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/sched/signal.h>
20 #include <linux/pagemap.h>
21 #include <linux/writeback.h>
22 #include <linux/blkdev.h>
23 #include <linux/sort.h>
24 #include <linux/rcupdate.h>
25 #include <linux/kthread.h>
26 #include <linux/slab.h>
27 #include <linux/ratelimit.h>
28 #include <linux/percpu_counter.h>
32 #include "print-tree.h"
36 #include "free-space-cache.h"
37 #include "free-space-tree.h"
42 #undef SCRAMBLE_DELAYED_REFS
45 * control flags for do_chunk_alloc's force field
46 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
47 * if we really need one.
49 * CHUNK_ALLOC_LIMITED means to only try and allocate one
50 * if we have very few chunks already allocated. This is
51 * used as part of the clustering code to help make sure
52 * we have a good pool of storage to cluster in, without
53 * filling the FS with empty chunks
55 * CHUNK_ALLOC_FORCE means it must try to allocate one
59 CHUNK_ALLOC_NO_FORCE
= 0,
60 CHUNK_ALLOC_LIMITED
= 1,
61 CHUNK_ALLOC_FORCE
= 2,
64 static int update_block_group(struct btrfs_trans_handle
*trans
,
65 struct btrfs_fs_info
*fs_info
, u64 bytenr
,
66 u64 num_bytes
, int alloc
);
67 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
68 struct btrfs_fs_info
*fs_info
,
69 struct btrfs_delayed_ref_node
*node
, u64 parent
,
70 u64 root_objectid
, u64 owner_objectid
,
71 u64 owner_offset
, int refs_to_drop
,
72 struct btrfs_delayed_extent_op
*extra_op
);
73 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
74 struct extent_buffer
*leaf
,
75 struct btrfs_extent_item
*ei
);
76 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
77 struct btrfs_fs_info
*fs_info
,
78 u64 parent
, u64 root_objectid
,
79 u64 flags
, u64 owner
, u64 offset
,
80 struct btrfs_key
*ins
, int ref_mod
);
81 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
82 struct btrfs_fs_info
*fs_info
,
83 u64 parent
, u64 root_objectid
,
84 u64 flags
, struct btrfs_disk_key
*key
,
85 int level
, struct btrfs_key
*ins
);
86 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
87 struct btrfs_fs_info
*fs_info
, u64 flags
,
89 static int find_next_key(struct btrfs_path
*path
, int level
,
90 struct btrfs_key
*key
);
91 static void dump_space_info(struct btrfs_fs_info
*fs_info
,
92 struct btrfs_space_info
*info
, u64 bytes
,
93 int dump_block_groups
);
94 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache
*cache
,
95 u64 ram_bytes
, u64 num_bytes
, int delalloc
);
96 static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache
*cache
,
97 u64 num_bytes
, int delalloc
);
98 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
100 static int __reserve_metadata_bytes(struct btrfs_fs_info
*fs_info
,
101 struct btrfs_space_info
*space_info
,
103 enum btrfs_reserve_flush_enum flush
,
105 static void space_info_add_new_bytes(struct btrfs_fs_info
*fs_info
,
106 struct btrfs_space_info
*space_info
,
108 static void space_info_add_old_bytes(struct btrfs_fs_info
*fs_info
,
109 struct btrfs_space_info
*space_info
,
113 block_group_cache_done(struct btrfs_block_group_cache
*cache
)
116 return cache
->cached
== BTRFS_CACHE_FINISHED
||
117 cache
->cached
== BTRFS_CACHE_ERROR
;
120 static int block_group_bits(struct btrfs_block_group_cache
*cache
, u64 bits
)
122 return (cache
->flags
& bits
) == bits
;
125 void btrfs_get_block_group(struct btrfs_block_group_cache
*cache
)
127 atomic_inc(&cache
->count
);
130 void btrfs_put_block_group(struct btrfs_block_group_cache
*cache
)
132 if (atomic_dec_and_test(&cache
->count
)) {
133 WARN_ON(cache
->pinned
> 0);
134 WARN_ON(cache
->reserved
> 0);
137 * If not empty, someone is still holding mutex of
138 * full_stripe_lock, which can only be released by caller.
139 * And it will definitely cause use-after-free when caller
140 * tries to release full stripe lock.
142 * No better way to resolve, but only to warn.
144 WARN_ON(!RB_EMPTY_ROOT(&cache
->full_stripe_locks_root
.root
));
145 kfree(cache
->free_space_ctl
);
151 * this adds the block group to the fs_info rb tree for the block group
154 static int btrfs_add_block_group_cache(struct btrfs_fs_info
*info
,
155 struct btrfs_block_group_cache
*block_group
)
158 struct rb_node
*parent
= NULL
;
159 struct btrfs_block_group_cache
*cache
;
161 spin_lock(&info
->block_group_cache_lock
);
162 p
= &info
->block_group_cache_tree
.rb_node
;
166 cache
= rb_entry(parent
, struct btrfs_block_group_cache
,
168 if (block_group
->key
.objectid
< cache
->key
.objectid
) {
170 } else if (block_group
->key
.objectid
> cache
->key
.objectid
) {
173 spin_unlock(&info
->block_group_cache_lock
);
178 rb_link_node(&block_group
->cache_node
, parent
, p
);
179 rb_insert_color(&block_group
->cache_node
,
180 &info
->block_group_cache_tree
);
182 if (info
->first_logical_byte
> block_group
->key
.objectid
)
183 info
->first_logical_byte
= block_group
->key
.objectid
;
185 spin_unlock(&info
->block_group_cache_lock
);
191 * This will return the block group at or after bytenr if contains is 0, else
192 * it will return the block group that contains the bytenr
194 static struct btrfs_block_group_cache
*
195 block_group_cache_tree_search(struct btrfs_fs_info
*info
, u64 bytenr
,
198 struct btrfs_block_group_cache
*cache
, *ret
= NULL
;
202 spin_lock(&info
->block_group_cache_lock
);
203 n
= info
->block_group_cache_tree
.rb_node
;
206 cache
= rb_entry(n
, struct btrfs_block_group_cache
,
208 end
= cache
->key
.objectid
+ cache
->key
.offset
- 1;
209 start
= cache
->key
.objectid
;
211 if (bytenr
< start
) {
212 if (!contains
&& (!ret
|| start
< ret
->key
.objectid
))
215 } else if (bytenr
> start
) {
216 if (contains
&& bytenr
<= end
) {
227 btrfs_get_block_group(ret
);
228 if (bytenr
== 0 && info
->first_logical_byte
> ret
->key
.objectid
)
229 info
->first_logical_byte
= ret
->key
.objectid
;
231 spin_unlock(&info
->block_group_cache_lock
);
236 static int add_excluded_extent(struct btrfs_fs_info
*fs_info
,
237 u64 start
, u64 num_bytes
)
239 u64 end
= start
+ num_bytes
- 1;
240 set_extent_bits(&fs_info
->freed_extents
[0],
241 start
, end
, EXTENT_UPTODATE
);
242 set_extent_bits(&fs_info
->freed_extents
[1],
243 start
, end
, EXTENT_UPTODATE
);
247 static void free_excluded_extents(struct btrfs_fs_info
*fs_info
,
248 struct btrfs_block_group_cache
*cache
)
252 start
= cache
->key
.objectid
;
253 end
= start
+ cache
->key
.offset
- 1;
255 clear_extent_bits(&fs_info
->freed_extents
[0],
256 start
, end
, EXTENT_UPTODATE
);
257 clear_extent_bits(&fs_info
->freed_extents
[1],
258 start
, end
, EXTENT_UPTODATE
);
261 static int exclude_super_stripes(struct btrfs_fs_info
*fs_info
,
262 struct btrfs_block_group_cache
*cache
)
269 if (cache
->key
.objectid
< BTRFS_SUPER_INFO_OFFSET
) {
270 stripe_len
= BTRFS_SUPER_INFO_OFFSET
- cache
->key
.objectid
;
271 cache
->bytes_super
+= stripe_len
;
272 ret
= add_excluded_extent(fs_info
, cache
->key
.objectid
,
278 for (i
= 0; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
279 bytenr
= btrfs_sb_offset(i
);
280 ret
= btrfs_rmap_block(fs_info
, cache
->key
.objectid
,
281 bytenr
, 0, &logical
, &nr
, &stripe_len
);
288 if (logical
[nr
] > cache
->key
.objectid
+
292 if (logical
[nr
] + stripe_len
<= cache
->key
.objectid
)
296 if (start
< cache
->key
.objectid
) {
297 start
= cache
->key
.objectid
;
298 len
= (logical
[nr
] + stripe_len
) - start
;
300 len
= min_t(u64
, stripe_len
,
301 cache
->key
.objectid
+
302 cache
->key
.offset
- start
);
305 cache
->bytes_super
+= len
;
306 ret
= add_excluded_extent(fs_info
, start
, len
);
318 static struct btrfs_caching_control
*
319 get_caching_control(struct btrfs_block_group_cache
*cache
)
321 struct btrfs_caching_control
*ctl
;
323 spin_lock(&cache
->lock
);
324 if (!cache
->caching_ctl
) {
325 spin_unlock(&cache
->lock
);
329 ctl
= cache
->caching_ctl
;
330 refcount_inc(&ctl
->count
);
331 spin_unlock(&cache
->lock
);
335 static void put_caching_control(struct btrfs_caching_control
*ctl
)
337 if (refcount_dec_and_test(&ctl
->count
))
341 #ifdef CONFIG_BTRFS_DEBUG
342 static void fragment_free_space(struct btrfs_block_group_cache
*block_group
)
344 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
345 u64 start
= block_group
->key
.objectid
;
346 u64 len
= block_group
->key
.offset
;
347 u64 chunk
= block_group
->flags
& BTRFS_BLOCK_GROUP_METADATA
?
348 fs_info
->nodesize
: fs_info
->sectorsize
;
349 u64 step
= chunk
<< 1;
351 while (len
> chunk
) {
352 btrfs_remove_free_space(block_group
, start
, chunk
);
363 * this is only called by cache_block_group, since we could have freed extents
364 * we need to check the pinned_extents for any extents that can't be used yet
365 * since their free space will be released as soon as the transaction commits.
367 u64
add_new_free_space(struct btrfs_block_group_cache
*block_group
,
368 struct btrfs_fs_info
*info
, u64 start
, u64 end
)
370 u64 extent_start
, extent_end
, size
, total_added
= 0;
373 while (start
< end
) {
374 ret
= find_first_extent_bit(info
->pinned_extents
, start
,
375 &extent_start
, &extent_end
,
376 EXTENT_DIRTY
| EXTENT_UPTODATE
,
381 if (extent_start
<= start
) {
382 start
= extent_end
+ 1;
383 } else if (extent_start
> start
&& extent_start
< end
) {
384 size
= extent_start
- start
;
386 ret
= btrfs_add_free_space(block_group
, start
,
388 BUG_ON(ret
); /* -ENOMEM or logic error */
389 start
= extent_end
+ 1;
398 ret
= btrfs_add_free_space(block_group
, start
, size
);
399 BUG_ON(ret
); /* -ENOMEM or logic error */
405 static int load_extent_tree_free(struct btrfs_caching_control
*caching_ctl
)
407 struct btrfs_block_group_cache
*block_group
= caching_ctl
->block_group
;
408 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
409 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
410 struct btrfs_path
*path
;
411 struct extent_buffer
*leaf
;
412 struct btrfs_key key
;
419 path
= btrfs_alloc_path();
423 last
= max_t(u64
, block_group
->key
.objectid
, BTRFS_SUPER_INFO_OFFSET
);
425 #ifdef CONFIG_BTRFS_DEBUG
427 * If we're fragmenting we don't want to make anybody think we can
428 * allocate from this block group until we've had a chance to fragment
431 if (btrfs_should_fragment_free_space(block_group
))
435 * We don't want to deadlock with somebody trying to allocate a new
436 * extent for the extent root while also trying to search the extent
437 * root to add free space. So we skip locking and search the commit
438 * root, since its read-only
440 path
->skip_locking
= 1;
441 path
->search_commit_root
= 1;
442 path
->reada
= READA_FORWARD
;
446 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
449 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
453 leaf
= path
->nodes
[0];
454 nritems
= btrfs_header_nritems(leaf
);
457 if (btrfs_fs_closing(fs_info
) > 1) {
462 if (path
->slots
[0] < nritems
) {
463 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
465 ret
= find_next_key(path
, 0, &key
);
469 if (need_resched() ||
470 rwsem_is_contended(&fs_info
->commit_root_sem
)) {
472 caching_ctl
->progress
= last
;
473 btrfs_release_path(path
);
474 up_read(&fs_info
->commit_root_sem
);
475 mutex_unlock(&caching_ctl
->mutex
);
477 mutex_lock(&caching_ctl
->mutex
);
478 down_read(&fs_info
->commit_root_sem
);
482 ret
= btrfs_next_leaf(extent_root
, path
);
487 leaf
= path
->nodes
[0];
488 nritems
= btrfs_header_nritems(leaf
);
492 if (key
.objectid
< last
) {
495 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
498 caching_ctl
->progress
= last
;
499 btrfs_release_path(path
);
503 if (key
.objectid
< block_group
->key
.objectid
) {
508 if (key
.objectid
>= block_group
->key
.objectid
+
509 block_group
->key
.offset
)
512 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
||
513 key
.type
== BTRFS_METADATA_ITEM_KEY
) {
514 total_found
+= add_new_free_space(block_group
,
517 if (key
.type
== BTRFS_METADATA_ITEM_KEY
)
518 last
= key
.objectid
+
521 last
= key
.objectid
+ key
.offset
;
523 if (total_found
> CACHING_CTL_WAKE_UP
) {
526 wake_up(&caching_ctl
->wait
);
533 total_found
+= add_new_free_space(block_group
, fs_info
, last
,
534 block_group
->key
.objectid
+
535 block_group
->key
.offset
);
536 caching_ctl
->progress
= (u64
)-1;
539 btrfs_free_path(path
);
543 static noinline
void caching_thread(struct btrfs_work
*work
)
545 struct btrfs_block_group_cache
*block_group
;
546 struct btrfs_fs_info
*fs_info
;
547 struct btrfs_caching_control
*caching_ctl
;
548 struct btrfs_root
*extent_root
;
551 caching_ctl
= container_of(work
, struct btrfs_caching_control
, work
);
552 block_group
= caching_ctl
->block_group
;
553 fs_info
= block_group
->fs_info
;
554 extent_root
= fs_info
->extent_root
;
556 mutex_lock(&caching_ctl
->mutex
);
557 down_read(&fs_info
->commit_root_sem
);
559 if (btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
))
560 ret
= load_free_space_tree(caching_ctl
);
562 ret
= load_extent_tree_free(caching_ctl
);
564 spin_lock(&block_group
->lock
);
565 block_group
->caching_ctl
= NULL
;
566 block_group
->cached
= ret
? BTRFS_CACHE_ERROR
: BTRFS_CACHE_FINISHED
;
567 spin_unlock(&block_group
->lock
);
569 #ifdef CONFIG_BTRFS_DEBUG
570 if (btrfs_should_fragment_free_space(block_group
)) {
573 spin_lock(&block_group
->space_info
->lock
);
574 spin_lock(&block_group
->lock
);
575 bytes_used
= block_group
->key
.offset
-
576 btrfs_block_group_used(&block_group
->item
);
577 block_group
->space_info
->bytes_used
+= bytes_used
>> 1;
578 spin_unlock(&block_group
->lock
);
579 spin_unlock(&block_group
->space_info
->lock
);
580 fragment_free_space(block_group
);
584 caching_ctl
->progress
= (u64
)-1;
586 up_read(&fs_info
->commit_root_sem
);
587 free_excluded_extents(fs_info
, block_group
);
588 mutex_unlock(&caching_ctl
->mutex
);
590 wake_up(&caching_ctl
->wait
);
592 put_caching_control(caching_ctl
);
593 btrfs_put_block_group(block_group
);
596 static int cache_block_group(struct btrfs_block_group_cache
*cache
,
600 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
601 struct btrfs_caching_control
*caching_ctl
;
604 caching_ctl
= kzalloc(sizeof(*caching_ctl
), GFP_NOFS
);
608 INIT_LIST_HEAD(&caching_ctl
->list
);
609 mutex_init(&caching_ctl
->mutex
);
610 init_waitqueue_head(&caching_ctl
->wait
);
611 caching_ctl
->block_group
= cache
;
612 caching_ctl
->progress
= cache
->key
.objectid
;
613 refcount_set(&caching_ctl
->count
, 1);
614 btrfs_init_work(&caching_ctl
->work
, btrfs_cache_helper
,
615 caching_thread
, NULL
, NULL
);
617 spin_lock(&cache
->lock
);
619 * This should be a rare occasion, but this could happen I think in the
620 * case where one thread starts to load the space cache info, and then
621 * some other thread starts a transaction commit which tries to do an
622 * allocation while the other thread is still loading the space cache
623 * info. The previous loop should have kept us from choosing this block
624 * group, but if we've moved to the state where we will wait on caching
625 * block groups we need to first check if we're doing a fast load here,
626 * so we can wait for it to finish, otherwise we could end up allocating
627 * from a block group who's cache gets evicted for one reason or
630 while (cache
->cached
== BTRFS_CACHE_FAST
) {
631 struct btrfs_caching_control
*ctl
;
633 ctl
= cache
->caching_ctl
;
634 refcount_inc(&ctl
->count
);
635 prepare_to_wait(&ctl
->wait
, &wait
, TASK_UNINTERRUPTIBLE
);
636 spin_unlock(&cache
->lock
);
640 finish_wait(&ctl
->wait
, &wait
);
641 put_caching_control(ctl
);
642 spin_lock(&cache
->lock
);
645 if (cache
->cached
!= BTRFS_CACHE_NO
) {
646 spin_unlock(&cache
->lock
);
650 WARN_ON(cache
->caching_ctl
);
651 cache
->caching_ctl
= caching_ctl
;
652 cache
->cached
= BTRFS_CACHE_FAST
;
653 spin_unlock(&cache
->lock
);
655 if (fs_info
->mount_opt
& BTRFS_MOUNT_SPACE_CACHE
) {
656 mutex_lock(&caching_ctl
->mutex
);
657 ret
= load_free_space_cache(fs_info
, cache
);
659 spin_lock(&cache
->lock
);
661 cache
->caching_ctl
= NULL
;
662 cache
->cached
= BTRFS_CACHE_FINISHED
;
663 cache
->last_byte_to_unpin
= (u64
)-1;
664 caching_ctl
->progress
= (u64
)-1;
666 if (load_cache_only
) {
667 cache
->caching_ctl
= NULL
;
668 cache
->cached
= BTRFS_CACHE_NO
;
670 cache
->cached
= BTRFS_CACHE_STARTED
;
671 cache
->has_caching_ctl
= 1;
674 spin_unlock(&cache
->lock
);
675 #ifdef CONFIG_BTRFS_DEBUG
677 btrfs_should_fragment_free_space(cache
)) {
680 spin_lock(&cache
->space_info
->lock
);
681 spin_lock(&cache
->lock
);
682 bytes_used
= cache
->key
.offset
-
683 btrfs_block_group_used(&cache
->item
);
684 cache
->space_info
->bytes_used
+= bytes_used
>> 1;
685 spin_unlock(&cache
->lock
);
686 spin_unlock(&cache
->space_info
->lock
);
687 fragment_free_space(cache
);
690 mutex_unlock(&caching_ctl
->mutex
);
692 wake_up(&caching_ctl
->wait
);
694 put_caching_control(caching_ctl
);
695 free_excluded_extents(fs_info
, cache
);
700 * We're either using the free space tree or no caching at all.
701 * Set cached to the appropriate value and wakeup any waiters.
703 spin_lock(&cache
->lock
);
704 if (load_cache_only
) {
705 cache
->caching_ctl
= NULL
;
706 cache
->cached
= BTRFS_CACHE_NO
;
708 cache
->cached
= BTRFS_CACHE_STARTED
;
709 cache
->has_caching_ctl
= 1;
711 spin_unlock(&cache
->lock
);
712 wake_up(&caching_ctl
->wait
);
715 if (load_cache_only
) {
716 put_caching_control(caching_ctl
);
720 down_write(&fs_info
->commit_root_sem
);
721 refcount_inc(&caching_ctl
->count
);
722 list_add_tail(&caching_ctl
->list
, &fs_info
->caching_block_groups
);
723 up_write(&fs_info
->commit_root_sem
);
725 btrfs_get_block_group(cache
);
727 btrfs_queue_work(fs_info
->caching_workers
, &caching_ctl
->work
);
733 * return the block group that starts at or after bytenr
735 static struct btrfs_block_group_cache
*
736 btrfs_lookup_first_block_group(struct btrfs_fs_info
*info
, u64 bytenr
)
738 return block_group_cache_tree_search(info
, bytenr
, 0);
742 * return the block group that contains the given bytenr
744 struct btrfs_block_group_cache
*btrfs_lookup_block_group(
745 struct btrfs_fs_info
*info
,
748 return block_group_cache_tree_search(info
, bytenr
, 1);
751 static struct btrfs_space_info
*__find_space_info(struct btrfs_fs_info
*info
,
754 struct list_head
*head
= &info
->space_info
;
755 struct btrfs_space_info
*found
;
757 flags
&= BTRFS_BLOCK_GROUP_TYPE_MASK
;
760 list_for_each_entry_rcu(found
, head
, list
) {
761 if (found
->flags
& flags
) {
770 static void add_pinned_bytes(struct btrfs_fs_info
*fs_info
, s64 num_bytes
,
771 u64 owner
, u64 root_objectid
)
773 struct btrfs_space_info
*space_info
;
776 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
777 if (root_objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
778 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
780 flags
= BTRFS_BLOCK_GROUP_METADATA
;
782 flags
= BTRFS_BLOCK_GROUP_DATA
;
785 space_info
= __find_space_info(fs_info
, flags
);
786 BUG_ON(!space_info
); /* Logic bug */
787 percpu_counter_add(&space_info
->total_bytes_pinned
, num_bytes
);
791 * after adding space to the filesystem, we need to clear the full flags
792 * on all the space infos.
794 void btrfs_clear_space_info_full(struct btrfs_fs_info
*info
)
796 struct list_head
*head
= &info
->space_info
;
797 struct btrfs_space_info
*found
;
800 list_for_each_entry_rcu(found
, head
, list
)
805 /* simple helper to search for an existing data extent at a given offset */
806 int btrfs_lookup_data_extent(struct btrfs_fs_info
*fs_info
, u64 start
, u64 len
)
809 struct btrfs_key key
;
810 struct btrfs_path
*path
;
812 path
= btrfs_alloc_path();
816 key
.objectid
= start
;
818 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
819 ret
= btrfs_search_slot(NULL
, fs_info
->extent_root
, &key
, path
, 0, 0);
820 btrfs_free_path(path
);
825 * helper function to lookup reference count and flags of a tree block.
827 * the head node for delayed ref is used to store the sum of all the
828 * reference count modifications queued up in the rbtree. the head
829 * node may also store the extent flags to set. This way you can check
830 * to see what the reference count and extent flags would be if all of
831 * the delayed refs are not processed.
833 int btrfs_lookup_extent_info(struct btrfs_trans_handle
*trans
,
834 struct btrfs_fs_info
*fs_info
, u64 bytenr
,
835 u64 offset
, int metadata
, u64
*refs
, u64
*flags
)
837 struct btrfs_delayed_ref_head
*head
;
838 struct btrfs_delayed_ref_root
*delayed_refs
;
839 struct btrfs_path
*path
;
840 struct btrfs_extent_item
*ei
;
841 struct extent_buffer
*leaf
;
842 struct btrfs_key key
;
849 * If we don't have skinny metadata, don't bother doing anything
852 if (metadata
&& !btrfs_fs_incompat(fs_info
, SKINNY_METADATA
)) {
853 offset
= fs_info
->nodesize
;
857 path
= btrfs_alloc_path();
862 path
->skip_locking
= 1;
863 path
->search_commit_root
= 1;
867 key
.objectid
= bytenr
;
870 key
.type
= BTRFS_METADATA_ITEM_KEY
;
872 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
874 ret
= btrfs_search_slot(trans
, fs_info
->extent_root
, &key
, path
, 0, 0);
878 if (ret
> 0 && metadata
&& key
.type
== BTRFS_METADATA_ITEM_KEY
) {
879 if (path
->slots
[0]) {
881 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
883 if (key
.objectid
== bytenr
&&
884 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
885 key
.offset
== fs_info
->nodesize
)
891 leaf
= path
->nodes
[0];
892 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
893 if (item_size
>= sizeof(*ei
)) {
894 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
895 struct btrfs_extent_item
);
896 num_refs
= btrfs_extent_refs(leaf
, ei
);
897 extent_flags
= btrfs_extent_flags(leaf
, ei
);
899 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
900 struct btrfs_extent_item_v0
*ei0
;
901 BUG_ON(item_size
!= sizeof(*ei0
));
902 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
903 struct btrfs_extent_item_v0
);
904 num_refs
= btrfs_extent_refs_v0(leaf
, ei0
);
905 /* FIXME: this isn't correct for data */
906 extent_flags
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
911 BUG_ON(num_refs
== 0);
921 delayed_refs
= &trans
->transaction
->delayed_refs
;
922 spin_lock(&delayed_refs
->lock
);
923 head
= btrfs_find_delayed_ref_head(delayed_refs
, bytenr
);
925 if (!mutex_trylock(&head
->mutex
)) {
926 refcount_inc(&head
->node
.refs
);
927 spin_unlock(&delayed_refs
->lock
);
929 btrfs_release_path(path
);
932 * Mutex was contended, block until it's released and try
935 mutex_lock(&head
->mutex
);
936 mutex_unlock(&head
->mutex
);
937 btrfs_put_delayed_ref(&head
->node
);
940 spin_lock(&head
->lock
);
941 if (head
->extent_op
&& head
->extent_op
->update_flags
)
942 extent_flags
|= head
->extent_op
->flags_to_set
;
944 BUG_ON(num_refs
== 0);
946 num_refs
+= head
->node
.ref_mod
;
947 spin_unlock(&head
->lock
);
948 mutex_unlock(&head
->mutex
);
950 spin_unlock(&delayed_refs
->lock
);
952 WARN_ON(num_refs
== 0);
956 *flags
= extent_flags
;
958 btrfs_free_path(path
);
963 * Back reference rules. Back refs have three main goals:
965 * 1) differentiate between all holders of references to an extent so that
966 * when a reference is dropped we can make sure it was a valid reference
967 * before freeing the extent.
969 * 2) Provide enough information to quickly find the holders of an extent
970 * if we notice a given block is corrupted or bad.
972 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
973 * maintenance. This is actually the same as #2, but with a slightly
974 * different use case.
976 * There are two kinds of back refs. The implicit back refs is optimized
977 * for pointers in non-shared tree blocks. For a given pointer in a block,
978 * back refs of this kind provide information about the block's owner tree
979 * and the pointer's key. These information allow us to find the block by
980 * b-tree searching. The full back refs is for pointers in tree blocks not
981 * referenced by their owner trees. The location of tree block is recorded
982 * in the back refs. Actually the full back refs is generic, and can be
983 * used in all cases the implicit back refs is used. The major shortcoming
984 * of the full back refs is its overhead. Every time a tree block gets
985 * COWed, we have to update back refs entry for all pointers in it.
987 * For a newly allocated tree block, we use implicit back refs for
988 * pointers in it. This means most tree related operations only involve
989 * implicit back refs. For a tree block created in old transaction, the
990 * only way to drop a reference to it is COW it. So we can detect the
991 * event that tree block loses its owner tree's reference and do the
992 * back refs conversion.
994 * When a tree block is COWed through a tree, there are four cases:
996 * The reference count of the block is one and the tree is the block's
997 * owner tree. Nothing to do in this case.
999 * The reference count of the block is one and the tree is not the
1000 * block's owner tree. In this case, full back refs is used for pointers
1001 * in the block. Remove these full back refs, add implicit back refs for
1002 * every pointers in the new block.
1004 * The reference count of the block is greater than one and the tree is
1005 * the block's owner tree. In this case, implicit back refs is used for
1006 * pointers in the block. Add full back refs for every pointers in the
1007 * block, increase lower level extents' reference counts. The original
1008 * implicit back refs are entailed to the new block.
1010 * The reference count of the block is greater than one and the tree is
1011 * not the block's owner tree. Add implicit back refs for every pointer in
1012 * the new block, increase lower level extents' reference count.
1014 * Back Reference Key composing:
1016 * The key objectid corresponds to the first byte in the extent,
1017 * The key type is used to differentiate between types of back refs.
1018 * There are different meanings of the key offset for different types
1021 * File extents can be referenced by:
1023 * - multiple snapshots, subvolumes, or different generations in one subvol
1024 * - different files inside a single subvolume
1025 * - different offsets inside a file (bookend extents in file.c)
1027 * The extent ref structure for the implicit back refs has fields for:
1029 * - Objectid of the subvolume root
1030 * - objectid of the file holding the reference
1031 * - original offset in the file
1032 * - how many bookend extents
1034 * The key offset for the implicit back refs is hash of the first
1037 * The extent ref structure for the full back refs has field for:
1039 * - number of pointers in the tree leaf
1041 * The key offset for the implicit back refs is the first byte of
1044 * When a file extent is allocated, The implicit back refs is used.
1045 * the fields are filled in:
1047 * (root_key.objectid, inode objectid, offset in file, 1)
1049 * When a file extent is removed file truncation, we find the
1050 * corresponding implicit back refs and check the following fields:
1052 * (btrfs_header_owner(leaf), inode objectid, offset in file)
1054 * Btree extents can be referenced by:
1056 * - Different subvolumes
1058 * Both the implicit back refs and the full back refs for tree blocks
1059 * only consist of key. The key offset for the implicit back refs is
1060 * objectid of block's owner tree. The key offset for the full back refs
1061 * is the first byte of parent block.
1063 * When implicit back refs is used, information about the lowest key and
1064 * level of the tree block are required. These information are stored in
1065 * tree block info structure.
1068 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1069 static int convert_extent_item_v0(struct btrfs_trans_handle
*trans
,
1070 struct btrfs_fs_info
*fs_info
,
1071 struct btrfs_path
*path
,
1072 u64 owner
, u32 extra_size
)
1074 struct btrfs_root
*root
= fs_info
->extent_root
;
1075 struct btrfs_extent_item
*item
;
1076 struct btrfs_extent_item_v0
*ei0
;
1077 struct btrfs_extent_ref_v0
*ref0
;
1078 struct btrfs_tree_block_info
*bi
;
1079 struct extent_buffer
*leaf
;
1080 struct btrfs_key key
;
1081 struct btrfs_key found_key
;
1082 u32 new_size
= sizeof(*item
);
1086 leaf
= path
->nodes
[0];
1087 BUG_ON(btrfs_item_size_nr(leaf
, path
->slots
[0]) != sizeof(*ei0
));
1089 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1090 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1091 struct btrfs_extent_item_v0
);
1092 refs
= btrfs_extent_refs_v0(leaf
, ei0
);
1094 if (owner
== (u64
)-1) {
1096 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1097 ret
= btrfs_next_leaf(root
, path
);
1100 BUG_ON(ret
> 0); /* Corruption */
1101 leaf
= path
->nodes
[0];
1103 btrfs_item_key_to_cpu(leaf
, &found_key
,
1105 BUG_ON(key
.objectid
!= found_key
.objectid
);
1106 if (found_key
.type
!= BTRFS_EXTENT_REF_V0_KEY
) {
1110 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1111 struct btrfs_extent_ref_v0
);
1112 owner
= btrfs_ref_objectid_v0(leaf
, ref0
);
1116 btrfs_release_path(path
);
1118 if (owner
< BTRFS_FIRST_FREE_OBJECTID
)
1119 new_size
+= sizeof(*bi
);
1121 new_size
-= sizeof(*ei0
);
1122 ret
= btrfs_search_slot(trans
, root
, &key
, path
,
1123 new_size
+ extra_size
, 1);
1126 BUG_ON(ret
); /* Corruption */
1128 btrfs_extend_item(fs_info
, path
, new_size
);
1130 leaf
= path
->nodes
[0];
1131 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1132 btrfs_set_extent_refs(leaf
, item
, refs
);
1133 /* FIXME: get real generation */
1134 btrfs_set_extent_generation(leaf
, item
, 0);
1135 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1136 btrfs_set_extent_flags(leaf
, item
,
1137 BTRFS_EXTENT_FLAG_TREE_BLOCK
|
1138 BTRFS_BLOCK_FLAG_FULL_BACKREF
);
1139 bi
= (struct btrfs_tree_block_info
*)(item
+ 1);
1140 /* FIXME: get first key of the block */
1141 memzero_extent_buffer(leaf
, (unsigned long)bi
, sizeof(*bi
));
1142 btrfs_set_tree_block_level(leaf
, bi
, (int)owner
);
1144 btrfs_set_extent_flags(leaf
, item
, BTRFS_EXTENT_FLAG_DATA
);
1146 btrfs_mark_buffer_dirty(leaf
);
1151 static u64
hash_extent_data_ref(u64 root_objectid
, u64 owner
, u64 offset
)
1153 u32 high_crc
= ~(u32
)0;
1154 u32 low_crc
= ~(u32
)0;
1157 lenum
= cpu_to_le64(root_objectid
);
1158 high_crc
= btrfs_crc32c(high_crc
, &lenum
, sizeof(lenum
));
1159 lenum
= cpu_to_le64(owner
);
1160 low_crc
= btrfs_crc32c(low_crc
, &lenum
, sizeof(lenum
));
1161 lenum
= cpu_to_le64(offset
);
1162 low_crc
= btrfs_crc32c(low_crc
, &lenum
, sizeof(lenum
));
1164 return ((u64
)high_crc
<< 31) ^ (u64
)low_crc
;
1167 static u64
hash_extent_data_ref_item(struct extent_buffer
*leaf
,
1168 struct btrfs_extent_data_ref
*ref
)
1170 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf
, ref
),
1171 btrfs_extent_data_ref_objectid(leaf
, ref
),
1172 btrfs_extent_data_ref_offset(leaf
, ref
));
1175 static int match_extent_data_ref(struct extent_buffer
*leaf
,
1176 struct btrfs_extent_data_ref
*ref
,
1177 u64 root_objectid
, u64 owner
, u64 offset
)
1179 if (btrfs_extent_data_ref_root(leaf
, ref
) != root_objectid
||
1180 btrfs_extent_data_ref_objectid(leaf
, ref
) != owner
||
1181 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
1186 static noinline
int lookup_extent_data_ref(struct btrfs_trans_handle
*trans
,
1187 struct btrfs_fs_info
*fs_info
,
1188 struct btrfs_path
*path
,
1189 u64 bytenr
, u64 parent
,
1191 u64 owner
, u64 offset
)
1193 struct btrfs_root
*root
= fs_info
->extent_root
;
1194 struct btrfs_key key
;
1195 struct btrfs_extent_data_ref
*ref
;
1196 struct extent_buffer
*leaf
;
1202 key
.objectid
= bytenr
;
1204 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1205 key
.offset
= parent
;
1207 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1208 key
.offset
= hash_extent_data_ref(root_objectid
,
1213 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1222 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1223 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1224 btrfs_release_path(path
);
1225 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1236 leaf
= path
->nodes
[0];
1237 nritems
= btrfs_header_nritems(leaf
);
1239 if (path
->slots
[0] >= nritems
) {
1240 ret
= btrfs_next_leaf(root
, path
);
1246 leaf
= path
->nodes
[0];
1247 nritems
= btrfs_header_nritems(leaf
);
1251 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1252 if (key
.objectid
!= bytenr
||
1253 key
.type
!= BTRFS_EXTENT_DATA_REF_KEY
)
1256 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1257 struct btrfs_extent_data_ref
);
1259 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1262 btrfs_release_path(path
);
1274 static noinline
int insert_extent_data_ref(struct btrfs_trans_handle
*trans
,
1275 struct btrfs_fs_info
*fs_info
,
1276 struct btrfs_path
*path
,
1277 u64 bytenr
, u64 parent
,
1278 u64 root_objectid
, u64 owner
,
1279 u64 offset
, int refs_to_add
)
1281 struct btrfs_root
*root
= fs_info
->extent_root
;
1282 struct btrfs_key key
;
1283 struct extent_buffer
*leaf
;
1288 key
.objectid
= bytenr
;
1290 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1291 key
.offset
= parent
;
1292 size
= sizeof(struct btrfs_shared_data_ref
);
1294 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1295 key
.offset
= hash_extent_data_ref(root_objectid
,
1297 size
= sizeof(struct btrfs_extent_data_ref
);
1300 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, size
);
1301 if (ret
&& ret
!= -EEXIST
)
1304 leaf
= path
->nodes
[0];
1306 struct btrfs_shared_data_ref
*ref
;
1307 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1308 struct btrfs_shared_data_ref
);
1310 btrfs_set_shared_data_ref_count(leaf
, ref
, refs_to_add
);
1312 num_refs
= btrfs_shared_data_ref_count(leaf
, ref
);
1313 num_refs
+= refs_to_add
;
1314 btrfs_set_shared_data_ref_count(leaf
, ref
, num_refs
);
1317 struct btrfs_extent_data_ref
*ref
;
1318 while (ret
== -EEXIST
) {
1319 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1320 struct btrfs_extent_data_ref
);
1321 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1324 btrfs_release_path(path
);
1326 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1328 if (ret
&& ret
!= -EEXIST
)
1331 leaf
= path
->nodes
[0];
1333 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1334 struct btrfs_extent_data_ref
);
1336 btrfs_set_extent_data_ref_root(leaf
, ref
,
1338 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
1339 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
1340 btrfs_set_extent_data_ref_count(leaf
, ref
, refs_to_add
);
1342 num_refs
= btrfs_extent_data_ref_count(leaf
, ref
);
1343 num_refs
+= refs_to_add
;
1344 btrfs_set_extent_data_ref_count(leaf
, ref
, num_refs
);
1347 btrfs_mark_buffer_dirty(leaf
);
1350 btrfs_release_path(path
);
1354 static noinline
int remove_extent_data_ref(struct btrfs_trans_handle
*trans
,
1355 struct btrfs_fs_info
*fs_info
,
1356 struct btrfs_path
*path
,
1357 int refs_to_drop
, int *last_ref
)
1359 struct btrfs_key key
;
1360 struct btrfs_extent_data_ref
*ref1
= NULL
;
1361 struct btrfs_shared_data_ref
*ref2
= NULL
;
1362 struct extent_buffer
*leaf
;
1366 leaf
= path
->nodes
[0];
1367 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1369 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1370 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1371 struct btrfs_extent_data_ref
);
1372 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1373 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1374 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1375 struct btrfs_shared_data_ref
);
1376 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1377 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1378 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1379 struct btrfs_extent_ref_v0
*ref0
;
1380 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1381 struct btrfs_extent_ref_v0
);
1382 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1388 BUG_ON(num_refs
< refs_to_drop
);
1389 num_refs
-= refs_to_drop
;
1391 if (num_refs
== 0) {
1392 ret
= btrfs_del_item(trans
, fs_info
->extent_root
, path
);
1395 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
)
1396 btrfs_set_extent_data_ref_count(leaf
, ref1
, num_refs
);
1397 else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
)
1398 btrfs_set_shared_data_ref_count(leaf
, ref2
, num_refs
);
1399 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1401 struct btrfs_extent_ref_v0
*ref0
;
1402 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1403 struct btrfs_extent_ref_v0
);
1404 btrfs_set_ref_count_v0(leaf
, ref0
, num_refs
);
1407 btrfs_mark_buffer_dirty(leaf
);
1412 static noinline u32
extent_data_ref_count(struct btrfs_path
*path
,
1413 struct btrfs_extent_inline_ref
*iref
)
1415 struct btrfs_key key
;
1416 struct extent_buffer
*leaf
;
1417 struct btrfs_extent_data_ref
*ref1
;
1418 struct btrfs_shared_data_ref
*ref2
;
1421 leaf
= path
->nodes
[0];
1422 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1424 if (btrfs_extent_inline_ref_type(leaf
, iref
) ==
1425 BTRFS_EXTENT_DATA_REF_KEY
) {
1426 ref1
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1427 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1429 ref2
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1430 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1432 } else if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1433 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1434 struct btrfs_extent_data_ref
);
1435 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1436 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1437 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1438 struct btrfs_shared_data_ref
);
1439 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1440 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1441 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1442 struct btrfs_extent_ref_v0
*ref0
;
1443 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1444 struct btrfs_extent_ref_v0
);
1445 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1453 static noinline
int lookup_tree_block_ref(struct btrfs_trans_handle
*trans
,
1454 struct btrfs_fs_info
*fs_info
,
1455 struct btrfs_path
*path
,
1456 u64 bytenr
, u64 parent
,
1459 struct btrfs_root
*root
= fs_info
->extent_root
;
1460 struct btrfs_key key
;
1463 key
.objectid
= bytenr
;
1465 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1466 key
.offset
= parent
;
1468 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1469 key
.offset
= root_objectid
;
1472 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1475 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1476 if (ret
== -ENOENT
&& parent
) {
1477 btrfs_release_path(path
);
1478 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1479 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1487 static noinline
int insert_tree_block_ref(struct btrfs_trans_handle
*trans
,
1488 struct btrfs_fs_info
*fs_info
,
1489 struct btrfs_path
*path
,
1490 u64 bytenr
, u64 parent
,
1493 struct btrfs_key key
;
1496 key
.objectid
= bytenr
;
1498 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1499 key
.offset
= parent
;
1501 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1502 key
.offset
= root_objectid
;
1505 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
,
1507 btrfs_release_path(path
);
1511 static inline int extent_ref_type(u64 parent
, u64 owner
)
1514 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1516 type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1518 type
= BTRFS_TREE_BLOCK_REF_KEY
;
1521 type
= BTRFS_SHARED_DATA_REF_KEY
;
1523 type
= BTRFS_EXTENT_DATA_REF_KEY
;
1528 static int find_next_key(struct btrfs_path
*path
, int level
,
1529 struct btrfs_key
*key
)
1532 for (; level
< BTRFS_MAX_LEVEL
; level
++) {
1533 if (!path
->nodes
[level
])
1535 if (path
->slots
[level
] + 1 >=
1536 btrfs_header_nritems(path
->nodes
[level
]))
1539 btrfs_item_key_to_cpu(path
->nodes
[level
], key
,
1540 path
->slots
[level
] + 1);
1542 btrfs_node_key_to_cpu(path
->nodes
[level
], key
,
1543 path
->slots
[level
] + 1);
1550 * look for inline back ref. if back ref is found, *ref_ret is set
1551 * to the address of inline back ref, and 0 is returned.
1553 * if back ref isn't found, *ref_ret is set to the address where it
1554 * should be inserted, and -ENOENT is returned.
1556 * if insert is true and there are too many inline back refs, the path
1557 * points to the extent item, and -EAGAIN is returned.
1559 * NOTE: inline back refs are ordered in the same way that back ref
1560 * items in the tree are ordered.
1562 static noinline_for_stack
1563 int lookup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1564 struct btrfs_fs_info
*fs_info
,
1565 struct btrfs_path
*path
,
1566 struct btrfs_extent_inline_ref
**ref_ret
,
1567 u64 bytenr
, u64 num_bytes
,
1568 u64 parent
, u64 root_objectid
,
1569 u64 owner
, u64 offset
, int insert
)
1571 struct btrfs_root
*root
= fs_info
->extent_root
;
1572 struct btrfs_key key
;
1573 struct extent_buffer
*leaf
;
1574 struct btrfs_extent_item
*ei
;
1575 struct btrfs_extent_inline_ref
*iref
;
1585 bool skinny_metadata
= btrfs_fs_incompat(fs_info
, SKINNY_METADATA
);
1587 key
.objectid
= bytenr
;
1588 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1589 key
.offset
= num_bytes
;
1591 want
= extent_ref_type(parent
, owner
);
1593 extra_size
= btrfs_extent_inline_ref_size(want
);
1594 path
->keep_locks
= 1;
1599 * Owner is our parent level, so we can just add one to get the level
1600 * for the block we are interested in.
1602 if (skinny_metadata
&& owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1603 key
.type
= BTRFS_METADATA_ITEM_KEY
;
1608 ret
= btrfs_search_slot(trans
, root
, &key
, path
, extra_size
, 1);
1615 * We may be a newly converted file system which still has the old fat
1616 * extent entries for metadata, so try and see if we have one of those.
1618 if (ret
> 0 && skinny_metadata
) {
1619 skinny_metadata
= false;
1620 if (path
->slots
[0]) {
1622 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
1624 if (key
.objectid
== bytenr
&&
1625 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
1626 key
.offset
== num_bytes
)
1630 key
.objectid
= bytenr
;
1631 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1632 key
.offset
= num_bytes
;
1633 btrfs_release_path(path
);
1638 if (ret
&& !insert
) {
1641 } else if (WARN_ON(ret
)) {
1646 leaf
= path
->nodes
[0];
1647 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1648 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1649 if (item_size
< sizeof(*ei
)) {
1654 ret
= convert_extent_item_v0(trans
, fs_info
, path
, owner
,
1660 leaf
= path
->nodes
[0];
1661 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1664 BUG_ON(item_size
< sizeof(*ei
));
1666 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1667 flags
= btrfs_extent_flags(leaf
, ei
);
1669 ptr
= (unsigned long)(ei
+ 1);
1670 end
= (unsigned long)ei
+ item_size
;
1672 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
&& !skinny_metadata
) {
1673 ptr
+= sizeof(struct btrfs_tree_block_info
);
1683 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1684 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1688 ptr
+= btrfs_extent_inline_ref_size(type
);
1692 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1693 struct btrfs_extent_data_ref
*dref
;
1694 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1695 if (match_extent_data_ref(leaf
, dref
, root_objectid
,
1700 if (hash_extent_data_ref_item(leaf
, dref
) <
1701 hash_extent_data_ref(root_objectid
, owner
, offset
))
1705 ref_offset
= btrfs_extent_inline_ref_offset(leaf
, iref
);
1707 if (parent
== ref_offset
) {
1711 if (ref_offset
< parent
)
1714 if (root_objectid
== ref_offset
) {
1718 if (ref_offset
< root_objectid
)
1722 ptr
+= btrfs_extent_inline_ref_size(type
);
1724 if (err
== -ENOENT
&& insert
) {
1725 if (item_size
+ extra_size
>=
1726 BTRFS_MAX_EXTENT_ITEM_SIZE(root
)) {
1731 * To add new inline back ref, we have to make sure
1732 * there is no corresponding back ref item.
1733 * For simplicity, we just do not add new inline back
1734 * ref if there is any kind of item for this block
1736 if (find_next_key(path
, 0, &key
) == 0 &&
1737 key
.objectid
== bytenr
&&
1738 key
.type
< BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1743 *ref_ret
= (struct btrfs_extent_inline_ref
*)ptr
;
1746 path
->keep_locks
= 0;
1747 btrfs_unlock_up_safe(path
, 1);
1753 * helper to add new inline back ref
1755 static noinline_for_stack
1756 void setup_inline_extent_backref(struct btrfs_fs_info
*fs_info
,
1757 struct btrfs_path
*path
,
1758 struct btrfs_extent_inline_ref
*iref
,
1759 u64 parent
, u64 root_objectid
,
1760 u64 owner
, u64 offset
, int refs_to_add
,
1761 struct btrfs_delayed_extent_op
*extent_op
)
1763 struct extent_buffer
*leaf
;
1764 struct btrfs_extent_item
*ei
;
1767 unsigned long item_offset
;
1772 leaf
= path
->nodes
[0];
1773 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1774 item_offset
= (unsigned long)iref
- (unsigned long)ei
;
1776 type
= extent_ref_type(parent
, owner
);
1777 size
= btrfs_extent_inline_ref_size(type
);
1779 btrfs_extend_item(fs_info
, path
, size
);
1781 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1782 refs
= btrfs_extent_refs(leaf
, ei
);
1783 refs
+= refs_to_add
;
1784 btrfs_set_extent_refs(leaf
, ei
, refs
);
1786 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1788 ptr
= (unsigned long)ei
+ item_offset
;
1789 end
= (unsigned long)ei
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
1790 if (ptr
< end
- size
)
1791 memmove_extent_buffer(leaf
, ptr
+ size
, ptr
,
1794 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1795 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
1796 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1797 struct btrfs_extent_data_ref
*dref
;
1798 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1799 btrfs_set_extent_data_ref_root(leaf
, dref
, root_objectid
);
1800 btrfs_set_extent_data_ref_objectid(leaf
, dref
, owner
);
1801 btrfs_set_extent_data_ref_offset(leaf
, dref
, offset
);
1802 btrfs_set_extent_data_ref_count(leaf
, dref
, refs_to_add
);
1803 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1804 struct btrfs_shared_data_ref
*sref
;
1805 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1806 btrfs_set_shared_data_ref_count(leaf
, sref
, refs_to_add
);
1807 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1808 } else if (type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
1809 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1811 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
1813 btrfs_mark_buffer_dirty(leaf
);
1816 static int lookup_extent_backref(struct btrfs_trans_handle
*trans
,
1817 struct btrfs_fs_info
*fs_info
,
1818 struct btrfs_path
*path
,
1819 struct btrfs_extent_inline_ref
**ref_ret
,
1820 u64 bytenr
, u64 num_bytes
, u64 parent
,
1821 u64 root_objectid
, u64 owner
, u64 offset
)
1825 ret
= lookup_inline_extent_backref(trans
, fs_info
, path
, ref_ret
,
1826 bytenr
, num_bytes
, parent
,
1827 root_objectid
, owner
, offset
, 0);
1831 btrfs_release_path(path
);
1834 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1835 ret
= lookup_tree_block_ref(trans
, fs_info
, path
, bytenr
,
1836 parent
, root_objectid
);
1838 ret
= lookup_extent_data_ref(trans
, fs_info
, path
, bytenr
,
1839 parent
, root_objectid
, owner
,
1846 * helper to update/remove inline back ref
1848 static noinline_for_stack
1849 void update_inline_extent_backref(struct btrfs_fs_info
*fs_info
,
1850 struct btrfs_path
*path
,
1851 struct btrfs_extent_inline_ref
*iref
,
1853 struct btrfs_delayed_extent_op
*extent_op
,
1856 struct extent_buffer
*leaf
;
1857 struct btrfs_extent_item
*ei
;
1858 struct btrfs_extent_data_ref
*dref
= NULL
;
1859 struct btrfs_shared_data_ref
*sref
= NULL
;
1867 leaf
= path
->nodes
[0];
1868 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1869 refs
= btrfs_extent_refs(leaf
, ei
);
1870 WARN_ON(refs_to_mod
< 0 && refs
+ refs_to_mod
<= 0);
1871 refs
+= refs_to_mod
;
1872 btrfs_set_extent_refs(leaf
, ei
, refs
);
1874 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1876 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1878 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1879 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1880 refs
= btrfs_extent_data_ref_count(leaf
, dref
);
1881 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1882 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1883 refs
= btrfs_shared_data_ref_count(leaf
, sref
);
1886 BUG_ON(refs_to_mod
!= -1);
1889 BUG_ON(refs_to_mod
< 0 && refs
< -refs_to_mod
);
1890 refs
+= refs_to_mod
;
1893 if (type
== BTRFS_EXTENT_DATA_REF_KEY
)
1894 btrfs_set_extent_data_ref_count(leaf
, dref
, refs
);
1896 btrfs_set_shared_data_ref_count(leaf
, sref
, refs
);
1899 size
= btrfs_extent_inline_ref_size(type
);
1900 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1901 ptr
= (unsigned long)iref
;
1902 end
= (unsigned long)ei
+ item_size
;
1903 if (ptr
+ size
< end
)
1904 memmove_extent_buffer(leaf
, ptr
, ptr
+ size
,
1907 btrfs_truncate_item(fs_info
, path
, item_size
, 1);
1909 btrfs_mark_buffer_dirty(leaf
);
1912 static noinline_for_stack
1913 int insert_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1914 struct btrfs_fs_info
*fs_info
,
1915 struct btrfs_path
*path
,
1916 u64 bytenr
, u64 num_bytes
, u64 parent
,
1917 u64 root_objectid
, u64 owner
,
1918 u64 offset
, int refs_to_add
,
1919 struct btrfs_delayed_extent_op
*extent_op
)
1921 struct btrfs_extent_inline_ref
*iref
;
1924 ret
= lookup_inline_extent_backref(trans
, fs_info
, path
, &iref
,
1925 bytenr
, num_bytes
, parent
,
1926 root_objectid
, owner
, offset
, 1);
1928 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
);
1929 update_inline_extent_backref(fs_info
, path
, iref
,
1930 refs_to_add
, extent_op
, NULL
);
1931 } else if (ret
== -ENOENT
) {
1932 setup_inline_extent_backref(fs_info
, path
, iref
, parent
,
1933 root_objectid
, owner
, offset
,
1934 refs_to_add
, extent_op
);
1940 static int insert_extent_backref(struct btrfs_trans_handle
*trans
,
1941 struct btrfs_fs_info
*fs_info
,
1942 struct btrfs_path
*path
,
1943 u64 bytenr
, u64 parent
, u64 root_objectid
,
1944 u64 owner
, u64 offset
, int refs_to_add
)
1947 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1948 BUG_ON(refs_to_add
!= 1);
1949 ret
= insert_tree_block_ref(trans
, fs_info
, path
, bytenr
,
1950 parent
, root_objectid
);
1952 ret
= insert_extent_data_ref(trans
, fs_info
, path
, bytenr
,
1953 parent
, root_objectid
,
1954 owner
, offset
, refs_to_add
);
1959 static int remove_extent_backref(struct btrfs_trans_handle
*trans
,
1960 struct btrfs_fs_info
*fs_info
,
1961 struct btrfs_path
*path
,
1962 struct btrfs_extent_inline_ref
*iref
,
1963 int refs_to_drop
, int is_data
, int *last_ref
)
1967 BUG_ON(!is_data
&& refs_to_drop
!= 1);
1969 update_inline_extent_backref(fs_info
, path
, iref
,
1970 -refs_to_drop
, NULL
, last_ref
);
1971 } else if (is_data
) {
1972 ret
= remove_extent_data_ref(trans
, fs_info
, path
, refs_to_drop
,
1976 ret
= btrfs_del_item(trans
, fs_info
->extent_root
, path
);
1981 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
1982 static int btrfs_issue_discard(struct block_device
*bdev
, u64 start
, u64 len
,
1983 u64
*discarded_bytes
)
1986 u64 bytes_left
, end
;
1987 u64 aligned_start
= ALIGN(start
, 1 << 9);
1989 if (WARN_ON(start
!= aligned_start
)) {
1990 len
-= aligned_start
- start
;
1991 len
= round_down(len
, 1 << 9);
1992 start
= aligned_start
;
1995 *discarded_bytes
= 0;
2003 /* Skip any superblocks on this device. */
2004 for (j
= 0; j
< BTRFS_SUPER_MIRROR_MAX
; j
++) {
2005 u64 sb_start
= btrfs_sb_offset(j
);
2006 u64 sb_end
= sb_start
+ BTRFS_SUPER_INFO_SIZE
;
2007 u64 size
= sb_start
- start
;
2009 if (!in_range(sb_start
, start
, bytes_left
) &&
2010 !in_range(sb_end
, start
, bytes_left
) &&
2011 !in_range(start
, sb_start
, BTRFS_SUPER_INFO_SIZE
))
2015 * Superblock spans beginning of range. Adjust start and
2018 if (sb_start
<= start
) {
2019 start
+= sb_end
- start
;
2024 bytes_left
= end
- start
;
2029 ret
= blkdev_issue_discard(bdev
, start
>> 9, size
>> 9,
2032 *discarded_bytes
+= size
;
2033 else if (ret
!= -EOPNOTSUPP
)
2042 bytes_left
= end
- start
;
2046 ret
= blkdev_issue_discard(bdev
, start
>> 9, bytes_left
>> 9,
2049 *discarded_bytes
+= bytes_left
;
2054 int btrfs_discard_extent(struct btrfs_fs_info
*fs_info
, u64 bytenr
,
2055 u64 num_bytes
, u64
*actual_bytes
)
2058 u64 discarded_bytes
= 0;
2059 struct btrfs_bio
*bbio
= NULL
;
2063 * Avoid races with device replace and make sure our bbio has devices
2064 * associated to its stripes that don't go away while we are discarding.
2066 btrfs_bio_counter_inc_blocked(fs_info
);
2067 /* Tell the block device(s) that the sectors can be discarded */
2068 ret
= btrfs_map_block(fs_info
, BTRFS_MAP_DISCARD
, bytenr
, &num_bytes
,
2070 /* Error condition is -ENOMEM */
2072 struct btrfs_bio_stripe
*stripe
= bbio
->stripes
;
2076 for (i
= 0; i
< bbio
->num_stripes
; i
++, stripe
++) {
2078 if (!stripe
->dev
->can_discard
)
2081 ret
= btrfs_issue_discard(stripe
->dev
->bdev
,
2086 discarded_bytes
+= bytes
;
2087 else if (ret
!= -EOPNOTSUPP
)
2088 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2091 * Just in case we get back EOPNOTSUPP for some reason,
2092 * just ignore the return value so we don't screw up
2093 * people calling discard_extent.
2097 btrfs_put_bbio(bbio
);
2099 btrfs_bio_counter_dec(fs_info
);
2102 *actual_bytes
= discarded_bytes
;
2105 if (ret
== -EOPNOTSUPP
)
2110 /* Can return -ENOMEM */
2111 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
2112 struct btrfs_fs_info
*fs_info
,
2113 u64 bytenr
, u64 num_bytes
, u64 parent
,
2114 u64 root_objectid
, u64 owner
, u64 offset
)
2118 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
&&
2119 root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
2121 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
2122 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
2124 parent
, root_objectid
, (int)owner
,
2125 BTRFS_ADD_DELAYED_REF
, NULL
);
2127 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
2128 num_bytes
, parent
, root_objectid
,
2130 BTRFS_ADD_DELAYED_REF
);
2135 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
2136 struct btrfs_fs_info
*fs_info
,
2137 struct btrfs_delayed_ref_node
*node
,
2138 u64 parent
, u64 root_objectid
,
2139 u64 owner
, u64 offset
, int refs_to_add
,
2140 struct btrfs_delayed_extent_op
*extent_op
)
2142 struct btrfs_path
*path
;
2143 struct extent_buffer
*leaf
;
2144 struct btrfs_extent_item
*item
;
2145 struct btrfs_key key
;
2146 u64 bytenr
= node
->bytenr
;
2147 u64 num_bytes
= node
->num_bytes
;
2151 path
= btrfs_alloc_path();
2155 path
->reada
= READA_FORWARD
;
2156 path
->leave_spinning
= 1;
2157 /* this will setup the path even if it fails to insert the back ref */
2158 ret
= insert_inline_extent_backref(trans
, fs_info
, path
, bytenr
,
2159 num_bytes
, parent
, root_objectid
,
2161 refs_to_add
, extent_op
);
2162 if ((ret
< 0 && ret
!= -EAGAIN
) || !ret
)
2166 * Ok we had -EAGAIN which means we didn't have space to insert and
2167 * inline extent ref, so just update the reference count and add a
2170 leaf
= path
->nodes
[0];
2171 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2172 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2173 refs
= btrfs_extent_refs(leaf
, item
);
2174 btrfs_set_extent_refs(leaf
, item
, refs
+ refs_to_add
);
2176 __run_delayed_extent_op(extent_op
, leaf
, item
);
2178 btrfs_mark_buffer_dirty(leaf
);
2179 btrfs_release_path(path
);
2181 path
->reada
= READA_FORWARD
;
2182 path
->leave_spinning
= 1;
2183 /* now insert the actual backref */
2184 ret
= insert_extent_backref(trans
, fs_info
, path
, bytenr
, parent
,
2185 root_objectid
, owner
, offset
, refs_to_add
);
2187 btrfs_abort_transaction(trans
, ret
);
2189 btrfs_free_path(path
);
2193 static int run_delayed_data_ref(struct btrfs_trans_handle
*trans
,
2194 struct btrfs_fs_info
*fs_info
,
2195 struct btrfs_delayed_ref_node
*node
,
2196 struct btrfs_delayed_extent_op
*extent_op
,
2197 int insert_reserved
)
2200 struct btrfs_delayed_data_ref
*ref
;
2201 struct btrfs_key ins
;
2206 ins
.objectid
= node
->bytenr
;
2207 ins
.offset
= node
->num_bytes
;
2208 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2210 ref
= btrfs_delayed_node_to_data_ref(node
);
2211 trace_run_delayed_data_ref(fs_info
, node
, ref
, node
->action
);
2213 if (node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2214 parent
= ref
->parent
;
2215 ref_root
= ref
->root
;
2217 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2219 flags
|= extent_op
->flags_to_set
;
2220 ret
= alloc_reserved_file_extent(trans
, fs_info
,
2221 parent
, ref_root
, flags
,
2222 ref
->objectid
, ref
->offset
,
2223 &ins
, node
->ref_mod
);
2224 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2225 ret
= __btrfs_inc_extent_ref(trans
, fs_info
, node
, parent
,
2226 ref_root
, ref
->objectid
,
2227 ref
->offset
, node
->ref_mod
,
2229 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2230 ret
= __btrfs_free_extent(trans
, fs_info
, node
, parent
,
2231 ref_root
, ref
->objectid
,
2232 ref
->offset
, node
->ref_mod
,
2240 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
2241 struct extent_buffer
*leaf
,
2242 struct btrfs_extent_item
*ei
)
2244 u64 flags
= btrfs_extent_flags(leaf
, ei
);
2245 if (extent_op
->update_flags
) {
2246 flags
|= extent_op
->flags_to_set
;
2247 btrfs_set_extent_flags(leaf
, ei
, flags
);
2250 if (extent_op
->update_key
) {
2251 struct btrfs_tree_block_info
*bi
;
2252 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
));
2253 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
2254 btrfs_set_tree_block_key(leaf
, bi
, &extent_op
->key
);
2258 static int run_delayed_extent_op(struct btrfs_trans_handle
*trans
,
2259 struct btrfs_fs_info
*fs_info
,
2260 struct btrfs_delayed_ref_node
*node
,
2261 struct btrfs_delayed_extent_op
*extent_op
)
2263 struct btrfs_key key
;
2264 struct btrfs_path
*path
;
2265 struct btrfs_extent_item
*ei
;
2266 struct extent_buffer
*leaf
;
2270 int metadata
= !extent_op
->is_data
;
2275 if (metadata
&& !btrfs_fs_incompat(fs_info
, SKINNY_METADATA
))
2278 path
= btrfs_alloc_path();
2282 key
.objectid
= node
->bytenr
;
2285 key
.type
= BTRFS_METADATA_ITEM_KEY
;
2286 key
.offset
= extent_op
->level
;
2288 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2289 key
.offset
= node
->num_bytes
;
2293 path
->reada
= READA_FORWARD
;
2294 path
->leave_spinning
= 1;
2295 ret
= btrfs_search_slot(trans
, fs_info
->extent_root
, &key
, path
, 0, 1);
2302 if (path
->slots
[0] > 0) {
2304 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
2306 if (key
.objectid
== node
->bytenr
&&
2307 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
2308 key
.offset
== node
->num_bytes
)
2312 btrfs_release_path(path
);
2315 key
.objectid
= node
->bytenr
;
2316 key
.offset
= node
->num_bytes
;
2317 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2326 leaf
= path
->nodes
[0];
2327 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2328 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2329 if (item_size
< sizeof(*ei
)) {
2330 ret
= convert_extent_item_v0(trans
, fs_info
, path
, (u64
)-1, 0);
2335 leaf
= path
->nodes
[0];
2336 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2339 BUG_ON(item_size
< sizeof(*ei
));
2340 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2341 __run_delayed_extent_op(extent_op
, leaf
, ei
);
2343 btrfs_mark_buffer_dirty(leaf
);
2345 btrfs_free_path(path
);
2349 static int run_delayed_tree_ref(struct btrfs_trans_handle
*trans
,
2350 struct btrfs_fs_info
*fs_info
,
2351 struct btrfs_delayed_ref_node
*node
,
2352 struct btrfs_delayed_extent_op
*extent_op
,
2353 int insert_reserved
)
2356 struct btrfs_delayed_tree_ref
*ref
;
2357 struct btrfs_key ins
;
2360 bool skinny_metadata
= btrfs_fs_incompat(fs_info
, SKINNY_METADATA
);
2362 ref
= btrfs_delayed_node_to_tree_ref(node
);
2363 trace_run_delayed_tree_ref(fs_info
, node
, ref
, node
->action
);
2365 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2366 parent
= ref
->parent
;
2367 ref_root
= ref
->root
;
2369 ins
.objectid
= node
->bytenr
;
2370 if (skinny_metadata
) {
2371 ins
.offset
= ref
->level
;
2372 ins
.type
= BTRFS_METADATA_ITEM_KEY
;
2374 ins
.offset
= node
->num_bytes
;
2375 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2378 if (node
->ref_mod
!= 1) {
2380 "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu",
2381 node
->bytenr
, node
->ref_mod
, node
->action
, ref_root
,
2385 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2386 BUG_ON(!extent_op
|| !extent_op
->update_flags
);
2387 ret
= alloc_reserved_tree_block(trans
, fs_info
,
2389 extent_op
->flags_to_set
,
2392 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2393 ret
= __btrfs_inc_extent_ref(trans
, fs_info
, node
,
2397 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2398 ret
= __btrfs_free_extent(trans
, fs_info
, node
,
2400 ref
->level
, 0, 1, extent_op
);
2407 /* helper function to actually process a single delayed ref entry */
2408 static int run_one_delayed_ref(struct btrfs_trans_handle
*trans
,
2409 struct btrfs_fs_info
*fs_info
,
2410 struct btrfs_delayed_ref_node
*node
,
2411 struct btrfs_delayed_extent_op
*extent_op
,
2412 int insert_reserved
)
2416 if (trans
->aborted
) {
2417 if (insert_reserved
)
2418 btrfs_pin_extent(fs_info
, node
->bytenr
,
2419 node
->num_bytes
, 1);
2423 if (btrfs_delayed_ref_is_head(node
)) {
2424 struct btrfs_delayed_ref_head
*head
;
2426 * we've hit the end of the chain and we were supposed
2427 * to insert this extent into the tree. But, it got
2428 * deleted before we ever needed to insert it, so all
2429 * we have to do is clean up the accounting
2432 head
= btrfs_delayed_node_to_head(node
);
2433 trace_run_delayed_ref_head(fs_info
, node
, head
, node
->action
);
2435 if (insert_reserved
) {
2436 btrfs_pin_extent(fs_info
, node
->bytenr
,
2437 node
->num_bytes
, 1);
2438 if (head
->is_data
) {
2439 ret
= btrfs_del_csums(trans
, fs_info
,
2445 /* Also free its reserved qgroup space */
2446 btrfs_qgroup_free_delayed_ref(fs_info
, head
->qgroup_ref_root
,
2447 head
->qgroup_reserved
);
2451 if (node
->type
== BTRFS_TREE_BLOCK_REF_KEY
||
2452 node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2453 ret
= run_delayed_tree_ref(trans
, fs_info
, node
, extent_op
,
2455 else if (node
->type
== BTRFS_EXTENT_DATA_REF_KEY
||
2456 node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2457 ret
= run_delayed_data_ref(trans
, fs_info
, node
, extent_op
,
2464 static inline struct btrfs_delayed_ref_node
*
2465 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
2467 struct btrfs_delayed_ref_node
*ref
;
2469 if (list_empty(&head
->ref_list
))
2473 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2474 * This is to prevent a ref count from going down to zero, which deletes
2475 * the extent item from the extent tree, when there still are references
2476 * to add, which would fail because they would not find the extent item.
2478 if (!list_empty(&head
->ref_add_list
))
2479 return list_first_entry(&head
->ref_add_list
,
2480 struct btrfs_delayed_ref_node
, add_list
);
2482 ref
= list_first_entry(&head
->ref_list
, struct btrfs_delayed_ref_node
,
2484 ASSERT(list_empty(&ref
->add_list
));
2489 * Returns 0 on success or if called with an already aborted transaction.
2490 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2492 static noinline
int __btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2493 struct btrfs_fs_info
*fs_info
,
2496 struct btrfs_delayed_ref_root
*delayed_refs
;
2497 struct btrfs_delayed_ref_node
*ref
;
2498 struct btrfs_delayed_ref_head
*locked_ref
= NULL
;
2499 struct btrfs_delayed_extent_op
*extent_op
;
2500 ktime_t start
= ktime_get();
2502 unsigned long count
= 0;
2503 unsigned long actual_count
= 0;
2504 int must_insert_reserved
= 0;
2506 delayed_refs
= &trans
->transaction
->delayed_refs
;
2512 spin_lock(&delayed_refs
->lock
);
2513 locked_ref
= btrfs_select_ref_head(trans
);
2515 spin_unlock(&delayed_refs
->lock
);
2519 /* grab the lock that says we are going to process
2520 * all the refs for this head */
2521 ret
= btrfs_delayed_ref_lock(trans
, locked_ref
);
2522 spin_unlock(&delayed_refs
->lock
);
2524 * we may have dropped the spin lock to get the head
2525 * mutex lock, and that might have given someone else
2526 * time to free the head. If that's true, it has been
2527 * removed from our list and we can move on.
2529 if (ret
== -EAGAIN
) {
2537 * We need to try and merge add/drops of the same ref since we
2538 * can run into issues with relocate dropping the implicit ref
2539 * and then it being added back again before the drop can
2540 * finish. If we merged anything we need to re-loop so we can
2542 * Or we can get node references of the same type that weren't
2543 * merged when created due to bumps in the tree mod seq, and
2544 * we need to merge them to prevent adding an inline extent
2545 * backref before dropping it (triggering a BUG_ON at
2546 * insert_inline_extent_backref()).
2548 spin_lock(&locked_ref
->lock
);
2549 btrfs_merge_delayed_refs(trans
, fs_info
, delayed_refs
,
2553 * locked_ref is the head node, so we have to go one
2554 * node back for any delayed ref updates
2556 ref
= select_delayed_ref(locked_ref
);
2558 if (ref
&& ref
->seq
&&
2559 btrfs_check_delayed_seq(fs_info
, delayed_refs
, ref
->seq
)) {
2560 spin_unlock(&locked_ref
->lock
);
2561 spin_lock(&delayed_refs
->lock
);
2562 locked_ref
->processing
= 0;
2563 delayed_refs
->num_heads_ready
++;
2564 spin_unlock(&delayed_refs
->lock
);
2565 btrfs_delayed_ref_unlock(locked_ref
);
2573 * record the must insert reserved flag before we
2574 * drop the spin lock.
2576 must_insert_reserved
= locked_ref
->must_insert_reserved
;
2577 locked_ref
->must_insert_reserved
= 0;
2579 extent_op
= locked_ref
->extent_op
;
2580 locked_ref
->extent_op
= NULL
;
2585 /* All delayed refs have been processed, Go ahead
2586 * and send the head node to run_one_delayed_ref,
2587 * so that any accounting fixes can happen
2589 ref
= &locked_ref
->node
;
2591 if (extent_op
&& must_insert_reserved
) {
2592 btrfs_free_delayed_extent_op(extent_op
);
2597 spin_unlock(&locked_ref
->lock
);
2598 ret
= run_delayed_extent_op(trans
, fs_info
,
2600 btrfs_free_delayed_extent_op(extent_op
);
2604 * Need to reset must_insert_reserved if
2605 * there was an error so the abort stuff
2606 * can cleanup the reserved space
2609 if (must_insert_reserved
)
2610 locked_ref
->must_insert_reserved
= 1;
2611 spin_lock(&delayed_refs
->lock
);
2612 locked_ref
->processing
= 0;
2613 delayed_refs
->num_heads_ready
++;
2614 spin_unlock(&delayed_refs
->lock
);
2615 btrfs_debug(fs_info
,
2616 "run_delayed_extent_op returned %d",
2618 btrfs_delayed_ref_unlock(locked_ref
);
2625 * Need to drop our head ref lock and re-acquire the
2626 * delayed ref lock and then re-check to make sure
2629 spin_unlock(&locked_ref
->lock
);
2630 spin_lock(&delayed_refs
->lock
);
2631 spin_lock(&locked_ref
->lock
);
2632 if (!list_empty(&locked_ref
->ref_list
) ||
2633 locked_ref
->extent_op
) {
2634 spin_unlock(&locked_ref
->lock
);
2635 spin_unlock(&delayed_refs
->lock
);
2639 delayed_refs
->num_heads
--;
2640 rb_erase(&locked_ref
->href_node
,
2641 &delayed_refs
->href_root
);
2642 spin_unlock(&delayed_refs
->lock
);
2646 list_del(&ref
->list
);
2647 if (!list_empty(&ref
->add_list
))
2648 list_del(&ref
->add_list
);
2650 atomic_dec(&delayed_refs
->num_entries
);
2652 if (!btrfs_delayed_ref_is_head(ref
)) {
2654 * when we play the delayed ref, also correct the
2657 switch (ref
->action
) {
2658 case BTRFS_ADD_DELAYED_REF
:
2659 case BTRFS_ADD_DELAYED_EXTENT
:
2660 locked_ref
->node
.ref_mod
-= ref
->ref_mod
;
2662 case BTRFS_DROP_DELAYED_REF
:
2663 locked_ref
->node
.ref_mod
+= ref
->ref_mod
;
2669 spin_unlock(&locked_ref
->lock
);
2671 ret
= run_one_delayed_ref(trans
, fs_info
, ref
, extent_op
,
2672 must_insert_reserved
);
2674 btrfs_free_delayed_extent_op(extent_op
);
2676 spin_lock(&delayed_refs
->lock
);
2677 locked_ref
->processing
= 0;
2678 delayed_refs
->num_heads_ready
++;
2679 spin_unlock(&delayed_refs
->lock
);
2680 btrfs_delayed_ref_unlock(locked_ref
);
2681 btrfs_put_delayed_ref(ref
);
2682 btrfs_debug(fs_info
, "run_one_delayed_ref returned %d",
2688 * If this node is a head, that means all the refs in this head
2689 * have been dealt with, and we will pick the next head to deal
2690 * with, so we must unlock the head and drop it from the cluster
2691 * list before we release it.
2693 if (btrfs_delayed_ref_is_head(ref
)) {
2694 if (locked_ref
->is_data
&&
2695 locked_ref
->total_ref_mod
< 0) {
2696 spin_lock(&delayed_refs
->lock
);
2697 delayed_refs
->pending_csums
-= ref
->num_bytes
;
2698 spin_unlock(&delayed_refs
->lock
);
2700 btrfs_delayed_ref_unlock(locked_ref
);
2703 btrfs_put_delayed_ref(ref
);
2709 * We don't want to include ref heads since we can have empty ref heads
2710 * and those will drastically skew our runtime down since we just do
2711 * accounting, no actual extent tree updates.
2713 if (actual_count
> 0) {
2714 u64 runtime
= ktime_to_ns(ktime_sub(ktime_get(), start
));
2718 * We weigh the current average higher than our current runtime
2719 * to avoid large swings in the average.
2721 spin_lock(&delayed_refs
->lock
);
2722 avg
= fs_info
->avg_delayed_ref_runtime
* 3 + runtime
;
2723 fs_info
->avg_delayed_ref_runtime
= avg
>> 2; /* div by 4 */
2724 spin_unlock(&delayed_refs
->lock
);
2729 #ifdef SCRAMBLE_DELAYED_REFS
2731 * Normally delayed refs get processed in ascending bytenr order. This
2732 * correlates in most cases to the order added. To expose dependencies on this
2733 * order, we start to process the tree in the middle instead of the beginning
2735 static u64
find_middle(struct rb_root
*root
)
2737 struct rb_node
*n
= root
->rb_node
;
2738 struct btrfs_delayed_ref_node
*entry
;
2741 u64 first
= 0, last
= 0;
2745 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2746 first
= entry
->bytenr
;
2750 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2751 last
= entry
->bytenr
;
2756 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2757 WARN_ON(!entry
->in_tree
);
2759 middle
= entry
->bytenr
;
2772 static inline u64
heads_to_leaves(struct btrfs_fs_info
*fs_info
, u64 heads
)
2776 num_bytes
= heads
* (sizeof(struct btrfs_extent_item
) +
2777 sizeof(struct btrfs_extent_inline_ref
));
2778 if (!btrfs_fs_incompat(fs_info
, SKINNY_METADATA
))
2779 num_bytes
+= heads
* sizeof(struct btrfs_tree_block_info
);
2782 * We don't ever fill up leaves all the way so multiply by 2 just to be
2783 * closer to what we're really going to want to use.
2785 return div_u64(num_bytes
, BTRFS_LEAF_DATA_SIZE(fs_info
));
2789 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2790 * would require to store the csums for that many bytes.
2792 u64
btrfs_csum_bytes_to_leaves(struct btrfs_fs_info
*fs_info
, u64 csum_bytes
)
2795 u64 num_csums_per_leaf
;
2798 csum_size
= BTRFS_MAX_ITEM_SIZE(fs_info
);
2799 num_csums_per_leaf
= div64_u64(csum_size
,
2800 (u64
)btrfs_super_csum_size(fs_info
->super_copy
));
2801 num_csums
= div64_u64(csum_bytes
, fs_info
->sectorsize
);
2802 num_csums
+= num_csums_per_leaf
- 1;
2803 num_csums
= div64_u64(num_csums
, num_csums_per_leaf
);
2807 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle
*trans
,
2808 struct btrfs_fs_info
*fs_info
)
2810 struct btrfs_block_rsv
*global_rsv
;
2811 u64 num_heads
= trans
->transaction
->delayed_refs
.num_heads_ready
;
2812 u64 csum_bytes
= trans
->transaction
->delayed_refs
.pending_csums
;
2813 u64 num_dirty_bgs
= trans
->transaction
->num_dirty_bgs
;
2814 u64 num_bytes
, num_dirty_bgs_bytes
;
2817 num_bytes
= btrfs_calc_trans_metadata_size(fs_info
, 1);
2818 num_heads
= heads_to_leaves(fs_info
, num_heads
);
2820 num_bytes
+= (num_heads
- 1) * fs_info
->nodesize
;
2822 num_bytes
+= btrfs_csum_bytes_to_leaves(fs_info
, csum_bytes
) *
2824 num_dirty_bgs_bytes
= btrfs_calc_trans_metadata_size(fs_info
,
2826 global_rsv
= &fs_info
->global_block_rsv
;
2829 * If we can't allocate any more chunks lets make sure we have _lots_ of
2830 * wiggle room since running delayed refs can create more delayed refs.
2832 if (global_rsv
->space_info
->full
) {
2833 num_dirty_bgs_bytes
<<= 1;
2837 spin_lock(&global_rsv
->lock
);
2838 if (global_rsv
->reserved
<= num_bytes
+ num_dirty_bgs_bytes
)
2840 spin_unlock(&global_rsv
->lock
);
2844 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle
*trans
,
2845 struct btrfs_fs_info
*fs_info
)
2848 atomic_read(&trans
->transaction
->delayed_refs
.num_entries
);
2853 avg_runtime
= fs_info
->avg_delayed_ref_runtime
;
2854 val
= num_entries
* avg_runtime
;
2855 if (val
>= NSEC_PER_SEC
)
2857 if (val
>= NSEC_PER_SEC
/ 2)
2860 return btrfs_check_space_for_delayed_refs(trans
, fs_info
);
2863 struct async_delayed_refs
{
2864 struct btrfs_root
*root
;
2869 struct completion wait
;
2870 struct btrfs_work work
;
2873 static inline struct async_delayed_refs
*
2874 to_async_delayed_refs(struct btrfs_work
*work
)
2876 return container_of(work
, struct async_delayed_refs
, work
);
2879 static void delayed_ref_async_start(struct btrfs_work
*work
)
2881 struct async_delayed_refs
*async
= to_async_delayed_refs(work
);
2882 struct btrfs_trans_handle
*trans
;
2883 struct btrfs_fs_info
*fs_info
= async
->root
->fs_info
;
2886 /* if the commit is already started, we don't need to wait here */
2887 if (btrfs_transaction_blocked(fs_info
))
2890 trans
= btrfs_join_transaction(async
->root
);
2891 if (IS_ERR(trans
)) {
2892 async
->error
= PTR_ERR(trans
);
2897 * trans->sync means that when we call end_transaction, we won't
2898 * wait on delayed refs
2902 /* Don't bother flushing if we got into a different transaction */
2903 if (trans
->transid
> async
->transid
)
2906 ret
= btrfs_run_delayed_refs(trans
, fs_info
, async
->count
);
2910 ret
= btrfs_end_transaction(trans
);
2911 if (ret
&& !async
->error
)
2915 complete(&async
->wait
);
2920 int btrfs_async_run_delayed_refs(struct btrfs_fs_info
*fs_info
,
2921 unsigned long count
, u64 transid
, int wait
)
2923 struct async_delayed_refs
*async
;
2926 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
2930 async
->root
= fs_info
->tree_root
;
2931 async
->count
= count
;
2933 async
->transid
= transid
;
2938 init_completion(&async
->wait
);
2940 btrfs_init_work(&async
->work
, btrfs_extent_refs_helper
,
2941 delayed_ref_async_start
, NULL
, NULL
);
2943 btrfs_queue_work(fs_info
->extent_workers
, &async
->work
);
2946 wait_for_completion(&async
->wait
);
2955 * this starts processing the delayed reference count updates and
2956 * extent insertions we have queued up so far. count can be
2957 * 0, which means to process everything in the tree at the start
2958 * of the run (but not newly added entries), or it can be some target
2959 * number you'd like to process.
2961 * Returns 0 on success or if called with an aborted transaction
2962 * Returns <0 on error and aborts the transaction
2964 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2965 struct btrfs_fs_info
*fs_info
, unsigned long count
)
2967 struct rb_node
*node
;
2968 struct btrfs_delayed_ref_root
*delayed_refs
;
2969 struct btrfs_delayed_ref_head
*head
;
2971 int run_all
= count
== (unsigned long)-1;
2972 bool can_flush_pending_bgs
= trans
->can_flush_pending_bgs
;
2974 /* We'll clean this up in btrfs_cleanup_transaction */
2978 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE
, &fs_info
->flags
))
2981 delayed_refs
= &trans
->transaction
->delayed_refs
;
2983 count
= atomic_read(&delayed_refs
->num_entries
) * 2;
2986 #ifdef SCRAMBLE_DELAYED_REFS
2987 delayed_refs
->run_delayed_start
= find_middle(&delayed_refs
->root
);
2989 trans
->can_flush_pending_bgs
= false;
2990 ret
= __btrfs_run_delayed_refs(trans
, fs_info
, count
);
2992 btrfs_abort_transaction(trans
, ret
);
2997 if (!list_empty(&trans
->new_bgs
))
2998 btrfs_create_pending_block_groups(trans
, fs_info
);
3000 spin_lock(&delayed_refs
->lock
);
3001 node
= rb_first(&delayed_refs
->href_root
);
3003 spin_unlock(&delayed_refs
->lock
);
3008 head
= rb_entry(node
, struct btrfs_delayed_ref_head
,
3010 if (btrfs_delayed_ref_is_head(&head
->node
)) {
3011 struct btrfs_delayed_ref_node
*ref
;
3014 refcount_inc(&ref
->refs
);
3016 spin_unlock(&delayed_refs
->lock
);
3018 * Mutex was contended, block until it's
3019 * released and try again
3021 mutex_lock(&head
->mutex
);
3022 mutex_unlock(&head
->mutex
);
3024 btrfs_put_delayed_ref(ref
);
3030 node
= rb_next(node
);
3032 spin_unlock(&delayed_refs
->lock
);
3037 trans
->can_flush_pending_bgs
= can_flush_pending_bgs
;
3041 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
3042 struct btrfs_fs_info
*fs_info
,
3043 u64 bytenr
, u64 num_bytes
, u64 flags
,
3044 int level
, int is_data
)
3046 struct btrfs_delayed_extent_op
*extent_op
;
3049 extent_op
= btrfs_alloc_delayed_extent_op();
3053 extent_op
->flags_to_set
= flags
;
3054 extent_op
->update_flags
= true;
3055 extent_op
->update_key
= false;
3056 extent_op
->is_data
= is_data
? true : false;
3057 extent_op
->level
= level
;
3059 ret
= btrfs_add_delayed_extent_op(fs_info
, trans
, bytenr
,
3060 num_bytes
, extent_op
);
3062 btrfs_free_delayed_extent_op(extent_op
);
3066 static noinline
int check_delayed_ref(struct btrfs_root
*root
,
3067 struct btrfs_path
*path
,
3068 u64 objectid
, u64 offset
, u64 bytenr
)
3070 struct btrfs_delayed_ref_head
*head
;
3071 struct btrfs_delayed_ref_node
*ref
;
3072 struct btrfs_delayed_data_ref
*data_ref
;
3073 struct btrfs_delayed_ref_root
*delayed_refs
;
3074 struct btrfs_transaction
*cur_trans
;
3077 cur_trans
= root
->fs_info
->running_transaction
;
3081 delayed_refs
= &cur_trans
->delayed_refs
;
3082 spin_lock(&delayed_refs
->lock
);
3083 head
= btrfs_find_delayed_ref_head(delayed_refs
, bytenr
);
3085 spin_unlock(&delayed_refs
->lock
);
3089 if (!mutex_trylock(&head
->mutex
)) {
3090 refcount_inc(&head
->node
.refs
);
3091 spin_unlock(&delayed_refs
->lock
);
3093 btrfs_release_path(path
);
3096 * Mutex was contended, block until it's released and let
3099 mutex_lock(&head
->mutex
);
3100 mutex_unlock(&head
->mutex
);
3101 btrfs_put_delayed_ref(&head
->node
);
3104 spin_unlock(&delayed_refs
->lock
);
3106 spin_lock(&head
->lock
);
3107 list_for_each_entry(ref
, &head
->ref_list
, list
) {
3108 /* If it's a shared ref we know a cross reference exists */
3109 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
) {
3114 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
3117 * If our ref doesn't match the one we're currently looking at
3118 * then we have a cross reference.
3120 if (data_ref
->root
!= root
->root_key
.objectid
||
3121 data_ref
->objectid
!= objectid
||
3122 data_ref
->offset
!= offset
) {
3127 spin_unlock(&head
->lock
);
3128 mutex_unlock(&head
->mutex
);
3132 static noinline
int check_committed_ref(struct btrfs_root
*root
,
3133 struct btrfs_path
*path
,
3134 u64 objectid
, u64 offset
, u64 bytenr
)
3136 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3137 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
3138 struct extent_buffer
*leaf
;
3139 struct btrfs_extent_data_ref
*ref
;
3140 struct btrfs_extent_inline_ref
*iref
;
3141 struct btrfs_extent_item
*ei
;
3142 struct btrfs_key key
;
3146 key
.objectid
= bytenr
;
3147 key
.offset
= (u64
)-1;
3148 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
3150 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
3153 BUG_ON(ret
== 0); /* Corruption */
3156 if (path
->slots
[0] == 0)
3160 leaf
= path
->nodes
[0];
3161 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
3163 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
3167 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
3168 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3169 if (item_size
< sizeof(*ei
)) {
3170 WARN_ON(item_size
!= sizeof(struct btrfs_extent_item_v0
));
3174 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
3176 if (item_size
!= sizeof(*ei
) +
3177 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
3180 if (btrfs_extent_generation(leaf
, ei
) <=
3181 btrfs_root_last_snapshot(&root
->root_item
))
3184 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
3185 if (btrfs_extent_inline_ref_type(leaf
, iref
) !=
3186 BTRFS_EXTENT_DATA_REF_KEY
)
3189 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
3190 if (btrfs_extent_refs(leaf
, ei
) !=
3191 btrfs_extent_data_ref_count(leaf
, ref
) ||
3192 btrfs_extent_data_ref_root(leaf
, ref
) !=
3193 root
->root_key
.objectid
||
3194 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
3195 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
3203 int btrfs_cross_ref_exist(struct btrfs_root
*root
, u64 objectid
, u64 offset
,
3206 struct btrfs_path
*path
;
3210 path
= btrfs_alloc_path();
3215 ret
= check_committed_ref(root
, path
, objectid
,
3217 if (ret
&& ret
!= -ENOENT
)
3220 ret2
= check_delayed_ref(root
, path
, objectid
,
3222 } while (ret2
== -EAGAIN
);
3224 if (ret2
&& ret2
!= -ENOENT
) {
3229 if (ret
!= -ENOENT
|| ret2
!= -ENOENT
)
3232 btrfs_free_path(path
);
3233 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
3238 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
3239 struct btrfs_root
*root
,
3240 struct extent_buffer
*buf
,
3241 int full_backref
, int inc
)
3243 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3249 struct btrfs_key key
;
3250 struct btrfs_file_extent_item
*fi
;
3254 int (*process_func
)(struct btrfs_trans_handle
*,
3255 struct btrfs_fs_info
*,
3256 u64
, u64
, u64
, u64
, u64
, u64
);
3259 if (btrfs_is_testing(fs_info
))
3262 ref_root
= btrfs_header_owner(buf
);
3263 nritems
= btrfs_header_nritems(buf
);
3264 level
= btrfs_header_level(buf
);
3266 if (!test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) && level
== 0)
3270 process_func
= btrfs_inc_extent_ref
;
3272 process_func
= btrfs_free_extent
;
3275 parent
= buf
->start
;
3279 for (i
= 0; i
< nritems
; i
++) {
3281 btrfs_item_key_to_cpu(buf
, &key
, i
);
3282 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
3284 fi
= btrfs_item_ptr(buf
, i
,
3285 struct btrfs_file_extent_item
);
3286 if (btrfs_file_extent_type(buf
, fi
) ==
3287 BTRFS_FILE_EXTENT_INLINE
)
3289 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
3293 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
3294 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
3295 ret
= process_func(trans
, fs_info
, bytenr
, num_bytes
,
3296 parent
, ref_root
, key
.objectid
,
3301 bytenr
= btrfs_node_blockptr(buf
, i
);
3302 num_bytes
= fs_info
->nodesize
;
3303 ret
= process_func(trans
, fs_info
, bytenr
, num_bytes
,
3304 parent
, ref_root
, level
- 1, 0);
3314 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3315 struct extent_buffer
*buf
, int full_backref
)
3317 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1);
3320 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3321 struct extent_buffer
*buf
, int full_backref
)
3323 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0);
3326 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
3327 struct btrfs_fs_info
*fs_info
,
3328 struct btrfs_path
*path
,
3329 struct btrfs_block_group_cache
*cache
)
3332 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
3334 struct extent_buffer
*leaf
;
3336 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
3343 leaf
= path
->nodes
[0];
3344 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
3345 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
3346 btrfs_mark_buffer_dirty(leaf
);
3348 btrfs_release_path(path
);
3353 static struct btrfs_block_group_cache
*
3354 next_block_group(struct btrfs_fs_info
*fs_info
,
3355 struct btrfs_block_group_cache
*cache
)
3357 struct rb_node
*node
;
3359 spin_lock(&fs_info
->block_group_cache_lock
);
3361 /* If our block group was removed, we need a full search. */
3362 if (RB_EMPTY_NODE(&cache
->cache_node
)) {
3363 const u64 next_bytenr
= cache
->key
.objectid
+ cache
->key
.offset
;
3365 spin_unlock(&fs_info
->block_group_cache_lock
);
3366 btrfs_put_block_group(cache
);
3367 cache
= btrfs_lookup_first_block_group(fs_info
, next_bytenr
); return cache
;
3369 node
= rb_next(&cache
->cache_node
);
3370 btrfs_put_block_group(cache
);
3372 cache
= rb_entry(node
, struct btrfs_block_group_cache
,
3374 btrfs_get_block_group(cache
);
3377 spin_unlock(&fs_info
->block_group_cache_lock
);
3381 static int cache_save_setup(struct btrfs_block_group_cache
*block_group
,
3382 struct btrfs_trans_handle
*trans
,
3383 struct btrfs_path
*path
)
3385 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
3386 struct btrfs_root
*root
= fs_info
->tree_root
;
3387 struct inode
*inode
= NULL
;
3389 int dcs
= BTRFS_DC_ERROR
;
3395 * If this block group is smaller than 100 megs don't bother caching the
3398 if (block_group
->key
.offset
< (100 * SZ_1M
)) {
3399 spin_lock(&block_group
->lock
);
3400 block_group
->disk_cache_state
= BTRFS_DC_WRITTEN
;
3401 spin_unlock(&block_group
->lock
);
3408 inode
= lookup_free_space_inode(fs_info
, block_group
, path
);
3409 if (IS_ERR(inode
) && PTR_ERR(inode
) != -ENOENT
) {
3410 ret
= PTR_ERR(inode
);
3411 btrfs_release_path(path
);
3415 if (IS_ERR(inode
)) {
3419 if (block_group
->ro
)
3422 ret
= create_free_space_inode(fs_info
, trans
, block_group
,
3429 /* We've already setup this transaction, go ahead and exit */
3430 if (block_group
->cache_generation
== trans
->transid
&&
3431 i_size_read(inode
)) {
3432 dcs
= BTRFS_DC_SETUP
;
3437 * We want to set the generation to 0, that way if anything goes wrong
3438 * from here on out we know not to trust this cache when we load up next
3441 BTRFS_I(inode
)->generation
= 0;
3442 ret
= btrfs_update_inode(trans
, root
, inode
);
3445 * So theoretically we could recover from this, simply set the
3446 * super cache generation to 0 so we know to invalidate the
3447 * cache, but then we'd have to keep track of the block groups
3448 * that fail this way so we know we _have_ to reset this cache
3449 * before the next commit or risk reading stale cache. So to
3450 * limit our exposure to horrible edge cases lets just abort the
3451 * transaction, this only happens in really bad situations
3454 btrfs_abort_transaction(trans
, ret
);
3459 if (i_size_read(inode
) > 0) {
3460 ret
= btrfs_check_trunc_cache_free_space(fs_info
,
3461 &fs_info
->global_block_rsv
);
3465 ret
= btrfs_truncate_free_space_cache(trans
, NULL
, inode
);
3470 spin_lock(&block_group
->lock
);
3471 if (block_group
->cached
!= BTRFS_CACHE_FINISHED
||
3472 !btrfs_test_opt(fs_info
, SPACE_CACHE
)) {
3474 * don't bother trying to write stuff out _if_
3475 * a) we're not cached,
3476 * b) we're with nospace_cache mount option,
3477 * c) we're with v2 space_cache (FREE_SPACE_TREE).
3479 dcs
= BTRFS_DC_WRITTEN
;
3480 spin_unlock(&block_group
->lock
);
3483 spin_unlock(&block_group
->lock
);
3486 * We hit an ENOSPC when setting up the cache in this transaction, just
3487 * skip doing the setup, we've already cleared the cache so we're safe.
3489 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC
, &trans
->transaction
->flags
)) {
3495 * Try to preallocate enough space based on how big the block group is.
3496 * Keep in mind this has to include any pinned space which could end up
3497 * taking up quite a bit since it's not folded into the other space
3500 num_pages
= div_u64(block_group
->key
.offset
, SZ_256M
);
3505 num_pages
*= PAGE_SIZE
;
3507 ret
= btrfs_check_data_free_space(inode
, 0, num_pages
);
3511 ret
= btrfs_prealloc_file_range_trans(inode
, trans
, 0, 0, num_pages
,
3512 num_pages
, num_pages
,
3515 * Our cache requires contiguous chunks so that we don't modify a bunch
3516 * of metadata or split extents when writing the cache out, which means
3517 * we can enospc if we are heavily fragmented in addition to just normal
3518 * out of space conditions. So if we hit this just skip setting up any
3519 * other block groups for this transaction, maybe we'll unpin enough
3520 * space the next time around.
3523 dcs
= BTRFS_DC_SETUP
;
3524 else if (ret
== -ENOSPC
)
3525 set_bit(BTRFS_TRANS_CACHE_ENOSPC
, &trans
->transaction
->flags
);
3530 btrfs_release_path(path
);
3532 spin_lock(&block_group
->lock
);
3533 if (!ret
&& dcs
== BTRFS_DC_SETUP
)
3534 block_group
->cache_generation
= trans
->transid
;
3535 block_group
->disk_cache_state
= dcs
;
3536 spin_unlock(&block_group
->lock
);
3541 int btrfs_setup_space_cache(struct btrfs_trans_handle
*trans
,
3542 struct btrfs_fs_info
*fs_info
)
3544 struct btrfs_block_group_cache
*cache
, *tmp
;
3545 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3546 struct btrfs_path
*path
;
3548 if (list_empty(&cur_trans
->dirty_bgs
) ||
3549 !btrfs_test_opt(fs_info
, SPACE_CACHE
))
3552 path
= btrfs_alloc_path();
3556 /* Could add new block groups, use _safe just in case */
3557 list_for_each_entry_safe(cache
, tmp
, &cur_trans
->dirty_bgs
,
3559 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
)
3560 cache_save_setup(cache
, trans
, path
);
3563 btrfs_free_path(path
);
3568 * transaction commit does final block group cache writeback during a
3569 * critical section where nothing is allowed to change the FS. This is
3570 * required in order for the cache to actually match the block group,
3571 * but can introduce a lot of latency into the commit.
3573 * So, btrfs_start_dirty_block_groups is here to kick off block group
3574 * cache IO. There's a chance we'll have to redo some of it if the
3575 * block group changes again during the commit, but it greatly reduces
3576 * the commit latency by getting rid of the easy block groups while
3577 * we're still allowing others to join the commit.
3579 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle
*trans
,
3580 struct btrfs_fs_info
*fs_info
)
3582 struct btrfs_block_group_cache
*cache
;
3583 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3586 struct btrfs_path
*path
= NULL
;
3588 struct list_head
*io
= &cur_trans
->io_bgs
;
3589 int num_started
= 0;
3592 spin_lock(&cur_trans
->dirty_bgs_lock
);
3593 if (list_empty(&cur_trans
->dirty_bgs
)) {
3594 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3597 list_splice_init(&cur_trans
->dirty_bgs
, &dirty
);
3598 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3602 * make sure all the block groups on our dirty list actually
3605 btrfs_create_pending_block_groups(trans
, fs_info
);
3608 path
= btrfs_alloc_path();
3614 * cache_write_mutex is here only to save us from balance or automatic
3615 * removal of empty block groups deleting this block group while we are
3616 * writing out the cache
3618 mutex_lock(&trans
->transaction
->cache_write_mutex
);
3619 while (!list_empty(&dirty
)) {
3620 cache
= list_first_entry(&dirty
,
3621 struct btrfs_block_group_cache
,
3624 * this can happen if something re-dirties a block
3625 * group that is already under IO. Just wait for it to
3626 * finish and then do it all again
3628 if (!list_empty(&cache
->io_list
)) {
3629 list_del_init(&cache
->io_list
);
3630 btrfs_wait_cache_io(trans
, cache
, path
);
3631 btrfs_put_block_group(cache
);
3636 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3637 * if it should update the cache_state. Don't delete
3638 * until after we wait.
3640 * Since we're not running in the commit critical section
3641 * we need the dirty_bgs_lock to protect from update_block_group
3643 spin_lock(&cur_trans
->dirty_bgs_lock
);
3644 list_del_init(&cache
->dirty_list
);
3645 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3649 cache_save_setup(cache
, trans
, path
);
3651 if (cache
->disk_cache_state
== BTRFS_DC_SETUP
) {
3652 cache
->io_ctl
.inode
= NULL
;
3653 ret
= btrfs_write_out_cache(fs_info
, trans
,
3655 if (ret
== 0 && cache
->io_ctl
.inode
) {
3660 * the cache_write_mutex is protecting
3663 list_add_tail(&cache
->io_list
, io
);
3666 * if we failed to write the cache, the
3667 * generation will be bad and life goes on
3673 ret
= write_one_cache_group(trans
, fs_info
,
3676 * Our block group might still be attached to the list
3677 * of new block groups in the transaction handle of some
3678 * other task (struct btrfs_trans_handle->new_bgs). This
3679 * means its block group item isn't yet in the extent
3680 * tree. If this happens ignore the error, as we will
3681 * try again later in the critical section of the
3682 * transaction commit.
3684 if (ret
== -ENOENT
) {
3686 spin_lock(&cur_trans
->dirty_bgs_lock
);
3687 if (list_empty(&cache
->dirty_list
)) {
3688 list_add_tail(&cache
->dirty_list
,
3689 &cur_trans
->dirty_bgs
);
3690 btrfs_get_block_group(cache
);
3692 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3694 btrfs_abort_transaction(trans
, ret
);
3698 /* if its not on the io list, we need to put the block group */
3700 btrfs_put_block_group(cache
);
3706 * Avoid blocking other tasks for too long. It might even save
3707 * us from writing caches for block groups that are going to be
3710 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
3711 mutex_lock(&trans
->transaction
->cache_write_mutex
);
3713 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
3716 * go through delayed refs for all the stuff we've just kicked off
3717 * and then loop back (just once)
3719 ret
= btrfs_run_delayed_refs(trans
, fs_info
, 0);
3720 if (!ret
&& loops
== 0) {
3722 spin_lock(&cur_trans
->dirty_bgs_lock
);
3723 list_splice_init(&cur_trans
->dirty_bgs
, &dirty
);
3725 * dirty_bgs_lock protects us from concurrent block group
3726 * deletes too (not just cache_write_mutex).
3728 if (!list_empty(&dirty
)) {
3729 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3732 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3733 } else if (ret
< 0) {
3734 btrfs_cleanup_dirty_bgs(cur_trans
, fs_info
);
3737 btrfs_free_path(path
);
3741 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
3742 struct btrfs_fs_info
*fs_info
)
3744 struct btrfs_block_group_cache
*cache
;
3745 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3748 struct btrfs_path
*path
;
3749 struct list_head
*io
= &cur_trans
->io_bgs
;
3750 int num_started
= 0;
3752 path
= btrfs_alloc_path();
3757 * Even though we are in the critical section of the transaction commit,
3758 * we can still have concurrent tasks adding elements to this
3759 * transaction's list of dirty block groups. These tasks correspond to
3760 * endio free space workers started when writeback finishes for a
3761 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3762 * allocate new block groups as a result of COWing nodes of the root
3763 * tree when updating the free space inode. The writeback for the space
3764 * caches is triggered by an earlier call to
3765 * btrfs_start_dirty_block_groups() and iterations of the following
3767 * Also we want to do the cache_save_setup first and then run the
3768 * delayed refs to make sure we have the best chance at doing this all
3771 spin_lock(&cur_trans
->dirty_bgs_lock
);
3772 while (!list_empty(&cur_trans
->dirty_bgs
)) {
3773 cache
= list_first_entry(&cur_trans
->dirty_bgs
,
3774 struct btrfs_block_group_cache
,
3778 * this can happen if cache_save_setup re-dirties a block
3779 * group that is already under IO. Just wait for it to
3780 * finish and then do it all again
3782 if (!list_empty(&cache
->io_list
)) {
3783 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3784 list_del_init(&cache
->io_list
);
3785 btrfs_wait_cache_io(trans
, cache
, path
);
3786 btrfs_put_block_group(cache
);
3787 spin_lock(&cur_trans
->dirty_bgs_lock
);
3791 * don't remove from the dirty list until after we've waited
3794 list_del_init(&cache
->dirty_list
);
3795 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3798 cache_save_setup(cache
, trans
, path
);
3801 ret
= btrfs_run_delayed_refs(trans
, fs_info
,
3802 (unsigned long) -1);
3804 if (!ret
&& cache
->disk_cache_state
== BTRFS_DC_SETUP
) {
3805 cache
->io_ctl
.inode
= NULL
;
3806 ret
= btrfs_write_out_cache(fs_info
, trans
,
3808 if (ret
== 0 && cache
->io_ctl
.inode
) {
3811 list_add_tail(&cache
->io_list
, io
);
3814 * if we failed to write the cache, the
3815 * generation will be bad and life goes on
3821 ret
= write_one_cache_group(trans
, fs_info
,
3824 * One of the free space endio workers might have
3825 * created a new block group while updating a free space
3826 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3827 * and hasn't released its transaction handle yet, in
3828 * which case the new block group is still attached to
3829 * its transaction handle and its creation has not
3830 * finished yet (no block group item in the extent tree
3831 * yet, etc). If this is the case, wait for all free
3832 * space endio workers to finish and retry. This is a
3833 * a very rare case so no need for a more efficient and
3836 if (ret
== -ENOENT
) {
3837 wait_event(cur_trans
->writer_wait
,
3838 atomic_read(&cur_trans
->num_writers
) == 1);
3839 ret
= write_one_cache_group(trans
, fs_info
,
3843 btrfs_abort_transaction(trans
, ret
);
3846 /* if its not on the io list, we need to put the block group */
3848 btrfs_put_block_group(cache
);
3849 spin_lock(&cur_trans
->dirty_bgs_lock
);
3851 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3853 while (!list_empty(io
)) {
3854 cache
= list_first_entry(io
, struct btrfs_block_group_cache
,
3856 list_del_init(&cache
->io_list
);
3857 btrfs_wait_cache_io(trans
, cache
, path
);
3858 btrfs_put_block_group(cache
);
3861 btrfs_free_path(path
);
3865 int btrfs_extent_readonly(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
3867 struct btrfs_block_group_cache
*block_group
;
3870 block_group
= btrfs_lookup_block_group(fs_info
, bytenr
);
3871 if (!block_group
|| block_group
->ro
)
3874 btrfs_put_block_group(block_group
);
3878 bool btrfs_inc_nocow_writers(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
3880 struct btrfs_block_group_cache
*bg
;
3883 bg
= btrfs_lookup_block_group(fs_info
, bytenr
);
3887 spin_lock(&bg
->lock
);
3891 atomic_inc(&bg
->nocow_writers
);
3892 spin_unlock(&bg
->lock
);
3894 /* no put on block group, done by btrfs_dec_nocow_writers */
3896 btrfs_put_block_group(bg
);
3902 void btrfs_dec_nocow_writers(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
3904 struct btrfs_block_group_cache
*bg
;
3906 bg
= btrfs_lookup_block_group(fs_info
, bytenr
);
3908 if (atomic_dec_and_test(&bg
->nocow_writers
))
3909 wake_up_atomic_t(&bg
->nocow_writers
);
3911 * Once for our lookup and once for the lookup done by a previous call
3912 * to btrfs_inc_nocow_writers()
3914 btrfs_put_block_group(bg
);
3915 btrfs_put_block_group(bg
);
3918 static int btrfs_wait_nocow_writers_atomic_t(atomic_t
*a
)
3924 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache
*bg
)
3926 wait_on_atomic_t(&bg
->nocow_writers
,
3927 btrfs_wait_nocow_writers_atomic_t
,
3928 TASK_UNINTERRUPTIBLE
);
3931 static const char *alloc_name(u64 flags
)
3934 case BTRFS_BLOCK_GROUP_METADATA
|BTRFS_BLOCK_GROUP_DATA
:
3936 case BTRFS_BLOCK_GROUP_METADATA
:
3938 case BTRFS_BLOCK_GROUP_DATA
:
3940 case BTRFS_BLOCK_GROUP_SYSTEM
:
3944 return "invalid-combination";
3948 static int create_space_info(struct btrfs_fs_info
*info
, u64 flags
,
3949 struct btrfs_space_info
**new)
3952 struct btrfs_space_info
*space_info
;
3956 space_info
= kzalloc(sizeof(*space_info
), GFP_NOFS
);
3960 ret
= percpu_counter_init(&space_info
->total_bytes_pinned
, 0,
3967 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
3968 INIT_LIST_HEAD(&space_info
->block_groups
[i
]);
3969 init_rwsem(&space_info
->groups_sem
);
3970 spin_lock_init(&space_info
->lock
);
3971 space_info
->flags
= flags
& BTRFS_BLOCK_GROUP_TYPE_MASK
;
3972 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3973 init_waitqueue_head(&space_info
->wait
);
3974 INIT_LIST_HEAD(&space_info
->ro_bgs
);
3975 INIT_LIST_HEAD(&space_info
->tickets
);
3976 INIT_LIST_HEAD(&space_info
->priority_tickets
);
3978 ret
= kobject_init_and_add(&space_info
->kobj
, &space_info_ktype
,
3979 info
->space_info_kobj
, "%s",
3980 alloc_name(space_info
->flags
));
3982 percpu_counter_destroy(&space_info
->total_bytes_pinned
);
3988 list_add_rcu(&space_info
->list
, &info
->space_info
);
3989 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3990 info
->data_sinfo
= space_info
;
3995 static void update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
3996 u64 total_bytes
, u64 bytes_used
,
3998 struct btrfs_space_info
**space_info
)
4000 struct btrfs_space_info
*found
;
4003 if (flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
4004 BTRFS_BLOCK_GROUP_RAID10
))
4009 found
= __find_space_info(info
, flags
);
4011 spin_lock(&found
->lock
);
4012 found
->total_bytes
+= total_bytes
;
4013 found
->disk_total
+= total_bytes
* factor
;
4014 found
->bytes_used
+= bytes_used
;
4015 found
->disk_used
+= bytes_used
* factor
;
4016 found
->bytes_readonly
+= bytes_readonly
;
4017 if (total_bytes
> 0)
4019 space_info_add_new_bytes(info
, found
, total_bytes
-
4020 bytes_used
- bytes_readonly
);
4021 spin_unlock(&found
->lock
);
4022 *space_info
= found
;
4025 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
4027 u64 extra_flags
= chunk_to_extended(flags
) &
4028 BTRFS_EXTENDED_PROFILE_MASK
;
4030 write_seqlock(&fs_info
->profiles_lock
);
4031 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
4032 fs_info
->avail_data_alloc_bits
|= extra_flags
;
4033 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
4034 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
4035 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
4036 fs_info
->avail_system_alloc_bits
|= extra_flags
;
4037 write_sequnlock(&fs_info
->profiles_lock
);
4041 * returns target flags in extended format or 0 if restripe for this
4042 * chunk_type is not in progress
4044 * should be called with either volume_mutex or balance_lock held
4046 static u64
get_restripe_target(struct btrfs_fs_info
*fs_info
, u64 flags
)
4048 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
4054 if (flags
& BTRFS_BLOCK_GROUP_DATA
&&
4055 bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
4056 target
= BTRFS_BLOCK_GROUP_DATA
| bctl
->data
.target
;
4057 } else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
&&
4058 bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
4059 target
= BTRFS_BLOCK_GROUP_SYSTEM
| bctl
->sys
.target
;
4060 } else if (flags
& BTRFS_BLOCK_GROUP_METADATA
&&
4061 bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
4062 target
= BTRFS_BLOCK_GROUP_METADATA
| bctl
->meta
.target
;
4069 * @flags: available profiles in extended format (see ctree.h)
4071 * Returns reduced profile in chunk format. If profile changing is in
4072 * progress (either running or paused) picks the target profile (if it's
4073 * already available), otherwise falls back to plain reducing.
4075 static u64
btrfs_reduce_alloc_profile(struct btrfs_fs_info
*fs_info
, u64 flags
)
4077 u64 num_devices
= fs_info
->fs_devices
->rw_devices
;
4083 * see if restripe for this chunk_type is in progress, if so
4084 * try to reduce to the target profile
4086 spin_lock(&fs_info
->balance_lock
);
4087 target
= get_restripe_target(fs_info
, flags
);
4089 /* pick target profile only if it's already available */
4090 if ((flags
& target
) & BTRFS_EXTENDED_PROFILE_MASK
) {
4091 spin_unlock(&fs_info
->balance_lock
);
4092 return extended_to_chunk(target
);
4095 spin_unlock(&fs_info
->balance_lock
);
4097 /* First, mask out the RAID levels which aren't possible */
4098 for (raid_type
= 0; raid_type
< BTRFS_NR_RAID_TYPES
; raid_type
++) {
4099 if (num_devices
>= btrfs_raid_array
[raid_type
].devs_min
)
4100 allowed
|= btrfs_raid_group
[raid_type
];
4104 if (allowed
& BTRFS_BLOCK_GROUP_RAID6
)
4105 allowed
= BTRFS_BLOCK_GROUP_RAID6
;
4106 else if (allowed
& BTRFS_BLOCK_GROUP_RAID5
)
4107 allowed
= BTRFS_BLOCK_GROUP_RAID5
;
4108 else if (allowed
& BTRFS_BLOCK_GROUP_RAID10
)
4109 allowed
= BTRFS_BLOCK_GROUP_RAID10
;
4110 else if (allowed
& BTRFS_BLOCK_GROUP_RAID1
)
4111 allowed
= BTRFS_BLOCK_GROUP_RAID1
;
4112 else if (allowed
& BTRFS_BLOCK_GROUP_RAID0
)
4113 allowed
= BTRFS_BLOCK_GROUP_RAID0
;
4115 flags
&= ~BTRFS_BLOCK_GROUP_PROFILE_MASK
;
4117 return extended_to_chunk(flags
| allowed
);
4120 static u64
get_alloc_profile(struct btrfs_fs_info
*fs_info
, u64 orig_flags
)
4127 seq
= read_seqbegin(&fs_info
->profiles_lock
);
4129 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
4130 flags
|= fs_info
->avail_data_alloc_bits
;
4131 else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
4132 flags
|= fs_info
->avail_system_alloc_bits
;
4133 else if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
4134 flags
|= fs_info
->avail_metadata_alloc_bits
;
4135 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
4137 return btrfs_reduce_alloc_profile(fs_info
, flags
);
4140 static u64
get_alloc_profile_by_root(struct btrfs_root
*root
, int data
)
4142 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4147 flags
= BTRFS_BLOCK_GROUP_DATA
;
4148 else if (root
== fs_info
->chunk_root
)
4149 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
4151 flags
= BTRFS_BLOCK_GROUP_METADATA
;
4153 ret
= get_alloc_profile(fs_info
, flags
);
4157 u64
btrfs_data_alloc_profile(struct btrfs_fs_info
*fs_info
)
4159 return get_alloc_profile(fs_info
, BTRFS_BLOCK_GROUP_DATA
);
4162 u64
btrfs_metadata_alloc_profile(struct btrfs_fs_info
*fs_info
)
4164 return get_alloc_profile(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4167 u64
btrfs_system_alloc_profile(struct btrfs_fs_info
*fs_info
)
4169 return get_alloc_profile(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
4172 static u64
btrfs_space_info_used(struct btrfs_space_info
*s_info
,
4173 bool may_use_included
)
4176 return s_info
->bytes_used
+ s_info
->bytes_reserved
+
4177 s_info
->bytes_pinned
+ s_info
->bytes_readonly
+
4178 (may_use_included
? s_info
->bytes_may_use
: 0);
4181 int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode
*inode
, u64 bytes
)
4183 struct btrfs_space_info
*data_sinfo
;
4184 struct btrfs_root
*root
= inode
->root
;
4185 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4188 int need_commit
= 2;
4189 int have_pinned_space
;
4191 /* make sure bytes are sectorsize aligned */
4192 bytes
= ALIGN(bytes
, fs_info
->sectorsize
);
4194 if (btrfs_is_free_space_inode(inode
)) {
4196 ASSERT(current
->journal_info
);
4199 data_sinfo
= fs_info
->data_sinfo
;
4204 /* make sure we have enough space to handle the data first */
4205 spin_lock(&data_sinfo
->lock
);
4206 used
= btrfs_space_info_used(data_sinfo
, true);
4208 if (used
+ bytes
> data_sinfo
->total_bytes
) {
4209 struct btrfs_trans_handle
*trans
;
4212 * if we don't have enough free bytes in this space then we need
4213 * to alloc a new chunk.
4215 if (!data_sinfo
->full
) {
4218 data_sinfo
->force_alloc
= CHUNK_ALLOC_FORCE
;
4219 spin_unlock(&data_sinfo
->lock
);
4221 alloc_target
= btrfs_data_alloc_profile(fs_info
);
4223 * It is ugly that we don't call nolock join
4224 * transaction for the free space inode case here.
4225 * But it is safe because we only do the data space
4226 * reservation for the free space cache in the
4227 * transaction context, the common join transaction
4228 * just increase the counter of the current transaction
4229 * handler, doesn't try to acquire the trans_lock of
4232 trans
= btrfs_join_transaction(root
);
4234 return PTR_ERR(trans
);
4236 ret
= do_chunk_alloc(trans
, fs_info
, alloc_target
,
4237 CHUNK_ALLOC_NO_FORCE
);
4238 btrfs_end_transaction(trans
);
4243 have_pinned_space
= 1;
4249 data_sinfo
= fs_info
->data_sinfo
;
4255 * If we don't have enough pinned space to deal with this
4256 * allocation, and no removed chunk in current transaction,
4257 * don't bother committing the transaction.
4259 have_pinned_space
= percpu_counter_compare(
4260 &data_sinfo
->total_bytes_pinned
,
4261 used
+ bytes
- data_sinfo
->total_bytes
);
4262 spin_unlock(&data_sinfo
->lock
);
4264 /* commit the current transaction and try again */
4267 !atomic_read(&fs_info
->open_ioctl_trans
)) {
4270 if (need_commit
> 0) {
4271 btrfs_start_delalloc_roots(fs_info
, 0, -1);
4272 btrfs_wait_ordered_roots(fs_info
, -1, 0,
4276 trans
= btrfs_join_transaction(root
);
4278 return PTR_ERR(trans
);
4279 if (have_pinned_space
>= 0 ||
4280 test_bit(BTRFS_TRANS_HAVE_FREE_BGS
,
4281 &trans
->transaction
->flags
) ||
4283 ret
= btrfs_commit_transaction(trans
);
4287 * The cleaner kthread might still be doing iput
4288 * operations. Wait for it to finish so that
4289 * more space is released.
4291 mutex_lock(&fs_info
->cleaner_delayed_iput_mutex
);
4292 mutex_unlock(&fs_info
->cleaner_delayed_iput_mutex
);
4295 btrfs_end_transaction(trans
);
4299 trace_btrfs_space_reservation(fs_info
,
4300 "space_info:enospc",
4301 data_sinfo
->flags
, bytes
, 1);
4304 data_sinfo
->bytes_may_use
+= bytes
;
4305 trace_btrfs_space_reservation(fs_info
, "space_info",
4306 data_sinfo
->flags
, bytes
, 1);
4307 spin_unlock(&data_sinfo
->lock
);
4313 * New check_data_free_space() with ability for precious data reservation
4314 * Will replace old btrfs_check_data_free_space(), but for patch split,
4315 * add a new function first and then replace it.
4317 int btrfs_check_data_free_space(struct inode
*inode
, u64 start
, u64 len
)
4319 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
4322 /* align the range */
4323 len
= round_up(start
+ len
, fs_info
->sectorsize
) -
4324 round_down(start
, fs_info
->sectorsize
);
4325 start
= round_down(start
, fs_info
->sectorsize
);
4327 ret
= btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode
), len
);
4331 /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */
4332 ret
= btrfs_qgroup_reserve_data(inode
, start
, len
);
4334 btrfs_free_reserved_data_space_noquota(inode
, start
, len
);
4339 * Called if we need to clear a data reservation for this inode
4340 * Normally in a error case.
4342 * This one will *NOT* use accurate qgroup reserved space API, just for case
4343 * which we can't sleep and is sure it won't affect qgroup reserved space.
4344 * Like clear_bit_hook().
4346 void btrfs_free_reserved_data_space_noquota(struct inode
*inode
, u64 start
,
4349 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
4350 struct btrfs_space_info
*data_sinfo
;
4352 /* Make sure the range is aligned to sectorsize */
4353 len
= round_up(start
+ len
, fs_info
->sectorsize
) -
4354 round_down(start
, fs_info
->sectorsize
);
4355 start
= round_down(start
, fs_info
->sectorsize
);
4357 data_sinfo
= fs_info
->data_sinfo
;
4358 spin_lock(&data_sinfo
->lock
);
4359 if (WARN_ON(data_sinfo
->bytes_may_use
< len
))
4360 data_sinfo
->bytes_may_use
= 0;
4362 data_sinfo
->bytes_may_use
-= len
;
4363 trace_btrfs_space_reservation(fs_info
, "space_info",
4364 data_sinfo
->flags
, len
, 0);
4365 spin_unlock(&data_sinfo
->lock
);
4369 * Called if we need to clear a data reservation for this inode
4370 * Normally in a error case.
4372 * This one will handle the per-inode data rsv map for accurate reserved
4375 void btrfs_free_reserved_data_space(struct inode
*inode
, u64 start
, u64 len
)
4377 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4379 /* Make sure the range is aligned to sectorsize */
4380 len
= round_up(start
+ len
, root
->fs_info
->sectorsize
) -
4381 round_down(start
, root
->fs_info
->sectorsize
);
4382 start
= round_down(start
, root
->fs_info
->sectorsize
);
4384 btrfs_free_reserved_data_space_noquota(inode
, start
, len
);
4385 btrfs_qgroup_free_data(inode
, start
, len
);
4388 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
4390 struct list_head
*head
= &info
->space_info
;
4391 struct btrfs_space_info
*found
;
4394 list_for_each_entry_rcu(found
, head
, list
) {
4395 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
4396 found
->force_alloc
= CHUNK_ALLOC_FORCE
;
4401 static inline u64
calc_global_rsv_need_space(struct btrfs_block_rsv
*global
)
4403 return (global
->size
<< 1);
4406 static int should_alloc_chunk(struct btrfs_fs_info
*fs_info
,
4407 struct btrfs_space_info
*sinfo
, int force
)
4409 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
4410 u64 num_bytes
= sinfo
->total_bytes
- sinfo
->bytes_readonly
;
4411 u64 num_allocated
= sinfo
->bytes_used
+ sinfo
->bytes_reserved
;
4414 if (force
== CHUNK_ALLOC_FORCE
)
4418 * We need to take into account the global rsv because for all intents
4419 * and purposes it's used space. Don't worry about locking the
4420 * global_rsv, it doesn't change except when the transaction commits.
4422 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
4423 num_allocated
+= calc_global_rsv_need_space(global_rsv
);
4426 * in limited mode, we want to have some free space up to
4427 * about 1% of the FS size.
4429 if (force
== CHUNK_ALLOC_LIMITED
) {
4430 thresh
= btrfs_super_total_bytes(fs_info
->super_copy
);
4431 thresh
= max_t(u64
, SZ_64M
, div_factor_fine(thresh
, 1));
4433 if (num_bytes
- num_allocated
< thresh
)
4437 if (num_allocated
+ SZ_2M
< div_factor(num_bytes
, 8))
4442 static u64
get_profile_num_devs(struct btrfs_fs_info
*fs_info
, u64 type
)
4446 if (type
& (BTRFS_BLOCK_GROUP_RAID10
|
4447 BTRFS_BLOCK_GROUP_RAID0
|
4448 BTRFS_BLOCK_GROUP_RAID5
|
4449 BTRFS_BLOCK_GROUP_RAID6
))
4450 num_dev
= fs_info
->fs_devices
->rw_devices
;
4451 else if (type
& BTRFS_BLOCK_GROUP_RAID1
)
4454 num_dev
= 1; /* DUP or single */
4460 * If @is_allocation is true, reserve space in the system space info necessary
4461 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4464 void check_system_chunk(struct btrfs_trans_handle
*trans
,
4465 struct btrfs_fs_info
*fs_info
, u64 type
)
4467 struct btrfs_space_info
*info
;
4474 * Needed because we can end up allocating a system chunk and for an
4475 * atomic and race free space reservation in the chunk block reserve.
4477 ASSERT(mutex_is_locked(&fs_info
->chunk_mutex
));
4479 info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
4480 spin_lock(&info
->lock
);
4481 left
= info
->total_bytes
- btrfs_space_info_used(info
, true);
4482 spin_unlock(&info
->lock
);
4484 num_devs
= get_profile_num_devs(fs_info
, type
);
4486 /* num_devs device items to update and 1 chunk item to add or remove */
4487 thresh
= btrfs_calc_trunc_metadata_size(fs_info
, num_devs
) +
4488 btrfs_calc_trans_metadata_size(fs_info
, 1);
4490 if (left
< thresh
&& btrfs_test_opt(fs_info
, ENOSPC_DEBUG
)) {
4491 btrfs_info(fs_info
, "left=%llu, need=%llu, flags=%llu",
4492 left
, thresh
, type
);
4493 dump_space_info(fs_info
, info
, 0, 0);
4496 if (left
< thresh
) {
4497 u64 flags
= btrfs_system_alloc_profile(fs_info
);
4500 * Ignore failure to create system chunk. We might end up not
4501 * needing it, as we might not need to COW all nodes/leafs from
4502 * the paths we visit in the chunk tree (they were already COWed
4503 * or created in the current transaction for example).
4505 ret
= btrfs_alloc_chunk(trans
, fs_info
, flags
);
4509 ret
= btrfs_block_rsv_add(fs_info
->chunk_root
,
4510 &fs_info
->chunk_block_rsv
,
4511 thresh
, BTRFS_RESERVE_NO_FLUSH
);
4513 trans
->chunk_bytes_reserved
+= thresh
;
4518 * If force is CHUNK_ALLOC_FORCE:
4519 * - return 1 if it successfully allocates a chunk,
4520 * - return errors including -ENOSPC otherwise.
4521 * If force is NOT CHUNK_ALLOC_FORCE:
4522 * - return 0 if it doesn't need to allocate a new chunk,
4523 * - return 1 if it successfully allocates a chunk,
4524 * - return errors including -ENOSPC otherwise.
4526 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
4527 struct btrfs_fs_info
*fs_info
, u64 flags
, int force
)
4529 struct btrfs_space_info
*space_info
;
4530 int wait_for_alloc
= 0;
4533 /* Don't re-enter if we're already allocating a chunk */
4534 if (trans
->allocating_chunk
)
4537 space_info
= __find_space_info(fs_info
, flags
);
4539 ret
= create_space_info(fs_info
, flags
, &space_info
);
4545 spin_lock(&space_info
->lock
);
4546 if (force
< space_info
->force_alloc
)
4547 force
= space_info
->force_alloc
;
4548 if (space_info
->full
) {
4549 if (should_alloc_chunk(fs_info
, space_info
, force
))
4553 spin_unlock(&space_info
->lock
);
4557 if (!should_alloc_chunk(fs_info
, space_info
, force
)) {
4558 spin_unlock(&space_info
->lock
);
4560 } else if (space_info
->chunk_alloc
) {
4563 space_info
->chunk_alloc
= 1;
4566 spin_unlock(&space_info
->lock
);
4568 mutex_lock(&fs_info
->chunk_mutex
);
4571 * The chunk_mutex is held throughout the entirety of a chunk
4572 * allocation, so once we've acquired the chunk_mutex we know that the
4573 * other guy is done and we need to recheck and see if we should
4576 if (wait_for_alloc
) {
4577 mutex_unlock(&fs_info
->chunk_mutex
);
4582 trans
->allocating_chunk
= true;
4585 * If we have mixed data/metadata chunks we want to make sure we keep
4586 * allocating mixed chunks instead of individual chunks.
4588 if (btrfs_mixed_space_info(space_info
))
4589 flags
|= (BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
);
4592 * if we're doing a data chunk, go ahead and make sure that
4593 * we keep a reasonable number of metadata chunks allocated in the
4596 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
4597 fs_info
->data_chunk_allocations
++;
4598 if (!(fs_info
->data_chunk_allocations
%
4599 fs_info
->metadata_ratio
))
4600 force_metadata_allocation(fs_info
);
4604 * Check if we have enough space in SYSTEM chunk because we may need
4605 * to update devices.
4607 check_system_chunk(trans
, fs_info
, flags
);
4609 ret
= btrfs_alloc_chunk(trans
, fs_info
, flags
);
4610 trans
->allocating_chunk
= false;
4612 spin_lock(&space_info
->lock
);
4613 if (ret
< 0 && ret
!= -ENOSPC
)
4616 space_info
->full
= 1;
4620 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
4622 space_info
->chunk_alloc
= 0;
4623 spin_unlock(&space_info
->lock
);
4624 mutex_unlock(&fs_info
->chunk_mutex
);
4626 * When we allocate a new chunk we reserve space in the chunk block
4627 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4628 * add new nodes/leafs to it if we end up needing to do it when
4629 * inserting the chunk item and updating device items as part of the
4630 * second phase of chunk allocation, performed by
4631 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4632 * large number of new block groups to create in our transaction
4633 * handle's new_bgs list to avoid exhausting the chunk block reserve
4634 * in extreme cases - like having a single transaction create many new
4635 * block groups when starting to write out the free space caches of all
4636 * the block groups that were made dirty during the lifetime of the
4639 if (trans
->can_flush_pending_bgs
&&
4640 trans
->chunk_bytes_reserved
>= (u64
)SZ_2M
) {
4641 btrfs_create_pending_block_groups(trans
, fs_info
);
4642 btrfs_trans_release_chunk_metadata(trans
);
4647 static int can_overcommit(struct btrfs_fs_info
*fs_info
,
4648 struct btrfs_space_info
*space_info
, u64 bytes
,
4649 enum btrfs_reserve_flush_enum flush
,
4652 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
4658 /* Don't overcommit when in mixed mode. */
4659 if (space_info
->flags
& BTRFS_BLOCK_GROUP_DATA
)
4663 profile
= btrfs_system_alloc_profile(fs_info
);
4665 profile
= btrfs_metadata_alloc_profile(fs_info
);
4667 used
= btrfs_space_info_used(space_info
, false);
4670 * We only want to allow over committing if we have lots of actual space
4671 * free, but if we don't have enough space to handle the global reserve
4672 * space then we could end up having a real enospc problem when trying
4673 * to allocate a chunk or some other such important allocation.
4675 spin_lock(&global_rsv
->lock
);
4676 space_size
= calc_global_rsv_need_space(global_rsv
);
4677 spin_unlock(&global_rsv
->lock
);
4678 if (used
+ space_size
>= space_info
->total_bytes
)
4681 used
+= space_info
->bytes_may_use
;
4683 avail
= atomic64_read(&fs_info
->free_chunk_space
);
4686 * If we have dup, raid1 or raid10 then only half of the free
4687 * space is actually useable. For raid56, the space info used
4688 * doesn't include the parity drive, so we don't have to
4691 if (profile
& (BTRFS_BLOCK_GROUP_DUP
|
4692 BTRFS_BLOCK_GROUP_RAID1
|
4693 BTRFS_BLOCK_GROUP_RAID10
))
4697 * If we aren't flushing all things, let us overcommit up to
4698 * 1/2th of the space. If we can flush, don't let us overcommit
4699 * too much, let it overcommit up to 1/8 of the space.
4701 if (flush
== BTRFS_RESERVE_FLUSH_ALL
)
4706 if (used
+ bytes
< space_info
->total_bytes
+ avail
)
4711 static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info
*fs_info
,
4712 unsigned long nr_pages
, int nr_items
)
4714 struct super_block
*sb
= fs_info
->sb
;
4716 if (down_read_trylock(&sb
->s_umount
)) {
4717 writeback_inodes_sb_nr(sb
, nr_pages
, WB_REASON_FS_FREE_SPACE
);
4718 up_read(&sb
->s_umount
);
4721 * We needn't worry the filesystem going from r/w to r/o though
4722 * we don't acquire ->s_umount mutex, because the filesystem
4723 * should guarantee the delalloc inodes list be empty after
4724 * the filesystem is readonly(all dirty pages are written to
4727 btrfs_start_delalloc_roots(fs_info
, 0, nr_items
);
4728 if (!current
->journal_info
)
4729 btrfs_wait_ordered_roots(fs_info
, nr_items
, 0, (u64
)-1);
4733 static inline int calc_reclaim_items_nr(struct btrfs_fs_info
*fs_info
,
4739 bytes
= btrfs_calc_trans_metadata_size(fs_info
, 1);
4740 nr
= (int)div64_u64(to_reclaim
, bytes
);
4746 #define EXTENT_SIZE_PER_ITEM SZ_256K
4749 * shrink metadata reservation for delalloc
4751 static void shrink_delalloc(struct btrfs_fs_info
*fs_info
, u64 to_reclaim
,
4752 u64 orig
, bool wait_ordered
)
4754 struct btrfs_block_rsv
*block_rsv
;
4755 struct btrfs_space_info
*space_info
;
4756 struct btrfs_trans_handle
*trans
;
4760 unsigned long nr_pages
;
4763 enum btrfs_reserve_flush_enum flush
;
4765 /* Calc the number of the pages we need flush for space reservation */
4766 items
= calc_reclaim_items_nr(fs_info
, to_reclaim
);
4767 to_reclaim
= (u64
)items
* EXTENT_SIZE_PER_ITEM
;
4769 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4770 block_rsv
= &fs_info
->delalloc_block_rsv
;
4771 space_info
= block_rsv
->space_info
;
4773 delalloc_bytes
= percpu_counter_sum_positive(
4774 &fs_info
->delalloc_bytes
);
4775 if (delalloc_bytes
== 0) {
4779 btrfs_wait_ordered_roots(fs_info
, items
, 0, (u64
)-1);
4784 while (delalloc_bytes
&& loops
< 3) {
4785 max_reclaim
= min(delalloc_bytes
, to_reclaim
);
4786 nr_pages
= max_reclaim
>> PAGE_SHIFT
;
4787 btrfs_writeback_inodes_sb_nr(fs_info
, nr_pages
, items
);
4789 * We need to wait for the async pages to actually start before
4792 max_reclaim
= atomic_read(&fs_info
->async_delalloc_pages
);
4796 if (max_reclaim
<= nr_pages
)
4799 max_reclaim
-= nr_pages
;
4801 wait_event(fs_info
->async_submit_wait
,
4802 atomic_read(&fs_info
->async_delalloc_pages
) <=
4806 flush
= BTRFS_RESERVE_FLUSH_ALL
;
4808 flush
= BTRFS_RESERVE_NO_FLUSH
;
4809 spin_lock(&space_info
->lock
);
4810 if (can_overcommit(fs_info
, space_info
, orig
, flush
, false)) {
4811 spin_unlock(&space_info
->lock
);
4814 if (list_empty(&space_info
->tickets
) &&
4815 list_empty(&space_info
->priority_tickets
)) {
4816 spin_unlock(&space_info
->lock
);
4819 spin_unlock(&space_info
->lock
);
4822 if (wait_ordered
&& !trans
) {
4823 btrfs_wait_ordered_roots(fs_info
, items
, 0, (u64
)-1);
4825 time_left
= schedule_timeout_killable(1);
4829 delalloc_bytes
= percpu_counter_sum_positive(
4830 &fs_info
->delalloc_bytes
);
4835 * maybe_commit_transaction - possibly commit the transaction if its ok to
4836 * @root - the root we're allocating for
4837 * @bytes - the number of bytes we want to reserve
4838 * @force - force the commit
4840 * This will check to make sure that committing the transaction will actually
4841 * get us somewhere and then commit the transaction if it does. Otherwise it
4842 * will return -ENOSPC.
4844 static int may_commit_transaction(struct btrfs_fs_info
*fs_info
,
4845 struct btrfs_space_info
*space_info
,
4846 u64 bytes
, int force
)
4848 struct btrfs_block_rsv
*delayed_rsv
= &fs_info
->delayed_block_rsv
;
4849 struct btrfs_trans_handle
*trans
;
4851 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4858 /* See if there is enough pinned space to make this reservation */
4859 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4864 * See if there is some space in the delayed insertion reservation for
4867 if (space_info
!= delayed_rsv
->space_info
)
4870 spin_lock(&delayed_rsv
->lock
);
4871 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4872 bytes
- delayed_rsv
->size
) < 0) {
4873 spin_unlock(&delayed_rsv
->lock
);
4876 spin_unlock(&delayed_rsv
->lock
);
4879 trans
= btrfs_join_transaction(fs_info
->extent_root
);
4883 return btrfs_commit_transaction(trans
);
4886 struct reserve_ticket
{
4889 struct list_head list
;
4890 wait_queue_head_t wait
;
4893 static int flush_space(struct btrfs_fs_info
*fs_info
,
4894 struct btrfs_space_info
*space_info
, u64 num_bytes
,
4895 u64 orig_bytes
, int state
)
4897 struct btrfs_root
*root
= fs_info
->extent_root
;
4898 struct btrfs_trans_handle
*trans
;
4903 case FLUSH_DELAYED_ITEMS_NR
:
4904 case FLUSH_DELAYED_ITEMS
:
4905 if (state
== FLUSH_DELAYED_ITEMS_NR
)
4906 nr
= calc_reclaim_items_nr(fs_info
, num_bytes
) * 2;
4910 trans
= btrfs_join_transaction(root
);
4911 if (IS_ERR(trans
)) {
4912 ret
= PTR_ERR(trans
);
4915 ret
= btrfs_run_delayed_items_nr(trans
, fs_info
, nr
);
4916 btrfs_end_transaction(trans
);
4918 case FLUSH_DELALLOC
:
4919 case FLUSH_DELALLOC_WAIT
:
4920 shrink_delalloc(fs_info
, num_bytes
* 2, orig_bytes
,
4921 state
== FLUSH_DELALLOC_WAIT
);
4924 trans
= btrfs_join_transaction(root
);
4925 if (IS_ERR(trans
)) {
4926 ret
= PTR_ERR(trans
);
4929 ret
= do_chunk_alloc(trans
, fs_info
,
4930 btrfs_metadata_alloc_profile(fs_info
),
4931 CHUNK_ALLOC_NO_FORCE
);
4932 btrfs_end_transaction(trans
);
4933 if (ret
> 0 || ret
== -ENOSPC
)
4937 ret
= may_commit_transaction(fs_info
, space_info
,
4945 trace_btrfs_flush_space(fs_info
, space_info
->flags
, num_bytes
,
4946 orig_bytes
, state
, ret
);
4951 btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info
*fs_info
,
4952 struct btrfs_space_info
*space_info
,
4955 struct reserve_ticket
*ticket
;
4960 list_for_each_entry(ticket
, &space_info
->tickets
, list
)
4961 to_reclaim
+= ticket
->bytes
;
4962 list_for_each_entry(ticket
, &space_info
->priority_tickets
, list
)
4963 to_reclaim
+= ticket
->bytes
;
4967 to_reclaim
= min_t(u64
, num_online_cpus() * SZ_1M
, SZ_16M
);
4968 if (can_overcommit(fs_info
, space_info
, to_reclaim
,
4969 BTRFS_RESERVE_FLUSH_ALL
, system_chunk
))
4972 used
= btrfs_space_info_used(space_info
, true);
4974 if (can_overcommit(fs_info
, space_info
, SZ_1M
,
4975 BTRFS_RESERVE_FLUSH_ALL
, system_chunk
))
4976 expected
= div_factor_fine(space_info
->total_bytes
, 95);
4978 expected
= div_factor_fine(space_info
->total_bytes
, 90);
4980 if (used
> expected
)
4981 to_reclaim
= used
- expected
;
4984 to_reclaim
= min(to_reclaim
, space_info
->bytes_may_use
+
4985 space_info
->bytes_reserved
);
4989 static inline int need_do_async_reclaim(struct btrfs_fs_info
*fs_info
,
4990 struct btrfs_space_info
*space_info
,
4991 u64 used
, bool system_chunk
)
4993 u64 thresh
= div_factor_fine(space_info
->total_bytes
, 98);
4995 /* If we're just plain full then async reclaim just slows us down. */
4996 if ((space_info
->bytes_used
+ space_info
->bytes_reserved
) >= thresh
)
4999 if (!btrfs_calc_reclaim_metadata_size(fs_info
, space_info
,
5003 return (used
>= thresh
&& !btrfs_fs_closing(fs_info
) &&
5004 !test_bit(BTRFS_FS_STATE_REMOUNTING
, &fs_info
->fs_state
));
5007 static void wake_all_tickets(struct list_head
*head
)
5009 struct reserve_ticket
*ticket
;
5011 while (!list_empty(head
)) {
5012 ticket
= list_first_entry(head
, struct reserve_ticket
, list
);
5013 list_del_init(&ticket
->list
);
5014 ticket
->error
= -ENOSPC
;
5015 wake_up(&ticket
->wait
);
5020 * This is for normal flushers, we can wait all goddamned day if we want to. We
5021 * will loop and continuously try to flush as long as we are making progress.
5022 * We count progress as clearing off tickets each time we have to loop.
5024 static void btrfs_async_reclaim_metadata_space(struct work_struct
*work
)
5026 struct btrfs_fs_info
*fs_info
;
5027 struct btrfs_space_info
*space_info
;
5030 int commit_cycles
= 0;
5031 u64 last_tickets_id
;
5033 fs_info
= container_of(work
, struct btrfs_fs_info
, async_reclaim_work
);
5034 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
5036 spin_lock(&space_info
->lock
);
5037 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
, space_info
,
5040 space_info
->flush
= 0;
5041 spin_unlock(&space_info
->lock
);
5044 last_tickets_id
= space_info
->tickets_id
;
5045 spin_unlock(&space_info
->lock
);
5047 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
5049 struct reserve_ticket
*ticket
;
5052 ret
= flush_space(fs_info
, space_info
, to_reclaim
, to_reclaim
,
5054 spin_lock(&space_info
->lock
);
5055 if (list_empty(&space_info
->tickets
)) {
5056 space_info
->flush
= 0;
5057 spin_unlock(&space_info
->lock
);
5060 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
,
5063 ticket
= list_first_entry(&space_info
->tickets
,
5064 struct reserve_ticket
, list
);
5065 if (last_tickets_id
== space_info
->tickets_id
) {
5068 last_tickets_id
= space_info
->tickets_id
;
5069 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
5074 if (flush_state
> COMMIT_TRANS
) {
5076 if (commit_cycles
> 2) {
5077 wake_all_tickets(&space_info
->tickets
);
5078 space_info
->flush
= 0;
5080 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
5083 spin_unlock(&space_info
->lock
);
5084 } while (flush_state
<= COMMIT_TRANS
);
5087 void btrfs_init_async_reclaim_work(struct work_struct
*work
)
5089 INIT_WORK(work
, btrfs_async_reclaim_metadata_space
);
5092 static void priority_reclaim_metadata_space(struct btrfs_fs_info
*fs_info
,
5093 struct btrfs_space_info
*space_info
,
5094 struct reserve_ticket
*ticket
)
5097 int flush_state
= FLUSH_DELAYED_ITEMS_NR
;
5099 spin_lock(&space_info
->lock
);
5100 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
, space_info
,
5103 spin_unlock(&space_info
->lock
);
5106 spin_unlock(&space_info
->lock
);
5109 flush_space(fs_info
, space_info
, to_reclaim
, to_reclaim
,
5112 spin_lock(&space_info
->lock
);
5113 if (ticket
->bytes
== 0) {
5114 spin_unlock(&space_info
->lock
);
5117 spin_unlock(&space_info
->lock
);
5120 * Priority flushers can't wait on delalloc without
5123 if (flush_state
== FLUSH_DELALLOC
||
5124 flush_state
== FLUSH_DELALLOC_WAIT
)
5125 flush_state
= ALLOC_CHUNK
;
5126 } while (flush_state
< COMMIT_TRANS
);
5129 static int wait_reserve_ticket(struct btrfs_fs_info
*fs_info
,
5130 struct btrfs_space_info
*space_info
,
5131 struct reserve_ticket
*ticket
, u64 orig_bytes
)
5137 spin_lock(&space_info
->lock
);
5138 while (ticket
->bytes
> 0 && ticket
->error
== 0) {
5139 ret
= prepare_to_wait_event(&ticket
->wait
, &wait
, TASK_KILLABLE
);
5144 spin_unlock(&space_info
->lock
);
5148 finish_wait(&ticket
->wait
, &wait
);
5149 spin_lock(&space_info
->lock
);
5152 ret
= ticket
->error
;
5153 if (!list_empty(&ticket
->list
))
5154 list_del_init(&ticket
->list
);
5155 if (ticket
->bytes
&& ticket
->bytes
< orig_bytes
) {
5156 u64 num_bytes
= orig_bytes
- ticket
->bytes
;
5157 space_info
->bytes_may_use
-= num_bytes
;
5158 trace_btrfs_space_reservation(fs_info
, "space_info",
5159 space_info
->flags
, num_bytes
, 0);
5161 spin_unlock(&space_info
->lock
);
5167 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5168 * @root - the root we're allocating for
5169 * @space_info - the space info we want to allocate from
5170 * @orig_bytes - the number of bytes we want
5171 * @flush - whether or not we can flush to make our reservation
5173 * This will reserve orig_bytes number of bytes from the space info associated
5174 * with the block_rsv. If there is not enough space it will make an attempt to
5175 * flush out space to make room. It will do this by flushing delalloc if
5176 * possible or committing the transaction. If flush is 0 then no attempts to
5177 * regain reservations will be made and this will fail if there is not enough
5180 static int __reserve_metadata_bytes(struct btrfs_fs_info
*fs_info
,
5181 struct btrfs_space_info
*space_info
,
5183 enum btrfs_reserve_flush_enum flush
,
5186 struct reserve_ticket ticket
;
5191 ASSERT(!current
->journal_info
|| flush
!= BTRFS_RESERVE_FLUSH_ALL
);
5193 spin_lock(&space_info
->lock
);
5195 used
= btrfs_space_info_used(space_info
, true);
5198 * If we have enough space then hooray, make our reservation and carry
5199 * on. If not see if we can overcommit, and if we can, hooray carry on.
5200 * If not things get more complicated.
5202 if (used
+ orig_bytes
<= space_info
->total_bytes
) {
5203 space_info
->bytes_may_use
+= orig_bytes
;
5204 trace_btrfs_space_reservation(fs_info
, "space_info",
5205 space_info
->flags
, orig_bytes
, 1);
5207 } else if (can_overcommit(fs_info
, space_info
, orig_bytes
, flush
,
5209 space_info
->bytes_may_use
+= orig_bytes
;
5210 trace_btrfs_space_reservation(fs_info
, "space_info",
5211 space_info
->flags
, orig_bytes
, 1);
5216 * If we couldn't make a reservation then setup our reservation ticket
5217 * and kick the async worker if it's not already running.
5219 * If we are a priority flusher then we just need to add our ticket to
5220 * the list and we will do our own flushing further down.
5222 if (ret
&& flush
!= BTRFS_RESERVE_NO_FLUSH
) {
5223 ticket
.bytes
= orig_bytes
;
5225 init_waitqueue_head(&ticket
.wait
);
5226 if (flush
== BTRFS_RESERVE_FLUSH_ALL
) {
5227 list_add_tail(&ticket
.list
, &space_info
->tickets
);
5228 if (!space_info
->flush
) {
5229 space_info
->flush
= 1;
5230 trace_btrfs_trigger_flush(fs_info
,
5234 queue_work(system_unbound_wq
,
5235 &fs_info
->async_reclaim_work
);
5238 list_add_tail(&ticket
.list
,
5239 &space_info
->priority_tickets
);
5241 } else if (!ret
&& space_info
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
5244 * We will do the space reservation dance during log replay,
5245 * which means we won't have fs_info->fs_root set, so don't do
5246 * the async reclaim as we will panic.
5248 if (!test_bit(BTRFS_FS_LOG_RECOVERING
, &fs_info
->flags
) &&
5249 need_do_async_reclaim(fs_info
, space_info
,
5250 used
, system_chunk
) &&
5251 !work_busy(&fs_info
->async_reclaim_work
)) {
5252 trace_btrfs_trigger_flush(fs_info
, space_info
->flags
,
5253 orig_bytes
, flush
, "preempt");
5254 queue_work(system_unbound_wq
,
5255 &fs_info
->async_reclaim_work
);
5258 spin_unlock(&space_info
->lock
);
5259 if (!ret
|| flush
== BTRFS_RESERVE_NO_FLUSH
)
5262 if (flush
== BTRFS_RESERVE_FLUSH_ALL
)
5263 return wait_reserve_ticket(fs_info
, space_info
, &ticket
,
5267 priority_reclaim_metadata_space(fs_info
, space_info
, &ticket
);
5268 spin_lock(&space_info
->lock
);
5270 if (ticket
.bytes
< orig_bytes
) {
5271 u64 num_bytes
= orig_bytes
- ticket
.bytes
;
5272 space_info
->bytes_may_use
-= num_bytes
;
5273 trace_btrfs_space_reservation(fs_info
, "space_info",
5278 list_del_init(&ticket
.list
);
5281 spin_unlock(&space_info
->lock
);
5282 ASSERT(list_empty(&ticket
.list
));
5287 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5288 * @root - the root we're allocating for
5289 * @block_rsv - the block_rsv we're allocating for
5290 * @orig_bytes - the number of bytes we want
5291 * @flush - whether or not we can flush to make our reservation
5293 * This will reserve orgi_bytes number of bytes from the space info associated
5294 * with the block_rsv. If there is not enough space it will make an attempt to
5295 * flush out space to make room. It will do this by flushing delalloc if
5296 * possible or committing the transaction. If flush is 0 then no attempts to
5297 * regain reservations will be made and this will fail if there is not enough
5300 static int reserve_metadata_bytes(struct btrfs_root
*root
,
5301 struct btrfs_block_rsv
*block_rsv
,
5303 enum btrfs_reserve_flush_enum flush
)
5305 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5306 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5308 bool system_chunk
= (root
== fs_info
->chunk_root
);
5310 ret
= __reserve_metadata_bytes(fs_info
, block_rsv
->space_info
,
5311 orig_bytes
, flush
, system_chunk
);
5312 if (ret
== -ENOSPC
&&
5313 unlikely(root
->orphan_cleanup_state
== ORPHAN_CLEANUP_STARTED
)) {
5314 if (block_rsv
!= global_rsv
&&
5315 !block_rsv_use_bytes(global_rsv
, orig_bytes
))
5319 trace_btrfs_space_reservation(fs_info
, "space_info:enospc",
5320 block_rsv
->space_info
->flags
,
5325 static struct btrfs_block_rsv
*get_block_rsv(
5326 const struct btrfs_trans_handle
*trans
,
5327 const struct btrfs_root
*root
)
5329 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5330 struct btrfs_block_rsv
*block_rsv
= NULL
;
5332 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
5333 (root
== fs_info
->csum_root
&& trans
->adding_csums
) ||
5334 (root
== fs_info
->uuid_root
))
5335 block_rsv
= trans
->block_rsv
;
5338 block_rsv
= root
->block_rsv
;
5341 block_rsv
= &fs_info
->empty_block_rsv
;
5346 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
5350 spin_lock(&block_rsv
->lock
);
5351 if (block_rsv
->reserved
>= num_bytes
) {
5352 block_rsv
->reserved
-= num_bytes
;
5353 if (block_rsv
->reserved
< block_rsv
->size
)
5354 block_rsv
->full
= 0;
5357 spin_unlock(&block_rsv
->lock
);
5361 static void block_rsv_add_bytes(struct btrfs_block_rsv
*block_rsv
,
5362 u64 num_bytes
, int update_size
)
5364 spin_lock(&block_rsv
->lock
);
5365 block_rsv
->reserved
+= num_bytes
;
5367 block_rsv
->size
+= num_bytes
;
5368 else if (block_rsv
->reserved
>= block_rsv
->size
)
5369 block_rsv
->full
= 1;
5370 spin_unlock(&block_rsv
->lock
);
5373 int btrfs_cond_migrate_bytes(struct btrfs_fs_info
*fs_info
,
5374 struct btrfs_block_rsv
*dest
, u64 num_bytes
,
5377 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5380 if (global_rsv
->space_info
!= dest
->space_info
)
5383 spin_lock(&global_rsv
->lock
);
5384 min_bytes
= div_factor(global_rsv
->size
, min_factor
);
5385 if (global_rsv
->reserved
< min_bytes
+ num_bytes
) {
5386 spin_unlock(&global_rsv
->lock
);
5389 global_rsv
->reserved
-= num_bytes
;
5390 if (global_rsv
->reserved
< global_rsv
->size
)
5391 global_rsv
->full
= 0;
5392 spin_unlock(&global_rsv
->lock
);
5394 block_rsv_add_bytes(dest
, num_bytes
, 1);
5399 * This is for space we already have accounted in space_info->bytes_may_use, so
5400 * basically when we're returning space from block_rsv's.
5402 static void space_info_add_old_bytes(struct btrfs_fs_info
*fs_info
,
5403 struct btrfs_space_info
*space_info
,
5406 struct reserve_ticket
*ticket
;
5407 struct list_head
*head
;
5409 enum btrfs_reserve_flush_enum flush
= BTRFS_RESERVE_NO_FLUSH
;
5410 bool check_overcommit
= false;
5412 spin_lock(&space_info
->lock
);
5413 head
= &space_info
->priority_tickets
;
5416 * If we are over our limit then we need to check and see if we can
5417 * overcommit, and if we can't then we just need to free up our space
5418 * and not satisfy any requests.
5420 used
= btrfs_space_info_used(space_info
, true);
5421 if (used
- num_bytes
>= space_info
->total_bytes
)
5422 check_overcommit
= true;
5424 while (!list_empty(head
) && num_bytes
) {
5425 ticket
= list_first_entry(head
, struct reserve_ticket
,
5428 * We use 0 bytes because this space is already reserved, so
5429 * adding the ticket space would be a double count.
5431 if (check_overcommit
&&
5432 !can_overcommit(fs_info
, space_info
, 0, flush
, false))
5434 if (num_bytes
>= ticket
->bytes
) {
5435 list_del_init(&ticket
->list
);
5436 num_bytes
-= ticket
->bytes
;
5438 space_info
->tickets_id
++;
5439 wake_up(&ticket
->wait
);
5441 ticket
->bytes
-= num_bytes
;
5446 if (num_bytes
&& head
== &space_info
->priority_tickets
) {
5447 head
= &space_info
->tickets
;
5448 flush
= BTRFS_RESERVE_FLUSH_ALL
;
5451 space_info
->bytes_may_use
-= num_bytes
;
5452 trace_btrfs_space_reservation(fs_info
, "space_info",
5453 space_info
->flags
, num_bytes
, 0);
5454 spin_unlock(&space_info
->lock
);
5458 * This is for newly allocated space that isn't accounted in
5459 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5460 * we use this helper.
5462 static void space_info_add_new_bytes(struct btrfs_fs_info
*fs_info
,
5463 struct btrfs_space_info
*space_info
,
5466 struct reserve_ticket
*ticket
;
5467 struct list_head
*head
= &space_info
->priority_tickets
;
5470 while (!list_empty(head
) && num_bytes
) {
5471 ticket
= list_first_entry(head
, struct reserve_ticket
,
5473 if (num_bytes
>= ticket
->bytes
) {
5474 trace_btrfs_space_reservation(fs_info
, "space_info",
5477 list_del_init(&ticket
->list
);
5478 num_bytes
-= ticket
->bytes
;
5479 space_info
->bytes_may_use
+= ticket
->bytes
;
5481 space_info
->tickets_id
++;
5482 wake_up(&ticket
->wait
);
5484 trace_btrfs_space_reservation(fs_info
, "space_info",
5487 space_info
->bytes_may_use
+= num_bytes
;
5488 ticket
->bytes
-= num_bytes
;
5493 if (num_bytes
&& head
== &space_info
->priority_tickets
) {
5494 head
= &space_info
->tickets
;
5499 static void block_rsv_release_bytes(struct btrfs_fs_info
*fs_info
,
5500 struct btrfs_block_rsv
*block_rsv
,
5501 struct btrfs_block_rsv
*dest
, u64 num_bytes
)
5503 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
5505 spin_lock(&block_rsv
->lock
);
5506 if (num_bytes
== (u64
)-1)
5507 num_bytes
= block_rsv
->size
;
5508 block_rsv
->size
-= num_bytes
;
5509 if (block_rsv
->reserved
>= block_rsv
->size
) {
5510 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
5511 block_rsv
->reserved
= block_rsv
->size
;
5512 block_rsv
->full
= 1;
5516 spin_unlock(&block_rsv
->lock
);
5518 if (num_bytes
> 0) {
5520 spin_lock(&dest
->lock
);
5524 bytes_to_add
= dest
->size
- dest
->reserved
;
5525 bytes_to_add
= min(num_bytes
, bytes_to_add
);
5526 dest
->reserved
+= bytes_to_add
;
5527 if (dest
->reserved
>= dest
->size
)
5529 num_bytes
-= bytes_to_add
;
5531 spin_unlock(&dest
->lock
);
5534 space_info_add_old_bytes(fs_info
, space_info
,
5539 int btrfs_block_rsv_migrate(struct btrfs_block_rsv
*src
,
5540 struct btrfs_block_rsv
*dst
, u64 num_bytes
,
5545 ret
= block_rsv_use_bytes(src
, num_bytes
);
5549 block_rsv_add_bytes(dst
, num_bytes
, update_size
);
5553 void btrfs_init_block_rsv(struct btrfs_block_rsv
*rsv
, unsigned short type
)
5555 memset(rsv
, 0, sizeof(*rsv
));
5556 spin_lock_init(&rsv
->lock
);
5560 struct btrfs_block_rsv
*btrfs_alloc_block_rsv(struct btrfs_fs_info
*fs_info
,
5561 unsigned short type
)
5563 struct btrfs_block_rsv
*block_rsv
;
5565 block_rsv
= kmalloc(sizeof(*block_rsv
), GFP_NOFS
);
5569 btrfs_init_block_rsv(block_rsv
, type
);
5570 block_rsv
->space_info
= __find_space_info(fs_info
,
5571 BTRFS_BLOCK_GROUP_METADATA
);
5575 void btrfs_free_block_rsv(struct btrfs_fs_info
*fs_info
,
5576 struct btrfs_block_rsv
*rsv
)
5580 btrfs_block_rsv_release(fs_info
, rsv
, (u64
)-1);
5584 void __btrfs_free_block_rsv(struct btrfs_block_rsv
*rsv
)
5589 int btrfs_block_rsv_add(struct btrfs_root
*root
,
5590 struct btrfs_block_rsv
*block_rsv
, u64 num_bytes
,
5591 enum btrfs_reserve_flush_enum flush
)
5598 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
5600 block_rsv_add_bytes(block_rsv
, num_bytes
, 1);
5607 int btrfs_block_rsv_check(struct btrfs_block_rsv
*block_rsv
, int min_factor
)
5615 spin_lock(&block_rsv
->lock
);
5616 num_bytes
= div_factor(block_rsv
->size
, min_factor
);
5617 if (block_rsv
->reserved
>= num_bytes
)
5619 spin_unlock(&block_rsv
->lock
);
5624 int btrfs_block_rsv_refill(struct btrfs_root
*root
,
5625 struct btrfs_block_rsv
*block_rsv
, u64 min_reserved
,
5626 enum btrfs_reserve_flush_enum flush
)
5634 spin_lock(&block_rsv
->lock
);
5635 num_bytes
= min_reserved
;
5636 if (block_rsv
->reserved
>= num_bytes
)
5639 num_bytes
-= block_rsv
->reserved
;
5640 spin_unlock(&block_rsv
->lock
);
5645 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
5647 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
5654 void btrfs_block_rsv_release(struct btrfs_fs_info
*fs_info
,
5655 struct btrfs_block_rsv
*block_rsv
,
5658 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5660 if (global_rsv
== block_rsv
||
5661 block_rsv
->space_info
!= global_rsv
->space_info
)
5663 block_rsv_release_bytes(fs_info
, block_rsv
, global_rsv
, num_bytes
);
5666 static void update_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5668 struct btrfs_block_rsv
*block_rsv
= &fs_info
->global_block_rsv
;
5669 struct btrfs_space_info
*sinfo
= block_rsv
->space_info
;
5673 * The global block rsv is based on the size of the extent tree, the
5674 * checksum tree and the root tree. If the fs is empty we want to set
5675 * it to a minimal amount for safety.
5677 num_bytes
= btrfs_root_used(&fs_info
->extent_root
->root_item
) +
5678 btrfs_root_used(&fs_info
->csum_root
->root_item
) +
5679 btrfs_root_used(&fs_info
->tree_root
->root_item
);
5680 num_bytes
= max_t(u64
, num_bytes
, SZ_16M
);
5682 spin_lock(&sinfo
->lock
);
5683 spin_lock(&block_rsv
->lock
);
5685 block_rsv
->size
= min_t(u64
, num_bytes
, SZ_512M
);
5687 if (block_rsv
->reserved
< block_rsv
->size
) {
5688 num_bytes
= btrfs_space_info_used(sinfo
, true);
5689 if (sinfo
->total_bytes
> num_bytes
) {
5690 num_bytes
= sinfo
->total_bytes
- num_bytes
;
5691 num_bytes
= min(num_bytes
,
5692 block_rsv
->size
- block_rsv
->reserved
);
5693 block_rsv
->reserved
+= num_bytes
;
5694 sinfo
->bytes_may_use
+= num_bytes
;
5695 trace_btrfs_space_reservation(fs_info
, "space_info",
5696 sinfo
->flags
, num_bytes
,
5699 } else if (block_rsv
->reserved
> block_rsv
->size
) {
5700 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
5701 sinfo
->bytes_may_use
-= num_bytes
;
5702 trace_btrfs_space_reservation(fs_info
, "space_info",
5703 sinfo
->flags
, num_bytes
, 0);
5704 block_rsv
->reserved
= block_rsv
->size
;
5707 if (block_rsv
->reserved
== block_rsv
->size
)
5708 block_rsv
->full
= 1;
5710 block_rsv
->full
= 0;
5712 spin_unlock(&block_rsv
->lock
);
5713 spin_unlock(&sinfo
->lock
);
5716 static void init_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5718 struct btrfs_space_info
*space_info
;
5720 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
5721 fs_info
->chunk_block_rsv
.space_info
= space_info
;
5723 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
5724 fs_info
->global_block_rsv
.space_info
= space_info
;
5725 fs_info
->delalloc_block_rsv
.space_info
= space_info
;
5726 fs_info
->trans_block_rsv
.space_info
= space_info
;
5727 fs_info
->empty_block_rsv
.space_info
= space_info
;
5728 fs_info
->delayed_block_rsv
.space_info
= space_info
;
5730 fs_info
->extent_root
->block_rsv
= &fs_info
->global_block_rsv
;
5731 fs_info
->csum_root
->block_rsv
= &fs_info
->global_block_rsv
;
5732 fs_info
->dev_root
->block_rsv
= &fs_info
->global_block_rsv
;
5733 fs_info
->tree_root
->block_rsv
= &fs_info
->global_block_rsv
;
5734 if (fs_info
->quota_root
)
5735 fs_info
->quota_root
->block_rsv
= &fs_info
->global_block_rsv
;
5736 fs_info
->chunk_root
->block_rsv
= &fs_info
->chunk_block_rsv
;
5738 update_global_block_rsv(fs_info
);
5741 static void release_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5743 block_rsv_release_bytes(fs_info
, &fs_info
->global_block_rsv
, NULL
,
5745 WARN_ON(fs_info
->delalloc_block_rsv
.size
> 0);
5746 WARN_ON(fs_info
->delalloc_block_rsv
.reserved
> 0);
5747 WARN_ON(fs_info
->trans_block_rsv
.size
> 0);
5748 WARN_ON(fs_info
->trans_block_rsv
.reserved
> 0);
5749 WARN_ON(fs_info
->chunk_block_rsv
.size
> 0);
5750 WARN_ON(fs_info
->chunk_block_rsv
.reserved
> 0);
5751 WARN_ON(fs_info
->delayed_block_rsv
.size
> 0);
5752 WARN_ON(fs_info
->delayed_block_rsv
.reserved
> 0);
5755 void btrfs_trans_release_metadata(struct btrfs_trans_handle
*trans
,
5756 struct btrfs_fs_info
*fs_info
)
5758 if (!trans
->block_rsv
)
5761 if (!trans
->bytes_reserved
)
5764 trace_btrfs_space_reservation(fs_info
, "transaction",
5765 trans
->transid
, trans
->bytes_reserved
, 0);
5766 btrfs_block_rsv_release(fs_info
, trans
->block_rsv
,
5767 trans
->bytes_reserved
);
5768 trans
->bytes_reserved
= 0;
5772 * To be called after all the new block groups attached to the transaction
5773 * handle have been created (btrfs_create_pending_block_groups()).
5775 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle
*trans
)
5777 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
5779 if (!trans
->chunk_bytes_reserved
)
5782 WARN_ON_ONCE(!list_empty(&trans
->new_bgs
));
5784 block_rsv_release_bytes(fs_info
, &fs_info
->chunk_block_rsv
, NULL
,
5785 trans
->chunk_bytes_reserved
);
5786 trans
->chunk_bytes_reserved
= 0;
5789 /* Can only return 0 or -ENOSPC */
5790 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle
*trans
,
5791 struct btrfs_inode
*inode
)
5793 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
5794 struct btrfs_root
*root
= inode
->root
;
5796 * We always use trans->block_rsv here as we will have reserved space
5797 * for our orphan when starting the transaction, using get_block_rsv()
5798 * here will sometimes make us choose the wrong block rsv as we could be
5799 * doing a reloc inode for a non refcounted root.
5801 struct btrfs_block_rsv
*src_rsv
= trans
->block_rsv
;
5802 struct btrfs_block_rsv
*dst_rsv
= root
->orphan_block_rsv
;
5805 * We need to hold space in order to delete our orphan item once we've
5806 * added it, so this takes the reservation so we can release it later
5807 * when we are truly done with the orphan item.
5809 u64 num_bytes
= btrfs_calc_trans_metadata_size(fs_info
, 1);
5811 trace_btrfs_space_reservation(fs_info
, "orphan", btrfs_ino(inode
),
5813 return btrfs_block_rsv_migrate(src_rsv
, dst_rsv
, num_bytes
, 1);
5816 void btrfs_orphan_release_metadata(struct btrfs_inode
*inode
)
5818 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
5819 struct btrfs_root
*root
= inode
->root
;
5820 u64 num_bytes
= btrfs_calc_trans_metadata_size(fs_info
, 1);
5822 trace_btrfs_space_reservation(fs_info
, "orphan", btrfs_ino(inode
),
5824 btrfs_block_rsv_release(fs_info
, root
->orphan_block_rsv
, num_bytes
);
5828 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5829 * root: the root of the parent directory
5830 * rsv: block reservation
5831 * items: the number of items that we need do reservation
5832 * qgroup_reserved: used to return the reserved size in qgroup
5834 * This function is used to reserve the space for snapshot/subvolume
5835 * creation and deletion. Those operations are different with the
5836 * common file/directory operations, they change two fs/file trees
5837 * and root tree, the number of items that the qgroup reserves is
5838 * different with the free space reservation. So we can not use
5839 * the space reservation mechanism in start_transaction().
5841 int btrfs_subvolume_reserve_metadata(struct btrfs_root
*root
,
5842 struct btrfs_block_rsv
*rsv
,
5844 u64
*qgroup_reserved
,
5845 bool use_global_rsv
)
5849 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5850 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5852 if (test_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
)) {
5853 /* One for parent inode, two for dir entries */
5854 num_bytes
= 3 * fs_info
->nodesize
;
5855 ret
= btrfs_qgroup_reserve_meta(root
, num_bytes
, true);
5862 *qgroup_reserved
= num_bytes
;
5864 num_bytes
= btrfs_calc_trans_metadata_size(fs_info
, items
);
5865 rsv
->space_info
= __find_space_info(fs_info
,
5866 BTRFS_BLOCK_GROUP_METADATA
);
5867 ret
= btrfs_block_rsv_add(root
, rsv
, num_bytes
,
5868 BTRFS_RESERVE_FLUSH_ALL
);
5870 if (ret
== -ENOSPC
&& use_global_rsv
)
5871 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
, num_bytes
, 1);
5873 if (ret
&& *qgroup_reserved
)
5874 btrfs_qgroup_free_meta(root
, *qgroup_reserved
);
5879 void btrfs_subvolume_release_metadata(struct btrfs_fs_info
*fs_info
,
5880 struct btrfs_block_rsv
*rsv
)
5882 btrfs_block_rsv_release(fs_info
, rsv
, (u64
)-1);
5886 * drop_outstanding_extent - drop an outstanding extent
5887 * @inode: the inode we're dropping the extent for
5888 * @num_bytes: the number of bytes we're releasing.
5890 * This is called when we are freeing up an outstanding extent, either called
5891 * after an error or after an extent is written. This will return the number of
5892 * reserved extents that need to be freed. This must be called with
5893 * BTRFS_I(inode)->lock held.
5895 static unsigned drop_outstanding_extent(struct btrfs_inode
*inode
,
5898 unsigned drop_inode_space
= 0;
5899 unsigned dropped_extents
= 0;
5900 unsigned num_extents
;
5902 num_extents
= count_max_extents(num_bytes
);
5903 ASSERT(num_extents
);
5904 ASSERT(inode
->outstanding_extents
>= num_extents
);
5905 inode
->outstanding_extents
-= num_extents
;
5907 if (inode
->outstanding_extents
== 0 &&
5908 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5909 &inode
->runtime_flags
))
5910 drop_inode_space
= 1;
5913 * If we have more or the same amount of outstanding extents than we have
5914 * reserved then we need to leave the reserved extents count alone.
5916 if (inode
->outstanding_extents
>= inode
->reserved_extents
)
5917 return drop_inode_space
;
5919 dropped_extents
= inode
->reserved_extents
- inode
->outstanding_extents
;
5920 inode
->reserved_extents
-= dropped_extents
;
5921 return dropped_extents
+ drop_inode_space
;
5925 * calc_csum_metadata_size - return the amount of metadata space that must be
5926 * reserved/freed for the given bytes.
5927 * @inode: the inode we're manipulating
5928 * @num_bytes: the number of bytes in question
5929 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5931 * This adjusts the number of csum_bytes in the inode and then returns the
5932 * correct amount of metadata that must either be reserved or freed. We
5933 * calculate how many checksums we can fit into one leaf and then divide the
5934 * number of bytes that will need to be checksumed by this value to figure out
5935 * how many checksums will be required. If we are adding bytes then the number
5936 * may go up and we will return the number of additional bytes that must be
5937 * reserved. If it is going down we will return the number of bytes that must
5940 * This must be called with BTRFS_I(inode)->lock held.
5942 static u64
calc_csum_metadata_size(struct btrfs_inode
*inode
, u64 num_bytes
,
5945 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
5946 u64 old_csums
, num_csums
;
5948 if (inode
->flags
& BTRFS_INODE_NODATASUM
&& inode
->csum_bytes
== 0)
5951 old_csums
= btrfs_csum_bytes_to_leaves(fs_info
, inode
->csum_bytes
);
5953 inode
->csum_bytes
+= num_bytes
;
5955 inode
->csum_bytes
-= num_bytes
;
5956 num_csums
= btrfs_csum_bytes_to_leaves(fs_info
, inode
->csum_bytes
);
5958 /* No change, no need to reserve more */
5959 if (old_csums
== num_csums
)
5963 return btrfs_calc_trans_metadata_size(fs_info
,
5964 num_csums
- old_csums
);
5966 return btrfs_calc_trans_metadata_size(fs_info
, old_csums
- num_csums
);
5969 int btrfs_delalloc_reserve_metadata(struct btrfs_inode
*inode
, u64 num_bytes
)
5971 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
5972 struct btrfs_root
*root
= inode
->root
;
5973 struct btrfs_block_rsv
*block_rsv
= &fs_info
->delalloc_block_rsv
;
5976 unsigned nr_extents
;
5977 enum btrfs_reserve_flush_enum flush
= BTRFS_RESERVE_FLUSH_ALL
;
5979 bool delalloc_lock
= true;
5982 bool release_extra
= false;
5984 /* If we are a free space inode we need to not flush since we will be in
5985 * the middle of a transaction commit. We also don't need the delalloc
5986 * mutex since we won't race with anybody. We need this mostly to make
5987 * lockdep shut its filthy mouth.
5989 * If we have a transaction open (can happen if we call truncate_block
5990 * from truncate), then we need FLUSH_LIMIT so we don't deadlock.
5992 if (btrfs_is_free_space_inode(inode
)) {
5993 flush
= BTRFS_RESERVE_NO_FLUSH
;
5994 delalloc_lock
= false;
5995 } else if (current
->journal_info
) {
5996 flush
= BTRFS_RESERVE_FLUSH_LIMIT
;
5999 if (flush
!= BTRFS_RESERVE_NO_FLUSH
&&
6000 btrfs_transaction_in_commit(fs_info
))
6001 schedule_timeout(1);
6004 mutex_lock(&inode
->delalloc_mutex
);
6006 num_bytes
= ALIGN(num_bytes
, fs_info
->sectorsize
);
6008 spin_lock(&inode
->lock
);
6009 nr_extents
= count_max_extents(num_bytes
);
6010 inode
->outstanding_extents
+= nr_extents
;
6013 if (inode
->outstanding_extents
> inode
->reserved_extents
)
6014 nr_extents
+= inode
->outstanding_extents
-
6015 inode
->reserved_extents
;
6017 /* We always want to reserve a slot for updating the inode. */
6018 to_reserve
= btrfs_calc_trans_metadata_size(fs_info
, nr_extents
+ 1);
6019 to_reserve
+= calc_csum_metadata_size(inode
, num_bytes
, 1);
6020 csum_bytes
= inode
->csum_bytes
;
6021 spin_unlock(&inode
->lock
);
6023 if (test_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
)) {
6024 ret
= btrfs_qgroup_reserve_meta(root
,
6025 nr_extents
* fs_info
->nodesize
, true);
6030 ret
= btrfs_block_rsv_add(root
, block_rsv
, to_reserve
, flush
);
6031 if (unlikely(ret
)) {
6032 btrfs_qgroup_free_meta(root
,
6033 nr_extents
* fs_info
->nodesize
);
6037 spin_lock(&inode
->lock
);
6038 if (test_and_set_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
6039 &inode
->runtime_flags
)) {
6040 to_reserve
-= btrfs_calc_trans_metadata_size(fs_info
, 1);
6041 release_extra
= true;
6043 inode
->reserved_extents
+= nr_extents
;
6044 spin_unlock(&inode
->lock
);
6047 mutex_unlock(&inode
->delalloc_mutex
);
6050 trace_btrfs_space_reservation(fs_info
, "delalloc",
6051 btrfs_ino(inode
), to_reserve
, 1);
6053 btrfs_block_rsv_release(fs_info
, block_rsv
,
6054 btrfs_calc_trans_metadata_size(fs_info
, 1));
6058 spin_lock(&inode
->lock
);
6059 dropped
= drop_outstanding_extent(inode
, num_bytes
);
6061 * If the inodes csum_bytes is the same as the original
6062 * csum_bytes then we know we haven't raced with any free()ers
6063 * so we can just reduce our inodes csum bytes and carry on.
6065 if (inode
->csum_bytes
== csum_bytes
) {
6066 calc_csum_metadata_size(inode
, num_bytes
, 0);
6068 u64 orig_csum_bytes
= inode
->csum_bytes
;
6072 * This is tricky, but first we need to figure out how much we
6073 * freed from any free-ers that occurred during this
6074 * reservation, so we reset ->csum_bytes to the csum_bytes
6075 * before we dropped our lock, and then call the free for the
6076 * number of bytes that were freed while we were trying our
6079 bytes
= csum_bytes
- inode
->csum_bytes
;
6080 inode
->csum_bytes
= csum_bytes
;
6081 to_free
= calc_csum_metadata_size(inode
, bytes
, 0);
6085 * Now we need to see how much we would have freed had we not
6086 * been making this reservation and our ->csum_bytes were not
6087 * artificially inflated.
6089 inode
->csum_bytes
= csum_bytes
- num_bytes
;
6090 bytes
= csum_bytes
- orig_csum_bytes
;
6091 bytes
= calc_csum_metadata_size(inode
, bytes
, 0);
6094 * Now reset ->csum_bytes to what it should be. If bytes is
6095 * more than to_free then we would have freed more space had we
6096 * not had an artificially high ->csum_bytes, so we need to free
6097 * the remainder. If bytes is the same or less then we don't
6098 * need to do anything, the other free-ers did the correct
6101 inode
->csum_bytes
= orig_csum_bytes
- num_bytes
;
6102 if (bytes
> to_free
)
6103 to_free
= bytes
- to_free
;
6107 spin_unlock(&inode
->lock
);
6109 to_free
+= btrfs_calc_trans_metadata_size(fs_info
, dropped
);
6112 btrfs_block_rsv_release(fs_info
, block_rsv
, to_free
);
6113 trace_btrfs_space_reservation(fs_info
, "delalloc",
6114 btrfs_ino(inode
), to_free
, 0);
6117 mutex_unlock(&inode
->delalloc_mutex
);
6122 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
6123 * @inode: the inode to release the reservation for
6124 * @num_bytes: the number of bytes we're releasing
6126 * This will release the metadata reservation for an inode. This can be called
6127 * once we complete IO for a given set of bytes to release their metadata
6130 void btrfs_delalloc_release_metadata(struct btrfs_inode
*inode
, u64 num_bytes
)
6132 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
6136 num_bytes
= ALIGN(num_bytes
, fs_info
->sectorsize
);
6137 spin_lock(&inode
->lock
);
6138 dropped
= drop_outstanding_extent(inode
, num_bytes
);
6141 to_free
= calc_csum_metadata_size(inode
, num_bytes
, 0);
6142 spin_unlock(&inode
->lock
);
6144 to_free
+= btrfs_calc_trans_metadata_size(fs_info
, dropped
);
6146 if (btrfs_is_testing(fs_info
))
6149 trace_btrfs_space_reservation(fs_info
, "delalloc", btrfs_ino(inode
),
6152 btrfs_block_rsv_release(fs_info
, &fs_info
->delalloc_block_rsv
, to_free
);
6156 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6158 * @inode: inode we're writing to
6159 * @start: start range we are writing to
6160 * @len: how long the range we are writing to
6162 * This will do the following things
6164 * o reserve space in data space info for num bytes
6165 * and reserve precious corresponding qgroup space
6166 * (Done in check_data_free_space)
6168 * o reserve space for metadata space, based on the number of outstanding
6169 * extents and how much csums will be needed
6170 * also reserve metadata space in a per root over-reserve method.
6171 * o add to the inodes->delalloc_bytes
6172 * o add it to the fs_info's delalloc inodes list.
6173 * (Above 3 all done in delalloc_reserve_metadata)
6175 * Return 0 for success
6176 * Return <0 for error(-ENOSPC or -EQUOT)
6178 int btrfs_delalloc_reserve_space(struct inode
*inode
, u64 start
, u64 len
)
6182 ret
= btrfs_check_data_free_space(inode
, start
, len
);
6185 ret
= btrfs_delalloc_reserve_metadata(BTRFS_I(inode
), len
);
6187 btrfs_free_reserved_data_space(inode
, start
, len
);
6192 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6193 * @inode: inode we're releasing space for
6194 * @start: start position of the space already reserved
6195 * @len: the len of the space already reserved
6197 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
6198 * called in the case that we don't need the metadata AND data reservations
6199 * anymore. So if there is an error or we insert an inline extent.
6201 * This function will release the metadata space that was not used and will
6202 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
6203 * list if there are no delalloc bytes left.
6204 * Also it will handle the qgroup reserved space.
6206 void btrfs_delalloc_release_space(struct inode
*inode
, u64 start
, u64 len
)
6208 btrfs_delalloc_release_metadata(BTRFS_I(inode
), len
);
6209 btrfs_free_reserved_data_space(inode
, start
, len
);
6212 static int update_block_group(struct btrfs_trans_handle
*trans
,
6213 struct btrfs_fs_info
*info
, u64 bytenr
,
6214 u64 num_bytes
, int alloc
)
6216 struct btrfs_block_group_cache
*cache
= NULL
;
6217 u64 total
= num_bytes
;
6222 /* block accounting for super block */
6223 spin_lock(&info
->delalloc_root_lock
);
6224 old_val
= btrfs_super_bytes_used(info
->super_copy
);
6226 old_val
+= num_bytes
;
6228 old_val
-= num_bytes
;
6229 btrfs_set_super_bytes_used(info
->super_copy
, old_val
);
6230 spin_unlock(&info
->delalloc_root_lock
);
6233 cache
= btrfs_lookup_block_group(info
, bytenr
);
6236 if (cache
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
6237 BTRFS_BLOCK_GROUP_RAID1
|
6238 BTRFS_BLOCK_GROUP_RAID10
))
6243 * If this block group has free space cache written out, we
6244 * need to make sure to load it if we are removing space. This
6245 * is because we need the unpinning stage to actually add the
6246 * space back to the block group, otherwise we will leak space.
6248 if (!alloc
&& cache
->cached
== BTRFS_CACHE_NO
)
6249 cache_block_group(cache
, 1);
6251 byte_in_group
= bytenr
- cache
->key
.objectid
;
6252 WARN_ON(byte_in_group
> cache
->key
.offset
);
6254 spin_lock(&cache
->space_info
->lock
);
6255 spin_lock(&cache
->lock
);
6257 if (btrfs_test_opt(info
, SPACE_CACHE
) &&
6258 cache
->disk_cache_state
< BTRFS_DC_CLEAR
)
6259 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
6261 old_val
= btrfs_block_group_used(&cache
->item
);
6262 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
6264 old_val
+= num_bytes
;
6265 btrfs_set_block_group_used(&cache
->item
, old_val
);
6266 cache
->reserved
-= num_bytes
;
6267 cache
->space_info
->bytes_reserved
-= num_bytes
;
6268 cache
->space_info
->bytes_used
+= num_bytes
;
6269 cache
->space_info
->disk_used
+= num_bytes
* factor
;
6270 spin_unlock(&cache
->lock
);
6271 spin_unlock(&cache
->space_info
->lock
);
6273 old_val
-= num_bytes
;
6274 btrfs_set_block_group_used(&cache
->item
, old_val
);
6275 cache
->pinned
+= num_bytes
;
6276 cache
->space_info
->bytes_pinned
+= num_bytes
;
6277 cache
->space_info
->bytes_used
-= num_bytes
;
6278 cache
->space_info
->disk_used
-= num_bytes
* factor
;
6279 spin_unlock(&cache
->lock
);
6280 spin_unlock(&cache
->space_info
->lock
);
6282 trace_btrfs_space_reservation(info
, "pinned",
6283 cache
->space_info
->flags
,
6285 set_extent_dirty(info
->pinned_extents
,
6286 bytenr
, bytenr
+ num_bytes
- 1,
6287 GFP_NOFS
| __GFP_NOFAIL
);
6290 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
6291 if (list_empty(&cache
->dirty_list
)) {
6292 list_add_tail(&cache
->dirty_list
,
6293 &trans
->transaction
->dirty_bgs
);
6294 trans
->transaction
->num_dirty_bgs
++;
6295 btrfs_get_block_group(cache
);
6297 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
6300 * No longer have used bytes in this block group, queue it for
6301 * deletion. We do this after adding the block group to the
6302 * dirty list to avoid races between cleaner kthread and space
6305 if (!alloc
&& old_val
== 0) {
6306 spin_lock(&info
->unused_bgs_lock
);
6307 if (list_empty(&cache
->bg_list
)) {
6308 btrfs_get_block_group(cache
);
6309 list_add_tail(&cache
->bg_list
,
6312 spin_unlock(&info
->unused_bgs_lock
);
6315 btrfs_put_block_group(cache
);
6317 bytenr
+= num_bytes
;
6322 static u64
first_logical_byte(struct btrfs_fs_info
*fs_info
, u64 search_start
)
6324 struct btrfs_block_group_cache
*cache
;
6327 spin_lock(&fs_info
->block_group_cache_lock
);
6328 bytenr
= fs_info
->first_logical_byte
;
6329 spin_unlock(&fs_info
->block_group_cache_lock
);
6331 if (bytenr
< (u64
)-1)
6334 cache
= btrfs_lookup_first_block_group(fs_info
, search_start
);
6338 bytenr
= cache
->key
.objectid
;
6339 btrfs_put_block_group(cache
);
6344 static int pin_down_extent(struct btrfs_fs_info
*fs_info
,
6345 struct btrfs_block_group_cache
*cache
,
6346 u64 bytenr
, u64 num_bytes
, int reserved
)
6348 spin_lock(&cache
->space_info
->lock
);
6349 spin_lock(&cache
->lock
);
6350 cache
->pinned
+= num_bytes
;
6351 cache
->space_info
->bytes_pinned
+= num_bytes
;
6353 cache
->reserved
-= num_bytes
;
6354 cache
->space_info
->bytes_reserved
-= num_bytes
;
6356 spin_unlock(&cache
->lock
);
6357 spin_unlock(&cache
->space_info
->lock
);
6359 trace_btrfs_space_reservation(fs_info
, "pinned",
6360 cache
->space_info
->flags
, num_bytes
, 1);
6361 set_extent_dirty(fs_info
->pinned_extents
, bytenr
,
6362 bytenr
+ num_bytes
- 1, GFP_NOFS
| __GFP_NOFAIL
);
6367 * this function must be called within transaction
6369 int btrfs_pin_extent(struct btrfs_fs_info
*fs_info
,
6370 u64 bytenr
, u64 num_bytes
, int reserved
)
6372 struct btrfs_block_group_cache
*cache
;
6374 cache
= btrfs_lookup_block_group(fs_info
, bytenr
);
6375 BUG_ON(!cache
); /* Logic error */
6377 pin_down_extent(fs_info
, cache
, bytenr
, num_bytes
, reserved
);
6379 btrfs_put_block_group(cache
);
6384 * this function must be called within transaction
6386 int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info
*fs_info
,
6387 u64 bytenr
, u64 num_bytes
)
6389 struct btrfs_block_group_cache
*cache
;
6392 cache
= btrfs_lookup_block_group(fs_info
, bytenr
);
6397 * pull in the free space cache (if any) so that our pin
6398 * removes the free space from the cache. We have load_only set
6399 * to one because the slow code to read in the free extents does check
6400 * the pinned extents.
6402 cache_block_group(cache
, 1);
6404 pin_down_extent(fs_info
, cache
, bytenr
, num_bytes
, 0);
6406 /* remove us from the free space cache (if we're there at all) */
6407 ret
= btrfs_remove_free_space(cache
, bytenr
, num_bytes
);
6408 btrfs_put_block_group(cache
);
6412 static int __exclude_logged_extent(struct btrfs_fs_info
*fs_info
,
6413 u64 start
, u64 num_bytes
)
6416 struct btrfs_block_group_cache
*block_group
;
6417 struct btrfs_caching_control
*caching_ctl
;
6419 block_group
= btrfs_lookup_block_group(fs_info
, start
);
6423 cache_block_group(block_group
, 0);
6424 caching_ctl
= get_caching_control(block_group
);
6428 BUG_ON(!block_group_cache_done(block_group
));
6429 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
6431 mutex_lock(&caching_ctl
->mutex
);
6433 if (start
>= caching_ctl
->progress
) {
6434 ret
= add_excluded_extent(fs_info
, start
, num_bytes
);
6435 } else if (start
+ num_bytes
<= caching_ctl
->progress
) {
6436 ret
= btrfs_remove_free_space(block_group
,
6439 num_bytes
= caching_ctl
->progress
- start
;
6440 ret
= btrfs_remove_free_space(block_group
,
6445 num_bytes
= (start
+ num_bytes
) -
6446 caching_ctl
->progress
;
6447 start
= caching_ctl
->progress
;
6448 ret
= add_excluded_extent(fs_info
, start
, num_bytes
);
6451 mutex_unlock(&caching_ctl
->mutex
);
6452 put_caching_control(caching_ctl
);
6454 btrfs_put_block_group(block_group
);
6458 int btrfs_exclude_logged_extents(struct btrfs_fs_info
*fs_info
,
6459 struct extent_buffer
*eb
)
6461 struct btrfs_file_extent_item
*item
;
6462 struct btrfs_key key
;
6466 if (!btrfs_fs_incompat(fs_info
, MIXED_GROUPS
))
6469 for (i
= 0; i
< btrfs_header_nritems(eb
); i
++) {
6470 btrfs_item_key_to_cpu(eb
, &key
, i
);
6471 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
6473 item
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
6474 found_type
= btrfs_file_extent_type(eb
, item
);
6475 if (found_type
== BTRFS_FILE_EXTENT_INLINE
)
6477 if (btrfs_file_extent_disk_bytenr(eb
, item
) == 0)
6479 key
.objectid
= btrfs_file_extent_disk_bytenr(eb
, item
);
6480 key
.offset
= btrfs_file_extent_disk_num_bytes(eb
, item
);
6481 __exclude_logged_extent(fs_info
, key
.objectid
, key
.offset
);
6488 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache
*bg
)
6490 atomic_inc(&bg
->reservations
);
6493 void btrfs_dec_block_group_reservations(struct btrfs_fs_info
*fs_info
,
6496 struct btrfs_block_group_cache
*bg
;
6498 bg
= btrfs_lookup_block_group(fs_info
, start
);
6500 if (atomic_dec_and_test(&bg
->reservations
))
6501 wake_up_atomic_t(&bg
->reservations
);
6502 btrfs_put_block_group(bg
);
6505 static int btrfs_wait_bg_reservations_atomic_t(atomic_t
*a
)
6511 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache
*bg
)
6513 struct btrfs_space_info
*space_info
= bg
->space_info
;
6517 if (!(bg
->flags
& BTRFS_BLOCK_GROUP_DATA
))
6521 * Our block group is read only but before we set it to read only,
6522 * some task might have had allocated an extent from it already, but it
6523 * has not yet created a respective ordered extent (and added it to a
6524 * root's list of ordered extents).
6525 * Therefore wait for any task currently allocating extents, since the
6526 * block group's reservations counter is incremented while a read lock
6527 * on the groups' semaphore is held and decremented after releasing
6528 * the read access on that semaphore and creating the ordered extent.
6530 down_write(&space_info
->groups_sem
);
6531 up_write(&space_info
->groups_sem
);
6533 wait_on_atomic_t(&bg
->reservations
,
6534 btrfs_wait_bg_reservations_atomic_t
,
6535 TASK_UNINTERRUPTIBLE
);
6539 * btrfs_add_reserved_bytes - update the block_group and space info counters
6540 * @cache: The cache we are manipulating
6541 * @ram_bytes: The number of bytes of file content, and will be same to
6542 * @num_bytes except for the compress path.
6543 * @num_bytes: The number of bytes in question
6544 * @delalloc: The blocks are allocated for the delalloc write
6546 * This is called by the allocator when it reserves space. If this is a
6547 * reservation and the block group has become read only we cannot make the
6548 * reservation and return -EAGAIN, otherwise this function always succeeds.
6550 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache
*cache
,
6551 u64 ram_bytes
, u64 num_bytes
, int delalloc
)
6553 struct btrfs_space_info
*space_info
= cache
->space_info
;
6556 spin_lock(&space_info
->lock
);
6557 spin_lock(&cache
->lock
);
6561 cache
->reserved
+= num_bytes
;
6562 space_info
->bytes_reserved
+= num_bytes
;
6564 trace_btrfs_space_reservation(cache
->fs_info
,
6565 "space_info", space_info
->flags
,
6567 space_info
->bytes_may_use
-= ram_bytes
;
6569 cache
->delalloc_bytes
+= num_bytes
;
6571 spin_unlock(&cache
->lock
);
6572 spin_unlock(&space_info
->lock
);
6577 * btrfs_free_reserved_bytes - update the block_group and space info counters
6578 * @cache: The cache we are manipulating
6579 * @num_bytes: The number of bytes in question
6580 * @delalloc: The blocks are allocated for the delalloc write
6582 * This is called by somebody who is freeing space that was never actually used
6583 * on disk. For example if you reserve some space for a new leaf in transaction
6584 * A and before transaction A commits you free that leaf, you call this with
6585 * reserve set to 0 in order to clear the reservation.
6588 static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache
*cache
,
6589 u64 num_bytes
, int delalloc
)
6591 struct btrfs_space_info
*space_info
= cache
->space_info
;
6594 spin_lock(&space_info
->lock
);
6595 spin_lock(&cache
->lock
);
6597 space_info
->bytes_readonly
+= num_bytes
;
6598 cache
->reserved
-= num_bytes
;
6599 space_info
->bytes_reserved
-= num_bytes
;
6602 cache
->delalloc_bytes
-= num_bytes
;
6603 spin_unlock(&cache
->lock
);
6604 spin_unlock(&space_info
->lock
);
6607 void btrfs_prepare_extent_commit(struct btrfs_fs_info
*fs_info
)
6609 struct btrfs_caching_control
*next
;
6610 struct btrfs_caching_control
*caching_ctl
;
6611 struct btrfs_block_group_cache
*cache
;
6613 down_write(&fs_info
->commit_root_sem
);
6615 list_for_each_entry_safe(caching_ctl
, next
,
6616 &fs_info
->caching_block_groups
, list
) {
6617 cache
= caching_ctl
->block_group
;
6618 if (block_group_cache_done(cache
)) {
6619 cache
->last_byte_to_unpin
= (u64
)-1;
6620 list_del_init(&caching_ctl
->list
);
6621 put_caching_control(caching_ctl
);
6623 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
6627 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
6628 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
6630 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
6632 up_write(&fs_info
->commit_root_sem
);
6634 update_global_block_rsv(fs_info
);
6638 * Returns the free cluster for the given space info and sets empty_cluster to
6639 * what it should be based on the mount options.
6641 static struct btrfs_free_cluster
*
6642 fetch_cluster_info(struct btrfs_fs_info
*fs_info
,
6643 struct btrfs_space_info
*space_info
, u64
*empty_cluster
)
6645 struct btrfs_free_cluster
*ret
= NULL
;
6646 bool ssd
= btrfs_test_opt(fs_info
, SSD
);
6649 if (btrfs_mixed_space_info(space_info
))
6653 *empty_cluster
= SZ_2M
;
6654 if (space_info
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
6655 ret
= &fs_info
->meta_alloc_cluster
;
6657 *empty_cluster
= SZ_64K
;
6658 } else if ((space_info
->flags
& BTRFS_BLOCK_GROUP_DATA
) && ssd
) {
6659 ret
= &fs_info
->data_alloc_cluster
;
6665 static int unpin_extent_range(struct btrfs_fs_info
*fs_info
,
6667 const bool return_free_space
)
6669 struct btrfs_block_group_cache
*cache
= NULL
;
6670 struct btrfs_space_info
*space_info
;
6671 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
6672 struct btrfs_free_cluster
*cluster
= NULL
;
6674 u64 total_unpinned
= 0;
6675 u64 empty_cluster
= 0;
6678 while (start
<= end
) {
6681 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
6683 btrfs_put_block_group(cache
);
6685 cache
= btrfs_lookup_block_group(fs_info
, start
);
6686 BUG_ON(!cache
); /* Logic error */
6688 cluster
= fetch_cluster_info(fs_info
,
6691 empty_cluster
<<= 1;
6694 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
6695 len
= min(len
, end
+ 1 - start
);
6697 if (start
< cache
->last_byte_to_unpin
) {
6698 len
= min(len
, cache
->last_byte_to_unpin
- start
);
6699 if (return_free_space
)
6700 btrfs_add_free_space(cache
, start
, len
);
6704 total_unpinned
+= len
;
6705 space_info
= cache
->space_info
;
6708 * If this space cluster has been marked as fragmented and we've
6709 * unpinned enough in this block group to potentially allow a
6710 * cluster to be created inside of it go ahead and clear the
6713 if (cluster
&& cluster
->fragmented
&&
6714 total_unpinned
> empty_cluster
) {
6715 spin_lock(&cluster
->lock
);
6716 cluster
->fragmented
= 0;
6717 spin_unlock(&cluster
->lock
);
6720 spin_lock(&space_info
->lock
);
6721 spin_lock(&cache
->lock
);
6722 cache
->pinned
-= len
;
6723 space_info
->bytes_pinned
-= len
;
6725 trace_btrfs_space_reservation(fs_info
, "pinned",
6726 space_info
->flags
, len
, 0);
6727 space_info
->max_extent_size
= 0;
6728 percpu_counter_add(&space_info
->total_bytes_pinned
, -len
);
6730 space_info
->bytes_readonly
+= len
;
6733 spin_unlock(&cache
->lock
);
6734 if (!readonly
&& return_free_space
&&
6735 global_rsv
->space_info
== space_info
) {
6737 WARN_ON(!return_free_space
);
6738 spin_lock(&global_rsv
->lock
);
6739 if (!global_rsv
->full
) {
6740 to_add
= min(len
, global_rsv
->size
-
6741 global_rsv
->reserved
);
6742 global_rsv
->reserved
+= to_add
;
6743 space_info
->bytes_may_use
+= to_add
;
6744 if (global_rsv
->reserved
>= global_rsv
->size
)
6745 global_rsv
->full
= 1;
6746 trace_btrfs_space_reservation(fs_info
,
6752 spin_unlock(&global_rsv
->lock
);
6753 /* Add to any tickets we may have */
6755 space_info_add_new_bytes(fs_info
, space_info
,
6758 spin_unlock(&space_info
->lock
);
6762 btrfs_put_block_group(cache
);
6766 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
6767 struct btrfs_fs_info
*fs_info
)
6769 struct btrfs_block_group_cache
*block_group
, *tmp
;
6770 struct list_head
*deleted_bgs
;
6771 struct extent_io_tree
*unpin
;
6776 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
6777 unpin
= &fs_info
->freed_extents
[1];
6779 unpin
= &fs_info
->freed_extents
[0];
6781 while (!trans
->aborted
) {
6782 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
6783 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
6784 EXTENT_DIRTY
, NULL
);
6786 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
6790 if (btrfs_test_opt(fs_info
, DISCARD
))
6791 ret
= btrfs_discard_extent(fs_info
, start
,
6792 end
+ 1 - start
, NULL
);
6794 clear_extent_dirty(unpin
, start
, end
);
6795 unpin_extent_range(fs_info
, start
, end
, true);
6796 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
6801 * Transaction is finished. We don't need the lock anymore. We
6802 * do need to clean up the block groups in case of a transaction
6805 deleted_bgs
= &trans
->transaction
->deleted_bgs
;
6806 list_for_each_entry_safe(block_group
, tmp
, deleted_bgs
, bg_list
) {
6810 if (!trans
->aborted
)
6811 ret
= btrfs_discard_extent(fs_info
,
6812 block_group
->key
.objectid
,
6813 block_group
->key
.offset
,
6816 list_del_init(&block_group
->bg_list
);
6817 btrfs_put_block_group_trimming(block_group
);
6818 btrfs_put_block_group(block_group
);
6821 const char *errstr
= btrfs_decode_error(ret
);
6823 "Discard failed while removing blockgroup: errno=%d %s\n",
6831 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
6832 struct btrfs_fs_info
*info
,
6833 struct btrfs_delayed_ref_node
*node
, u64 parent
,
6834 u64 root_objectid
, u64 owner_objectid
,
6835 u64 owner_offset
, int refs_to_drop
,
6836 struct btrfs_delayed_extent_op
*extent_op
)
6838 struct btrfs_key key
;
6839 struct btrfs_path
*path
;
6840 struct btrfs_root
*extent_root
= info
->extent_root
;
6841 struct extent_buffer
*leaf
;
6842 struct btrfs_extent_item
*ei
;
6843 struct btrfs_extent_inline_ref
*iref
;
6846 int extent_slot
= 0;
6847 int found_extent
= 0;
6851 u64 bytenr
= node
->bytenr
;
6852 u64 num_bytes
= node
->num_bytes
;
6854 bool skinny_metadata
= btrfs_fs_incompat(info
, SKINNY_METADATA
);
6856 path
= btrfs_alloc_path();
6860 path
->reada
= READA_FORWARD
;
6861 path
->leave_spinning
= 1;
6863 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
6864 BUG_ON(!is_data
&& refs_to_drop
!= 1);
6867 skinny_metadata
= 0;
6869 ret
= lookup_extent_backref(trans
, info
, path
, &iref
,
6870 bytenr
, num_bytes
, parent
,
6871 root_objectid
, owner_objectid
,
6874 extent_slot
= path
->slots
[0];
6875 while (extent_slot
>= 0) {
6876 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
6878 if (key
.objectid
!= bytenr
)
6880 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
6881 key
.offset
== num_bytes
) {
6885 if (key
.type
== BTRFS_METADATA_ITEM_KEY
&&
6886 key
.offset
== owner_objectid
) {
6890 if (path
->slots
[0] - extent_slot
> 5)
6894 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6895 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
6896 if (found_extent
&& item_size
< sizeof(*ei
))
6899 if (!found_extent
) {
6901 ret
= remove_extent_backref(trans
, info
, path
, NULL
,
6903 is_data
, &last_ref
);
6905 btrfs_abort_transaction(trans
, ret
);
6908 btrfs_release_path(path
);
6909 path
->leave_spinning
= 1;
6911 key
.objectid
= bytenr
;
6912 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6913 key
.offset
= num_bytes
;
6915 if (!is_data
&& skinny_metadata
) {
6916 key
.type
= BTRFS_METADATA_ITEM_KEY
;
6917 key
.offset
= owner_objectid
;
6920 ret
= btrfs_search_slot(trans
, extent_root
,
6922 if (ret
> 0 && skinny_metadata
&& path
->slots
[0]) {
6924 * Couldn't find our skinny metadata item,
6925 * see if we have ye olde extent item.
6928 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
6930 if (key
.objectid
== bytenr
&&
6931 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
6932 key
.offset
== num_bytes
)
6936 if (ret
> 0 && skinny_metadata
) {
6937 skinny_metadata
= false;
6938 key
.objectid
= bytenr
;
6939 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6940 key
.offset
= num_bytes
;
6941 btrfs_release_path(path
);
6942 ret
= btrfs_search_slot(trans
, extent_root
,
6948 "umm, got %d back from search, was looking for %llu",
6951 btrfs_print_leaf(info
, path
->nodes
[0]);
6954 btrfs_abort_transaction(trans
, ret
);
6957 extent_slot
= path
->slots
[0];
6959 } else if (WARN_ON(ret
== -ENOENT
)) {
6960 btrfs_print_leaf(info
, path
->nodes
[0]);
6962 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6963 bytenr
, parent
, root_objectid
, owner_objectid
,
6965 btrfs_abort_transaction(trans
, ret
);
6968 btrfs_abort_transaction(trans
, ret
);
6972 leaf
= path
->nodes
[0];
6973 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
6974 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6975 if (item_size
< sizeof(*ei
)) {
6976 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
6977 ret
= convert_extent_item_v0(trans
, info
, path
, owner_objectid
,
6980 btrfs_abort_transaction(trans
, ret
);
6984 btrfs_release_path(path
);
6985 path
->leave_spinning
= 1;
6987 key
.objectid
= bytenr
;
6988 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6989 key
.offset
= num_bytes
;
6991 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
6995 "umm, got %d back from search, was looking for %llu",
6997 btrfs_print_leaf(info
, path
->nodes
[0]);
7000 btrfs_abort_transaction(trans
, ret
);
7004 extent_slot
= path
->slots
[0];
7005 leaf
= path
->nodes
[0];
7006 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
7009 BUG_ON(item_size
< sizeof(*ei
));
7010 ei
= btrfs_item_ptr(leaf
, extent_slot
,
7011 struct btrfs_extent_item
);
7012 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
7013 key
.type
== BTRFS_EXTENT_ITEM_KEY
) {
7014 struct btrfs_tree_block_info
*bi
;
7015 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
7016 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
7017 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
7020 refs
= btrfs_extent_refs(leaf
, ei
);
7021 if (refs
< refs_to_drop
) {
7023 "trying to drop %d refs but we only have %Lu for bytenr %Lu",
7024 refs_to_drop
, refs
, bytenr
);
7026 btrfs_abort_transaction(trans
, ret
);
7029 refs
-= refs_to_drop
;
7033 __run_delayed_extent_op(extent_op
, leaf
, ei
);
7035 * In the case of inline back ref, reference count will
7036 * be updated by remove_extent_backref
7039 BUG_ON(!found_extent
);
7041 btrfs_set_extent_refs(leaf
, ei
, refs
);
7042 btrfs_mark_buffer_dirty(leaf
);
7045 ret
= remove_extent_backref(trans
, info
, path
,
7047 is_data
, &last_ref
);
7049 btrfs_abort_transaction(trans
, ret
);
7053 add_pinned_bytes(info
, -num_bytes
, owner_objectid
,
7057 BUG_ON(is_data
&& refs_to_drop
!=
7058 extent_data_ref_count(path
, iref
));
7060 BUG_ON(path
->slots
[0] != extent_slot
);
7062 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
7063 path
->slots
[0] = extent_slot
;
7069 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
7072 btrfs_abort_transaction(trans
, ret
);
7075 btrfs_release_path(path
);
7078 ret
= btrfs_del_csums(trans
, info
, bytenr
, num_bytes
);
7080 btrfs_abort_transaction(trans
, ret
);
7085 ret
= add_to_free_space_tree(trans
, info
, bytenr
, num_bytes
);
7087 btrfs_abort_transaction(trans
, ret
);
7091 ret
= update_block_group(trans
, info
, bytenr
, num_bytes
, 0);
7093 btrfs_abort_transaction(trans
, ret
);
7097 btrfs_release_path(path
);
7100 btrfs_free_path(path
);
7105 * when we free an block, it is possible (and likely) that we free the last
7106 * delayed ref for that extent as well. This searches the delayed ref tree for
7107 * a given extent, and if there are no other delayed refs to be processed, it
7108 * removes it from the tree.
7110 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
7113 struct btrfs_delayed_ref_head
*head
;
7114 struct btrfs_delayed_ref_root
*delayed_refs
;
7117 delayed_refs
= &trans
->transaction
->delayed_refs
;
7118 spin_lock(&delayed_refs
->lock
);
7119 head
= btrfs_find_delayed_ref_head(delayed_refs
, bytenr
);
7121 goto out_delayed_unlock
;
7123 spin_lock(&head
->lock
);
7124 if (!list_empty(&head
->ref_list
))
7127 if (head
->extent_op
) {
7128 if (!head
->must_insert_reserved
)
7130 btrfs_free_delayed_extent_op(head
->extent_op
);
7131 head
->extent_op
= NULL
;
7135 * waiting for the lock here would deadlock. If someone else has it
7136 * locked they are already in the process of dropping it anyway
7138 if (!mutex_trylock(&head
->mutex
))
7142 * at this point we have a head with no other entries. Go
7143 * ahead and process it.
7145 head
->node
.in_tree
= 0;
7146 rb_erase(&head
->href_node
, &delayed_refs
->href_root
);
7148 atomic_dec(&delayed_refs
->num_entries
);
7151 * we don't take a ref on the node because we're removing it from the
7152 * tree, so we just steal the ref the tree was holding.
7154 delayed_refs
->num_heads
--;
7155 if (head
->processing
== 0)
7156 delayed_refs
->num_heads_ready
--;
7157 head
->processing
= 0;
7158 spin_unlock(&head
->lock
);
7159 spin_unlock(&delayed_refs
->lock
);
7161 BUG_ON(head
->extent_op
);
7162 if (head
->must_insert_reserved
)
7165 mutex_unlock(&head
->mutex
);
7166 btrfs_put_delayed_ref(&head
->node
);
7169 spin_unlock(&head
->lock
);
7172 spin_unlock(&delayed_refs
->lock
);
7176 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
7177 struct btrfs_root
*root
,
7178 struct extent_buffer
*buf
,
7179 u64 parent
, int last_ref
)
7181 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7185 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
7186 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
,
7187 buf
->start
, buf
->len
,
7189 root
->root_key
.objectid
,
7190 btrfs_header_level(buf
),
7191 BTRFS_DROP_DELAYED_REF
, NULL
);
7192 BUG_ON(ret
); /* -ENOMEM */
7198 if (btrfs_header_generation(buf
) == trans
->transid
) {
7199 struct btrfs_block_group_cache
*cache
;
7201 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
7202 ret
= check_ref_cleanup(trans
, buf
->start
);
7207 cache
= btrfs_lookup_block_group(fs_info
, buf
->start
);
7209 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
7210 pin_down_extent(fs_info
, cache
, buf
->start
,
7212 btrfs_put_block_group(cache
);
7216 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
7218 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
7219 btrfs_free_reserved_bytes(cache
, buf
->len
, 0);
7220 btrfs_put_block_group(cache
);
7221 trace_btrfs_reserved_extent_free(fs_info
, buf
->start
, buf
->len
);
7226 add_pinned_bytes(fs_info
, buf
->len
, btrfs_header_level(buf
),
7227 root
->root_key
.objectid
);
7230 * Deleting the buffer, clear the corrupt flag since it doesn't matter
7233 clear_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
);
7236 /* Can return -ENOMEM */
7237 int btrfs_free_extent(struct btrfs_trans_handle
*trans
,
7238 struct btrfs_fs_info
*fs_info
,
7239 u64 bytenr
, u64 num_bytes
, u64 parent
, u64 root_objectid
,
7240 u64 owner
, u64 offset
)
7244 if (btrfs_is_testing(fs_info
))
7247 add_pinned_bytes(fs_info
, num_bytes
, owner
, root_objectid
);
7250 * tree log blocks never actually go into the extent allocation
7251 * tree, just update pinning info and exit early.
7253 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
7254 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
7255 /* unlocks the pinned mutex */
7256 btrfs_pin_extent(fs_info
, bytenr
, num_bytes
, 1);
7258 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
7259 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
7261 parent
, root_objectid
, (int)owner
,
7262 BTRFS_DROP_DELAYED_REF
, NULL
);
7264 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
7266 parent
, root_objectid
, owner
,
7268 BTRFS_DROP_DELAYED_REF
);
7274 * when we wait for progress in the block group caching, its because
7275 * our allocation attempt failed at least once. So, we must sleep
7276 * and let some progress happen before we try again.
7278 * This function will sleep at least once waiting for new free space to
7279 * show up, and then it will check the block group free space numbers
7280 * for our min num_bytes. Another option is to have it go ahead
7281 * and look in the rbtree for a free extent of a given size, but this
7284 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7285 * any of the information in this block group.
7287 static noinline
void
7288 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
7291 struct btrfs_caching_control
*caching_ctl
;
7293 caching_ctl
= get_caching_control(cache
);
7297 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
7298 (cache
->free_space_ctl
->free_space
>= num_bytes
));
7300 put_caching_control(caching_ctl
);
7304 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
7306 struct btrfs_caching_control
*caching_ctl
;
7309 caching_ctl
= get_caching_control(cache
);
7311 return (cache
->cached
== BTRFS_CACHE_ERROR
) ? -EIO
: 0;
7313 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
7314 if (cache
->cached
== BTRFS_CACHE_ERROR
)
7316 put_caching_control(caching_ctl
);
7320 int __get_raid_index(u64 flags
)
7322 if (flags
& BTRFS_BLOCK_GROUP_RAID10
)
7323 return BTRFS_RAID_RAID10
;
7324 else if (flags
& BTRFS_BLOCK_GROUP_RAID1
)
7325 return BTRFS_RAID_RAID1
;
7326 else if (flags
& BTRFS_BLOCK_GROUP_DUP
)
7327 return BTRFS_RAID_DUP
;
7328 else if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
7329 return BTRFS_RAID_RAID0
;
7330 else if (flags
& BTRFS_BLOCK_GROUP_RAID5
)
7331 return BTRFS_RAID_RAID5
;
7332 else if (flags
& BTRFS_BLOCK_GROUP_RAID6
)
7333 return BTRFS_RAID_RAID6
;
7335 return BTRFS_RAID_SINGLE
; /* BTRFS_BLOCK_GROUP_SINGLE */
7338 int get_block_group_index(struct btrfs_block_group_cache
*cache
)
7340 return __get_raid_index(cache
->flags
);
7343 static const char *btrfs_raid_type_names
[BTRFS_NR_RAID_TYPES
] = {
7344 [BTRFS_RAID_RAID10
] = "raid10",
7345 [BTRFS_RAID_RAID1
] = "raid1",
7346 [BTRFS_RAID_DUP
] = "dup",
7347 [BTRFS_RAID_RAID0
] = "raid0",
7348 [BTRFS_RAID_SINGLE
] = "single",
7349 [BTRFS_RAID_RAID5
] = "raid5",
7350 [BTRFS_RAID_RAID6
] = "raid6",
7353 static const char *get_raid_name(enum btrfs_raid_types type
)
7355 if (type
>= BTRFS_NR_RAID_TYPES
)
7358 return btrfs_raid_type_names
[type
];
7361 enum btrfs_loop_type
{
7362 LOOP_CACHING_NOWAIT
= 0,
7363 LOOP_CACHING_WAIT
= 1,
7364 LOOP_ALLOC_CHUNK
= 2,
7365 LOOP_NO_EMPTY_SIZE
= 3,
7369 btrfs_lock_block_group(struct btrfs_block_group_cache
*cache
,
7373 down_read(&cache
->data_rwsem
);
7377 btrfs_grab_block_group(struct btrfs_block_group_cache
*cache
,
7380 btrfs_get_block_group(cache
);
7382 down_read(&cache
->data_rwsem
);
7385 static struct btrfs_block_group_cache
*
7386 btrfs_lock_cluster(struct btrfs_block_group_cache
*block_group
,
7387 struct btrfs_free_cluster
*cluster
,
7390 struct btrfs_block_group_cache
*used_bg
= NULL
;
7392 spin_lock(&cluster
->refill_lock
);
7394 used_bg
= cluster
->block_group
;
7398 if (used_bg
== block_group
)
7401 btrfs_get_block_group(used_bg
);
7406 if (down_read_trylock(&used_bg
->data_rwsem
))
7409 spin_unlock(&cluster
->refill_lock
);
7411 /* We should only have one-level nested. */
7412 down_read_nested(&used_bg
->data_rwsem
, SINGLE_DEPTH_NESTING
);
7414 spin_lock(&cluster
->refill_lock
);
7415 if (used_bg
== cluster
->block_group
)
7418 up_read(&used_bg
->data_rwsem
);
7419 btrfs_put_block_group(used_bg
);
7424 btrfs_release_block_group(struct btrfs_block_group_cache
*cache
,
7428 up_read(&cache
->data_rwsem
);
7429 btrfs_put_block_group(cache
);
7433 * walks the btree of allocated extents and find a hole of a given size.
7434 * The key ins is changed to record the hole:
7435 * ins->objectid == start position
7436 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7437 * ins->offset == the size of the hole.
7438 * Any available blocks before search_start are skipped.
7440 * If there is no suitable free space, we will record the max size of
7441 * the free space extent currently.
7443 static noinline
int find_free_extent(struct btrfs_fs_info
*fs_info
,
7444 u64 ram_bytes
, u64 num_bytes
, u64 empty_size
,
7445 u64 hint_byte
, struct btrfs_key
*ins
,
7446 u64 flags
, int delalloc
)
7449 struct btrfs_root
*root
= fs_info
->extent_root
;
7450 struct btrfs_free_cluster
*last_ptr
= NULL
;
7451 struct btrfs_block_group_cache
*block_group
= NULL
;
7452 u64 search_start
= 0;
7453 u64 max_extent_size
= 0;
7454 u64 empty_cluster
= 0;
7455 struct btrfs_space_info
*space_info
;
7457 int index
= __get_raid_index(flags
);
7458 bool failed_cluster_refill
= false;
7459 bool failed_alloc
= false;
7460 bool use_cluster
= true;
7461 bool have_caching_bg
= false;
7462 bool orig_have_caching_bg
= false;
7463 bool full_search
= false;
7465 WARN_ON(num_bytes
< fs_info
->sectorsize
);
7466 ins
->type
= BTRFS_EXTENT_ITEM_KEY
;
7470 trace_find_free_extent(fs_info
, num_bytes
, empty_size
, flags
);
7472 space_info
= __find_space_info(fs_info
, flags
);
7474 btrfs_err(fs_info
, "No space info for %llu", flags
);
7479 * If our free space is heavily fragmented we may not be able to make
7480 * big contiguous allocations, so instead of doing the expensive search
7481 * for free space, simply return ENOSPC with our max_extent_size so we
7482 * can go ahead and search for a more manageable chunk.
7484 * If our max_extent_size is large enough for our allocation simply
7485 * disable clustering since we will likely not be able to find enough
7486 * space to create a cluster and induce latency trying.
7488 if (unlikely(space_info
->max_extent_size
)) {
7489 spin_lock(&space_info
->lock
);
7490 if (space_info
->max_extent_size
&&
7491 num_bytes
> space_info
->max_extent_size
) {
7492 ins
->offset
= space_info
->max_extent_size
;
7493 spin_unlock(&space_info
->lock
);
7495 } else if (space_info
->max_extent_size
) {
7496 use_cluster
= false;
7498 spin_unlock(&space_info
->lock
);
7501 last_ptr
= fetch_cluster_info(fs_info
, space_info
, &empty_cluster
);
7503 spin_lock(&last_ptr
->lock
);
7504 if (last_ptr
->block_group
)
7505 hint_byte
= last_ptr
->window_start
;
7506 if (last_ptr
->fragmented
) {
7508 * We still set window_start so we can keep track of the
7509 * last place we found an allocation to try and save
7512 hint_byte
= last_ptr
->window_start
;
7513 use_cluster
= false;
7515 spin_unlock(&last_ptr
->lock
);
7518 search_start
= max(search_start
, first_logical_byte(fs_info
, 0));
7519 search_start
= max(search_start
, hint_byte
);
7520 if (search_start
== hint_byte
) {
7521 block_group
= btrfs_lookup_block_group(fs_info
, search_start
);
7523 * we don't want to use the block group if it doesn't match our
7524 * allocation bits, or if its not cached.
7526 * However if we are re-searching with an ideal block group
7527 * picked out then we don't care that the block group is cached.
7529 if (block_group
&& block_group_bits(block_group
, flags
) &&
7530 block_group
->cached
!= BTRFS_CACHE_NO
) {
7531 down_read(&space_info
->groups_sem
);
7532 if (list_empty(&block_group
->list
) ||
7535 * someone is removing this block group,
7536 * we can't jump into the have_block_group
7537 * target because our list pointers are not
7540 btrfs_put_block_group(block_group
);
7541 up_read(&space_info
->groups_sem
);
7543 index
= get_block_group_index(block_group
);
7544 btrfs_lock_block_group(block_group
, delalloc
);
7545 goto have_block_group
;
7547 } else if (block_group
) {
7548 btrfs_put_block_group(block_group
);
7552 have_caching_bg
= false;
7553 if (index
== 0 || index
== __get_raid_index(flags
))
7555 down_read(&space_info
->groups_sem
);
7556 list_for_each_entry(block_group
, &space_info
->block_groups
[index
],
7561 btrfs_grab_block_group(block_group
, delalloc
);
7562 search_start
= block_group
->key
.objectid
;
7565 * this can happen if we end up cycling through all the
7566 * raid types, but we want to make sure we only allocate
7567 * for the proper type.
7569 if (!block_group_bits(block_group
, flags
)) {
7570 u64 extra
= BTRFS_BLOCK_GROUP_DUP
|
7571 BTRFS_BLOCK_GROUP_RAID1
|
7572 BTRFS_BLOCK_GROUP_RAID5
|
7573 BTRFS_BLOCK_GROUP_RAID6
|
7574 BTRFS_BLOCK_GROUP_RAID10
;
7577 * if they asked for extra copies and this block group
7578 * doesn't provide them, bail. This does allow us to
7579 * fill raid0 from raid1.
7581 if ((flags
& extra
) && !(block_group
->flags
& extra
))
7586 cached
= block_group_cache_done(block_group
);
7587 if (unlikely(!cached
)) {
7588 have_caching_bg
= true;
7589 ret
= cache_block_group(block_group
, 0);
7594 if (unlikely(block_group
->cached
== BTRFS_CACHE_ERROR
))
7596 if (unlikely(block_group
->ro
))
7600 * Ok we want to try and use the cluster allocator, so
7603 if (last_ptr
&& use_cluster
) {
7604 struct btrfs_block_group_cache
*used_block_group
;
7605 unsigned long aligned_cluster
;
7607 * the refill lock keeps out other
7608 * people trying to start a new cluster
7610 used_block_group
= btrfs_lock_cluster(block_group
,
7613 if (!used_block_group
)
7614 goto refill_cluster
;
7616 if (used_block_group
!= block_group
&&
7617 (used_block_group
->ro
||
7618 !block_group_bits(used_block_group
, flags
)))
7619 goto release_cluster
;
7621 offset
= btrfs_alloc_from_cluster(used_block_group
,
7624 used_block_group
->key
.objectid
,
7627 /* we have a block, we're done */
7628 spin_unlock(&last_ptr
->refill_lock
);
7629 trace_btrfs_reserve_extent_cluster(fs_info
,
7631 search_start
, num_bytes
);
7632 if (used_block_group
!= block_group
) {
7633 btrfs_release_block_group(block_group
,
7635 block_group
= used_block_group
;
7640 WARN_ON(last_ptr
->block_group
!= used_block_group
);
7642 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7643 * set up a new clusters, so lets just skip it
7644 * and let the allocator find whatever block
7645 * it can find. If we reach this point, we
7646 * will have tried the cluster allocator
7647 * plenty of times and not have found
7648 * anything, so we are likely way too
7649 * fragmented for the clustering stuff to find
7652 * However, if the cluster is taken from the
7653 * current block group, release the cluster
7654 * first, so that we stand a better chance of
7655 * succeeding in the unclustered
7657 if (loop
>= LOOP_NO_EMPTY_SIZE
&&
7658 used_block_group
!= block_group
) {
7659 spin_unlock(&last_ptr
->refill_lock
);
7660 btrfs_release_block_group(used_block_group
,
7662 goto unclustered_alloc
;
7666 * this cluster didn't work out, free it and
7669 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
7671 if (used_block_group
!= block_group
)
7672 btrfs_release_block_group(used_block_group
,
7675 if (loop
>= LOOP_NO_EMPTY_SIZE
) {
7676 spin_unlock(&last_ptr
->refill_lock
);
7677 goto unclustered_alloc
;
7680 aligned_cluster
= max_t(unsigned long,
7681 empty_cluster
+ empty_size
,
7682 block_group
->full_stripe_len
);
7684 /* allocate a cluster in this block group */
7685 ret
= btrfs_find_space_cluster(fs_info
, block_group
,
7686 last_ptr
, search_start
,
7691 * now pull our allocation out of this
7694 offset
= btrfs_alloc_from_cluster(block_group
,
7700 /* we found one, proceed */
7701 spin_unlock(&last_ptr
->refill_lock
);
7702 trace_btrfs_reserve_extent_cluster(fs_info
,
7703 block_group
, search_start
,
7707 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
7708 && !failed_cluster_refill
) {
7709 spin_unlock(&last_ptr
->refill_lock
);
7711 failed_cluster_refill
= true;
7712 wait_block_group_cache_progress(block_group
,
7713 num_bytes
+ empty_cluster
+ empty_size
);
7714 goto have_block_group
;
7718 * at this point we either didn't find a cluster
7719 * or we weren't able to allocate a block from our
7720 * cluster. Free the cluster we've been trying
7721 * to use, and go to the next block group
7723 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
7724 spin_unlock(&last_ptr
->refill_lock
);
7730 * We are doing an unclustered alloc, set the fragmented flag so
7731 * we don't bother trying to setup a cluster again until we get
7734 if (unlikely(last_ptr
)) {
7735 spin_lock(&last_ptr
->lock
);
7736 last_ptr
->fragmented
= 1;
7737 spin_unlock(&last_ptr
->lock
);
7740 struct btrfs_free_space_ctl
*ctl
=
7741 block_group
->free_space_ctl
;
7743 spin_lock(&ctl
->tree_lock
);
7744 if (ctl
->free_space
<
7745 num_bytes
+ empty_cluster
+ empty_size
) {
7746 if (ctl
->free_space
> max_extent_size
)
7747 max_extent_size
= ctl
->free_space
;
7748 spin_unlock(&ctl
->tree_lock
);
7751 spin_unlock(&ctl
->tree_lock
);
7754 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
7755 num_bytes
, empty_size
,
7758 * If we didn't find a chunk, and we haven't failed on this
7759 * block group before, and this block group is in the middle of
7760 * caching and we are ok with waiting, then go ahead and wait
7761 * for progress to be made, and set failed_alloc to true.
7763 * If failed_alloc is true then we've already waited on this
7764 * block group once and should move on to the next block group.
7766 if (!offset
&& !failed_alloc
&& !cached
&&
7767 loop
> LOOP_CACHING_NOWAIT
) {
7768 wait_block_group_cache_progress(block_group
,
7769 num_bytes
+ empty_size
);
7770 failed_alloc
= true;
7771 goto have_block_group
;
7772 } else if (!offset
) {
7776 search_start
= ALIGN(offset
, fs_info
->stripesize
);
7778 /* move on to the next group */
7779 if (search_start
+ num_bytes
>
7780 block_group
->key
.objectid
+ block_group
->key
.offset
) {
7781 btrfs_add_free_space(block_group
, offset
, num_bytes
);
7785 if (offset
< search_start
)
7786 btrfs_add_free_space(block_group
, offset
,
7787 search_start
- offset
);
7788 BUG_ON(offset
> search_start
);
7790 ret
= btrfs_add_reserved_bytes(block_group
, ram_bytes
,
7791 num_bytes
, delalloc
);
7792 if (ret
== -EAGAIN
) {
7793 btrfs_add_free_space(block_group
, offset
, num_bytes
);
7796 btrfs_inc_block_group_reservations(block_group
);
7798 /* we are all good, lets return */
7799 ins
->objectid
= search_start
;
7800 ins
->offset
= num_bytes
;
7802 trace_btrfs_reserve_extent(fs_info
, block_group
,
7803 search_start
, num_bytes
);
7804 btrfs_release_block_group(block_group
, delalloc
);
7807 failed_cluster_refill
= false;
7808 failed_alloc
= false;
7809 BUG_ON(index
!= get_block_group_index(block_group
));
7810 btrfs_release_block_group(block_group
, delalloc
);
7812 up_read(&space_info
->groups_sem
);
7814 if ((loop
== LOOP_CACHING_NOWAIT
) && have_caching_bg
7815 && !orig_have_caching_bg
)
7816 orig_have_caching_bg
= true;
7818 if (!ins
->objectid
&& loop
>= LOOP_CACHING_WAIT
&& have_caching_bg
)
7821 if (!ins
->objectid
&& ++index
< BTRFS_NR_RAID_TYPES
)
7825 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7826 * caching kthreads as we move along
7827 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7828 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7829 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7832 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
) {
7834 if (loop
== LOOP_CACHING_NOWAIT
) {
7836 * We want to skip the LOOP_CACHING_WAIT step if we
7837 * don't have any uncached bgs and we've already done a
7838 * full search through.
7840 if (orig_have_caching_bg
|| !full_search
)
7841 loop
= LOOP_CACHING_WAIT
;
7843 loop
= LOOP_ALLOC_CHUNK
;
7848 if (loop
== LOOP_ALLOC_CHUNK
) {
7849 struct btrfs_trans_handle
*trans
;
7852 trans
= current
->journal_info
;
7856 trans
= btrfs_join_transaction(root
);
7858 if (IS_ERR(trans
)) {
7859 ret
= PTR_ERR(trans
);
7863 ret
= do_chunk_alloc(trans
, fs_info
, flags
,
7867 * If we can't allocate a new chunk we've already looped
7868 * through at least once, move on to the NO_EMPTY_SIZE
7872 loop
= LOOP_NO_EMPTY_SIZE
;
7875 * Do not bail out on ENOSPC since we
7876 * can do more things.
7878 if (ret
< 0 && ret
!= -ENOSPC
)
7879 btrfs_abort_transaction(trans
, ret
);
7883 btrfs_end_transaction(trans
);
7888 if (loop
== LOOP_NO_EMPTY_SIZE
) {
7890 * Don't loop again if we already have no empty_size and
7893 if (empty_size
== 0 &&
7894 empty_cluster
== 0) {
7903 } else if (!ins
->objectid
) {
7905 } else if (ins
->objectid
) {
7906 if (!use_cluster
&& last_ptr
) {
7907 spin_lock(&last_ptr
->lock
);
7908 last_ptr
->window_start
= ins
->objectid
;
7909 spin_unlock(&last_ptr
->lock
);
7914 if (ret
== -ENOSPC
) {
7915 spin_lock(&space_info
->lock
);
7916 space_info
->max_extent_size
= max_extent_size
;
7917 spin_unlock(&space_info
->lock
);
7918 ins
->offset
= max_extent_size
;
7923 static void dump_space_info(struct btrfs_fs_info
*fs_info
,
7924 struct btrfs_space_info
*info
, u64 bytes
,
7925 int dump_block_groups
)
7927 struct btrfs_block_group_cache
*cache
;
7930 spin_lock(&info
->lock
);
7931 btrfs_info(fs_info
, "space_info %llu has %llu free, is %sfull",
7933 info
->total_bytes
- btrfs_space_info_used(info
, true),
7934 info
->full
? "" : "not ");
7936 "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu",
7937 info
->total_bytes
, info
->bytes_used
, info
->bytes_pinned
,
7938 info
->bytes_reserved
, info
->bytes_may_use
,
7939 info
->bytes_readonly
);
7940 spin_unlock(&info
->lock
);
7942 if (!dump_block_groups
)
7945 down_read(&info
->groups_sem
);
7947 list_for_each_entry(cache
, &info
->block_groups
[index
], list
) {
7948 spin_lock(&cache
->lock
);
7950 "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s",
7951 cache
->key
.objectid
, cache
->key
.offset
,
7952 btrfs_block_group_used(&cache
->item
), cache
->pinned
,
7953 cache
->reserved
, cache
->ro
? "[readonly]" : "");
7954 btrfs_dump_free_space(cache
, bytes
);
7955 spin_unlock(&cache
->lock
);
7957 if (++index
< BTRFS_NR_RAID_TYPES
)
7959 up_read(&info
->groups_sem
);
7962 int btrfs_reserve_extent(struct btrfs_root
*root
, u64 ram_bytes
,
7963 u64 num_bytes
, u64 min_alloc_size
,
7964 u64 empty_size
, u64 hint_byte
,
7965 struct btrfs_key
*ins
, int is_data
, int delalloc
)
7967 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7968 bool final_tried
= num_bytes
== min_alloc_size
;
7972 flags
= get_alloc_profile_by_root(root
, is_data
);
7974 WARN_ON(num_bytes
< fs_info
->sectorsize
);
7975 ret
= find_free_extent(fs_info
, ram_bytes
, num_bytes
, empty_size
,
7976 hint_byte
, ins
, flags
, delalloc
);
7977 if (!ret
&& !is_data
) {
7978 btrfs_dec_block_group_reservations(fs_info
, ins
->objectid
);
7979 } else if (ret
== -ENOSPC
) {
7980 if (!final_tried
&& ins
->offset
) {
7981 num_bytes
= min(num_bytes
>> 1, ins
->offset
);
7982 num_bytes
= round_down(num_bytes
,
7983 fs_info
->sectorsize
);
7984 num_bytes
= max(num_bytes
, min_alloc_size
);
7985 ram_bytes
= num_bytes
;
7986 if (num_bytes
== min_alloc_size
)
7989 } else if (btrfs_test_opt(fs_info
, ENOSPC_DEBUG
)) {
7990 struct btrfs_space_info
*sinfo
;
7992 sinfo
= __find_space_info(fs_info
, flags
);
7994 "allocation failed flags %llu, wanted %llu",
7997 dump_space_info(fs_info
, sinfo
, num_bytes
, 1);
8004 static int __btrfs_free_reserved_extent(struct btrfs_fs_info
*fs_info
,
8006 int pin
, int delalloc
)
8008 struct btrfs_block_group_cache
*cache
;
8011 cache
= btrfs_lookup_block_group(fs_info
, start
);
8013 btrfs_err(fs_info
, "Unable to find block group for %llu",
8019 pin_down_extent(fs_info
, cache
, start
, len
, 1);
8021 if (btrfs_test_opt(fs_info
, DISCARD
))
8022 ret
= btrfs_discard_extent(fs_info
, start
, len
, NULL
);
8023 btrfs_add_free_space(cache
, start
, len
);
8024 btrfs_free_reserved_bytes(cache
, len
, delalloc
);
8025 trace_btrfs_reserved_extent_free(fs_info
, start
, len
);
8028 btrfs_put_block_group(cache
);
8032 int btrfs_free_reserved_extent(struct btrfs_fs_info
*fs_info
,
8033 u64 start
, u64 len
, int delalloc
)
8035 return __btrfs_free_reserved_extent(fs_info
, start
, len
, 0, delalloc
);
8038 int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info
*fs_info
,
8041 return __btrfs_free_reserved_extent(fs_info
, start
, len
, 1, 0);
8044 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
8045 struct btrfs_fs_info
*fs_info
,
8046 u64 parent
, u64 root_objectid
,
8047 u64 flags
, u64 owner
, u64 offset
,
8048 struct btrfs_key
*ins
, int ref_mod
)
8051 struct btrfs_extent_item
*extent_item
;
8052 struct btrfs_extent_inline_ref
*iref
;
8053 struct btrfs_path
*path
;
8054 struct extent_buffer
*leaf
;
8059 type
= BTRFS_SHARED_DATA_REF_KEY
;
8061 type
= BTRFS_EXTENT_DATA_REF_KEY
;
8063 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
8065 path
= btrfs_alloc_path();
8069 path
->leave_spinning
= 1;
8070 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
8073 btrfs_free_path(path
);
8077 leaf
= path
->nodes
[0];
8078 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
8079 struct btrfs_extent_item
);
8080 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
8081 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
8082 btrfs_set_extent_flags(leaf
, extent_item
,
8083 flags
| BTRFS_EXTENT_FLAG_DATA
);
8085 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
8086 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
8088 struct btrfs_shared_data_ref
*ref
;
8089 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
8090 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
8091 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
8093 struct btrfs_extent_data_ref
*ref
;
8094 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
8095 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
8096 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
8097 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
8098 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
8101 btrfs_mark_buffer_dirty(path
->nodes
[0]);
8102 btrfs_free_path(path
);
8104 ret
= remove_from_free_space_tree(trans
, fs_info
, ins
->objectid
,
8109 ret
= update_block_group(trans
, fs_info
, ins
->objectid
, ins
->offset
, 1);
8110 if (ret
) { /* -ENOENT, logic error */
8111 btrfs_err(fs_info
, "update block group failed for %llu %llu",
8112 ins
->objectid
, ins
->offset
);
8115 trace_btrfs_reserved_extent_alloc(fs_info
, ins
->objectid
, ins
->offset
);
8119 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
8120 struct btrfs_fs_info
*fs_info
,
8121 u64 parent
, u64 root_objectid
,
8122 u64 flags
, struct btrfs_disk_key
*key
,
8123 int level
, struct btrfs_key
*ins
)
8126 struct btrfs_extent_item
*extent_item
;
8127 struct btrfs_tree_block_info
*block_info
;
8128 struct btrfs_extent_inline_ref
*iref
;
8129 struct btrfs_path
*path
;
8130 struct extent_buffer
*leaf
;
8131 u32 size
= sizeof(*extent_item
) + sizeof(*iref
);
8132 u64 num_bytes
= ins
->offset
;
8133 bool skinny_metadata
= btrfs_fs_incompat(fs_info
, SKINNY_METADATA
);
8135 if (!skinny_metadata
)
8136 size
+= sizeof(*block_info
);
8138 path
= btrfs_alloc_path();
8140 btrfs_free_and_pin_reserved_extent(fs_info
, ins
->objectid
,
8145 path
->leave_spinning
= 1;
8146 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
8149 btrfs_free_path(path
);
8150 btrfs_free_and_pin_reserved_extent(fs_info
, ins
->objectid
,
8155 leaf
= path
->nodes
[0];
8156 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
8157 struct btrfs_extent_item
);
8158 btrfs_set_extent_refs(leaf
, extent_item
, 1);
8159 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
8160 btrfs_set_extent_flags(leaf
, extent_item
,
8161 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
8163 if (skinny_metadata
) {
8164 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
8165 num_bytes
= fs_info
->nodesize
;
8167 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
8168 btrfs_set_tree_block_key(leaf
, block_info
, key
);
8169 btrfs_set_tree_block_level(leaf
, block_info
, level
);
8170 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
8174 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
8175 btrfs_set_extent_inline_ref_type(leaf
, iref
,
8176 BTRFS_SHARED_BLOCK_REF_KEY
);
8177 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
8179 btrfs_set_extent_inline_ref_type(leaf
, iref
,
8180 BTRFS_TREE_BLOCK_REF_KEY
);
8181 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
8184 btrfs_mark_buffer_dirty(leaf
);
8185 btrfs_free_path(path
);
8187 ret
= remove_from_free_space_tree(trans
, fs_info
, ins
->objectid
,
8192 ret
= update_block_group(trans
, fs_info
, ins
->objectid
,
8193 fs_info
->nodesize
, 1);
8194 if (ret
) { /* -ENOENT, logic error */
8195 btrfs_err(fs_info
, "update block group failed for %llu %llu",
8196 ins
->objectid
, ins
->offset
);
8200 trace_btrfs_reserved_extent_alloc(fs_info
, ins
->objectid
,
8205 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
8206 u64 root_objectid
, u64 owner
,
8207 u64 offset
, u64 ram_bytes
,
8208 struct btrfs_key
*ins
)
8210 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
8213 BUG_ON(root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
8215 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, ins
->objectid
,
8217 root_objectid
, owner
, offset
,
8218 ram_bytes
, BTRFS_ADD_DELAYED_EXTENT
);
8223 * this is used by the tree logging recovery code. It records that
8224 * an extent has been allocated and makes sure to clear the free
8225 * space cache bits as well
8227 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
8228 struct btrfs_fs_info
*fs_info
,
8229 u64 root_objectid
, u64 owner
, u64 offset
,
8230 struct btrfs_key
*ins
)
8233 struct btrfs_block_group_cache
*block_group
;
8234 struct btrfs_space_info
*space_info
;
8237 * Mixed block groups will exclude before processing the log so we only
8238 * need to do the exclude dance if this fs isn't mixed.
8240 if (!btrfs_fs_incompat(fs_info
, MIXED_GROUPS
)) {
8241 ret
= __exclude_logged_extent(fs_info
, ins
->objectid
,
8247 block_group
= btrfs_lookup_block_group(fs_info
, ins
->objectid
);
8251 space_info
= block_group
->space_info
;
8252 spin_lock(&space_info
->lock
);
8253 spin_lock(&block_group
->lock
);
8254 space_info
->bytes_reserved
+= ins
->offset
;
8255 block_group
->reserved
+= ins
->offset
;
8256 spin_unlock(&block_group
->lock
);
8257 spin_unlock(&space_info
->lock
);
8259 ret
= alloc_reserved_file_extent(trans
, fs_info
, 0, root_objectid
,
8260 0, owner
, offset
, ins
, 1);
8261 btrfs_put_block_group(block_group
);
8265 static struct extent_buffer
*
8266 btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
8267 u64 bytenr
, int level
)
8269 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8270 struct extent_buffer
*buf
;
8272 buf
= btrfs_find_create_tree_block(fs_info
, bytenr
);
8276 btrfs_set_header_generation(buf
, trans
->transid
);
8277 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, buf
, level
);
8278 btrfs_tree_lock(buf
);
8279 clean_tree_block(fs_info
, buf
);
8280 clear_bit(EXTENT_BUFFER_STALE
, &buf
->bflags
);
8282 btrfs_set_lock_blocking(buf
);
8283 set_extent_buffer_uptodate(buf
);
8285 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
8286 buf
->log_index
= root
->log_transid
% 2;
8288 * we allow two log transactions at a time, use different
8289 * EXENT bit to differentiate dirty pages.
8291 if (buf
->log_index
== 0)
8292 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
8293 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
8295 set_extent_new(&root
->dirty_log_pages
, buf
->start
,
8296 buf
->start
+ buf
->len
- 1);
8298 buf
->log_index
= -1;
8299 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
8300 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
8302 trans
->dirty
= true;
8303 /* this returns a buffer locked for blocking */
8307 static struct btrfs_block_rsv
*
8308 use_block_rsv(struct btrfs_trans_handle
*trans
,
8309 struct btrfs_root
*root
, u32 blocksize
)
8311 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8312 struct btrfs_block_rsv
*block_rsv
;
8313 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
8315 bool global_updated
= false;
8317 block_rsv
= get_block_rsv(trans
, root
);
8319 if (unlikely(block_rsv
->size
== 0))
8322 ret
= block_rsv_use_bytes(block_rsv
, blocksize
);
8326 if (block_rsv
->failfast
)
8327 return ERR_PTR(ret
);
8329 if (block_rsv
->type
== BTRFS_BLOCK_RSV_GLOBAL
&& !global_updated
) {
8330 global_updated
= true;
8331 update_global_block_rsv(fs_info
);
8335 if (btrfs_test_opt(fs_info
, ENOSPC_DEBUG
)) {
8336 static DEFINE_RATELIMIT_STATE(_rs
,
8337 DEFAULT_RATELIMIT_INTERVAL
* 10,
8338 /*DEFAULT_RATELIMIT_BURST*/ 1);
8339 if (__ratelimit(&_rs
))
8341 "BTRFS: block rsv returned %d\n", ret
);
8344 ret
= reserve_metadata_bytes(root
, block_rsv
, blocksize
,
8345 BTRFS_RESERVE_NO_FLUSH
);
8349 * If we couldn't reserve metadata bytes try and use some from
8350 * the global reserve if its space type is the same as the global
8353 if (block_rsv
->type
!= BTRFS_BLOCK_RSV_GLOBAL
&&
8354 block_rsv
->space_info
== global_rsv
->space_info
) {
8355 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
8359 return ERR_PTR(ret
);
8362 static void unuse_block_rsv(struct btrfs_fs_info
*fs_info
,
8363 struct btrfs_block_rsv
*block_rsv
, u32 blocksize
)
8365 block_rsv_add_bytes(block_rsv
, blocksize
, 0);
8366 block_rsv_release_bytes(fs_info
, block_rsv
, NULL
, 0);
8370 * finds a free extent and does all the dirty work required for allocation
8371 * returns the tree buffer or an ERR_PTR on error.
8373 struct extent_buffer
*btrfs_alloc_tree_block(struct btrfs_trans_handle
*trans
,
8374 struct btrfs_root
*root
,
8375 u64 parent
, u64 root_objectid
,
8376 const struct btrfs_disk_key
*key
,
8377 int level
, u64 hint
,
8380 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8381 struct btrfs_key ins
;
8382 struct btrfs_block_rsv
*block_rsv
;
8383 struct extent_buffer
*buf
;
8384 struct btrfs_delayed_extent_op
*extent_op
;
8387 u32 blocksize
= fs_info
->nodesize
;
8388 bool skinny_metadata
= btrfs_fs_incompat(fs_info
, SKINNY_METADATA
);
8390 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8391 if (btrfs_is_testing(fs_info
)) {
8392 buf
= btrfs_init_new_buffer(trans
, root
, root
->alloc_bytenr
,
8395 root
->alloc_bytenr
+= blocksize
;
8400 block_rsv
= use_block_rsv(trans
, root
, blocksize
);
8401 if (IS_ERR(block_rsv
))
8402 return ERR_CAST(block_rsv
);
8404 ret
= btrfs_reserve_extent(root
, blocksize
, blocksize
, blocksize
,
8405 empty_size
, hint
, &ins
, 0, 0);
8409 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
, level
);
8412 goto out_free_reserved
;
8415 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
8417 parent
= ins
.objectid
;
8418 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
8422 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
8423 extent_op
= btrfs_alloc_delayed_extent_op();
8429 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
8431 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
8432 extent_op
->flags_to_set
= flags
;
8433 extent_op
->update_key
= skinny_metadata
? false : true;
8434 extent_op
->update_flags
= true;
8435 extent_op
->is_data
= false;
8436 extent_op
->level
= level
;
8438 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
,
8439 ins
.objectid
, ins
.offset
,
8440 parent
, root_objectid
, level
,
8441 BTRFS_ADD_DELAYED_EXTENT
,
8444 goto out_free_delayed
;
8449 btrfs_free_delayed_extent_op(extent_op
);
8451 free_extent_buffer(buf
);
8453 btrfs_free_reserved_extent(fs_info
, ins
.objectid
, ins
.offset
, 0);
8455 unuse_block_rsv(fs_info
, block_rsv
, blocksize
);
8456 return ERR_PTR(ret
);
8459 struct walk_control
{
8460 u64 refs
[BTRFS_MAX_LEVEL
];
8461 u64 flags
[BTRFS_MAX_LEVEL
];
8462 struct btrfs_key update_progress
;
8473 #define DROP_REFERENCE 1
8474 #define UPDATE_BACKREF 2
8476 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
8477 struct btrfs_root
*root
,
8478 struct walk_control
*wc
,
8479 struct btrfs_path
*path
)
8481 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8487 struct btrfs_key key
;
8488 struct extent_buffer
*eb
;
8493 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
8494 wc
->reada_count
= wc
->reada_count
* 2 / 3;
8495 wc
->reada_count
= max(wc
->reada_count
, 2);
8497 wc
->reada_count
= wc
->reada_count
* 3 / 2;
8498 wc
->reada_count
= min_t(int, wc
->reada_count
,
8499 BTRFS_NODEPTRS_PER_BLOCK(fs_info
));
8502 eb
= path
->nodes
[wc
->level
];
8503 nritems
= btrfs_header_nritems(eb
);
8505 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
8506 if (nread
>= wc
->reada_count
)
8510 bytenr
= btrfs_node_blockptr(eb
, slot
);
8511 generation
= btrfs_node_ptr_generation(eb
, slot
);
8513 if (slot
== path
->slots
[wc
->level
])
8516 if (wc
->stage
== UPDATE_BACKREF
&&
8517 generation
<= root
->root_key
.offset
)
8520 /* We don't lock the tree block, it's OK to be racy here */
8521 ret
= btrfs_lookup_extent_info(trans
, fs_info
, bytenr
,
8522 wc
->level
- 1, 1, &refs
,
8524 /* We don't care about errors in readahead. */
8529 if (wc
->stage
== DROP_REFERENCE
) {
8533 if (wc
->level
== 1 &&
8534 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8536 if (!wc
->update_ref
||
8537 generation
<= root
->root_key
.offset
)
8539 btrfs_node_key_to_cpu(eb
, &key
, slot
);
8540 ret
= btrfs_comp_cpu_keys(&key
,
8541 &wc
->update_progress
);
8545 if (wc
->level
== 1 &&
8546 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8550 readahead_tree_block(fs_info
, bytenr
);
8553 wc
->reada_slot
= slot
;
8557 * helper to process tree block while walking down the tree.
8559 * when wc->stage == UPDATE_BACKREF, this function updates
8560 * back refs for pointers in the block.
8562 * NOTE: return value 1 means we should stop walking down.
8564 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
8565 struct btrfs_root
*root
,
8566 struct btrfs_path
*path
,
8567 struct walk_control
*wc
, int lookup_info
)
8569 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8570 int level
= wc
->level
;
8571 struct extent_buffer
*eb
= path
->nodes
[level
];
8572 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
8575 if (wc
->stage
== UPDATE_BACKREF
&&
8576 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
8580 * when reference count of tree block is 1, it won't increase
8581 * again. once full backref flag is set, we never clear it.
8584 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
8585 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
8586 BUG_ON(!path
->locks
[level
]);
8587 ret
= btrfs_lookup_extent_info(trans
, fs_info
,
8588 eb
->start
, level
, 1,
8591 BUG_ON(ret
== -ENOMEM
);
8594 BUG_ON(wc
->refs
[level
] == 0);
8597 if (wc
->stage
== DROP_REFERENCE
) {
8598 if (wc
->refs
[level
] > 1)
8601 if (path
->locks
[level
] && !wc
->keep_locks
) {
8602 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8603 path
->locks
[level
] = 0;
8608 /* wc->stage == UPDATE_BACKREF */
8609 if (!(wc
->flags
[level
] & flag
)) {
8610 BUG_ON(!path
->locks
[level
]);
8611 ret
= btrfs_inc_ref(trans
, root
, eb
, 1);
8612 BUG_ON(ret
); /* -ENOMEM */
8613 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
8614 BUG_ON(ret
); /* -ENOMEM */
8615 ret
= btrfs_set_disk_extent_flags(trans
, fs_info
, eb
->start
,
8617 btrfs_header_level(eb
), 0);
8618 BUG_ON(ret
); /* -ENOMEM */
8619 wc
->flags
[level
] |= flag
;
8623 * the block is shared by multiple trees, so it's not good to
8624 * keep the tree lock
8626 if (path
->locks
[level
] && level
> 0) {
8627 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8628 path
->locks
[level
] = 0;
8634 * helper to process tree block pointer.
8636 * when wc->stage == DROP_REFERENCE, this function checks
8637 * reference count of the block pointed to. if the block
8638 * is shared and we need update back refs for the subtree
8639 * rooted at the block, this function changes wc->stage to
8640 * UPDATE_BACKREF. if the block is shared and there is no
8641 * need to update back, this function drops the reference
8644 * NOTE: return value 1 means we should stop walking down.
8646 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
8647 struct btrfs_root
*root
,
8648 struct btrfs_path
*path
,
8649 struct walk_control
*wc
, int *lookup_info
)
8651 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8656 struct btrfs_key key
;
8657 struct extent_buffer
*next
;
8658 int level
= wc
->level
;
8661 bool need_account
= false;
8663 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
8664 path
->slots
[level
]);
8666 * if the lower level block was created before the snapshot
8667 * was created, we know there is no need to update back refs
8670 if (wc
->stage
== UPDATE_BACKREF
&&
8671 generation
<= root
->root_key
.offset
) {
8676 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
8677 blocksize
= fs_info
->nodesize
;
8679 next
= find_extent_buffer(fs_info
, bytenr
);
8681 next
= btrfs_find_create_tree_block(fs_info
, bytenr
);
8683 return PTR_ERR(next
);
8685 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, next
,
8689 btrfs_tree_lock(next
);
8690 btrfs_set_lock_blocking(next
);
8692 ret
= btrfs_lookup_extent_info(trans
, fs_info
, bytenr
, level
- 1, 1,
8693 &wc
->refs
[level
- 1],
8694 &wc
->flags
[level
- 1]);
8698 if (unlikely(wc
->refs
[level
- 1] == 0)) {
8699 btrfs_err(fs_info
, "Missing references.");
8705 if (wc
->stage
== DROP_REFERENCE
) {
8706 if (wc
->refs
[level
- 1] > 1) {
8707 need_account
= true;
8709 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8712 if (!wc
->update_ref
||
8713 generation
<= root
->root_key
.offset
)
8716 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
8717 path
->slots
[level
]);
8718 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
8722 wc
->stage
= UPDATE_BACKREF
;
8723 wc
->shared_level
= level
- 1;
8727 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8731 if (!btrfs_buffer_uptodate(next
, generation
, 0)) {
8732 btrfs_tree_unlock(next
);
8733 free_extent_buffer(next
);
8739 if (reada
&& level
== 1)
8740 reada_walk_down(trans
, root
, wc
, path
);
8741 next
= read_tree_block(fs_info
, bytenr
, generation
);
8743 return PTR_ERR(next
);
8744 } else if (!extent_buffer_uptodate(next
)) {
8745 free_extent_buffer(next
);
8748 btrfs_tree_lock(next
);
8749 btrfs_set_lock_blocking(next
);
8753 ASSERT(level
== btrfs_header_level(next
));
8754 if (level
!= btrfs_header_level(next
)) {
8755 btrfs_err(root
->fs_info
, "mismatched level");
8759 path
->nodes
[level
] = next
;
8760 path
->slots
[level
] = 0;
8761 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8767 wc
->refs
[level
- 1] = 0;
8768 wc
->flags
[level
- 1] = 0;
8769 if (wc
->stage
== DROP_REFERENCE
) {
8770 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
8771 parent
= path
->nodes
[level
]->start
;
8773 ASSERT(root
->root_key
.objectid
==
8774 btrfs_header_owner(path
->nodes
[level
]));
8775 if (root
->root_key
.objectid
!=
8776 btrfs_header_owner(path
->nodes
[level
])) {
8777 btrfs_err(root
->fs_info
,
8778 "mismatched block owner");
8786 ret
= btrfs_qgroup_trace_subtree(trans
, root
, next
,
8787 generation
, level
- 1);
8789 btrfs_err_rl(fs_info
,
8790 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
8794 ret
= btrfs_free_extent(trans
, fs_info
, bytenr
, blocksize
,
8795 parent
, root
->root_key
.objectid
,
8805 btrfs_tree_unlock(next
);
8806 free_extent_buffer(next
);
8812 * helper to process tree block while walking up the tree.
8814 * when wc->stage == DROP_REFERENCE, this function drops
8815 * reference count on the block.
8817 * when wc->stage == UPDATE_BACKREF, this function changes
8818 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8819 * to UPDATE_BACKREF previously while processing the block.
8821 * NOTE: return value 1 means we should stop walking up.
8823 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
8824 struct btrfs_root
*root
,
8825 struct btrfs_path
*path
,
8826 struct walk_control
*wc
)
8828 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8830 int level
= wc
->level
;
8831 struct extent_buffer
*eb
= path
->nodes
[level
];
8834 if (wc
->stage
== UPDATE_BACKREF
) {
8835 BUG_ON(wc
->shared_level
< level
);
8836 if (level
< wc
->shared_level
)
8839 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
8843 wc
->stage
= DROP_REFERENCE
;
8844 wc
->shared_level
= -1;
8845 path
->slots
[level
] = 0;
8848 * check reference count again if the block isn't locked.
8849 * we should start walking down the tree again if reference
8852 if (!path
->locks
[level
]) {
8854 btrfs_tree_lock(eb
);
8855 btrfs_set_lock_blocking(eb
);
8856 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8858 ret
= btrfs_lookup_extent_info(trans
, fs_info
,
8859 eb
->start
, level
, 1,
8863 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8864 path
->locks
[level
] = 0;
8867 BUG_ON(wc
->refs
[level
] == 0);
8868 if (wc
->refs
[level
] == 1) {
8869 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8870 path
->locks
[level
] = 0;
8876 /* wc->stage == DROP_REFERENCE */
8877 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
8879 if (wc
->refs
[level
] == 1) {
8881 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8882 ret
= btrfs_dec_ref(trans
, root
, eb
, 1);
8884 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
8885 BUG_ON(ret
); /* -ENOMEM */
8886 ret
= btrfs_qgroup_trace_leaf_items(trans
, fs_info
, eb
);
8888 btrfs_err_rl(fs_info
,
8889 "error %d accounting leaf items. Quota is out of sync, rescan required.",
8893 /* make block locked assertion in clean_tree_block happy */
8894 if (!path
->locks
[level
] &&
8895 btrfs_header_generation(eb
) == trans
->transid
) {
8896 btrfs_tree_lock(eb
);
8897 btrfs_set_lock_blocking(eb
);
8898 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8900 clean_tree_block(fs_info
, eb
);
8903 if (eb
== root
->node
) {
8904 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8907 BUG_ON(root
->root_key
.objectid
!=
8908 btrfs_header_owner(eb
));
8910 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8911 parent
= path
->nodes
[level
+ 1]->start
;
8913 BUG_ON(root
->root_key
.objectid
!=
8914 btrfs_header_owner(path
->nodes
[level
+ 1]));
8917 btrfs_free_tree_block(trans
, root
, eb
, parent
, wc
->refs
[level
] == 1);
8919 wc
->refs
[level
] = 0;
8920 wc
->flags
[level
] = 0;
8924 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
8925 struct btrfs_root
*root
,
8926 struct btrfs_path
*path
,
8927 struct walk_control
*wc
)
8929 int level
= wc
->level
;
8930 int lookup_info
= 1;
8933 while (level
>= 0) {
8934 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
8941 if (path
->slots
[level
] >=
8942 btrfs_header_nritems(path
->nodes
[level
]))
8945 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
8947 path
->slots
[level
]++;
8956 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
8957 struct btrfs_root
*root
,
8958 struct btrfs_path
*path
,
8959 struct walk_control
*wc
, int max_level
)
8961 int level
= wc
->level
;
8964 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
8965 while (level
< max_level
&& path
->nodes
[level
]) {
8967 if (path
->slots
[level
] + 1 <
8968 btrfs_header_nritems(path
->nodes
[level
])) {
8969 path
->slots
[level
]++;
8972 ret
= walk_up_proc(trans
, root
, path
, wc
);
8976 if (path
->locks
[level
]) {
8977 btrfs_tree_unlock_rw(path
->nodes
[level
],
8978 path
->locks
[level
]);
8979 path
->locks
[level
] = 0;
8981 free_extent_buffer(path
->nodes
[level
]);
8982 path
->nodes
[level
] = NULL
;
8990 * drop a subvolume tree.
8992 * this function traverses the tree freeing any blocks that only
8993 * referenced by the tree.
8995 * when a shared tree block is found. this function decreases its
8996 * reference count by one. if update_ref is true, this function
8997 * also make sure backrefs for the shared block and all lower level
8998 * blocks are properly updated.
9000 * If called with for_reloc == 0, may exit early with -EAGAIN
9002 int btrfs_drop_snapshot(struct btrfs_root
*root
,
9003 struct btrfs_block_rsv
*block_rsv
, int update_ref
,
9006 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
9007 struct btrfs_path
*path
;
9008 struct btrfs_trans_handle
*trans
;
9009 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
9010 struct btrfs_root_item
*root_item
= &root
->root_item
;
9011 struct walk_control
*wc
;
9012 struct btrfs_key key
;
9016 bool root_dropped
= false;
9018 btrfs_debug(fs_info
, "Drop subvolume %llu", root
->objectid
);
9020 path
= btrfs_alloc_path();
9026 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
9028 btrfs_free_path(path
);
9033 trans
= btrfs_start_transaction(tree_root
, 0);
9034 if (IS_ERR(trans
)) {
9035 err
= PTR_ERR(trans
);
9040 trans
->block_rsv
= block_rsv
;
9042 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
9043 level
= btrfs_header_level(root
->node
);
9044 path
->nodes
[level
] = btrfs_lock_root_node(root
);
9045 btrfs_set_lock_blocking(path
->nodes
[level
]);
9046 path
->slots
[level
] = 0;
9047 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
9048 memset(&wc
->update_progress
, 0,
9049 sizeof(wc
->update_progress
));
9051 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
9052 memcpy(&wc
->update_progress
, &key
,
9053 sizeof(wc
->update_progress
));
9055 level
= root_item
->drop_level
;
9057 path
->lowest_level
= level
;
9058 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
9059 path
->lowest_level
= 0;
9067 * unlock our path, this is safe because only this
9068 * function is allowed to delete this snapshot
9070 btrfs_unlock_up_safe(path
, 0);
9072 level
= btrfs_header_level(root
->node
);
9074 btrfs_tree_lock(path
->nodes
[level
]);
9075 btrfs_set_lock_blocking(path
->nodes
[level
]);
9076 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
9078 ret
= btrfs_lookup_extent_info(trans
, fs_info
,
9079 path
->nodes
[level
]->start
,
9080 level
, 1, &wc
->refs
[level
],
9086 BUG_ON(wc
->refs
[level
] == 0);
9088 if (level
== root_item
->drop_level
)
9091 btrfs_tree_unlock(path
->nodes
[level
]);
9092 path
->locks
[level
] = 0;
9093 WARN_ON(wc
->refs
[level
] != 1);
9099 wc
->shared_level
= -1;
9100 wc
->stage
= DROP_REFERENCE
;
9101 wc
->update_ref
= update_ref
;
9103 wc
->for_reloc
= for_reloc
;
9104 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(fs_info
);
9108 ret
= walk_down_tree(trans
, root
, path
, wc
);
9114 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
9121 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
9125 if (wc
->stage
== DROP_REFERENCE
) {
9127 btrfs_node_key(path
->nodes
[level
],
9128 &root_item
->drop_progress
,
9129 path
->slots
[level
]);
9130 root_item
->drop_level
= level
;
9133 BUG_ON(wc
->level
== 0);
9134 if (btrfs_should_end_transaction(trans
) ||
9135 (!for_reloc
&& btrfs_need_cleaner_sleep(fs_info
))) {
9136 ret
= btrfs_update_root(trans
, tree_root
,
9140 btrfs_abort_transaction(trans
, ret
);
9145 btrfs_end_transaction_throttle(trans
);
9146 if (!for_reloc
&& btrfs_need_cleaner_sleep(fs_info
)) {
9147 btrfs_debug(fs_info
,
9148 "drop snapshot early exit");
9153 trans
= btrfs_start_transaction(tree_root
, 0);
9154 if (IS_ERR(trans
)) {
9155 err
= PTR_ERR(trans
);
9159 trans
->block_rsv
= block_rsv
;
9162 btrfs_release_path(path
);
9166 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
9168 btrfs_abort_transaction(trans
, ret
);
9172 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
9173 ret
= btrfs_find_root(tree_root
, &root
->root_key
, path
,
9176 btrfs_abort_transaction(trans
, ret
);
9179 } else if (ret
> 0) {
9180 /* if we fail to delete the orphan item this time
9181 * around, it'll get picked up the next time.
9183 * The most common failure here is just -ENOENT.
9185 btrfs_del_orphan_item(trans
, tree_root
,
9186 root
->root_key
.objectid
);
9190 if (test_bit(BTRFS_ROOT_IN_RADIX
, &root
->state
)) {
9191 btrfs_add_dropped_root(trans
, root
);
9193 free_extent_buffer(root
->node
);
9194 free_extent_buffer(root
->commit_root
);
9195 btrfs_put_fs_root(root
);
9197 root_dropped
= true;
9199 btrfs_end_transaction_throttle(trans
);
9202 btrfs_free_path(path
);
9205 * So if we need to stop dropping the snapshot for whatever reason we
9206 * need to make sure to add it back to the dead root list so that we
9207 * keep trying to do the work later. This also cleans up roots if we
9208 * don't have it in the radix (like when we recover after a power fail
9209 * or unmount) so we don't leak memory.
9211 if (!for_reloc
&& root_dropped
== false)
9212 btrfs_add_dead_root(root
);
9213 if (err
&& err
!= -EAGAIN
)
9214 btrfs_handle_fs_error(fs_info
, err
, NULL
);
9219 * drop subtree rooted at tree block 'node'.
9221 * NOTE: this function will unlock and release tree block 'node'
9222 * only used by relocation code
9224 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
9225 struct btrfs_root
*root
,
9226 struct extent_buffer
*node
,
9227 struct extent_buffer
*parent
)
9229 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
9230 struct btrfs_path
*path
;
9231 struct walk_control
*wc
;
9237 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
9239 path
= btrfs_alloc_path();
9243 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
9245 btrfs_free_path(path
);
9249 btrfs_assert_tree_locked(parent
);
9250 parent_level
= btrfs_header_level(parent
);
9251 extent_buffer_get(parent
);
9252 path
->nodes
[parent_level
] = parent
;
9253 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
9255 btrfs_assert_tree_locked(node
);
9256 level
= btrfs_header_level(node
);
9257 path
->nodes
[level
] = node
;
9258 path
->slots
[level
] = 0;
9259 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
9261 wc
->refs
[parent_level
] = 1;
9262 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
9264 wc
->shared_level
= -1;
9265 wc
->stage
= DROP_REFERENCE
;
9269 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(fs_info
);
9272 wret
= walk_down_tree(trans
, root
, path
, wc
);
9278 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
9286 btrfs_free_path(path
);
9290 static u64
update_block_group_flags(struct btrfs_fs_info
*fs_info
, u64 flags
)
9296 * if restripe for this chunk_type is on pick target profile and
9297 * return, otherwise do the usual balance
9299 stripped
= get_restripe_target(fs_info
, flags
);
9301 return extended_to_chunk(stripped
);
9303 num_devices
= fs_info
->fs_devices
->rw_devices
;
9305 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
9306 BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
|
9307 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
9309 if (num_devices
== 1) {
9310 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
9311 stripped
= flags
& ~stripped
;
9313 /* turn raid0 into single device chunks */
9314 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
9317 /* turn mirroring into duplication */
9318 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
9319 BTRFS_BLOCK_GROUP_RAID10
))
9320 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
9322 /* they already had raid on here, just return */
9323 if (flags
& stripped
)
9326 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
9327 stripped
= flags
& ~stripped
;
9329 /* switch duplicated blocks with raid1 */
9330 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
9331 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
9333 /* this is drive concat, leave it alone */
9339 static int inc_block_group_ro(struct btrfs_block_group_cache
*cache
, int force
)
9341 struct btrfs_space_info
*sinfo
= cache
->space_info
;
9343 u64 min_allocable_bytes
;
9347 * We need some metadata space and system metadata space for
9348 * allocating chunks in some corner cases until we force to set
9349 * it to be readonly.
9352 (BTRFS_BLOCK_GROUP_SYSTEM
| BTRFS_BLOCK_GROUP_METADATA
)) &&
9354 min_allocable_bytes
= SZ_1M
;
9356 min_allocable_bytes
= 0;
9358 spin_lock(&sinfo
->lock
);
9359 spin_lock(&cache
->lock
);
9367 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
9368 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
9370 if (btrfs_space_info_used(sinfo
, true) + num_bytes
+
9371 min_allocable_bytes
<= sinfo
->total_bytes
) {
9372 sinfo
->bytes_readonly
+= num_bytes
;
9374 list_add_tail(&cache
->ro_list
, &sinfo
->ro_bgs
);
9378 spin_unlock(&cache
->lock
);
9379 spin_unlock(&sinfo
->lock
);
9383 int btrfs_inc_block_group_ro(struct btrfs_fs_info
*fs_info
,
9384 struct btrfs_block_group_cache
*cache
)
9387 struct btrfs_trans_handle
*trans
;
9392 trans
= btrfs_join_transaction(fs_info
->extent_root
);
9394 return PTR_ERR(trans
);
9397 * we're not allowed to set block groups readonly after the dirty
9398 * block groups cache has started writing. If it already started,
9399 * back off and let this transaction commit
9401 mutex_lock(&fs_info
->ro_block_group_mutex
);
9402 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN
, &trans
->transaction
->flags
)) {
9403 u64 transid
= trans
->transid
;
9405 mutex_unlock(&fs_info
->ro_block_group_mutex
);
9406 btrfs_end_transaction(trans
);
9408 ret
= btrfs_wait_for_commit(fs_info
, transid
);
9415 * if we are changing raid levels, try to allocate a corresponding
9416 * block group with the new raid level.
9418 alloc_flags
= update_block_group_flags(fs_info
, cache
->flags
);
9419 if (alloc_flags
!= cache
->flags
) {
9420 ret
= do_chunk_alloc(trans
, fs_info
, alloc_flags
,
9423 * ENOSPC is allowed here, we may have enough space
9424 * already allocated at the new raid level to
9433 ret
= inc_block_group_ro(cache
, 0);
9436 alloc_flags
= get_alloc_profile(fs_info
, cache
->space_info
->flags
);
9437 ret
= do_chunk_alloc(trans
, fs_info
, alloc_flags
,
9441 ret
= inc_block_group_ro(cache
, 0);
9443 if (cache
->flags
& BTRFS_BLOCK_GROUP_SYSTEM
) {
9444 alloc_flags
= update_block_group_flags(fs_info
, cache
->flags
);
9445 mutex_lock(&fs_info
->chunk_mutex
);
9446 check_system_chunk(trans
, fs_info
, alloc_flags
);
9447 mutex_unlock(&fs_info
->chunk_mutex
);
9449 mutex_unlock(&fs_info
->ro_block_group_mutex
);
9451 btrfs_end_transaction(trans
);
9455 int btrfs_force_chunk_alloc(struct btrfs_trans_handle
*trans
,
9456 struct btrfs_fs_info
*fs_info
, u64 type
)
9458 u64 alloc_flags
= get_alloc_profile(fs_info
, type
);
9460 return do_chunk_alloc(trans
, fs_info
, alloc_flags
, CHUNK_ALLOC_FORCE
);
9464 * helper to account the unused space of all the readonly block group in the
9465 * space_info. takes mirrors into account.
9467 u64
btrfs_account_ro_block_groups_free_space(struct btrfs_space_info
*sinfo
)
9469 struct btrfs_block_group_cache
*block_group
;
9473 /* It's df, we don't care if it's racy */
9474 if (list_empty(&sinfo
->ro_bgs
))
9477 spin_lock(&sinfo
->lock
);
9478 list_for_each_entry(block_group
, &sinfo
->ro_bgs
, ro_list
) {
9479 spin_lock(&block_group
->lock
);
9481 if (!block_group
->ro
) {
9482 spin_unlock(&block_group
->lock
);
9486 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_RAID1
|
9487 BTRFS_BLOCK_GROUP_RAID10
|
9488 BTRFS_BLOCK_GROUP_DUP
))
9493 free_bytes
+= (block_group
->key
.offset
-
9494 btrfs_block_group_used(&block_group
->item
)) *
9497 spin_unlock(&block_group
->lock
);
9499 spin_unlock(&sinfo
->lock
);
9504 void btrfs_dec_block_group_ro(struct btrfs_block_group_cache
*cache
)
9506 struct btrfs_space_info
*sinfo
= cache
->space_info
;
9511 spin_lock(&sinfo
->lock
);
9512 spin_lock(&cache
->lock
);
9514 num_bytes
= cache
->key
.offset
- cache
->reserved
-
9515 cache
->pinned
- cache
->bytes_super
-
9516 btrfs_block_group_used(&cache
->item
);
9517 sinfo
->bytes_readonly
-= num_bytes
;
9518 list_del_init(&cache
->ro_list
);
9520 spin_unlock(&cache
->lock
);
9521 spin_unlock(&sinfo
->lock
);
9525 * checks to see if its even possible to relocate this block group.
9527 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9528 * ok to go ahead and try.
9530 int btrfs_can_relocate(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
9532 struct btrfs_root
*root
= fs_info
->extent_root
;
9533 struct btrfs_block_group_cache
*block_group
;
9534 struct btrfs_space_info
*space_info
;
9535 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
9536 struct btrfs_device
*device
;
9537 struct btrfs_trans_handle
*trans
;
9547 debug
= btrfs_test_opt(fs_info
, ENOSPC_DEBUG
);
9549 block_group
= btrfs_lookup_block_group(fs_info
, bytenr
);
9551 /* odd, couldn't find the block group, leave it alone */
9555 "can't find block group for bytenr %llu",
9560 min_free
= btrfs_block_group_used(&block_group
->item
);
9562 /* no bytes used, we're good */
9566 space_info
= block_group
->space_info
;
9567 spin_lock(&space_info
->lock
);
9569 full
= space_info
->full
;
9572 * if this is the last block group we have in this space, we can't
9573 * relocate it unless we're able to allocate a new chunk below.
9575 * Otherwise, we need to make sure we have room in the space to handle
9576 * all of the extents from this block group. If we can, we're good
9578 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
9579 (btrfs_space_info_used(space_info
, false) + min_free
<
9580 space_info
->total_bytes
)) {
9581 spin_unlock(&space_info
->lock
);
9584 spin_unlock(&space_info
->lock
);
9587 * ok we don't have enough space, but maybe we have free space on our
9588 * devices to allocate new chunks for relocation, so loop through our
9589 * alloc devices and guess if we have enough space. if this block
9590 * group is going to be restriped, run checks against the target
9591 * profile instead of the current one.
9603 target
= get_restripe_target(fs_info
, block_group
->flags
);
9605 index
= __get_raid_index(extended_to_chunk(target
));
9608 * this is just a balance, so if we were marked as full
9609 * we know there is no space for a new chunk
9614 "no space to alloc new chunk for block group %llu",
9615 block_group
->key
.objectid
);
9619 index
= get_block_group_index(block_group
);
9622 if (index
== BTRFS_RAID_RAID10
) {
9626 } else if (index
== BTRFS_RAID_RAID1
) {
9628 } else if (index
== BTRFS_RAID_DUP
) {
9631 } else if (index
== BTRFS_RAID_RAID0
) {
9632 dev_min
= fs_devices
->rw_devices
;
9633 min_free
= div64_u64(min_free
, dev_min
);
9636 /* We need to do this so that we can look at pending chunks */
9637 trans
= btrfs_join_transaction(root
);
9638 if (IS_ERR(trans
)) {
9639 ret
= PTR_ERR(trans
);
9643 mutex_lock(&fs_info
->chunk_mutex
);
9644 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
9648 * check to make sure we can actually find a chunk with enough
9649 * space to fit our block group in.
9651 if (device
->total_bytes
> device
->bytes_used
+ min_free
&&
9652 !device
->is_tgtdev_for_dev_replace
) {
9653 ret
= find_free_dev_extent(trans
, device
, min_free
,
9658 if (dev_nr
>= dev_min
)
9664 if (debug
&& ret
== -1)
9666 "no space to allocate a new chunk for block group %llu",
9667 block_group
->key
.objectid
);
9668 mutex_unlock(&fs_info
->chunk_mutex
);
9669 btrfs_end_transaction(trans
);
9671 btrfs_put_block_group(block_group
);
9675 static int find_first_block_group(struct btrfs_fs_info
*fs_info
,
9676 struct btrfs_path
*path
,
9677 struct btrfs_key
*key
)
9679 struct btrfs_root
*root
= fs_info
->extent_root
;
9681 struct btrfs_key found_key
;
9682 struct extent_buffer
*leaf
;
9685 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
9690 slot
= path
->slots
[0];
9691 leaf
= path
->nodes
[0];
9692 if (slot
>= btrfs_header_nritems(leaf
)) {
9693 ret
= btrfs_next_leaf(root
, path
);
9700 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
9702 if (found_key
.objectid
>= key
->objectid
&&
9703 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
9704 struct extent_map_tree
*em_tree
;
9705 struct extent_map
*em
;
9707 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
9708 read_lock(&em_tree
->lock
);
9709 em
= lookup_extent_mapping(em_tree
, found_key
.objectid
,
9711 read_unlock(&em_tree
->lock
);
9714 "logical %llu len %llu found bg but no related chunk",
9715 found_key
.objectid
, found_key
.offset
);
9720 free_extent_map(em
);
9729 void btrfs_put_block_group_cache(struct btrfs_fs_info
*info
)
9731 struct btrfs_block_group_cache
*block_group
;
9735 struct inode
*inode
;
9737 block_group
= btrfs_lookup_first_block_group(info
, last
);
9738 while (block_group
) {
9739 spin_lock(&block_group
->lock
);
9740 if (block_group
->iref
)
9742 spin_unlock(&block_group
->lock
);
9743 block_group
= next_block_group(info
, block_group
);
9752 inode
= block_group
->inode
;
9753 block_group
->iref
= 0;
9754 block_group
->inode
= NULL
;
9755 spin_unlock(&block_group
->lock
);
9756 ASSERT(block_group
->io_ctl
.inode
== NULL
);
9758 last
= block_group
->key
.objectid
+ block_group
->key
.offset
;
9759 btrfs_put_block_group(block_group
);
9764 * Must be called only after stopping all workers, since we could have block
9765 * group caching kthreads running, and therefore they could race with us if we
9766 * freed the block groups before stopping them.
9768 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
9770 struct btrfs_block_group_cache
*block_group
;
9771 struct btrfs_space_info
*space_info
;
9772 struct btrfs_caching_control
*caching_ctl
;
9775 down_write(&info
->commit_root_sem
);
9776 while (!list_empty(&info
->caching_block_groups
)) {
9777 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
9778 struct btrfs_caching_control
, list
);
9779 list_del(&caching_ctl
->list
);
9780 put_caching_control(caching_ctl
);
9782 up_write(&info
->commit_root_sem
);
9784 spin_lock(&info
->unused_bgs_lock
);
9785 while (!list_empty(&info
->unused_bgs
)) {
9786 block_group
= list_first_entry(&info
->unused_bgs
,
9787 struct btrfs_block_group_cache
,
9789 list_del_init(&block_group
->bg_list
);
9790 btrfs_put_block_group(block_group
);
9792 spin_unlock(&info
->unused_bgs_lock
);
9794 spin_lock(&info
->block_group_cache_lock
);
9795 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
9796 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
9798 rb_erase(&block_group
->cache_node
,
9799 &info
->block_group_cache_tree
);
9800 RB_CLEAR_NODE(&block_group
->cache_node
);
9801 spin_unlock(&info
->block_group_cache_lock
);
9803 down_write(&block_group
->space_info
->groups_sem
);
9804 list_del(&block_group
->list
);
9805 up_write(&block_group
->space_info
->groups_sem
);
9808 * We haven't cached this block group, which means we could
9809 * possibly have excluded extents on this block group.
9811 if (block_group
->cached
== BTRFS_CACHE_NO
||
9812 block_group
->cached
== BTRFS_CACHE_ERROR
)
9813 free_excluded_extents(info
, block_group
);
9815 btrfs_remove_free_space_cache(block_group
);
9816 ASSERT(block_group
->cached
!= BTRFS_CACHE_STARTED
);
9817 ASSERT(list_empty(&block_group
->dirty_list
));
9818 ASSERT(list_empty(&block_group
->io_list
));
9819 ASSERT(list_empty(&block_group
->bg_list
));
9820 ASSERT(atomic_read(&block_group
->count
) == 1);
9821 btrfs_put_block_group(block_group
);
9823 spin_lock(&info
->block_group_cache_lock
);
9825 spin_unlock(&info
->block_group_cache_lock
);
9827 /* now that all the block groups are freed, go through and
9828 * free all the space_info structs. This is only called during
9829 * the final stages of unmount, and so we know nobody is
9830 * using them. We call synchronize_rcu() once before we start,
9831 * just to be on the safe side.
9835 release_global_block_rsv(info
);
9837 while (!list_empty(&info
->space_info
)) {
9840 space_info
= list_entry(info
->space_info
.next
,
9841 struct btrfs_space_info
,
9845 * Do not hide this behind enospc_debug, this is actually
9846 * important and indicates a real bug if this happens.
9848 if (WARN_ON(space_info
->bytes_pinned
> 0 ||
9849 space_info
->bytes_reserved
> 0 ||
9850 space_info
->bytes_may_use
> 0))
9851 dump_space_info(info
, space_info
, 0, 0);
9852 list_del(&space_info
->list
);
9853 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++) {
9854 struct kobject
*kobj
;
9855 kobj
= space_info
->block_group_kobjs
[i
];
9856 space_info
->block_group_kobjs
[i
] = NULL
;
9862 kobject_del(&space_info
->kobj
);
9863 kobject_put(&space_info
->kobj
);
9868 static void __link_block_group(struct btrfs_space_info
*space_info
,
9869 struct btrfs_block_group_cache
*cache
)
9871 int index
= get_block_group_index(cache
);
9874 down_write(&space_info
->groups_sem
);
9875 if (list_empty(&space_info
->block_groups
[index
]))
9877 list_add_tail(&cache
->list
, &space_info
->block_groups
[index
]);
9878 up_write(&space_info
->groups_sem
);
9881 struct raid_kobject
*rkobj
;
9884 rkobj
= kzalloc(sizeof(*rkobj
), GFP_NOFS
);
9887 rkobj
->raid_type
= index
;
9888 kobject_init(&rkobj
->kobj
, &btrfs_raid_ktype
);
9889 ret
= kobject_add(&rkobj
->kobj
, &space_info
->kobj
,
9890 "%s", get_raid_name(index
));
9892 kobject_put(&rkobj
->kobj
);
9895 space_info
->block_group_kobjs
[index
] = &rkobj
->kobj
;
9900 btrfs_warn(cache
->fs_info
,
9901 "failed to add kobject for block cache, ignoring");
9904 static struct btrfs_block_group_cache
*
9905 btrfs_create_block_group_cache(struct btrfs_fs_info
*fs_info
,
9906 u64 start
, u64 size
)
9908 struct btrfs_block_group_cache
*cache
;
9910 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
9914 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
9916 if (!cache
->free_space_ctl
) {
9921 cache
->key
.objectid
= start
;
9922 cache
->key
.offset
= size
;
9923 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
9925 cache
->sectorsize
= fs_info
->sectorsize
;
9926 cache
->fs_info
= fs_info
;
9927 cache
->full_stripe_len
= btrfs_full_stripe_len(fs_info
,
9928 &fs_info
->mapping_tree
,
9930 set_free_space_tree_thresholds(cache
);
9932 atomic_set(&cache
->count
, 1);
9933 spin_lock_init(&cache
->lock
);
9934 init_rwsem(&cache
->data_rwsem
);
9935 INIT_LIST_HEAD(&cache
->list
);
9936 INIT_LIST_HEAD(&cache
->cluster_list
);
9937 INIT_LIST_HEAD(&cache
->bg_list
);
9938 INIT_LIST_HEAD(&cache
->ro_list
);
9939 INIT_LIST_HEAD(&cache
->dirty_list
);
9940 INIT_LIST_HEAD(&cache
->io_list
);
9941 btrfs_init_free_space_ctl(cache
);
9942 atomic_set(&cache
->trimming
, 0);
9943 mutex_init(&cache
->free_space_lock
);
9944 btrfs_init_full_stripe_locks_tree(&cache
->full_stripe_locks_root
);
9949 int btrfs_read_block_groups(struct btrfs_fs_info
*info
)
9951 struct btrfs_path
*path
;
9953 struct btrfs_block_group_cache
*cache
;
9954 struct btrfs_space_info
*space_info
;
9955 struct btrfs_key key
;
9956 struct btrfs_key found_key
;
9957 struct extent_buffer
*leaf
;
9963 feature
= btrfs_super_incompat_flags(info
->super_copy
);
9964 mixed
= !!(feature
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
);
9968 key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
9969 path
= btrfs_alloc_path();
9972 path
->reada
= READA_FORWARD
;
9974 cache_gen
= btrfs_super_cache_generation(info
->super_copy
);
9975 if (btrfs_test_opt(info
, SPACE_CACHE
) &&
9976 btrfs_super_generation(info
->super_copy
) != cache_gen
)
9978 if (btrfs_test_opt(info
, CLEAR_CACHE
))
9982 ret
= find_first_block_group(info
, path
, &key
);
9988 leaf
= path
->nodes
[0];
9989 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
9991 cache
= btrfs_create_block_group_cache(info
, found_key
.objectid
,
10000 * When we mount with old space cache, we need to
10001 * set BTRFS_DC_CLEAR and set dirty flag.
10003 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
10004 * truncate the old free space cache inode and
10006 * b) Setting 'dirty flag' makes sure that we flush
10007 * the new space cache info onto disk.
10009 if (btrfs_test_opt(info
, SPACE_CACHE
))
10010 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
10013 read_extent_buffer(leaf
, &cache
->item
,
10014 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
10015 sizeof(cache
->item
));
10016 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
10018 ((cache
->flags
& BTRFS_BLOCK_GROUP_METADATA
) &&
10019 (cache
->flags
& BTRFS_BLOCK_GROUP_DATA
))) {
10021 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
10022 cache
->key
.objectid
);
10027 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
10028 btrfs_release_path(path
);
10031 * We need to exclude the super stripes now so that the space
10032 * info has super bytes accounted for, otherwise we'll think
10033 * we have more space than we actually do.
10035 ret
= exclude_super_stripes(info
, cache
);
10038 * We may have excluded something, so call this just in
10041 free_excluded_extents(info
, cache
);
10042 btrfs_put_block_group(cache
);
10047 * check for two cases, either we are full, and therefore
10048 * don't need to bother with the caching work since we won't
10049 * find any space, or we are empty, and we can just add all
10050 * the space in and be done with it. This saves us _alot_ of
10051 * time, particularly in the full case.
10053 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
10054 cache
->last_byte_to_unpin
= (u64
)-1;
10055 cache
->cached
= BTRFS_CACHE_FINISHED
;
10056 free_excluded_extents(info
, cache
);
10057 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
10058 cache
->last_byte_to_unpin
= (u64
)-1;
10059 cache
->cached
= BTRFS_CACHE_FINISHED
;
10060 add_new_free_space(cache
, info
,
10061 found_key
.objectid
,
10062 found_key
.objectid
+
10064 free_excluded_extents(info
, cache
);
10067 ret
= btrfs_add_block_group_cache(info
, cache
);
10069 btrfs_remove_free_space_cache(cache
);
10070 btrfs_put_block_group(cache
);
10074 trace_btrfs_add_block_group(info
, cache
, 0);
10075 update_space_info(info
, cache
->flags
, found_key
.offset
,
10076 btrfs_block_group_used(&cache
->item
),
10077 cache
->bytes_super
, &space_info
);
10079 cache
->space_info
= space_info
;
10081 __link_block_group(space_info
, cache
);
10083 set_avail_alloc_bits(info
, cache
->flags
);
10084 if (btrfs_chunk_readonly(info
, cache
->key
.objectid
)) {
10085 inc_block_group_ro(cache
, 1);
10086 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
10087 spin_lock(&info
->unused_bgs_lock
);
10088 /* Should always be true but just in case. */
10089 if (list_empty(&cache
->bg_list
)) {
10090 btrfs_get_block_group(cache
);
10091 list_add_tail(&cache
->bg_list
,
10092 &info
->unused_bgs
);
10094 spin_unlock(&info
->unused_bgs_lock
);
10098 list_for_each_entry_rcu(space_info
, &info
->space_info
, list
) {
10099 if (!(get_alloc_profile(info
, space_info
->flags
) &
10100 (BTRFS_BLOCK_GROUP_RAID10
|
10101 BTRFS_BLOCK_GROUP_RAID1
|
10102 BTRFS_BLOCK_GROUP_RAID5
|
10103 BTRFS_BLOCK_GROUP_RAID6
|
10104 BTRFS_BLOCK_GROUP_DUP
)))
10107 * avoid allocating from un-mirrored block group if there are
10108 * mirrored block groups.
10110 list_for_each_entry(cache
,
10111 &space_info
->block_groups
[BTRFS_RAID_RAID0
],
10113 inc_block_group_ro(cache
, 1);
10114 list_for_each_entry(cache
,
10115 &space_info
->block_groups
[BTRFS_RAID_SINGLE
],
10117 inc_block_group_ro(cache
, 1);
10120 init_global_block_rsv(info
);
10123 btrfs_free_path(path
);
10127 void btrfs_create_pending_block_groups(struct btrfs_trans_handle
*trans
,
10128 struct btrfs_fs_info
*fs_info
)
10130 struct btrfs_block_group_cache
*block_group
, *tmp
;
10131 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
10132 struct btrfs_block_group_item item
;
10133 struct btrfs_key key
;
10135 bool can_flush_pending_bgs
= trans
->can_flush_pending_bgs
;
10137 trans
->can_flush_pending_bgs
= false;
10138 list_for_each_entry_safe(block_group
, tmp
, &trans
->new_bgs
, bg_list
) {
10142 spin_lock(&block_group
->lock
);
10143 memcpy(&item
, &block_group
->item
, sizeof(item
));
10144 memcpy(&key
, &block_group
->key
, sizeof(key
));
10145 spin_unlock(&block_group
->lock
);
10147 ret
= btrfs_insert_item(trans
, extent_root
, &key
, &item
,
10150 btrfs_abort_transaction(trans
, ret
);
10151 ret
= btrfs_finish_chunk_alloc(trans
, fs_info
, key
.objectid
,
10154 btrfs_abort_transaction(trans
, ret
);
10155 add_block_group_free_space(trans
, fs_info
, block_group
);
10156 /* already aborted the transaction if it failed. */
10158 list_del_init(&block_group
->bg_list
);
10160 trans
->can_flush_pending_bgs
= can_flush_pending_bgs
;
10163 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
10164 struct btrfs_fs_info
*fs_info
, u64 bytes_used
,
10165 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
10168 struct btrfs_block_group_cache
*cache
;
10171 btrfs_set_log_full_commit(fs_info
, trans
);
10173 cache
= btrfs_create_block_group_cache(fs_info
, chunk_offset
, size
);
10177 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
10178 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
10179 btrfs_set_block_group_flags(&cache
->item
, type
);
10181 cache
->flags
= type
;
10182 cache
->last_byte_to_unpin
= (u64
)-1;
10183 cache
->cached
= BTRFS_CACHE_FINISHED
;
10184 cache
->needs_free_space
= 1;
10185 ret
= exclude_super_stripes(fs_info
, cache
);
10188 * We may have excluded something, so call this just in
10191 free_excluded_extents(fs_info
, cache
);
10192 btrfs_put_block_group(cache
);
10196 add_new_free_space(cache
, fs_info
, chunk_offset
, chunk_offset
+ size
);
10198 free_excluded_extents(fs_info
, cache
);
10200 #ifdef CONFIG_BTRFS_DEBUG
10201 if (btrfs_should_fragment_free_space(cache
)) {
10202 u64 new_bytes_used
= size
- bytes_used
;
10204 bytes_used
+= new_bytes_used
>> 1;
10205 fragment_free_space(cache
);
10209 * Ensure the corresponding space_info object is created and
10210 * assigned to our block group. We want our bg to be added to the rbtree
10211 * with its ->space_info set.
10213 cache
->space_info
= __find_space_info(fs_info
, cache
->flags
);
10214 if (!cache
->space_info
) {
10215 ret
= create_space_info(fs_info
, cache
->flags
,
10216 &cache
->space_info
);
10218 btrfs_remove_free_space_cache(cache
);
10219 btrfs_put_block_group(cache
);
10224 ret
= btrfs_add_block_group_cache(fs_info
, cache
);
10226 btrfs_remove_free_space_cache(cache
);
10227 btrfs_put_block_group(cache
);
10232 * Now that our block group has its ->space_info set and is inserted in
10233 * the rbtree, update the space info's counters.
10235 trace_btrfs_add_block_group(fs_info
, cache
, 1);
10236 update_space_info(fs_info
, cache
->flags
, size
, bytes_used
,
10237 cache
->bytes_super
, &cache
->space_info
);
10238 update_global_block_rsv(fs_info
);
10240 __link_block_group(cache
->space_info
, cache
);
10242 list_add_tail(&cache
->bg_list
, &trans
->new_bgs
);
10244 set_avail_alloc_bits(fs_info
, type
);
10248 static void clear_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
10250 u64 extra_flags
= chunk_to_extended(flags
) &
10251 BTRFS_EXTENDED_PROFILE_MASK
;
10253 write_seqlock(&fs_info
->profiles_lock
);
10254 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
10255 fs_info
->avail_data_alloc_bits
&= ~extra_flags
;
10256 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
10257 fs_info
->avail_metadata_alloc_bits
&= ~extra_flags
;
10258 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
10259 fs_info
->avail_system_alloc_bits
&= ~extra_flags
;
10260 write_sequnlock(&fs_info
->profiles_lock
);
10263 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
10264 struct btrfs_fs_info
*fs_info
, u64 group_start
,
10265 struct extent_map
*em
)
10267 struct btrfs_root
*root
= fs_info
->extent_root
;
10268 struct btrfs_path
*path
;
10269 struct btrfs_block_group_cache
*block_group
;
10270 struct btrfs_free_cluster
*cluster
;
10271 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
10272 struct btrfs_key key
;
10273 struct inode
*inode
;
10274 struct kobject
*kobj
= NULL
;
10278 struct btrfs_caching_control
*caching_ctl
= NULL
;
10281 block_group
= btrfs_lookup_block_group(fs_info
, group_start
);
10282 BUG_ON(!block_group
);
10283 BUG_ON(!block_group
->ro
);
10286 * Free the reserved super bytes from this block group before
10289 free_excluded_extents(fs_info
, block_group
);
10291 memcpy(&key
, &block_group
->key
, sizeof(key
));
10292 index
= get_block_group_index(block_group
);
10293 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
10294 BTRFS_BLOCK_GROUP_RAID1
|
10295 BTRFS_BLOCK_GROUP_RAID10
))
10300 /* make sure this block group isn't part of an allocation cluster */
10301 cluster
= &fs_info
->data_alloc_cluster
;
10302 spin_lock(&cluster
->refill_lock
);
10303 btrfs_return_cluster_to_free_space(block_group
, cluster
);
10304 spin_unlock(&cluster
->refill_lock
);
10307 * make sure this block group isn't part of a metadata
10308 * allocation cluster
10310 cluster
= &fs_info
->meta_alloc_cluster
;
10311 spin_lock(&cluster
->refill_lock
);
10312 btrfs_return_cluster_to_free_space(block_group
, cluster
);
10313 spin_unlock(&cluster
->refill_lock
);
10315 path
= btrfs_alloc_path();
10322 * get the inode first so any iput calls done for the io_list
10323 * aren't the final iput (no unlinks allowed now)
10325 inode
= lookup_free_space_inode(fs_info
, block_group
, path
);
10327 mutex_lock(&trans
->transaction
->cache_write_mutex
);
10329 * make sure our free spache cache IO is done before remove the
10332 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
10333 if (!list_empty(&block_group
->io_list
)) {
10334 list_del_init(&block_group
->io_list
);
10336 WARN_ON(!IS_ERR(inode
) && inode
!= block_group
->io_ctl
.inode
);
10338 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
10339 btrfs_wait_cache_io(trans
, block_group
, path
);
10340 btrfs_put_block_group(block_group
);
10341 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
10344 if (!list_empty(&block_group
->dirty_list
)) {
10345 list_del_init(&block_group
->dirty_list
);
10346 btrfs_put_block_group(block_group
);
10348 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
10349 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
10351 if (!IS_ERR(inode
)) {
10352 ret
= btrfs_orphan_add(trans
, BTRFS_I(inode
));
10354 btrfs_add_delayed_iput(inode
);
10357 clear_nlink(inode
);
10358 /* One for the block groups ref */
10359 spin_lock(&block_group
->lock
);
10360 if (block_group
->iref
) {
10361 block_group
->iref
= 0;
10362 block_group
->inode
= NULL
;
10363 spin_unlock(&block_group
->lock
);
10366 spin_unlock(&block_group
->lock
);
10368 /* One for our lookup ref */
10369 btrfs_add_delayed_iput(inode
);
10372 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
10373 key
.offset
= block_group
->key
.objectid
;
10376 ret
= btrfs_search_slot(trans
, tree_root
, &key
, path
, -1, 1);
10380 btrfs_release_path(path
);
10382 ret
= btrfs_del_item(trans
, tree_root
, path
);
10385 btrfs_release_path(path
);
10388 spin_lock(&fs_info
->block_group_cache_lock
);
10389 rb_erase(&block_group
->cache_node
,
10390 &fs_info
->block_group_cache_tree
);
10391 RB_CLEAR_NODE(&block_group
->cache_node
);
10393 if (fs_info
->first_logical_byte
== block_group
->key
.objectid
)
10394 fs_info
->first_logical_byte
= (u64
)-1;
10395 spin_unlock(&fs_info
->block_group_cache_lock
);
10397 down_write(&block_group
->space_info
->groups_sem
);
10399 * we must use list_del_init so people can check to see if they
10400 * are still on the list after taking the semaphore
10402 list_del_init(&block_group
->list
);
10403 if (list_empty(&block_group
->space_info
->block_groups
[index
])) {
10404 kobj
= block_group
->space_info
->block_group_kobjs
[index
];
10405 block_group
->space_info
->block_group_kobjs
[index
] = NULL
;
10406 clear_avail_alloc_bits(fs_info
, block_group
->flags
);
10408 up_write(&block_group
->space_info
->groups_sem
);
10414 if (block_group
->has_caching_ctl
)
10415 caching_ctl
= get_caching_control(block_group
);
10416 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
10417 wait_block_group_cache_done(block_group
);
10418 if (block_group
->has_caching_ctl
) {
10419 down_write(&fs_info
->commit_root_sem
);
10420 if (!caching_ctl
) {
10421 struct btrfs_caching_control
*ctl
;
10423 list_for_each_entry(ctl
,
10424 &fs_info
->caching_block_groups
, list
)
10425 if (ctl
->block_group
== block_group
) {
10427 refcount_inc(&caching_ctl
->count
);
10432 list_del_init(&caching_ctl
->list
);
10433 up_write(&fs_info
->commit_root_sem
);
10435 /* Once for the caching bgs list and once for us. */
10436 put_caching_control(caching_ctl
);
10437 put_caching_control(caching_ctl
);
10441 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
10442 if (!list_empty(&block_group
->dirty_list
)) {
10445 if (!list_empty(&block_group
->io_list
)) {
10448 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
10449 btrfs_remove_free_space_cache(block_group
);
10451 spin_lock(&block_group
->space_info
->lock
);
10452 list_del_init(&block_group
->ro_list
);
10454 if (btrfs_test_opt(fs_info
, ENOSPC_DEBUG
)) {
10455 WARN_ON(block_group
->space_info
->total_bytes
10456 < block_group
->key
.offset
);
10457 WARN_ON(block_group
->space_info
->bytes_readonly
10458 < block_group
->key
.offset
);
10459 WARN_ON(block_group
->space_info
->disk_total
10460 < block_group
->key
.offset
* factor
);
10462 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
10463 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
10464 block_group
->space_info
->disk_total
-= block_group
->key
.offset
* factor
;
10466 spin_unlock(&block_group
->space_info
->lock
);
10468 memcpy(&key
, &block_group
->key
, sizeof(key
));
10470 mutex_lock(&fs_info
->chunk_mutex
);
10471 if (!list_empty(&em
->list
)) {
10472 /* We're in the transaction->pending_chunks list. */
10473 free_extent_map(em
);
10475 spin_lock(&block_group
->lock
);
10476 block_group
->removed
= 1;
10478 * At this point trimming can't start on this block group, because we
10479 * removed the block group from the tree fs_info->block_group_cache_tree
10480 * so no one can't find it anymore and even if someone already got this
10481 * block group before we removed it from the rbtree, they have already
10482 * incremented block_group->trimming - if they didn't, they won't find
10483 * any free space entries because we already removed them all when we
10484 * called btrfs_remove_free_space_cache().
10486 * And we must not remove the extent map from the fs_info->mapping_tree
10487 * to prevent the same logical address range and physical device space
10488 * ranges from being reused for a new block group. This is because our
10489 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10490 * completely transactionless, so while it is trimming a range the
10491 * currently running transaction might finish and a new one start,
10492 * allowing for new block groups to be created that can reuse the same
10493 * physical device locations unless we take this special care.
10495 * There may also be an implicit trim operation if the file system
10496 * is mounted with -odiscard. The same protections must remain
10497 * in place until the extents have been discarded completely when
10498 * the transaction commit has completed.
10500 remove_em
= (atomic_read(&block_group
->trimming
) == 0);
10502 * Make sure a trimmer task always sees the em in the pinned_chunks list
10503 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10504 * before checking block_group->removed).
10508 * Our em might be in trans->transaction->pending_chunks which
10509 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10510 * and so is the fs_info->pinned_chunks list.
10512 * So at this point we must be holding the chunk_mutex to avoid
10513 * any races with chunk allocation (more specifically at
10514 * volumes.c:contains_pending_extent()), to ensure it always
10515 * sees the em, either in the pending_chunks list or in the
10516 * pinned_chunks list.
10518 list_move_tail(&em
->list
, &fs_info
->pinned_chunks
);
10520 spin_unlock(&block_group
->lock
);
10523 struct extent_map_tree
*em_tree
;
10525 em_tree
= &fs_info
->mapping_tree
.map_tree
;
10526 write_lock(&em_tree
->lock
);
10528 * The em might be in the pending_chunks list, so make sure the
10529 * chunk mutex is locked, since remove_extent_mapping() will
10530 * delete us from that list.
10532 remove_extent_mapping(em_tree
, em
);
10533 write_unlock(&em_tree
->lock
);
10534 /* once for the tree */
10535 free_extent_map(em
);
10538 mutex_unlock(&fs_info
->chunk_mutex
);
10540 ret
= remove_block_group_free_space(trans
, fs_info
, block_group
);
10544 btrfs_put_block_group(block_group
);
10545 btrfs_put_block_group(block_group
);
10547 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
10553 ret
= btrfs_del_item(trans
, root
, path
);
10555 btrfs_free_path(path
);
10559 struct btrfs_trans_handle
*
10560 btrfs_start_trans_remove_block_group(struct btrfs_fs_info
*fs_info
,
10561 const u64 chunk_offset
)
10563 struct extent_map_tree
*em_tree
= &fs_info
->mapping_tree
.map_tree
;
10564 struct extent_map
*em
;
10565 struct map_lookup
*map
;
10566 unsigned int num_items
;
10568 read_lock(&em_tree
->lock
);
10569 em
= lookup_extent_mapping(em_tree
, chunk_offset
, 1);
10570 read_unlock(&em_tree
->lock
);
10571 ASSERT(em
&& em
->start
== chunk_offset
);
10574 * We need to reserve 3 + N units from the metadata space info in order
10575 * to remove a block group (done at btrfs_remove_chunk() and at
10576 * btrfs_remove_block_group()), which are used for:
10578 * 1 unit for adding the free space inode's orphan (located in the tree
10580 * 1 unit for deleting the block group item (located in the extent
10582 * 1 unit for deleting the free space item (located in tree of tree
10584 * N units for deleting N device extent items corresponding to each
10585 * stripe (located in the device tree).
10587 * In order to remove a block group we also need to reserve units in the
10588 * system space info in order to update the chunk tree (update one or
10589 * more device items and remove one chunk item), but this is done at
10590 * btrfs_remove_chunk() through a call to check_system_chunk().
10592 map
= em
->map_lookup
;
10593 num_items
= 3 + map
->num_stripes
;
10594 free_extent_map(em
);
10596 return btrfs_start_transaction_fallback_global_rsv(fs_info
->extent_root
,
10601 * Process the unused_bgs list and remove any that don't have any allocated
10602 * space inside of them.
10604 void btrfs_delete_unused_bgs(struct btrfs_fs_info
*fs_info
)
10606 struct btrfs_block_group_cache
*block_group
;
10607 struct btrfs_space_info
*space_info
;
10608 struct btrfs_trans_handle
*trans
;
10611 if (!test_bit(BTRFS_FS_OPEN
, &fs_info
->flags
))
10614 spin_lock(&fs_info
->unused_bgs_lock
);
10615 while (!list_empty(&fs_info
->unused_bgs
)) {
10619 block_group
= list_first_entry(&fs_info
->unused_bgs
,
10620 struct btrfs_block_group_cache
,
10622 list_del_init(&block_group
->bg_list
);
10624 space_info
= block_group
->space_info
;
10626 if (ret
|| btrfs_mixed_space_info(space_info
)) {
10627 btrfs_put_block_group(block_group
);
10630 spin_unlock(&fs_info
->unused_bgs_lock
);
10632 mutex_lock(&fs_info
->delete_unused_bgs_mutex
);
10634 /* Don't want to race with allocators so take the groups_sem */
10635 down_write(&space_info
->groups_sem
);
10636 spin_lock(&block_group
->lock
);
10637 if (block_group
->reserved
||
10638 btrfs_block_group_used(&block_group
->item
) ||
10640 list_is_singular(&block_group
->list
)) {
10642 * We want to bail if we made new allocations or have
10643 * outstanding allocations in this block group. We do
10644 * the ro check in case balance is currently acting on
10645 * this block group.
10647 spin_unlock(&block_group
->lock
);
10648 up_write(&space_info
->groups_sem
);
10651 spin_unlock(&block_group
->lock
);
10653 /* We don't want to force the issue, only flip if it's ok. */
10654 ret
= inc_block_group_ro(block_group
, 0);
10655 up_write(&space_info
->groups_sem
);
10662 * Want to do this before we do anything else so we can recover
10663 * properly if we fail to join the transaction.
10665 trans
= btrfs_start_trans_remove_block_group(fs_info
,
10666 block_group
->key
.objectid
);
10667 if (IS_ERR(trans
)) {
10668 btrfs_dec_block_group_ro(block_group
);
10669 ret
= PTR_ERR(trans
);
10674 * We could have pending pinned extents for this block group,
10675 * just delete them, we don't care about them anymore.
10677 start
= block_group
->key
.objectid
;
10678 end
= start
+ block_group
->key
.offset
- 1;
10680 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10681 * btrfs_finish_extent_commit(). If we are at transaction N,
10682 * another task might be running finish_extent_commit() for the
10683 * previous transaction N - 1, and have seen a range belonging
10684 * to the block group in freed_extents[] before we were able to
10685 * clear the whole block group range from freed_extents[]. This
10686 * means that task can lookup for the block group after we
10687 * unpinned it from freed_extents[] and removed it, leading to
10688 * a BUG_ON() at btrfs_unpin_extent_range().
10690 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
10691 ret
= clear_extent_bits(&fs_info
->freed_extents
[0], start
, end
,
10694 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10695 btrfs_dec_block_group_ro(block_group
);
10698 ret
= clear_extent_bits(&fs_info
->freed_extents
[1], start
, end
,
10701 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10702 btrfs_dec_block_group_ro(block_group
);
10705 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10707 /* Reset pinned so btrfs_put_block_group doesn't complain */
10708 spin_lock(&space_info
->lock
);
10709 spin_lock(&block_group
->lock
);
10711 space_info
->bytes_pinned
-= block_group
->pinned
;
10712 space_info
->bytes_readonly
+= block_group
->pinned
;
10713 percpu_counter_add(&space_info
->total_bytes_pinned
,
10714 -block_group
->pinned
);
10715 block_group
->pinned
= 0;
10717 spin_unlock(&block_group
->lock
);
10718 spin_unlock(&space_info
->lock
);
10720 /* DISCARD can flip during remount */
10721 trimming
= btrfs_test_opt(fs_info
, DISCARD
);
10723 /* Implicit trim during transaction commit. */
10725 btrfs_get_block_group_trimming(block_group
);
10728 * Btrfs_remove_chunk will abort the transaction if things go
10731 ret
= btrfs_remove_chunk(trans
, fs_info
,
10732 block_group
->key
.objectid
);
10736 btrfs_put_block_group_trimming(block_group
);
10741 * If we're not mounted with -odiscard, we can just forget
10742 * about this block group. Otherwise we'll need to wait
10743 * until transaction commit to do the actual discard.
10746 spin_lock(&fs_info
->unused_bgs_lock
);
10748 * A concurrent scrub might have added us to the list
10749 * fs_info->unused_bgs, so use a list_move operation
10750 * to add the block group to the deleted_bgs list.
10752 list_move(&block_group
->bg_list
,
10753 &trans
->transaction
->deleted_bgs
);
10754 spin_unlock(&fs_info
->unused_bgs_lock
);
10755 btrfs_get_block_group(block_group
);
10758 btrfs_end_transaction(trans
);
10760 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
10761 btrfs_put_block_group(block_group
);
10762 spin_lock(&fs_info
->unused_bgs_lock
);
10764 spin_unlock(&fs_info
->unused_bgs_lock
);
10767 int btrfs_init_space_info(struct btrfs_fs_info
*fs_info
)
10769 struct btrfs_space_info
*space_info
;
10770 struct btrfs_super_block
*disk_super
;
10776 disk_super
= fs_info
->super_copy
;
10777 if (!btrfs_super_root(disk_super
))
10780 features
= btrfs_super_incompat_flags(disk_super
);
10781 if (features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
10784 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
10785 ret
= create_space_info(fs_info
, flags
, &space_info
);
10790 flags
= BTRFS_BLOCK_GROUP_METADATA
| BTRFS_BLOCK_GROUP_DATA
;
10791 ret
= create_space_info(fs_info
, flags
, &space_info
);
10793 flags
= BTRFS_BLOCK_GROUP_METADATA
;
10794 ret
= create_space_info(fs_info
, flags
, &space_info
);
10798 flags
= BTRFS_BLOCK_GROUP_DATA
;
10799 ret
= create_space_info(fs_info
, flags
, &space_info
);
10805 int btrfs_error_unpin_extent_range(struct btrfs_fs_info
*fs_info
,
10806 u64 start
, u64 end
)
10808 return unpin_extent_range(fs_info
, start
, end
, false);
10812 * It used to be that old block groups would be left around forever.
10813 * Iterating over them would be enough to trim unused space. Since we
10814 * now automatically remove them, we also need to iterate over unallocated
10817 * We don't want a transaction for this since the discard may take a
10818 * substantial amount of time. We don't require that a transaction be
10819 * running, but we do need to take a running transaction into account
10820 * to ensure that we're not discarding chunks that were released in
10821 * the current transaction.
10823 * Holding the chunks lock will prevent other threads from allocating
10824 * or releasing chunks, but it won't prevent a running transaction
10825 * from committing and releasing the memory that the pending chunks
10826 * list head uses. For that, we need to take a reference to the
10829 static int btrfs_trim_free_extents(struct btrfs_device
*device
,
10830 u64 minlen
, u64
*trimmed
)
10832 u64 start
= 0, len
= 0;
10837 /* Not writeable = nothing to do. */
10838 if (!device
->writeable
)
10841 /* No free space = nothing to do. */
10842 if (device
->total_bytes
<= device
->bytes_used
)
10848 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
10849 struct btrfs_transaction
*trans
;
10852 ret
= mutex_lock_interruptible(&fs_info
->chunk_mutex
);
10856 down_read(&fs_info
->commit_root_sem
);
10858 spin_lock(&fs_info
->trans_lock
);
10859 trans
= fs_info
->running_transaction
;
10861 refcount_inc(&trans
->use_count
);
10862 spin_unlock(&fs_info
->trans_lock
);
10864 ret
= find_free_dev_extent_start(trans
, device
, minlen
, start
,
10867 btrfs_put_transaction(trans
);
10870 up_read(&fs_info
->commit_root_sem
);
10871 mutex_unlock(&fs_info
->chunk_mutex
);
10872 if (ret
== -ENOSPC
)
10877 ret
= btrfs_issue_discard(device
->bdev
, start
, len
, &bytes
);
10878 up_read(&fs_info
->commit_root_sem
);
10879 mutex_unlock(&fs_info
->chunk_mutex
);
10887 if (fatal_signal_pending(current
)) {
10888 ret
= -ERESTARTSYS
;
10898 int btrfs_trim_fs(struct btrfs_fs_info
*fs_info
, struct fstrim_range
*range
)
10900 struct btrfs_block_group_cache
*cache
= NULL
;
10901 struct btrfs_device
*device
;
10902 struct list_head
*devices
;
10907 u64 total_bytes
= btrfs_super_total_bytes(fs_info
->super_copy
);
10911 * try to trim all FS space, our block group may start from non-zero.
10913 if (range
->len
== total_bytes
)
10914 cache
= btrfs_lookup_first_block_group(fs_info
, range
->start
);
10916 cache
= btrfs_lookup_block_group(fs_info
, range
->start
);
10919 if (cache
->key
.objectid
>= (range
->start
+ range
->len
)) {
10920 btrfs_put_block_group(cache
);
10924 start
= max(range
->start
, cache
->key
.objectid
);
10925 end
= min(range
->start
+ range
->len
,
10926 cache
->key
.objectid
+ cache
->key
.offset
);
10928 if (end
- start
>= range
->minlen
) {
10929 if (!block_group_cache_done(cache
)) {
10930 ret
= cache_block_group(cache
, 0);
10932 btrfs_put_block_group(cache
);
10935 ret
= wait_block_group_cache_done(cache
);
10937 btrfs_put_block_group(cache
);
10941 ret
= btrfs_trim_block_group(cache
,
10947 trimmed
+= group_trimmed
;
10949 btrfs_put_block_group(cache
);
10954 cache
= next_block_group(fs_info
, cache
);
10957 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
10958 devices
= &fs_info
->fs_devices
->alloc_list
;
10959 list_for_each_entry(device
, devices
, dev_alloc_list
) {
10960 ret
= btrfs_trim_free_extents(device
, range
->minlen
,
10965 trimmed
+= group_trimmed
;
10967 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
10969 range
->len
= trimmed
;
10974 * btrfs_{start,end}_write_no_snapshoting() are similar to
10975 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10976 * data into the page cache through nocow before the subvolume is snapshoted,
10977 * but flush the data into disk after the snapshot creation, or to prevent
10978 * operations while snapshoting is ongoing and that cause the snapshot to be
10979 * inconsistent (writes followed by expanding truncates for example).
10981 void btrfs_end_write_no_snapshoting(struct btrfs_root
*root
)
10983 percpu_counter_dec(&root
->subv_writers
->counter
);
10985 * Make sure counter is updated before we wake up waiters.
10988 if (waitqueue_active(&root
->subv_writers
->wait
))
10989 wake_up(&root
->subv_writers
->wait
);
10992 int btrfs_start_write_no_snapshoting(struct btrfs_root
*root
)
10994 if (atomic_read(&root
->will_be_snapshoted
))
10997 percpu_counter_inc(&root
->subv_writers
->counter
);
10999 * Make sure counter is updated before we check for snapshot creation.
11002 if (atomic_read(&root
->will_be_snapshoted
)) {
11003 btrfs_end_write_no_snapshoting(root
);
11009 static int wait_snapshoting_atomic_t(atomic_t
*a
)
11015 void btrfs_wait_for_snapshot_creation(struct btrfs_root
*root
)
11020 ret
= btrfs_start_write_no_snapshoting(root
);
11023 wait_on_atomic_t(&root
->will_be_snapshoted
,
11024 wait_snapshoting_atomic_t
,
11025 TASK_UNINTERRUPTIBLE
);