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"
38 #undef SCRAMBLE_DELAYED_REFS
41 * control flags for do_chunk_alloc's force field
42 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
43 * if we really need one.
45 * CHUNK_ALLOC_LIMITED means to only try and allocate one
46 * if we have very few chunks already allocated. This is
47 * used as part of the clustering code to help make sure
48 * we have a good pool of storage to cluster in, without
49 * filling the FS with empty chunks
51 * CHUNK_ALLOC_FORCE means it must try to allocate one
55 CHUNK_ALLOC_NO_FORCE
= 0,
56 CHUNK_ALLOC_LIMITED
= 1,
57 CHUNK_ALLOC_FORCE
= 2,
61 * Control how reservations are dealt with.
63 * RESERVE_FREE - freeing a reservation.
64 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
66 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
67 * bytes_may_use as the ENOSPC accounting is done elsewhere
72 RESERVE_ALLOC_NO_ACCOUNT
= 2,
75 static int update_block_group(struct btrfs_root
*root
,
76 u64 bytenr
, u64 num_bytes
, int alloc
);
77 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
78 struct btrfs_root
*root
,
79 u64 bytenr
, u64 num_bytes
, u64 parent
,
80 u64 root_objectid
, u64 owner_objectid
,
81 u64 owner_offset
, int refs_to_drop
,
82 struct btrfs_delayed_extent_op
*extra_op
);
83 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
84 struct extent_buffer
*leaf
,
85 struct btrfs_extent_item
*ei
);
86 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
87 struct btrfs_root
*root
,
88 u64 parent
, u64 root_objectid
,
89 u64 flags
, u64 owner
, u64 offset
,
90 struct btrfs_key
*ins
, int ref_mod
);
91 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
92 struct btrfs_root
*root
,
93 u64 parent
, u64 root_objectid
,
94 u64 flags
, struct btrfs_disk_key
*key
,
95 int level
, struct btrfs_key
*ins
);
96 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
97 struct btrfs_root
*extent_root
, u64 flags
,
99 static int find_next_key(struct btrfs_path
*path
, int level
,
100 struct btrfs_key
*key
);
101 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
102 int dump_block_groups
);
103 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
104 u64 num_bytes
, int reserve
);
107 block_group_cache_done(struct btrfs_block_group_cache
*cache
)
110 return cache
->cached
== BTRFS_CACHE_FINISHED
;
113 static int block_group_bits(struct btrfs_block_group_cache
*cache
, u64 bits
)
115 return (cache
->flags
& bits
) == bits
;
118 static void btrfs_get_block_group(struct btrfs_block_group_cache
*cache
)
120 atomic_inc(&cache
->count
);
123 void btrfs_put_block_group(struct btrfs_block_group_cache
*cache
)
125 if (atomic_dec_and_test(&cache
->count
)) {
126 WARN_ON(cache
->pinned
> 0);
127 WARN_ON(cache
->reserved
> 0);
128 kfree(cache
->free_space_ctl
);
134 * this adds the block group to the fs_info rb tree for the block group
137 static int btrfs_add_block_group_cache(struct btrfs_fs_info
*info
,
138 struct btrfs_block_group_cache
*block_group
)
141 struct rb_node
*parent
= NULL
;
142 struct btrfs_block_group_cache
*cache
;
144 spin_lock(&info
->block_group_cache_lock
);
145 p
= &info
->block_group_cache_tree
.rb_node
;
149 cache
= rb_entry(parent
, struct btrfs_block_group_cache
,
151 if (block_group
->key
.objectid
< cache
->key
.objectid
) {
153 } else if (block_group
->key
.objectid
> cache
->key
.objectid
) {
156 spin_unlock(&info
->block_group_cache_lock
);
161 rb_link_node(&block_group
->cache_node
, parent
, p
);
162 rb_insert_color(&block_group
->cache_node
,
163 &info
->block_group_cache_tree
);
165 if (info
->first_logical_byte
> block_group
->key
.objectid
)
166 info
->first_logical_byte
= block_group
->key
.objectid
;
168 spin_unlock(&info
->block_group_cache_lock
);
174 * This will return the block group at or after bytenr if contains is 0, else
175 * it will return the block group that contains the bytenr
177 static struct btrfs_block_group_cache
*
178 block_group_cache_tree_search(struct btrfs_fs_info
*info
, u64 bytenr
,
181 struct btrfs_block_group_cache
*cache
, *ret
= NULL
;
185 spin_lock(&info
->block_group_cache_lock
);
186 n
= info
->block_group_cache_tree
.rb_node
;
189 cache
= rb_entry(n
, struct btrfs_block_group_cache
,
191 end
= cache
->key
.objectid
+ cache
->key
.offset
- 1;
192 start
= cache
->key
.objectid
;
194 if (bytenr
< start
) {
195 if (!contains
&& (!ret
|| start
< ret
->key
.objectid
))
198 } else if (bytenr
> start
) {
199 if (contains
&& bytenr
<= end
) {
210 btrfs_get_block_group(ret
);
211 if (bytenr
== 0 && info
->first_logical_byte
> ret
->key
.objectid
)
212 info
->first_logical_byte
= ret
->key
.objectid
;
214 spin_unlock(&info
->block_group_cache_lock
);
219 static int add_excluded_extent(struct btrfs_root
*root
,
220 u64 start
, u64 num_bytes
)
222 u64 end
= start
+ num_bytes
- 1;
223 set_extent_bits(&root
->fs_info
->freed_extents
[0],
224 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
225 set_extent_bits(&root
->fs_info
->freed_extents
[1],
226 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
230 static void free_excluded_extents(struct btrfs_root
*root
,
231 struct btrfs_block_group_cache
*cache
)
235 start
= cache
->key
.objectid
;
236 end
= start
+ cache
->key
.offset
- 1;
238 clear_extent_bits(&root
->fs_info
->freed_extents
[0],
239 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
240 clear_extent_bits(&root
->fs_info
->freed_extents
[1],
241 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
244 static int exclude_super_stripes(struct btrfs_root
*root
,
245 struct btrfs_block_group_cache
*cache
)
252 if (cache
->key
.objectid
< BTRFS_SUPER_INFO_OFFSET
) {
253 stripe_len
= BTRFS_SUPER_INFO_OFFSET
- cache
->key
.objectid
;
254 cache
->bytes_super
+= stripe_len
;
255 ret
= add_excluded_extent(root
, cache
->key
.objectid
,
257 BUG_ON(ret
); /* -ENOMEM */
260 for (i
= 0; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
261 bytenr
= btrfs_sb_offset(i
);
262 ret
= btrfs_rmap_block(&root
->fs_info
->mapping_tree
,
263 cache
->key
.objectid
, bytenr
,
264 0, &logical
, &nr
, &stripe_len
);
265 BUG_ON(ret
); /* -ENOMEM */
268 cache
->bytes_super
+= stripe_len
;
269 ret
= add_excluded_extent(root
, logical
[nr
],
271 BUG_ON(ret
); /* -ENOMEM */
279 static struct btrfs_caching_control
*
280 get_caching_control(struct btrfs_block_group_cache
*cache
)
282 struct btrfs_caching_control
*ctl
;
284 spin_lock(&cache
->lock
);
285 if (cache
->cached
!= BTRFS_CACHE_STARTED
) {
286 spin_unlock(&cache
->lock
);
290 /* We're loading it the fast way, so we don't have a caching_ctl. */
291 if (!cache
->caching_ctl
) {
292 spin_unlock(&cache
->lock
);
296 ctl
= cache
->caching_ctl
;
297 atomic_inc(&ctl
->count
);
298 spin_unlock(&cache
->lock
);
302 static void put_caching_control(struct btrfs_caching_control
*ctl
)
304 if (atomic_dec_and_test(&ctl
->count
))
309 * this is only called by cache_block_group, since we could have freed extents
310 * we need to check the pinned_extents for any extents that can't be used yet
311 * since their free space will be released as soon as the transaction commits.
313 static u64
add_new_free_space(struct btrfs_block_group_cache
*block_group
,
314 struct btrfs_fs_info
*info
, u64 start
, u64 end
)
316 u64 extent_start
, extent_end
, size
, total_added
= 0;
319 while (start
< end
) {
320 ret
= find_first_extent_bit(info
->pinned_extents
, start
,
321 &extent_start
, &extent_end
,
322 EXTENT_DIRTY
| EXTENT_UPTODATE
,
327 if (extent_start
<= start
) {
328 start
= extent_end
+ 1;
329 } else if (extent_start
> start
&& extent_start
< end
) {
330 size
= extent_start
- start
;
332 ret
= btrfs_add_free_space(block_group
, start
,
334 BUG_ON(ret
); /* -ENOMEM or logic error */
335 start
= extent_end
+ 1;
344 ret
= btrfs_add_free_space(block_group
, start
, size
);
345 BUG_ON(ret
); /* -ENOMEM or logic error */
351 static noinline
void caching_thread(struct btrfs_work
*work
)
353 struct btrfs_block_group_cache
*block_group
;
354 struct btrfs_fs_info
*fs_info
;
355 struct btrfs_caching_control
*caching_ctl
;
356 struct btrfs_root
*extent_root
;
357 struct btrfs_path
*path
;
358 struct extent_buffer
*leaf
;
359 struct btrfs_key key
;
365 caching_ctl
= container_of(work
, struct btrfs_caching_control
, work
);
366 block_group
= caching_ctl
->block_group
;
367 fs_info
= block_group
->fs_info
;
368 extent_root
= fs_info
->extent_root
;
370 path
= btrfs_alloc_path();
374 last
= max_t(u64
, block_group
->key
.objectid
, BTRFS_SUPER_INFO_OFFSET
);
377 * We don't want to deadlock with somebody trying to allocate a new
378 * extent for the extent root while also trying to search the extent
379 * root to add free space. So we skip locking and search the commit
380 * root, since its read-only
382 path
->skip_locking
= 1;
383 path
->search_commit_root
= 1;
388 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
390 mutex_lock(&caching_ctl
->mutex
);
391 /* need to make sure the commit_root doesn't disappear */
392 down_read(&fs_info
->extent_commit_sem
);
394 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
398 leaf
= path
->nodes
[0];
399 nritems
= btrfs_header_nritems(leaf
);
402 if (btrfs_fs_closing(fs_info
) > 1) {
407 if (path
->slots
[0] < nritems
) {
408 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
410 ret
= find_next_key(path
, 0, &key
);
414 if (need_resched() ||
415 btrfs_next_leaf(extent_root
, path
)) {
416 caching_ctl
->progress
= last
;
417 btrfs_release_path(path
);
418 up_read(&fs_info
->extent_commit_sem
);
419 mutex_unlock(&caching_ctl
->mutex
);
423 leaf
= path
->nodes
[0];
424 nritems
= btrfs_header_nritems(leaf
);
428 if (key
.objectid
< block_group
->key
.objectid
) {
433 if (key
.objectid
>= block_group
->key
.objectid
+
434 block_group
->key
.offset
)
437 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
) {
438 total_found
+= add_new_free_space(block_group
,
441 last
= key
.objectid
+ key
.offset
;
443 if (total_found
> (1024 * 1024 * 2)) {
445 wake_up(&caching_ctl
->wait
);
452 total_found
+= add_new_free_space(block_group
, fs_info
, last
,
453 block_group
->key
.objectid
+
454 block_group
->key
.offset
);
455 caching_ctl
->progress
= (u64
)-1;
457 spin_lock(&block_group
->lock
);
458 block_group
->caching_ctl
= NULL
;
459 block_group
->cached
= BTRFS_CACHE_FINISHED
;
460 spin_unlock(&block_group
->lock
);
463 btrfs_free_path(path
);
464 up_read(&fs_info
->extent_commit_sem
);
466 free_excluded_extents(extent_root
, block_group
);
468 mutex_unlock(&caching_ctl
->mutex
);
470 wake_up(&caching_ctl
->wait
);
472 put_caching_control(caching_ctl
);
473 btrfs_put_block_group(block_group
);
476 static int cache_block_group(struct btrfs_block_group_cache
*cache
,
480 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
481 struct btrfs_caching_control
*caching_ctl
;
484 caching_ctl
= kzalloc(sizeof(*caching_ctl
), GFP_NOFS
);
488 INIT_LIST_HEAD(&caching_ctl
->list
);
489 mutex_init(&caching_ctl
->mutex
);
490 init_waitqueue_head(&caching_ctl
->wait
);
491 caching_ctl
->block_group
= cache
;
492 caching_ctl
->progress
= cache
->key
.objectid
;
493 atomic_set(&caching_ctl
->count
, 1);
494 caching_ctl
->work
.func
= caching_thread
;
496 spin_lock(&cache
->lock
);
498 * This should be a rare occasion, but this could happen I think in the
499 * case where one thread starts to load the space cache info, and then
500 * some other thread starts a transaction commit which tries to do an
501 * allocation while the other thread is still loading the space cache
502 * info. The previous loop should have kept us from choosing this block
503 * group, but if we've moved to the state where we will wait on caching
504 * block groups we need to first check if we're doing a fast load here,
505 * so we can wait for it to finish, otherwise we could end up allocating
506 * from a block group who's cache gets evicted for one reason or
509 while (cache
->cached
== BTRFS_CACHE_FAST
) {
510 struct btrfs_caching_control
*ctl
;
512 ctl
= cache
->caching_ctl
;
513 atomic_inc(&ctl
->count
);
514 prepare_to_wait(&ctl
->wait
, &wait
, TASK_UNINTERRUPTIBLE
);
515 spin_unlock(&cache
->lock
);
519 finish_wait(&ctl
->wait
, &wait
);
520 put_caching_control(ctl
);
521 spin_lock(&cache
->lock
);
524 if (cache
->cached
!= BTRFS_CACHE_NO
) {
525 spin_unlock(&cache
->lock
);
529 WARN_ON(cache
->caching_ctl
);
530 cache
->caching_ctl
= caching_ctl
;
531 cache
->cached
= BTRFS_CACHE_FAST
;
532 spin_unlock(&cache
->lock
);
534 if (fs_info
->mount_opt
& BTRFS_MOUNT_SPACE_CACHE
) {
535 ret
= load_free_space_cache(fs_info
, cache
);
537 spin_lock(&cache
->lock
);
539 cache
->caching_ctl
= NULL
;
540 cache
->cached
= BTRFS_CACHE_FINISHED
;
541 cache
->last_byte_to_unpin
= (u64
)-1;
543 if (load_cache_only
) {
544 cache
->caching_ctl
= NULL
;
545 cache
->cached
= BTRFS_CACHE_NO
;
547 cache
->cached
= BTRFS_CACHE_STARTED
;
550 spin_unlock(&cache
->lock
);
551 wake_up(&caching_ctl
->wait
);
553 put_caching_control(caching_ctl
);
554 free_excluded_extents(fs_info
->extent_root
, cache
);
559 * We are not going to do the fast caching, set cached to the
560 * appropriate value and wakeup any waiters.
562 spin_lock(&cache
->lock
);
563 if (load_cache_only
) {
564 cache
->caching_ctl
= NULL
;
565 cache
->cached
= BTRFS_CACHE_NO
;
567 cache
->cached
= BTRFS_CACHE_STARTED
;
569 spin_unlock(&cache
->lock
);
570 wake_up(&caching_ctl
->wait
);
573 if (load_cache_only
) {
574 put_caching_control(caching_ctl
);
578 down_write(&fs_info
->extent_commit_sem
);
579 atomic_inc(&caching_ctl
->count
);
580 list_add_tail(&caching_ctl
->list
, &fs_info
->caching_block_groups
);
581 up_write(&fs_info
->extent_commit_sem
);
583 btrfs_get_block_group(cache
);
585 btrfs_queue_worker(&fs_info
->caching_workers
, &caching_ctl
->work
);
591 * return the block group that starts at or after bytenr
593 static struct btrfs_block_group_cache
*
594 btrfs_lookup_first_block_group(struct btrfs_fs_info
*info
, u64 bytenr
)
596 struct btrfs_block_group_cache
*cache
;
598 cache
= block_group_cache_tree_search(info
, bytenr
, 0);
604 * return the block group that contains the given bytenr
606 struct btrfs_block_group_cache
*btrfs_lookup_block_group(
607 struct btrfs_fs_info
*info
,
610 struct btrfs_block_group_cache
*cache
;
612 cache
= block_group_cache_tree_search(info
, bytenr
, 1);
617 static struct btrfs_space_info
*__find_space_info(struct btrfs_fs_info
*info
,
620 struct list_head
*head
= &info
->space_info
;
621 struct btrfs_space_info
*found
;
623 flags
&= BTRFS_BLOCK_GROUP_TYPE_MASK
;
626 list_for_each_entry_rcu(found
, head
, list
) {
627 if (found
->flags
& flags
) {
637 * after adding space to the filesystem, we need to clear the full flags
638 * on all the space infos.
640 void btrfs_clear_space_info_full(struct btrfs_fs_info
*info
)
642 struct list_head
*head
= &info
->space_info
;
643 struct btrfs_space_info
*found
;
646 list_for_each_entry_rcu(found
, head
, list
)
651 u64
btrfs_find_block_group(struct btrfs_root
*root
,
652 u64 search_start
, u64 search_hint
, int owner
)
654 struct btrfs_block_group_cache
*cache
;
656 u64 last
= max(search_hint
, search_start
);
663 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
667 spin_lock(&cache
->lock
);
668 last
= cache
->key
.objectid
+ cache
->key
.offset
;
669 used
= btrfs_block_group_used(&cache
->item
);
671 if ((full_search
|| !cache
->ro
) &&
672 block_group_bits(cache
, BTRFS_BLOCK_GROUP_METADATA
)) {
673 if (used
+ cache
->pinned
+ cache
->reserved
<
674 div_factor(cache
->key
.offset
, factor
)) {
675 group_start
= cache
->key
.objectid
;
676 spin_unlock(&cache
->lock
);
677 btrfs_put_block_group(cache
);
681 spin_unlock(&cache
->lock
);
682 btrfs_put_block_group(cache
);
690 if (!full_search
&& factor
< 10) {
700 /* simple helper to search for an existing extent at a given offset */
701 int btrfs_lookup_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
704 struct btrfs_key key
;
705 struct btrfs_path
*path
;
707 path
= btrfs_alloc_path();
711 key
.objectid
= start
;
713 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
714 ret
= btrfs_search_slot(NULL
, root
->fs_info
->extent_root
, &key
, path
,
716 btrfs_free_path(path
);
721 * helper function to lookup reference count and flags of extent.
723 * the head node for delayed ref is used to store the sum of all the
724 * reference count modifications queued up in the rbtree. the head
725 * node may also store the extent flags to set. This way you can check
726 * to see what the reference count and extent flags would be if all of
727 * the delayed refs are not processed.
729 int btrfs_lookup_extent_info(struct btrfs_trans_handle
*trans
,
730 struct btrfs_root
*root
, u64 bytenr
,
731 u64 num_bytes
, u64
*refs
, u64
*flags
)
733 struct btrfs_delayed_ref_head
*head
;
734 struct btrfs_delayed_ref_root
*delayed_refs
;
735 struct btrfs_path
*path
;
736 struct btrfs_extent_item
*ei
;
737 struct extent_buffer
*leaf
;
738 struct btrfs_key key
;
744 path
= btrfs_alloc_path();
748 key
.objectid
= bytenr
;
749 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
750 key
.offset
= num_bytes
;
752 path
->skip_locking
= 1;
753 path
->search_commit_root
= 1;
756 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
,
762 leaf
= path
->nodes
[0];
763 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
764 if (item_size
>= sizeof(*ei
)) {
765 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
766 struct btrfs_extent_item
);
767 num_refs
= btrfs_extent_refs(leaf
, ei
);
768 extent_flags
= btrfs_extent_flags(leaf
, ei
);
770 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
771 struct btrfs_extent_item_v0
*ei0
;
772 BUG_ON(item_size
!= sizeof(*ei0
));
773 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
774 struct btrfs_extent_item_v0
);
775 num_refs
= btrfs_extent_refs_v0(leaf
, ei0
);
776 /* FIXME: this isn't correct for data */
777 extent_flags
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
782 BUG_ON(num_refs
== 0);
792 delayed_refs
= &trans
->transaction
->delayed_refs
;
793 spin_lock(&delayed_refs
->lock
);
794 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
796 if (!mutex_trylock(&head
->mutex
)) {
797 atomic_inc(&head
->node
.refs
);
798 spin_unlock(&delayed_refs
->lock
);
800 btrfs_release_path(path
);
803 * Mutex was contended, block until it's released and try
806 mutex_lock(&head
->mutex
);
807 mutex_unlock(&head
->mutex
);
808 btrfs_put_delayed_ref(&head
->node
);
811 if (head
->extent_op
&& head
->extent_op
->update_flags
)
812 extent_flags
|= head
->extent_op
->flags_to_set
;
814 BUG_ON(num_refs
== 0);
816 num_refs
+= head
->node
.ref_mod
;
817 mutex_unlock(&head
->mutex
);
819 spin_unlock(&delayed_refs
->lock
);
821 WARN_ON(num_refs
== 0);
825 *flags
= extent_flags
;
827 btrfs_free_path(path
);
832 * Back reference rules. Back refs have three main goals:
834 * 1) differentiate between all holders of references to an extent so that
835 * when a reference is dropped we can make sure it was a valid reference
836 * before freeing the extent.
838 * 2) Provide enough information to quickly find the holders of an extent
839 * if we notice a given block is corrupted or bad.
841 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
842 * maintenance. This is actually the same as #2, but with a slightly
843 * different use case.
845 * There are two kinds of back refs. The implicit back refs is optimized
846 * for pointers in non-shared tree blocks. For a given pointer in a block,
847 * back refs of this kind provide information about the block's owner tree
848 * and the pointer's key. These information allow us to find the block by
849 * b-tree searching. The full back refs is for pointers in tree blocks not
850 * referenced by their owner trees. The location of tree block is recorded
851 * in the back refs. Actually the full back refs is generic, and can be
852 * used in all cases the implicit back refs is used. The major shortcoming
853 * of the full back refs is its overhead. Every time a tree block gets
854 * COWed, we have to update back refs entry for all pointers in it.
856 * For a newly allocated tree block, we use implicit back refs for
857 * pointers in it. This means most tree related operations only involve
858 * implicit back refs. For a tree block created in old transaction, the
859 * only way to drop a reference to it is COW it. So we can detect the
860 * event that tree block loses its owner tree's reference and do the
861 * back refs conversion.
863 * When a tree block is COW'd through a tree, there are four cases:
865 * The reference count of the block is one and the tree is the block's
866 * owner tree. Nothing to do in this case.
868 * The reference count of the block is one and the tree is not the
869 * block's owner tree. In this case, full back refs is used for pointers
870 * in the block. Remove these full back refs, add implicit back refs for
871 * every pointers in the new block.
873 * The reference count of the block is greater than one and the tree is
874 * the block's owner tree. In this case, implicit back refs is used for
875 * pointers in the block. Add full back refs for every pointers in the
876 * block, increase lower level extents' reference counts. The original
877 * implicit back refs are entailed to the new block.
879 * The reference count of the block is greater than one and the tree is
880 * not the block's owner tree. Add implicit back refs for every pointer in
881 * the new block, increase lower level extents' reference count.
883 * Back Reference Key composing:
885 * The key objectid corresponds to the first byte in the extent,
886 * The key type is used to differentiate between types of back refs.
887 * There are different meanings of the key offset for different types
890 * File extents can be referenced by:
892 * - multiple snapshots, subvolumes, or different generations in one subvol
893 * - different files inside a single subvolume
894 * - different offsets inside a file (bookend extents in file.c)
896 * The extent ref structure for the implicit back refs has fields for:
898 * - Objectid of the subvolume root
899 * - objectid of the file holding the reference
900 * - original offset in the file
901 * - how many bookend extents
903 * The key offset for the implicit back refs is hash of the first
906 * The extent ref structure for the full back refs has field for:
908 * - number of pointers in the tree leaf
910 * The key offset for the implicit back refs is the first byte of
913 * When a file extent is allocated, The implicit back refs is used.
914 * the fields are filled in:
916 * (root_key.objectid, inode objectid, offset in file, 1)
918 * When a file extent is removed file truncation, we find the
919 * corresponding implicit back refs and check the following fields:
921 * (btrfs_header_owner(leaf), inode objectid, offset in file)
923 * Btree extents can be referenced by:
925 * - Different subvolumes
927 * Both the implicit back refs and the full back refs for tree blocks
928 * only consist of key. The key offset for the implicit back refs is
929 * objectid of block's owner tree. The key offset for the full back refs
930 * is the first byte of parent block.
932 * When implicit back refs is used, information about the lowest key and
933 * level of the tree block are required. These information are stored in
934 * tree block info structure.
937 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
938 static int convert_extent_item_v0(struct btrfs_trans_handle
*trans
,
939 struct btrfs_root
*root
,
940 struct btrfs_path
*path
,
941 u64 owner
, u32 extra_size
)
943 struct btrfs_extent_item
*item
;
944 struct btrfs_extent_item_v0
*ei0
;
945 struct btrfs_extent_ref_v0
*ref0
;
946 struct btrfs_tree_block_info
*bi
;
947 struct extent_buffer
*leaf
;
948 struct btrfs_key key
;
949 struct btrfs_key found_key
;
950 u32 new_size
= sizeof(*item
);
954 leaf
= path
->nodes
[0];
955 BUG_ON(btrfs_item_size_nr(leaf
, path
->slots
[0]) != sizeof(*ei0
));
957 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
958 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
959 struct btrfs_extent_item_v0
);
960 refs
= btrfs_extent_refs_v0(leaf
, ei0
);
962 if (owner
== (u64
)-1) {
964 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
965 ret
= btrfs_next_leaf(root
, path
);
968 BUG_ON(ret
> 0); /* Corruption */
969 leaf
= path
->nodes
[0];
971 btrfs_item_key_to_cpu(leaf
, &found_key
,
973 BUG_ON(key
.objectid
!= found_key
.objectid
);
974 if (found_key
.type
!= BTRFS_EXTENT_REF_V0_KEY
) {
978 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
979 struct btrfs_extent_ref_v0
);
980 owner
= btrfs_ref_objectid_v0(leaf
, ref0
);
984 btrfs_release_path(path
);
986 if (owner
< BTRFS_FIRST_FREE_OBJECTID
)
987 new_size
+= sizeof(*bi
);
989 new_size
-= sizeof(*ei0
);
990 ret
= btrfs_search_slot(trans
, root
, &key
, path
,
991 new_size
+ extra_size
, 1);
994 BUG_ON(ret
); /* Corruption */
996 btrfs_extend_item(trans
, root
, path
, new_size
);
998 leaf
= path
->nodes
[0];
999 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1000 btrfs_set_extent_refs(leaf
, item
, refs
);
1001 /* FIXME: get real generation */
1002 btrfs_set_extent_generation(leaf
, item
, 0);
1003 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1004 btrfs_set_extent_flags(leaf
, item
,
1005 BTRFS_EXTENT_FLAG_TREE_BLOCK
|
1006 BTRFS_BLOCK_FLAG_FULL_BACKREF
);
1007 bi
= (struct btrfs_tree_block_info
*)(item
+ 1);
1008 /* FIXME: get first key of the block */
1009 memset_extent_buffer(leaf
, 0, (unsigned long)bi
, sizeof(*bi
));
1010 btrfs_set_tree_block_level(leaf
, bi
, (int)owner
);
1012 btrfs_set_extent_flags(leaf
, item
, BTRFS_EXTENT_FLAG_DATA
);
1014 btrfs_mark_buffer_dirty(leaf
);
1019 static u64
hash_extent_data_ref(u64 root_objectid
, u64 owner
, u64 offset
)
1021 u32 high_crc
= ~(u32
)0;
1022 u32 low_crc
= ~(u32
)0;
1025 lenum
= cpu_to_le64(root_objectid
);
1026 high_crc
= crc32c(high_crc
, &lenum
, sizeof(lenum
));
1027 lenum
= cpu_to_le64(owner
);
1028 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
1029 lenum
= cpu_to_le64(offset
);
1030 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
1032 return ((u64
)high_crc
<< 31) ^ (u64
)low_crc
;
1035 static u64
hash_extent_data_ref_item(struct extent_buffer
*leaf
,
1036 struct btrfs_extent_data_ref
*ref
)
1038 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf
, ref
),
1039 btrfs_extent_data_ref_objectid(leaf
, ref
),
1040 btrfs_extent_data_ref_offset(leaf
, ref
));
1043 static int match_extent_data_ref(struct extent_buffer
*leaf
,
1044 struct btrfs_extent_data_ref
*ref
,
1045 u64 root_objectid
, u64 owner
, u64 offset
)
1047 if (btrfs_extent_data_ref_root(leaf
, ref
) != root_objectid
||
1048 btrfs_extent_data_ref_objectid(leaf
, ref
) != owner
||
1049 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
1054 static noinline
int lookup_extent_data_ref(struct btrfs_trans_handle
*trans
,
1055 struct btrfs_root
*root
,
1056 struct btrfs_path
*path
,
1057 u64 bytenr
, u64 parent
,
1059 u64 owner
, u64 offset
)
1061 struct btrfs_key key
;
1062 struct btrfs_extent_data_ref
*ref
;
1063 struct extent_buffer
*leaf
;
1069 key
.objectid
= bytenr
;
1071 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1072 key
.offset
= parent
;
1074 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1075 key
.offset
= hash_extent_data_ref(root_objectid
,
1080 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1089 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1090 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1091 btrfs_release_path(path
);
1092 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1103 leaf
= path
->nodes
[0];
1104 nritems
= btrfs_header_nritems(leaf
);
1106 if (path
->slots
[0] >= nritems
) {
1107 ret
= btrfs_next_leaf(root
, path
);
1113 leaf
= path
->nodes
[0];
1114 nritems
= btrfs_header_nritems(leaf
);
1118 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1119 if (key
.objectid
!= bytenr
||
1120 key
.type
!= BTRFS_EXTENT_DATA_REF_KEY
)
1123 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1124 struct btrfs_extent_data_ref
);
1126 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1129 btrfs_release_path(path
);
1141 static noinline
int insert_extent_data_ref(struct btrfs_trans_handle
*trans
,
1142 struct btrfs_root
*root
,
1143 struct btrfs_path
*path
,
1144 u64 bytenr
, u64 parent
,
1145 u64 root_objectid
, u64 owner
,
1146 u64 offset
, int refs_to_add
)
1148 struct btrfs_key key
;
1149 struct extent_buffer
*leaf
;
1154 key
.objectid
= bytenr
;
1156 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1157 key
.offset
= parent
;
1158 size
= sizeof(struct btrfs_shared_data_ref
);
1160 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1161 key
.offset
= hash_extent_data_ref(root_objectid
,
1163 size
= sizeof(struct btrfs_extent_data_ref
);
1166 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, size
);
1167 if (ret
&& ret
!= -EEXIST
)
1170 leaf
= path
->nodes
[0];
1172 struct btrfs_shared_data_ref
*ref
;
1173 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1174 struct btrfs_shared_data_ref
);
1176 btrfs_set_shared_data_ref_count(leaf
, ref
, refs_to_add
);
1178 num_refs
= btrfs_shared_data_ref_count(leaf
, ref
);
1179 num_refs
+= refs_to_add
;
1180 btrfs_set_shared_data_ref_count(leaf
, ref
, num_refs
);
1183 struct btrfs_extent_data_ref
*ref
;
1184 while (ret
== -EEXIST
) {
1185 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1186 struct btrfs_extent_data_ref
);
1187 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1190 btrfs_release_path(path
);
1192 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1194 if (ret
&& ret
!= -EEXIST
)
1197 leaf
= path
->nodes
[0];
1199 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1200 struct btrfs_extent_data_ref
);
1202 btrfs_set_extent_data_ref_root(leaf
, ref
,
1204 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
1205 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
1206 btrfs_set_extent_data_ref_count(leaf
, ref
, refs_to_add
);
1208 num_refs
= btrfs_extent_data_ref_count(leaf
, ref
);
1209 num_refs
+= refs_to_add
;
1210 btrfs_set_extent_data_ref_count(leaf
, ref
, num_refs
);
1213 btrfs_mark_buffer_dirty(leaf
);
1216 btrfs_release_path(path
);
1220 static noinline
int remove_extent_data_ref(struct btrfs_trans_handle
*trans
,
1221 struct btrfs_root
*root
,
1222 struct btrfs_path
*path
,
1225 struct btrfs_key key
;
1226 struct btrfs_extent_data_ref
*ref1
= NULL
;
1227 struct btrfs_shared_data_ref
*ref2
= NULL
;
1228 struct extent_buffer
*leaf
;
1232 leaf
= path
->nodes
[0];
1233 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1235 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1236 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1237 struct btrfs_extent_data_ref
);
1238 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1239 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1240 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1241 struct btrfs_shared_data_ref
);
1242 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1243 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1244 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1245 struct btrfs_extent_ref_v0
*ref0
;
1246 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1247 struct btrfs_extent_ref_v0
);
1248 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1254 BUG_ON(num_refs
< refs_to_drop
);
1255 num_refs
-= refs_to_drop
;
1257 if (num_refs
== 0) {
1258 ret
= btrfs_del_item(trans
, root
, path
);
1260 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
)
1261 btrfs_set_extent_data_ref_count(leaf
, ref1
, num_refs
);
1262 else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
)
1263 btrfs_set_shared_data_ref_count(leaf
, ref2
, num_refs
);
1264 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1266 struct btrfs_extent_ref_v0
*ref0
;
1267 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1268 struct btrfs_extent_ref_v0
);
1269 btrfs_set_ref_count_v0(leaf
, ref0
, num_refs
);
1272 btrfs_mark_buffer_dirty(leaf
);
1277 static noinline u32
extent_data_ref_count(struct btrfs_root
*root
,
1278 struct btrfs_path
*path
,
1279 struct btrfs_extent_inline_ref
*iref
)
1281 struct btrfs_key key
;
1282 struct extent_buffer
*leaf
;
1283 struct btrfs_extent_data_ref
*ref1
;
1284 struct btrfs_shared_data_ref
*ref2
;
1287 leaf
= path
->nodes
[0];
1288 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1290 if (btrfs_extent_inline_ref_type(leaf
, iref
) ==
1291 BTRFS_EXTENT_DATA_REF_KEY
) {
1292 ref1
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1293 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1295 ref2
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1296 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1298 } else if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1299 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1300 struct btrfs_extent_data_ref
);
1301 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1302 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1303 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1304 struct btrfs_shared_data_ref
);
1305 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1306 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1307 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1308 struct btrfs_extent_ref_v0
*ref0
;
1309 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1310 struct btrfs_extent_ref_v0
);
1311 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1319 static noinline
int lookup_tree_block_ref(struct btrfs_trans_handle
*trans
,
1320 struct btrfs_root
*root
,
1321 struct btrfs_path
*path
,
1322 u64 bytenr
, u64 parent
,
1325 struct btrfs_key key
;
1328 key
.objectid
= bytenr
;
1330 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1331 key
.offset
= parent
;
1333 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1334 key
.offset
= root_objectid
;
1337 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1340 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1341 if (ret
== -ENOENT
&& parent
) {
1342 btrfs_release_path(path
);
1343 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1344 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1352 static noinline
int insert_tree_block_ref(struct btrfs_trans_handle
*trans
,
1353 struct btrfs_root
*root
,
1354 struct btrfs_path
*path
,
1355 u64 bytenr
, u64 parent
,
1358 struct btrfs_key key
;
1361 key
.objectid
= bytenr
;
1363 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1364 key
.offset
= parent
;
1366 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1367 key
.offset
= root_objectid
;
1370 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
1371 btrfs_release_path(path
);
1375 static inline int extent_ref_type(u64 parent
, u64 owner
)
1378 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1380 type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1382 type
= BTRFS_TREE_BLOCK_REF_KEY
;
1385 type
= BTRFS_SHARED_DATA_REF_KEY
;
1387 type
= BTRFS_EXTENT_DATA_REF_KEY
;
1392 static int find_next_key(struct btrfs_path
*path
, int level
,
1393 struct btrfs_key
*key
)
1396 for (; level
< BTRFS_MAX_LEVEL
; level
++) {
1397 if (!path
->nodes
[level
])
1399 if (path
->slots
[level
] + 1 >=
1400 btrfs_header_nritems(path
->nodes
[level
]))
1403 btrfs_item_key_to_cpu(path
->nodes
[level
], key
,
1404 path
->slots
[level
] + 1);
1406 btrfs_node_key_to_cpu(path
->nodes
[level
], key
,
1407 path
->slots
[level
] + 1);
1414 * look for inline back ref. if back ref is found, *ref_ret is set
1415 * to the address of inline back ref, and 0 is returned.
1417 * if back ref isn't found, *ref_ret is set to the address where it
1418 * should be inserted, and -ENOENT is returned.
1420 * if insert is true and there are too many inline back refs, the path
1421 * points to the extent item, and -EAGAIN is returned.
1423 * NOTE: inline back refs are ordered in the same way that back ref
1424 * items in the tree are ordered.
1426 static noinline_for_stack
1427 int lookup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1428 struct btrfs_root
*root
,
1429 struct btrfs_path
*path
,
1430 struct btrfs_extent_inline_ref
**ref_ret
,
1431 u64 bytenr
, u64 num_bytes
,
1432 u64 parent
, u64 root_objectid
,
1433 u64 owner
, u64 offset
, int insert
)
1435 struct btrfs_key key
;
1436 struct extent_buffer
*leaf
;
1437 struct btrfs_extent_item
*ei
;
1438 struct btrfs_extent_inline_ref
*iref
;
1449 key
.objectid
= bytenr
;
1450 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1451 key
.offset
= num_bytes
;
1453 want
= extent_ref_type(parent
, owner
);
1455 extra_size
= btrfs_extent_inline_ref_size(want
);
1456 path
->keep_locks
= 1;
1459 ret
= btrfs_search_slot(trans
, root
, &key
, path
, extra_size
, 1);
1464 if (ret
&& !insert
) {
1468 BUG_ON(ret
); /* Corruption */
1470 leaf
= path
->nodes
[0];
1471 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1472 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1473 if (item_size
< sizeof(*ei
)) {
1478 ret
= convert_extent_item_v0(trans
, root
, path
, owner
,
1484 leaf
= path
->nodes
[0];
1485 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1488 BUG_ON(item_size
< sizeof(*ei
));
1490 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1491 flags
= btrfs_extent_flags(leaf
, ei
);
1493 ptr
= (unsigned long)(ei
+ 1);
1494 end
= (unsigned long)ei
+ item_size
;
1496 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
) {
1497 ptr
+= sizeof(struct btrfs_tree_block_info
);
1500 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_DATA
));
1509 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1510 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1514 ptr
+= btrfs_extent_inline_ref_size(type
);
1518 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1519 struct btrfs_extent_data_ref
*dref
;
1520 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1521 if (match_extent_data_ref(leaf
, dref
, root_objectid
,
1526 if (hash_extent_data_ref_item(leaf
, dref
) <
1527 hash_extent_data_ref(root_objectid
, owner
, offset
))
1531 ref_offset
= btrfs_extent_inline_ref_offset(leaf
, iref
);
1533 if (parent
== ref_offset
) {
1537 if (ref_offset
< parent
)
1540 if (root_objectid
== ref_offset
) {
1544 if (ref_offset
< root_objectid
)
1548 ptr
+= btrfs_extent_inline_ref_size(type
);
1550 if (err
== -ENOENT
&& insert
) {
1551 if (item_size
+ extra_size
>=
1552 BTRFS_MAX_EXTENT_ITEM_SIZE(root
)) {
1557 * To add new inline back ref, we have to make sure
1558 * there is no corresponding back ref item.
1559 * For simplicity, we just do not add new inline back
1560 * ref if there is any kind of item for this block
1562 if (find_next_key(path
, 0, &key
) == 0 &&
1563 key
.objectid
== bytenr
&&
1564 key
.type
< BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1569 *ref_ret
= (struct btrfs_extent_inline_ref
*)ptr
;
1572 path
->keep_locks
= 0;
1573 btrfs_unlock_up_safe(path
, 1);
1579 * helper to add new inline back ref
1581 static noinline_for_stack
1582 void setup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1583 struct btrfs_root
*root
,
1584 struct btrfs_path
*path
,
1585 struct btrfs_extent_inline_ref
*iref
,
1586 u64 parent
, u64 root_objectid
,
1587 u64 owner
, u64 offset
, int refs_to_add
,
1588 struct btrfs_delayed_extent_op
*extent_op
)
1590 struct extent_buffer
*leaf
;
1591 struct btrfs_extent_item
*ei
;
1594 unsigned long item_offset
;
1599 leaf
= path
->nodes
[0];
1600 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1601 item_offset
= (unsigned long)iref
- (unsigned long)ei
;
1603 type
= extent_ref_type(parent
, owner
);
1604 size
= btrfs_extent_inline_ref_size(type
);
1606 btrfs_extend_item(trans
, root
, path
, size
);
1608 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1609 refs
= btrfs_extent_refs(leaf
, ei
);
1610 refs
+= refs_to_add
;
1611 btrfs_set_extent_refs(leaf
, ei
, refs
);
1613 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1615 ptr
= (unsigned long)ei
+ item_offset
;
1616 end
= (unsigned long)ei
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
1617 if (ptr
< end
- size
)
1618 memmove_extent_buffer(leaf
, ptr
+ size
, ptr
,
1621 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1622 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
1623 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1624 struct btrfs_extent_data_ref
*dref
;
1625 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1626 btrfs_set_extent_data_ref_root(leaf
, dref
, root_objectid
);
1627 btrfs_set_extent_data_ref_objectid(leaf
, dref
, owner
);
1628 btrfs_set_extent_data_ref_offset(leaf
, dref
, offset
);
1629 btrfs_set_extent_data_ref_count(leaf
, dref
, refs_to_add
);
1630 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1631 struct btrfs_shared_data_ref
*sref
;
1632 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1633 btrfs_set_shared_data_ref_count(leaf
, sref
, refs_to_add
);
1634 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1635 } else if (type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
1636 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1638 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
1640 btrfs_mark_buffer_dirty(leaf
);
1643 static int lookup_extent_backref(struct btrfs_trans_handle
*trans
,
1644 struct btrfs_root
*root
,
1645 struct btrfs_path
*path
,
1646 struct btrfs_extent_inline_ref
**ref_ret
,
1647 u64 bytenr
, u64 num_bytes
, u64 parent
,
1648 u64 root_objectid
, u64 owner
, u64 offset
)
1652 ret
= lookup_inline_extent_backref(trans
, root
, path
, ref_ret
,
1653 bytenr
, num_bytes
, parent
,
1654 root_objectid
, owner
, offset
, 0);
1658 btrfs_release_path(path
);
1661 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1662 ret
= lookup_tree_block_ref(trans
, root
, path
, bytenr
, parent
,
1665 ret
= lookup_extent_data_ref(trans
, root
, path
, bytenr
, parent
,
1666 root_objectid
, owner
, offset
);
1672 * helper to update/remove inline back ref
1674 static noinline_for_stack
1675 void update_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1676 struct btrfs_root
*root
,
1677 struct btrfs_path
*path
,
1678 struct btrfs_extent_inline_ref
*iref
,
1680 struct btrfs_delayed_extent_op
*extent_op
)
1682 struct extent_buffer
*leaf
;
1683 struct btrfs_extent_item
*ei
;
1684 struct btrfs_extent_data_ref
*dref
= NULL
;
1685 struct btrfs_shared_data_ref
*sref
= NULL
;
1693 leaf
= path
->nodes
[0];
1694 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1695 refs
= btrfs_extent_refs(leaf
, ei
);
1696 WARN_ON(refs_to_mod
< 0 && refs
+ refs_to_mod
<= 0);
1697 refs
+= refs_to_mod
;
1698 btrfs_set_extent_refs(leaf
, ei
, refs
);
1700 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1702 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1704 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1705 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1706 refs
= btrfs_extent_data_ref_count(leaf
, dref
);
1707 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1708 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1709 refs
= btrfs_shared_data_ref_count(leaf
, sref
);
1712 BUG_ON(refs_to_mod
!= -1);
1715 BUG_ON(refs_to_mod
< 0 && refs
< -refs_to_mod
);
1716 refs
+= refs_to_mod
;
1719 if (type
== BTRFS_EXTENT_DATA_REF_KEY
)
1720 btrfs_set_extent_data_ref_count(leaf
, dref
, refs
);
1722 btrfs_set_shared_data_ref_count(leaf
, sref
, refs
);
1724 size
= btrfs_extent_inline_ref_size(type
);
1725 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1726 ptr
= (unsigned long)iref
;
1727 end
= (unsigned long)ei
+ item_size
;
1728 if (ptr
+ size
< end
)
1729 memmove_extent_buffer(leaf
, ptr
, ptr
+ size
,
1732 btrfs_truncate_item(trans
, root
, path
, item_size
, 1);
1734 btrfs_mark_buffer_dirty(leaf
);
1737 static noinline_for_stack
1738 int insert_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1739 struct btrfs_root
*root
,
1740 struct btrfs_path
*path
,
1741 u64 bytenr
, u64 num_bytes
, u64 parent
,
1742 u64 root_objectid
, u64 owner
,
1743 u64 offset
, int refs_to_add
,
1744 struct btrfs_delayed_extent_op
*extent_op
)
1746 struct btrfs_extent_inline_ref
*iref
;
1749 ret
= lookup_inline_extent_backref(trans
, root
, path
, &iref
,
1750 bytenr
, num_bytes
, parent
,
1751 root_objectid
, owner
, offset
, 1);
1753 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
);
1754 update_inline_extent_backref(trans
, root
, path
, iref
,
1755 refs_to_add
, extent_op
);
1756 } else if (ret
== -ENOENT
) {
1757 setup_inline_extent_backref(trans
, root
, path
, iref
, parent
,
1758 root_objectid
, owner
, offset
,
1759 refs_to_add
, extent_op
);
1765 static int insert_extent_backref(struct btrfs_trans_handle
*trans
,
1766 struct btrfs_root
*root
,
1767 struct btrfs_path
*path
,
1768 u64 bytenr
, u64 parent
, u64 root_objectid
,
1769 u64 owner
, u64 offset
, int refs_to_add
)
1772 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1773 BUG_ON(refs_to_add
!= 1);
1774 ret
= insert_tree_block_ref(trans
, root
, path
, bytenr
,
1775 parent
, root_objectid
);
1777 ret
= insert_extent_data_ref(trans
, root
, path
, bytenr
,
1778 parent
, root_objectid
,
1779 owner
, offset
, refs_to_add
);
1784 static int remove_extent_backref(struct btrfs_trans_handle
*trans
,
1785 struct btrfs_root
*root
,
1786 struct btrfs_path
*path
,
1787 struct btrfs_extent_inline_ref
*iref
,
1788 int refs_to_drop
, int is_data
)
1792 BUG_ON(!is_data
&& refs_to_drop
!= 1);
1794 update_inline_extent_backref(trans
, root
, path
, iref
,
1795 -refs_to_drop
, NULL
);
1796 } else if (is_data
) {
1797 ret
= remove_extent_data_ref(trans
, root
, path
, refs_to_drop
);
1799 ret
= btrfs_del_item(trans
, root
, path
);
1804 static int btrfs_issue_discard(struct block_device
*bdev
,
1807 return blkdev_issue_discard(bdev
, start
>> 9, len
>> 9, GFP_NOFS
, 0);
1810 static int btrfs_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
1811 u64 num_bytes
, u64
*actual_bytes
)
1814 u64 discarded_bytes
= 0;
1815 struct btrfs_bio
*bbio
= NULL
;
1818 /* Tell the block device(s) that the sectors can be discarded */
1819 ret
= btrfs_map_block(root
->fs_info
, REQ_DISCARD
,
1820 bytenr
, &num_bytes
, &bbio
, 0);
1821 /* Error condition is -ENOMEM */
1823 struct btrfs_bio_stripe
*stripe
= bbio
->stripes
;
1827 for (i
= 0; i
< bbio
->num_stripes
; i
++, stripe
++) {
1828 if (!stripe
->dev
->can_discard
)
1831 ret
= btrfs_issue_discard(stripe
->dev
->bdev
,
1835 discarded_bytes
+= stripe
->length
;
1836 else if (ret
!= -EOPNOTSUPP
)
1837 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1840 * Just in case we get back EOPNOTSUPP for some reason,
1841 * just ignore the return value so we don't screw up
1842 * people calling discard_extent.
1850 *actual_bytes
= discarded_bytes
;
1856 /* Can return -ENOMEM */
1857 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1858 struct btrfs_root
*root
,
1859 u64 bytenr
, u64 num_bytes
, u64 parent
,
1860 u64 root_objectid
, u64 owner
, u64 offset
, int for_cow
)
1863 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1865 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
&&
1866 root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
1868 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1869 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
1871 parent
, root_objectid
, (int)owner
,
1872 BTRFS_ADD_DELAYED_REF
, NULL
, for_cow
);
1874 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
1876 parent
, root_objectid
, owner
, offset
,
1877 BTRFS_ADD_DELAYED_REF
, NULL
, for_cow
);
1882 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1883 struct btrfs_root
*root
,
1884 u64 bytenr
, u64 num_bytes
,
1885 u64 parent
, u64 root_objectid
,
1886 u64 owner
, u64 offset
, int refs_to_add
,
1887 struct btrfs_delayed_extent_op
*extent_op
)
1889 struct btrfs_path
*path
;
1890 struct extent_buffer
*leaf
;
1891 struct btrfs_extent_item
*item
;
1896 path
= btrfs_alloc_path();
1901 path
->leave_spinning
= 1;
1902 /* this will setup the path even if it fails to insert the back ref */
1903 ret
= insert_inline_extent_backref(trans
, root
->fs_info
->extent_root
,
1904 path
, bytenr
, num_bytes
, parent
,
1905 root_objectid
, owner
, offset
,
1906 refs_to_add
, extent_op
);
1910 if (ret
!= -EAGAIN
) {
1915 leaf
= path
->nodes
[0];
1916 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1917 refs
= btrfs_extent_refs(leaf
, item
);
1918 btrfs_set_extent_refs(leaf
, item
, refs
+ refs_to_add
);
1920 __run_delayed_extent_op(extent_op
, leaf
, item
);
1922 btrfs_mark_buffer_dirty(leaf
);
1923 btrfs_release_path(path
);
1926 path
->leave_spinning
= 1;
1928 /* now insert the actual backref */
1929 ret
= insert_extent_backref(trans
, root
->fs_info
->extent_root
,
1930 path
, bytenr
, parent
, root_objectid
,
1931 owner
, offset
, refs_to_add
);
1933 btrfs_abort_transaction(trans
, root
, ret
);
1935 btrfs_free_path(path
);
1939 static int run_delayed_data_ref(struct btrfs_trans_handle
*trans
,
1940 struct btrfs_root
*root
,
1941 struct btrfs_delayed_ref_node
*node
,
1942 struct btrfs_delayed_extent_op
*extent_op
,
1943 int insert_reserved
)
1946 struct btrfs_delayed_data_ref
*ref
;
1947 struct btrfs_key ins
;
1952 ins
.objectid
= node
->bytenr
;
1953 ins
.offset
= node
->num_bytes
;
1954 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1956 ref
= btrfs_delayed_node_to_data_ref(node
);
1957 if (node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
1958 parent
= ref
->parent
;
1960 ref_root
= ref
->root
;
1962 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
1964 BUG_ON(extent_op
->update_key
);
1965 flags
|= extent_op
->flags_to_set
;
1967 ret
= alloc_reserved_file_extent(trans
, root
,
1968 parent
, ref_root
, flags
,
1969 ref
->objectid
, ref
->offset
,
1970 &ins
, node
->ref_mod
);
1971 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
1972 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
1973 node
->num_bytes
, parent
,
1974 ref_root
, ref
->objectid
,
1975 ref
->offset
, node
->ref_mod
,
1977 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
1978 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
1979 node
->num_bytes
, parent
,
1980 ref_root
, ref
->objectid
,
1981 ref
->offset
, node
->ref_mod
,
1989 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
1990 struct extent_buffer
*leaf
,
1991 struct btrfs_extent_item
*ei
)
1993 u64 flags
= btrfs_extent_flags(leaf
, ei
);
1994 if (extent_op
->update_flags
) {
1995 flags
|= extent_op
->flags_to_set
;
1996 btrfs_set_extent_flags(leaf
, ei
, flags
);
1999 if (extent_op
->update_key
) {
2000 struct btrfs_tree_block_info
*bi
;
2001 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
));
2002 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
2003 btrfs_set_tree_block_key(leaf
, bi
, &extent_op
->key
);
2007 static int run_delayed_extent_op(struct btrfs_trans_handle
*trans
,
2008 struct btrfs_root
*root
,
2009 struct btrfs_delayed_ref_node
*node
,
2010 struct btrfs_delayed_extent_op
*extent_op
)
2012 struct btrfs_key key
;
2013 struct btrfs_path
*path
;
2014 struct btrfs_extent_item
*ei
;
2015 struct extent_buffer
*leaf
;
2023 path
= btrfs_alloc_path();
2027 key
.objectid
= node
->bytenr
;
2028 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2029 key
.offset
= node
->num_bytes
;
2032 path
->leave_spinning
= 1;
2033 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
,
2044 leaf
= path
->nodes
[0];
2045 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2046 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2047 if (item_size
< sizeof(*ei
)) {
2048 ret
= convert_extent_item_v0(trans
, root
->fs_info
->extent_root
,
2054 leaf
= path
->nodes
[0];
2055 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2058 BUG_ON(item_size
< sizeof(*ei
));
2059 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2060 __run_delayed_extent_op(extent_op
, leaf
, ei
);
2062 btrfs_mark_buffer_dirty(leaf
);
2064 btrfs_free_path(path
);
2068 static int run_delayed_tree_ref(struct btrfs_trans_handle
*trans
,
2069 struct btrfs_root
*root
,
2070 struct btrfs_delayed_ref_node
*node
,
2071 struct btrfs_delayed_extent_op
*extent_op
,
2072 int insert_reserved
)
2075 struct btrfs_delayed_tree_ref
*ref
;
2076 struct btrfs_key ins
;
2080 ins
.objectid
= node
->bytenr
;
2081 ins
.offset
= node
->num_bytes
;
2082 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2084 ref
= btrfs_delayed_node_to_tree_ref(node
);
2085 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2086 parent
= ref
->parent
;
2088 ref_root
= ref
->root
;
2090 BUG_ON(node
->ref_mod
!= 1);
2091 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2092 BUG_ON(!extent_op
|| !extent_op
->update_flags
||
2093 !extent_op
->update_key
);
2094 ret
= alloc_reserved_tree_block(trans
, root
,
2096 extent_op
->flags_to_set
,
2099 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2100 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
2101 node
->num_bytes
, parent
, ref_root
,
2102 ref
->level
, 0, 1, extent_op
);
2103 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2104 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
2105 node
->num_bytes
, parent
, ref_root
,
2106 ref
->level
, 0, 1, extent_op
);
2113 /* helper function to actually process a single delayed ref entry */
2114 static int run_one_delayed_ref(struct btrfs_trans_handle
*trans
,
2115 struct btrfs_root
*root
,
2116 struct btrfs_delayed_ref_node
*node
,
2117 struct btrfs_delayed_extent_op
*extent_op
,
2118 int insert_reserved
)
2125 if (btrfs_delayed_ref_is_head(node
)) {
2126 struct btrfs_delayed_ref_head
*head
;
2128 * we've hit the end of the chain and we were supposed
2129 * to insert this extent into the tree. But, it got
2130 * deleted before we ever needed to insert it, so all
2131 * we have to do is clean up the accounting
2134 head
= btrfs_delayed_node_to_head(node
);
2135 if (insert_reserved
) {
2136 btrfs_pin_extent(root
, node
->bytenr
,
2137 node
->num_bytes
, 1);
2138 if (head
->is_data
) {
2139 ret
= btrfs_del_csums(trans
, root
,
2147 if (node
->type
== BTRFS_TREE_BLOCK_REF_KEY
||
2148 node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2149 ret
= run_delayed_tree_ref(trans
, root
, node
, extent_op
,
2151 else if (node
->type
== BTRFS_EXTENT_DATA_REF_KEY
||
2152 node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2153 ret
= run_delayed_data_ref(trans
, root
, node
, extent_op
,
2160 static noinline
struct btrfs_delayed_ref_node
*
2161 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
2163 struct rb_node
*node
;
2164 struct btrfs_delayed_ref_node
*ref
;
2165 int action
= BTRFS_ADD_DELAYED_REF
;
2168 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2169 * this prevents ref count from going down to zero when
2170 * there still are pending delayed ref.
2172 node
= rb_prev(&head
->node
.rb_node
);
2176 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
2178 if (ref
->bytenr
!= head
->node
.bytenr
)
2180 if (ref
->action
== action
)
2182 node
= rb_prev(node
);
2184 if (action
== BTRFS_ADD_DELAYED_REF
) {
2185 action
= BTRFS_DROP_DELAYED_REF
;
2192 * Returns 0 on success or if called with an already aborted transaction.
2193 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
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
;
2203 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2206 int must_insert_reserved
= 0;
2208 delayed_refs
= &trans
->transaction
->delayed_refs
;
2211 /* pick a new head ref from the cluster list */
2212 if (list_empty(cluster
))
2215 locked_ref
= list_entry(cluster
->next
,
2216 struct btrfs_delayed_ref_head
, cluster
);
2218 /* grab the lock that says we are going to process
2219 * all the refs for this head */
2220 ret
= btrfs_delayed_ref_lock(trans
, locked_ref
);
2223 * we may have dropped the spin lock to get the head
2224 * mutex lock, and that might have given someone else
2225 * time to free the head. If that's true, it has been
2226 * removed from our list and we can move on.
2228 if (ret
== -EAGAIN
) {
2236 * We need to try and merge add/drops of the same ref since we
2237 * can run into issues with relocate dropping the implicit ref
2238 * and then it being added back again before the drop can
2239 * finish. If we merged anything we need to re-loop so we can
2242 btrfs_merge_delayed_refs(trans
, fs_info
, delayed_refs
,
2246 * locked_ref is the head node, so we have to go one
2247 * node back for any delayed ref updates
2249 ref
= select_delayed_ref(locked_ref
);
2251 if (ref
&& ref
->seq
&&
2252 btrfs_check_delayed_seq(fs_info
, delayed_refs
, ref
->seq
)) {
2254 * there are still refs with lower seq numbers in the
2255 * process of being added. Don't run this ref yet.
2257 list_del_init(&locked_ref
->cluster
);
2258 btrfs_delayed_ref_unlock(locked_ref
);
2260 delayed_refs
->num_heads_ready
++;
2261 spin_unlock(&delayed_refs
->lock
);
2263 spin_lock(&delayed_refs
->lock
);
2268 * record the must insert reserved flag before we
2269 * drop the spin lock.
2271 must_insert_reserved
= locked_ref
->must_insert_reserved
;
2272 locked_ref
->must_insert_reserved
= 0;
2274 extent_op
= locked_ref
->extent_op
;
2275 locked_ref
->extent_op
= NULL
;
2278 /* All delayed refs have been processed, Go ahead
2279 * and send the head node to run_one_delayed_ref,
2280 * so that any accounting fixes can happen
2282 ref
= &locked_ref
->node
;
2284 if (extent_op
&& must_insert_reserved
) {
2285 btrfs_free_delayed_extent_op(extent_op
);
2290 spin_unlock(&delayed_refs
->lock
);
2292 ret
= run_delayed_extent_op(trans
, root
,
2294 btrfs_free_delayed_extent_op(extent_op
);
2298 "btrfs: run_delayed_extent_op "
2299 "returned %d\n", ret
);
2300 spin_lock(&delayed_refs
->lock
);
2301 btrfs_delayed_ref_unlock(locked_ref
);
2310 rb_erase(&ref
->rb_node
, &delayed_refs
->root
);
2311 delayed_refs
->num_entries
--;
2312 if (!btrfs_delayed_ref_is_head(ref
)) {
2314 * when we play the delayed ref, also correct the
2317 switch (ref
->action
) {
2318 case BTRFS_ADD_DELAYED_REF
:
2319 case BTRFS_ADD_DELAYED_EXTENT
:
2320 locked_ref
->node
.ref_mod
-= ref
->ref_mod
;
2322 case BTRFS_DROP_DELAYED_REF
:
2323 locked_ref
->node
.ref_mod
+= ref
->ref_mod
;
2329 spin_unlock(&delayed_refs
->lock
);
2331 ret
= run_one_delayed_ref(trans
, root
, ref
, extent_op
,
2332 must_insert_reserved
);
2334 btrfs_free_delayed_extent_op(extent_op
);
2336 btrfs_delayed_ref_unlock(locked_ref
);
2337 btrfs_put_delayed_ref(ref
);
2339 "btrfs: run_one_delayed_ref returned %d\n", ret
);
2340 spin_lock(&delayed_refs
->lock
);
2345 * If this node is a head, that means all the refs in this head
2346 * have been dealt with, and we will pick the next head to deal
2347 * with, so we must unlock the head and drop it from the cluster
2348 * list before we release it.
2350 if (btrfs_delayed_ref_is_head(ref
)) {
2351 list_del_init(&locked_ref
->cluster
);
2352 btrfs_delayed_ref_unlock(locked_ref
);
2355 btrfs_put_delayed_ref(ref
);
2359 spin_lock(&delayed_refs
->lock
);
2364 #ifdef SCRAMBLE_DELAYED_REFS
2366 * Normally delayed refs get processed in ascending bytenr order. This
2367 * correlates in most cases to the order added. To expose dependencies on this
2368 * order, we start to process the tree in the middle instead of the beginning
2370 static u64
find_middle(struct rb_root
*root
)
2372 struct rb_node
*n
= root
->rb_node
;
2373 struct btrfs_delayed_ref_node
*entry
;
2376 u64 first
= 0, last
= 0;
2380 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2381 first
= entry
->bytenr
;
2385 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2386 last
= entry
->bytenr
;
2391 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2392 WARN_ON(!entry
->in_tree
);
2394 middle
= entry
->bytenr
;
2407 int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle
*trans
,
2408 struct btrfs_fs_info
*fs_info
)
2410 struct qgroup_update
*qgroup_update
;
2413 if (list_empty(&trans
->qgroup_ref_list
) !=
2414 !trans
->delayed_ref_elem
.seq
) {
2415 /* list without seq or seq without list */
2416 printk(KERN_ERR
"btrfs: qgroup accounting update error, list is%s empty, seq is %llu\n",
2417 list_empty(&trans
->qgroup_ref_list
) ? "" : " not",
2418 trans
->delayed_ref_elem
.seq
);
2422 if (!trans
->delayed_ref_elem
.seq
)
2425 while (!list_empty(&trans
->qgroup_ref_list
)) {
2426 qgroup_update
= list_first_entry(&trans
->qgroup_ref_list
,
2427 struct qgroup_update
, list
);
2428 list_del(&qgroup_update
->list
);
2430 ret
= btrfs_qgroup_account_ref(
2431 trans
, fs_info
, qgroup_update
->node
,
2432 qgroup_update
->extent_op
);
2433 kfree(qgroup_update
);
2436 btrfs_put_tree_mod_seq(fs_info
, &trans
->delayed_ref_elem
);
2442 * this starts processing the delayed reference count updates and
2443 * extent insertions we have queued up so far. count can be
2444 * 0, which means to process everything in the tree at the start
2445 * of the run (but not newly added entries), or it can be some target
2446 * number you'd like to process.
2448 * Returns 0 on success or if called with an aborted transaction
2449 * Returns <0 on error and aborts the transaction
2451 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2452 struct btrfs_root
*root
, unsigned long count
)
2454 struct rb_node
*node
;
2455 struct btrfs_delayed_ref_root
*delayed_refs
;
2456 struct btrfs_delayed_ref_node
*ref
;
2457 struct list_head cluster
;
2460 int run_all
= count
== (unsigned long)-1;
2464 /* We'll clean this up in btrfs_cleanup_transaction */
2468 if (root
== root
->fs_info
->extent_root
)
2469 root
= root
->fs_info
->tree_root
;
2471 btrfs_delayed_refs_qgroup_accounting(trans
, root
->fs_info
);
2473 delayed_refs
= &trans
->transaction
->delayed_refs
;
2474 INIT_LIST_HEAD(&cluster
);
2477 spin_lock(&delayed_refs
->lock
);
2479 #ifdef SCRAMBLE_DELAYED_REFS
2480 delayed_refs
->run_delayed_start
= find_middle(&delayed_refs
->root
);
2484 count
= delayed_refs
->num_entries
* 2;
2488 if (!(run_all
|| run_most
) &&
2489 delayed_refs
->num_heads_ready
< 64)
2493 * go find something we can process in the rbtree. We start at
2494 * the beginning of the tree, and then build a cluster
2495 * of refs to process starting at the first one we are able to
2498 delayed_start
= delayed_refs
->run_delayed_start
;
2499 ret
= btrfs_find_ref_cluster(trans
, &cluster
,
2500 delayed_refs
->run_delayed_start
);
2504 ret
= run_clustered_refs(trans
, root
, &cluster
);
2506 btrfs_release_ref_cluster(&cluster
);
2507 spin_unlock(&delayed_refs
->lock
);
2508 btrfs_abort_transaction(trans
, root
, ret
);
2512 count
-= min_t(unsigned long, ret
, count
);
2517 if (delayed_start
>= delayed_refs
->run_delayed_start
) {
2520 * btrfs_find_ref_cluster looped. let's do one
2521 * more cycle. if we don't run any delayed ref
2522 * during that cycle (because we can't because
2523 * all of them are blocked), bail out.
2528 * no runnable refs left, stop trying
2535 /* refs were run, let's reset staleness detection */
2541 if (!list_empty(&trans
->new_bgs
)) {
2542 spin_unlock(&delayed_refs
->lock
);
2543 btrfs_create_pending_block_groups(trans
, root
);
2544 spin_lock(&delayed_refs
->lock
);
2547 node
= rb_first(&delayed_refs
->root
);
2550 count
= (unsigned long)-1;
2553 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
2555 if (btrfs_delayed_ref_is_head(ref
)) {
2556 struct btrfs_delayed_ref_head
*head
;
2558 head
= btrfs_delayed_node_to_head(ref
);
2559 atomic_inc(&ref
->refs
);
2561 spin_unlock(&delayed_refs
->lock
);
2563 * Mutex was contended, block until it's
2564 * released and try again
2566 mutex_lock(&head
->mutex
);
2567 mutex_unlock(&head
->mutex
);
2569 btrfs_put_delayed_ref(ref
);
2573 node
= rb_next(node
);
2575 spin_unlock(&delayed_refs
->lock
);
2576 schedule_timeout(1);
2580 spin_unlock(&delayed_refs
->lock
);
2581 assert_qgroups_uptodate(trans
);
2585 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
2586 struct btrfs_root
*root
,
2587 u64 bytenr
, u64 num_bytes
, u64 flags
,
2590 struct btrfs_delayed_extent_op
*extent_op
;
2593 extent_op
= btrfs_alloc_delayed_extent_op();
2597 extent_op
->flags_to_set
= flags
;
2598 extent_op
->update_flags
= 1;
2599 extent_op
->update_key
= 0;
2600 extent_op
->is_data
= is_data
? 1 : 0;
2602 ret
= btrfs_add_delayed_extent_op(root
->fs_info
, trans
, bytenr
,
2603 num_bytes
, extent_op
);
2605 btrfs_free_delayed_extent_op(extent_op
);
2609 static noinline
int check_delayed_ref(struct btrfs_trans_handle
*trans
,
2610 struct btrfs_root
*root
,
2611 struct btrfs_path
*path
,
2612 u64 objectid
, u64 offset
, u64 bytenr
)
2614 struct btrfs_delayed_ref_head
*head
;
2615 struct btrfs_delayed_ref_node
*ref
;
2616 struct btrfs_delayed_data_ref
*data_ref
;
2617 struct btrfs_delayed_ref_root
*delayed_refs
;
2618 struct rb_node
*node
;
2622 delayed_refs
= &trans
->transaction
->delayed_refs
;
2623 spin_lock(&delayed_refs
->lock
);
2624 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
2628 if (!mutex_trylock(&head
->mutex
)) {
2629 atomic_inc(&head
->node
.refs
);
2630 spin_unlock(&delayed_refs
->lock
);
2632 btrfs_release_path(path
);
2635 * Mutex was contended, block until it's released and let
2638 mutex_lock(&head
->mutex
);
2639 mutex_unlock(&head
->mutex
);
2640 btrfs_put_delayed_ref(&head
->node
);
2644 node
= rb_prev(&head
->node
.rb_node
);
2648 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2650 if (ref
->bytenr
!= bytenr
)
2654 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
)
2657 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
2659 node
= rb_prev(node
);
2663 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2664 if (ref
->bytenr
== bytenr
&& ref
->seq
== seq
)
2668 if (data_ref
->root
!= root
->root_key
.objectid
||
2669 data_ref
->objectid
!= objectid
|| data_ref
->offset
!= offset
)
2674 mutex_unlock(&head
->mutex
);
2676 spin_unlock(&delayed_refs
->lock
);
2680 static noinline
int check_committed_ref(struct btrfs_trans_handle
*trans
,
2681 struct btrfs_root
*root
,
2682 struct btrfs_path
*path
,
2683 u64 objectid
, u64 offset
, u64 bytenr
)
2685 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2686 struct extent_buffer
*leaf
;
2687 struct btrfs_extent_data_ref
*ref
;
2688 struct btrfs_extent_inline_ref
*iref
;
2689 struct btrfs_extent_item
*ei
;
2690 struct btrfs_key key
;
2694 key
.objectid
= bytenr
;
2695 key
.offset
= (u64
)-1;
2696 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2698 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
2701 BUG_ON(ret
== 0); /* Corruption */
2704 if (path
->slots
[0] == 0)
2708 leaf
= path
->nodes
[0];
2709 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2711 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
2715 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2716 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2717 if (item_size
< sizeof(*ei
)) {
2718 WARN_ON(item_size
!= sizeof(struct btrfs_extent_item_v0
));
2722 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2724 if (item_size
!= sizeof(*ei
) +
2725 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
2728 if (btrfs_extent_generation(leaf
, ei
) <=
2729 btrfs_root_last_snapshot(&root
->root_item
))
2732 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
2733 if (btrfs_extent_inline_ref_type(leaf
, iref
) !=
2734 BTRFS_EXTENT_DATA_REF_KEY
)
2737 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
2738 if (btrfs_extent_refs(leaf
, ei
) !=
2739 btrfs_extent_data_ref_count(leaf
, ref
) ||
2740 btrfs_extent_data_ref_root(leaf
, ref
) !=
2741 root
->root_key
.objectid
||
2742 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
2743 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
2751 int btrfs_cross_ref_exist(struct btrfs_trans_handle
*trans
,
2752 struct btrfs_root
*root
,
2753 u64 objectid
, u64 offset
, u64 bytenr
)
2755 struct btrfs_path
*path
;
2759 path
= btrfs_alloc_path();
2764 ret
= check_committed_ref(trans
, root
, path
, objectid
,
2766 if (ret
&& ret
!= -ENOENT
)
2769 ret2
= check_delayed_ref(trans
, root
, path
, objectid
,
2771 } while (ret2
== -EAGAIN
);
2773 if (ret2
&& ret2
!= -ENOENT
) {
2778 if (ret
!= -ENOENT
|| ret2
!= -ENOENT
)
2781 btrfs_free_path(path
);
2782 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
2787 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
2788 struct btrfs_root
*root
,
2789 struct extent_buffer
*buf
,
2790 int full_backref
, int inc
, int for_cow
)
2797 struct btrfs_key key
;
2798 struct btrfs_file_extent_item
*fi
;
2802 int (*process_func
)(struct btrfs_trans_handle
*, struct btrfs_root
*,
2803 u64
, u64
, u64
, u64
, u64
, u64
, int);
2805 ref_root
= btrfs_header_owner(buf
);
2806 nritems
= btrfs_header_nritems(buf
);
2807 level
= btrfs_header_level(buf
);
2809 if (!root
->ref_cows
&& level
== 0)
2813 process_func
= btrfs_inc_extent_ref
;
2815 process_func
= btrfs_free_extent
;
2818 parent
= buf
->start
;
2822 for (i
= 0; i
< nritems
; i
++) {
2824 btrfs_item_key_to_cpu(buf
, &key
, i
);
2825 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
2827 fi
= btrfs_item_ptr(buf
, i
,
2828 struct btrfs_file_extent_item
);
2829 if (btrfs_file_extent_type(buf
, fi
) ==
2830 BTRFS_FILE_EXTENT_INLINE
)
2832 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
2836 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
2837 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
2838 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
2839 parent
, ref_root
, key
.objectid
,
2840 key
.offset
, for_cow
);
2844 bytenr
= btrfs_node_blockptr(buf
, i
);
2845 num_bytes
= btrfs_level_size(root
, level
- 1);
2846 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
2847 parent
, ref_root
, level
- 1, 0,
2858 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2859 struct extent_buffer
*buf
, int full_backref
, int for_cow
)
2861 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1, for_cow
);
2864 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2865 struct extent_buffer
*buf
, int full_backref
, int for_cow
)
2867 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0, for_cow
);
2870 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
2871 struct btrfs_root
*root
,
2872 struct btrfs_path
*path
,
2873 struct btrfs_block_group_cache
*cache
)
2876 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2878 struct extent_buffer
*leaf
;
2880 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
2883 BUG_ON(ret
); /* Corruption */
2885 leaf
= path
->nodes
[0];
2886 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
2887 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
2888 btrfs_mark_buffer_dirty(leaf
);
2889 btrfs_release_path(path
);
2892 btrfs_abort_transaction(trans
, root
, ret
);
2899 static struct btrfs_block_group_cache
*
2900 next_block_group(struct btrfs_root
*root
,
2901 struct btrfs_block_group_cache
*cache
)
2903 struct rb_node
*node
;
2904 spin_lock(&root
->fs_info
->block_group_cache_lock
);
2905 node
= rb_next(&cache
->cache_node
);
2906 btrfs_put_block_group(cache
);
2908 cache
= rb_entry(node
, struct btrfs_block_group_cache
,
2910 btrfs_get_block_group(cache
);
2913 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
2917 static int cache_save_setup(struct btrfs_block_group_cache
*block_group
,
2918 struct btrfs_trans_handle
*trans
,
2919 struct btrfs_path
*path
)
2921 struct btrfs_root
*root
= block_group
->fs_info
->tree_root
;
2922 struct inode
*inode
= NULL
;
2924 int dcs
= BTRFS_DC_ERROR
;
2930 * If this block group is smaller than 100 megs don't bother caching the
2933 if (block_group
->key
.offset
< (100 * 1024 * 1024)) {
2934 spin_lock(&block_group
->lock
);
2935 block_group
->disk_cache_state
= BTRFS_DC_WRITTEN
;
2936 spin_unlock(&block_group
->lock
);
2941 inode
= lookup_free_space_inode(root
, block_group
, path
);
2942 if (IS_ERR(inode
) && PTR_ERR(inode
) != -ENOENT
) {
2943 ret
= PTR_ERR(inode
);
2944 btrfs_release_path(path
);
2948 if (IS_ERR(inode
)) {
2952 if (block_group
->ro
)
2955 ret
= create_free_space_inode(root
, trans
, block_group
, path
);
2961 /* We've already setup this transaction, go ahead and exit */
2962 if (block_group
->cache_generation
== trans
->transid
&&
2963 i_size_read(inode
)) {
2964 dcs
= BTRFS_DC_SETUP
;
2969 * We want to set the generation to 0, that way if anything goes wrong
2970 * from here on out we know not to trust this cache when we load up next
2973 BTRFS_I(inode
)->generation
= 0;
2974 ret
= btrfs_update_inode(trans
, root
, inode
);
2977 if (i_size_read(inode
) > 0) {
2978 ret
= btrfs_truncate_free_space_cache(root
, trans
, path
,
2984 spin_lock(&block_group
->lock
);
2985 if (block_group
->cached
!= BTRFS_CACHE_FINISHED
||
2986 !btrfs_test_opt(root
, SPACE_CACHE
)) {
2988 * don't bother trying to write stuff out _if_
2989 * a) we're not cached,
2990 * b) we're with nospace_cache mount option.
2992 dcs
= BTRFS_DC_WRITTEN
;
2993 spin_unlock(&block_group
->lock
);
2996 spin_unlock(&block_group
->lock
);
2999 * Try to preallocate enough space based on how big the block group is.
3000 * Keep in mind this has to include any pinned space which could end up
3001 * taking up quite a bit since it's not folded into the other space
3004 num_pages
= (int)div64_u64(block_group
->key
.offset
, 256 * 1024 * 1024);
3009 num_pages
*= PAGE_CACHE_SIZE
;
3011 ret
= btrfs_check_data_free_space(inode
, num_pages
);
3015 ret
= btrfs_prealloc_file_range_trans(inode
, trans
, 0, 0, num_pages
,
3016 num_pages
, num_pages
,
3019 dcs
= BTRFS_DC_SETUP
;
3020 btrfs_free_reserved_data_space(inode
, num_pages
);
3025 btrfs_release_path(path
);
3027 spin_lock(&block_group
->lock
);
3028 if (!ret
&& dcs
== BTRFS_DC_SETUP
)
3029 block_group
->cache_generation
= trans
->transid
;
3030 block_group
->disk_cache_state
= dcs
;
3031 spin_unlock(&block_group
->lock
);
3036 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
3037 struct btrfs_root
*root
)
3039 struct btrfs_block_group_cache
*cache
;
3041 struct btrfs_path
*path
;
3044 path
= btrfs_alloc_path();
3050 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
3052 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
)
3054 cache
= next_block_group(root
, cache
);
3062 err
= cache_save_setup(cache
, trans
, path
);
3063 last
= cache
->key
.objectid
+ cache
->key
.offset
;
3064 btrfs_put_block_group(cache
);
3069 err
= btrfs_run_delayed_refs(trans
, root
,
3071 if (err
) /* File system offline */
3075 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
3077 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
) {
3078 btrfs_put_block_group(cache
);
3084 cache
= next_block_group(root
, cache
);
3093 if (cache
->disk_cache_state
== BTRFS_DC_SETUP
)
3094 cache
->disk_cache_state
= BTRFS_DC_NEED_WRITE
;
3096 last
= cache
->key
.objectid
+ cache
->key
.offset
;
3098 err
= write_one_cache_group(trans
, root
, path
, cache
);
3099 if (err
) /* File system offline */
3102 btrfs_put_block_group(cache
);
3107 * I don't think this is needed since we're just marking our
3108 * preallocated extent as written, but just in case it can't
3112 err
= btrfs_run_delayed_refs(trans
, root
,
3114 if (err
) /* File system offline */
3118 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
3121 * Really this shouldn't happen, but it could if we
3122 * couldn't write the entire preallocated extent and
3123 * splitting the extent resulted in a new block.
3126 btrfs_put_block_group(cache
);
3129 if (cache
->disk_cache_state
== BTRFS_DC_NEED_WRITE
)
3131 cache
= next_block_group(root
, cache
);
3140 err
= btrfs_write_out_cache(root
, trans
, cache
, path
);
3143 * If we didn't have an error then the cache state is still
3144 * NEED_WRITE, so we can set it to WRITTEN.
3146 if (!err
&& cache
->disk_cache_state
== BTRFS_DC_NEED_WRITE
)
3147 cache
->disk_cache_state
= BTRFS_DC_WRITTEN
;
3148 last
= cache
->key
.objectid
+ cache
->key
.offset
;
3149 btrfs_put_block_group(cache
);
3153 btrfs_free_path(path
);
3157 int btrfs_extent_readonly(struct btrfs_root
*root
, u64 bytenr
)
3159 struct btrfs_block_group_cache
*block_group
;
3162 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
3163 if (!block_group
|| block_group
->ro
)
3166 btrfs_put_block_group(block_group
);
3170 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
3171 u64 total_bytes
, u64 bytes_used
,
3172 struct btrfs_space_info
**space_info
)
3174 struct btrfs_space_info
*found
;
3178 if (flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3179 BTRFS_BLOCK_GROUP_RAID10
))
3184 found
= __find_space_info(info
, flags
);
3186 spin_lock(&found
->lock
);
3187 found
->total_bytes
+= total_bytes
;
3188 found
->disk_total
+= total_bytes
* factor
;
3189 found
->bytes_used
+= bytes_used
;
3190 found
->disk_used
+= bytes_used
* factor
;
3192 spin_unlock(&found
->lock
);
3193 *space_info
= found
;
3196 found
= kzalloc(sizeof(*found
), GFP_NOFS
);
3200 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
3201 INIT_LIST_HEAD(&found
->block_groups
[i
]);
3202 init_rwsem(&found
->groups_sem
);
3203 spin_lock_init(&found
->lock
);
3204 found
->flags
= flags
& BTRFS_BLOCK_GROUP_TYPE_MASK
;
3205 found
->total_bytes
= total_bytes
;
3206 found
->disk_total
= total_bytes
* factor
;
3207 found
->bytes_used
= bytes_used
;
3208 found
->disk_used
= bytes_used
* factor
;
3209 found
->bytes_pinned
= 0;
3210 found
->bytes_reserved
= 0;
3211 found
->bytes_readonly
= 0;
3212 found
->bytes_may_use
= 0;
3214 found
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3215 found
->chunk_alloc
= 0;
3217 init_waitqueue_head(&found
->wait
);
3218 *space_info
= found
;
3219 list_add_rcu(&found
->list
, &info
->space_info
);
3220 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3221 info
->data_sinfo
= found
;
3225 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
3227 u64 extra_flags
= chunk_to_extended(flags
) &
3228 BTRFS_EXTENDED_PROFILE_MASK
;
3230 write_seqlock(&fs_info
->profiles_lock
);
3231 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3232 fs_info
->avail_data_alloc_bits
|= extra_flags
;
3233 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3234 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
3235 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3236 fs_info
->avail_system_alloc_bits
|= extra_flags
;
3237 write_sequnlock(&fs_info
->profiles_lock
);
3241 * returns target flags in extended format or 0 if restripe for this
3242 * chunk_type is not in progress
3244 * should be called with either volume_mutex or balance_lock held
3246 static u64
get_restripe_target(struct btrfs_fs_info
*fs_info
, u64 flags
)
3248 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3254 if (flags
& BTRFS_BLOCK_GROUP_DATA
&&
3255 bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3256 target
= BTRFS_BLOCK_GROUP_DATA
| bctl
->data
.target
;
3257 } else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
&&
3258 bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3259 target
= BTRFS_BLOCK_GROUP_SYSTEM
| bctl
->sys
.target
;
3260 } else if (flags
& BTRFS_BLOCK_GROUP_METADATA
&&
3261 bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3262 target
= BTRFS_BLOCK_GROUP_METADATA
| bctl
->meta
.target
;
3269 * @flags: available profiles in extended format (see ctree.h)
3271 * Returns reduced profile in chunk format. If profile changing is in
3272 * progress (either running or paused) picks the target profile (if it's
3273 * already available), otherwise falls back to plain reducing.
3275 u64
btrfs_reduce_alloc_profile(struct btrfs_root
*root
, u64 flags
)
3278 * we add in the count of missing devices because we want
3279 * to make sure that any RAID levels on a degraded FS
3280 * continue to be honored.
3282 u64 num_devices
= root
->fs_info
->fs_devices
->rw_devices
+
3283 root
->fs_info
->fs_devices
->missing_devices
;
3287 * see if restripe for this chunk_type is in progress, if so
3288 * try to reduce to the target profile
3290 spin_lock(&root
->fs_info
->balance_lock
);
3291 target
= get_restripe_target(root
->fs_info
, flags
);
3293 /* pick target profile only if it's already available */
3294 if ((flags
& target
) & BTRFS_EXTENDED_PROFILE_MASK
) {
3295 spin_unlock(&root
->fs_info
->balance_lock
);
3296 return extended_to_chunk(target
);
3299 spin_unlock(&root
->fs_info
->balance_lock
);
3301 if (num_devices
== 1)
3302 flags
&= ~(BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID0
);
3303 if (num_devices
< 4)
3304 flags
&= ~BTRFS_BLOCK_GROUP_RAID10
;
3306 if ((flags
& BTRFS_BLOCK_GROUP_DUP
) &&
3307 (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
3308 BTRFS_BLOCK_GROUP_RAID10
))) {
3309 flags
&= ~BTRFS_BLOCK_GROUP_DUP
;
3312 if ((flags
& BTRFS_BLOCK_GROUP_RAID1
) &&
3313 (flags
& BTRFS_BLOCK_GROUP_RAID10
)) {
3314 flags
&= ~BTRFS_BLOCK_GROUP_RAID1
;
3317 if ((flags
& BTRFS_BLOCK_GROUP_RAID0
) &&
3318 ((flags
& BTRFS_BLOCK_GROUP_RAID1
) |
3319 (flags
& BTRFS_BLOCK_GROUP_RAID10
) |
3320 (flags
& BTRFS_BLOCK_GROUP_DUP
))) {
3321 flags
&= ~BTRFS_BLOCK_GROUP_RAID0
;
3324 return extended_to_chunk(flags
);
3327 static u64
get_alloc_profile(struct btrfs_root
*root
, u64 flags
)
3332 seq
= read_seqbegin(&root
->fs_info
->profiles_lock
);
3334 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3335 flags
|= root
->fs_info
->avail_data_alloc_bits
;
3336 else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3337 flags
|= root
->fs_info
->avail_system_alloc_bits
;
3338 else if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3339 flags
|= root
->fs_info
->avail_metadata_alloc_bits
;
3340 } while (read_seqretry(&root
->fs_info
->profiles_lock
, seq
));
3342 return btrfs_reduce_alloc_profile(root
, flags
);
3345 u64
btrfs_get_alloc_profile(struct btrfs_root
*root
, int data
)
3350 flags
= BTRFS_BLOCK_GROUP_DATA
;
3351 else if (root
== root
->fs_info
->chunk_root
)
3352 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
3354 flags
= BTRFS_BLOCK_GROUP_METADATA
;
3356 return get_alloc_profile(root
, flags
);
3360 * This will check the space that the inode allocates from to make sure we have
3361 * enough space for bytes.
3363 int btrfs_check_data_free_space(struct inode
*inode
, u64 bytes
)
3365 struct btrfs_space_info
*data_sinfo
;
3366 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3367 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3369 int ret
= 0, committed
= 0, alloc_chunk
= 1;
3371 /* make sure bytes are sectorsize aligned */
3372 bytes
= (bytes
+ root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
3374 if (root
== root
->fs_info
->tree_root
||
3375 BTRFS_I(inode
)->location
.objectid
== BTRFS_FREE_INO_OBJECTID
) {
3380 data_sinfo
= fs_info
->data_sinfo
;
3385 /* make sure we have enough space to handle the data first */
3386 spin_lock(&data_sinfo
->lock
);
3387 used
= data_sinfo
->bytes_used
+ data_sinfo
->bytes_reserved
+
3388 data_sinfo
->bytes_pinned
+ data_sinfo
->bytes_readonly
+
3389 data_sinfo
->bytes_may_use
;
3391 if (used
+ bytes
> data_sinfo
->total_bytes
) {
3392 struct btrfs_trans_handle
*trans
;
3395 * if we don't have enough free bytes in this space then we need
3396 * to alloc a new chunk.
3398 if (!data_sinfo
->full
&& alloc_chunk
) {
3401 data_sinfo
->force_alloc
= CHUNK_ALLOC_FORCE
;
3402 spin_unlock(&data_sinfo
->lock
);
3404 alloc_target
= btrfs_get_alloc_profile(root
, 1);
3405 trans
= btrfs_join_transaction(root
);
3407 return PTR_ERR(trans
);
3409 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
3411 CHUNK_ALLOC_NO_FORCE
);
3412 btrfs_end_transaction(trans
, root
);
3421 data_sinfo
= fs_info
->data_sinfo
;
3427 * If we have less pinned bytes than we want to allocate then
3428 * don't bother committing the transaction, it won't help us.
3430 if (data_sinfo
->bytes_pinned
< bytes
)
3432 spin_unlock(&data_sinfo
->lock
);
3434 /* commit the current transaction and try again */
3437 !atomic_read(&root
->fs_info
->open_ioctl_trans
)) {
3439 trans
= btrfs_join_transaction(root
);
3441 return PTR_ERR(trans
);
3442 ret
= btrfs_commit_transaction(trans
, root
);
3450 data_sinfo
->bytes_may_use
+= bytes
;
3451 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
3452 data_sinfo
->flags
, bytes
, 1);
3453 spin_unlock(&data_sinfo
->lock
);
3459 * Called if we need to clear a data reservation for this inode.
3461 void btrfs_free_reserved_data_space(struct inode
*inode
, u64 bytes
)
3463 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3464 struct btrfs_space_info
*data_sinfo
;
3466 /* make sure bytes are sectorsize aligned */
3467 bytes
= (bytes
+ root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
3469 data_sinfo
= root
->fs_info
->data_sinfo
;
3470 spin_lock(&data_sinfo
->lock
);
3471 data_sinfo
->bytes_may_use
-= bytes
;
3472 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
3473 data_sinfo
->flags
, bytes
, 0);
3474 spin_unlock(&data_sinfo
->lock
);
3477 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
3479 struct list_head
*head
= &info
->space_info
;
3480 struct btrfs_space_info
*found
;
3483 list_for_each_entry_rcu(found
, head
, list
) {
3484 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
3485 found
->force_alloc
= CHUNK_ALLOC_FORCE
;
3490 static int should_alloc_chunk(struct btrfs_root
*root
,
3491 struct btrfs_space_info
*sinfo
, int force
)
3493 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
3494 u64 num_bytes
= sinfo
->total_bytes
- sinfo
->bytes_readonly
;
3495 u64 num_allocated
= sinfo
->bytes_used
+ sinfo
->bytes_reserved
;
3498 if (force
== CHUNK_ALLOC_FORCE
)
3502 * We need to take into account the global rsv because for all intents
3503 * and purposes it's used space. Don't worry about locking the
3504 * global_rsv, it doesn't change except when the transaction commits.
3506 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
3507 num_allocated
+= global_rsv
->size
;
3510 * in limited mode, we want to have some free space up to
3511 * about 1% of the FS size.
3513 if (force
== CHUNK_ALLOC_LIMITED
) {
3514 thresh
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
3515 thresh
= max_t(u64
, 64 * 1024 * 1024,
3516 div_factor_fine(thresh
, 1));
3518 if (num_bytes
- num_allocated
< thresh
)
3522 if (num_allocated
+ 2 * 1024 * 1024 < div_factor(num_bytes
, 8))
3527 static u64
get_system_chunk_thresh(struct btrfs_root
*root
, u64 type
)
3531 if (type
& BTRFS_BLOCK_GROUP_RAID10
||
3532 type
& BTRFS_BLOCK_GROUP_RAID0
)
3533 num_dev
= root
->fs_info
->fs_devices
->rw_devices
;
3534 else if (type
& BTRFS_BLOCK_GROUP_RAID1
)
3537 num_dev
= 1; /* DUP or single */
3539 /* metadata for updaing devices and chunk tree */
3540 return btrfs_calc_trans_metadata_size(root
, num_dev
+ 1);
3543 static void check_system_chunk(struct btrfs_trans_handle
*trans
,
3544 struct btrfs_root
*root
, u64 type
)
3546 struct btrfs_space_info
*info
;
3550 info
= __find_space_info(root
->fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
3551 spin_lock(&info
->lock
);
3552 left
= info
->total_bytes
- info
->bytes_used
- info
->bytes_pinned
-
3553 info
->bytes_reserved
- info
->bytes_readonly
;
3554 spin_unlock(&info
->lock
);
3556 thresh
= get_system_chunk_thresh(root
, type
);
3557 if (left
< thresh
&& btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
3558 printk(KERN_INFO
"left=%llu, need=%llu, flags=%llu\n",
3559 left
, thresh
, type
);
3560 dump_space_info(info
, 0, 0);
3563 if (left
< thresh
) {
3566 flags
= btrfs_get_alloc_profile(root
->fs_info
->chunk_root
, 0);
3567 btrfs_alloc_chunk(trans
, root
, flags
);
3571 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
3572 struct btrfs_root
*extent_root
, u64 flags
, int force
)
3574 struct btrfs_space_info
*space_info
;
3575 struct btrfs_fs_info
*fs_info
= extent_root
->fs_info
;
3576 int wait_for_alloc
= 0;
3579 /* Don't re-enter if we're already allocating a chunk */
3580 if (trans
->allocating_chunk
)
3583 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
3585 ret
= update_space_info(extent_root
->fs_info
, flags
,
3587 BUG_ON(ret
); /* -ENOMEM */
3589 BUG_ON(!space_info
); /* Logic error */
3592 spin_lock(&space_info
->lock
);
3593 if (force
< space_info
->force_alloc
)
3594 force
= space_info
->force_alloc
;
3595 if (space_info
->full
) {
3596 spin_unlock(&space_info
->lock
);
3600 if (!should_alloc_chunk(extent_root
, space_info
, force
)) {
3601 spin_unlock(&space_info
->lock
);
3603 } else if (space_info
->chunk_alloc
) {
3606 space_info
->chunk_alloc
= 1;
3609 spin_unlock(&space_info
->lock
);
3611 mutex_lock(&fs_info
->chunk_mutex
);
3614 * The chunk_mutex is held throughout the entirety of a chunk
3615 * allocation, so once we've acquired the chunk_mutex we know that the
3616 * other guy is done and we need to recheck and see if we should
3619 if (wait_for_alloc
) {
3620 mutex_unlock(&fs_info
->chunk_mutex
);
3625 trans
->allocating_chunk
= true;
3628 * If we have mixed data/metadata chunks we want to make sure we keep
3629 * allocating mixed chunks instead of individual chunks.
3631 if (btrfs_mixed_space_info(space_info
))
3632 flags
|= (BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
);
3635 * if we're doing a data chunk, go ahead and make sure that
3636 * we keep a reasonable number of metadata chunks allocated in the
3639 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
3640 fs_info
->data_chunk_allocations
++;
3641 if (!(fs_info
->data_chunk_allocations
%
3642 fs_info
->metadata_ratio
))
3643 force_metadata_allocation(fs_info
);
3647 * Check if we have enough space in SYSTEM chunk because we may need
3648 * to update devices.
3650 check_system_chunk(trans
, extent_root
, flags
);
3652 ret
= btrfs_alloc_chunk(trans
, extent_root
, flags
);
3653 trans
->allocating_chunk
= false;
3654 if (ret
< 0 && ret
!= -ENOSPC
)
3657 spin_lock(&space_info
->lock
);
3659 space_info
->full
= 1;
3663 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3664 space_info
->chunk_alloc
= 0;
3665 spin_unlock(&space_info
->lock
);
3667 mutex_unlock(&fs_info
->chunk_mutex
);
3671 static int can_overcommit(struct btrfs_root
*root
,
3672 struct btrfs_space_info
*space_info
, u64 bytes
,
3673 enum btrfs_reserve_flush_enum flush
)
3675 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
3676 u64 profile
= btrfs_get_alloc_profile(root
, 0);
3681 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
3682 space_info
->bytes_pinned
+ space_info
->bytes_readonly
;
3684 spin_lock(&global_rsv
->lock
);
3685 rsv_size
= global_rsv
->size
;
3686 spin_unlock(&global_rsv
->lock
);
3689 * We only want to allow over committing if we have lots of actual space
3690 * free, but if we don't have enough space to handle the global reserve
3691 * space then we could end up having a real enospc problem when trying
3692 * to allocate a chunk or some other such important allocation.
3695 if (used
+ rsv_size
>= space_info
->total_bytes
)
3698 used
+= space_info
->bytes_may_use
;
3700 spin_lock(&root
->fs_info
->free_chunk_lock
);
3701 avail
= root
->fs_info
->free_chunk_space
;
3702 spin_unlock(&root
->fs_info
->free_chunk_lock
);
3705 * If we have dup, raid1 or raid10 then only half of the free
3706 * space is actually useable.
3708 if (profile
& (BTRFS_BLOCK_GROUP_DUP
|
3709 BTRFS_BLOCK_GROUP_RAID1
|
3710 BTRFS_BLOCK_GROUP_RAID10
))
3714 * If we aren't flushing all things, let us overcommit up to
3715 * 1/2th of the space. If we can flush, don't let us overcommit
3716 * too much, let it overcommit up to 1/8 of the space.
3718 if (flush
== BTRFS_RESERVE_FLUSH_ALL
)
3723 if (used
+ bytes
< space_info
->total_bytes
+ avail
)
3728 static inline int writeback_inodes_sb_nr_if_idle_safe(struct super_block
*sb
,
3729 unsigned long nr_pages
,
3730 enum wb_reason reason
)
3732 /* the flusher is dealing with the dirty inodes now. */
3733 if (writeback_in_progress(sb
->s_bdi
))
3736 if (down_read_trylock(&sb
->s_umount
)) {
3737 writeback_inodes_sb_nr(sb
, nr_pages
, reason
);
3738 up_read(&sb
->s_umount
);
3745 void btrfs_writeback_inodes_sb_nr(struct btrfs_root
*root
,
3746 unsigned long nr_pages
)
3748 struct super_block
*sb
= root
->fs_info
->sb
;
3751 /* If we can not start writeback, just sync all the delalloc file. */
3752 started
= writeback_inodes_sb_nr_if_idle_safe(sb
, nr_pages
,
3753 WB_REASON_FS_FREE_SPACE
);
3756 * We needn't worry the filesystem going from r/w to r/o though
3757 * we don't acquire ->s_umount mutex, because the filesystem
3758 * should guarantee the delalloc inodes list be empty after
3759 * the filesystem is readonly(all dirty pages are written to
3762 btrfs_start_delalloc_inodes(root
, 0);
3763 btrfs_wait_ordered_extents(root
, 0);
3768 * shrink metadata reservation for delalloc
3770 static void shrink_delalloc(struct btrfs_root
*root
, u64 to_reclaim
, u64 orig
,
3773 struct btrfs_block_rsv
*block_rsv
;
3774 struct btrfs_space_info
*space_info
;
3775 struct btrfs_trans_handle
*trans
;
3779 unsigned long nr_pages
= (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT
;
3781 enum btrfs_reserve_flush_enum flush
;
3783 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
3784 block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
3785 space_info
= block_rsv
->space_info
;
3788 delalloc_bytes
= percpu_counter_sum_positive(
3789 &root
->fs_info
->delalloc_bytes
);
3790 if (delalloc_bytes
== 0) {
3793 btrfs_wait_ordered_extents(root
, 0);
3797 while (delalloc_bytes
&& loops
< 3) {
3798 max_reclaim
= min(delalloc_bytes
, to_reclaim
);
3799 nr_pages
= max_reclaim
>> PAGE_CACHE_SHIFT
;
3800 btrfs_writeback_inodes_sb_nr(root
, nr_pages
);
3802 * We need to wait for the async pages to actually start before
3805 wait_event(root
->fs_info
->async_submit_wait
,
3806 !atomic_read(&root
->fs_info
->async_delalloc_pages
));
3809 flush
= BTRFS_RESERVE_FLUSH_ALL
;
3811 flush
= BTRFS_RESERVE_NO_FLUSH
;
3812 spin_lock(&space_info
->lock
);
3813 if (can_overcommit(root
, space_info
, orig
, flush
)) {
3814 spin_unlock(&space_info
->lock
);
3817 spin_unlock(&space_info
->lock
);
3820 if (wait_ordered
&& !trans
) {
3821 btrfs_wait_ordered_extents(root
, 0);
3823 time_left
= schedule_timeout_killable(1);
3828 delalloc_bytes
= percpu_counter_sum_positive(
3829 &root
->fs_info
->delalloc_bytes
);
3834 * maybe_commit_transaction - possibly commit the transaction if its ok to
3835 * @root - the root we're allocating for
3836 * @bytes - the number of bytes we want to reserve
3837 * @force - force the commit
3839 * This will check to make sure that committing the transaction will actually
3840 * get us somewhere and then commit the transaction if it does. Otherwise it
3841 * will return -ENOSPC.
3843 static int may_commit_transaction(struct btrfs_root
*root
,
3844 struct btrfs_space_info
*space_info
,
3845 u64 bytes
, int force
)
3847 struct btrfs_block_rsv
*delayed_rsv
= &root
->fs_info
->delayed_block_rsv
;
3848 struct btrfs_trans_handle
*trans
;
3850 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
3857 /* See if there is enough pinned space to make this reservation */
3858 spin_lock(&space_info
->lock
);
3859 if (space_info
->bytes_pinned
>= bytes
) {
3860 spin_unlock(&space_info
->lock
);
3863 spin_unlock(&space_info
->lock
);
3866 * See if there is some space in the delayed insertion reservation for
3869 if (space_info
!= delayed_rsv
->space_info
)
3872 spin_lock(&space_info
->lock
);
3873 spin_lock(&delayed_rsv
->lock
);
3874 if (space_info
->bytes_pinned
+ delayed_rsv
->size
< bytes
) {
3875 spin_unlock(&delayed_rsv
->lock
);
3876 spin_unlock(&space_info
->lock
);
3879 spin_unlock(&delayed_rsv
->lock
);
3880 spin_unlock(&space_info
->lock
);
3883 trans
= btrfs_join_transaction(root
);
3887 return btrfs_commit_transaction(trans
, root
);
3891 FLUSH_DELAYED_ITEMS_NR
= 1,
3892 FLUSH_DELAYED_ITEMS
= 2,
3894 FLUSH_DELALLOC_WAIT
= 4,
3899 static int flush_space(struct btrfs_root
*root
,
3900 struct btrfs_space_info
*space_info
, u64 num_bytes
,
3901 u64 orig_bytes
, int state
)
3903 struct btrfs_trans_handle
*trans
;
3908 case FLUSH_DELAYED_ITEMS_NR
:
3909 case FLUSH_DELAYED_ITEMS
:
3910 if (state
== FLUSH_DELAYED_ITEMS_NR
) {
3911 u64 bytes
= btrfs_calc_trans_metadata_size(root
, 1);
3913 nr
= (int)div64_u64(num_bytes
, bytes
);
3920 trans
= btrfs_join_transaction(root
);
3921 if (IS_ERR(trans
)) {
3922 ret
= PTR_ERR(trans
);
3925 ret
= btrfs_run_delayed_items_nr(trans
, root
, nr
);
3926 btrfs_end_transaction(trans
, root
);
3928 case FLUSH_DELALLOC
:
3929 case FLUSH_DELALLOC_WAIT
:
3930 shrink_delalloc(root
, num_bytes
, orig_bytes
,
3931 state
== FLUSH_DELALLOC_WAIT
);
3934 trans
= btrfs_join_transaction(root
);
3935 if (IS_ERR(trans
)) {
3936 ret
= PTR_ERR(trans
);
3939 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
3940 btrfs_get_alloc_profile(root
, 0),
3941 CHUNK_ALLOC_NO_FORCE
);
3942 btrfs_end_transaction(trans
, root
);
3947 ret
= may_commit_transaction(root
, space_info
, orig_bytes
, 0);
3957 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3958 * @root - the root we're allocating for
3959 * @block_rsv - the block_rsv we're allocating for
3960 * @orig_bytes - the number of bytes we want
3961 * @flush - wether or not we can flush to make our reservation
3963 * This will reserve orgi_bytes number of bytes from the space info associated
3964 * with the block_rsv. If there is not enough space it will make an attempt to
3965 * flush out space to make room. It will do this by flushing delalloc if
3966 * possible or committing the transaction. If flush is 0 then no attempts to
3967 * regain reservations will be made and this will fail if there is not enough
3970 static int reserve_metadata_bytes(struct btrfs_root
*root
,
3971 struct btrfs_block_rsv
*block_rsv
,
3973 enum btrfs_reserve_flush_enum flush
)
3975 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
3977 u64 num_bytes
= orig_bytes
;
3978 int flush_state
= FLUSH_DELAYED_ITEMS_NR
;
3980 bool flushing
= false;
3984 spin_lock(&space_info
->lock
);
3986 * We only want to wait if somebody other than us is flushing and we
3987 * are actually allowed to flush all things.
3989 while (flush
== BTRFS_RESERVE_FLUSH_ALL
&& !flushing
&&
3990 space_info
->flush
) {
3991 spin_unlock(&space_info
->lock
);
3993 * If we have a trans handle we can't wait because the flusher
3994 * may have to commit the transaction, which would mean we would
3995 * deadlock since we are waiting for the flusher to finish, but
3996 * hold the current transaction open.
3998 if (current
->journal_info
)
4000 ret
= wait_event_killable(space_info
->wait
, !space_info
->flush
);
4001 /* Must have been killed, return */
4005 spin_lock(&space_info
->lock
);
4009 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4010 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4011 space_info
->bytes_may_use
;
4014 * The idea here is that we've not already over-reserved the block group
4015 * then we can go ahead and save our reservation first and then start
4016 * flushing if we need to. Otherwise if we've already overcommitted
4017 * lets start flushing stuff first and then come back and try to make
4020 if (used
<= space_info
->total_bytes
) {
4021 if (used
+ orig_bytes
<= space_info
->total_bytes
) {
4022 space_info
->bytes_may_use
+= orig_bytes
;
4023 trace_btrfs_space_reservation(root
->fs_info
,
4024 "space_info", space_info
->flags
, orig_bytes
, 1);
4028 * Ok set num_bytes to orig_bytes since we aren't
4029 * overocmmitted, this way we only try and reclaim what
4032 num_bytes
= orig_bytes
;
4036 * Ok we're over committed, set num_bytes to the overcommitted
4037 * amount plus the amount of bytes that we need for this
4040 num_bytes
= used
- space_info
->total_bytes
+
4044 if (ret
&& can_overcommit(root
, space_info
, orig_bytes
, flush
)) {
4045 space_info
->bytes_may_use
+= orig_bytes
;
4046 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
4047 space_info
->flags
, orig_bytes
,
4053 * Couldn't make our reservation, save our place so while we're trying
4054 * to reclaim space we can actually use it instead of somebody else
4055 * stealing it from us.
4057 * We make the other tasks wait for the flush only when we can flush
4060 if (ret
&& flush
!= BTRFS_RESERVE_NO_FLUSH
) {
4062 space_info
->flush
= 1;
4065 spin_unlock(&space_info
->lock
);
4067 if (!ret
|| flush
== BTRFS_RESERVE_NO_FLUSH
)
4070 ret
= flush_space(root
, space_info
, num_bytes
, orig_bytes
,
4075 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4076 * would happen. So skip delalloc flush.
4078 if (flush
== BTRFS_RESERVE_FLUSH_LIMIT
&&
4079 (flush_state
== FLUSH_DELALLOC
||
4080 flush_state
== FLUSH_DELALLOC_WAIT
))
4081 flush_state
= ALLOC_CHUNK
;
4085 else if (flush
== BTRFS_RESERVE_FLUSH_LIMIT
&&
4086 flush_state
< COMMIT_TRANS
)
4088 else if (flush
== BTRFS_RESERVE_FLUSH_ALL
&&
4089 flush_state
<= COMMIT_TRANS
)
4094 spin_lock(&space_info
->lock
);
4095 space_info
->flush
= 0;
4096 wake_up_all(&space_info
->wait
);
4097 spin_unlock(&space_info
->lock
);
4102 static struct btrfs_block_rsv
*get_block_rsv(
4103 const struct btrfs_trans_handle
*trans
,
4104 const struct btrfs_root
*root
)
4106 struct btrfs_block_rsv
*block_rsv
= NULL
;
4109 block_rsv
= trans
->block_rsv
;
4111 if (root
== root
->fs_info
->csum_root
&& trans
->adding_csums
)
4112 block_rsv
= trans
->block_rsv
;
4115 block_rsv
= root
->block_rsv
;
4118 block_rsv
= &root
->fs_info
->empty_block_rsv
;
4123 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
4127 spin_lock(&block_rsv
->lock
);
4128 if (block_rsv
->reserved
>= num_bytes
) {
4129 block_rsv
->reserved
-= num_bytes
;
4130 if (block_rsv
->reserved
< block_rsv
->size
)
4131 block_rsv
->full
= 0;
4134 spin_unlock(&block_rsv
->lock
);
4138 static void block_rsv_add_bytes(struct btrfs_block_rsv
*block_rsv
,
4139 u64 num_bytes
, int update_size
)
4141 spin_lock(&block_rsv
->lock
);
4142 block_rsv
->reserved
+= num_bytes
;
4144 block_rsv
->size
+= num_bytes
;
4145 else if (block_rsv
->reserved
>= block_rsv
->size
)
4146 block_rsv
->full
= 1;
4147 spin_unlock(&block_rsv
->lock
);
4150 static void block_rsv_release_bytes(struct btrfs_fs_info
*fs_info
,
4151 struct btrfs_block_rsv
*block_rsv
,
4152 struct btrfs_block_rsv
*dest
, u64 num_bytes
)
4154 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
4156 spin_lock(&block_rsv
->lock
);
4157 if (num_bytes
== (u64
)-1)
4158 num_bytes
= block_rsv
->size
;
4159 block_rsv
->size
-= num_bytes
;
4160 if (block_rsv
->reserved
>= block_rsv
->size
) {
4161 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
4162 block_rsv
->reserved
= block_rsv
->size
;
4163 block_rsv
->full
= 1;
4167 spin_unlock(&block_rsv
->lock
);
4169 if (num_bytes
> 0) {
4171 spin_lock(&dest
->lock
);
4175 bytes_to_add
= dest
->size
- dest
->reserved
;
4176 bytes_to_add
= min(num_bytes
, bytes_to_add
);
4177 dest
->reserved
+= bytes_to_add
;
4178 if (dest
->reserved
>= dest
->size
)
4180 num_bytes
-= bytes_to_add
;
4182 spin_unlock(&dest
->lock
);
4185 spin_lock(&space_info
->lock
);
4186 space_info
->bytes_may_use
-= num_bytes
;
4187 trace_btrfs_space_reservation(fs_info
, "space_info",
4188 space_info
->flags
, num_bytes
, 0);
4189 space_info
->reservation_progress
++;
4190 spin_unlock(&space_info
->lock
);
4195 static int block_rsv_migrate_bytes(struct btrfs_block_rsv
*src
,
4196 struct btrfs_block_rsv
*dst
, u64 num_bytes
)
4200 ret
= block_rsv_use_bytes(src
, num_bytes
);
4204 block_rsv_add_bytes(dst
, num_bytes
, 1);
4208 void btrfs_init_block_rsv(struct btrfs_block_rsv
*rsv
, unsigned short type
)
4210 memset(rsv
, 0, sizeof(*rsv
));
4211 spin_lock_init(&rsv
->lock
);
4215 struct btrfs_block_rsv
*btrfs_alloc_block_rsv(struct btrfs_root
*root
,
4216 unsigned short type
)
4218 struct btrfs_block_rsv
*block_rsv
;
4219 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4221 block_rsv
= kmalloc(sizeof(*block_rsv
), GFP_NOFS
);
4225 btrfs_init_block_rsv(block_rsv
, type
);
4226 block_rsv
->space_info
= __find_space_info(fs_info
,
4227 BTRFS_BLOCK_GROUP_METADATA
);
4231 void btrfs_free_block_rsv(struct btrfs_root
*root
,
4232 struct btrfs_block_rsv
*rsv
)
4236 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
4240 int btrfs_block_rsv_add(struct btrfs_root
*root
,
4241 struct btrfs_block_rsv
*block_rsv
, u64 num_bytes
,
4242 enum btrfs_reserve_flush_enum flush
)
4249 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
4251 block_rsv_add_bytes(block_rsv
, num_bytes
, 1);
4258 int btrfs_block_rsv_check(struct btrfs_root
*root
,
4259 struct btrfs_block_rsv
*block_rsv
, int min_factor
)
4267 spin_lock(&block_rsv
->lock
);
4268 num_bytes
= div_factor(block_rsv
->size
, min_factor
);
4269 if (block_rsv
->reserved
>= num_bytes
)
4271 spin_unlock(&block_rsv
->lock
);
4276 int btrfs_block_rsv_refill(struct btrfs_root
*root
,
4277 struct btrfs_block_rsv
*block_rsv
, u64 min_reserved
,
4278 enum btrfs_reserve_flush_enum flush
)
4286 spin_lock(&block_rsv
->lock
);
4287 num_bytes
= min_reserved
;
4288 if (block_rsv
->reserved
>= num_bytes
)
4291 num_bytes
-= block_rsv
->reserved
;
4292 spin_unlock(&block_rsv
->lock
);
4297 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
4299 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
4306 int btrfs_block_rsv_migrate(struct btrfs_block_rsv
*src_rsv
,
4307 struct btrfs_block_rsv
*dst_rsv
,
4310 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
4313 void btrfs_block_rsv_release(struct btrfs_root
*root
,
4314 struct btrfs_block_rsv
*block_rsv
,
4317 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
4318 if (global_rsv
->full
|| global_rsv
== block_rsv
||
4319 block_rsv
->space_info
!= global_rsv
->space_info
)
4321 block_rsv_release_bytes(root
->fs_info
, block_rsv
, global_rsv
,
4326 * helper to calculate size of global block reservation.
4327 * the desired value is sum of space used by extent tree,
4328 * checksum tree and root tree
4330 static u64
calc_global_metadata_size(struct btrfs_fs_info
*fs_info
)
4332 struct btrfs_space_info
*sinfo
;
4336 int csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
4338 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_DATA
);
4339 spin_lock(&sinfo
->lock
);
4340 data_used
= sinfo
->bytes_used
;
4341 spin_unlock(&sinfo
->lock
);
4343 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4344 spin_lock(&sinfo
->lock
);
4345 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_DATA
)
4347 meta_used
= sinfo
->bytes_used
;
4348 spin_unlock(&sinfo
->lock
);
4350 num_bytes
= (data_used
>> fs_info
->sb
->s_blocksize_bits
) *
4352 num_bytes
+= div64_u64(data_used
+ meta_used
, 50);
4354 if (num_bytes
* 3 > meta_used
)
4355 num_bytes
= div64_u64(meta_used
, 3);
4357 return ALIGN(num_bytes
, fs_info
->extent_root
->leafsize
<< 10);
4360 static void update_global_block_rsv(struct btrfs_fs_info
*fs_info
)
4362 struct btrfs_block_rsv
*block_rsv
= &fs_info
->global_block_rsv
;
4363 struct btrfs_space_info
*sinfo
= block_rsv
->space_info
;
4366 num_bytes
= calc_global_metadata_size(fs_info
);
4368 spin_lock(&sinfo
->lock
);
4369 spin_lock(&block_rsv
->lock
);
4371 block_rsv
->size
= num_bytes
;
4373 num_bytes
= sinfo
->bytes_used
+ sinfo
->bytes_pinned
+
4374 sinfo
->bytes_reserved
+ sinfo
->bytes_readonly
+
4375 sinfo
->bytes_may_use
;
4377 if (sinfo
->total_bytes
> num_bytes
) {
4378 num_bytes
= sinfo
->total_bytes
- num_bytes
;
4379 block_rsv
->reserved
+= num_bytes
;
4380 sinfo
->bytes_may_use
+= num_bytes
;
4381 trace_btrfs_space_reservation(fs_info
, "space_info",
4382 sinfo
->flags
, num_bytes
, 1);
4385 if (block_rsv
->reserved
>= block_rsv
->size
) {
4386 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
4387 sinfo
->bytes_may_use
-= num_bytes
;
4388 trace_btrfs_space_reservation(fs_info
, "space_info",
4389 sinfo
->flags
, num_bytes
, 0);
4390 sinfo
->reservation_progress
++;
4391 block_rsv
->reserved
= block_rsv
->size
;
4392 block_rsv
->full
= 1;
4395 spin_unlock(&block_rsv
->lock
);
4396 spin_unlock(&sinfo
->lock
);
4399 static void init_global_block_rsv(struct btrfs_fs_info
*fs_info
)
4401 struct btrfs_space_info
*space_info
;
4403 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
4404 fs_info
->chunk_block_rsv
.space_info
= space_info
;
4406 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4407 fs_info
->global_block_rsv
.space_info
= space_info
;
4408 fs_info
->delalloc_block_rsv
.space_info
= space_info
;
4409 fs_info
->trans_block_rsv
.space_info
= space_info
;
4410 fs_info
->empty_block_rsv
.space_info
= space_info
;
4411 fs_info
->delayed_block_rsv
.space_info
= space_info
;
4413 fs_info
->extent_root
->block_rsv
= &fs_info
->global_block_rsv
;
4414 fs_info
->csum_root
->block_rsv
= &fs_info
->global_block_rsv
;
4415 fs_info
->dev_root
->block_rsv
= &fs_info
->global_block_rsv
;
4416 fs_info
->tree_root
->block_rsv
= &fs_info
->global_block_rsv
;
4417 fs_info
->chunk_root
->block_rsv
= &fs_info
->chunk_block_rsv
;
4419 update_global_block_rsv(fs_info
);
4422 static void release_global_block_rsv(struct btrfs_fs_info
*fs_info
)
4424 block_rsv_release_bytes(fs_info
, &fs_info
->global_block_rsv
, NULL
,
4426 WARN_ON(fs_info
->delalloc_block_rsv
.size
> 0);
4427 WARN_ON(fs_info
->delalloc_block_rsv
.reserved
> 0);
4428 WARN_ON(fs_info
->trans_block_rsv
.size
> 0);
4429 WARN_ON(fs_info
->trans_block_rsv
.reserved
> 0);
4430 WARN_ON(fs_info
->chunk_block_rsv
.size
> 0);
4431 WARN_ON(fs_info
->chunk_block_rsv
.reserved
> 0);
4432 WARN_ON(fs_info
->delayed_block_rsv
.size
> 0);
4433 WARN_ON(fs_info
->delayed_block_rsv
.reserved
> 0);
4436 void btrfs_trans_release_metadata(struct btrfs_trans_handle
*trans
,
4437 struct btrfs_root
*root
)
4439 if (!trans
->block_rsv
)
4442 if (!trans
->bytes_reserved
)
4445 trace_btrfs_space_reservation(root
->fs_info
, "transaction",
4446 trans
->transid
, trans
->bytes_reserved
, 0);
4447 btrfs_block_rsv_release(root
, trans
->block_rsv
, trans
->bytes_reserved
);
4448 trans
->bytes_reserved
= 0;
4451 /* Can only return 0 or -ENOSPC */
4452 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle
*trans
,
4453 struct inode
*inode
)
4455 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4456 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
4457 struct btrfs_block_rsv
*dst_rsv
= root
->orphan_block_rsv
;
4460 * We need to hold space in order to delete our orphan item once we've
4461 * added it, so this takes the reservation so we can release it later
4462 * when we are truly done with the orphan item.
4464 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
4465 trace_btrfs_space_reservation(root
->fs_info
, "orphan",
4466 btrfs_ino(inode
), num_bytes
, 1);
4467 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
4470 void btrfs_orphan_release_metadata(struct inode
*inode
)
4472 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4473 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
4474 trace_btrfs_space_reservation(root
->fs_info
, "orphan",
4475 btrfs_ino(inode
), num_bytes
, 0);
4476 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
, num_bytes
);
4479 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle
*trans
,
4480 struct btrfs_pending_snapshot
*pending
)
4482 struct btrfs_root
*root
= pending
->root
;
4483 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
4484 struct btrfs_block_rsv
*dst_rsv
= &pending
->block_rsv
;
4486 * two for root back/forward refs, two for directory entries,
4487 * one for root of the snapshot and one for parent inode.
4489 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 6);
4490 dst_rsv
->space_info
= src_rsv
->space_info
;
4491 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
4495 * drop_outstanding_extent - drop an outstanding extent
4496 * @inode: the inode we're dropping the extent for
4498 * This is called when we are freeing up an outstanding extent, either called
4499 * after an error or after an extent is written. This will return the number of
4500 * reserved extents that need to be freed. This must be called with
4501 * BTRFS_I(inode)->lock held.
4503 static unsigned drop_outstanding_extent(struct inode
*inode
)
4505 unsigned drop_inode_space
= 0;
4506 unsigned dropped_extents
= 0;
4508 BUG_ON(!BTRFS_I(inode
)->outstanding_extents
);
4509 BTRFS_I(inode
)->outstanding_extents
--;
4511 if (BTRFS_I(inode
)->outstanding_extents
== 0 &&
4512 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
4513 &BTRFS_I(inode
)->runtime_flags
))
4514 drop_inode_space
= 1;
4517 * If we have more or the same amount of outsanding extents than we have
4518 * reserved then we need to leave the reserved extents count alone.
4520 if (BTRFS_I(inode
)->outstanding_extents
>=
4521 BTRFS_I(inode
)->reserved_extents
)
4522 return drop_inode_space
;
4524 dropped_extents
= BTRFS_I(inode
)->reserved_extents
-
4525 BTRFS_I(inode
)->outstanding_extents
;
4526 BTRFS_I(inode
)->reserved_extents
-= dropped_extents
;
4527 return dropped_extents
+ drop_inode_space
;
4531 * calc_csum_metadata_size - return the amount of metada space that must be
4532 * reserved/free'd for the given bytes.
4533 * @inode: the inode we're manipulating
4534 * @num_bytes: the number of bytes in question
4535 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4537 * This adjusts the number of csum_bytes in the inode and then returns the
4538 * correct amount of metadata that must either be reserved or freed. We
4539 * calculate how many checksums we can fit into one leaf and then divide the
4540 * number of bytes that will need to be checksumed by this value to figure out
4541 * how many checksums will be required. If we are adding bytes then the number
4542 * may go up and we will return the number of additional bytes that must be
4543 * reserved. If it is going down we will return the number of bytes that must
4546 * This must be called with BTRFS_I(inode)->lock held.
4548 static u64
calc_csum_metadata_size(struct inode
*inode
, u64 num_bytes
,
4551 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4553 int num_csums_per_leaf
;
4557 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
&&
4558 BTRFS_I(inode
)->csum_bytes
== 0)
4561 old_csums
= (int)div64_u64(BTRFS_I(inode
)->csum_bytes
, root
->sectorsize
);
4563 BTRFS_I(inode
)->csum_bytes
+= num_bytes
;
4565 BTRFS_I(inode
)->csum_bytes
-= num_bytes
;
4566 csum_size
= BTRFS_LEAF_DATA_SIZE(root
) - sizeof(struct btrfs_item
);
4567 num_csums_per_leaf
= (int)div64_u64(csum_size
,
4568 sizeof(struct btrfs_csum_item
) +
4569 sizeof(struct btrfs_disk_key
));
4570 num_csums
= (int)div64_u64(BTRFS_I(inode
)->csum_bytes
, root
->sectorsize
);
4571 num_csums
= num_csums
+ num_csums_per_leaf
- 1;
4572 num_csums
= num_csums
/ num_csums_per_leaf
;
4574 old_csums
= old_csums
+ num_csums_per_leaf
- 1;
4575 old_csums
= old_csums
/ num_csums_per_leaf
;
4577 /* No change, no need to reserve more */
4578 if (old_csums
== num_csums
)
4582 return btrfs_calc_trans_metadata_size(root
,
4583 num_csums
- old_csums
);
4585 return btrfs_calc_trans_metadata_size(root
, old_csums
- num_csums
);
4588 int btrfs_delalloc_reserve_metadata(struct inode
*inode
, u64 num_bytes
)
4590 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4591 struct btrfs_block_rsv
*block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
4594 unsigned nr_extents
= 0;
4595 int extra_reserve
= 0;
4596 enum btrfs_reserve_flush_enum flush
= BTRFS_RESERVE_FLUSH_ALL
;
4598 bool delalloc_lock
= true;
4600 /* If we are a free space inode we need to not flush since we will be in
4601 * the middle of a transaction commit. We also don't need the delalloc
4602 * mutex since we won't race with anybody. We need this mostly to make
4603 * lockdep shut its filthy mouth.
4605 if (btrfs_is_free_space_inode(inode
)) {
4606 flush
= BTRFS_RESERVE_NO_FLUSH
;
4607 delalloc_lock
= false;
4610 if (flush
!= BTRFS_RESERVE_NO_FLUSH
&&
4611 btrfs_transaction_in_commit(root
->fs_info
))
4612 schedule_timeout(1);
4615 mutex_lock(&BTRFS_I(inode
)->delalloc_mutex
);
4617 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
4619 spin_lock(&BTRFS_I(inode
)->lock
);
4620 BTRFS_I(inode
)->outstanding_extents
++;
4622 if (BTRFS_I(inode
)->outstanding_extents
>
4623 BTRFS_I(inode
)->reserved_extents
)
4624 nr_extents
= BTRFS_I(inode
)->outstanding_extents
-
4625 BTRFS_I(inode
)->reserved_extents
;
4628 * Add an item to reserve for updating the inode when we complete the
4631 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
4632 &BTRFS_I(inode
)->runtime_flags
)) {
4637 to_reserve
= btrfs_calc_trans_metadata_size(root
, nr_extents
);
4638 to_reserve
+= calc_csum_metadata_size(inode
, num_bytes
, 1);
4639 csum_bytes
= BTRFS_I(inode
)->csum_bytes
;
4640 spin_unlock(&BTRFS_I(inode
)->lock
);
4642 if (root
->fs_info
->quota_enabled
)
4643 ret
= btrfs_qgroup_reserve(root
, num_bytes
+
4644 nr_extents
* root
->leafsize
);
4647 * ret != 0 here means the qgroup reservation failed, we go straight to
4648 * the shared error handling then.
4651 ret
= reserve_metadata_bytes(root
, block_rsv
,
4658 spin_lock(&BTRFS_I(inode
)->lock
);
4659 dropped
= drop_outstanding_extent(inode
);
4661 * If the inodes csum_bytes is the same as the original
4662 * csum_bytes then we know we haven't raced with any free()ers
4663 * so we can just reduce our inodes csum bytes and carry on.
4664 * Otherwise we have to do the normal free thing to account for
4665 * the case that the free side didn't free up its reserve
4666 * because of this outstanding reservation.
4668 if (BTRFS_I(inode
)->csum_bytes
== csum_bytes
)
4669 calc_csum_metadata_size(inode
, num_bytes
, 0);
4671 to_free
= calc_csum_metadata_size(inode
, num_bytes
, 0);
4672 spin_unlock(&BTRFS_I(inode
)->lock
);
4674 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
4677 btrfs_block_rsv_release(root
, block_rsv
, to_free
);
4678 trace_btrfs_space_reservation(root
->fs_info
,
4683 if (root
->fs_info
->quota_enabled
) {
4684 btrfs_qgroup_free(root
, num_bytes
+
4685 nr_extents
* root
->leafsize
);
4688 mutex_unlock(&BTRFS_I(inode
)->delalloc_mutex
);
4692 spin_lock(&BTRFS_I(inode
)->lock
);
4693 if (extra_reserve
) {
4694 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
4695 &BTRFS_I(inode
)->runtime_flags
);
4698 BTRFS_I(inode
)->reserved_extents
+= nr_extents
;
4699 spin_unlock(&BTRFS_I(inode
)->lock
);
4702 mutex_unlock(&BTRFS_I(inode
)->delalloc_mutex
);
4705 trace_btrfs_space_reservation(root
->fs_info
,"delalloc",
4706 btrfs_ino(inode
), to_reserve
, 1);
4707 block_rsv_add_bytes(block_rsv
, to_reserve
, 1);
4713 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4714 * @inode: the inode to release the reservation for
4715 * @num_bytes: the number of bytes we're releasing
4717 * This will release the metadata reservation for an inode. This can be called
4718 * once we complete IO for a given set of bytes to release their metadata
4721 void btrfs_delalloc_release_metadata(struct inode
*inode
, u64 num_bytes
)
4723 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4727 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
4728 spin_lock(&BTRFS_I(inode
)->lock
);
4729 dropped
= drop_outstanding_extent(inode
);
4731 to_free
= calc_csum_metadata_size(inode
, num_bytes
, 0);
4732 spin_unlock(&BTRFS_I(inode
)->lock
);
4734 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
4736 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
4737 btrfs_ino(inode
), to_free
, 0);
4738 if (root
->fs_info
->quota_enabled
) {
4739 btrfs_qgroup_free(root
, num_bytes
+
4740 dropped
* root
->leafsize
);
4743 btrfs_block_rsv_release(root
, &root
->fs_info
->delalloc_block_rsv
,
4748 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4749 * @inode: inode we're writing to
4750 * @num_bytes: the number of bytes we want to allocate
4752 * This will do the following things
4754 * o reserve space in the data space info for num_bytes
4755 * o reserve space in the metadata space info based on number of outstanding
4756 * extents and how much csums will be needed
4757 * o add to the inodes ->delalloc_bytes
4758 * o add it to the fs_info's delalloc inodes list.
4760 * This will return 0 for success and -ENOSPC if there is no space left.
4762 int btrfs_delalloc_reserve_space(struct inode
*inode
, u64 num_bytes
)
4766 ret
= btrfs_check_data_free_space(inode
, num_bytes
);
4770 ret
= btrfs_delalloc_reserve_metadata(inode
, num_bytes
);
4772 btrfs_free_reserved_data_space(inode
, num_bytes
);
4780 * btrfs_delalloc_release_space - release data and metadata space for delalloc
4781 * @inode: inode we're releasing space for
4782 * @num_bytes: the number of bytes we want to free up
4784 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
4785 * called in the case that we don't need the metadata AND data reservations
4786 * anymore. So if there is an error or we insert an inline extent.
4788 * This function will release the metadata space that was not used and will
4789 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4790 * list if there are no delalloc bytes left.
4792 void btrfs_delalloc_release_space(struct inode
*inode
, u64 num_bytes
)
4794 btrfs_delalloc_release_metadata(inode
, num_bytes
);
4795 btrfs_free_reserved_data_space(inode
, num_bytes
);
4798 static int update_block_group(struct btrfs_root
*root
,
4799 u64 bytenr
, u64 num_bytes
, int alloc
)
4801 struct btrfs_block_group_cache
*cache
= NULL
;
4802 struct btrfs_fs_info
*info
= root
->fs_info
;
4803 u64 total
= num_bytes
;
4808 /* block accounting for super block */
4809 spin_lock(&info
->delalloc_lock
);
4810 old_val
= btrfs_super_bytes_used(info
->super_copy
);
4812 old_val
+= num_bytes
;
4814 old_val
-= num_bytes
;
4815 btrfs_set_super_bytes_used(info
->super_copy
, old_val
);
4816 spin_unlock(&info
->delalloc_lock
);
4819 cache
= btrfs_lookup_block_group(info
, bytenr
);
4822 if (cache
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
4823 BTRFS_BLOCK_GROUP_RAID1
|
4824 BTRFS_BLOCK_GROUP_RAID10
))
4829 * If this block group has free space cache written out, we
4830 * need to make sure to load it if we are removing space. This
4831 * is because we need the unpinning stage to actually add the
4832 * space back to the block group, otherwise we will leak space.
4834 if (!alloc
&& cache
->cached
== BTRFS_CACHE_NO
)
4835 cache_block_group(cache
, 1);
4837 byte_in_group
= bytenr
- cache
->key
.objectid
;
4838 WARN_ON(byte_in_group
> cache
->key
.offset
);
4840 spin_lock(&cache
->space_info
->lock
);
4841 spin_lock(&cache
->lock
);
4843 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
4844 cache
->disk_cache_state
< BTRFS_DC_CLEAR
)
4845 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
4848 old_val
= btrfs_block_group_used(&cache
->item
);
4849 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
4851 old_val
+= num_bytes
;
4852 btrfs_set_block_group_used(&cache
->item
, old_val
);
4853 cache
->reserved
-= num_bytes
;
4854 cache
->space_info
->bytes_reserved
-= num_bytes
;
4855 cache
->space_info
->bytes_used
+= num_bytes
;
4856 cache
->space_info
->disk_used
+= num_bytes
* factor
;
4857 spin_unlock(&cache
->lock
);
4858 spin_unlock(&cache
->space_info
->lock
);
4860 old_val
-= num_bytes
;
4861 btrfs_set_block_group_used(&cache
->item
, old_val
);
4862 cache
->pinned
+= num_bytes
;
4863 cache
->space_info
->bytes_pinned
+= num_bytes
;
4864 cache
->space_info
->bytes_used
-= num_bytes
;
4865 cache
->space_info
->disk_used
-= num_bytes
* factor
;
4866 spin_unlock(&cache
->lock
);
4867 spin_unlock(&cache
->space_info
->lock
);
4869 set_extent_dirty(info
->pinned_extents
,
4870 bytenr
, bytenr
+ num_bytes
- 1,
4871 GFP_NOFS
| __GFP_NOFAIL
);
4873 btrfs_put_block_group(cache
);
4875 bytenr
+= num_bytes
;
4880 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
4882 struct btrfs_block_group_cache
*cache
;
4885 spin_lock(&root
->fs_info
->block_group_cache_lock
);
4886 bytenr
= root
->fs_info
->first_logical_byte
;
4887 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
4889 if (bytenr
< (u64
)-1)
4892 cache
= btrfs_lookup_first_block_group(root
->fs_info
, search_start
);
4896 bytenr
= cache
->key
.objectid
;
4897 btrfs_put_block_group(cache
);
4902 static int pin_down_extent(struct btrfs_root
*root
,
4903 struct btrfs_block_group_cache
*cache
,
4904 u64 bytenr
, u64 num_bytes
, int reserved
)
4906 spin_lock(&cache
->space_info
->lock
);
4907 spin_lock(&cache
->lock
);
4908 cache
->pinned
+= num_bytes
;
4909 cache
->space_info
->bytes_pinned
+= num_bytes
;
4911 cache
->reserved
-= num_bytes
;
4912 cache
->space_info
->bytes_reserved
-= num_bytes
;
4914 spin_unlock(&cache
->lock
);
4915 spin_unlock(&cache
->space_info
->lock
);
4917 set_extent_dirty(root
->fs_info
->pinned_extents
, bytenr
,
4918 bytenr
+ num_bytes
- 1, GFP_NOFS
| __GFP_NOFAIL
);
4923 * this function must be called within transaction
4925 int btrfs_pin_extent(struct btrfs_root
*root
,
4926 u64 bytenr
, u64 num_bytes
, int reserved
)
4928 struct btrfs_block_group_cache
*cache
;
4930 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
4931 BUG_ON(!cache
); /* Logic error */
4933 pin_down_extent(root
, cache
, bytenr
, num_bytes
, reserved
);
4935 btrfs_put_block_group(cache
);
4940 * this function must be called within transaction
4942 int btrfs_pin_extent_for_log_replay(struct btrfs_root
*root
,
4943 u64 bytenr
, u64 num_bytes
)
4945 struct btrfs_block_group_cache
*cache
;
4947 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
4948 BUG_ON(!cache
); /* Logic error */
4951 * pull in the free space cache (if any) so that our pin
4952 * removes the free space from the cache. We have load_only set
4953 * to one because the slow code to read in the free extents does check
4954 * the pinned extents.
4956 cache_block_group(cache
, 1);
4958 pin_down_extent(root
, cache
, bytenr
, num_bytes
, 0);
4960 /* remove us from the free space cache (if we're there at all) */
4961 btrfs_remove_free_space(cache
, bytenr
, num_bytes
);
4962 btrfs_put_block_group(cache
);
4967 * btrfs_update_reserved_bytes - update the block_group and space info counters
4968 * @cache: The cache we are manipulating
4969 * @num_bytes: The number of bytes in question
4970 * @reserve: One of the reservation enums
4972 * This is called by the allocator when it reserves space, or by somebody who is
4973 * freeing space that was never actually used on disk. For example if you
4974 * reserve some space for a new leaf in transaction A and before transaction A
4975 * commits you free that leaf, you call this with reserve set to 0 in order to
4976 * clear the reservation.
4978 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4979 * ENOSPC accounting. For data we handle the reservation through clearing the
4980 * delalloc bits in the io_tree. We have to do this since we could end up
4981 * allocating less disk space for the amount of data we have reserved in the
4982 * case of compression.
4984 * If this is a reservation and the block group has become read only we cannot
4985 * make the reservation and return -EAGAIN, otherwise this function always
4988 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
4989 u64 num_bytes
, int reserve
)
4991 struct btrfs_space_info
*space_info
= cache
->space_info
;
4994 spin_lock(&space_info
->lock
);
4995 spin_lock(&cache
->lock
);
4996 if (reserve
!= RESERVE_FREE
) {
5000 cache
->reserved
+= num_bytes
;
5001 space_info
->bytes_reserved
+= num_bytes
;
5002 if (reserve
== RESERVE_ALLOC
) {
5003 trace_btrfs_space_reservation(cache
->fs_info
,
5004 "space_info", space_info
->flags
,
5006 space_info
->bytes_may_use
-= num_bytes
;
5011 space_info
->bytes_readonly
+= num_bytes
;
5012 cache
->reserved
-= num_bytes
;
5013 space_info
->bytes_reserved
-= num_bytes
;
5014 space_info
->reservation_progress
++;
5016 spin_unlock(&cache
->lock
);
5017 spin_unlock(&space_info
->lock
);
5021 void btrfs_prepare_extent_commit(struct btrfs_trans_handle
*trans
,
5022 struct btrfs_root
*root
)
5024 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5025 struct btrfs_caching_control
*next
;
5026 struct btrfs_caching_control
*caching_ctl
;
5027 struct btrfs_block_group_cache
*cache
;
5029 down_write(&fs_info
->extent_commit_sem
);
5031 list_for_each_entry_safe(caching_ctl
, next
,
5032 &fs_info
->caching_block_groups
, list
) {
5033 cache
= caching_ctl
->block_group
;
5034 if (block_group_cache_done(cache
)) {
5035 cache
->last_byte_to_unpin
= (u64
)-1;
5036 list_del_init(&caching_ctl
->list
);
5037 put_caching_control(caching_ctl
);
5039 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
5043 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
5044 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
5046 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
5048 up_write(&fs_info
->extent_commit_sem
);
5050 update_global_block_rsv(fs_info
);
5053 static int unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
5055 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5056 struct btrfs_block_group_cache
*cache
= NULL
;
5057 struct btrfs_space_info
*space_info
;
5058 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5062 while (start
<= end
) {
5065 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
5067 btrfs_put_block_group(cache
);
5068 cache
= btrfs_lookup_block_group(fs_info
, start
);
5069 BUG_ON(!cache
); /* Logic error */
5072 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
5073 len
= min(len
, end
+ 1 - start
);
5075 if (start
< cache
->last_byte_to_unpin
) {
5076 len
= min(len
, cache
->last_byte_to_unpin
- start
);
5077 btrfs_add_free_space(cache
, start
, len
);
5081 space_info
= cache
->space_info
;
5083 spin_lock(&space_info
->lock
);
5084 spin_lock(&cache
->lock
);
5085 cache
->pinned
-= len
;
5086 space_info
->bytes_pinned
-= len
;
5088 space_info
->bytes_readonly
+= len
;
5091 spin_unlock(&cache
->lock
);
5092 if (!readonly
&& global_rsv
->space_info
== space_info
) {
5093 spin_lock(&global_rsv
->lock
);
5094 if (!global_rsv
->full
) {
5095 len
= min(len
, global_rsv
->size
-
5096 global_rsv
->reserved
);
5097 global_rsv
->reserved
+= len
;
5098 space_info
->bytes_may_use
+= len
;
5099 if (global_rsv
->reserved
>= global_rsv
->size
)
5100 global_rsv
->full
= 1;
5102 spin_unlock(&global_rsv
->lock
);
5104 spin_unlock(&space_info
->lock
);
5108 btrfs_put_block_group(cache
);
5112 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
5113 struct btrfs_root
*root
)
5115 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5116 struct extent_io_tree
*unpin
;
5124 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
5125 unpin
= &fs_info
->freed_extents
[1];
5127 unpin
= &fs_info
->freed_extents
[0];
5130 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
5131 EXTENT_DIRTY
, NULL
);
5135 if (btrfs_test_opt(root
, DISCARD
))
5136 ret
= btrfs_discard_extent(root
, start
,
5137 end
+ 1 - start
, NULL
);
5139 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
5140 unpin_extent_range(root
, start
, end
);
5147 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
5148 struct btrfs_root
*root
,
5149 u64 bytenr
, u64 num_bytes
, u64 parent
,
5150 u64 root_objectid
, u64 owner_objectid
,
5151 u64 owner_offset
, int refs_to_drop
,
5152 struct btrfs_delayed_extent_op
*extent_op
)
5154 struct btrfs_key key
;
5155 struct btrfs_path
*path
;
5156 struct btrfs_fs_info
*info
= root
->fs_info
;
5157 struct btrfs_root
*extent_root
= info
->extent_root
;
5158 struct extent_buffer
*leaf
;
5159 struct btrfs_extent_item
*ei
;
5160 struct btrfs_extent_inline_ref
*iref
;
5163 int extent_slot
= 0;
5164 int found_extent
= 0;
5169 path
= btrfs_alloc_path();
5174 path
->leave_spinning
= 1;
5176 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
5177 BUG_ON(!is_data
&& refs_to_drop
!= 1);
5179 ret
= lookup_extent_backref(trans
, extent_root
, path
, &iref
,
5180 bytenr
, num_bytes
, parent
,
5181 root_objectid
, owner_objectid
,
5184 extent_slot
= path
->slots
[0];
5185 while (extent_slot
>= 0) {
5186 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
5188 if (key
.objectid
!= bytenr
)
5190 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
5191 key
.offset
== num_bytes
) {
5195 if (path
->slots
[0] - extent_slot
> 5)
5199 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5200 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
5201 if (found_extent
&& item_size
< sizeof(*ei
))
5204 if (!found_extent
) {
5206 ret
= remove_extent_backref(trans
, extent_root
, path
,
5210 btrfs_abort_transaction(trans
, extent_root
, ret
);
5213 btrfs_release_path(path
);
5214 path
->leave_spinning
= 1;
5216 key
.objectid
= bytenr
;
5217 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
5218 key
.offset
= num_bytes
;
5220 ret
= btrfs_search_slot(trans
, extent_root
,
5223 printk(KERN_ERR
"umm, got %d back from search"
5224 ", was looking for %llu\n", ret
,
5225 (unsigned long long)bytenr
);
5227 btrfs_print_leaf(extent_root
,
5231 btrfs_abort_transaction(trans
, extent_root
, ret
);
5234 extent_slot
= path
->slots
[0];
5236 } else if (ret
== -ENOENT
) {
5237 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
5239 printk(KERN_ERR
"btrfs unable to find ref byte nr %llu "
5240 "parent %llu root %llu owner %llu offset %llu\n",
5241 (unsigned long long)bytenr
,
5242 (unsigned long long)parent
,
5243 (unsigned long long)root_objectid
,
5244 (unsigned long long)owner_objectid
,
5245 (unsigned long long)owner_offset
);
5247 btrfs_abort_transaction(trans
, extent_root
, ret
);
5251 leaf
= path
->nodes
[0];
5252 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
5253 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5254 if (item_size
< sizeof(*ei
)) {
5255 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
5256 ret
= convert_extent_item_v0(trans
, extent_root
, path
,
5259 btrfs_abort_transaction(trans
, extent_root
, ret
);
5263 btrfs_release_path(path
);
5264 path
->leave_spinning
= 1;
5266 key
.objectid
= bytenr
;
5267 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
5268 key
.offset
= num_bytes
;
5270 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
5273 printk(KERN_ERR
"umm, got %d back from search"
5274 ", was looking for %llu\n", ret
,
5275 (unsigned long long)bytenr
);
5276 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
5279 btrfs_abort_transaction(trans
, extent_root
, ret
);
5283 extent_slot
= path
->slots
[0];
5284 leaf
= path
->nodes
[0];
5285 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
5288 BUG_ON(item_size
< sizeof(*ei
));
5289 ei
= btrfs_item_ptr(leaf
, extent_slot
,
5290 struct btrfs_extent_item
);
5291 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
) {
5292 struct btrfs_tree_block_info
*bi
;
5293 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
5294 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
5295 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
5298 refs
= btrfs_extent_refs(leaf
, ei
);
5299 BUG_ON(refs
< refs_to_drop
);
5300 refs
-= refs_to_drop
;
5304 __run_delayed_extent_op(extent_op
, leaf
, ei
);
5306 * In the case of inline back ref, reference count will
5307 * be updated by remove_extent_backref
5310 BUG_ON(!found_extent
);
5312 btrfs_set_extent_refs(leaf
, ei
, refs
);
5313 btrfs_mark_buffer_dirty(leaf
);
5316 ret
= remove_extent_backref(trans
, extent_root
, path
,
5320 btrfs_abort_transaction(trans
, extent_root
, ret
);
5326 BUG_ON(is_data
&& refs_to_drop
!=
5327 extent_data_ref_count(root
, path
, iref
));
5329 BUG_ON(path
->slots
[0] != extent_slot
);
5331 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
5332 path
->slots
[0] = extent_slot
;
5337 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
5340 btrfs_abort_transaction(trans
, extent_root
, ret
);
5343 btrfs_release_path(path
);
5346 ret
= btrfs_del_csums(trans
, root
, bytenr
, num_bytes
);
5348 btrfs_abort_transaction(trans
, extent_root
, ret
);
5353 ret
= update_block_group(root
, bytenr
, num_bytes
, 0);
5355 btrfs_abort_transaction(trans
, extent_root
, ret
);
5360 btrfs_free_path(path
);
5365 * when we free an block, it is possible (and likely) that we free the last
5366 * delayed ref for that extent as well. This searches the delayed ref tree for
5367 * a given extent, and if there are no other delayed refs to be processed, it
5368 * removes it from the tree.
5370 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
5371 struct btrfs_root
*root
, u64 bytenr
)
5373 struct btrfs_delayed_ref_head
*head
;
5374 struct btrfs_delayed_ref_root
*delayed_refs
;
5375 struct btrfs_delayed_ref_node
*ref
;
5376 struct rb_node
*node
;
5379 delayed_refs
= &trans
->transaction
->delayed_refs
;
5380 spin_lock(&delayed_refs
->lock
);
5381 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
5385 node
= rb_prev(&head
->node
.rb_node
);
5389 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
5391 /* there are still entries for this ref, we can't drop it */
5392 if (ref
->bytenr
== bytenr
)
5395 if (head
->extent_op
) {
5396 if (!head
->must_insert_reserved
)
5398 btrfs_free_delayed_extent_op(head
->extent_op
);
5399 head
->extent_op
= NULL
;
5403 * waiting for the lock here would deadlock. If someone else has it
5404 * locked they are already in the process of dropping it anyway
5406 if (!mutex_trylock(&head
->mutex
))
5410 * at this point we have a head with no other entries. Go
5411 * ahead and process it.
5413 head
->node
.in_tree
= 0;
5414 rb_erase(&head
->node
.rb_node
, &delayed_refs
->root
);
5416 delayed_refs
->num_entries
--;
5419 * we don't take a ref on the node because we're removing it from the
5420 * tree, so we just steal the ref the tree was holding.
5422 delayed_refs
->num_heads
--;
5423 if (list_empty(&head
->cluster
))
5424 delayed_refs
->num_heads_ready
--;
5426 list_del_init(&head
->cluster
);
5427 spin_unlock(&delayed_refs
->lock
);
5429 BUG_ON(head
->extent_op
);
5430 if (head
->must_insert_reserved
)
5433 mutex_unlock(&head
->mutex
);
5434 btrfs_put_delayed_ref(&head
->node
);
5437 spin_unlock(&delayed_refs
->lock
);
5441 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
5442 struct btrfs_root
*root
,
5443 struct extent_buffer
*buf
,
5444 u64 parent
, int last_ref
)
5446 struct btrfs_block_group_cache
*cache
= NULL
;
5449 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
5450 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
5451 buf
->start
, buf
->len
,
5452 parent
, root
->root_key
.objectid
,
5453 btrfs_header_level(buf
),
5454 BTRFS_DROP_DELAYED_REF
, NULL
, 0);
5455 BUG_ON(ret
); /* -ENOMEM */
5461 cache
= btrfs_lookup_block_group(root
->fs_info
, buf
->start
);
5463 if (btrfs_header_generation(buf
) == trans
->transid
) {
5464 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
5465 ret
= check_ref_cleanup(trans
, root
, buf
->start
);
5470 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
5471 pin_down_extent(root
, cache
, buf
->start
, buf
->len
, 1);
5475 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
5477 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
5478 btrfs_update_reserved_bytes(cache
, buf
->len
, RESERVE_FREE
);
5482 * Deleting the buffer, clear the corrupt flag since it doesn't matter
5485 clear_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
);
5486 btrfs_put_block_group(cache
);
5489 /* Can return -ENOMEM */
5490 int btrfs_free_extent(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
5491 u64 bytenr
, u64 num_bytes
, u64 parent
, u64 root_objectid
,
5492 u64 owner
, u64 offset
, int for_cow
)
5495 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5498 * tree log blocks never actually go into the extent allocation
5499 * tree, just update pinning info and exit early.
5501 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
5502 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
5503 /* unlocks the pinned mutex */
5504 btrfs_pin_extent(root
, bytenr
, num_bytes
, 1);
5506 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
5507 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
5509 parent
, root_objectid
, (int)owner
,
5510 BTRFS_DROP_DELAYED_REF
, NULL
, for_cow
);
5512 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
5514 parent
, root_objectid
, owner
,
5515 offset
, BTRFS_DROP_DELAYED_REF
,
5521 static u64
stripe_align(struct btrfs_root
*root
, u64 val
)
5523 u64 mask
= ((u64
)root
->stripesize
- 1);
5524 u64 ret
= (val
+ mask
) & ~mask
;
5529 * when we wait for progress in the block group caching, its because
5530 * our allocation attempt failed at least once. So, we must sleep
5531 * and let some progress happen before we try again.
5533 * This function will sleep at least once waiting for new free space to
5534 * show up, and then it will check the block group free space numbers
5535 * for our min num_bytes. Another option is to have it go ahead
5536 * and look in the rbtree for a free extent of a given size, but this
5540 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
5543 struct btrfs_caching_control
*caching_ctl
;
5546 caching_ctl
= get_caching_control(cache
);
5550 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
5551 (cache
->free_space_ctl
->free_space
>= num_bytes
));
5553 put_caching_control(caching_ctl
);
5558 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
5560 struct btrfs_caching_control
*caching_ctl
;
5563 caching_ctl
= get_caching_control(cache
);
5567 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
5569 put_caching_control(caching_ctl
);
5573 int __get_raid_index(u64 flags
)
5575 if (flags
& BTRFS_BLOCK_GROUP_RAID10
)
5576 return BTRFS_RAID_RAID10
;
5577 else if (flags
& BTRFS_BLOCK_GROUP_RAID1
)
5578 return BTRFS_RAID_RAID1
;
5579 else if (flags
& BTRFS_BLOCK_GROUP_DUP
)
5580 return BTRFS_RAID_DUP
;
5581 else if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
5582 return BTRFS_RAID_RAID0
;
5584 return BTRFS_RAID_SINGLE
;
5587 static int get_block_group_index(struct btrfs_block_group_cache
*cache
)
5589 return __get_raid_index(cache
->flags
);
5592 enum btrfs_loop_type
{
5593 LOOP_CACHING_NOWAIT
= 0,
5594 LOOP_CACHING_WAIT
= 1,
5595 LOOP_ALLOC_CHUNK
= 2,
5596 LOOP_NO_EMPTY_SIZE
= 3,
5600 * walks the btree of allocated extents and find a hole of a given size.
5601 * The key ins is changed to record the hole:
5602 * ins->objectid == block start
5603 * ins->flags = BTRFS_EXTENT_ITEM_KEY
5604 * ins->offset == number of blocks
5605 * Any available blocks before search_start are skipped.
5607 static noinline
int find_free_extent(struct btrfs_trans_handle
*trans
,
5608 struct btrfs_root
*orig_root
,
5609 u64 num_bytes
, u64 empty_size
,
5610 u64 hint_byte
, struct btrfs_key
*ins
,
5614 struct btrfs_root
*root
= orig_root
->fs_info
->extent_root
;
5615 struct btrfs_free_cluster
*last_ptr
= NULL
;
5616 struct btrfs_block_group_cache
*block_group
= NULL
;
5617 struct btrfs_block_group_cache
*used_block_group
;
5618 u64 search_start
= 0;
5619 int empty_cluster
= 2 * 1024 * 1024;
5620 struct btrfs_space_info
*space_info
;
5622 int index
= __get_raid_index(data
);
5623 int alloc_type
= (data
& BTRFS_BLOCK_GROUP_DATA
) ?
5624 RESERVE_ALLOC_NO_ACCOUNT
: RESERVE_ALLOC
;
5625 bool found_uncached_bg
= false;
5626 bool failed_cluster_refill
= false;
5627 bool failed_alloc
= false;
5628 bool use_cluster
= true;
5629 bool have_caching_bg
= false;
5631 WARN_ON(num_bytes
< root
->sectorsize
);
5632 btrfs_set_key_type(ins
, BTRFS_EXTENT_ITEM_KEY
);
5636 trace_find_free_extent(orig_root
, num_bytes
, empty_size
, data
);
5638 space_info
= __find_space_info(root
->fs_info
, data
);
5640 printk(KERN_ERR
"No space info for %llu\n", data
);
5645 * If the space info is for both data and metadata it means we have a
5646 * small filesystem and we can't use the clustering stuff.
5648 if (btrfs_mixed_space_info(space_info
))
5649 use_cluster
= false;
5651 if (data
& BTRFS_BLOCK_GROUP_METADATA
&& use_cluster
) {
5652 last_ptr
= &root
->fs_info
->meta_alloc_cluster
;
5653 if (!btrfs_test_opt(root
, SSD
))
5654 empty_cluster
= 64 * 1024;
5657 if ((data
& BTRFS_BLOCK_GROUP_DATA
) && use_cluster
&&
5658 btrfs_test_opt(root
, SSD
)) {
5659 last_ptr
= &root
->fs_info
->data_alloc_cluster
;
5663 spin_lock(&last_ptr
->lock
);
5664 if (last_ptr
->block_group
)
5665 hint_byte
= last_ptr
->window_start
;
5666 spin_unlock(&last_ptr
->lock
);
5669 search_start
= max(search_start
, first_logical_byte(root
, 0));
5670 search_start
= max(search_start
, hint_byte
);
5675 if (search_start
== hint_byte
) {
5676 block_group
= btrfs_lookup_block_group(root
->fs_info
,
5678 used_block_group
= block_group
;
5680 * we don't want to use the block group if it doesn't match our
5681 * allocation bits, or if its not cached.
5683 * However if we are re-searching with an ideal block group
5684 * picked out then we don't care that the block group is cached.
5686 if (block_group
&& block_group_bits(block_group
, data
) &&
5687 block_group
->cached
!= BTRFS_CACHE_NO
) {
5688 down_read(&space_info
->groups_sem
);
5689 if (list_empty(&block_group
->list
) ||
5692 * someone is removing this block group,
5693 * we can't jump into the have_block_group
5694 * target because our list pointers are not
5697 btrfs_put_block_group(block_group
);
5698 up_read(&space_info
->groups_sem
);
5700 index
= get_block_group_index(block_group
);
5701 goto have_block_group
;
5703 } else if (block_group
) {
5704 btrfs_put_block_group(block_group
);
5708 have_caching_bg
= false;
5709 down_read(&space_info
->groups_sem
);
5710 list_for_each_entry(block_group
, &space_info
->block_groups
[index
],
5715 used_block_group
= block_group
;
5716 btrfs_get_block_group(block_group
);
5717 search_start
= block_group
->key
.objectid
;
5720 * this can happen if we end up cycling through all the
5721 * raid types, but we want to make sure we only allocate
5722 * for the proper type.
5724 if (!block_group_bits(block_group
, data
)) {
5725 u64 extra
= BTRFS_BLOCK_GROUP_DUP
|
5726 BTRFS_BLOCK_GROUP_RAID1
|
5727 BTRFS_BLOCK_GROUP_RAID10
;
5730 * if they asked for extra copies and this block group
5731 * doesn't provide them, bail. This does allow us to
5732 * fill raid0 from raid1.
5734 if ((data
& extra
) && !(block_group
->flags
& extra
))
5739 cached
= block_group_cache_done(block_group
);
5740 if (unlikely(!cached
)) {
5741 found_uncached_bg
= true;
5742 ret
= cache_block_group(block_group
, 0);
5747 if (unlikely(block_group
->ro
))
5751 * Ok we want to try and use the cluster allocator, so
5756 * the refill lock keeps out other
5757 * people trying to start a new cluster
5759 spin_lock(&last_ptr
->refill_lock
);
5760 used_block_group
= last_ptr
->block_group
;
5761 if (used_block_group
!= block_group
&&
5762 (!used_block_group
||
5763 used_block_group
->ro
||
5764 !block_group_bits(used_block_group
, data
))) {
5765 used_block_group
= block_group
;
5766 goto refill_cluster
;
5769 if (used_block_group
!= block_group
)
5770 btrfs_get_block_group(used_block_group
);
5772 offset
= btrfs_alloc_from_cluster(used_block_group
,
5773 last_ptr
, num_bytes
, used_block_group
->key
.objectid
);
5775 /* we have a block, we're done */
5776 spin_unlock(&last_ptr
->refill_lock
);
5777 trace_btrfs_reserve_extent_cluster(root
,
5778 block_group
, search_start
, num_bytes
);
5782 WARN_ON(last_ptr
->block_group
!= used_block_group
);
5783 if (used_block_group
!= block_group
) {
5784 btrfs_put_block_group(used_block_group
);
5785 used_block_group
= block_group
;
5788 BUG_ON(used_block_group
!= block_group
);
5789 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
5790 * set up a new clusters, so lets just skip it
5791 * and let the allocator find whatever block
5792 * it can find. If we reach this point, we
5793 * will have tried the cluster allocator
5794 * plenty of times and not have found
5795 * anything, so we are likely way too
5796 * fragmented for the clustering stuff to find
5799 * However, if the cluster is taken from the
5800 * current block group, release the cluster
5801 * first, so that we stand a better chance of
5802 * succeeding in the unclustered
5804 if (loop
>= LOOP_NO_EMPTY_SIZE
&&
5805 last_ptr
->block_group
!= block_group
) {
5806 spin_unlock(&last_ptr
->refill_lock
);
5807 goto unclustered_alloc
;
5811 * this cluster didn't work out, free it and
5814 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
5816 if (loop
>= LOOP_NO_EMPTY_SIZE
) {
5817 spin_unlock(&last_ptr
->refill_lock
);
5818 goto unclustered_alloc
;
5821 /* allocate a cluster in this block group */
5822 ret
= btrfs_find_space_cluster(trans
, root
,
5823 block_group
, last_ptr
,
5824 search_start
, num_bytes
,
5825 empty_cluster
+ empty_size
);
5828 * now pull our allocation out of this
5831 offset
= btrfs_alloc_from_cluster(block_group
,
5832 last_ptr
, num_bytes
,
5835 /* we found one, proceed */
5836 spin_unlock(&last_ptr
->refill_lock
);
5837 trace_btrfs_reserve_extent_cluster(root
,
5838 block_group
, search_start
,
5842 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
5843 && !failed_cluster_refill
) {
5844 spin_unlock(&last_ptr
->refill_lock
);
5846 failed_cluster_refill
= true;
5847 wait_block_group_cache_progress(block_group
,
5848 num_bytes
+ empty_cluster
+ empty_size
);
5849 goto have_block_group
;
5853 * at this point we either didn't find a cluster
5854 * or we weren't able to allocate a block from our
5855 * cluster. Free the cluster we've been trying
5856 * to use, and go to the next block group
5858 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
5859 spin_unlock(&last_ptr
->refill_lock
);
5864 spin_lock(&block_group
->free_space_ctl
->tree_lock
);
5866 block_group
->free_space_ctl
->free_space
<
5867 num_bytes
+ empty_cluster
+ empty_size
) {
5868 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
5871 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
5873 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
5874 num_bytes
, empty_size
);
5876 * If we didn't find a chunk, and we haven't failed on this
5877 * block group before, and this block group is in the middle of
5878 * caching and we are ok with waiting, then go ahead and wait
5879 * for progress to be made, and set failed_alloc to true.
5881 * If failed_alloc is true then we've already waited on this
5882 * block group once and should move on to the next block group.
5884 if (!offset
&& !failed_alloc
&& !cached
&&
5885 loop
> LOOP_CACHING_NOWAIT
) {
5886 wait_block_group_cache_progress(block_group
,
5887 num_bytes
+ empty_size
);
5888 failed_alloc
= true;
5889 goto have_block_group
;
5890 } else if (!offset
) {
5892 have_caching_bg
= true;
5896 search_start
= stripe_align(root
, offset
);
5898 /* move on to the next group */
5899 if (search_start
+ num_bytes
>
5900 used_block_group
->key
.objectid
+ used_block_group
->key
.offset
) {
5901 btrfs_add_free_space(used_block_group
, offset
, num_bytes
);
5905 if (offset
< search_start
)
5906 btrfs_add_free_space(used_block_group
, offset
,
5907 search_start
- offset
);
5908 BUG_ON(offset
> search_start
);
5910 ret
= btrfs_update_reserved_bytes(used_block_group
, num_bytes
,
5912 if (ret
== -EAGAIN
) {
5913 btrfs_add_free_space(used_block_group
, offset
, num_bytes
);
5917 /* we are all good, lets return */
5918 ins
->objectid
= search_start
;
5919 ins
->offset
= num_bytes
;
5921 trace_btrfs_reserve_extent(orig_root
, block_group
,
5922 search_start
, num_bytes
);
5923 if (used_block_group
!= block_group
)
5924 btrfs_put_block_group(used_block_group
);
5925 btrfs_put_block_group(block_group
);
5928 failed_cluster_refill
= false;
5929 failed_alloc
= false;
5930 BUG_ON(index
!= get_block_group_index(block_group
));
5931 if (used_block_group
!= block_group
)
5932 btrfs_put_block_group(used_block_group
);
5933 btrfs_put_block_group(block_group
);
5935 up_read(&space_info
->groups_sem
);
5937 if (!ins
->objectid
&& loop
>= LOOP_CACHING_WAIT
&& have_caching_bg
)
5940 if (!ins
->objectid
&& ++index
< BTRFS_NR_RAID_TYPES
)
5944 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5945 * caching kthreads as we move along
5946 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5947 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5948 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5951 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
) {
5954 if (loop
== LOOP_ALLOC_CHUNK
) {
5955 ret
= do_chunk_alloc(trans
, root
, data
,
5958 * Do not bail out on ENOSPC since we
5959 * can do more things.
5961 if (ret
< 0 && ret
!= -ENOSPC
) {
5962 btrfs_abort_transaction(trans
,
5968 if (loop
== LOOP_NO_EMPTY_SIZE
) {
5974 } else if (!ins
->objectid
) {
5976 } else if (ins
->objectid
) {
5984 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
5985 int dump_block_groups
)
5987 struct btrfs_block_group_cache
*cache
;
5990 spin_lock(&info
->lock
);
5991 printk(KERN_INFO
"space_info %llu has %llu free, is %sfull\n",
5992 (unsigned long long)info
->flags
,
5993 (unsigned long long)(info
->total_bytes
- info
->bytes_used
-
5994 info
->bytes_pinned
- info
->bytes_reserved
-
5995 info
->bytes_readonly
),
5996 (info
->full
) ? "" : "not ");
5997 printk(KERN_INFO
"space_info total=%llu, used=%llu, pinned=%llu, "
5998 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5999 (unsigned long long)info
->total_bytes
,
6000 (unsigned long long)info
->bytes_used
,
6001 (unsigned long long)info
->bytes_pinned
,
6002 (unsigned long long)info
->bytes_reserved
,
6003 (unsigned long long)info
->bytes_may_use
,
6004 (unsigned long long)info
->bytes_readonly
);
6005 spin_unlock(&info
->lock
);
6007 if (!dump_block_groups
)
6010 down_read(&info
->groups_sem
);
6012 list_for_each_entry(cache
, &info
->block_groups
[index
], list
) {
6013 spin_lock(&cache
->lock
);
6014 printk(KERN_INFO
"block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n",
6015 (unsigned long long)cache
->key
.objectid
,
6016 (unsigned long long)cache
->key
.offset
,
6017 (unsigned long long)btrfs_block_group_used(&cache
->item
),
6018 (unsigned long long)cache
->pinned
,
6019 (unsigned long long)cache
->reserved
,
6020 cache
->ro
? "[readonly]" : "");
6021 btrfs_dump_free_space(cache
, bytes
);
6022 spin_unlock(&cache
->lock
);
6024 if (++index
< BTRFS_NR_RAID_TYPES
)
6026 up_read(&info
->groups_sem
);
6029 int btrfs_reserve_extent(struct btrfs_trans_handle
*trans
,
6030 struct btrfs_root
*root
,
6031 u64 num_bytes
, u64 min_alloc_size
,
6032 u64 empty_size
, u64 hint_byte
,
6033 struct btrfs_key
*ins
, u64 data
)
6035 bool final_tried
= false;
6038 data
= btrfs_get_alloc_profile(root
, data
);
6040 WARN_ON(num_bytes
< root
->sectorsize
);
6041 ret
= find_free_extent(trans
, root
, num_bytes
, empty_size
,
6042 hint_byte
, ins
, data
);
6044 if (ret
== -ENOSPC
) {
6046 num_bytes
= num_bytes
>> 1;
6047 num_bytes
= num_bytes
& ~(root
->sectorsize
- 1);
6048 num_bytes
= max(num_bytes
, min_alloc_size
);
6049 if (num_bytes
== min_alloc_size
)
6052 } else if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
6053 struct btrfs_space_info
*sinfo
;
6055 sinfo
= __find_space_info(root
->fs_info
, data
);
6056 printk(KERN_ERR
"btrfs allocation failed flags %llu, "
6057 "wanted %llu\n", (unsigned long long)data
,
6058 (unsigned long long)num_bytes
);
6060 dump_space_info(sinfo
, num_bytes
, 1);
6064 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, ins
->offset
);
6069 static int __btrfs_free_reserved_extent(struct btrfs_root
*root
,
6070 u64 start
, u64 len
, int pin
)
6072 struct btrfs_block_group_cache
*cache
;
6075 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
6077 printk(KERN_ERR
"Unable to find block group for %llu\n",
6078 (unsigned long long)start
);
6082 if (btrfs_test_opt(root
, DISCARD
))
6083 ret
= btrfs_discard_extent(root
, start
, len
, NULL
);
6086 pin_down_extent(root
, cache
, start
, len
, 1);
6088 btrfs_add_free_space(cache
, start
, len
);
6089 btrfs_update_reserved_bytes(cache
, len
, RESERVE_FREE
);
6091 btrfs_put_block_group(cache
);
6093 trace_btrfs_reserved_extent_free(root
, start
, len
);
6098 int btrfs_free_reserved_extent(struct btrfs_root
*root
,
6101 return __btrfs_free_reserved_extent(root
, start
, len
, 0);
6104 int btrfs_free_and_pin_reserved_extent(struct btrfs_root
*root
,
6107 return __btrfs_free_reserved_extent(root
, start
, len
, 1);
6110 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
6111 struct btrfs_root
*root
,
6112 u64 parent
, u64 root_objectid
,
6113 u64 flags
, u64 owner
, u64 offset
,
6114 struct btrfs_key
*ins
, int ref_mod
)
6117 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6118 struct btrfs_extent_item
*extent_item
;
6119 struct btrfs_extent_inline_ref
*iref
;
6120 struct btrfs_path
*path
;
6121 struct extent_buffer
*leaf
;
6126 type
= BTRFS_SHARED_DATA_REF_KEY
;
6128 type
= BTRFS_EXTENT_DATA_REF_KEY
;
6130 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
6132 path
= btrfs_alloc_path();
6136 path
->leave_spinning
= 1;
6137 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
6140 btrfs_free_path(path
);
6144 leaf
= path
->nodes
[0];
6145 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
6146 struct btrfs_extent_item
);
6147 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
6148 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
6149 btrfs_set_extent_flags(leaf
, extent_item
,
6150 flags
| BTRFS_EXTENT_FLAG_DATA
);
6152 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
6153 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
6155 struct btrfs_shared_data_ref
*ref
;
6156 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
6157 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
6158 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
6160 struct btrfs_extent_data_ref
*ref
;
6161 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
6162 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
6163 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
6164 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
6165 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
6168 btrfs_mark_buffer_dirty(path
->nodes
[0]);
6169 btrfs_free_path(path
);
6171 ret
= update_block_group(root
, ins
->objectid
, ins
->offset
, 1);
6172 if (ret
) { /* -ENOENT, logic error */
6173 printk(KERN_ERR
"btrfs update block group failed for %llu "
6174 "%llu\n", (unsigned long long)ins
->objectid
,
6175 (unsigned long long)ins
->offset
);
6181 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
6182 struct btrfs_root
*root
,
6183 u64 parent
, u64 root_objectid
,
6184 u64 flags
, struct btrfs_disk_key
*key
,
6185 int level
, struct btrfs_key
*ins
)
6188 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6189 struct btrfs_extent_item
*extent_item
;
6190 struct btrfs_tree_block_info
*block_info
;
6191 struct btrfs_extent_inline_ref
*iref
;
6192 struct btrfs_path
*path
;
6193 struct extent_buffer
*leaf
;
6194 u32 size
= sizeof(*extent_item
) + sizeof(*block_info
) + sizeof(*iref
);
6196 path
= btrfs_alloc_path();
6200 path
->leave_spinning
= 1;
6201 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
6204 btrfs_free_path(path
);
6208 leaf
= path
->nodes
[0];
6209 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
6210 struct btrfs_extent_item
);
6211 btrfs_set_extent_refs(leaf
, extent_item
, 1);
6212 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
6213 btrfs_set_extent_flags(leaf
, extent_item
,
6214 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
6215 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
6217 btrfs_set_tree_block_key(leaf
, block_info
, key
);
6218 btrfs_set_tree_block_level(leaf
, block_info
, level
);
6220 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
6222 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
6223 btrfs_set_extent_inline_ref_type(leaf
, iref
,
6224 BTRFS_SHARED_BLOCK_REF_KEY
);
6225 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
6227 btrfs_set_extent_inline_ref_type(leaf
, iref
,
6228 BTRFS_TREE_BLOCK_REF_KEY
);
6229 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
6232 btrfs_mark_buffer_dirty(leaf
);
6233 btrfs_free_path(path
);
6235 ret
= update_block_group(root
, ins
->objectid
, ins
->offset
, 1);
6236 if (ret
) { /* -ENOENT, logic error */
6237 printk(KERN_ERR
"btrfs update block group failed for %llu "
6238 "%llu\n", (unsigned long long)ins
->objectid
,
6239 (unsigned long long)ins
->offset
);
6245 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
6246 struct btrfs_root
*root
,
6247 u64 root_objectid
, u64 owner
,
6248 u64 offset
, struct btrfs_key
*ins
)
6252 BUG_ON(root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
6254 ret
= btrfs_add_delayed_data_ref(root
->fs_info
, trans
, ins
->objectid
,
6256 root_objectid
, owner
, offset
,
6257 BTRFS_ADD_DELAYED_EXTENT
, NULL
, 0);
6262 * this is used by the tree logging recovery code. It records that
6263 * an extent has been allocated and makes sure to clear the free
6264 * space cache bits as well
6266 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
6267 struct btrfs_root
*root
,
6268 u64 root_objectid
, u64 owner
, u64 offset
,
6269 struct btrfs_key
*ins
)
6272 struct btrfs_block_group_cache
*block_group
;
6273 struct btrfs_caching_control
*caching_ctl
;
6274 u64 start
= ins
->objectid
;
6275 u64 num_bytes
= ins
->offset
;
6277 block_group
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
6278 cache_block_group(block_group
, 0);
6279 caching_ctl
= get_caching_control(block_group
);
6282 BUG_ON(!block_group_cache_done(block_group
));
6283 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
6284 BUG_ON(ret
); /* -ENOMEM */
6286 mutex_lock(&caching_ctl
->mutex
);
6288 if (start
>= caching_ctl
->progress
) {
6289 ret
= add_excluded_extent(root
, start
, num_bytes
);
6290 BUG_ON(ret
); /* -ENOMEM */
6291 } else if (start
+ num_bytes
<= caching_ctl
->progress
) {
6292 ret
= btrfs_remove_free_space(block_group
,
6294 BUG_ON(ret
); /* -ENOMEM */
6296 num_bytes
= caching_ctl
->progress
- start
;
6297 ret
= btrfs_remove_free_space(block_group
,
6299 BUG_ON(ret
); /* -ENOMEM */
6301 start
= caching_ctl
->progress
;
6302 num_bytes
= ins
->objectid
+ ins
->offset
-
6303 caching_ctl
->progress
;
6304 ret
= add_excluded_extent(root
, start
, num_bytes
);
6305 BUG_ON(ret
); /* -ENOMEM */
6308 mutex_unlock(&caching_ctl
->mutex
);
6309 put_caching_control(caching_ctl
);
6312 ret
= btrfs_update_reserved_bytes(block_group
, ins
->offset
,
6313 RESERVE_ALLOC_NO_ACCOUNT
);
6314 BUG_ON(ret
); /* logic error */
6315 btrfs_put_block_group(block_group
);
6316 ret
= alloc_reserved_file_extent(trans
, root
, 0, root_objectid
,
6317 0, owner
, offset
, ins
, 1);
6321 struct extent_buffer
*btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
,
6322 struct btrfs_root
*root
,
6323 u64 bytenr
, u32 blocksize
,
6326 struct extent_buffer
*buf
;
6328 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
6330 return ERR_PTR(-ENOMEM
);
6331 btrfs_set_header_generation(buf
, trans
->transid
);
6332 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, buf
, level
);
6333 btrfs_tree_lock(buf
);
6334 clean_tree_block(trans
, root
, buf
);
6335 clear_bit(EXTENT_BUFFER_STALE
, &buf
->bflags
);
6337 btrfs_set_lock_blocking(buf
);
6338 btrfs_set_buffer_uptodate(buf
);
6340 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
6342 * we allow two log transactions at a time, use different
6343 * EXENT bit to differentiate dirty pages.
6345 if (root
->log_transid
% 2 == 0)
6346 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
6347 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
6349 set_extent_new(&root
->dirty_log_pages
, buf
->start
,
6350 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
6352 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
6353 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
6355 trans
->blocks_used
++;
6356 /* this returns a buffer locked for blocking */
6360 static struct btrfs_block_rsv
*
6361 use_block_rsv(struct btrfs_trans_handle
*trans
,
6362 struct btrfs_root
*root
, u32 blocksize
)
6364 struct btrfs_block_rsv
*block_rsv
;
6365 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
6368 block_rsv
= get_block_rsv(trans
, root
);
6370 if (block_rsv
->size
== 0) {
6371 ret
= reserve_metadata_bytes(root
, block_rsv
, blocksize
,
6372 BTRFS_RESERVE_NO_FLUSH
);
6374 * If we couldn't reserve metadata bytes try and use some from
6375 * the global reserve.
6377 if (ret
&& block_rsv
!= global_rsv
) {
6378 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
6381 return ERR_PTR(ret
);
6383 return ERR_PTR(ret
);
6388 ret
= block_rsv_use_bytes(block_rsv
, blocksize
);
6391 if (ret
&& !block_rsv
->failfast
) {
6392 static DEFINE_RATELIMIT_STATE(_rs
,
6393 DEFAULT_RATELIMIT_INTERVAL
,
6394 /*DEFAULT_RATELIMIT_BURST*/ 2);
6395 if (__ratelimit(&_rs
))
6396 WARN(1, KERN_DEBUG
"btrfs: block rsv returned %d\n",
6398 ret
= reserve_metadata_bytes(root
, block_rsv
, blocksize
,
6399 BTRFS_RESERVE_NO_FLUSH
);
6402 } else if (ret
&& block_rsv
!= global_rsv
) {
6403 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
6409 return ERR_PTR(-ENOSPC
);
6412 static void unuse_block_rsv(struct btrfs_fs_info
*fs_info
,
6413 struct btrfs_block_rsv
*block_rsv
, u32 blocksize
)
6415 block_rsv_add_bytes(block_rsv
, blocksize
, 0);
6416 block_rsv_release_bytes(fs_info
, block_rsv
, NULL
, 0);
6420 * finds a free extent and does all the dirty work required for allocation
6421 * returns the key for the extent through ins, and a tree buffer for
6422 * the first block of the extent through buf.
6424 * returns the tree buffer or NULL.
6426 struct extent_buffer
*btrfs_alloc_free_block(struct btrfs_trans_handle
*trans
,
6427 struct btrfs_root
*root
, u32 blocksize
,
6428 u64 parent
, u64 root_objectid
,
6429 struct btrfs_disk_key
*key
, int level
,
6430 u64 hint
, u64 empty_size
)
6432 struct btrfs_key ins
;
6433 struct btrfs_block_rsv
*block_rsv
;
6434 struct extent_buffer
*buf
;
6439 block_rsv
= use_block_rsv(trans
, root
, blocksize
);
6440 if (IS_ERR(block_rsv
))
6441 return ERR_CAST(block_rsv
);
6443 ret
= btrfs_reserve_extent(trans
, root
, blocksize
, blocksize
,
6444 empty_size
, hint
, &ins
, 0);
6446 unuse_block_rsv(root
->fs_info
, block_rsv
, blocksize
);
6447 return ERR_PTR(ret
);
6450 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
,
6452 BUG_ON(IS_ERR(buf
)); /* -ENOMEM */
6454 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
6456 parent
= ins
.objectid
;
6457 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
6461 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
6462 struct btrfs_delayed_extent_op
*extent_op
;
6463 extent_op
= btrfs_alloc_delayed_extent_op();
6464 BUG_ON(!extent_op
); /* -ENOMEM */
6466 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
6468 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
6469 extent_op
->flags_to_set
= flags
;
6470 extent_op
->update_key
= 1;
6471 extent_op
->update_flags
= 1;
6472 extent_op
->is_data
= 0;
6474 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
6476 ins
.offset
, parent
, root_objectid
,
6477 level
, BTRFS_ADD_DELAYED_EXTENT
,
6479 BUG_ON(ret
); /* -ENOMEM */
6484 struct walk_control
{
6485 u64 refs
[BTRFS_MAX_LEVEL
];
6486 u64 flags
[BTRFS_MAX_LEVEL
];
6487 struct btrfs_key update_progress
;
6498 #define DROP_REFERENCE 1
6499 #define UPDATE_BACKREF 2
6501 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
6502 struct btrfs_root
*root
,
6503 struct walk_control
*wc
,
6504 struct btrfs_path
*path
)
6512 struct btrfs_key key
;
6513 struct extent_buffer
*eb
;
6518 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
6519 wc
->reada_count
= wc
->reada_count
* 2 / 3;
6520 wc
->reada_count
= max(wc
->reada_count
, 2);
6522 wc
->reada_count
= wc
->reada_count
* 3 / 2;
6523 wc
->reada_count
= min_t(int, wc
->reada_count
,
6524 BTRFS_NODEPTRS_PER_BLOCK(root
));
6527 eb
= path
->nodes
[wc
->level
];
6528 nritems
= btrfs_header_nritems(eb
);
6529 blocksize
= btrfs_level_size(root
, wc
->level
- 1);
6531 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
6532 if (nread
>= wc
->reada_count
)
6536 bytenr
= btrfs_node_blockptr(eb
, slot
);
6537 generation
= btrfs_node_ptr_generation(eb
, slot
);
6539 if (slot
== path
->slots
[wc
->level
])
6542 if (wc
->stage
== UPDATE_BACKREF
&&
6543 generation
<= root
->root_key
.offset
)
6546 /* We don't lock the tree block, it's OK to be racy here */
6547 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, blocksize
,
6549 /* We don't care about errors in readahead. */
6554 if (wc
->stage
== DROP_REFERENCE
) {
6558 if (wc
->level
== 1 &&
6559 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
6561 if (!wc
->update_ref
||
6562 generation
<= root
->root_key
.offset
)
6564 btrfs_node_key_to_cpu(eb
, &key
, slot
);
6565 ret
= btrfs_comp_cpu_keys(&key
,
6566 &wc
->update_progress
);
6570 if (wc
->level
== 1 &&
6571 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
6575 ret
= readahead_tree_block(root
, bytenr
, blocksize
,
6581 wc
->reada_slot
= slot
;
6585 * hepler to process tree block while walking down the tree.
6587 * when wc->stage == UPDATE_BACKREF, this function updates
6588 * back refs for pointers in the block.
6590 * NOTE: return value 1 means we should stop walking down.
6592 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
6593 struct btrfs_root
*root
,
6594 struct btrfs_path
*path
,
6595 struct walk_control
*wc
, int lookup_info
)
6597 int level
= wc
->level
;
6598 struct extent_buffer
*eb
= path
->nodes
[level
];
6599 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
6602 if (wc
->stage
== UPDATE_BACKREF
&&
6603 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
6607 * when reference count of tree block is 1, it won't increase
6608 * again. once full backref flag is set, we never clear it.
6611 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
6612 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
6613 BUG_ON(!path
->locks
[level
]);
6614 ret
= btrfs_lookup_extent_info(trans
, root
,
6618 BUG_ON(ret
== -ENOMEM
);
6621 BUG_ON(wc
->refs
[level
] == 0);
6624 if (wc
->stage
== DROP_REFERENCE
) {
6625 if (wc
->refs
[level
] > 1)
6628 if (path
->locks
[level
] && !wc
->keep_locks
) {
6629 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
6630 path
->locks
[level
] = 0;
6635 /* wc->stage == UPDATE_BACKREF */
6636 if (!(wc
->flags
[level
] & flag
)) {
6637 BUG_ON(!path
->locks
[level
]);
6638 ret
= btrfs_inc_ref(trans
, root
, eb
, 1, wc
->for_reloc
);
6639 BUG_ON(ret
); /* -ENOMEM */
6640 ret
= btrfs_dec_ref(trans
, root
, eb
, 0, wc
->for_reloc
);
6641 BUG_ON(ret
); /* -ENOMEM */
6642 ret
= btrfs_set_disk_extent_flags(trans
, root
, eb
->start
,
6644 BUG_ON(ret
); /* -ENOMEM */
6645 wc
->flags
[level
] |= flag
;
6649 * the block is shared by multiple trees, so it's not good to
6650 * keep the tree lock
6652 if (path
->locks
[level
] && level
> 0) {
6653 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
6654 path
->locks
[level
] = 0;
6660 * hepler to process tree block pointer.
6662 * when wc->stage == DROP_REFERENCE, this function checks
6663 * reference count of the block pointed to. if the block
6664 * is shared and we need update back refs for the subtree
6665 * rooted at the block, this function changes wc->stage to
6666 * UPDATE_BACKREF. if the block is shared and there is no
6667 * need to update back, this function drops the reference
6670 * NOTE: return value 1 means we should stop walking down.
6672 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
6673 struct btrfs_root
*root
,
6674 struct btrfs_path
*path
,
6675 struct walk_control
*wc
, int *lookup_info
)
6681 struct btrfs_key key
;
6682 struct extent_buffer
*next
;
6683 int level
= wc
->level
;
6687 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
6688 path
->slots
[level
]);
6690 * if the lower level block was created before the snapshot
6691 * was created, we know there is no need to update back refs
6694 if (wc
->stage
== UPDATE_BACKREF
&&
6695 generation
<= root
->root_key
.offset
) {
6700 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
6701 blocksize
= btrfs_level_size(root
, level
- 1);
6703 next
= btrfs_find_tree_block(root
, bytenr
, blocksize
);
6705 next
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
6710 btrfs_tree_lock(next
);
6711 btrfs_set_lock_blocking(next
);
6713 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, blocksize
,
6714 &wc
->refs
[level
- 1],
6715 &wc
->flags
[level
- 1]);
6717 btrfs_tree_unlock(next
);
6721 BUG_ON(wc
->refs
[level
- 1] == 0);
6724 if (wc
->stage
== DROP_REFERENCE
) {
6725 if (wc
->refs
[level
- 1] > 1) {
6727 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
6730 if (!wc
->update_ref
||
6731 generation
<= root
->root_key
.offset
)
6734 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
6735 path
->slots
[level
]);
6736 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
6740 wc
->stage
= UPDATE_BACKREF
;
6741 wc
->shared_level
= level
- 1;
6745 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
6749 if (!btrfs_buffer_uptodate(next
, generation
, 0)) {
6750 btrfs_tree_unlock(next
);
6751 free_extent_buffer(next
);
6757 if (reada
&& level
== 1)
6758 reada_walk_down(trans
, root
, wc
, path
);
6759 next
= read_tree_block(root
, bytenr
, blocksize
, generation
);
6762 btrfs_tree_lock(next
);
6763 btrfs_set_lock_blocking(next
);
6767 BUG_ON(level
!= btrfs_header_level(next
));
6768 path
->nodes
[level
] = next
;
6769 path
->slots
[level
] = 0;
6770 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
6776 wc
->refs
[level
- 1] = 0;
6777 wc
->flags
[level
- 1] = 0;
6778 if (wc
->stage
== DROP_REFERENCE
) {
6779 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
6780 parent
= path
->nodes
[level
]->start
;
6782 BUG_ON(root
->root_key
.objectid
!=
6783 btrfs_header_owner(path
->nodes
[level
]));
6787 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
, parent
,
6788 root
->root_key
.objectid
, level
- 1, 0, 0);
6789 BUG_ON(ret
); /* -ENOMEM */
6791 btrfs_tree_unlock(next
);
6792 free_extent_buffer(next
);
6798 * hepler to process tree block while walking up the tree.
6800 * when wc->stage == DROP_REFERENCE, this function drops
6801 * reference count on the block.
6803 * when wc->stage == UPDATE_BACKREF, this function changes
6804 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6805 * to UPDATE_BACKREF previously while processing the block.
6807 * NOTE: return value 1 means we should stop walking up.
6809 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
6810 struct btrfs_root
*root
,
6811 struct btrfs_path
*path
,
6812 struct walk_control
*wc
)
6815 int level
= wc
->level
;
6816 struct extent_buffer
*eb
= path
->nodes
[level
];
6819 if (wc
->stage
== UPDATE_BACKREF
) {
6820 BUG_ON(wc
->shared_level
< level
);
6821 if (level
< wc
->shared_level
)
6824 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
6828 wc
->stage
= DROP_REFERENCE
;
6829 wc
->shared_level
= -1;
6830 path
->slots
[level
] = 0;
6833 * check reference count again if the block isn't locked.
6834 * we should start walking down the tree again if reference
6837 if (!path
->locks
[level
]) {
6839 btrfs_tree_lock(eb
);
6840 btrfs_set_lock_blocking(eb
);
6841 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
6843 ret
= btrfs_lookup_extent_info(trans
, root
,
6848 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
6849 path
->locks
[level
] = 0;
6852 BUG_ON(wc
->refs
[level
] == 0);
6853 if (wc
->refs
[level
] == 1) {
6854 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
6855 path
->locks
[level
] = 0;
6861 /* wc->stage == DROP_REFERENCE */
6862 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
6864 if (wc
->refs
[level
] == 1) {
6866 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
6867 ret
= btrfs_dec_ref(trans
, root
, eb
, 1,
6870 ret
= btrfs_dec_ref(trans
, root
, eb
, 0,
6872 BUG_ON(ret
); /* -ENOMEM */
6874 /* make block locked assertion in clean_tree_block happy */
6875 if (!path
->locks
[level
] &&
6876 btrfs_header_generation(eb
) == trans
->transid
) {
6877 btrfs_tree_lock(eb
);
6878 btrfs_set_lock_blocking(eb
);
6879 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
6881 clean_tree_block(trans
, root
, eb
);
6884 if (eb
== root
->node
) {
6885 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
6888 BUG_ON(root
->root_key
.objectid
!=
6889 btrfs_header_owner(eb
));
6891 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
6892 parent
= path
->nodes
[level
+ 1]->start
;
6894 BUG_ON(root
->root_key
.objectid
!=
6895 btrfs_header_owner(path
->nodes
[level
+ 1]));
6898 btrfs_free_tree_block(trans
, root
, eb
, parent
, wc
->refs
[level
] == 1);
6900 wc
->refs
[level
] = 0;
6901 wc
->flags
[level
] = 0;
6905 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
6906 struct btrfs_root
*root
,
6907 struct btrfs_path
*path
,
6908 struct walk_control
*wc
)
6910 int level
= wc
->level
;
6911 int lookup_info
= 1;
6914 while (level
>= 0) {
6915 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
6922 if (path
->slots
[level
] >=
6923 btrfs_header_nritems(path
->nodes
[level
]))
6926 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
6928 path
->slots
[level
]++;
6937 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
6938 struct btrfs_root
*root
,
6939 struct btrfs_path
*path
,
6940 struct walk_control
*wc
, int max_level
)
6942 int level
= wc
->level
;
6945 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
6946 while (level
< max_level
&& path
->nodes
[level
]) {
6948 if (path
->slots
[level
] + 1 <
6949 btrfs_header_nritems(path
->nodes
[level
])) {
6950 path
->slots
[level
]++;
6953 ret
= walk_up_proc(trans
, root
, path
, wc
);
6957 if (path
->locks
[level
]) {
6958 btrfs_tree_unlock_rw(path
->nodes
[level
],
6959 path
->locks
[level
]);
6960 path
->locks
[level
] = 0;
6962 free_extent_buffer(path
->nodes
[level
]);
6963 path
->nodes
[level
] = NULL
;
6971 * drop a subvolume tree.
6973 * this function traverses the tree freeing any blocks that only
6974 * referenced by the tree.
6976 * when a shared tree block is found. this function decreases its
6977 * reference count by one. if update_ref is true, this function
6978 * also make sure backrefs for the shared block and all lower level
6979 * blocks are properly updated.
6981 int btrfs_drop_snapshot(struct btrfs_root
*root
,
6982 struct btrfs_block_rsv
*block_rsv
, int update_ref
,
6985 struct btrfs_path
*path
;
6986 struct btrfs_trans_handle
*trans
;
6987 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
6988 struct btrfs_root_item
*root_item
= &root
->root_item
;
6989 struct walk_control
*wc
;
6990 struct btrfs_key key
;
6995 path
= btrfs_alloc_path();
7001 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
7003 btrfs_free_path(path
);
7008 trans
= btrfs_start_transaction(tree_root
, 0);
7009 if (IS_ERR(trans
)) {
7010 err
= PTR_ERR(trans
);
7015 trans
->block_rsv
= block_rsv
;
7017 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
7018 level
= btrfs_header_level(root
->node
);
7019 path
->nodes
[level
] = btrfs_lock_root_node(root
);
7020 btrfs_set_lock_blocking(path
->nodes
[level
]);
7021 path
->slots
[level
] = 0;
7022 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
7023 memset(&wc
->update_progress
, 0,
7024 sizeof(wc
->update_progress
));
7026 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
7027 memcpy(&wc
->update_progress
, &key
,
7028 sizeof(wc
->update_progress
));
7030 level
= root_item
->drop_level
;
7032 path
->lowest_level
= level
;
7033 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
7034 path
->lowest_level
= 0;
7042 * unlock our path, this is safe because only this
7043 * function is allowed to delete this snapshot
7045 btrfs_unlock_up_safe(path
, 0);
7047 level
= btrfs_header_level(root
->node
);
7049 btrfs_tree_lock(path
->nodes
[level
]);
7050 btrfs_set_lock_blocking(path
->nodes
[level
]);
7052 ret
= btrfs_lookup_extent_info(trans
, root
,
7053 path
->nodes
[level
]->start
,
7054 path
->nodes
[level
]->len
,
7061 BUG_ON(wc
->refs
[level
] == 0);
7063 if (level
== root_item
->drop_level
)
7066 btrfs_tree_unlock(path
->nodes
[level
]);
7067 WARN_ON(wc
->refs
[level
] != 1);
7073 wc
->shared_level
= -1;
7074 wc
->stage
= DROP_REFERENCE
;
7075 wc
->update_ref
= update_ref
;
7077 wc
->for_reloc
= for_reloc
;
7078 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
7081 ret
= walk_down_tree(trans
, root
, path
, wc
);
7087 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
7094 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
7098 if (wc
->stage
== DROP_REFERENCE
) {
7100 btrfs_node_key(path
->nodes
[level
],
7101 &root_item
->drop_progress
,
7102 path
->slots
[level
]);
7103 root_item
->drop_level
= level
;
7106 BUG_ON(wc
->level
== 0);
7107 if (btrfs_should_end_transaction(trans
, tree_root
)) {
7108 ret
= btrfs_update_root(trans
, tree_root
,
7112 btrfs_abort_transaction(trans
, tree_root
, ret
);
7117 btrfs_end_transaction_throttle(trans
, tree_root
);
7118 trans
= btrfs_start_transaction(tree_root
, 0);
7119 if (IS_ERR(trans
)) {
7120 err
= PTR_ERR(trans
);
7124 trans
->block_rsv
= block_rsv
;
7127 btrfs_release_path(path
);
7131 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
7133 btrfs_abort_transaction(trans
, tree_root
, ret
);
7137 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
7138 ret
= btrfs_find_last_root(tree_root
, root
->root_key
.objectid
,
7141 btrfs_abort_transaction(trans
, tree_root
, ret
);
7144 } else if (ret
> 0) {
7145 /* if we fail to delete the orphan item this time
7146 * around, it'll get picked up the next time.
7148 * The most common failure here is just -ENOENT.
7150 btrfs_del_orphan_item(trans
, tree_root
,
7151 root
->root_key
.objectid
);
7155 if (root
->in_radix
) {
7156 btrfs_free_fs_root(tree_root
->fs_info
, root
);
7158 free_extent_buffer(root
->node
);
7159 free_extent_buffer(root
->commit_root
);
7163 btrfs_end_transaction_throttle(trans
, tree_root
);
7166 btrfs_free_path(path
);
7169 btrfs_std_error(root
->fs_info
, err
);
7174 * drop subtree rooted at tree block 'node'.
7176 * NOTE: this function will unlock and release tree block 'node'
7177 * only used by relocation code
7179 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
7180 struct btrfs_root
*root
,
7181 struct extent_buffer
*node
,
7182 struct extent_buffer
*parent
)
7184 struct btrfs_path
*path
;
7185 struct walk_control
*wc
;
7191 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
7193 path
= btrfs_alloc_path();
7197 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
7199 btrfs_free_path(path
);
7203 btrfs_assert_tree_locked(parent
);
7204 parent_level
= btrfs_header_level(parent
);
7205 extent_buffer_get(parent
);
7206 path
->nodes
[parent_level
] = parent
;
7207 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
7209 btrfs_assert_tree_locked(node
);
7210 level
= btrfs_header_level(node
);
7211 path
->nodes
[level
] = node
;
7212 path
->slots
[level
] = 0;
7213 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
7215 wc
->refs
[parent_level
] = 1;
7216 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
7218 wc
->shared_level
= -1;
7219 wc
->stage
= DROP_REFERENCE
;
7223 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
7226 wret
= walk_down_tree(trans
, root
, path
, wc
);
7232 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
7240 btrfs_free_path(path
);
7244 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
7250 * if restripe for this chunk_type is on pick target profile and
7251 * return, otherwise do the usual balance
7253 stripped
= get_restripe_target(root
->fs_info
, flags
);
7255 return extended_to_chunk(stripped
);
7258 * we add in the count of missing devices because we want
7259 * to make sure that any RAID levels on a degraded FS
7260 * continue to be honored.
7262 num_devices
= root
->fs_info
->fs_devices
->rw_devices
+
7263 root
->fs_info
->fs_devices
->missing_devices
;
7265 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
7266 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
7268 if (num_devices
== 1) {
7269 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
7270 stripped
= flags
& ~stripped
;
7272 /* turn raid0 into single device chunks */
7273 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
7276 /* turn mirroring into duplication */
7277 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
7278 BTRFS_BLOCK_GROUP_RAID10
))
7279 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
7281 /* they already had raid on here, just return */
7282 if (flags
& stripped
)
7285 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
7286 stripped
= flags
& ~stripped
;
7288 /* switch duplicated blocks with raid1 */
7289 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
7290 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
7292 /* this is drive concat, leave it alone */
7298 static int set_block_group_ro(struct btrfs_block_group_cache
*cache
, int force
)
7300 struct btrfs_space_info
*sinfo
= cache
->space_info
;
7302 u64 min_allocable_bytes
;
7307 * We need some metadata space and system metadata space for
7308 * allocating chunks in some corner cases until we force to set
7309 * it to be readonly.
7312 (BTRFS_BLOCK_GROUP_SYSTEM
| BTRFS_BLOCK_GROUP_METADATA
)) &&
7314 min_allocable_bytes
= 1 * 1024 * 1024;
7316 min_allocable_bytes
= 0;
7318 spin_lock(&sinfo
->lock
);
7319 spin_lock(&cache
->lock
);
7326 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
7327 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
7329 if (sinfo
->bytes_used
+ sinfo
->bytes_reserved
+ sinfo
->bytes_pinned
+
7330 sinfo
->bytes_may_use
+ sinfo
->bytes_readonly
+ num_bytes
+
7331 min_allocable_bytes
<= sinfo
->total_bytes
) {
7332 sinfo
->bytes_readonly
+= num_bytes
;
7337 spin_unlock(&cache
->lock
);
7338 spin_unlock(&sinfo
->lock
);
7342 int btrfs_set_block_group_ro(struct btrfs_root
*root
,
7343 struct btrfs_block_group_cache
*cache
)
7346 struct btrfs_trans_handle
*trans
;
7352 trans
= btrfs_join_transaction(root
);
7354 return PTR_ERR(trans
);
7356 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
7357 if (alloc_flags
!= cache
->flags
) {
7358 ret
= do_chunk_alloc(trans
, root
, alloc_flags
,
7364 ret
= set_block_group_ro(cache
, 0);
7367 alloc_flags
= get_alloc_profile(root
, cache
->space_info
->flags
);
7368 ret
= do_chunk_alloc(trans
, root
, alloc_flags
,
7372 ret
= set_block_group_ro(cache
, 0);
7374 btrfs_end_transaction(trans
, root
);
7378 int btrfs_force_chunk_alloc(struct btrfs_trans_handle
*trans
,
7379 struct btrfs_root
*root
, u64 type
)
7381 u64 alloc_flags
= get_alloc_profile(root
, type
);
7382 return do_chunk_alloc(trans
, root
, alloc_flags
,
7387 * helper to account the unused space of all the readonly block group in the
7388 * list. takes mirrors into account.
7390 static u64
__btrfs_get_ro_block_group_free_space(struct list_head
*groups_list
)
7392 struct btrfs_block_group_cache
*block_group
;
7396 list_for_each_entry(block_group
, groups_list
, list
) {
7397 spin_lock(&block_group
->lock
);
7399 if (!block_group
->ro
) {
7400 spin_unlock(&block_group
->lock
);
7404 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_RAID1
|
7405 BTRFS_BLOCK_GROUP_RAID10
|
7406 BTRFS_BLOCK_GROUP_DUP
))
7411 free_bytes
+= (block_group
->key
.offset
-
7412 btrfs_block_group_used(&block_group
->item
)) *
7415 spin_unlock(&block_group
->lock
);
7422 * helper to account the unused space of all the readonly block group in the
7423 * space_info. takes mirrors into account.
7425 u64
btrfs_account_ro_block_groups_free_space(struct btrfs_space_info
*sinfo
)
7430 spin_lock(&sinfo
->lock
);
7432 for(i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
7433 if (!list_empty(&sinfo
->block_groups
[i
]))
7434 free_bytes
+= __btrfs_get_ro_block_group_free_space(
7435 &sinfo
->block_groups
[i
]);
7437 spin_unlock(&sinfo
->lock
);
7442 void btrfs_set_block_group_rw(struct btrfs_root
*root
,
7443 struct btrfs_block_group_cache
*cache
)
7445 struct btrfs_space_info
*sinfo
= cache
->space_info
;
7450 spin_lock(&sinfo
->lock
);
7451 spin_lock(&cache
->lock
);
7452 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
7453 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
7454 sinfo
->bytes_readonly
-= num_bytes
;
7456 spin_unlock(&cache
->lock
);
7457 spin_unlock(&sinfo
->lock
);
7461 * checks to see if its even possible to relocate this block group.
7463 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7464 * ok to go ahead and try.
7466 int btrfs_can_relocate(struct btrfs_root
*root
, u64 bytenr
)
7468 struct btrfs_block_group_cache
*block_group
;
7469 struct btrfs_space_info
*space_info
;
7470 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
7471 struct btrfs_device
*device
;
7480 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
7482 /* odd, couldn't find the block group, leave it alone */
7486 min_free
= btrfs_block_group_used(&block_group
->item
);
7488 /* no bytes used, we're good */
7492 space_info
= block_group
->space_info
;
7493 spin_lock(&space_info
->lock
);
7495 full
= space_info
->full
;
7498 * if this is the last block group we have in this space, we can't
7499 * relocate it unless we're able to allocate a new chunk below.
7501 * Otherwise, we need to make sure we have room in the space to handle
7502 * all of the extents from this block group. If we can, we're good
7504 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
7505 (space_info
->bytes_used
+ space_info
->bytes_reserved
+
7506 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
7507 min_free
< space_info
->total_bytes
)) {
7508 spin_unlock(&space_info
->lock
);
7511 spin_unlock(&space_info
->lock
);
7514 * ok we don't have enough space, but maybe we have free space on our
7515 * devices to allocate new chunks for relocation, so loop through our
7516 * alloc devices and guess if we have enough space. if this block
7517 * group is going to be restriped, run checks against the target
7518 * profile instead of the current one.
7530 target
= get_restripe_target(root
->fs_info
, block_group
->flags
);
7532 index
= __get_raid_index(extended_to_chunk(target
));
7535 * this is just a balance, so if we were marked as full
7536 * we know there is no space for a new chunk
7541 index
= get_block_group_index(block_group
);
7544 if (index
== BTRFS_RAID_RAID10
) {
7548 } else if (index
== BTRFS_RAID_RAID1
) {
7550 } else if (index
== BTRFS_RAID_DUP
) {
7553 } else if (index
== BTRFS_RAID_RAID0
) {
7554 dev_min
= fs_devices
->rw_devices
;
7555 do_div(min_free
, dev_min
);
7558 mutex_lock(&root
->fs_info
->chunk_mutex
);
7559 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
7563 * check to make sure we can actually find a chunk with enough
7564 * space to fit our block group in.
7566 if (device
->total_bytes
> device
->bytes_used
+ min_free
&&
7567 !device
->is_tgtdev_for_dev_replace
) {
7568 ret
= find_free_dev_extent(device
, min_free
,
7573 if (dev_nr
>= dev_min
)
7579 mutex_unlock(&root
->fs_info
->chunk_mutex
);
7581 btrfs_put_block_group(block_group
);
7585 static int find_first_block_group(struct btrfs_root
*root
,
7586 struct btrfs_path
*path
, struct btrfs_key
*key
)
7589 struct btrfs_key found_key
;
7590 struct extent_buffer
*leaf
;
7593 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
7598 slot
= path
->slots
[0];
7599 leaf
= path
->nodes
[0];
7600 if (slot
>= btrfs_header_nritems(leaf
)) {
7601 ret
= btrfs_next_leaf(root
, path
);
7608 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
7610 if (found_key
.objectid
>= key
->objectid
&&
7611 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
7621 void btrfs_put_block_group_cache(struct btrfs_fs_info
*info
)
7623 struct btrfs_block_group_cache
*block_group
;
7627 struct inode
*inode
;
7629 block_group
= btrfs_lookup_first_block_group(info
, last
);
7630 while (block_group
) {
7631 spin_lock(&block_group
->lock
);
7632 if (block_group
->iref
)
7634 spin_unlock(&block_group
->lock
);
7635 block_group
= next_block_group(info
->tree_root
,
7645 inode
= block_group
->inode
;
7646 block_group
->iref
= 0;
7647 block_group
->inode
= NULL
;
7648 spin_unlock(&block_group
->lock
);
7650 last
= block_group
->key
.objectid
+ block_group
->key
.offset
;
7651 btrfs_put_block_group(block_group
);
7655 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
7657 struct btrfs_block_group_cache
*block_group
;
7658 struct btrfs_space_info
*space_info
;
7659 struct btrfs_caching_control
*caching_ctl
;
7662 down_write(&info
->extent_commit_sem
);
7663 while (!list_empty(&info
->caching_block_groups
)) {
7664 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
7665 struct btrfs_caching_control
, list
);
7666 list_del(&caching_ctl
->list
);
7667 put_caching_control(caching_ctl
);
7669 up_write(&info
->extent_commit_sem
);
7671 spin_lock(&info
->block_group_cache_lock
);
7672 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
7673 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
7675 rb_erase(&block_group
->cache_node
,
7676 &info
->block_group_cache_tree
);
7677 spin_unlock(&info
->block_group_cache_lock
);
7679 down_write(&block_group
->space_info
->groups_sem
);
7680 list_del(&block_group
->list
);
7681 up_write(&block_group
->space_info
->groups_sem
);
7683 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
7684 wait_block_group_cache_done(block_group
);
7687 * We haven't cached this block group, which means we could
7688 * possibly have excluded extents on this block group.
7690 if (block_group
->cached
== BTRFS_CACHE_NO
)
7691 free_excluded_extents(info
->extent_root
, block_group
);
7693 btrfs_remove_free_space_cache(block_group
);
7694 btrfs_put_block_group(block_group
);
7696 spin_lock(&info
->block_group_cache_lock
);
7698 spin_unlock(&info
->block_group_cache_lock
);
7700 /* now that all the block groups are freed, go through and
7701 * free all the space_info structs. This is only called during
7702 * the final stages of unmount, and so we know nobody is
7703 * using them. We call synchronize_rcu() once before we start,
7704 * just to be on the safe side.
7708 release_global_block_rsv(info
);
7710 while(!list_empty(&info
->space_info
)) {
7711 space_info
= list_entry(info
->space_info
.next
,
7712 struct btrfs_space_info
,
7714 if (space_info
->bytes_pinned
> 0 ||
7715 space_info
->bytes_reserved
> 0 ||
7716 space_info
->bytes_may_use
> 0) {
7718 dump_space_info(space_info
, 0, 0);
7720 list_del(&space_info
->list
);
7726 static void __link_block_group(struct btrfs_space_info
*space_info
,
7727 struct btrfs_block_group_cache
*cache
)
7729 int index
= get_block_group_index(cache
);
7731 down_write(&space_info
->groups_sem
);
7732 list_add_tail(&cache
->list
, &space_info
->block_groups
[index
]);
7733 up_write(&space_info
->groups_sem
);
7736 int btrfs_read_block_groups(struct btrfs_root
*root
)
7738 struct btrfs_path
*path
;
7740 struct btrfs_block_group_cache
*cache
;
7741 struct btrfs_fs_info
*info
= root
->fs_info
;
7742 struct btrfs_space_info
*space_info
;
7743 struct btrfs_key key
;
7744 struct btrfs_key found_key
;
7745 struct extent_buffer
*leaf
;
7749 root
= info
->extent_root
;
7752 btrfs_set_key_type(&key
, BTRFS_BLOCK_GROUP_ITEM_KEY
);
7753 path
= btrfs_alloc_path();
7758 cache_gen
= btrfs_super_cache_generation(root
->fs_info
->super_copy
);
7759 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
7760 btrfs_super_generation(root
->fs_info
->super_copy
) != cache_gen
)
7762 if (btrfs_test_opt(root
, CLEAR_CACHE
))
7766 ret
= find_first_block_group(root
, path
, &key
);
7771 leaf
= path
->nodes
[0];
7772 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
7773 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
7778 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
7780 if (!cache
->free_space_ctl
) {
7786 atomic_set(&cache
->count
, 1);
7787 spin_lock_init(&cache
->lock
);
7788 cache
->fs_info
= info
;
7789 INIT_LIST_HEAD(&cache
->list
);
7790 INIT_LIST_HEAD(&cache
->cluster_list
);
7794 * When we mount with old space cache, we need to
7795 * set BTRFS_DC_CLEAR and set dirty flag.
7797 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
7798 * truncate the old free space cache inode and
7800 * b) Setting 'dirty flag' makes sure that we flush
7801 * the new space cache info onto disk.
7803 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
7804 if (btrfs_test_opt(root
, SPACE_CACHE
))
7808 read_extent_buffer(leaf
, &cache
->item
,
7809 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
7810 sizeof(cache
->item
));
7811 memcpy(&cache
->key
, &found_key
, sizeof(found_key
));
7813 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
7814 btrfs_release_path(path
);
7815 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
7816 cache
->sectorsize
= root
->sectorsize
;
7818 btrfs_init_free_space_ctl(cache
);
7821 * We need to exclude the super stripes now so that the space
7822 * info has super bytes accounted for, otherwise we'll think
7823 * we have more space than we actually do.
7825 exclude_super_stripes(root
, cache
);
7828 * check for two cases, either we are full, and therefore
7829 * don't need to bother with the caching work since we won't
7830 * find any space, or we are empty, and we can just add all
7831 * the space in and be done with it. This saves us _alot_ of
7832 * time, particularly in the full case.
7834 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
7835 cache
->last_byte_to_unpin
= (u64
)-1;
7836 cache
->cached
= BTRFS_CACHE_FINISHED
;
7837 free_excluded_extents(root
, cache
);
7838 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
7839 cache
->last_byte_to_unpin
= (u64
)-1;
7840 cache
->cached
= BTRFS_CACHE_FINISHED
;
7841 add_new_free_space(cache
, root
->fs_info
,
7843 found_key
.objectid
+
7845 free_excluded_extents(root
, cache
);
7848 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
7849 btrfs_block_group_used(&cache
->item
),
7851 BUG_ON(ret
); /* -ENOMEM */
7852 cache
->space_info
= space_info
;
7853 spin_lock(&cache
->space_info
->lock
);
7854 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
7855 spin_unlock(&cache
->space_info
->lock
);
7857 __link_block_group(space_info
, cache
);
7859 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
7860 BUG_ON(ret
); /* Logic error */
7862 set_avail_alloc_bits(root
->fs_info
, cache
->flags
);
7863 if (btrfs_chunk_readonly(root
, cache
->key
.objectid
))
7864 set_block_group_ro(cache
, 1);
7867 list_for_each_entry_rcu(space_info
, &root
->fs_info
->space_info
, list
) {
7868 if (!(get_alloc_profile(root
, space_info
->flags
) &
7869 (BTRFS_BLOCK_GROUP_RAID10
|
7870 BTRFS_BLOCK_GROUP_RAID1
|
7871 BTRFS_BLOCK_GROUP_DUP
)))
7874 * avoid allocating from un-mirrored block group if there are
7875 * mirrored block groups.
7877 list_for_each_entry(cache
, &space_info
->block_groups
[3], list
)
7878 set_block_group_ro(cache
, 1);
7879 list_for_each_entry(cache
, &space_info
->block_groups
[4], list
)
7880 set_block_group_ro(cache
, 1);
7883 init_global_block_rsv(info
);
7886 btrfs_free_path(path
);
7890 void btrfs_create_pending_block_groups(struct btrfs_trans_handle
*trans
,
7891 struct btrfs_root
*root
)
7893 struct btrfs_block_group_cache
*block_group
, *tmp
;
7894 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
7895 struct btrfs_block_group_item item
;
7896 struct btrfs_key key
;
7899 list_for_each_entry_safe(block_group
, tmp
, &trans
->new_bgs
,
7901 list_del_init(&block_group
->new_bg_list
);
7906 spin_lock(&block_group
->lock
);
7907 memcpy(&item
, &block_group
->item
, sizeof(item
));
7908 memcpy(&key
, &block_group
->key
, sizeof(key
));
7909 spin_unlock(&block_group
->lock
);
7911 ret
= btrfs_insert_item(trans
, extent_root
, &key
, &item
,
7914 btrfs_abort_transaction(trans
, extent_root
, ret
);
7918 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
7919 struct btrfs_root
*root
, u64 bytes_used
,
7920 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
7924 struct btrfs_root
*extent_root
;
7925 struct btrfs_block_group_cache
*cache
;
7927 extent_root
= root
->fs_info
->extent_root
;
7929 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
7931 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
7934 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
7936 if (!cache
->free_space_ctl
) {
7941 cache
->key
.objectid
= chunk_offset
;
7942 cache
->key
.offset
= size
;
7943 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
7944 cache
->sectorsize
= root
->sectorsize
;
7945 cache
->fs_info
= root
->fs_info
;
7947 atomic_set(&cache
->count
, 1);
7948 spin_lock_init(&cache
->lock
);
7949 INIT_LIST_HEAD(&cache
->list
);
7950 INIT_LIST_HEAD(&cache
->cluster_list
);
7951 INIT_LIST_HEAD(&cache
->new_bg_list
);
7953 btrfs_init_free_space_ctl(cache
);
7955 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
7956 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
7957 cache
->flags
= type
;
7958 btrfs_set_block_group_flags(&cache
->item
, type
);
7960 cache
->last_byte_to_unpin
= (u64
)-1;
7961 cache
->cached
= BTRFS_CACHE_FINISHED
;
7962 exclude_super_stripes(root
, cache
);
7964 add_new_free_space(cache
, root
->fs_info
, chunk_offset
,
7965 chunk_offset
+ size
);
7967 free_excluded_extents(root
, cache
);
7969 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
7970 &cache
->space_info
);
7971 BUG_ON(ret
); /* -ENOMEM */
7972 update_global_block_rsv(root
->fs_info
);
7974 spin_lock(&cache
->space_info
->lock
);
7975 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
7976 spin_unlock(&cache
->space_info
->lock
);
7978 __link_block_group(cache
->space_info
, cache
);
7980 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
7981 BUG_ON(ret
); /* Logic error */
7983 list_add_tail(&cache
->new_bg_list
, &trans
->new_bgs
);
7985 set_avail_alloc_bits(extent_root
->fs_info
, type
);
7990 static void clear_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
7992 u64 extra_flags
= chunk_to_extended(flags
) &
7993 BTRFS_EXTENDED_PROFILE_MASK
;
7995 write_seqlock(&fs_info
->profiles_lock
);
7996 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
7997 fs_info
->avail_data_alloc_bits
&= ~extra_flags
;
7998 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
7999 fs_info
->avail_metadata_alloc_bits
&= ~extra_flags
;
8000 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
8001 fs_info
->avail_system_alloc_bits
&= ~extra_flags
;
8002 write_sequnlock(&fs_info
->profiles_lock
);
8005 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
8006 struct btrfs_root
*root
, u64 group_start
)
8008 struct btrfs_path
*path
;
8009 struct btrfs_block_group_cache
*block_group
;
8010 struct btrfs_free_cluster
*cluster
;
8011 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
8012 struct btrfs_key key
;
8013 struct inode
*inode
;
8018 root
= root
->fs_info
->extent_root
;
8020 block_group
= btrfs_lookup_block_group(root
->fs_info
, group_start
);
8021 BUG_ON(!block_group
);
8022 BUG_ON(!block_group
->ro
);
8025 * Free the reserved super bytes from this block group before
8028 free_excluded_extents(root
, block_group
);
8030 memcpy(&key
, &block_group
->key
, sizeof(key
));
8031 index
= get_block_group_index(block_group
);
8032 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
8033 BTRFS_BLOCK_GROUP_RAID1
|
8034 BTRFS_BLOCK_GROUP_RAID10
))
8039 /* make sure this block group isn't part of an allocation cluster */
8040 cluster
= &root
->fs_info
->data_alloc_cluster
;
8041 spin_lock(&cluster
->refill_lock
);
8042 btrfs_return_cluster_to_free_space(block_group
, cluster
);
8043 spin_unlock(&cluster
->refill_lock
);
8046 * make sure this block group isn't part of a metadata
8047 * allocation cluster
8049 cluster
= &root
->fs_info
->meta_alloc_cluster
;
8050 spin_lock(&cluster
->refill_lock
);
8051 btrfs_return_cluster_to_free_space(block_group
, cluster
);
8052 spin_unlock(&cluster
->refill_lock
);
8054 path
= btrfs_alloc_path();
8060 inode
= lookup_free_space_inode(tree_root
, block_group
, path
);
8061 if (!IS_ERR(inode
)) {
8062 ret
= btrfs_orphan_add(trans
, inode
);
8064 btrfs_add_delayed_iput(inode
);
8068 /* One for the block groups ref */
8069 spin_lock(&block_group
->lock
);
8070 if (block_group
->iref
) {
8071 block_group
->iref
= 0;
8072 block_group
->inode
= NULL
;
8073 spin_unlock(&block_group
->lock
);
8076 spin_unlock(&block_group
->lock
);
8078 /* One for our lookup ref */
8079 btrfs_add_delayed_iput(inode
);
8082 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
8083 key
.offset
= block_group
->key
.objectid
;
8086 ret
= btrfs_search_slot(trans
, tree_root
, &key
, path
, -1, 1);
8090 btrfs_release_path(path
);
8092 ret
= btrfs_del_item(trans
, tree_root
, path
);
8095 btrfs_release_path(path
);
8098 spin_lock(&root
->fs_info
->block_group_cache_lock
);
8099 rb_erase(&block_group
->cache_node
,
8100 &root
->fs_info
->block_group_cache_tree
);
8102 if (root
->fs_info
->first_logical_byte
== block_group
->key
.objectid
)
8103 root
->fs_info
->first_logical_byte
= (u64
)-1;
8104 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
8106 down_write(&block_group
->space_info
->groups_sem
);
8108 * we must use list_del_init so people can check to see if they
8109 * are still on the list after taking the semaphore
8111 list_del_init(&block_group
->list
);
8112 if (list_empty(&block_group
->space_info
->block_groups
[index
]))
8113 clear_avail_alloc_bits(root
->fs_info
, block_group
->flags
);
8114 up_write(&block_group
->space_info
->groups_sem
);
8116 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
8117 wait_block_group_cache_done(block_group
);
8119 btrfs_remove_free_space_cache(block_group
);
8121 spin_lock(&block_group
->space_info
->lock
);
8122 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
8123 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
8124 block_group
->space_info
->disk_total
-= block_group
->key
.offset
* factor
;
8125 spin_unlock(&block_group
->space_info
->lock
);
8127 memcpy(&key
, &block_group
->key
, sizeof(key
));
8129 btrfs_clear_space_info_full(root
->fs_info
);
8131 btrfs_put_block_group(block_group
);
8132 btrfs_put_block_group(block_group
);
8134 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
8140 ret
= btrfs_del_item(trans
, root
, path
);
8142 btrfs_free_path(path
);
8146 int btrfs_init_space_info(struct btrfs_fs_info
*fs_info
)
8148 struct btrfs_space_info
*space_info
;
8149 struct btrfs_super_block
*disk_super
;
8155 disk_super
= fs_info
->super_copy
;
8156 if (!btrfs_super_root(disk_super
))
8159 features
= btrfs_super_incompat_flags(disk_super
);
8160 if (features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
8163 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
8164 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
8169 flags
= BTRFS_BLOCK_GROUP_METADATA
| BTRFS_BLOCK_GROUP_DATA
;
8170 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
8172 flags
= BTRFS_BLOCK_GROUP_METADATA
;
8173 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
8177 flags
= BTRFS_BLOCK_GROUP_DATA
;
8178 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
8184 int btrfs_error_unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
8186 return unpin_extent_range(root
, start
, end
);
8189 int btrfs_error_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
8190 u64 num_bytes
, u64
*actual_bytes
)
8192 return btrfs_discard_extent(root
, bytenr
, num_bytes
, actual_bytes
);
8195 int btrfs_trim_fs(struct btrfs_root
*root
, struct fstrim_range
*range
)
8197 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8198 struct btrfs_block_group_cache
*cache
= NULL
;
8203 u64 total_bytes
= btrfs_super_total_bytes(fs_info
->super_copy
);
8207 * try to trim all FS space, our block group may start from non-zero.
8209 if (range
->len
== total_bytes
)
8210 cache
= btrfs_lookup_first_block_group(fs_info
, range
->start
);
8212 cache
= btrfs_lookup_block_group(fs_info
, range
->start
);
8215 if (cache
->key
.objectid
>= (range
->start
+ range
->len
)) {
8216 btrfs_put_block_group(cache
);
8220 start
= max(range
->start
, cache
->key
.objectid
);
8221 end
= min(range
->start
+ range
->len
,
8222 cache
->key
.objectid
+ cache
->key
.offset
);
8224 if (end
- start
>= range
->minlen
) {
8225 if (!block_group_cache_done(cache
)) {
8226 ret
= cache_block_group(cache
, 0);
8228 wait_block_group_cache_done(cache
);
8230 ret
= btrfs_trim_block_group(cache
,
8236 trimmed
+= group_trimmed
;
8238 btrfs_put_block_group(cache
);
8243 cache
= next_block_group(fs_info
->tree_root
, cache
);
8246 range
->len
= trimmed
;