EOPNOTSUPP for some reason,
1856 * just ignore the return value so we don't screw up
1857 * people calling discard_extent.
1865 *actual_bytes
= discarded_bytes
;
1871 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1872 struct btrfs_root
*root
,
1873 u64 bytenr
, u64 num_bytes
, u64 parent
,
1874 u64 root_objectid
, u64 owner
, u64 offset
)
1877 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
&&
1878 root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
1880 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1881 ret
= btrfs_add_delayed_tree_ref(trans
, bytenr
, num_bytes
,
1882 parent
, root_objectid
, (int)owner
,
1883 BTRFS_ADD_DELAYED_REF
, NULL
);
1885 ret
= btrfs_add_delayed_data_ref(trans
, bytenr
, num_bytes
,
1886 parent
, root_objectid
, owner
, offset
,
1887 BTRFS_ADD_DELAYED_REF
, NULL
);
1892 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1893 struct btrfs_root
*root
,
1894 u64 bytenr
, u64 num_bytes
,
1895 u64 parent
, u64 root_objectid
,
1896 u64 owner
, u64 offset
, int refs_to_add
,
1897 struct btrfs_delayed_extent_op
*extent_op
)
1899 struct btrfs_path
*path
;
1900 struct extent_buffer
*leaf
;
1901 struct btrfs_extent_item
*item
;
1906 path
= btrfs_alloc_path();
1911 path
->leave_spinning
= 1;
1912 /* this will setup the path even if it fails to insert the back ref */
1913 ret
= insert_inline_extent_backref(trans
, root
->fs_info
->extent_root
,
1914 path
, bytenr
, num_bytes
, parent
,
1915 root_objectid
, owner
, offset
,
1916 refs_to_add
, extent_op
);
1920 if (ret
!= -EAGAIN
) {
1925 leaf
= path
->nodes
[0];
1926 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1927 refs
= btrfs_extent_refs(leaf
, item
);
1928 btrfs_set_extent_refs(leaf
, item
, refs
+ refs_to_add
);
1930 __run_delayed_extent_op(extent_op
, leaf
, item
);
1932 btrfs_mark_buffer_dirty(leaf
);
1933 btrfs_release_path(path
);
1936 path
->leave_spinning
= 1;
1938 /* now insert the actual backref */
1939 ret
= insert_extent_backref(trans
, root
->fs_info
->extent_root
,
1940 path
, bytenr
, parent
, root_objectid
,
1941 owner
, offset
, refs_to_add
);
1944 btrfs_free_path(path
);
1948 static int run_delayed_data_ref(struct btrfs_trans_handle
*trans
,
1949 struct btrfs_root
*root
,
1950 struct btrfs_delayed_ref_node
*node
,
1951 struct btrfs_delayed_extent_op
*extent_op
,
1952 int insert_reserved
)
1955 struct btrfs_delayed_data_ref
*ref
;
1956 struct btrfs_key ins
;
1961 ins
.objectid
= node
->bytenr
;
1962 ins
.offset
= node
->num_bytes
;
1963 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1965 ref
= btrfs_delayed_node_to_data_ref(node
);
1966 if (node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
1967 parent
= ref
->parent
;
1969 ref_root
= ref
->root
;
1971 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
1973 BUG_ON(extent_op
->update_key
);
1974 flags
|= extent_op
->flags_to_set
;
1976 ret
= alloc_reserved_file_extent(trans
, root
,
1977 parent
, ref_root
, flags
,
1978 ref
->objectid
, ref
->offset
,
1979 &ins
, node
->ref_mod
);
1980 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
1981 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
1982 node
->num_bytes
, parent
,
1983 ref_root
, ref
->objectid
,
1984 ref
->offset
, node
->ref_mod
,
1986 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
1987 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
1988 node
->num_bytes
, parent
,
1989 ref_root
, ref
->objectid
,
1990 ref
->offset
, node
->ref_mod
,
1998 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
1999 struct extent_buffer
*leaf
,
2000 struct btrfs_extent_item
*ei
)
2002 u64 flags
= btrfs_extent_flags(leaf
, ei
);
2003 if (extent_op
->update_flags
) {
2004 flags
|= extent_op
->flags_to_set
;
2005 btrfs_set_extent_flags(leaf
, ei
, flags
);
2008 if (extent_op
->update_key
) {
2009 struct btrfs_tree_block_info
*bi
;
2010 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
));
2011 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
2012 btrfs_set_tree_block_key(leaf
, bi
, &extent_op
->key
);
2016 static int run_delayed_extent_op(struct btrfs_trans_handle
*trans
,
2017 struct btrfs_root
*root
,
2018 struct btrfs_delayed_ref_node
*node
,
2019 struct btrfs_delayed_extent_op
*extent_op
)
2021 struct btrfs_key key
;
2022 struct btrfs_path
*path
;
2023 struct btrfs_extent_item
*ei
;
2024 struct extent_buffer
*leaf
;
2029 path
= btrfs_alloc_path();
2033 key
.objectid
= node
->bytenr
;
2034 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2035 key
.offset
= node
->num_bytes
;
2038 path
->leave_spinning
= 1;
2039 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
,
2050 leaf
= path
->nodes
[0];
2051 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2052 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2053 if (item_size
< sizeof(*ei
)) {
2054 ret
= convert_extent_item_v0(trans
, root
->fs_info
->extent_root
,
2060 leaf
= path
->nodes
[0];
2061 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2064 BUG_ON(item_size
< sizeof(*ei
));
2065 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2066 __run_delayed_extent_op(extent_op
, leaf
, ei
);
2068 btrfs_mark_buffer_dirty(leaf
);
2070 btrfs_free_path(path
);
2074 static int run_delayed_tree_ref(struct btrfs_trans_handle
*trans
,
2075 struct btrfs_root
*root
,
2076 struct btrfs_delayed_ref_node
*node
,
2077 struct btrfs_delayed_extent_op
*extent_op
,
2078 int insert_reserved
)
2081 struct btrfs_delayed_tree_ref
*ref
;
2082 struct btrfs_key ins
;
2086 ins
.objectid
= node
->bytenr
;
2087 ins
.offset
= node
->num_bytes
;
2088 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2090 ref
= btrfs_delayed_node_to_tree_ref(node
);
2091 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2092 parent
= ref
->parent
;
2094 ref_root
= ref
->root
;
2096 BUG_ON(node
->ref_mod
!= 1);
2097 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2098 BUG_ON(!extent_op
|| !extent_op
->update_flags
||
2099 !extent_op
->update_key
);
2100 ret
= alloc_reserved_tree_block(trans
, root
,
2102 extent_op
->flags_to_set
,
2105 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2106 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
2107 node
->num_bytes
, parent
, ref_root
,
2108 ref
->level
, 0, 1, extent_op
);
2109 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2110 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
2111 node
->num_bytes
, parent
, ref_root
,
2112 ref
->level
, 0, 1, extent_op
);
2119 /* helper function to actually process a single delayed ref entry */
2120 static int run_one_delayed_ref(struct btrfs_trans_handle
*trans
,
2121 struct btrfs_root
*root
,
2122 struct btrfs_delayed_ref_node
*node
,
2123 struct btrfs_delayed_extent_op
*extent_op
,
2124 int insert_reserved
)
2127 if (btrfs_delayed_ref_is_head(node
)) {
2128 struct btrfs_delayed_ref_head
*head
;
2130 * we've hit the end of the chain and we were supposed
2131 * to insert this extent into the tree. But, it got
2132 * deleted before we ever needed to insert it, so all
2133 * we have to do is clean up the accounting
2136 head
= btrfs_delayed_node_to_head(node
);
2137 if (insert_reserved
) {
2138 btrfs_pin_extent(root
, node
->bytenr
,
2139 node
->num_bytes
, 1);
2140 if (head
->is_data
) {
2141 ret
= btrfs_del_csums(trans
, root
,
2147 mutex_unlock(&head
->mutex
);
2151 if (node
->type
== BTRFS_TREE_BLOCK_REF_KEY
||
2152 node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2153 ret
= run_delayed_tree_ref(trans
, root
, node
, extent_op
,
2155 else if (node
->type
== BTRFS_EXTENT_DATA_REF_KEY
||
2156 node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2157 ret
= run_delayed_data_ref(trans
, root
, node
, extent_op
,
2164 static noinline
struct btrfs_delayed_ref_node
*
2165 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
2167 struct rb_node
*node
;
2168 struct btrfs_delayed_ref_node
*ref
;
2169 int action
= BTRFS_ADD_DELAYED_REF
;
2172 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2173 * this prevents ref count from going down to zero when
2174 * there still are pending delayed ref.
2176 node
= rb_prev(&head
->node
.rb_node
);
2180 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
2182 if (ref
->bytenr
!= head
->node
.bytenr
)
2184 if (ref
->action
== action
)
2186 node
= rb_prev(node
);
2188 if (action
== BTRFS_ADD_DELAYED_REF
) {
2189 action
= BTRFS_DROP_DELAYED_REF
;
2195 static noinline
int run_clustered_refs(struct btrfs_trans_handle
*trans
,
2196 struct btrfs_root
*root
,
2197 struct list_head
*cluster
)
2199 struct btrfs_delayed_ref_root
*delayed_refs
;
2200 struct btrfs_delayed_ref_node
*ref
;
2201 struct btrfs_delayed_ref_head
*locked_ref
= NULL
;
2202 struct btrfs_delayed_extent_op
*extent_op
;
2205 int must_insert_reserved
= 0;
2207 delayed_refs
= &trans
->transaction
->delayed_refs
;
2210 /* pick a new head ref from the cluster list */
2211 if (list_empty(cluster
))
2214 locked_ref
= list_entry(cluster
->next
,
2215 struct btrfs_delayed_ref_head
, cluster
);
2217 /* grab the lock that says we are going to process
2218 * all the refs for this head */
2219 ret
= btrfs_delayed_ref_lock(trans
, locked_ref
);
2222 * we may have dropped the spin lock to get the head
2223 * mutex lock, and that might have given someone else
2224 * time to free the head. If that's true, it has been
2225 * removed from our list and we can move on.
2227 if (ret
== -EAGAIN
) {
2235 * record the must insert reserved flag before we
2236 * drop the spin lock.
2238 must_insert_reserved
= locked_ref
->must_insert_reserved
;
2239 locked_ref
->must_insert_reserved
= 0;
2241 extent_op
= locked_ref
->extent_op
;
2242 locked_ref
->extent_op
= NULL
;
2245 * locked_ref is the head node, so we have to go one
2246 * node back for any delayed ref updates
2248 ref
= select_delayed_ref(locked_ref
);
2250 /* All delayed refs have been processed, Go ahead
2251 * and send the head node to run_one_delayed_ref,
2252 * so that any accounting fixes can happen
2254 ref
= &locked_ref
->node
;
2256 if (extent_op
&& must_insert_reserved
) {
2262 spin_unlock(&delayed_refs
->lock
);
2264 ret
= run_delayed_extent_op(trans
, root
,
2270 spin_lock(&delayed_refs
->lock
);
2274 list_del_init(&locked_ref
->cluster
);
2279 rb_erase(&ref
->rb_node
, &delayed_refs
->root
);
2280 delayed_refs
->num_entries
--;
2282 spin_unlock(&delayed_refs
->lock
);
2284 ret
= run_one_delayed_ref(trans
, root
, ref
, extent_op
,
2285 must_insert_reserved
);
2288 btrfs_put_delayed_ref(ref
);
2293 spin_lock(&delayed_refs
->lock
);
2299 * this starts processing the delayed reference count updates and
2300 * extent insertions we have queued up so far. count can be
2301 * 0, which means to process everything in the tree at the start
2302 * of the run (but not newly added entries), or it can be some target
2303 * number you'd like to process.
2305 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2306 struct btrfs_root
*root
, unsigned long count
)
2308 struct rb_node
*node
;
2309 struct btrfs_delayed_ref_root
*delayed_refs
;
2310 struct btrfs_delayed_ref_node
*ref
;
2311 struct list_head cluster
;
2313 int run_all
= count
== (unsigned long)-1;
2316 if (root
== root
->fs_info
->extent_root
)
2317 root
= root
->fs_info
->tree_root
;
2319 delayed_refs
= &trans
->transaction
->delayed_refs
;
2320 INIT_LIST_HEAD(&cluster
);
2322 spin_lock(&delayed_refs
->lock
);
2324 count
= delayed_refs
->num_entries
* 2;
2328 if (!(run_all
|| run_most
) &&
2329 delayed_refs
->num_heads_ready
< 64)
2333 * go find something we can process in the rbtree. We start at
2334 * the beginning of the tree, and then build a cluster
2335 * of refs to process starting at the first one we are able to
2338 ret
= btrfs_find_ref_cluster(trans
, &cluster
,
2339 delayed_refs
->run_delayed_start
);
2343 ret
= run_clustered_refs(trans
, root
, &cluster
);
2346 count
-= min_t(unsigned long, ret
, count
);
2353 node
= rb_first(&delayed_refs
->root
);
2356 count
= (unsigned long)-1;
2359 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
2361 if (btrfs_delayed_ref_is_head(ref
)) {
2362 struct btrfs_delayed_ref_head
*head
;
2364 head
= btrfs_delayed_node_to_head(ref
);
2365 atomic_inc(&ref
->refs
);
2367 spin_unlock(&delayed_refs
->lock
);
2369 * Mutex was contended, block until it's
2370 * released and try again
2372 mutex_lock(&head
->mutex
);
2373 mutex_unlock(&head
->mutex
);
2375 btrfs_put_delayed_ref(ref
);
2379 node
= rb_next(node
);
2381 spin_unlock(&delayed_refs
->lock
);
2382 schedule_timeout(1);
2386 spin_unlock(&delayed_refs
->lock
);
2390 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
2391 struct btrfs_root
*root
,
2392 u64 bytenr
, u64 num_bytes
, u64 flags
,
2395 struct btrfs_delayed_extent_op
*extent_op
;
2398 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
2402 extent_op
->flags_to_set
= flags
;
2403 extent_op
->update_flags
= 1;
2404 extent_op
->update_key
= 0;
2405 extent_op
->is_data
= is_data
? 1 : 0;
2407 ret
= btrfs_add_delayed_extent_op(trans
, bytenr
, num_bytes
, extent_op
);
2413 static noinline
int check_delayed_ref(struct btrfs_trans_handle
*trans
,
2414 struct btrfs_root
*root
,
2415 struct btrfs_path
*path
,
2416 u64 objectid
, u64 offset
, u64 bytenr
)
2418 struct btrfs_delayed_ref_head
*head
;
2419 struct btrfs_delayed_ref_node
*ref
;
2420 struct btrfs_delayed_data_ref
*data_ref
;
2421 struct btrfs_delayed_ref_root
*delayed_refs
;
2422 struct rb_node
*node
;
2426 delayed_refs
= &trans
->transaction
->delayed_refs
;
2427 spin_lock(&delayed_refs
->lock
);
2428 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
2432 if (!mutex_trylock(&head
->mutex
)) {
2433 atomic_inc(&head
->node
.refs
);
2434 spin_unlock(&delayed_refs
->lock
);
2436 btrfs_release_path(path
);
2439 * Mutex was contended, block until it's released and let
2442 mutex_lock(&head
->mutex
);
2443 mutex_unlock(&head
->mutex
);
2444 btrfs_put_delayed_ref(&head
->node
);
2448 node
= rb_prev(&head
->node
.rb_node
);
2452 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2454 if (ref
->bytenr
!= bytenr
)
2458 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
)
2461 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
2463 node
= rb_prev(node
);
2465 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2466 if (ref
->bytenr
== bytenr
)
2470 if (data_ref
->root
!= root
->root_key
.objectid
||
2471 data_ref
->objectid
!= objectid
|| data_ref
->offset
!= offset
)
2476 mutex_unlock(&head
->mutex
);
2478 spin_unlock(&delayed_refs
->lock
);
2482 static noinline
int check_committed_ref(struct btrfs_trans_handle
*trans
,
2483 struct btrfs_root
*root
,
2484 struct btrfs_path
*path
,
2485 u64 objectid
, u64 offset
, u64 bytenr
)
2487 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2488 struct extent_buffer
*leaf
;
2489 struct btrfs_extent_data_ref
*ref
;
2490 struct btrfs_extent_inline_ref
*iref
;
2491 struct btrfs_extent_item
*ei
;
2492 struct btrfs_key key
;
2496 key
.objectid
= bytenr
;
2497 key
.offset
= (u64
)-1;
2498 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2500 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
2506 if (path
->slots
[0] == 0)
2510 leaf
= path
->nodes
[0];
2511 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2513 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
2517 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2518 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2519 if (item_size
< sizeof(*ei
)) {
2520 WARN_ON(item_size
!= sizeof(struct btrfs_extent_item_v0
));
2524 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2526 if (item_size
!= sizeof(*ei
) +
2527 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
2530 if (btrfs_extent_generation(leaf
, ei
) <=
2531 btrfs_root_last_snapshot(&root
->root_item
))
2534 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
2535 if (btrfs_extent_inline_ref_type(leaf
, iref
) !=
2536 BTRFS_EXTENT_DATA_REF_KEY
)
2539 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
2540 if (btrfs_extent_refs(leaf
, ei
) !=
2541 btrfs_extent_data_ref_count(leaf
, ref
) ||
2542 btrfs_extent_data_ref_root(leaf
, ref
) !=
2543 root
->root_key
.objectid
||
2544 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
2545 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
2553 int btrfs_cross_ref_exist(struct btrfs_trans_handle
*trans
,
2554 struct btrfs_root
*root
,
2555 u64 objectid
, u64 offset
, u64 bytenr
)
2557 struct btrfs_path
*path
;
2561 path
= btrfs_alloc_path();
2566 ret
= check_committed_ref(trans
, root
, path
, objectid
,
2568 if (ret
&& ret
!= -ENOENT
)
2571 ret2
= check_delayed_ref(trans
, root
, path
, objectid
,
2573 } while (ret2
== -EAGAIN
);
2575 if (ret2
&& ret2
!= -ENOENT
) {
2580 if (ret
!= -ENOENT
|| ret2
!= -ENOENT
)
2583 btrfs_free_path(path
);
2584 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
2589 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
2590 struct btrfs_root
*root
,
2591 struct extent_buffer
*buf
,
2592 int full_backref
, int inc
)
2599 struct btrfs_key key
;
2600 struct btrfs_file_extent_item
*fi
;
2604 int (*process_func
)(struct btrfs_trans_handle
*, struct btrfs_root
*,
2605 u64
, u64
, u64
, u64
, u64
, u64
);
2607 ref_root
= btrfs_header_owner(buf
);
2608 nritems
= btrfs_header_nritems(buf
);
2609 level
= btrfs_header_level(buf
);
2611 if (!root
->ref_cows
&& level
== 0)
2615 process_func
= btrfs_inc_extent_ref
;
2617 process_func
= btrfs_free_extent
;
2620 parent
= buf
->start
;
2624 for (i
= 0; i
< nritems
; i
++) {
2626 btrfs_item_key_to_cpu(buf
, &key
, i
);
2627 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
2629 fi
= btrfs_item_ptr(buf
, i
,
2630 struct btrfs_file_extent_item
);
2631 if (btrfs_file_extent_type(buf
, fi
) ==
2632 BTRFS_FILE_EXTENT_INLINE
)
2634 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
2638 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
2639 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
2640 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
2641 parent
, ref_root
, key
.objectid
,
2646 bytenr
= btrfs_node_blockptr(buf
, i
);
2647 num_bytes
= btrfs_level_size(root
, level
- 1);
2648 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
2649 parent
, ref_root
, level
- 1, 0);
2660 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2661 struct extent_buffer
*buf
, int full_backref
)
2663 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1);
2666 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2667 struct extent_buffer
*buf
, int full_backref
)
2669 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0);
2672 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
2673 struct btrfs_root
*root
,
2674 struct btrfs_path
*path
,
2675 struct btrfs_block_group_cache
*cache
)
2678 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2680 struct extent_buffer
*leaf
;
2682 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
2687 leaf
= path
->nodes
[0];
2688 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
2689 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
2690 btrfs_mark_buffer_dirty(leaf
);
2691 btrfs_release_path(path
);
2699 static struct btrfs_block_group_cache
*
2700 next_block_group(struct btrfs_root
*root
,
2701 struct btrfs_block_group_cache
*cache
)
2703 struct rb_node
*node
;
2704 spin_lock(&root
->fs_info
->block_group_cache_lock
);
2705 node
= rb_next(&cache
->cache_node
);
2706 btrfs_put_block_group(cache
);
2708 cache
= rb_entry(node
, struct btrfs_block_group_cache
,
2710 btrfs_get_block_group(cache
);
2713 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
2717 static int cache_save_setup(struct btrfs_block_group_cache
*block_group
,
2718 struct btrfs_trans_handle
*trans
,
2719 struct btrfs_path
*path
)
2721 struct btrfs_root
*root
= block_group
->fs_info
->tree_root
;
2722 struct inode
*inode
= NULL
;
2724 int dcs
= BTRFS_DC_ERROR
;
2730 * If this block group is smaller than 100 megs don't bother caching the
2733 if (block_group
->key
.offset
< (100 * 1024 * 1024)) {
2734 spin_lock(&block_group
->lock
);
2735 block_group
->disk_cache_state
= BTRFS_DC_WRITTEN
;
2736 spin_unlock(&block_group
->lock
);
2741 inode
= lookup_free_space_inode(root
, block_group
, path
);
2742 if (IS_ERR(inode
) && PTR_ERR(inode
) != -ENOENT
) {
2743 ret
= PTR_ERR(inode
);
2744 btrfs_release_path(path
);
2748 if (IS_ERR(inode
)) {
2752 if (block_group
->ro
)
2755 ret
= create_free_space_inode(root
, trans
, block_group
, path
);
2761 /* We've already setup this transaction, go ahead and exit */
2762 if (block_group
->cache_generation
== trans
->transid
&&
2763 i_size_read(inode
)) {
2764 dcs
= BTRFS_DC_SETUP
;
2769 * We want to set the generation to 0, that way if anything goes wrong
2770 * from here on out we know not to trust this cache when we load up next
2773 BTRFS_I(inode
)->generation
= 0;
2774 ret
= btrfs_update_inode(trans
, root
, inode
);
2777 if (i_size_read(inode
) > 0) {
2778 ret
= btrfs_truncate_free_space_cache(root
, trans
, path
,
2784 spin_lock(&block_group
->lock
);
2785 if (block_group
->cached
!= BTRFS_CACHE_FINISHED
) {
2786 /* We're not cached, don't bother trying to write stuff out */
2787 dcs
= BTRFS_DC_WRITTEN
;
2788 spin_unlock(&block_group
->lock
);
2791 spin_unlock(&block_group
->lock
);
2793 num_pages
= (int)div64_u64(block_group
->key
.offset
, 1024 * 1024 * 1024);
2798 * Just to make absolutely sure we have enough space, we're going to
2799 * preallocate 12 pages worth of space for each block group. In
2800 * practice we ought to use at most 8, but we need extra space so we can
2801 * add our header and have a terminator between the extents and the
2805 num_pages
*= PAGE_CACHE_SIZE
;
2807 ret
= btrfs_check_data_free_space(inode
, num_pages
);
2811 ret
= btrfs_prealloc_file_range_trans(inode
, trans
, 0, 0, num_pages
,
2812 num_pages
, num_pages
,
2815 dcs
= BTRFS_DC_SETUP
;
2816 btrfs_free_reserved_data_space(inode
, num_pages
);
2821 btrfs_release_path(path
);
2823 spin_lock(&block_group
->lock
);
2824 if (!ret
&& dcs
== BTRFS_DC_SETUP
)
2825 block_group
->cache_generation
= trans
->transid
;
2826 block_group
->disk_cache_state
= dcs
;
2827 spin_unlock(&block_group
->lock
);
2832 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
2833 struct btrfs_root
*root
)
2835 struct btrfs_block_group_cache
*cache
;
2837 struct btrfs_path
*path
;
2840 path
= btrfs_alloc_path();
2846 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
2848 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
)
2850 cache
= next_block_group(root
, cache
);
2858 err
= cache_save_setup(cache
, trans
, path
);
2859 last
= cache
->key
.objectid
+ cache
->key
.offset
;
2860 btrfs_put_block_group(cache
);
2865 err
= btrfs_run_delayed_refs(trans
, root
,
2870 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
2872 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
) {
2873 btrfs_put_block_group(cache
);
2879 cache
= next_block_group(root
, cache
);
2888 if (cache
->disk_cache_state
== BTRFS_DC_SETUP
)
2889 cache
->disk_cache_state
= BTRFS_DC_NEED_WRITE
;
2891 last
= cache
->key
.objectid
+ cache
->key
.offset
;
2893 err
= write_one_cache_group(trans
, root
, path
, cache
);
2895 btrfs_put_block_group(cache
);
2900 * I don't think this is needed since we're just marking our
2901 * preallocated extent as written, but just in case it can't
2905 err
= btrfs_run_delayed_refs(trans
, root
,
2910 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
2913 * Really this shouldn't happen, but it could if we
2914 * couldn't write the entire preallocated extent and
2915 * splitting the extent resulted in a new block.
2918 btrfs_put_block_group(cache
);
2921 if (cache
->disk_cache_state
== BTRFS_DC_NEED_WRITE
)
2923 cache
= next_block_group(root
, cache
);
2932 btrfs_write_out_cache(root
, trans
, cache
, path
);
2935 * If we didn't have an error then the cache state is still
2936 * NEED_WRITE, so we can set it to WRITTEN.
2938 if (cache
->disk_cache_state
== BTRFS_DC_NEED_WRITE
)
2939 cache
->disk_cache_state
= BTRFS_DC_WRITTEN
;
2940 last
= cache
->key
.objectid
+ cache
->key
.offset
;
2941 btrfs_put_block_group(cache
);
2944 btrfs_free_path(path
);
2948 int btrfs_extent_readonly(struct btrfs_root
*root
, u64 bytenr
)
2950 struct btrfs_block_group_cache
*block_group
;
2953 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
2954 if (!block_group
|| block_group
->ro
)
2957 btrfs_put_block_group(block_group
);
2961 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
2962 u64 total_bytes
, u64 bytes_used
,
2963 struct btrfs_space_info
**space_info
)
2965 struct btrfs_space_info
*found
;
2969 if (flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
2970 BTRFS_BLOCK_GROUP_RAID10
))
2975 found
= __find_space_info(info
, flags
);
2977 spin_lock(&found
->lock
);
2978 found
->total_bytes
+= total_bytes
;
2979 found
->disk_total
+= total_bytes
* factor
;
2980 found
->bytes_used
+= bytes_used
;
2981 found
->disk_used
+= bytes_used
* factor
;
2983 spin_unlock(&found
->lock
);
2984 *space_info
= found
;
2987 found
= kzalloc(sizeof(*found
), GFP_NOFS
);
2991 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
2992 INIT_LIST_HEAD(&found
->block_groups
[i
]);
2993 init_rwsem(&found
->groups_sem
);
2994 spin_lock_init(&found
->lock
);
2995 found
->flags
= flags
& BTRFS_BLOCK_GROUP_TYPE_MASK
;
2996 found
->total_bytes
= total_bytes
;
2997 found
->disk_total
= total_bytes
* factor
;
2998 found
->bytes_used
= bytes_used
;
2999 found
->disk_used
= bytes_used
* factor
;
3000 found
->bytes_pinned
= 0;
3001 found
->bytes_reserved
= 0;
3002 found
->bytes_readonly
= 0;
3003 found
->bytes_may_use
= 0;
3005 found
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3006 found
->chunk_alloc
= 0;
3008 init_waitqueue_head(&found
->wait
);
3009 *space_info
= found
;
3010 list_add_rcu(&found
->list
, &info
->space_info
);
3014 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
3016 u64 extra_flags
= flags
& BTRFS_BLOCK_GROUP_PROFILE_MASK
;
3018 /* chunk -> extended profile */
3019 if (extra_flags
== 0)
3020 extra_flags
= BTRFS_AVAIL_ALLOC_BIT_SINGLE
;
3022 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3023 fs_info
->avail_data_alloc_bits
|= extra_flags
;
3024 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3025 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
3026 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3027 fs_info
->avail_system_alloc_bits
|= extra_flags
;
3031 * @flags: available profiles in extended format (see ctree.h)
3033 * Returns reduced profile in chunk format.
3035 u64
btrfs_reduce_alloc_profile(struct btrfs_root
*root
, u64 flags
)
3038 * we add in the count of missing devices because we want
3039 * to make sure that any RAID levels on a degraded FS
3040 * continue to be honored.
3042 u64 num_devices
= root
->fs_info
->fs_devices
->rw_devices
+
3043 root
->fs_info
->fs_devices
->missing_devices
;
3045 if (num_devices
== 1)
3046 flags
&= ~(BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID0
);
3047 if (num_devices
< 4)
3048 flags
&= ~BTRFS_BLOCK_GROUP_RAID10
;
3050 if ((flags
& BTRFS_BLOCK_GROUP_DUP
) &&
3051 (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
3052 BTRFS_BLOCK_GROUP_RAID10
))) {
3053 flags
&= ~BTRFS_BLOCK_GROUP_DUP
;
3056 if ((flags
& BTRFS_BLOCK_GROUP_RAID1
) &&
3057 (flags
& BTRFS_BLOCK_GROUP_RAID10
)) {
3058 flags
&= ~BTRFS_BLOCK_GROUP_RAID1
;
3061 if ((flags
& BTRFS_BLOCK_GROUP_RAID0
) &&
3062 ((flags
& BTRFS_BLOCK_GROUP_RAID1
) |
3063 (flags
& BTRFS_BLOCK_GROUP_RAID10
) |
3064 (flags
& BTRFS_BLOCK_GROUP_DUP
))) {
3065 flags
&= ~BTRFS_BLOCK_GROUP_RAID0
;
3068 /* extended -> chunk profile */
3069 flags
&= ~BTRFS_AVAIL_ALLOC_BIT_SINGLE
;
3073 static u64
get_alloc_profile(struct btrfs_root
*root
, u64 flags
)
3075 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3076 flags
|= root
->fs_info
->avail_data_alloc_bits
;
3077 else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3078 flags
|= root
->fs_info
->avail_system_alloc_bits
;
3079 else if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3080 flags
|= root
->fs_info
->avail_metadata_alloc_bits
;
3082 return btrfs_reduce_alloc_profile(root
, flags
);
3085 u64
btrfs_get_alloc_profile(struct btrfs_root
*root
, int data
)
3090 flags
= BTRFS_BLOCK_GROUP_DATA
;
3091 else if (root
== root
->fs_info
->chunk_root
)
3092 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
3094 flags
= BTRFS_BLOCK_GROUP_METADATA
;
3096 return get_alloc_profile(root
, flags
);
3099 void btrfs_set_inode_space_info(struct btrfs_root
*root
, struct inode
*inode
)
3101 BTRFS_I(inode
)->space_info
= __find_space_info(root
->fs_info
,
3102 BTRFS_BLOCK_GROUP_DATA
);
3106 * This will check the space that the inode allocates from to make sure we have
3107 * enough space for bytes.
3109 int btrfs_check_data_free_space(struct inode
*inode
, u64 bytes
)
3111 struct btrfs_space_info
*data_sinfo
;
3112 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3114 int ret
= 0, committed
= 0, alloc_chunk
= 1;
3116 /* make sure bytes are sectorsize aligned */
3117 bytes
= (bytes
+ root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
3119 if (root
== root
->fs_info
->tree_root
||
3120 BTRFS_I(inode
)->location
.objectid
== BTRFS_FREE_INO_OBJECTID
) {
3125 data_sinfo
= BTRFS_I(inode
)->space_info
;
3130 /* make sure we have enough space to handle the data first */
3131 spin_lock(&data_sinfo
->lock
);
3132 used
= data_sinfo
->bytes_used
+ data_sinfo
->bytes_reserved
+
3133 data_sinfo
->bytes_pinned
+ data_sinfo
->bytes_readonly
+
3134 data_sinfo
->bytes_may_use
;
3136 if (used
+ bytes
> data_sinfo
->total_bytes
) {
3137 struct btrfs_trans_handle
*trans
;
3140 * if we don't have enough free bytes in this space then we need
3141 * to alloc a new chunk.
3143 if (!data_sinfo
->full
&& alloc_chunk
) {
3146 data_sinfo
->force_alloc
= CHUNK_ALLOC_FORCE
;
3147 spin_unlock(&data_sinfo
->lock
);
3149 alloc_target
= btrfs_get_alloc_profile(root
, 1);
3150 trans
= btrfs_join_transaction(root
);
3152 return PTR_ERR(trans
);
3154 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
3155 bytes
+ 2 * 1024 * 1024,
3157 CHUNK_ALLOC_NO_FORCE
);
3158 btrfs_end_transaction(trans
, root
);
3167 btrfs_set_inode_space_info(root
, inode
);
3168 data_sinfo
= BTRFS_I(inode
)->space_info
;
3174 * If we have less pinned bytes than we want to allocate then
3175 * don't bother committing the transaction, it won't help us.
3177 if (data_sinfo
->bytes_pinned
< bytes
)
3179 spin_unlock(&data_sinfo
->lock
);
3181 /* commit the current transaction and try again */
3184 !atomic_read(&root
->fs_info
->open_ioctl_trans
)) {
3186 trans
= btrfs_join_transaction(root
);
3188 return PTR_ERR(trans
);
3189 ret
= btrfs_commit_transaction(trans
, root
);
3197 data_sinfo
->bytes_may_use
+= bytes
;
3198 spin_unlock(&data_sinfo
->lock
);
3204 * Called if we need to clear a data reservation for this inode.
3206 void btrfs_free_reserved_data_space(struct inode
*inode
, u64 bytes
)
3208 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3209 struct btrfs_space_info
*data_sinfo
;
3211 /* make sure bytes are sectorsize aligned */
3212 bytes
= (bytes
+ root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
3214 data_sinfo
= BTRFS_I(inode
)->space_info
;
3215 spin_lock(&data_sinfo
->lock
);
3216 data_sinfo
->bytes_may_use
-= bytes
;
3217 spin_unlock(&data_sinfo
->lock
);
3220 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
3222 struct list_head
*head
= &info
->space_info
;
3223 struct btrfs_space_info
*found
;
3226 list_for_each_entry_rcu(found
, head
, list
) {
3227 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
3228 found
->force_alloc
= CHUNK_ALLOC_FORCE
;
3233 static int should_alloc_chunk(struct btrfs_root
*root
,
3234 struct btrfs_space_info
*sinfo
, u64 alloc_bytes
,
3237 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
3238 u64 num_bytes
= sinfo
->total_bytes
- sinfo
->bytes_readonly
;
3239 u64 num_allocated
= sinfo
->bytes_used
+ sinfo
->bytes_reserved
;
3242 if (force
== CHUNK_ALLOC_FORCE
)
3246 * We need to take into account the global rsv because for all intents
3247 * and purposes it's used space. Don't worry about locking the
3248 * global_rsv, it doesn't change except when the transaction commits.
3250 num_allocated
+= global_rsv
->size
;
3253 * in limited mode, we want to have some free space up to
3254 * about 1% of the FS size.
3256 if (force
== CHUNK_ALLOC_LIMITED
) {
3257 thresh
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
3258 thresh
= max_t(u64
, 64 * 1024 * 1024,
3259 div_factor_fine(thresh
, 1));
3261 if (num_bytes
- num_allocated
< thresh
)
3266 * we have two similar checks here, one based on percentage
3267 * and once based on a hard number of 256MB. The idea
3268 * is that if we have a good amount of free
3269 * room, don't allocate a chunk. A good mount is
3270 * less than 80% utilized of the chunks we have allocated,
3271 * or more than 256MB free
3273 if (num_allocated
+ alloc_bytes
+ 256 * 1024 * 1024 < num_bytes
)
3276 if (num_allocated
+ alloc_bytes
< div_factor(num_bytes
, 8))
3279 thresh
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
3281 /* 256MB or 5% of the FS */
3282 thresh
= max_t(u64
, 256 * 1024 * 1024, div_factor_fine(thresh
, 5));
3284 if (num_bytes
> thresh
&& sinfo
->bytes_used
< div_factor(num_bytes
, 3))
3289 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
3290 struct btrfs_root
*extent_root
, u64 alloc_bytes
,
3291 u64 flags
, int force
)
3293 struct btrfs_space_info
*space_info
;
3294 struct btrfs_fs_info
*fs_info
= extent_root
->fs_info
;
3295 int wait_for_alloc
= 0;
3298 BUG_ON(!profile_is_valid(flags
, 0));
3300 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
3302 ret
= update_space_info(extent_root
->fs_info
, flags
,
3306 BUG_ON(!space_info
);
3309 spin_lock(&space_info
->lock
);
3310 if (space_info
->force_alloc
)
3311 force
= space_info
->force_alloc
;
3312 if (space_info
->full
) {
3313 spin_unlock(&space_info
->lock
);
3317 if (!should_alloc_chunk(extent_root
, space_info
, alloc_bytes
, force
)) {
3318 spin_unlock(&space_info
->lock
);
3320 } else if (space_info
->chunk_alloc
) {
3323 space_info
->chunk_alloc
= 1;
3326 spin_unlock(&space_info
->lock
);
3328 mutex_lock(&fs_info
->chunk_mutex
);
3331 * The chunk_mutex is held throughout the entirety of a chunk
3332 * allocation, so once we've acquired the chunk_mutex we know that the
3333 * other guy is done and we need to recheck and see if we should
3336 if (wait_for_alloc
) {
3337 mutex_unlock(&fs_info
->chunk_mutex
);
3343 * If we have mixed data/metadata chunks we want to make sure we keep
3344 * allocating mixed chunks instead of individual chunks.
3346 if (btrfs_mixed_space_info(space_info
))
3347 flags
|= (BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
);
3350 * if we're doing a data chunk, go ahead and make sure that
3351 * we keep a reasonable number of metadata chunks allocated in the
3354 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
3355 fs_info
->data_chunk_allocations
++;
3356 if (!(fs_info
->data_chunk_allocations
%
3357 fs_info
->metadata_ratio
))
3358 force_metadata_allocation(fs_info
);
3361 ret
= btrfs_alloc_chunk(trans
, extent_root
, flags
);
3362 if (ret
< 0 && ret
!= -ENOSPC
)
3365 spin_lock(&space_info
->lock
);
3367 space_info
->full
= 1;
3371 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3372 space_info
->chunk_alloc
= 0;
3373 spin_unlock(&space_info
->lock
);
3375 mutex_unlock(&extent_root
->fs_info
->chunk_mutex
);
3380 * shrink metadata reservation for delalloc
3382 static int shrink_delalloc(struct btrfs_root
*root
, u64 to_reclaim
,
3385 struct btrfs_block_rsv
*block_rsv
;
3386 struct btrfs_space_info
*space_info
;
3387 struct btrfs_trans_handle
*trans
;
3392 unsigned long nr_pages
= (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT
;
3394 unsigned long progress
;
3396 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
3397 block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
3398 space_info
= block_rsv
->space_info
;
3401 reserved
= space_info
->bytes_may_use
;
3402 progress
= space_info
->reservation_progress
;
3408 if (root
->fs_info
->delalloc_bytes
== 0) {
3411 btrfs_wait_ordered_extents(root
, 0, 0);
3415 max_reclaim
= min(reserved
, to_reclaim
);
3416 nr_pages
= max_t(unsigned long, nr_pages
,
3417 max_reclaim
>> PAGE_CACHE_SHIFT
);
3418 while (loops
< 1024) {
3419 /* have the flusher threads jump in and do some IO */
3421 nr_pages
= min_t(unsigned long, nr_pages
,
3422 root
->fs_info
->delalloc_bytes
>> PAGE_CACHE_SHIFT
);
3423 writeback_inodes_sb_nr_if_idle(root
->fs_info
->sb
, nr_pages
);
3425 spin_lock(&space_info
->lock
);
3426 if (reserved
> space_info
->bytes_may_use
)
3427 reclaimed
+= reserved
- space_info
->bytes_may_use
;
3428 reserved
= space_info
->bytes_may_use
;
3429 spin_unlock(&space_info
->lock
);
3433 if (reserved
== 0 || reclaimed
>= max_reclaim
)
3436 if (trans
&& trans
->transaction
->blocked
)
3439 if (wait_ordered
&& !trans
) {
3440 btrfs_wait_ordered_extents(root
, 0, 0);
3442 time_left
= schedule_timeout_interruptible(1);
3444 /* We were interrupted, exit */
3449 /* we've kicked the IO a few times, if anything has been freed,
3450 * exit. There is no sense in looping here for a long time
3451 * when we really need to commit the transaction, or there are
3452 * just too many writers without enough free space
3457 if (progress
!= space_info
->reservation_progress
)
3463 return reclaimed
>= to_reclaim
;
3467 * maybe_commit_transaction - possibly commit the transaction if its ok to
3468 * @root - the root we're allocating for
3469 * @bytes - the number of bytes we want to reserve
3470 * @force - force the commit
3472 * This will check to make sure that committing the transaction will actually
3473 * get us somewhere and then commit the transaction if it does. Otherwise it
3474 * will return -ENOSPC.
3476 static int may_commit_transaction(struct btrfs_root
*root
,
3477 struct btrfs_space_info
*space_info
,
3478 u64 bytes
, int force
)
3480 struct btrfs_block_rsv
*delayed_rsv
= &root
->fs_info
->delayed_block_rsv
;
3481 struct btrfs_trans_handle
*trans
;
3483 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
3490 /* See if there is enough pinned space to make this reservation */
3491 spin_lock(&space_info
->lock
);
3492 if (space_info
->bytes_pinned
>= bytes
) {
3493 spin_unlock(&space_info
->lock
);
3496 spin_unlock(&space_info
->lock
);
3499 * See if there is some space in the delayed insertion reservation for
3502 if (space_info
!= delayed_rsv
->space_info
)
3505 spin_lock(&delayed_rsv
->lock
);
3506 if (delayed_rsv
->size
< bytes
) {
3507 spin_unlock(&delayed_rsv
->lock
);
3510 spin_unlock(&delayed_rsv
->lock
);
3513 trans
= btrfs_join_transaction(root
);
3517 return btrfs_commit_transaction(trans
, root
);
3521 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3522 * @root - the root we're allocating for
3523 * @block_rsv - the block_rsv we're allocating for
3524 * @orig_bytes - the number of bytes we want
3525 * @flush - wether or not we can flush to make our reservation
3527 * This will reserve orgi_bytes number of bytes from the space info associated
3528 * with the block_rsv. If there is not enough space it will make an attempt to
3529 * flush out space to make room. It will do this by flushing delalloc if
3530 * possible or committing the transaction. If flush is 0 then no attempts to
3531 * regain reservations will be made and this will fail if there is not enough
3534 static int reserve_metadata_bytes(struct btrfs_root
*root
,
3535 struct btrfs_block_rsv
*block_rsv
,
3536 u64 orig_bytes
, int flush
)
3538 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
3540 u64 num_bytes
= orig_bytes
;
3543 bool committed
= false;
3544 bool flushing
= false;
3545 bool wait_ordered
= false;
3549 spin_lock(&space_info
->lock
);
3551 * We only want to wait if somebody other than us is flushing and we are
3552 * actually alloed to flush.
3554 while (flush
&& !flushing
&& space_info
->flush
) {
3555 spin_unlock(&space_info
->lock
);
3557 * If we have a trans handle we can't wait because the flusher
3558 * may have to commit the transaction, which would mean we would
3559 * deadlock since we are waiting for the flusher to finish, but
3560 * hold the current transaction open.
3562 if (current
->journal_info
)
3564 ret
= wait_event_interruptible(space_info
->wait
,
3565 !space_info
->flush
);
3566 /* Must have been interrupted, return */
3570 spin_lock(&space_info
->lock
);
3574 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
3575 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
3576 space_info
->bytes_may_use
;
3579 * The idea here is that we've not already over-reserved the block group
3580 * then we can go ahead and save our reservation first and then start
3581 * flushing if we need to. Otherwise if we've already overcommitted
3582 * lets start flushing stuff first and then come back and try to make
3585 if (used
<= space_info
->total_bytes
) {
3586 if (used
+ orig_bytes
<= space_info
->total_bytes
) {
3587 space_info
->bytes_may_use
+= orig_bytes
;
3591 * Ok set num_bytes to orig_bytes since we aren't
3592 * overocmmitted, this way we only try and reclaim what
3595 num_bytes
= orig_bytes
;
3599 * Ok we're over committed, set num_bytes to the overcommitted
3600 * amount plus the amount of bytes that we need for this
3603 wait_ordered
= true;
3604 num_bytes
= used
- space_info
->total_bytes
+
3605 (orig_bytes
* (retries
+ 1));
3609 u64 profile
= btrfs_get_alloc_profile(root
, 0);
3613 * If we have a lot of space that's pinned, don't bother doing
3614 * the overcommit dance yet and just commit the transaction.
3616 avail
= (space_info
->total_bytes
- space_info
->bytes_used
) * 8;
3618 if (space_info
->bytes_pinned
>= avail
&& flush
&& !committed
) {
3619 space_info
->flush
= 1;
3621 spin_unlock(&space_info
->lock
);
3622 ret
= may_commit_transaction(root
, space_info
,
3630 spin_lock(&root
->fs_info
->free_chunk_lock
);
3631 avail
= root
->fs_info
->free_chunk_space
;
3634 * If we have dup, raid1 or raid10 then only half of the free
3635 * space is actually useable.
3637 if (profile
& (BTRFS_BLOCK_GROUP_DUP
|
3638 BTRFS_BLOCK_GROUP_RAID1
|
3639 BTRFS_BLOCK_GROUP_RAID10
))
3643 * If we aren't flushing don't let us overcommit too much, say
3644 * 1/8th of the space. If we can flush, let it overcommit up to
3651 spin_unlock(&root
->fs_info
->free_chunk_lock
);
3653 if (used
+ num_bytes
< space_info
->total_bytes
+ avail
) {
3654 space_info
->bytes_may_use
+= orig_bytes
;
3657 wait_ordered
= true;
3662 * Couldn't make our reservation, save our place so while we're trying
3663 * to reclaim space we can actually use it instead of somebody else
3664 * stealing it from us.
3668 space_info
->flush
= 1;
3671 spin_unlock(&space_info
->lock
);
3677 * We do synchronous shrinking since we don't actually unreserve
3678 * metadata until after the IO is completed.
3680 ret
= shrink_delalloc(root
, num_bytes
, wait_ordered
);
3687 * So if we were overcommitted it's possible that somebody else flushed
3688 * out enough space and we simply didn't have enough space to reclaim,
3689 * so go back around and try again.
3692 wait_ordered
= true;
3701 ret
= may_commit_transaction(root
, space_info
, orig_bytes
, 0);
3709 spin_lock(&space_info
->lock
);
3710 space_info
->flush
= 0;
3711 wake_up_all(&space_info
->wait
);
3712 spin_unlock(&space_info
->lock
);
3717 static struct btrfs_block_rsv
*get_block_rsv(struct btrfs_trans_handle
*trans
,
3718 struct btrfs_root
*root
)
3720 struct btrfs_block_rsv
*block_rsv
= NULL
;
3722 if (root
->ref_cows
|| root
== root
->fs_info
->csum_root
)
3723 block_rsv
= trans
->block_rsv
;
3726 block_rsv
= root
->block_rsv
;
3729 block_rsv
= &root
->fs_info
->empty_block_rsv
;
3734 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
3738 spin_lock(&block_rsv
->lock
);
3739 if (block_rsv
->reserved
>= num_bytes
) {
3740 block_rsv
->reserved
-= num_bytes
;
3741 if (block_rsv
->reserved
< block_rsv
->size
)
3742 block_rsv
->full
= 0;
3745 spin_unlock(&block_rsv
->lock
);
3749 static void block_rsv_add_bytes(struct btrfs_block_rsv
*block_rsv
,
3750 u64 num_bytes
, int update_size
)
3752 spin_lock(&block_rsv
->lock
);
3753 block_rsv
->reserved
+= num_bytes
;
3755 block_rsv
->size
+= num_bytes
;
3756 else if (block_rsv
->reserved
>= block_rsv
->size
)
3757 block_rsv
->full
= 1;
3758 spin_unlock(&block_rsv
->lock
);
3761 static void block_rsv_release_bytes(struct btrfs_block_rsv
*block_rsv
,
3762 struct btrfs_block_rsv
*dest
, u64 num_bytes
)
3764 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
3766 spin_lock(&block_rsv
->lock
);
3767 if (num_bytes
== (u64
)-1)
3768 num_bytes
= block_rsv
->size
;
3769 block_rsv
->size
-= num_bytes
;
3770 if (block_rsv
->reserved
>= block_rsv
->size
) {
3771 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
3772 block_rsv
->reserved
= block_rsv
->size
;
3773 block_rsv
->full
= 1;
3777 spin_unlock(&block_rsv
->lock
);
3779 if (num_bytes
> 0) {
3781 spin_lock(&dest
->lock
);
3785 bytes_to_add
= dest
->size
- dest
->reserved
;
3786 bytes_to_add
= min(num_bytes
, bytes_to_add
);
3787 dest
->reserved
+= bytes_to_add
;
3788 if (dest
->reserved
>= dest
->size
)
3790 num_bytes
-= bytes_to_add
;
3792 spin_unlock(&dest
->lock
);
3795 spin_lock(&space_info
->lock
);
3796 space_info
->bytes_may_use
-= num_bytes
;
3797 space_info
->reservation_progress
++;
3798 spin_unlock(&space_info
->lock
);
3803 static int block_rsv_migrate_bytes(struct btrfs_block_rsv
*src
,
3804 struct btrfs_block_rsv
*dst
, u64 num_bytes
)
3808 ret
= block_rsv_use_bytes(src
, num_bytes
);
3812 block_rsv_add_bytes(dst
, num_bytes
, 1);
3816 void btrfs_init_block_rsv(struct btrfs_block_rsv
*rsv
)
3818 memset(rsv
, 0, sizeof(*rsv
));
3819 spin_lock_init(&rsv
->lock
);
3822 struct btrfs_block_rsv
*btrfs_alloc_block_rsv(struct btrfs_root
*root
)
3824 struct btrfs_block_rsv
*block_rsv
;
3825 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3827 block_rsv
= kmalloc(sizeof(*block_rsv
), GFP_NOFS
);
3831 btrfs_init_block_rsv(block_rsv
);
3832 block_rsv
->space_info
= __find_space_info(fs_info
,
3833 BTRFS_BLOCK_GROUP_METADATA
);
3837 void btrfs_free_block_rsv(struct btrfs_root
*root
,
3838 struct btrfs_block_rsv
*rsv
)
3840 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
3844 static inline int __block_rsv_add(struct btrfs_root
*root
,
3845 struct btrfs_block_rsv
*block_rsv
,
3846 u64 num_bytes
, int flush
)
3853 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
3855 block_rsv_add_bytes(block_rsv
, num_bytes
, 1);
3862 int btrfs_block_rsv_add(struct btrfs_root
*root
,
3863 struct btrfs_block_rsv
*block_rsv
,
3866 return __block_rsv_add(root
, block_rsv
, num_bytes
, 1);
3869 int btrfs_block_rsv_add_noflush(struct btrfs_root
*root
,
3870 struct btrfs_block_rsv
*block_rsv
,
3873 return __block_rsv_add(root
, block_rsv
, num_bytes
, 0);
3876 int btrfs_block_rsv_check(struct btrfs_root
*root
,
3877 struct btrfs_block_rsv
*block_rsv
, int min_factor
)
3885 spin_lock(&block_rsv
->lock
);
3886 num_bytes
= div_factor(block_rsv
->size
, min_factor
);
3887 if (block_rsv
->reserved
>= num_bytes
)
3889 spin_unlock(&block_rsv
->lock
);
3894 static inline int __btrfs_block_rsv_refill(struct btrfs_root
*root
,
3895 struct btrfs_block_rsv
*block_rsv
,
3896 u64 min_reserved
, int flush
)
3904 spin_lock(&block_rsv
->lock
);
3905 num_bytes
= min_reserved
;
3906 if (block_rsv
->reserved
>= num_bytes
)
3909 num_bytes
-= block_rsv
->reserved
;
3910 spin_unlock(&block_rsv
->lock
);
3915 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
3917 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
3924 int btrfs_block_rsv_refill(struct btrfs_root
*root
,
3925 struct btrfs_block_rsv
*block_rsv
,
3928 return __btrfs_block_rsv_refill(root
, block_rsv
, min_reserved
, 1);
3931 int btrfs_block_rsv_refill_noflush(struct btrfs_root
*root
,
3932 struct btrfs_block_rsv
*block_rsv
,
3935 return __btrfs_block_rsv_refill(root
, block_rsv
, min_reserved
, 0);
3938 int btrfs_block_rsv_migrate(struct btrfs_block_rsv
*src_rsv
,
3939 struct btrfs_block_rsv
*dst_rsv
,
3942 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
3945 void btrfs_block_rsv_release(struct btrfs_root
*root
,
3946 struct btrfs_block_rsv
*block_rsv
,
3949 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
3950 if (global_rsv
->full
|| global_rsv
== block_rsv
||
3951 block_rsv
->space_info
!= global_rsv
->space_info
)
3953 block_rsv_release_bytes(block_rsv
, global_rsv
, num_bytes
);
3957 * helper to calculate size of global block reservation.
3958 * the desired value is sum of space used by extent tree,
3959 * checksum tree and root tree
3961 static u64
calc_global_metadata_size(struct btrfs_fs_info
*fs_info
)
3963 struct btrfs_space_info
*sinfo
;
3967 int csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
3969 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_DATA
);
3970 spin_lock(&sinfo
->lock
);
3971 data_used
= sinfo
->bytes_used
;
3972 spin_unlock(&sinfo
->lock
);
3974 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
3975 spin_lock(&sinfo
->lock
);
3976 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_DATA
)
3978 meta_used
= sinfo
->bytes_used
;
3979 spin_unlock(&sinfo
->lock
);
3981 num_bytes
= (data_used
>> fs_info
->sb
->s_blocksize_bits
) *
3983 num_bytes
+= div64_u64(data_used
+ meta_used
, 50);
3985 if (num_bytes
* 3 > meta_used
)
3986 num_bytes
= div64_u64(meta_used
, 3);
3988 return ALIGN(num_bytes
, fs_info
->extent_root
->leafsize
<< 10);
3991 static void update_global_block_rsv(struct btrfs_fs_info
*fs_info
)
3993 struct btrfs_block_rsv
*block_rsv
= &fs_info
->global_block_rsv
;
3994 struct btrfs_space_info
*sinfo
= block_rsv
->space_info
;
3997 num_bytes
= calc_global_metadata_size(fs_info
);
3999 spin_lock(&block_rsv
->lock
);
4000 spin_lock(&sinfo
->lock
);
4002 block_rsv
->size
= num_bytes
;
4004 num_bytes
= sinfo
->bytes_used
+ sinfo
->bytes_pinned
+
4005 sinfo
->bytes_reserved
+ sinfo
->bytes_readonly
+
4006 sinfo
->bytes_may_use
;
4008 if (sinfo
->total_bytes
> num_bytes
) {
4009 num_bytes
= sinfo
->total_bytes
- num_bytes
;
4010 block_rsv
->reserved
+= num_bytes
;
4011 sinfo
->bytes_may_use
+= num_bytes
;
4014 if (block_rsv
->reserved
>= block_rsv
->size
) {
4015 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
4016 sinfo
->bytes_may_use
-= num_bytes
;
4017 sinfo
->reservation_progress
++;
4018 block_rsv
->reserved
= block_rsv
->size
;
4019 block_rsv
->full
= 1;
4022 spin_unlock(&sinfo
->lock
);
4023 spin_unlock(&block_rsv
->lock
);
4026 static void init_global_block_rsv(struct btrfs_fs_info
*fs_info
)
4028 struct btrfs_space_info
*space_info
;
4030 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
4031 fs_info
->chunk_block_rsv
.space_info
= space_info
;
4033 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4034 fs_info
->global_block_rsv
.space_info
= space_info
;
4035 fs_info
->delalloc_block_rsv
.space_info
= space_info
;
4036 fs_info
->trans_block_rsv
.space_info
= space_info
;
4037 fs_info
->empty_block_rsv
.space_info
= space_info
;
4038 fs_info
->delayed_block_rsv
.space_info
= space_info
;
4040 fs_info
->extent_root
->block_rsv
= &fs_info
->global_block_rsv
;
4041 fs_info
->csum_root
->block_rsv
= &fs_info
->global_block_rsv
;
4042 fs_info
->dev_root
->block_rsv
= &fs_info
->global_block_rsv
;
4043 fs_info
->tree_root
->block_rsv
= &fs_info
->global_block_rsv
;
4044 fs_info
->chunk_root
->block_rsv
= &fs_info
->chunk_block_rsv
;
4046 update_global_block_rsv(fs_info
);
4049 static void release_global_block_rsv(struct btrfs_fs_info
*fs_info
)
4051 block_rsv_release_bytes(&fs_info
->global_block_rsv
, NULL
, (u64
)-1);
4052 WARN_ON(fs_info
->delalloc_block_rsv
.size
> 0);
4053 WARN_ON(fs_info
->delalloc_block_rsv
.reserved
> 0);
4054 WARN_ON(fs_info
->trans_block_rsv
.size
> 0);
4055 WARN_ON(fs_info
->trans_block_rsv
.reserved
> 0);
4056 WARN_ON(fs_info
->chunk_block_rsv
.size
> 0);
4057 WARN_ON(fs_info
->chunk_block_rsv
.reserved
> 0);
4058 WARN_ON(fs_info
->delayed_block_rsv
.size
> 0);
4059 WARN_ON(fs_info
->delayed_block_rsv
.reserved
> 0);
4062 void btrfs_trans_release_metadata(struct btrfs_trans_handle
*trans
,
4063 struct btrfs_root
*root
)
4065 if (!trans
->bytes_reserved
)
4068 btrfs_block_rsv_release(root
, trans
->block_rsv
, trans
->bytes_reserved
);
4069 trans
->bytes_reserved
= 0;
4072 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle
*trans
,
4073 struct inode
*inode
)
4075 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4076 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
4077 struct btrfs_block_rsv
*dst_rsv
= root
->orphan_block_rsv
;
4080 * We need to hold space in order to delete our orphan item once we've
4081 * added it, so this takes the reservation so we can release it later
4082 * when we are truly done with the orphan item.
4084 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
4085 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
4088 void btrfs_orphan_release_metadata(struct inode
*inode
)
4090 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4091 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
4092 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
, num_bytes
);
4095 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle
*trans
,
4096 struct btrfs_pending_snapshot
*pending
)
4098 struct btrfs_root
*root
= pending
->root
;
4099 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
4100 struct btrfs_block_rsv
*dst_rsv
= &pending
->block_rsv
;
4102 * two for root back/forward refs, two for directory entries
4103 * and one for root of the snapshot.
4105 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 5);
4106 dst_rsv
->space_info
= src_rsv
->space_info
;
4107 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
4111 * drop_outstanding_extent - drop an outstanding extent
4112 * @inode: the inode we're dropping the extent for
4114 * This is called when we are freeing up an outstanding extent, either called
4115 * after an error or after an extent is written. This will return the number of
4116 * reserved extents that need to be freed. This must be called with
4117 * BTRFS_I(inode)->lock held.
4119 static unsigned drop_outstanding_extent(struct inode
*inode
)
4121 unsigned drop_inode_space
= 0;
4122 unsigned dropped_extents
= 0;
4124 BUG_ON(!BTRFS_I(inode
)->outstanding_extents
);
4125 BTRFS_I(inode
)->outstanding_extents
--;
4127 if (BTRFS_I(inode
)->outstanding_extents
== 0 &&
4128 BTRFS_I(inode
)->delalloc_meta_reserved
) {
4129 drop_inode_space
= 1;
4130 BTRFS_I(inode
)->delalloc_meta_reserved
= 0;
4134 * If we have more or the same amount of outsanding extents than we have
4135 * reserved then we need to leave the reserved extents count alone.
4137 if (BTRFS_I(inode
)->outstanding_extents
>=
4138 BTRFS_I(inode
)->reserved_extents
)
4139 return drop_inode_space
;
4141 dropped_extents
= BTRFS_I(inode
)->reserved_extents
-
4142 BTRFS_I(inode
)->outstanding_extents
;
4143 BTRFS_I(inode
)->reserved_extents
-= dropped_extents
;
4144 return dropped_extents
+ drop_inode_space
;
4148 * calc_csum_metadata_size - return the amount of metada space that must be
4149 * reserved/free'd for the given bytes.
4150 * @inode: the inode we're manipulating
4151 * @num_bytes: the number of bytes in question
4152 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4154 * This adjusts the number of csum_bytes in the inode and then returns the
4155 * correct amount of metadata that must either be reserved or freed. We
4156 * calculate how many checksums we can fit into one leaf and then divide the
4157 * number of bytes that will need to be checksumed by this value to figure out
4158 * how many checksums will be required. If we are adding bytes then the number
4159 * may go up and we will return the number of additional bytes that must be
4160 * reserved. If it is going down we will return the number of bytes that must
4163 * This must be called with BTRFS_I(inode)->lock held.
4165 static u64
calc_csum_metadata_size(struct inode
*inode
, u64 num_bytes
,
4168 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4170 int num_csums_per_leaf
;
4174 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
&&
4175 BTRFS_I(inode
)->csum_bytes
== 0)
4178 old_csums
= (int)div64_u64(BTRFS_I(inode
)->csum_bytes
, root
->sectorsize
);
4180 BTRFS_I(inode
)->csum_bytes
+= num_bytes
;
4182 BTRFS_I(inode
)->csum_bytes
-= num_bytes
;
4183 csum_size
= BTRFS_LEAF_DATA_SIZE(root
) - sizeof(struct btrfs_item
);
4184 num_csums_per_leaf
= (int)div64_u64(csum_size
,
4185 sizeof(struct btrfs_csum_item
) +
4186 sizeof(struct btrfs_disk_key
));
4187 num_csums
= (int)div64_u64(BTRFS_I(inode
)->csum_bytes
, root
->sectorsize
);
4188 num_csums
= num_csums
+ num_csums_per_leaf
- 1;
4189 num_csums
= num_csums
/ num_csums_per_leaf
;
4191 old_csums
= old_csums
+ num_csums_per_leaf
- 1;
4192 old_csums
= old_csums
/ num_csums_per_leaf
;
4194 /* No change, no need to reserve more */
4195 if (old_csums
== num_csums
)
4199 return btrfs_calc_trans_metadata_size(root
,
4200 num_csums
- old_csums
);
4202 return btrfs_calc_trans_metadata_size(root
, old_csums
- num_csums
);
4205 int btrfs_delalloc_reserve_metadata(struct inode
*inode
, u64 num_bytes
)
4207 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4208 struct btrfs_block_rsv
*block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
4211 unsigned nr_extents
= 0;
4212 int extra_reserve
= 0;
4216 /* Need to be holding the i_mutex here if we aren't free space cache */
4217 if (btrfs_is_free_space_inode(root
, inode
))
4220 WARN_ON(!mutex_is_locked(&inode
->i_mutex
));
4222 if (flush
&& btrfs_transaction_in_commit(root
->fs_info
))
4223 schedule_timeout(1);
4225 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
4227 spin_lock(&BTRFS_I(inode
)->lock
);
4228 BTRFS_I(inode
)->outstanding_extents
++;
4230 if (BTRFS_I(inode
)->outstanding_extents
>
4231 BTRFS_I(inode
)->reserved_extents
)
4232 nr_extents
= BTRFS_I(inode
)->outstanding_extents
-
4233 BTRFS_I(inode
)->reserved_extents
;
4236 * Add an item to reserve for updating the inode when we complete the
4239 if (!BTRFS_I(inode
)->delalloc_meta_reserved
) {
4244 to_reserve
= btrfs_calc_trans_metadata_size(root
, nr_extents
);
4245 to_reserve
+= calc_csum_metadata_size(inode
, num_bytes
, 1);
4246 csum_bytes
= BTRFS_I(inode
)->csum_bytes
;
4247 spin_unlock(&BTRFS_I(inode
)->lock
);
4249 ret
= reserve_metadata_bytes(root
, block_rsv
, to_reserve
, flush
);
4254 spin_lock(&BTRFS_I(inode
)->lock
);
4255 dropped
= drop_outstanding_extent(inode
);
4257 * If the inodes csum_bytes is the same as the original
4258 * csum_bytes then we know we haven't raced with any free()ers
4259 * so we can just reduce our inodes csum bytes and carry on.
4260 * Otherwise we have to do the normal free thing to account for
4261 * the case that the free side didn't free up its reserve
4262 * because of this outstanding reservation.
4264 if (BTRFS_I(inode
)->csum_bytes
== csum_bytes
)
4265 calc_csum_metadata_size(inode
, num_bytes
, 0);
4267 to_free
= calc_csum_metadata_size(inode
, num_bytes
, 0);
4268 spin_unlock(&BTRFS_I(inode
)->lock
);
4270 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
4273 btrfs_block_rsv_release(root
, block_rsv
, to_free
);
4277 spin_lock(&BTRFS_I(inode
)->lock
);
4278 if (extra_reserve
) {
4279 BTRFS_I(inode
)->delalloc_meta_reserved
= 1;
4282 BTRFS_I(inode
)->reserved_extents
+= nr_extents
;
4283 spin_unlock(&BTRFS_I(inode
)->lock
);
4285 block_rsv_add_bytes(block_rsv
, to_reserve
, 1);
4291 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4292 * @inode: the inode to release the reservation for
4293 * @num_bytes: the number of bytes we're releasing
4295 * This will release the metadata reservation for an inode. This can be called
4296 * once we complete IO for a given set of bytes to release their metadata
4299 void btrfs_delalloc_release_metadata(struct inode
*inode
, u64 num_bytes
)
4301 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4305 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
4306 spin_lock(&BTRFS_I(inode
)->lock
);
4307 dropped
= drop_outstanding_extent(inode
);
4309 to_free
= calc_csum_metadata_size(inode
, num_bytes
, 0);
4310 spin_unlock(&BTRFS_I(inode
)->lock
);
4312 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
4314 btrfs_block_rsv_release(root
, &root
->fs_info
->delalloc_block_rsv
,
4319 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4320 * @inode: inode we're writing to
4321 * @num_bytes: the number of bytes we want to allocate
4323 * This will do the following things
4325 * o reserve space in the data space info for num_bytes
4326 * o reserve space in the metadata space info based on number of outstanding
4327 * extents and how much csums will be needed
4328 * o add to the inodes ->delalloc_bytes
4329 * o add it to the fs_info's delalloc inodes list.
4331 * This will return 0 for success and -ENOSPC if there is no space left.
4333 int btrfs_delalloc_reserve_space(struct inode
*inode
, u64 num_bytes
)
4337 ret
= btrfs_check_data_free_space(inode
, num_bytes
);
4341 ret
= btrfs_delalloc_reserve_metadata(inode
, num_bytes
);
4343 btrfs_free_reserved_data_space(inode
, num_bytes
);
4351 * btrfs_delalloc_release_space - release data and metadata space for delalloc
4352 * @inode: inode we're releasing space for
4353 * @num_bytes: the number of bytes we want to free up
4355 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
4356 * called in the case that we don't need the metadata AND data reservations
4357 * anymore. So if there is an error or we insert an inline extent.
4359 * This function will release the metadata space that was not used and will
4360 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4361 * list if there are no delalloc bytes left.
4363 void btrfs_delalloc_release_space(struct inode
*inode
, u64 num_bytes
)
4365 btrfs_delalloc_release_metadata(inode
, num_bytes
);
4366 btrfs_free_reserved_data_space(inode
, num_bytes
);
4369 static int update_block_group(struct btrfs_trans_handle
*trans
,
4370 struct btrfs_root
*root
,
4371 u64 bytenr
, u64 num_bytes
, int alloc
)
4373 struct btrfs_block_group_cache
*cache
= NULL
;
4374 struct btrfs_fs_info
*info
= root
->fs_info
;
4375 u64 total
= num_bytes
;
4380 /* block accounting for super block */
4381 spin_lock(&info
->delalloc_lock
);
4382 old_val
= btrfs_super_bytes_used(info
->super_copy
);
4384 old_val
+= num_bytes
;
4386 old_val
-= num_bytes
;
4387 btrfs_set_super_bytes_used(info
->super_copy
, old_val
);
4388 spin_unlock(&info
->delalloc_lock
);
4391 cache
= btrfs_lookup_block_group(info
, bytenr
);
4394 if (cache
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
4395 BTRFS_BLOCK_GROUP_RAID1
|
4396 BTRFS_BLOCK_GROUP_RAID10
))
4401 * If this block group has free space cache written out, we
4402 * need to make sure to load it if we are removing space. This
4403 * is because we need the unpinning stage to actually add the
4404 * space back to the block group, otherwise we will leak space.
4406 if (!alloc
&& cache
->cached
== BTRFS_CACHE_NO
)
4407 cache_block_group(cache
, trans
, NULL
, 1);
4409 byte_in_group
= bytenr
- cache
->key
.objectid
;
4410 WARN_ON(byte_in_group
> cache
->key
.offset
);
4412 spin_lock(&cache
->space_info
->lock
);
4413 spin_lock(&cache
->lock
);
4415 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
4416 cache
->disk_cache_state
< BTRFS_DC_CLEAR
)
4417 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
4420 old_val
= btrfs_block_group_used(&cache
->item
);
4421 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
4423 old_val
+= num_bytes
;
4424 btrfs_set_block_group_used(&cache
->item
, old_val
);
4425 cache
->reserved
-= num_bytes
;
4426 cache
->space_info
->bytes_reserved
-= num_bytes
;
4427 cache
->space_info
->bytes_used
+= num_bytes
;
4428 cache
->space_info
->disk_used
+= num_bytes
* factor
;
4429 spin_unlock(&cache
->lock
);
4430 spin_unlock(&cache
->space_info
->lock
);
4432 old_val
-= num_bytes
;
4433 btrfs_set_block_group_used(&cache
->item
, old_val
);
4434 cache
->pinned
+= num_bytes
;
4435 cache
->space_info
->bytes_pinned
+= num_bytes
;
4436 cache
->space_info
->bytes_used
-= num_bytes
;
4437 cache
->space_info
->disk_used
-= num_bytes
* factor
;
4438 spin_unlock(&cache
->lock
);
4439 spin_unlock(&cache
->space_info
->lock
);
4441 set_extent_dirty(info
->pinned_extents
,
4442 bytenr
, bytenr
+ num_bytes
- 1,
4443 GFP_NOFS
| __GFP_NOFAIL
);
4445 btrfs_put_block_group(cache
);
4447 bytenr
+= num_bytes
;
4452 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
4454 struct btrfs_block_group_cache
*cache
;
4457 cache
= btrfs_lookup_first_block_group(root
->fs_info
, search_start
);
4461 bytenr
= cache
->key
.objectid
;
4462 btrfs_put_block_group(cache
);
4467 static int pin_down_extent(struct btrfs_root
*root
,
4468 struct btrfs_block_group_cache
*cache
,
4469 u64 bytenr
, u64 num_bytes
, int reserved
)
4471 spin_lock(&cache
->space_info
->lock
);
4472 spin_lock(&cache
->lock
);
4473 cache
->pinned
+= num_bytes
;
4474 cache
->space_info
->bytes_pinned
+= num_bytes
;
4476 cache
->reserved
-= num_bytes
;
4477 cache
->space_info
->bytes_reserved
-= num_bytes
;
4479 spin_unlock(&cache
->lock
);
4480 spin_unlock(&cache
->space_info
->lock
);
4482 set_extent_dirty(root
->fs_info
->pinned_extents
, bytenr
,
4483 bytenr
+ num_bytes
- 1, GFP_NOFS
| __GFP_NOFAIL
);
4488 * this function must be called within transaction
4490 int btrfs_pin_extent(struct btrfs_root
*root
,
4491 u64 bytenr
, u64 num_bytes
, int reserved
)
4493 struct btrfs_block_group_cache
*cache
;
4495 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
4498 pin_down_extent(root
, cache
, bytenr
, num_bytes
, reserved
);
4500 btrfs_put_block_group(cache
);
4505 * this function must be called within transaction
4507 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle
*trans
,
4508 struct btrfs_root
*root
,
4509 u64 bytenr
, u64 num_bytes
)
4511 struct btrfs_block_group_cache
*cache
;
4513 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
4517 * pull in the free space cache (if any) so that our pin
4518 * removes the free space from the cache. We have load_only set
4519 * to one because the slow code to read in the free extents does check
4520 * the pinned extents.
4522 cache_block_group(cache
, trans
, root
, 1);
4524 pin_down_extent(root
, cache
, bytenr
, num_bytes
, 0);
4526 /* remove us from the free space cache (if we're there at all) */
4527 btrfs_remove_free_space(cache
, bytenr
, num_bytes
);
4528 btrfs_put_block_group(cache
);
4533 * btrfs_update_reserved_bytes - update the block_group and space info counters
4534 * @cache: The cache we are manipulating
4535 * @num_bytes: The number of bytes in question
4536 * @reserve: One of the reservation enums
4538 * This is called by the allocator when it reserves space, or by somebody who is
4539 * freeing space that was never actually used on disk. For example if you
4540 * reserve some space for a new leaf in transaction A and before transaction A
4541 * commits you free that leaf, you call this with reserve set to 0 in order to
4542 * clear the reservation.
4544 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4545 * ENOSPC accounting. For data we handle the reservation through clearing the
4546 * delalloc bits in the io_tree. We have to do this since we could end up
4547 * allocating less disk space for the amount of data we have reserved in the
4548 * case of compression.
4550 * If this is a reservation and the block group has become read only we cannot
4551 * make the reservation and return -EAGAIN, otherwise this function always
4554 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
4555 u64 num_bytes
, int reserve
)
4557 struct btrfs_space_info
*space_info
= cache
->space_info
;
4559 spin_lock(&space_info
->lock
);
4560 spin_lock(&cache
->lock
);
4561 if (reserve
!= RESERVE_FREE
) {
4565 cache
->reserved
+= num_bytes
;
4566 space_info
->bytes_reserved
+= num_bytes
;
4567 if (reserve
== RESERVE_ALLOC
) {
4568 BUG_ON(space_info
->bytes_may_use
< num_bytes
);
4569 space_info
->bytes_may_use
-= num_bytes
;
4574 space_info
->bytes_readonly
+= num_bytes
;
4575 cache
->reserved
-= num_bytes
;
4576 space_info
->bytes_reserved
-= num_bytes
;
4577 space_info
->reservation_progress
++;
4579 spin_unlock(&cache
->lock
);
4580 spin_unlock(&space_info
->lock
);
4584 int btrfs_prepare_extent_commit(struct btrfs_trans_handle
*trans
,
4585 struct btrfs_root
*root
)
4587 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4588 struct btrfs_caching_control
*next
;
4589 struct btrfs_caching_control
*caching_ctl
;
4590 struct btrfs_block_group_cache
*cache
;
4592 down_write(&fs_info
->extent_commit_sem
);
4594 list_for_each_entry_safe(caching_ctl
, next
,
4595 &fs_info
->caching_block_groups
, list
) {
4596 cache
= caching_ctl
->block_group
;
4597 if (block_group_cache_done(cache
)) {
4598 cache
->last_byte_to_unpin
= (u64
)-1;
4599 list_del_init(&caching_ctl
->list
);
4600 put_caching_control(caching_ctl
);
4602 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
4606 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
4607 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
4609 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
4611 up_write(&fs_info
->extent_commit_sem
);
4613 update_global_block_rsv(fs_info
);
4617 static int unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
4619 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4620 struct btrfs_block_group_cache
*cache
= NULL
;
4623 while (start
<= end
) {
4625 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
4627 btrfs_put_block_group(cache
);
4628 cache
= btrfs_lookup_block_group(fs_info
, start
);
4632 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
4633 len
= min(len
, end
+ 1 - start
);
4635 if (start
< cache
->last_byte_to_unpin
) {
4636 len
= min(len
, cache
->last_byte_to_unpin
- start
);
4637 btrfs_add_free_space(cache
, start
, len
);
4642 spin_lock(&cache
->space_info
->lock
);
4643 spin_lock(&cache
->lock
);
4644 cache
->pinned
-= len
;
4645 cache
->space_info
->bytes_pinned
-= len
;
4647 cache
->space_info
->bytes_readonly
+= len
;
4648 spin_unlock(&cache
->lock
);
4649 spin_unlock(&cache
->space_info
->lock
);
4653 btrfs_put_block_group(cache
);
4657 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
4658 struct btrfs_root
*root
)
4660 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4661 struct extent_io_tree
*unpin
;
4666 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
4667 unpin
= &fs_info
->freed_extents
[1];
4669 unpin
= &fs_info
->freed_extents
[0];
4672 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
4677 if (btrfs_test_opt(root
, DISCARD
))
4678 ret
= btrfs_discard_extent(root
, start
,
4679 end
+ 1 - start
, NULL
);
4681 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
4682 unpin_extent_range(root
, start
, end
);
4689 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
4690 struct btrfs_root
*root
,
4691 u64 bytenr
, u64 num_bytes
, u64 parent
,
4692 u64 root_objectid
, u64 owner_objectid
,
4693 u64 owner_offset
, int refs_to_drop
,
4694 struct btrfs_delayed_extent_op
*extent_op
)
4696 struct btrfs_key key
;
4697 struct btrfs_path
*path
;
4698 struct btrfs_fs_info
*info
= root
->fs_info
;
4699 struct btrfs_root
*extent_root
= info
->extent_root
;
4700 struct extent_buffer
*leaf
;
4701 struct btrfs_extent_item
*ei
;
4702 struct btrfs_extent_inline_ref
*iref
;
4705 int extent_slot
= 0;
4706 int found_extent
= 0;
4711 path
= btrfs_alloc_path();
4716 path
->leave_spinning
= 1;
4718 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
4719 BUG_ON(!is_data
&& refs_to_drop
!= 1);
4721 ret
= lookup_extent_backref(trans
, extent_root
, path
, &iref
,
4722 bytenr
, num_bytes
, parent
,
4723 root_objectid
, owner_objectid
,
4726 extent_slot
= path
->slots
[0];
4727 while (extent_slot
>= 0) {
4728 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
4730 if (key
.objectid
!= bytenr
)
4732 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
4733 key
.offset
== num_bytes
) {
4737 if (path
->slots
[0] - extent_slot
> 5)
4741 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4742 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
4743 if (found_extent
&& item_size
< sizeof(*ei
))
4746 if (!found_extent
) {
4748 ret
= remove_extent_backref(trans
, extent_root
, path
,
4752 btrfs_release_path(path
);
4753 path
->leave_spinning
= 1;
4755 key
.objectid
= bytenr
;
4756 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
4757 key
.offset
= num_bytes
;
4759 ret
= btrfs_search_slot(trans
, extent_root
,
4762 printk(KERN_ERR
"umm, got %d back from search"
4763 ", was looking for %llu\n", ret
,
4764 (unsigned long long)bytenr
);
4766 btrfs_print_leaf(extent_root
,
4770 extent_slot
= path
->slots
[0];
4773 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
4775 printk(KERN_ERR
"btrfs unable to find ref byte nr %llu "
4776 "parent %llu root %llu owner %llu offset %llu\n",
4777 (unsigned long long)bytenr
,
4778 (unsigned long long)parent
,
4779 (unsigned long long)root_objectid
,
4780 (unsigned long long)owner_objectid
,
4781 (unsigned long long)owner_offset
);
4784 leaf
= path
->nodes
[0];
4785 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
4786 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4787 if (item_size
< sizeof(*ei
)) {
4788 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
4789 ret
= convert_extent_item_v0(trans
, extent_root
, path
,
4793 btrfs_release_path(path
);
4794 path
->leave_spinning
= 1;
4796 key
.objectid
= bytenr
;
4797 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
4798 key
.offset
= num_bytes
;
4800 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
4803 printk(KERN_ERR
"umm, got %d back from search"
4804 ", was looking for %llu\n", ret
,
4805 (unsigned long long)bytenr
);
4806 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
4809 extent_slot
= path
->slots
[0];
4810 leaf
= path
->nodes
[0];
4811 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
4814 BUG_ON(item_size
< sizeof(*ei
));
4815 ei
= btrfs_item_ptr(leaf
, extent_slot
,
4816 struct btrfs_extent_item
);
4817 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
) {
4818 struct btrfs_tree_block_info
*bi
;
4819 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
4820 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
4821 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
4824 refs
= btrfs_extent_refs(leaf
, ei
);
4825 BUG_ON(refs
< refs_to_drop
);
4826 refs
-= refs_to_drop
;
4830 __run_delayed_extent_op(extent_op
, leaf
, ei
);
4832 * In the case of inline back ref, reference count will
4833 * be updated by remove_extent_backref
4836 BUG_ON(!found_extent
);
4838 btrfs_set_extent_refs(leaf
, ei
, refs
);
4839 btrfs_mark_buffer_dirty(leaf
);
4842 ret
= remove_extent_backref(trans
, extent_root
, path
,
4849 BUG_ON(is_data
&& refs_to_drop
!=
4850 extent_data_ref_count(root
, path
, iref
));
4852 BUG_ON(path
->slots
[0] != extent_slot
);
4854 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
4855 path
->slots
[0] = extent_slot
;
4860 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
4863 btrfs_release_path(path
);
4866 ret
= btrfs_del_csums(trans
, root
, bytenr
, num_bytes
);
4869 invalidate_mapping_pages(info
->btree_inode
->i_mapping
,
4870 bytenr
>> PAGE_CACHE_SHIFT
,
4871 (bytenr
+ num_bytes
- 1) >> PAGE_CACHE_SHIFT
);
4874 ret
= update_block_group(trans
, root
, bytenr
, num_bytes
, 0);
4877 btrfs_free_path(path
);
4882 * when we free an block, it is possible (and likely) that we free the last
4883 * delayed ref for that extent as well. This searches the delayed ref tree for
4884 * a given extent, and if there are no other delayed refs to be processed, it
4885 * removes it from the tree.
4887 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
4888 struct btrfs_root
*root
, u64 bytenr
)
4890 struct btrfs_delayed_ref_head
*head
;
4891 struct btrfs_delayed_ref_root
*delayed_refs
;
4892 struct btrfs_delayed_ref_node
*ref
;
4893 struct rb_node
*node
;
4896 delayed_refs
= &trans
->transaction
->delayed_refs
;
4897 spin_lock(&delayed_refs
->lock
);
4898 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
4902 node
= rb_prev(&head
->node
.rb_node
);
4906 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
4908 /* there are still entries for this ref, we can't drop it */
4909 if (ref
->bytenr
== bytenr
)
4912 if (head
->extent_op
) {
4913 if (!head
->must_insert_reserved
)
4915 kfree(head
->extent_op
);
4916 head
->extent_op
= NULL
;
4920 * waiting for the lock here would deadlock. If someone else has it
4921 * locked they are already in the process of dropping it anyway
4923 if (!mutex_trylock(&head
->mutex
))
4927 * at this point we have a head with no other entries. Go
4928 * ahead and process it.
4930 head
->node
.in_tree
= 0;
4931 rb_erase(&head
->node
.rb_node
, &delayed_refs
->root
);
4933 delayed_refs
->num_entries
--;
4936 * we don't take a ref on the node because we're removing it from the
4937 * tree, so we just steal the ref the tree was holding.
4939 delayed_refs
->num_heads
--;
4940 if (list_empty(&head
->cluster
))
4941 delayed_refs
->num_heads_ready
--;
4943 list_del_init(&head
->cluster
);
4944 spin_unlock(&delayed_refs
->lock
);
4946 BUG_ON(head
->extent_op
);
4947 if (head
->must_insert_reserved
)
4950 mutex_unlock(&head
->mutex
);
4951 btrfs_put_delayed_ref(&head
->node
);
4954 spin_unlock(&delayed_refs
->lock
);
4958 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
4959 struct btrfs_root
*root
,
4960 struct extent_buffer
*buf
,
4961 u64 parent
, int last_ref
)
4963 struct btrfs_block_group_cache
*cache
= NULL
;
4966 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
4967 ret
= btrfs_add_delayed_tree_ref(trans
, buf
->start
, buf
->len
,
4968 parent
, root
->root_key
.objectid
,
4969 btrfs_header_level(buf
),
4970 BTRFS_DROP_DELAYED_REF
, NULL
);
4977 cache
= btrfs_lookup_block_group(root
->fs_info
, buf
->start
);
4979 if (btrfs_header_generation(buf
) == trans
->transid
) {
4980 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
4981 ret
= check_ref_cleanup(trans
, root
, buf
->start
);
4986 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
4987 pin_down_extent(root
, cache
, buf
->start
, buf
->len
, 1);
4991 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
4993 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
4994 btrfs_update_reserved_bytes(cache
, buf
->len
, RESERVE_FREE
);
4998 * Deleting the buffer, clear the corrupt flag since it doesn't matter
5001 clear_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
);
5002 btrfs_put_block_group(cache
);
5005 int btrfs_free_extent(struct btrfs_trans_handle
*trans
,
5006 struct btrfs_root
*root
,
5007 u64 bytenr
, u64 num_bytes
, u64 parent
,
5008 u64 root_objectid
, u64 owner
, u64 offset
)
5013 * tree log blocks never actually go into the extent allocation
5014 * tree, just update pinning info and exit early.
5016 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
5017 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
5018 /* unlocks the pinned mutex */
5019 btrfs_pin_extent(root
, bytenr
, num_bytes
, 1);
5021 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
5022 ret
= btrfs_add_delayed_tree_ref(trans
, bytenr
, num_bytes
,
5023 parent
, root_objectid
, (int)owner
,
5024 BTRFS_DROP_DELAYED_REF
, NULL
);
5027 ret
= btrfs_add_delayed_data_ref(trans
, bytenr
, num_bytes
,
5028 parent
, root_objectid
, owner
,
5029 offset
, BTRFS_DROP_DELAYED_REF
, NULL
);
5035 static u64
stripe_align(struct btrfs_root
*root
, u64 val
)
5037 u64 mask
= ((u64
)root
->stripesize
- 1);
5038 u64 ret
= (val
+ mask
) & ~mask
;
5043 * when we wait for progress in the block group caching, its because
5044 * our allocation attempt failed at least once. So, we must sleep
5045 * and let some progress happen before we try again.
5047 * This function will sleep at least once waiting for new free space to
5048 * show up, and then it will check the block group free space numbers
5049 * for our min num_bytes. Another option is to have it go ahead
5050 * and look in the rbtree for a free extent of a given size, but this
5054 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
5057 struct btrfs_caching_control
*caching_ctl
;
5060 caching_ctl
= get_caching_control(cache
);
5064 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
5065 (cache
->free_space_ctl
->free_space
>= num_bytes
));
5067 put_caching_control(caching_ctl
);
5072 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
5074 struct btrfs_caching_control
*caching_ctl
;
5077 caching_ctl
= get_caching_control(cache
);
5081 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
5083 put_caching_control(caching_ctl
);
5087 static int get_block_group_index(struct btrfs_block_group_cache
*cache
)
5090 if (cache
->flags
& BTRFS_BLOCK_GROUP_RAID10
)
5092 else if (cache
->flags
& BTRFS_BLOCK_GROUP_RAID1
)
5094 else if (cache
->flags
& BTRFS_BLOCK_GROUP_DUP
)
5096 else if (cache
->flags
& BTRFS_BLOCK_GROUP_RAID0
)
5103 enum btrfs_loop_type
{
5104 LOOP_FIND_IDEAL
= 0,
5105 LOOP_CACHING_NOWAIT
= 1,
5106 LOOP_CACHING_WAIT
= 2,
5107 LOOP_ALLOC_CHUNK
= 3,
5108 LOOP_NO_EMPTY_SIZE
= 4,
5112 * walks the btree of allocated extents and find a hole of a given size.
5113 * The key ins is changed to record the hole:
5114 * ins->objectid == block start
5115 * ins->flags = BTRFS_EXTENT_ITEM_KEY
5116 * ins->offset == number of blocks
5117 * Any available blocks before search_start are skipped.
5119 static noinline
int find_free_extent(struct btrfs_trans_handle
*trans
,
5120 struct btrfs_root
*orig_root
,
5121 u64 num_bytes
, u64 empty_size
,
5122 u64 search_start
, u64 search_end
,
5123 u64 hint_byte
, struct btrfs_key
*ins
,
5127 struct btrfs_root
*root
= orig_root
->fs_info
->extent_root
;
5128 struct btrfs_free_cluster
*last_ptr
= NULL
;
5129 struct btrfs_block_group_cache
*block_group
= NULL
;
5130 struct btrfs_block_group_cache
*used_block_group
;
5131 int empty_cluster
= 2 * 1024 * 1024;
5132 int allowed_chunk_alloc
= 0;
5133 int done_chunk_alloc
= 0;
5134 struct btrfs_space_info
*space_info
;
5137 int alloc_type
= (data
& BTRFS_BLOCK_GROUP_DATA
) ?
5138 RESERVE_ALLOC_NO_ACCOUNT
: RESERVE_ALLOC
;
5139 bool found_uncached_bg
= false;
5140 bool failed_cluster_refill
= false;
5141 bool failed_alloc
= false;
5142 bool use_cluster
= true;
5143 bool have_caching_bg
= false;
5144 u64 ideal_cache_percent
= 0;
5145 u64 ideal_cache_offset
= 0;
5147 WARN_ON(num_bytes
< root
->sectorsize
);
5148 btrfs_set_key_type(ins
, BTRFS_EXTENT_ITEM_KEY
);
5152 space_info
= __find_space_info(root
->fs_info
, data
);
5154 printk(KERN_ERR
"No space info for %llu\n", data
);
5159 * If the space info is for both data and metadata it means we have a
5160 * small filesystem and we can't use the clustering stuff.
5162 if (btrfs_mixed_space_info(space_info
))
5163 use_cluster
= false;
5165 if (orig_root
->ref_cows
|| empty_size
)
5166 allowed_chunk_alloc
= 1;
5168 if (data
& BTRFS_BLOCK_GROUP_METADATA
&& use_cluster
) {
5169 last_ptr
= &root
->fs_info
->meta_alloc_cluster
;
5170 if (!btrfs_test_opt(root
, SSD
))
5171 empty_cluster
= 64 * 1024;
5174 if ((data
& BTRFS_BLOCK_GROUP_DATA
) && use_cluster
&&
5175 btrfs_test_opt(root
, SSD
)) {
5176 last_ptr
= &root
->fs_info
->data_alloc_cluster
;
5180 spin_lock(&last_ptr
->lock
);
5181 if (last_ptr
->block_group
)
5182 hint_byte
= last_ptr
->window_start
;
5183 spin_unlock(&last_ptr
->lock
);
5186 search_start
= max(search_start
, first_logical_byte(root
, 0));
5187 search_start
= max(search_start
, hint_byte
);
5192 if (search_start
== hint_byte
) {
5194 block_group
= btrfs_lookup_block_group(root
->fs_info
,
5196 used_block_group
= block_group
;
5198 * we don't want to use the block group if it doesn't match our
5199 * allocation bits, or if its not cached.
5201 * However if we are re-searching with an ideal block group
5202 * picked out then we don't care that the block group is cached.
5204 if (block_group
&& block_group_bits(block_group
, data
) &&
5205 (block_group
->cached
!= BTRFS_CACHE_NO
||
5206 search_start
== ideal_cache_offset
)) {
5207 down_read(&space_info
->groups_sem
);
5208 if (list_empty(&block_group
->list
) ||
5211 * someone is removing this block group,
5212 * we can't jump into the have_block_group
5213 * target because our list pointers are not
5216 btrfs_put_block_group(block_group
);
5217 up_read(&space_info
->groups_sem
);
5219 index
= get_block_group_index(block_group
);
5220 goto have_block_group
;
5222 } else if (block_group
) {
5223 btrfs_put_block_group(block_group
);
5227 have_caching_bg
= false;
5228 down_read(&space_info
->groups_sem
);
5229 list_for_each_entry(block_group
, &space_info
->block_groups
[index
],
5234 used_block_group
= block_group
;
5235 btrfs_get_block_group(block_group
);
5236 search_start
= block_group
->key
.objectid
;
5239 * this can happen if we end up cycling through all the
5240 * raid types, but we want to make sure we only allocate
5241 * for the proper type.
5243 if (!block_group_bits(block_group
, data
)) {
5244 u64 extra
= BTRFS_BLOCK_GROUP_DUP
|
5245 BTRFS_BLOCK_GROUP_RAID1
|
5246 BTRFS_BLOCK_GROUP_RAID10
;
5249 * if they asked for extra copies and this block group
5250 * doesn't provide them, bail. This does allow us to
5251 * fill raid0 from raid1.
5253 if ((data
& extra
) && !(block_group
->flags
& extra
))
5258 cached
= block_group_cache_done(block_group
);
5259 if (unlikely(!cached
)) {
5262 found_uncached_bg
= true;
5263 ret
= cache_block_group(block_group
, trans
,
5265 if (block_group
->cached
== BTRFS_CACHE_FINISHED
)
5268 free_percent
= btrfs_block_group_used(&block_group
->item
);
5269 free_percent
*= 100;
5270 free_percent
= div64_u64(free_percent
,
5271 block_group
->key
.offset
);
5272 free_percent
= 100 - free_percent
;
5273 if (free_percent
> ideal_cache_percent
&&
5274 likely(!block_group
->ro
)) {
5275 ideal_cache_offset
= block_group
->key
.objectid
;
5276 ideal_cache_percent
= free_percent
;
5280 * The caching workers are limited to 2 threads, so we
5281 * can queue as much work as we care to.
5283 if (loop
> LOOP_FIND_IDEAL
) {
5284 ret
= cache_block_group(block_group
, trans
,
5290 * If loop is set for cached only, try the next block
5293 if (loop
== LOOP_FIND_IDEAL
)
5298 if (unlikely(block_group
->ro
))
5301 spin_lock(&block_group
->free_space_ctl
->tree_lock
);
5303 block_group
->free_space_ctl
->free_space
<
5304 num_bytes
+ empty_cluster
+ empty_size
) {
5305 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
5308 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
5311 * Ok we want to try and use the cluster allocator, so
5316 * the refill lock keeps out other
5317 * people trying to start a new cluster
5319 spin_lock(&last_ptr
->refill_lock
);
5320 used_block_group
= last_ptr
->block_group
;
5321 if (used_block_group
!= block_group
&&
5322 (!used_block_group
||
5323 used_block_group
->ro
||
5324 !block_group_bits(used_block_group
, data
))) {
5325 used_block_group
= block_group
;
5326 goto refill_cluster
;
5329 if (used_block_group
!= block_group
)
5330 btrfs_get_block_group(used_block_group
);
5332 offset
= btrfs_alloc_from_cluster(used_block_group
,
5333 last_ptr
, num_bytes
, used_block_group
->key
.objectid
);
5335 /* we have a block, we're done */
5336 spin_unlock(&last_ptr
->refill_lock
);
5340 WARN_ON(last_ptr
->block_group
!= used_block_group
);
5341 if (used_block_group
!= block_group
) {
5342 btrfs_put_block_group(used_block_group
);
5343 used_block_group
= block_group
;
5346 BUG_ON(used_block_group
!= block_group
);
5347 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
5348 * set up a new clusters, so lets just skip it
5349 * and let the allocator find whatever block
5350 * it can find. If we reach this point, we
5351 * will have tried the cluster allocator
5352 * plenty of times and not have found
5353 * anything, so we are likely way too
5354 * fragmented for the clustering stuff to find
5356 if (loop
>= LOOP_NO_EMPTY_SIZE
) {
5357 spin_unlock(&last_ptr
->refill_lock
);
5358 goto unclustered_alloc
;
5362 * this cluster didn't work out, free it and
5365 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
5367 /* allocate a cluster in this block group */
5368 ret
= btrfs_find_space_cluster(trans
, root
,
5369 block_group
, last_ptr
,
5370 search_start
, num_bytes
,
5371 empty_cluster
+ empty_size
);
5374 * now pull our allocation out of this
5377 offset
= btrfs_alloc_from_cluster(block_group
,
5378 last_ptr
, num_bytes
,
5381 /* we found one, proceed */
5382 spin_unlock(&last_ptr
->refill_lock
);
5385 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
5386 && !failed_cluster_refill
) {
5387 spin_unlock(&last_ptr
->refill_lock
);
5389 failed_cluster_refill
= true;
5390 wait_block_group_cache_progress(block_group
,
5391 num_bytes
+ empty_cluster
+ empty_size
);
5392 goto have_block_group
;
5396 * at this point we either didn't find a cluster
5397 * or we weren't able to allocate a block from our
5398 * cluster. Free the cluster we've been trying
5399 * to use, and go to the next block group
5401 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
5402 spin_unlock(&last_ptr
->refill_lock
);
5407 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
5408 num_bytes
, empty_size
);
5410 * If we didn't find a chunk, and we haven't failed on this
5411 * block group before, and this block group is in the middle of
5412 * caching and we are ok with waiting, then go ahead and wait
5413 * for progress to be made, and set failed_alloc to true.
5415 * If failed_alloc is true then we've already waited on this
5416 * block group once and should move on to the next block group.
5418 if (!offset
&& !failed_alloc
&& !cached
&&
5419 loop
> LOOP_CACHING_NOWAIT
) {
5420 wait_block_group_cache_progress(block_group
,
5421 num_bytes
+ empty_size
);
5422 failed_alloc
= true;
5423 goto have_block_group
;
5424 } else if (!offset
) {
5426 have_caching_bg
= true;
5430 search_start
= stripe_align(root
, offset
);
5431 /* move on to the next group */
5432 if (search_start
+ num_bytes
>= search_end
) {
5433 btrfs_add_free_space(used_block_group
, offset
, num_bytes
);
5437 /* move on to the next group */
5438 if (search_start
+ num_bytes
>
5439 used_block_group
->key
.objectid
+ used_block_group
->key
.offset
) {
5440 btrfs_add_free_space(used_block_group
, offset
, num_bytes
);
5444 ins
->objectid
= search_start
;
5445 ins
->offset
= num_bytes
;
5447 if (offset
< search_start
)
5448 btrfs_add_free_space(used_block_group
, offset
,
5449 search_start
- offset
);
5450 BUG_ON(offset
> search_start
);
5452 ret
= btrfs_update_reserved_bytes(used_block_group
, num_bytes
,
5454 if (ret
== -EAGAIN
) {
5455 btrfs_add_free_space(used_block_group
, offset
, num_bytes
);
5459 /* we are all good, lets return */
5460 ins
->objectid
= search_start
;
5461 ins
->offset
= num_bytes
;
5463 if (offset
< search_start
)
5464 btrfs_add_free_space(used_block_group
, offset
,
5465 search_start
- offset
);
5466 BUG_ON(offset
> search_start
);
5467 if (used_block_group
!= block_group
)
5468 btrfs_put_block_group(used_block_group
);
5469 btrfs_put_block_group(block_group
);
5472 failed_cluster_refill
= false;
5473 failed_alloc
= false;
5474 BUG_ON(index
!= get_block_group_index(block_group
));
5475 if (used_block_group
!= block_group
)
5476 btrfs_put_block_group(used_block_group
);
5477 btrfs_put_block_group(block_group
);
5479 up_read(&space_info
->groups_sem
);
5481 if (!ins
->objectid
&& loop
>= LOOP_CACHING_WAIT
&& have_caching_bg
)
5484 if (!ins
->objectid
&& ++index
< BTRFS_NR_RAID_TYPES
)
5487 /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5488 * for them to make caching progress. Also
5489 * determine the best possible bg to cache
5490 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5491 * caching kthreads as we move along
5492 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5493 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5494 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5497 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
) {
5499 if (loop
== LOOP_FIND_IDEAL
&& found_uncached_bg
) {
5500 found_uncached_bg
= false;
5502 if (!ideal_cache_percent
)
5506 * 1 of the following 2 things have happened so far
5508 * 1) We found an ideal block group for caching that
5509 * is mostly full and will cache quickly, so we might
5510 * as well wait for it.
5512 * 2) We searched for cached only and we didn't find
5513 * anything, and we didn't start any caching kthreads
5514 * either, so chances are we will loop through and
5515 * start a couple caching kthreads, and then come back
5516 * around and just wait for them. This will be slower
5517 * because we will have 2 caching kthreads reading at
5518 * the same time when we could have just started one
5519 * and waited for it to get far enough to give us an
5520 * allocation, so go ahead and go to the wait caching
5523 loop
= LOOP_CACHING_WAIT
;
5524 search_start
= ideal_cache_offset
;
5525 ideal_cache_percent
= 0;
5527 } else if (loop
== LOOP_FIND_IDEAL
) {
5529 * Didn't find a uncached bg, wait on anything we find
5532 loop
= LOOP_CACHING_WAIT
;
5538 if (loop
== LOOP_ALLOC_CHUNK
) {
5539 if (allowed_chunk_alloc
) {
5540 ret
= do_chunk_alloc(trans
, root
, num_bytes
+
5541 2 * 1024 * 1024, data
,
5542 CHUNK_ALLOC_LIMITED
);
5543 allowed_chunk_alloc
= 0;
5545 done_chunk_alloc
= 1;
5546 } else if (!done_chunk_alloc
&&
5547 space_info
->force_alloc
==
5548 CHUNK_ALLOC_NO_FORCE
) {
5549 space_info
->force_alloc
= CHUNK_ALLOC_LIMITED
;
5553 * We didn't allocate a chunk, go ahead and drop the
5554 * empty size and loop again.
5556 if (!done_chunk_alloc
)
5557 loop
= LOOP_NO_EMPTY_SIZE
;
5560 if (loop
== LOOP_NO_EMPTY_SIZE
) {
5566 } else if (!ins
->objectid
) {
5568 } else if (ins
->objectid
) {
5575 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
5576 int dump_block_groups
)
5578 struct btrfs_block_group_cache
*cache
;
5581 spin_lock(&info
->lock
);
5582 printk(KERN_INFO
"space_info %llu has %llu free, is %sfull\n",
5583 (unsigned long long)info
->flags
,
5584 (unsigned long long)(info
->total_bytes
- info
->bytes_used
-
5585 info
->bytes_pinned
- info
->bytes_reserved
-
5586 info
->bytes_readonly
),
5587 (info
->full
) ? "" : "not ");
5588 printk(KERN_INFO
"space_info total=%llu, used=%llu, pinned=%llu, "
5589 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5590 (unsigned long long)info
->total_bytes
,
5591 (unsigned long long)info
->bytes_used
,
5592 (unsigned long long)info
->bytes_pinned
,
5593 (unsigned long long)info
->bytes_reserved
,
5594 (unsigned long long)info
->bytes_may_use
,
5595 (unsigned long long)info
->bytes_readonly
);
5596 spin_unlock(&info
->lock
);
5598 if (!dump_block_groups
)
5601 down_read(&info
->groups_sem
);
5603 list_for_each_entry(cache
, &info
->block_groups
[index
], list
) {
5604 spin_lock(&cache
->lock
);
5605 printk(KERN_INFO
"block group %llu has %llu bytes, %llu used "
5606 "%llu pinned %llu reserved\n",
5607 (unsigned long long)cache
->key
.objectid
,
5608 (unsigned long long)cache
->key
.offset
,
5609 (unsigned long long)btrfs_block_group_used(&cache
->item
),
5610 (unsigned long long)cache
->pinned
,
5611 (unsigned long long)cache
->reserved
);
5612 btrfs_dump_free_space(cache
, bytes
);
5613 spin_unlock(&cache
->lock
);
5615 if (++index
< BTRFS_NR_RAID_TYPES
)
5617 up_read(&info
->groups_sem
);
5620 int btrfs_reserve_extent(struct btrfs_trans_handle
*trans
,
5621 struct btrfs_root
*root
,
5622 u64 num_bytes
, u64 min_alloc_size
,
5623 u64 empty_size
, u64 hint_byte
,
5624 u64 search_end
, struct btrfs_key
*ins
,
5628 u64 search_start
= 0;
5630 data
= btrfs_get_alloc_profile(root
, data
);
5633 * the only place that sets empty_size is btrfs_realloc_node, which
5634 * is not called recursively on allocations
5636 if (empty_size
|| root
->ref_cows
)
5637 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
5638 num_bytes
+ 2 * 1024 * 1024, data
,
5639 CHUNK_ALLOC_NO_FORCE
);
5641 WARN_ON(num_bytes
< root
->sectorsize
);
5642 ret
= find_free_extent(trans
, root
, num_bytes
, empty_size
,
5643 search_start
, search_end
, hint_byte
,
5646 if (ret
== -ENOSPC
&& num_bytes
> min_alloc_size
) {
5647 num_bytes
= num_bytes
>> 1;
5648 num_bytes
= num_bytes
& ~(root
->sectorsize
- 1);
5649 num_bytes
= max(num_bytes
, min_alloc_size
);
5650 do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
5651 num_bytes
, data
, CHUNK_ALLOC_FORCE
);
5654 if (ret
== -ENOSPC
&& btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
5655 struct btrfs_space_info
*sinfo
;
5657 sinfo
= __find_space_info(root
->fs_info
, data
);
5658 printk(KERN_ERR
"btrfs allocation failed flags %llu, "
5659 "wanted %llu\n", (unsigned long long)data
,
5660 (unsigned long long)num_bytes
);
5661 dump_space_info(sinfo
, num_bytes
, 1);
5664 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, ins
->offset
);
5669 static int __btrfs_free_reserved_extent(struct btrfs_root
*root
,
5670 u64 start
, u64 len
, int pin
)
5672 struct btrfs_block_group_cache
*cache
;
5675 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
5677 printk(KERN_ERR
"Unable to find block group for %llu\n",
5678 (unsigned long long)start
);
5682 if (btrfs_test_opt(root
, DISCARD
))
5683 ret
= btrfs_discard_extent(root
, start
, len
, NULL
);
5686 pin_down_extent(root
, cache
, start
, len
, 1);
5688 btrfs_add_free_space(cache
, start
, len
);
5689 btrfs_update_reserved_bytes(cache
, len
, RESERVE_FREE
);
5691 btrfs_put_block_group(cache
);
5693 trace_btrfs_reserved_extent_free(root
, start
, len
);
5698 int btrfs_free_reserved_extent(struct btrfs_root
*root
,
5701 return __btrfs_free_reserved_extent(root
, start
, len
, 0);
5704 int btrfs_free_and_pin_reserved_extent(struct btrfs_root
*root
,
5707 return __btrfs_free_reserved_extent(root
, start
, len
, 1);
5710 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
5711 struct btrfs_root
*root
,
5712 u64 parent
, u64 root_objectid
,
5713 u64 flags
, u64 owner
, u64 offset
,
5714 struct btrfs_key
*ins
, int ref_mod
)
5717 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5718 struct btrfs_extent_item
*extent_item
;
5719 struct btrfs_extent_inline_ref
*iref
;
5720 struct btrfs_path
*path
;
5721 struct extent_buffer
*leaf
;
5726 type
= BTRFS_SHARED_DATA_REF_KEY
;
5728 type
= BTRFS_EXTENT_DATA_REF_KEY
;
5730 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
5732 path
= btrfs_alloc_path();
5736 path
->leave_spinning
= 1;
5737 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
5741 leaf
= path
->nodes
[0];
5742 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
5743 struct btrfs_extent_item
);
5744 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
5745 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
5746 btrfs_set_extent_flags(leaf
, extent_item
,
5747 flags
| BTRFS_EXTENT_FLAG_DATA
);
5749 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
5750 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
5752 struct btrfs_shared_data_ref
*ref
;
5753 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
5754 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
5755 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
5757 struct btrfs_extent_data_ref
*ref
;
5758 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
5759 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
5760 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
5761 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
5762 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
5765 btrfs_mark_buffer_dirty(path
->nodes
[0]);
5766 btrfs_free_path(path
);
5768 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1);
5770 printk(KERN_ERR
"btrfs update block group failed for %llu "
5771 "%llu\n", (unsigned long long)ins
->objectid
,
5772 (unsigned long long)ins
->offset
);
5778 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
5779 struct btrfs_root
*root
,
5780 u64 parent
, u64 root_objectid
,
5781 u64 flags
, struct btrfs_disk_key
*key
,
5782 int level
, struct btrfs_key
*ins
)
5785 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5786 struct btrfs_extent_item
*extent_item
;
5787 struct btrfs_tree_block_info
*block_info
;
5788 struct btrfs_extent_inline_ref
*iref
;
5789 struct btrfs_path
*path
;
5790 struct extent_buffer
*leaf
;
5791 u32 size
= sizeof(*extent_item
) + sizeof(*block_info
) + sizeof(*iref
);
5793 path
= btrfs_alloc_path();
5797 path
->leave_spinning
= 1;
5798 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
5802 leaf
= path
->nodes
[0];
5803 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
5804 struct btrfs_extent_item
);
5805 btrfs_set_extent_refs(leaf
, extent_item
, 1);
5806 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
5807 btrfs_set_extent_flags(leaf
, extent_item
,
5808 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
5809 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
5811 btrfs_set_tree_block_key(leaf
, block_info
, key
);
5812 btrfs_set_tree_block_level(leaf
, block_info
, level
);
5814 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
5816 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
5817 btrfs_set_extent_inline_ref_type(leaf
, iref
,
5818 BTRFS_SHARED_BLOCK_REF_KEY
);
5819 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
5821 btrfs_set_extent_inline_ref_type(leaf
, iref
,
5822 BTRFS_TREE_BLOCK_REF_KEY
);
5823 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
5826 btrfs_mark_buffer_dirty(leaf
);
5827 btrfs_free_path(path
);
5829 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1);
5831 printk(KERN_ERR
"btrfs update block group failed for %llu "
5832 "%llu\n", (unsigned long long)ins
->objectid
,
5833 (unsigned long long)ins
->offset
);
5839 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
5840 struct btrfs_root
*root
,
5841 u64 root_objectid
, u64 owner
,
5842 u64 offset
, struct btrfs_key
*ins
)
5846 BUG_ON(root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
5848 ret
= btrfs_add_delayed_data_ref(trans
, ins
->objectid
, ins
->offset
,
5849 0, root_objectid
, owner
, offset
,
5850 BTRFS_ADD_DELAYED_EXTENT
, NULL
);
5855 * this is used by the tree logging recovery code. It records that
5856 * an extent has been allocated and makes sure to clear the free
5857 * space cache bits as well
5859 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
5860 struct btrfs_root
*root
,
5861 u64 root_objectid
, u64 owner
, u64 offset
,
5862 struct btrfs_key
*ins
)
5865 struct btrfs_block_group_cache
*block_group
;
5866 struct btrfs_caching_control
*caching_ctl
;
5867 u64 start
= ins
->objectid
;
5868 u64 num_bytes
= ins
->offset
;
5870 block_group
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
5871 cache_block_group(block_group
, trans
, NULL
, 0);
5872 caching_ctl
= get_caching_control(block_group
);
5875 BUG_ON(!block_group_cache_done(block_group
));
5876 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
5879 mutex_lock(&caching_ctl
->mutex
);
5881 if (start
>= caching_ctl
->progress
) {
5882 ret
= add_excluded_extent(root
, start
, num_bytes
);
5884 } else if (start
+ num_bytes
<= caching_ctl
->progress
) {
5885 ret
= btrfs_remove_free_space(block_group
,
5889 num_bytes
= caching_ctl
->progress
- start
;
5890 ret
= btrfs_remove_free_space(block_group
,
5894 start
= caching_ctl
->progress
;
5895 num_bytes
= ins
->objectid
+ ins
->offset
-
5896 caching_ctl
->progress
;
5897 ret
= add_excluded_extent(root
, start
, num_bytes
);
5901 mutex_unlock(&caching_ctl
->mutex
);
5902 put_caching_control(caching_ctl
);
5905 ret
= btrfs_update_reserved_bytes(block_group
, ins
->offset
,
5906 RESERVE_ALLOC_NO_ACCOUNT
);
5908 btrfs_put_block_group(block_group
);
5909 ret
= alloc_reserved_file_extent(trans
, root
, 0, root_objectid
,
5910 0, owner
, offset
, ins
, 1);
5914 struct extent_buffer
*btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
,
5915 struct btrfs_root
*root
,
5916 u64 bytenr
, u32 blocksize
,
5919 struct extent_buffer
*buf
;
5921 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
5923 return ERR_PTR(-ENOMEM
);
5924 btrfs_set_header_generation(buf
, trans
->transid
);
5925 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, buf
, level
);
5926 btrfs_tree_lock(buf
);
5927 clean_tree_block(trans
, root
, buf
);
5929 btrfs_set_lock_blocking(buf
);
5930 btrfs_set_buffer_uptodate(buf
);
5932 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
5934 * we allow two log transactions at a time, use different
5935 * EXENT bit to differentiate dirty pages.
5937 if (root
->log_transid
% 2 == 0)
5938 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
5939 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
5941 set_extent_new(&root
->dirty_log_pages
, buf
->start
,
5942 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
5944 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
5945 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
5947 trans
->blocks_used
++;
5948 /* this returns a buffer locked for blocking */
5952 static struct btrfs_block_rsv
*
5953 use_block_rsv(struct btrfs_trans_handle
*trans
,
5954 struct btrfs_root
*root
, u32 blocksize
)
5956 struct btrfs_block_rsv
*block_rsv
;
5957 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
5960 block_rsv
= get_block_rsv(trans
, root
);
5962 if (block_rsv
->size
== 0) {
5963 ret
= reserve_metadata_bytes(root
, block_rsv
, blocksize
, 0);
5965 * If we couldn't reserve metadata bytes try and use some from
5966 * the global reserve.
5968 if (ret
&& block_rsv
!= global_rsv
) {
5969 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
5972 return ERR_PTR(ret
);
5974 return ERR_PTR(ret
);
5979 ret
= block_rsv_use_bytes(block_rsv
, blocksize
);
5983 static DEFINE_RATELIMIT_STATE(_rs
,
5984 DEFAULT_RATELIMIT_INTERVAL
,
5985 /*DEFAULT_RATELIMIT_BURST*/ 2);
5986 if (__ratelimit(&_rs
)) {
5987 printk(KERN_DEBUG
"btrfs: block rsv returned %d\n", ret
);
5990 ret
= reserve_metadata_bytes(root
, block_rsv
, blocksize
, 0);
5993 } else if (ret
&& block_rsv
!= global_rsv
) {
5994 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
6000 return ERR_PTR(-ENOSPC
);
6003 static void unuse_block_rsv(struct btrfs_block_rsv
*block_rsv
, u32 blocksize
)
6005 block_rsv_add_bytes(block_rsv
, blocksize
, 0);
6006 block_rsv_release_bytes(block_rsv
, NULL
, 0);
6010 * finds a free extent and does all the dirty work required for allocation
6011 * returns the key for the extent through ins, and a tree buffer for
6012 * the first block of the extent through buf.
6014 * returns the tree buffer or NULL.
6016 struct extent_buffer
*btrfs_alloc_free_block(struct btrfs_trans_handle
*trans
,
6017 struct btrfs_root
*root
, u32 blocksize
,
6018 u64 parent
, u64 root_objectid
,
6019 struct btrfs_disk_key
*key
, int level
,
6020 u64 hint
, u64 empty_size
)
6022 struct btrfs_key ins
;
6023 struct btrfs_block_rsv
*block_rsv
;
6024 struct extent_buffer
*buf
;
6029 block_rsv
= use_block_rsv(trans
, root
, blocksize
);
6030 if (IS_ERR(block_rsv
))
6031 return ERR_CAST(block_rsv
);
6033 ret
= btrfs_reserve_extent(trans
, root
, blocksize
, blocksize
,
6034 empty_size
, hint
, (u64
)-1, &ins
, 0);
6036 unuse_block_rsv(block_rsv
, blocksize
);
6037 return ERR_PTR(ret
);
6040 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
,
6042 BUG_ON(IS_ERR(buf
));
6044 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
6046 parent
= ins
.objectid
;
6047 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
6051 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
6052 struct btrfs_delayed_extent_op
*extent_op
;
6053 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
6056 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
6058 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
6059 extent_op
->flags_to_set
= flags
;
6060 extent_op
->update_key
= 1;
6061 extent_op
->update_flags
= 1;
6062 extent_op
->is_data
= 0;
6064 ret
= btrfs_add_delayed_tree_ref(trans
, ins
.objectid
,
6065 ins
.offset
, parent
, root_objectid
,
6066 level
, BTRFS_ADD_DELAYED_EXTENT
,
6073 struct walk_control
{
6074 u64 refs
[BTRFS_MAX_LEVEL
];
6075 u64 flags
[BTRFS_MAX_LEVEL
];
6076 struct btrfs_key update_progress
;
6086 #define DROP_REFERENCE 1
6087 #define UPDATE_BACKREF 2
6089 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
6090 struct btrfs_root
*root
,
6091 struct walk_control
*wc
,
6092 struct btrfs_path
*path
)
6100 struct btrfs_key key
;
6101 struct extent_buffer
*eb
;
6106 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
6107 wc
->reada_count
= wc
->reada_count
* 2 / 3;
6108 wc
->reada_count
= max(wc
->reada_count
, 2);
6110 wc
->reada_count
= wc
->reada_count
* 3 / 2;
6111 wc
->reada_count
= min_t(int, wc
->reada_count
,
6112 BTRFS_NODEPTRS_PER_BLOCK(root
));
6115 eb
= path
->nodes
[wc
->level
];
6116 nritems
= btrfs_header_nritems(eb
);
6117 blocksize
= btrfs_level_size(root
, wc
->level
- 1);
6119 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
6120 if (nread
>= wc
->reada_count
)
6124 bytenr
= btrfs_node_blockptr(eb
, slot
);
6125 generation
= btrfs_node_ptr_generation(eb
, slot
);
6127 if (slot
== path
->slots
[wc
->level
])
6130 if (wc
->stage
== UPDATE_BACKREF
&&
6131 generation
<= root
->root_key
.offset
)
6134 /* We don't lock the tree block, it's OK to be racy here */
6135 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, blocksize
,
6140 if (wc
->stage
== DROP_REFERENCE
) {
6144 if (wc
->level
== 1 &&
6145 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
6147 if (!wc
->update_ref
||
6148 generation
<= root
->root_key
.offset
)
6150 btrfs_node_key_to_cpu(eb
, &key
, slot
);
6151 ret
= btrfs_comp_cpu_keys(&key
,
6152 &wc
->update_progress
);
6156 if (wc
->level
== 1 &&
6157 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
6161 ret
= readahead_tree_block(root
, bytenr
, blocksize
,
6167 wc
->reada_slot
= slot
;
6171 * hepler to process tree block while walking down the tree.
6173 * when wc->stage == UPDATE_BACKREF, this function updates
6174 * back refs for pointers in the block.
6176 * NOTE: return value 1 means we should stop walking down.
6178 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
6179 struct btrfs_root
*root
,
6180 struct btrfs_path
*path
,
6181 struct walk_control
*wc
, int lookup_info
)
6183 int level
= wc
->level
;
6184 struct extent_buffer
*eb
= path
->nodes
[level
];
6185 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
6188 if (wc
->stage
== UPDATE_BACKREF
&&
6189 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
6193 * when reference count of tree block is 1, it won't increase
6194 * again. once full backref flag is set, we never clear it.
6197 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
6198 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
6199 BUG_ON(!path
->locks
[level
]);
6200 ret
= btrfs_lookup_extent_info(trans
, root
,
6205 BUG_ON(wc
->refs
[level
] == 0);
6208 if (wc
->stage
== DROP_REFERENCE
) {
6209 if (wc
->refs
[level
] > 1)
6212 if (path
->locks
[level
] && !wc
->keep_locks
) {
6213 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
6214 path
->locks
[level
] = 0;
6219 /* wc->stage == UPDATE_BACKREF */
6220 if (!(wc
->flags
[level
] & flag
)) {
6221 BUG_ON(!path
->locks
[level
]);
6222 ret
= btrfs_inc_ref(trans
, root
, eb
, 1);
6224 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
6226 ret
= btrfs_set_disk_extent_flags(trans
, root
, eb
->start
,
6229 wc
->flags
[level
] |= flag
;
6233 * the block is shared by multiple trees, so it's not good to
6234 * keep the tree lock
6236 if (path
->locks
[level
] && level
> 0) {
6237 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
6238 path
->locks
[level
] = 0;
6244 * hepler to process tree block pointer.
6246 * when wc->stage == DROP_REFERENCE, this function checks
6247 * reference count of the block pointed to. if the block
6248 * is shared and we need update back refs for the subtree
6249 * rooted at the block, this function changes wc->stage to
6250 * UPDATE_BACKREF. if the block is shared and there is no
6251 * need to update back, this function drops the reference
6254 * NOTE: return value 1 means we should stop walking down.
6256 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
6257 struct btrfs_root
*root
,
6258 struct btrfs_path
*path
,
6259 struct walk_control
*wc
, int *lookup_info
)
6265 struct btrfs_key key
;
6266 struct extent_buffer
*next
;
6267 int level
= wc
->level
;
6271 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
6272 path
->slots
[level
]);
6274 * if the lower level block was created before the snapshot
6275 * was created, we know there is no need to update back refs
6278 if (wc
->stage
== UPDATE_BACKREF
&&
6279 generation
<= root
->root_key
.offset
) {
6284 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
6285 blocksize
= btrfs_level_size(root
, level
- 1);
6287 next
= btrfs_find_tree_block(root
, bytenr
, blocksize
);
6289 next
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
6294 btrfs_tree_lock(next
);
6295 btrfs_set_lock_blocking(next
);
6297 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, blocksize
,
6298 &wc
->refs
[level
- 1],
6299 &wc
->flags
[level
- 1]);
6301 BUG_ON(wc
->refs
[level
- 1] == 0);
6304 if (wc
->stage
== DROP_REFERENCE
) {
6305 if (wc
->refs
[level
- 1] > 1) {
6307 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
6310 if (!wc
->update_ref
||
6311 generation
<= root
->root_key
.offset
)
6314 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
6315 path
->slots
[level
]);
6316 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
6320 wc
->stage
= UPDATE_BACKREF
;
6321 wc
->shared_level
= level
- 1;
6325 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
6329 if (!btrfs_buffer_uptodate(next
, generation
)) {
6330 btrfs_tree_unlock(next
);
6331 free_extent_buffer(next
);
6337 if (reada
&& level
== 1)
6338 reada_walk_down(trans
, root
, wc
, path
);
6339 next
= read_tree_block(root
, bytenr
, blocksize
, generation
);
6342 btrfs_tree_lock(next
);
6343 btrfs_set_lock_blocking(next
);
6347 BUG_ON(level
!= btrfs_header_level(next
));
6348 path
->nodes
[level
] = next
;
6349 path
->slots
[level
] = 0;
6350 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
6356 wc
->refs
[level
- 1] = 0;
6357 wc
->flags
[level
- 1] = 0;
6358 if (wc
->stage
== DROP_REFERENCE
) {
6359 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
6360 parent
= path
->nodes
[level
]->start
;
6362 BUG_ON(root
->root_key
.objectid
!=
6363 btrfs_header_owner(path
->nodes
[level
]));
6367 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
, parent
,
6368 root
->root_key
.objectid
, level
- 1, 0);
6371 btrfs_tree_unlock(next
);
6372 free_extent_buffer(next
);
6378 * hepler to process tree block while walking up the tree.
6380 * when wc->stage == DROP_REFERENCE, this function drops
6381 * reference count on the block.
6383 * when wc->stage == UPDATE_BACKREF, this function changes
6384 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6385 * to UPDATE_BACKREF previously while processing the block.
6387 * NOTE: return value 1 means we should stop walking up.
6389 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
6390 struct btrfs_root
*root
,
6391 struct btrfs_path
*path
,
6392 struct walk_control
*wc
)
6395 int level
= wc
->level
;
6396 struct extent_buffer
*eb
= path
->nodes
[level
];
6399 if (wc
->stage
== UPDATE_BACKREF
) {
6400 BUG_ON(wc
->shared_level
< level
);
6401 if (level
< wc
->shared_level
)
6404 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
6408 wc
->stage
= DROP_REFERENCE
;
6409 wc
->shared_level
= -1;
6410 path
->slots
[level
] = 0;
6413 * check reference count again if the block isn't locked.
6414 * we should start walking down the tree again if reference
6417 if (!path
->locks
[level
]) {
6419 btrfs_tree_lock(eb
);
6420 btrfs_set_lock_blocking(eb
);
6421 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
6423 ret
= btrfs_lookup_extent_info(trans
, root
,
6428 BUG_ON(wc
->refs
[level
] == 0);
6429 if (wc
->refs
[level
] == 1) {
6430 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
6436 /* wc->stage == DROP_REFERENCE */
6437 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
6439 if (wc
->refs
[level
] == 1) {
6441 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
6442 ret
= btrfs_dec_ref(trans
, root
, eb
, 1);
6444 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
6447 /* make block locked assertion in clean_tree_block happy */
6448 if (!path
->locks
[level
] &&
6449 btrfs_header_generation(eb
) == trans
->transid
) {
6450 btrfs_tree_lock(eb
);
6451 btrfs_set_lock_blocking(eb
);
6452 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
6454 clean_tree_block(trans
, root
, eb
);
6457 if (eb
== root
->node
) {
6458 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
6461 BUG_ON(root
->root_key
.objectid
!=
6462 btrfs_header_owner(eb
));
6464 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
6465 parent
= path
->nodes
[level
+ 1]->start
;
6467 BUG_ON(root
->root_key
.objectid
!=
6468 btrfs_header_owner(path
->nodes
[level
+ 1]));
6471 btrfs_free_tree_block(trans
, root
, eb
, parent
, wc
->refs
[level
] == 1);
6473 wc
->refs
[level
] = 0;
6474 wc
->flags
[level
] = 0;
6478 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
6479 struct btrfs_root
*root
,
6480 struct btrfs_path
*path
,
6481 struct walk_control
*wc
)
6483 int level
= wc
->level
;
6484 int lookup_info
= 1;
6487 while (level
>= 0) {
6488 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
6495 if (path
->slots
[level
] >=
6496 btrfs_header_nritems(path
->nodes
[level
]))
6499 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
6501 path
->slots
[level
]++;
6510 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
6511 struct btrfs_root
*root
,
6512 struct btrfs_path
*path
,
6513 struct walk_control
*wc
, int max_level
)
6515 int level
= wc
->level
;
6518 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
6519 while (level
< max_level
&& path
->nodes
[level
]) {
6521 if (path
->slots
[level
] + 1 <
6522 btrfs_header_nritems(path
->nodes
[level
])) {
6523 path
->slots
[level
]++;
6526 ret
= walk_up_proc(trans
, root
, path
, wc
);
6530 if (path
->locks
[level
]) {
6531 btrfs_tree_unlock_rw(path
->nodes
[level
],
6532 path
->locks
[level
]);
6533 path
->locks
[level
] = 0;
6535 free_extent_buffer(path
->nodes
[level
]);
6536 path
->nodes
[level
] = NULL
;
6544 * drop a subvolume tree.
6546 * this function traverses the tree freeing any blocks that only
6547 * referenced by the tree.
6549 * when a shared tree block is found. this function decreases its
6550 * reference count by one. if update_ref is true, this function
6551 * also make sure backrefs for the shared block and all lower level
6552 * blocks are properly updated.
6554 void btrfs_drop_snapshot(struct btrfs_root
*root
,
6555 struct btrfs_block_rsv
*block_rsv
, int update_ref
)
6557 struct btrfs_path
*path
;
6558 struct btrfs_trans_handle
*trans
;
6559 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
6560 struct btrfs_root_item
*root_item
= &root
->root_item
;
6561 struct walk_control
*wc
;
6562 struct btrfs_key key
;
6567 path
= btrfs_alloc_path();
6573 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
6575 btrfs_free_path(path
);
6580 trans
= btrfs_start_transaction(tree_root
, 0);
6581 BUG_ON(IS_ERR(trans
));
6584 trans
->block_rsv
= block_rsv
;
6586 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
6587 level
= btrfs_header_level(root
->node
);
6588 path
->nodes
[level
] = btrfs_lock_root_node(root
);
6589 btrfs_set_lock_blocking(path
->nodes
[level
]);
6590 path
->slots
[level
] = 0;
6591 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
6592 memset(&wc
->update_progress
, 0,
6593 sizeof(wc
->update_progress
));
6595 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
6596 memcpy(&wc
->update_progress
, &key
,
6597 sizeof(wc
->update_progress
));
6599 level
= root_item
->drop_level
;
6601 path
->lowest_level
= level
;
6602 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6603 path
->lowest_level
= 0;
6611 * unlock our path, this is safe because only this
6612 * function is allowed to delete this snapshot
6614 btrfs_unlock_up_safe(path
, 0);
6616 level
= btrfs_header_level(root
->node
);
6618 btrfs_tree_lock(path
->nodes
[level
]);
6619 btrfs_set_lock_blocking(path
->nodes
[level
]);
6621 ret
= btrfs_lookup_extent_info(trans
, root
,
6622 path
->nodes
[level
]->start
,
6623 path
->nodes
[level
]->len
,
6627 BUG_ON(wc
->refs
[level
] == 0);
6629 if (level
== root_item
->drop_level
)
6632 btrfs_tree_unlock(path
->nodes
[level
]);
6633 WARN_ON(wc
->refs
[level
] != 1);
6639 wc
->shared_level
= -1;
6640 wc
->stage
= DROP_REFERENCE
;
6641 wc
->update_ref
= update_ref
;
6643 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
6646 ret
= walk_down_tree(trans
, root
, path
, wc
);
6652 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
6659 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
6663 if (wc
->stage
== DROP_REFERENCE
) {
6665 btrfs_node_key(path
->nodes
[level
],
6666 &root_item
->drop_progress
,
6667 path
->slots
[level
]);
6668 root_item
->drop_level
= level
;
6671 BUG_ON(wc
->level
== 0);
6672 if (btrfs_should_end_transaction(trans
, tree_root
)) {
6673 ret
= btrfs_update_root(trans
, tree_root
,
6678 btrfs_end_transaction_throttle(trans
, tree_root
);
6679 trans
= btrfs_start_transaction(tree_root
, 0);
6680 BUG_ON(IS_ERR(trans
));
6682 trans
->block_rsv
= block_rsv
;
6685 btrfs_release_path(path
);
6688 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
6691 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
6692 ret
= btrfs_find_last_root(tree_root
, root
->root_key
.objectid
,
6696 /* if we fail to delete the orphan item this time
6697 * around, it'll get picked up the next time.
6699 * The most common failure here is just -ENOENT.
6701 btrfs_del_orphan_item(trans
, tree_root
,
6702 root
->root_key
.objectid
);
6706 if (root
->in_radix
) {
6707 btrfs_free_fs_root(tree_root
->fs_info
, root
);
6709 free_extent_buffer(root
->node
);
6710 free_extent_buffer(root
->commit_root
);
6714 btrfs_end_transaction_throttle(trans
, tree_root
);
6716 btrfs_free_path(path
);
6719 btrfs_std_error(root
->fs_info
, err
);
6724 * drop subtree rooted at tree block 'node'.
6726 * NOTE: this function will unlock and release tree block 'node'
6728 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
6729 struct btrfs_root
*root
,
6730 struct extent_buffer
*node
,
6731 struct extent_buffer
*parent
)
6733 struct btrfs_path
*path
;
6734 struct walk_control
*wc
;
6740 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
6742 path
= btrfs_alloc_path();
6746 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
6748 btrfs_free_path(path
);
6752 btrfs_assert_tree_locked(parent
);
6753 parent_level
= btrfs_header_level(parent
);
6754 extent_buffer_get(parent
);
6755 path
->nodes
[parent_level
] = parent
;
6756 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
6758 btrfs_assert_tree_locked(node
);
6759 level
= btrfs_header_level(node
);
6760 path
->nodes
[level
] = node
;
6761 path
->slots
[level
] = 0;
6762 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
6764 wc
->refs
[parent_level
] = 1;
6765 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
6767 wc
->shared_level
= -1;
6768 wc
->stage
= DROP_REFERENCE
;
6771 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
6774 wret
= walk_down_tree(trans
, root
, path
, wc
);
6780 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
6788 btrfs_free_path(path
);
6792 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
6795 u64 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
6796 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
6799 * we add in the count of missing devices because we want
6800 * to make sure that any RAID levels on a degraded FS
6801 * continue to be honored.
6803 num_devices
= root
->fs_info
->fs_devices
->rw_devices
+
6804 root
->fs_info
->fs_devices
->missing_devices
;
6806 if (num_devices
== 1) {
6807 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
6808 stripped
= flags
& ~stripped
;
6810 /* turn raid0 into single device chunks */
6811 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
6814 /* turn mirroring into duplication */
6815 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
6816 BTRFS_BLOCK_GROUP_RAID10
))
6817 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
6820 /* they already had raid on here, just return */
6821 if (flags
& stripped
)
6824 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
6825 stripped
= flags
& ~stripped
;
6827 /* switch duplicated blocks with raid1 */
6828 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
6829 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
6831 /* turn single device chunks into raid0 */
6832 return stripped
| BTRFS_BLOCK_GROUP_RAID0
;
6837 static int set_block_group_ro(struct btrfs_block_group_cache
*cache
, int force
)
6839 struct btrfs_space_info
*sinfo
= cache
->space_info
;
6841 u64 min_allocable_bytes
;
6846 * We need some metadata space and system metadata space for
6847 * allocating chunks in some corner cases until we force to set
6848 * it to be readonly.
6851 (BTRFS_BLOCK_GROUP_SYSTEM
| BTRFS_BLOCK_GROUP_METADATA
)) &&
6853 min_allocable_bytes
= 1 * 1024 * 1024;
6855 min_allocable_bytes
= 0;
6857 spin_lock(&sinfo
->lock
);
6858 spin_lock(&cache
->lock
);
6865 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
6866 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
6868 if (sinfo
->bytes_used
+ sinfo
->bytes_reserved
+ sinfo
->bytes_pinned
+
6869 sinfo
->bytes_may_use
+ sinfo
->bytes_readonly
+ num_bytes
+
6870 min_allocable_bytes
<= sinfo
->total_bytes
) {
6871 sinfo
->bytes_readonly
+= num_bytes
;
6876 spin_unlock(&cache
->lock
);
6877 spin_unlock(&sinfo
->lock
);
6881 int btrfs_set_block_group_ro(struct btrfs_root
*root
,
6882 struct btrfs_block_group_cache
*cache
)
6885 struct btrfs_trans_handle
*trans
;
6891 trans
= btrfs_join_transaction(root
);
6892 BUG_ON(IS_ERR(trans
));
6894 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
6895 if (alloc_flags
!= cache
->flags
)
6896 do_chunk_alloc(trans
, root
, 2 * 1024 * 1024, alloc_flags
,
6899 ret
= set_block_group_ro(cache
, 0);
6902 alloc_flags
= get_alloc_profile(root
, cache
->space_info
->flags
);
6903 ret
= do_chunk_alloc(trans
, root
, 2 * 1024 * 1024, alloc_flags
,
6907 ret
= set_block_group_ro(cache
, 0);
6909 btrfs_end_transaction(trans
, root
);
6913 int btrfs_force_chunk_alloc(struct btrfs_trans_handle
*trans
,
6914 struct btrfs_root
*root
, u64 type
)
6916 u64 alloc_flags
= get_alloc_profile(root
, type
);
6917 return do_chunk_alloc(trans
, root
, 2 * 1024 * 1024, alloc_flags
,
6922 * helper to account the unused space of all the readonly block group in the
6923 * list. takes mirrors into account.
6925 static u64
__btrfs_get_ro_block_group_free_space(struct list_head
*groups_list
)
6927 struct btrfs_block_group_cache
*block_group
;
6931 list_for_each_entry(block_group
, groups_list
, list
) {
6932 spin_lock(&block_group
->lock
);
6934 if (!block_group
->ro
) {
6935 spin_unlock(&block_group
->lock
);
6939 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_RAID1
|
6940 BTRFS_BLOCK_GROUP_RAID10
|
6941 BTRFS_BLOCK_GROUP_DUP
))
6946 free_bytes
+= (block_group
->key
.offset
-
6947 btrfs_block_group_used(&block_group
->item
)) *
6950 spin_unlock(&block_group
->lock
);
6957 * helper to account the unused space of all the readonly block group in the
6958 * space_info. takes mirrors into account.
6960 u64
btrfs_account_ro_block_groups_free_space(struct btrfs_space_info
*sinfo
)
6965 spin_lock(&sinfo
->lock
);
6967 for(i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
6968 if (!list_empty(&sinfo
->block_groups
[i
]))
6969 free_bytes
+= __btrfs_get_ro_block_group_free_space(
6970 &sinfo
->block_groups
[i
]);
6972 spin_unlock(&sinfo
->lock
);
6977 int btrfs_set_block_group_rw(struct btrfs_root
*root
,
6978 struct btrfs_block_group_cache
*cache
)
6980 struct btrfs_space_info
*sinfo
= cache
->space_info
;
6985 spin_lock(&sinfo
->lock
);
6986 spin_lock(&cache
->lock
);
6987 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
6988 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
6989 sinfo
->bytes_readonly
-= num_bytes
;
6991 spin_unlock(&cache
->lock
);
6992 spin_unlock(&sinfo
->lock
);
6997 * checks to see if its even possible to relocate this block group.
6999 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7000 * ok to go ahead and try.
7002 int btrfs_can_relocate(struct btrfs_root
*root
, u64 bytenr
)
7004 struct btrfs_block_group_cache
*block_group
;
7005 struct btrfs_space_info
*space_info
;
7006 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
7007 struct btrfs_device
*device
;
7015 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
7017 /* odd, couldn't find the block group, leave it alone */
7021 min_free
= btrfs_block_group_used(&block_group
->item
);
7023 /* no bytes used, we're good */
7027 space_info
= block_group
->space_info
;
7028 spin_lock(&space_info
->lock
);
7030 full
= space_info
->full
;
7033 * if this is the last block group we have in this space, we can't
7034 * relocate it unless we're able to allocate a new chunk below.
7036 * Otherwise, we need to make sure we have room in the space to handle
7037 * all of the extents from this block group. If we can, we're good
7039 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
7040 (space_info
->bytes_used
+ space_info
->bytes_reserved
+
7041 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
7042 min_free
< space_info
->total_bytes
)) {
7043 spin_unlock(&space_info
->lock
);
7046 spin_unlock(&space_info
->lock
);
7049 * ok we don't have enough space, but maybe we have free space on our
7050 * devices to allocate new chunks for relocation, so loop through our
7051 * alloc devices and guess if we have enough space. However, if we
7052 * were marked as full, then we know there aren't enough chunks, and we
7067 index
= get_block_group_index(block_group
);
7072 } else if (index
== 1) {
7074 } else if (index
== 2) {
7077 } else if (index
== 3) {
7078 dev_min
= fs_devices
->rw_devices
;
7079 do_div(min_free
, dev_min
);
7082 mutex_lock(&root
->fs_info
->chunk_mutex
);
7083 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
7087 * check to make sure we can actually find a chunk with enough
7088 * space to fit our block group in.
7090 if (device
->total_bytes
> device
->bytes_used
+ min_free
) {
7091 ret
= find_free_dev_extent(NULL
, device
, min_free
,
7096 if (dev_nr
>= dev_min
)
7102 mutex_unlock(&root
->fs_info
->chunk_mutex
);
7104 btrfs_put_block_group(block_group
);
7108 static int find_first_block_group(struct btrfs_root
*root
,
7109 struct btrfs_path
*path
, struct btrfs_key
*key
)
7112 struct btrfs_key found_key
;
7113 struct extent_buffer
*leaf
;
7116 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
7121 slot
= path
->slots
[0];
7122 leaf
= path
->nodes
[0];
7123 if (slot
>= btrfs_header_nritems(leaf
)) {
7124 ret
= btrfs_next_leaf(root
, path
);
7131 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
7133 if (found_key
.objectid
>= key
->objectid
&&
7134 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
7144 void btrfs_put_block_group_cache(struct btrfs_fs_info
*info
)
7146 struct btrfs_block_group_cache
*block_group
;
7150 struct inode
*inode
;
7152 block_group
= btrfs_lookup_first_block_group(info
, last
);
7153 while (block_group
) {
7154 spin_lock(&block_group
->lock
);
7155 if (block_group
->iref
)
7157 spin_unlock(&block_group
->lock
);
7158 block_group
= next_block_group(info
->tree_root
,
7168 inode
= block_group
->inode
;
7169 block_group
->iref
= 0;
7170 block_group
->inode
= NULL
;
7171 spin_unlock(&block_group
->lock
);
7173 last
= block_group
->key
.objectid
+ block_group
->key
.offset
;
7174 btrfs_put_block_group(block_group
);
7178 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
7180 struct btrfs_block_group_cache
*block_group
;
7181 struct btrfs_space_info
*space_info
;
7182 struct btrfs_caching_control
*caching_ctl
;
7185 down_write(&info
->extent_commit_sem
);
7186 while (!list_empty(&info
->caching_block_groups
)) {
7187 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
7188 struct btrfs_caching_control
, list
);
7189 list_del(&caching_ctl
->list
);
7190 put_caching_control(caching_ctl
);
7192 up_write(&info
->extent_commit_sem
);
7194 spin_lock(&info
->block_group_cache_lock
);
7195 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
7196 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
7198 rb_erase(&block_group
->cache_node
,
7199 &info
->block_group_cache_tree
);
7200 spin_unlock(&info
->block_group_cache_lock
);
7202 down_write(&block_group
->space_info
->groups_sem
);
7203 list_del(&block_group
->list
);
7204 up_write(&block_group
->space_info
->groups_sem
);
7206 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
7207 wait_block_group_cache_done(block_group
);
7210 * We haven't cached this block group, which means we could
7211 * possibly have excluded extents on this block group.
7213 if (block_group
->cached
== BTRFS_CACHE_NO
)
7214 free_excluded_extents(info
->extent_root
, block_group
);
7216 btrfs_remove_free_space_cache(block_group
);
7217 btrfs_put_block_group(block_group
);
7219 spin_lock(&info
->block_group_cache_lock
);
7221 spin_unlock(&info
->block_group_cache_lock
);
7223 /* now that all the block groups are freed, go through and
7224 * free all the space_info structs. This is only called during
7225 * the final stages of unmount, and so we know nobody is
7226 * using them. We call synchronize_rcu() once before we start,
7227 * just to be on the safe side.
7231 release_global_block_rsv(info
);
7233 while(!list_empty(&info
->space_info
)) {
7234 space_info
= list_entry(info
->space_info
.next
,
7235 struct btrfs_space_info
,
7237 if (space_info
->bytes_pinned
> 0 ||
7238 space_info
->bytes_reserved
> 0 ||
7239 space_info
->bytes_may_use
> 0) {
7241 dump_space_info(space_info
, 0, 0);
7243 list_del(&space_info
->list
);
7249 static void __link_block_group(struct btrfs_space_info
*space_info
,
7250 struct btrfs_block_group_cache
*cache
)
7252 int index
= get_block_group_index(cache
);
7254 down_write(&space_info
->groups_sem
);
7255 list_add_tail(&cache
->list
, &space_info
->block_groups
[index
]);
7256 up_write(&space_info
->groups_sem
);
7259 int btrfs_read_block_groups(struct btrfs_root
*root
)
7261 struct btrfs_path
*path
;
7263 struct btrfs_block_group_cache
*cache
;
7264 struct btrfs_fs_info
*info
= root
->fs_info
;
7265 struct btrfs_space_info
*space_info
;
7266 struct btrfs_key key
;
7267 struct btrfs_key found_key
;
7268 struct extent_buffer
*leaf
;
7272 root
= info
->extent_root
;
7275 btrfs_set_key_type(&key
, BTRFS_BLOCK_GROUP_ITEM_KEY
);
7276 path
= btrfs_alloc_path();
7281 cache_gen
= btrfs_super_cache_generation(root
->fs_info
->super_copy
);
7282 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
7283 btrfs_super_generation(root
->fs_info
->super_copy
) != cache_gen
)
7285 if (btrfs_test_opt(root
, CLEAR_CACHE
))
7289 ret
= find_first_block_group(root
, path
, &key
);
7294 leaf
= path
->nodes
[0];
7295 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
7296 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
7301 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
7303 if (!cache
->free_space_ctl
) {
7309 atomic_set(&cache
->count
, 1);
7310 spin_lock_init(&cache
->lock
);
7311 cache
->fs_info
= info
;
7312 INIT_LIST_HEAD(&cache
->list
);
7313 INIT_LIST_HEAD(&cache
->cluster_list
);
7316 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
7318 read_extent_buffer(leaf
, &cache
->item
,
7319 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
7320 sizeof(cache
->item
));
7321 memcpy(&cache
->key
, &found_key
, sizeof(found_key
));
7323 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
7324 btrfs_release_path(path
);
7325 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
7326 cache
->sectorsize
= root
->sectorsize
;
7328 btrfs_init_free_space_ctl(cache
);
7331 * We need to exclude the super stripes now so that the space
7332 * info has super bytes accounted for, otherwise we'll think
7333 * we have more space than we actually do.
7335 exclude_super_stripes(root
, cache
);
7338 * check for two cases, either we are full, and therefore
7339 * don't need to bother with the caching work since we won't
7340 * find any space, or we are empty, and we can just add all
7341 * the space in and be done with it. This saves us _alot_ of
7342 * time, particularly in the full case.
7344 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
7345 cache
->last_byte_to_unpin
= (u64
)-1;
7346 cache
->cached
= BTRFS_CACHE_FINISHED
;
7347 free_excluded_extents(root
, cache
);
7348 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
7349 cache
->last_byte_to_unpin
= (u64
)-1;
7350 cache
->cached
= BTRFS_CACHE_FINISHED
;
7351 add_new_free_space(cache
, root
->fs_info
,
7353 found_key
.objectid
+
7355 free_excluded_extents(root
, cache
);
7358 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
7359 btrfs_block_group_used(&cache
->item
),
7362 cache
->space_info
= space_info
;
7363 spin_lock(&cache
->space_info
->lock
);
7364 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
7365 spin_unlock(&cache
->space_info
->lock
);
7367 __link_block_group(space_info
, cache
);
7369 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
7372 set_avail_alloc_bits(root
->fs_info
, cache
->flags
);
7373 if (btrfs_chunk_readonly(root
, cache
->key
.objectid
))
7374 set_block_group_ro(cache
, 1);
7377 list_for_each_entry_rcu(space_info
, &root
->fs_info
->space_info
, list
) {
7378 if (!(get_alloc_profile(root
, space_info
->flags
) &
7379 (BTRFS_BLOCK_GROUP_RAID10
|
7380 BTRFS_BLOCK_GROUP_RAID1
|
7381 BTRFS_BLOCK_GROUP_DUP
)))
7384 * avoid allocating from un-mirrored block group if there are
7385 * mirrored block groups.
7387 list_for_each_entry(cache
, &space_info
->block_groups
[3], list
)
7388 set_block_group_ro(cache
, 1);
7389 list_for_each_entry(cache
, &space_info
->block_groups
[4], list
)
7390 set_block_group_ro(cache
, 1);
7393 init_global_block_rsv(info
);
7396 btrfs_free_path(path
);
7400 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
7401 struct btrfs_root
*root
, u64 bytes_used
,
7402 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
7406 struct btrfs_root
*extent_root
;
7407 struct btrfs_block_group_cache
*cache
;
7409 extent_root
= root
->fs_info
->extent_root
;
7411 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
7413 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
7416 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
7418 if (!cache
->free_space_ctl
) {
7423 cache
->key
.objectid
= chunk_offset
;
7424 cache
->key
.offset
= size
;
7425 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
7426 cache
->sectorsize
= root
->sectorsize
;
7427 cache
->fs_info
= root
->fs_info
;
7429 atomic_set(&cache
->count
, 1);
7430 spin_lock_init(&cache
->lock
);
7431 INIT_LIST_HEAD(&cache
->list
);
7432 INIT_LIST_HEAD(&cache
->cluster_list
);
7434 btrfs_init_free_space_ctl(cache
);
7436 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
7437 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
7438 cache
->flags
= type
;
7439 btrfs_set_block_group_flags(&cache
->item
, type
);
7441 cache
->last_byte_to_unpin
= (u64
)-1;
7442 cache
->cached
= BTRFS_CACHE_FINISHED
;
7443 exclude_super_stripes(root
, cache
);
7445 add_new_free_space(cache
, root
->fs_info
, chunk_offset
,
7446 chunk_offset
+ size
);
7448 free_excluded_extents(root
, cache
);
7450 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
7451 &cache
->space_info
);
7454 spin_lock(&cache
->space_info
->lock
);
7455 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
7456 spin_unlock(&cache
->space_info
->lock
);
7458 __link_block_group(cache
->space_info
, cache
);
7460 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
7463 ret
= btrfs_insert_item(trans
, extent_root
, &cache
->key
, &cache
->item
,
7464 sizeof(cache
->item
));
7467 set_avail_alloc_bits(extent_root
->fs_info
, type
);
7472 static void clear_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
7474 u64 extra_flags
= flags
& BTRFS_BLOCK_GROUP_PROFILE_MASK
;
7476 /* chunk -> extended profile */
7477 if (extra_flags
== 0)
7478 extra_flags
= BTRFS_AVAIL_ALLOC_BIT_SINGLE
;
7480 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
7481 fs_info
->avail_data_alloc_bits
&= ~extra_flags
;
7482 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
7483 fs_info
->avail_metadata_alloc_bits
&= ~extra_flags
;
7484 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
7485 fs_info
->avail_system_alloc_bits
&= ~extra_flags
;
7488 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
7489 struct btrfs_root
*root
, u64 group_start
)
7491 struct btrfs_path
*path
;
7492 struct btrfs_block_group_cache
*block_group
;
7493 struct btrfs_free_cluster
*cluster
;
7494 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
7495 struct btrfs_key key
;
7496 struct inode
*inode
;
7501 root
= root
->fs_info
->extent_root
;
7503 block_group
= btrfs_lookup_block_group(root
->fs_info
, group_start
);
7504 BUG_ON(!block_group
);
7505 BUG_ON(!block_group
->ro
);
7508 * Free the reserved super bytes from this block group before
7511 free_excluded_extents(root
, block_group
);
7513 memcpy(&key
, &block_group
->key
, sizeof(key
));
7514 index
= get_block_group_index(block_group
);
7515 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
7516 BTRFS_BLOCK_GROUP_RAID1
|
7517 BTRFS_BLOCK_GROUP_RAID10
))
7522 /* make sure this block group isn't part of an allocation cluster */
7523 cluster
= &root
->fs_info
->data_alloc_cluster
;
7524 spin_lock(&cluster
->refill_lock
);
7525 btrfs_return_cluster_to_free_space(block_group
, cluster
);
7526 spin_unlock(&cluster
->refill_lock
);
7529 * make sure this block group isn't part of a metadata
7530 * allocation cluster
7532 cluster
= &root
->fs_info
->meta_alloc_cluster
;
7533 spin_lock(&cluster
->refill_lock
);
7534 btrfs_return_cluster_to_free_space(block_group
, cluster
);
7535 spin_unlock(&cluster
->refill_lock
);
7537 path
= btrfs_alloc_path();
7543 inode
= lookup_free_space_inode(tree_root
, block_group
, path
);
7544 if (!IS_ERR(inode
)) {
7545 ret
= btrfs_orphan_add(trans
, inode
);
7548 /* One for the block groups ref */
7549 spin_lock(&block_group
->lock
);
7550 if (block_group
->iref
) {
7551 block_group
->iref
= 0;
7552 block_group
->inode
= NULL
;
7553 spin_unlock(&block_group
->lock
);
7556 spin_unlock(&block_group
->lock
);
7558 /* One for our lookup ref */
7559 btrfs_add_delayed_iput(inode
);
7562 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
7563 key
.offset
= block_group
->key
.objectid
;
7566 ret
= btrfs_search_slot(trans
, tree_root
, &key
, path
, -1, 1);
7570 btrfs_release_path(path
);
7572 ret
= btrfs_del_item(trans
, tree_root
, path
);
7575 btrfs_release_path(path
);
7578 spin_lock(&root
->fs_info
->block_group_cache_lock
);
7579 rb_erase(&block_group
->cache_node
,
7580 &root
->fs_info
->block_group_cache_tree
);
7581 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
7583 down_write(&block_group
->space_info
->groups_sem
);
7585 * we must use list_del_init so people can check to see if they
7586 * are still on the list after taking the semaphore
7588 list_del_init(&block_group
->list
);
7589 if (list_empty(&block_group
->space_info
->block_groups
[index
]))
7590 clear_avail_alloc_bits(root
->fs_info
, block_group
->flags
);
7591 up_write(&block_group
->space_info
->groups_sem
);
7593 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
7594 wait_block_group_cache_done(block_group
);
7596 btrfs_remove_free_space_cache(block_group
);
7598 spin_lock(&block_group
->space_info
->lock
);
7599 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
7600 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
7601 block_group
->space_info
->disk_total
-= block_group
->key
.offset
* factor
;
7602 spin_unlock(&block_group
->space_info
->lock
);
7604 memcpy(&key
, &block_group
->key
, sizeof(key
));
7606 btrfs_clear_space_info_full(root
->fs_info
);
7608 btrfs_put_block_group(block_group
);
7609 btrfs_put_block_group(block_group
);
7611 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
7617 ret
= btrfs_del_item(trans
, root
, path
);
7619 btrfs_free_path(path
);
7623 int btrfs_init_space_info(struct btrfs_fs_info
*fs_info
)
7625 struct btrfs_space_info
*space_info
;
7626 struct btrfs_super_block
*disk_super
;
7632 disk_super
= fs_info
->super_copy
;
7633 if (!btrfs_super_root(disk_super
))
7636 features
= btrfs_super_incompat_flags(disk_super
);
7637 if (features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
7640 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
7641 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
7646 flags
= BTRFS_BLOCK_GROUP_METADATA
| BTRFS_BLOCK_GROUP_DATA
;
7647 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
7649 flags
= BTRFS_BLOCK_GROUP_METADATA
;
7650 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
7654 flags
= BTRFS_BLOCK_GROUP_DATA
;
7655 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
7661 int btrfs_error_unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
7663 return unpin_extent_range(root
, start
, end
);
7666 int btrfs_error_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
7667 u64 num_bytes
, u64
*actual_bytes
)
7669 return btrfs_discard_extent(root
, bytenr
, num_bytes
, actual_bytes
);
7672 int btrfs_trim_fs(struct btrfs_root
*root
, struct fstrim_range
*range
)
7674 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7675 struct btrfs_block_group_cache
*cache
= NULL
;
7682 cache
= btrfs_lookup_block_group(fs_info
, range
->start
);
7685 if (cache
->key
.objectid
>= (range
->start
+ range
->len
)) {
7686 btrfs_put_block_group(cache
);
7690 start
= max(range
->start
, cache
->key
.objectid
);
7691 end
= min(range
->start
+ range
->len
,
7692 cache
->key
.objectid
+ cache
->key
.offset
);
7694 if (end
- start
>= range
->minlen
) {
7695 if (!block_group_cache_done(cache
)) {
7696 ret
= cache_block_group(cache
, NULL
, root
, 0);
7698 wait_block_group_cache_done(cache
);
7700 ret
= btrfs_trim_block_group(cache
,
7706 trimmed
+= group_trimmed
;
7708 btrfs_put_block_group(cache
);
7713 cache
= next_block_group(fs_info
->tree_root
, cache
);
7716 range
->len
= trimmed
;