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
) {
771 * after adding space to the filesystem, we need to clear the full flags
772 * on all the space infos.
774 void btrfs_clear_space_info_full(struct btrfs_fs_info
*info
)
776 struct list_head
*head
= &info
->space_info
;
777 struct btrfs_space_info
*found
;
780 list_for_each_entry_rcu(found
, head
, list
)
785 /* simple helper to search for an existing data extent at a given offset */
786 int btrfs_lookup_data_extent(struct btrfs_fs_info
*fs_info
, u64 start
, u64 len
)
789 struct btrfs_key key
;
790 struct btrfs_path
*path
;
792 path
= btrfs_alloc_path();
796 key
.objectid
= start
;
798 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
799 ret
= btrfs_search_slot(NULL
, fs_info
->extent_root
, &key
, path
, 0, 0);
800 btrfs_free_path(path
);
805 * helper function to lookup reference count and flags of a tree block.
807 * the head node for delayed ref is used to store the sum of all the
808 * reference count modifications queued up in the rbtree. the head
809 * node may also store the extent flags to set. This way you can check
810 * to see what the reference count and extent flags would be if all of
811 * the delayed refs are not processed.
813 int btrfs_lookup_extent_info(struct btrfs_trans_handle
*trans
,
814 struct btrfs_fs_info
*fs_info
, u64 bytenr
,
815 u64 offset
, int metadata
, u64
*refs
, u64
*flags
)
817 struct btrfs_delayed_ref_head
*head
;
818 struct btrfs_delayed_ref_root
*delayed_refs
;
819 struct btrfs_path
*path
;
820 struct btrfs_extent_item
*ei
;
821 struct extent_buffer
*leaf
;
822 struct btrfs_key key
;
829 * If we don't have skinny metadata, don't bother doing anything
832 if (metadata
&& !btrfs_fs_incompat(fs_info
, SKINNY_METADATA
)) {
833 offset
= fs_info
->nodesize
;
837 path
= btrfs_alloc_path();
842 path
->skip_locking
= 1;
843 path
->search_commit_root
= 1;
847 key
.objectid
= bytenr
;
850 key
.type
= BTRFS_METADATA_ITEM_KEY
;
852 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
854 ret
= btrfs_search_slot(trans
, fs_info
->extent_root
, &key
, path
, 0, 0);
858 if (ret
> 0 && metadata
&& key
.type
== BTRFS_METADATA_ITEM_KEY
) {
859 if (path
->slots
[0]) {
861 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
863 if (key
.objectid
== bytenr
&&
864 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
865 key
.offset
== fs_info
->nodesize
)
871 leaf
= path
->nodes
[0];
872 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
873 if (item_size
>= sizeof(*ei
)) {
874 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
875 struct btrfs_extent_item
);
876 num_refs
= btrfs_extent_refs(leaf
, ei
);
877 extent_flags
= btrfs_extent_flags(leaf
, ei
);
879 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
880 struct btrfs_extent_item_v0
*ei0
;
881 BUG_ON(item_size
!= sizeof(*ei0
));
882 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
883 struct btrfs_extent_item_v0
);
884 num_refs
= btrfs_extent_refs_v0(leaf
, ei0
);
885 /* FIXME: this isn't correct for data */
886 extent_flags
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
891 BUG_ON(num_refs
== 0);
901 delayed_refs
= &trans
->transaction
->delayed_refs
;
902 spin_lock(&delayed_refs
->lock
);
903 head
= btrfs_find_delayed_ref_head(delayed_refs
, bytenr
);
905 if (!mutex_trylock(&head
->mutex
)) {
906 refcount_inc(&head
->node
.refs
);
907 spin_unlock(&delayed_refs
->lock
);
909 btrfs_release_path(path
);
912 * Mutex was contended, block until it's released and try
915 mutex_lock(&head
->mutex
);
916 mutex_unlock(&head
->mutex
);
917 btrfs_put_delayed_ref(&head
->node
);
920 spin_lock(&head
->lock
);
921 if (head
->extent_op
&& head
->extent_op
->update_flags
)
922 extent_flags
|= head
->extent_op
->flags_to_set
;
924 BUG_ON(num_refs
== 0);
926 num_refs
+= head
->node
.ref_mod
;
927 spin_unlock(&head
->lock
);
928 mutex_unlock(&head
->mutex
);
930 spin_unlock(&delayed_refs
->lock
);
932 WARN_ON(num_refs
== 0);
936 *flags
= extent_flags
;
938 btrfs_free_path(path
);
943 * Back reference rules. Back refs have three main goals:
945 * 1) differentiate between all holders of references to an extent so that
946 * when a reference is dropped we can make sure it was a valid reference
947 * before freeing the extent.
949 * 2) Provide enough information to quickly find the holders of an extent
950 * if we notice a given block is corrupted or bad.
952 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
953 * maintenance. This is actually the same as #2, but with a slightly
954 * different use case.
956 * There are two kinds of back refs. The implicit back refs is optimized
957 * for pointers in non-shared tree blocks. For a given pointer in a block,
958 * back refs of this kind provide information about the block's owner tree
959 * and the pointer's key. These information allow us to find the block by
960 * b-tree searching. The full back refs is for pointers in tree blocks not
961 * referenced by their owner trees. The location of tree block is recorded
962 * in the back refs. Actually the full back refs is generic, and can be
963 * used in all cases the implicit back refs is used. The major shortcoming
964 * of the full back refs is its overhead. Every time a tree block gets
965 * COWed, we have to update back refs entry for all pointers in it.
967 * For a newly allocated tree block, we use implicit back refs for
968 * pointers in it. This means most tree related operations only involve
969 * implicit back refs. For a tree block created in old transaction, the
970 * only way to drop a reference to it is COW it. So we can detect the
971 * event that tree block loses its owner tree's reference and do the
972 * back refs conversion.
974 * When a tree block is COWed through a tree, there are four cases:
976 * The reference count of the block is one and the tree is the block's
977 * owner tree. Nothing to do in this case.
979 * The reference count of the block is one and the tree is not the
980 * block's owner tree. In this case, full back refs is used for pointers
981 * in the block. Remove these full back refs, add implicit back refs for
982 * every pointers in the new block.
984 * The reference count of the block is greater than one and the tree is
985 * the block's owner tree. In this case, implicit back refs is used for
986 * pointers in the block. Add full back refs for every pointers in the
987 * block, increase lower level extents' reference counts. The original
988 * implicit back refs are entailed to the new block.
990 * The reference count of the block is greater than one and the tree is
991 * not the block's owner tree. Add implicit back refs for every pointer in
992 * the new block, increase lower level extents' reference count.
994 * Back Reference Key composing:
996 * The key objectid corresponds to the first byte in the extent,
997 * The key type is used to differentiate between types of back refs.
998 * There are different meanings of the key offset for different types
1001 * File extents can be referenced by:
1003 * - multiple snapshots, subvolumes, or different generations in one subvol
1004 * - different files inside a single subvolume
1005 * - different offsets inside a file (bookend extents in file.c)
1007 * The extent ref structure for the implicit back refs has fields for:
1009 * - Objectid of the subvolume root
1010 * - objectid of the file holding the reference
1011 * - original offset in the file
1012 * - how many bookend extents
1014 * The key offset for the implicit back refs is hash of the first
1017 * The extent ref structure for the full back refs has field for:
1019 * - number of pointers in the tree leaf
1021 * The key offset for the implicit back refs is the first byte of
1024 * When a file extent is allocated, The implicit back refs is used.
1025 * the fields are filled in:
1027 * (root_key.objectid, inode objectid, offset in file, 1)
1029 * When a file extent is removed file truncation, we find the
1030 * corresponding implicit back refs and check the following fields:
1032 * (btrfs_header_owner(leaf), inode objectid, offset in file)
1034 * Btree extents can be referenced by:
1036 * - Different subvolumes
1038 * Both the implicit back refs and the full back refs for tree blocks
1039 * only consist of key. The key offset for the implicit back refs is
1040 * objectid of block's owner tree. The key offset for the full back refs
1041 * is the first byte of parent block.
1043 * When implicit back refs is used, information about the lowest key and
1044 * level of the tree block are required. These information are stored in
1045 * tree block info structure.
1048 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1049 static int convert_extent_item_v0(struct btrfs_trans_handle
*trans
,
1050 struct btrfs_fs_info
*fs_info
,
1051 struct btrfs_path
*path
,
1052 u64 owner
, u32 extra_size
)
1054 struct btrfs_root
*root
= fs_info
->extent_root
;
1055 struct btrfs_extent_item
*item
;
1056 struct btrfs_extent_item_v0
*ei0
;
1057 struct btrfs_extent_ref_v0
*ref0
;
1058 struct btrfs_tree_block_info
*bi
;
1059 struct extent_buffer
*leaf
;
1060 struct btrfs_key key
;
1061 struct btrfs_key found_key
;
1062 u32 new_size
= sizeof(*item
);
1066 leaf
= path
->nodes
[0];
1067 BUG_ON(btrfs_item_size_nr(leaf
, path
->slots
[0]) != sizeof(*ei0
));
1069 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1070 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1071 struct btrfs_extent_item_v0
);
1072 refs
= btrfs_extent_refs_v0(leaf
, ei0
);
1074 if (owner
== (u64
)-1) {
1076 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1077 ret
= btrfs_next_leaf(root
, path
);
1080 BUG_ON(ret
> 0); /* Corruption */
1081 leaf
= path
->nodes
[0];
1083 btrfs_item_key_to_cpu(leaf
, &found_key
,
1085 BUG_ON(key
.objectid
!= found_key
.objectid
);
1086 if (found_key
.type
!= BTRFS_EXTENT_REF_V0_KEY
) {
1090 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1091 struct btrfs_extent_ref_v0
);
1092 owner
= btrfs_ref_objectid_v0(leaf
, ref0
);
1096 btrfs_release_path(path
);
1098 if (owner
< BTRFS_FIRST_FREE_OBJECTID
)
1099 new_size
+= sizeof(*bi
);
1101 new_size
-= sizeof(*ei0
);
1102 ret
= btrfs_search_slot(trans
, root
, &key
, path
,
1103 new_size
+ extra_size
, 1);
1106 BUG_ON(ret
); /* Corruption */
1108 btrfs_extend_item(fs_info
, path
, new_size
);
1110 leaf
= path
->nodes
[0];
1111 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1112 btrfs_set_extent_refs(leaf
, item
, refs
);
1113 /* FIXME: get real generation */
1114 btrfs_set_extent_generation(leaf
, item
, 0);
1115 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1116 btrfs_set_extent_flags(leaf
, item
,
1117 BTRFS_EXTENT_FLAG_TREE_BLOCK
|
1118 BTRFS_BLOCK_FLAG_FULL_BACKREF
);
1119 bi
= (struct btrfs_tree_block_info
*)(item
+ 1);
1120 /* FIXME: get first key of the block */
1121 memzero_extent_buffer(leaf
, (unsigned long)bi
, sizeof(*bi
));
1122 btrfs_set_tree_block_level(leaf
, bi
, (int)owner
);
1124 btrfs_set_extent_flags(leaf
, item
, BTRFS_EXTENT_FLAG_DATA
);
1126 btrfs_mark_buffer_dirty(leaf
);
1131 static u64
hash_extent_data_ref(u64 root_objectid
, u64 owner
, u64 offset
)
1133 u32 high_crc
= ~(u32
)0;
1134 u32 low_crc
= ~(u32
)0;
1137 lenum
= cpu_to_le64(root_objectid
);
1138 high_crc
= btrfs_crc32c(high_crc
, &lenum
, sizeof(lenum
));
1139 lenum
= cpu_to_le64(owner
);
1140 low_crc
= btrfs_crc32c(low_crc
, &lenum
, sizeof(lenum
));
1141 lenum
= cpu_to_le64(offset
);
1142 low_crc
= btrfs_crc32c(low_crc
, &lenum
, sizeof(lenum
));
1144 return ((u64
)high_crc
<< 31) ^ (u64
)low_crc
;
1147 static u64
hash_extent_data_ref_item(struct extent_buffer
*leaf
,
1148 struct btrfs_extent_data_ref
*ref
)
1150 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf
, ref
),
1151 btrfs_extent_data_ref_objectid(leaf
, ref
),
1152 btrfs_extent_data_ref_offset(leaf
, ref
));
1155 static int match_extent_data_ref(struct extent_buffer
*leaf
,
1156 struct btrfs_extent_data_ref
*ref
,
1157 u64 root_objectid
, u64 owner
, u64 offset
)
1159 if (btrfs_extent_data_ref_root(leaf
, ref
) != root_objectid
||
1160 btrfs_extent_data_ref_objectid(leaf
, ref
) != owner
||
1161 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
1166 static noinline
int lookup_extent_data_ref(struct btrfs_trans_handle
*trans
,
1167 struct btrfs_fs_info
*fs_info
,
1168 struct btrfs_path
*path
,
1169 u64 bytenr
, u64 parent
,
1171 u64 owner
, u64 offset
)
1173 struct btrfs_root
*root
= fs_info
->extent_root
;
1174 struct btrfs_key key
;
1175 struct btrfs_extent_data_ref
*ref
;
1176 struct extent_buffer
*leaf
;
1182 key
.objectid
= bytenr
;
1184 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1185 key
.offset
= parent
;
1187 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1188 key
.offset
= hash_extent_data_ref(root_objectid
,
1193 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1202 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1203 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1204 btrfs_release_path(path
);
1205 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1216 leaf
= path
->nodes
[0];
1217 nritems
= btrfs_header_nritems(leaf
);
1219 if (path
->slots
[0] >= nritems
) {
1220 ret
= btrfs_next_leaf(root
, path
);
1226 leaf
= path
->nodes
[0];
1227 nritems
= btrfs_header_nritems(leaf
);
1231 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1232 if (key
.objectid
!= bytenr
||
1233 key
.type
!= BTRFS_EXTENT_DATA_REF_KEY
)
1236 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1237 struct btrfs_extent_data_ref
);
1239 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1242 btrfs_release_path(path
);
1254 static noinline
int insert_extent_data_ref(struct btrfs_trans_handle
*trans
,
1255 struct btrfs_fs_info
*fs_info
,
1256 struct btrfs_path
*path
,
1257 u64 bytenr
, u64 parent
,
1258 u64 root_objectid
, u64 owner
,
1259 u64 offset
, int refs_to_add
)
1261 struct btrfs_root
*root
= fs_info
->extent_root
;
1262 struct btrfs_key key
;
1263 struct extent_buffer
*leaf
;
1268 key
.objectid
= bytenr
;
1270 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1271 key
.offset
= parent
;
1272 size
= sizeof(struct btrfs_shared_data_ref
);
1274 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1275 key
.offset
= hash_extent_data_ref(root_objectid
,
1277 size
= sizeof(struct btrfs_extent_data_ref
);
1280 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, size
);
1281 if (ret
&& ret
!= -EEXIST
)
1284 leaf
= path
->nodes
[0];
1286 struct btrfs_shared_data_ref
*ref
;
1287 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1288 struct btrfs_shared_data_ref
);
1290 btrfs_set_shared_data_ref_count(leaf
, ref
, refs_to_add
);
1292 num_refs
= btrfs_shared_data_ref_count(leaf
, ref
);
1293 num_refs
+= refs_to_add
;
1294 btrfs_set_shared_data_ref_count(leaf
, ref
, num_refs
);
1297 struct btrfs_extent_data_ref
*ref
;
1298 while (ret
== -EEXIST
) {
1299 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1300 struct btrfs_extent_data_ref
);
1301 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1304 btrfs_release_path(path
);
1306 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1308 if (ret
&& ret
!= -EEXIST
)
1311 leaf
= path
->nodes
[0];
1313 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1314 struct btrfs_extent_data_ref
);
1316 btrfs_set_extent_data_ref_root(leaf
, ref
,
1318 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
1319 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
1320 btrfs_set_extent_data_ref_count(leaf
, ref
, refs_to_add
);
1322 num_refs
= btrfs_extent_data_ref_count(leaf
, ref
);
1323 num_refs
+= refs_to_add
;
1324 btrfs_set_extent_data_ref_count(leaf
, ref
, num_refs
);
1327 btrfs_mark_buffer_dirty(leaf
);
1330 btrfs_release_path(path
);
1334 static noinline
int remove_extent_data_ref(struct btrfs_trans_handle
*trans
,
1335 struct btrfs_fs_info
*fs_info
,
1336 struct btrfs_path
*path
,
1337 int refs_to_drop
, int *last_ref
)
1339 struct btrfs_key key
;
1340 struct btrfs_extent_data_ref
*ref1
= NULL
;
1341 struct btrfs_shared_data_ref
*ref2
= NULL
;
1342 struct extent_buffer
*leaf
;
1346 leaf
= path
->nodes
[0];
1347 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1349 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1350 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1351 struct btrfs_extent_data_ref
);
1352 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1353 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1354 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1355 struct btrfs_shared_data_ref
);
1356 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1357 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1358 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1359 struct btrfs_extent_ref_v0
*ref0
;
1360 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1361 struct btrfs_extent_ref_v0
);
1362 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1368 BUG_ON(num_refs
< refs_to_drop
);
1369 num_refs
-= refs_to_drop
;
1371 if (num_refs
== 0) {
1372 ret
= btrfs_del_item(trans
, fs_info
->extent_root
, path
);
1375 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
)
1376 btrfs_set_extent_data_ref_count(leaf
, ref1
, num_refs
);
1377 else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
)
1378 btrfs_set_shared_data_ref_count(leaf
, ref2
, num_refs
);
1379 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1381 struct btrfs_extent_ref_v0
*ref0
;
1382 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1383 struct btrfs_extent_ref_v0
);
1384 btrfs_set_ref_count_v0(leaf
, ref0
, num_refs
);
1387 btrfs_mark_buffer_dirty(leaf
);
1392 static noinline u32
extent_data_ref_count(struct btrfs_path
*path
,
1393 struct btrfs_extent_inline_ref
*iref
)
1395 struct btrfs_key key
;
1396 struct extent_buffer
*leaf
;
1397 struct btrfs_extent_data_ref
*ref1
;
1398 struct btrfs_shared_data_ref
*ref2
;
1401 leaf
= path
->nodes
[0];
1402 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1404 if (btrfs_extent_inline_ref_type(leaf
, iref
) ==
1405 BTRFS_EXTENT_DATA_REF_KEY
) {
1406 ref1
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1407 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1409 ref2
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1410 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1412 } else if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1413 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1414 struct btrfs_extent_data_ref
);
1415 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1416 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1417 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1418 struct btrfs_shared_data_ref
);
1419 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1420 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1421 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1422 struct btrfs_extent_ref_v0
*ref0
;
1423 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1424 struct btrfs_extent_ref_v0
);
1425 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1433 static noinline
int lookup_tree_block_ref(struct btrfs_trans_handle
*trans
,
1434 struct btrfs_fs_info
*fs_info
,
1435 struct btrfs_path
*path
,
1436 u64 bytenr
, u64 parent
,
1439 struct btrfs_root
*root
= fs_info
->extent_root
;
1440 struct btrfs_key key
;
1443 key
.objectid
= bytenr
;
1445 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1446 key
.offset
= parent
;
1448 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1449 key
.offset
= root_objectid
;
1452 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1455 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1456 if (ret
== -ENOENT
&& parent
) {
1457 btrfs_release_path(path
);
1458 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1459 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1467 static noinline
int insert_tree_block_ref(struct btrfs_trans_handle
*trans
,
1468 struct btrfs_fs_info
*fs_info
,
1469 struct btrfs_path
*path
,
1470 u64 bytenr
, u64 parent
,
1473 struct btrfs_key key
;
1476 key
.objectid
= bytenr
;
1478 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1479 key
.offset
= parent
;
1481 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1482 key
.offset
= root_objectid
;
1485 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
,
1487 btrfs_release_path(path
);
1491 static inline int extent_ref_type(u64 parent
, u64 owner
)
1494 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1496 type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1498 type
= BTRFS_TREE_BLOCK_REF_KEY
;
1501 type
= BTRFS_SHARED_DATA_REF_KEY
;
1503 type
= BTRFS_EXTENT_DATA_REF_KEY
;
1508 static int find_next_key(struct btrfs_path
*path
, int level
,
1509 struct btrfs_key
*key
)
1512 for (; level
< BTRFS_MAX_LEVEL
; level
++) {
1513 if (!path
->nodes
[level
])
1515 if (path
->slots
[level
] + 1 >=
1516 btrfs_header_nritems(path
->nodes
[level
]))
1519 btrfs_item_key_to_cpu(path
->nodes
[level
], key
,
1520 path
->slots
[level
] + 1);
1522 btrfs_node_key_to_cpu(path
->nodes
[level
], key
,
1523 path
->slots
[level
] + 1);
1530 * look for inline back ref. if back ref is found, *ref_ret is set
1531 * to the address of inline back ref, and 0 is returned.
1533 * if back ref isn't found, *ref_ret is set to the address where it
1534 * should be inserted, and -ENOENT is returned.
1536 * if insert is true and there are too many inline back refs, the path
1537 * points to the extent item, and -EAGAIN is returned.
1539 * NOTE: inline back refs are ordered in the same way that back ref
1540 * items in the tree are ordered.
1542 static noinline_for_stack
1543 int lookup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1544 struct btrfs_fs_info
*fs_info
,
1545 struct btrfs_path
*path
,
1546 struct btrfs_extent_inline_ref
**ref_ret
,
1547 u64 bytenr
, u64 num_bytes
,
1548 u64 parent
, u64 root_objectid
,
1549 u64 owner
, u64 offset
, int insert
)
1551 struct btrfs_root
*root
= fs_info
->extent_root
;
1552 struct btrfs_key key
;
1553 struct extent_buffer
*leaf
;
1554 struct btrfs_extent_item
*ei
;
1555 struct btrfs_extent_inline_ref
*iref
;
1565 bool skinny_metadata
= btrfs_fs_incompat(fs_info
, SKINNY_METADATA
);
1567 key
.objectid
= bytenr
;
1568 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1569 key
.offset
= num_bytes
;
1571 want
= extent_ref_type(parent
, owner
);
1573 extra_size
= btrfs_extent_inline_ref_size(want
);
1574 path
->keep_locks
= 1;
1579 * Owner is our parent level, so we can just add one to get the level
1580 * for the block we are interested in.
1582 if (skinny_metadata
&& owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1583 key
.type
= BTRFS_METADATA_ITEM_KEY
;
1588 ret
= btrfs_search_slot(trans
, root
, &key
, path
, extra_size
, 1);
1595 * We may be a newly converted file system which still has the old fat
1596 * extent entries for metadata, so try and see if we have one of those.
1598 if (ret
> 0 && skinny_metadata
) {
1599 skinny_metadata
= false;
1600 if (path
->slots
[0]) {
1602 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
1604 if (key
.objectid
== bytenr
&&
1605 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
1606 key
.offset
== num_bytes
)
1610 key
.objectid
= bytenr
;
1611 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1612 key
.offset
= num_bytes
;
1613 btrfs_release_path(path
);
1618 if (ret
&& !insert
) {
1621 } else if (WARN_ON(ret
)) {
1626 leaf
= path
->nodes
[0];
1627 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1628 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1629 if (item_size
< sizeof(*ei
)) {
1634 ret
= convert_extent_item_v0(trans
, fs_info
, path
, owner
,
1640 leaf
= path
->nodes
[0];
1641 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1644 BUG_ON(item_size
< sizeof(*ei
));
1646 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1647 flags
= btrfs_extent_flags(leaf
, ei
);
1649 ptr
= (unsigned long)(ei
+ 1);
1650 end
= (unsigned long)ei
+ item_size
;
1652 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
&& !skinny_metadata
) {
1653 ptr
+= sizeof(struct btrfs_tree_block_info
);
1663 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1664 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1668 ptr
+= btrfs_extent_inline_ref_size(type
);
1672 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1673 struct btrfs_extent_data_ref
*dref
;
1674 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1675 if (match_extent_data_ref(leaf
, dref
, root_objectid
,
1680 if (hash_extent_data_ref_item(leaf
, dref
) <
1681 hash_extent_data_ref(root_objectid
, owner
, offset
))
1685 ref_offset
= btrfs_extent_inline_ref_offset(leaf
, iref
);
1687 if (parent
== ref_offset
) {
1691 if (ref_offset
< parent
)
1694 if (root_objectid
== ref_offset
) {
1698 if (ref_offset
< root_objectid
)
1702 ptr
+= btrfs_extent_inline_ref_size(type
);
1704 if (err
== -ENOENT
&& insert
) {
1705 if (item_size
+ extra_size
>=
1706 BTRFS_MAX_EXTENT_ITEM_SIZE(root
)) {
1711 * To add new inline back ref, we have to make sure
1712 * there is no corresponding back ref item.
1713 * For simplicity, we just do not add new inline back
1714 * ref if there is any kind of item for this block
1716 if (find_next_key(path
, 0, &key
) == 0 &&
1717 key
.objectid
== bytenr
&&
1718 key
.type
< BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1723 *ref_ret
= (struct btrfs_extent_inline_ref
*)ptr
;
1726 path
->keep_locks
= 0;
1727 btrfs_unlock_up_safe(path
, 1);
1733 * helper to add new inline back ref
1735 static noinline_for_stack
1736 void setup_inline_extent_backref(struct btrfs_fs_info
*fs_info
,
1737 struct btrfs_path
*path
,
1738 struct btrfs_extent_inline_ref
*iref
,
1739 u64 parent
, u64 root_objectid
,
1740 u64 owner
, u64 offset
, int refs_to_add
,
1741 struct btrfs_delayed_extent_op
*extent_op
)
1743 struct extent_buffer
*leaf
;
1744 struct btrfs_extent_item
*ei
;
1747 unsigned long item_offset
;
1752 leaf
= path
->nodes
[0];
1753 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1754 item_offset
= (unsigned long)iref
- (unsigned long)ei
;
1756 type
= extent_ref_type(parent
, owner
);
1757 size
= btrfs_extent_inline_ref_size(type
);
1759 btrfs_extend_item(fs_info
, path
, size
);
1761 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1762 refs
= btrfs_extent_refs(leaf
, ei
);
1763 refs
+= refs_to_add
;
1764 btrfs_set_extent_refs(leaf
, ei
, refs
);
1766 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1768 ptr
= (unsigned long)ei
+ item_offset
;
1769 end
= (unsigned long)ei
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
1770 if (ptr
< end
- size
)
1771 memmove_extent_buffer(leaf
, ptr
+ size
, ptr
,
1774 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1775 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
1776 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1777 struct btrfs_extent_data_ref
*dref
;
1778 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1779 btrfs_set_extent_data_ref_root(leaf
, dref
, root_objectid
);
1780 btrfs_set_extent_data_ref_objectid(leaf
, dref
, owner
);
1781 btrfs_set_extent_data_ref_offset(leaf
, dref
, offset
);
1782 btrfs_set_extent_data_ref_count(leaf
, dref
, refs_to_add
);
1783 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1784 struct btrfs_shared_data_ref
*sref
;
1785 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1786 btrfs_set_shared_data_ref_count(leaf
, sref
, refs_to_add
);
1787 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1788 } else if (type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
1789 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1791 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
1793 btrfs_mark_buffer_dirty(leaf
);
1796 static int lookup_extent_backref(struct btrfs_trans_handle
*trans
,
1797 struct btrfs_fs_info
*fs_info
,
1798 struct btrfs_path
*path
,
1799 struct btrfs_extent_inline_ref
**ref_ret
,
1800 u64 bytenr
, u64 num_bytes
, u64 parent
,
1801 u64 root_objectid
, u64 owner
, u64 offset
)
1805 ret
= lookup_inline_extent_backref(trans
, fs_info
, path
, ref_ret
,
1806 bytenr
, num_bytes
, parent
,
1807 root_objectid
, owner
, offset
, 0);
1811 btrfs_release_path(path
);
1814 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1815 ret
= lookup_tree_block_ref(trans
, fs_info
, path
, bytenr
,
1816 parent
, root_objectid
);
1818 ret
= lookup_extent_data_ref(trans
, fs_info
, path
, bytenr
,
1819 parent
, root_objectid
, owner
,
1826 * helper to update/remove inline back ref
1828 static noinline_for_stack
1829 void update_inline_extent_backref(struct btrfs_fs_info
*fs_info
,
1830 struct btrfs_path
*path
,
1831 struct btrfs_extent_inline_ref
*iref
,
1833 struct btrfs_delayed_extent_op
*extent_op
,
1836 struct extent_buffer
*leaf
;
1837 struct btrfs_extent_item
*ei
;
1838 struct btrfs_extent_data_ref
*dref
= NULL
;
1839 struct btrfs_shared_data_ref
*sref
= NULL
;
1847 leaf
= path
->nodes
[0];
1848 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1849 refs
= btrfs_extent_refs(leaf
, ei
);
1850 WARN_ON(refs_to_mod
< 0 && refs
+ refs_to_mod
<= 0);
1851 refs
+= refs_to_mod
;
1852 btrfs_set_extent_refs(leaf
, ei
, refs
);
1854 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1856 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1858 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1859 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1860 refs
= btrfs_extent_data_ref_count(leaf
, dref
);
1861 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1862 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1863 refs
= btrfs_shared_data_ref_count(leaf
, sref
);
1866 BUG_ON(refs_to_mod
!= -1);
1869 BUG_ON(refs_to_mod
< 0 && refs
< -refs_to_mod
);
1870 refs
+= refs_to_mod
;
1873 if (type
== BTRFS_EXTENT_DATA_REF_KEY
)
1874 btrfs_set_extent_data_ref_count(leaf
, dref
, refs
);
1876 btrfs_set_shared_data_ref_count(leaf
, sref
, refs
);
1879 size
= btrfs_extent_inline_ref_size(type
);
1880 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1881 ptr
= (unsigned long)iref
;
1882 end
= (unsigned long)ei
+ item_size
;
1883 if (ptr
+ size
< end
)
1884 memmove_extent_buffer(leaf
, ptr
, ptr
+ size
,
1887 btrfs_truncate_item(fs_info
, path
, item_size
, 1);
1889 btrfs_mark_buffer_dirty(leaf
);
1892 static noinline_for_stack
1893 int insert_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1894 struct btrfs_fs_info
*fs_info
,
1895 struct btrfs_path
*path
,
1896 u64 bytenr
, u64 num_bytes
, u64 parent
,
1897 u64 root_objectid
, u64 owner
,
1898 u64 offset
, int refs_to_add
,
1899 struct btrfs_delayed_extent_op
*extent_op
)
1901 struct btrfs_extent_inline_ref
*iref
;
1904 ret
= lookup_inline_extent_backref(trans
, fs_info
, path
, &iref
,
1905 bytenr
, num_bytes
, parent
,
1906 root_objectid
, owner
, offset
, 1);
1908 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
);
1909 update_inline_extent_backref(fs_info
, path
, iref
,
1910 refs_to_add
, extent_op
, NULL
);
1911 } else if (ret
== -ENOENT
) {
1912 setup_inline_extent_backref(fs_info
, path
, iref
, parent
,
1913 root_objectid
, owner
, offset
,
1914 refs_to_add
, extent_op
);
1920 static int insert_extent_backref(struct btrfs_trans_handle
*trans
,
1921 struct btrfs_fs_info
*fs_info
,
1922 struct btrfs_path
*path
,
1923 u64 bytenr
, u64 parent
, u64 root_objectid
,
1924 u64 owner
, u64 offset
, int refs_to_add
)
1927 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1928 BUG_ON(refs_to_add
!= 1);
1929 ret
= insert_tree_block_ref(trans
, fs_info
, path
, bytenr
,
1930 parent
, root_objectid
);
1932 ret
= insert_extent_data_ref(trans
, fs_info
, path
, bytenr
,
1933 parent
, root_objectid
,
1934 owner
, offset
, refs_to_add
);
1939 static int remove_extent_backref(struct btrfs_trans_handle
*trans
,
1940 struct btrfs_fs_info
*fs_info
,
1941 struct btrfs_path
*path
,
1942 struct btrfs_extent_inline_ref
*iref
,
1943 int refs_to_drop
, int is_data
, int *last_ref
)
1947 BUG_ON(!is_data
&& refs_to_drop
!= 1);
1949 update_inline_extent_backref(fs_info
, path
, iref
,
1950 -refs_to_drop
, NULL
, last_ref
);
1951 } else if (is_data
) {
1952 ret
= remove_extent_data_ref(trans
, fs_info
, path
, refs_to_drop
,
1956 ret
= btrfs_del_item(trans
, fs_info
->extent_root
, path
);
1961 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
1962 static int btrfs_issue_discard(struct block_device
*bdev
, u64 start
, u64 len
,
1963 u64
*discarded_bytes
)
1966 u64 bytes_left
, end
;
1967 u64 aligned_start
= ALIGN(start
, 1 << 9);
1969 if (WARN_ON(start
!= aligned_start
)) {
1970 len
-= aligned_start
- start
;
1971 len
= round_down(len
, 1 << 9);
1972 start
= aligned_start
;
1975 *discarded_bytes
= 0;
1983 /* Skip any superblocks on this device. */
1984 for (j
= 0; j
< BTRFS_SUPER_MIRROR_MAX
; j
++) {
1985 u64 sb_start
= btrfs_sb_offset(j
);
1986 u64 sb_end
= sb_start
+ BTRFS_SUPER_INFO_SIZE
;
1987 u64 size
= sb_start
- start
;
1989 if (!in_range(sb_start
, start
, bytes_left
) &&
1990 !in_range(sb_end
, start
, bytes_left
) &&
1991 !in_range(start
, sb_start
, BTRFS_SUPER_INFO_SIZE
))
1995 * Superblock spans beginning of range. Adjust start and
1998 if (sb_start
<= start
) {
1999 start
+= sb_end
- start
;
2004 bytes_left
= end
- start
;
2009 ret
= blkdev_issue_discard(bdev
, start
>> 9, size
>> 9,
2012 *discarded_bytes
+= size
;
2013 else if (ret
!= -EOPNOTSUPP
)
2022 bytes_left
= end
- start
;
2026 ret
= blkdev_issue_discard(bdev
, start
>> 9, bytes_left
>> 9,
2029 *discarded_bytes
+= bytes_left
;
2034 int btrfs_discard_extent(struct btrfs_fs_info
*fs_info
, u64 bytenr
,
2035 u64 num_bytes
, u64
*actual_bytes
)
2038 u64 discarded_bytes
= 0;
2039 struct btrfs_bio
*bbio
= NULL
;
2043 * Avoid races with device replace and make sure our bbio has devices
2044 * associated to its stripes that don't go away while we are discarding.
2046 btrfs_bio_counter_inc_blocked(fs_info
);
2047 /* Tell the block device(s) that the sectors can be discarded */
2048 ret
= btrfs_map_block(fs_info
, BTRFS_MAP_DISCARD
, bytenr
, &num_bytes
,
2050 /* Error condition is -ENOMEM */
2052 struct btrfs_bio_stripe
*stripe
= bbio
->stripes
;
2056 for (i
= 0; i
< bbio
->num_stripes
; i
++, stripe
++) {
2058 if (!stripe
->dev
->can_discard
)
2061 ret
= btrfs_issue_discard(stripe
->dev
->bdev
,
2066 discarded_bytes
+= bytes
;
2067 else if (ret
!= -EOPNOTSUPP
)
2068 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2071 * Just in case we get back EOPNOTSUPP for some reason,
2072 * just ignore the return value so we don't screw up
2073 * people calling discard_extent.
2077 btrfs_put_bbio(bbio
);
2079 btrfs_bio_counter_dec(fs_info
);
2082 *actual_bytes
= discarded_bytes
;
2085 if (ret
== -EOPNOTSUPP
)
2090 /* Can return -ENOMEM */
2091 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
2092 struct btrfs_fs_info
*fs_info
,
2093 u64 bytenr
, u64 num_bytes
, u64 parent
,
2094 u64 root_objectid
, u64 owner
, u64 offset
)
2098 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
&&
2099 root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
2101 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
2102 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
2104 parent
, root_objectid
, (int)owner
,
2105 BTRFS_ADD_DELAYED_REF
, NULL
);
2107 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
2108 num_bytes
, parent
, root_objectid
,
2110 BTRFS_ADD_DELAYED_REF
);
2115 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
2116 struct btrfs_fs_info
*fs_info
,
2117 struct btrfs_delayed_ref_node
*node
,
2118 u64 parent
, u64 root_objectid
,
2119 u64 owner
, u64 offset
, int refs_to_add
,
2120 struct btrfs_delayed_extent_op
*extent_op
)
2122 struct btrfs_path
*path
;
2123 struct extent_buffer
*leaf
;
2124 struct btrfs_extent_item
*item
;
2125 struct btrfs_key key
;
2126 u64 bytenr
= node
->bytenr
;
2127 u64 num_bytes
= node
->num_bytes
;
2131 path
= btrfs_alloc_path();
2135 path
->reada
= READA_FORWARD
;
2136 path
->leave_spinning
= 1;
2137 /* this will setup the path even if it fails to insert the back ref */
2138 ret
= insert_inline_extent_backref(trans
, fs_info
, path
, bytenr
,
2139 num_bytes
, parent
, root_objectid
,
2141 refs_to_add
, extent_op
);
2142 if ((ret
< 0 && ret
!= -EAGAIN
) || !ret
)
2146 * Ok we had -EAGAIN which means we didn't have space to insert and
2147 * inline extent ref, so just update the reference count and add a
2150 leaf
= path
->nodes
[0];
2151 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2152 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2153 refs
= btrfs_extent_refs(leaf
, item
);
2154 btrfs_set_extent_refs(leaf
, item
, refs
+ refs_to_add
);
2156 __run_delayed_extent_op(extent_op
, leaf
, item
);
2158 btrfs_mark_buffer_dirty(leaf
);
2159 btrfs_release_path(path
);
2161 path
->reada
= READA_FORWARD
;
2162 path
->leave_spinning
= 1;
2163 /* now insert the actual backref */
2164 ret
= insert_extent_backref(trans
, fs_info
, path
, bytenr
, parent
,
2165 root_objectid
, owner
, offset
, refs_to_add
);
2167 btrfs_abort_transaction(trans
, ret
);
2169 btrfs_free_path(path
);
2173 static int run_delayed_data_ref(struct btrfs_trans_handle
*trans
,
2174 struct btrfs_fs_info
*fs_info
,
2175 struct btrfs_delayed_ref_node
*node
,
2176 struct btrfs_delayed_extent_op
*extent_op
,
2177 int insert_reserved
)
2180 struct btrfs_delayed_data_ref
*ref
;
2181 struct btrfs_key ins
;
2186 ins
.objectid
= node
->bytenr
;
2187 ins
.offset
= node
->num_bytes
;
2188 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2190 ref
= btrfs_delayed_node_to_data_ref(node
);
2191 trace_run_delayed_data_ref(fs_info
, node
, ref
, node
->action
);
2193 if (node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2194 parent
= ref
->parent
;
2195 ref_root
= ref
->root
;
2197 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2199 flags
|= extent_op
->flags_to_set
;
2200 ret
= alloc_reserved_file_extent(trans
, fs_info
,
2201 parent
, ref_root
, flags
,
2202 ref
->objectid
, ref
->offset
,
2203 &ins
, node
->ref_mod
);
2204 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2205 ret
= __btrfs_inc_extent_ref(trans
, fs_info
, node
, parent
,
2206 ref_root
, ref
->objectid
,
2207 ref
->offset
, node
->ref_mod
,
2209 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2210 ret
= __btrfs_free_extent(trans
, fs_info
, node
, parent
,
2211 ref_root
, ref
->objectid
,
2212 ref
->offset
, node
->ref_mod
,
2220 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
2221 struct extent_buffer
*leaf
,
2222 struct btrfs_extent_item
*ei
)
2224 u64 flags
= btrfs_extent_flags(leaf
, ei
);
2225 if (extent_op
->update_flags
) {
2226 flags
|= extent_op
->flags_to_set
;
2227 btrfs_set_extent_flags(leaf
, ei
, flags
);
2230 if (extent_op
->update_key
) {
2231 struct btrfs_tree_block_info
*bi
;
2232 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
));
2233 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
2234 btrfs_set_tree_block_key(leaf
, bi
, &extent_op
->key
);
2238 static int run_delayed_extent_op(struct btrfs_trans_handle
*trans
,
2239 struct btrfs_fs_info
*fs_info
,
2240 struct btrfs_delayed_ref_node
*node
,
2241 struct btrfs_delayed_extent_op
*extent_op
)
2243 struct btrfs_key key
;
2244 struct btrfs_path
*path
;
2245 struct btrfs_extent_item
*ei
;
2246 struct extent_buffer
*leaf
;
2250 int metadata
= !extent_op
->is_data
;
2255 if (metadata
&& !btrfs_fs_incompat(fs_info
, SKINNY_METADATA
))
2258 path
= btrfs_alloc_path();
2262 key
.objectid
= node
->bytenr
;
2265 key
.type
= BTRFS_METADATA_ITEM_KEY
;
2266 key
.offset
= extent_op
->level
;
2268 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2269 key
.offset
= node
->num_bytes
;
2273 path
->reada
= READA_FORWARD
;
2274 path
->leave_spinning
= 1;
2275 ret
= btrfs_search_slot(trans
, fs_info
->extent_root
, &key
, path
, 0, 1);
2282 if (path
->slots
[0] > 0) {
2284 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
2286 if (key
.objectid
== node
->bytenr
&&
2287 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
2288 key
.offset
== node
->num_bytes
)
2292 btrfs_release_path(path
);
2295 key
.objectid
= node
->bytenr
;
2296 key
.offset
= node
->num_bytes
;
2297 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2306 leaf
= path
->nodes
[0];
2307 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2308 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2309 if (item_size
< sizeof(*ei
)) {
2310 ret
= convert_extent_item_v0(trans
, fs_info
, path
, (u64
)-1, 0);
2315 leaf
= path
->nodes
[0];
2316 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2319 BUG_ON(item_size
< sizeof(*ei
));
2320 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2321 __run_delayed_extent_op(extent_op
, leaf
, ei
);
2323 btrfs_mark_buffer_dirty(leaf
);
2325 btrfs_free_path(path
);
2329 static int run_delayed_tree_ref(struct btrfs_trans_handle
*trans
,
2330 struct btrfs_fs_info
*fs_info
,
2331 struct btrfs_delayed_ref_node
*node
,
2332 struct btrfs_delayed_extent_op
*extent_op
,
2333 int insert_reserved
)
2336 struct btrfs_delayed_tree_ref
*ref
;
2337 struct btrfs_key ins
;
2340 bool skinny_metadata
= btrfs_fs_incompat(fs_info
, SKINNY_METADATA
);
2342 ref
= btrfs_delayed_node_to_tree_ref(node
);
2343 trace_run_delayed_tree_ref(fs_info
, node
, ref
, node
->action
);
2345 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2346 parent
= ref
->parent
;
2347 ref_root
= ref
->root
;
2349 ins
.objectid
= node
->bytenr
;
2350 if (skinny_metadata
) {
2351 ins
.offset
= ref
->level
;
2352 ins
.type
= BTRFS_METADATA_ITEM_KEY
;
2354 ins
.offset
= node
->num_bytes
;
2355 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2358 if (node
->ref_mod
!= 1) {
2360 "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu",
2361 node
->bytenr
, node
->ref_mod
, node
->action
, ref_root
,
2365 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2366 BUG_ON(!extent_op
|| !extent_op
->update_flags
);
2367 ret
= alloc_reserved_tree_block(trans
, fs_info
,
2369 extent_op
->flags_to_set
,
2372 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2373 ret
= __btrfs_inc_extent_ref(trans
, fs_info
, node
,
2377 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2378 ret
= __btrfs_free_extent(trans
, fs_info
, node
,
2380 ref
->level
, 0, 1, extent_op
);
2387 /* helper function to actually process a single delayed ref entry */
2388 static int run_one_delayed_ref(struct btrfs_trans_handle
*trans
,
2389 struct btrfs_fs_info
*fs_info
,
2390 struct btrfs_delayed_ref_node
*node
,
2391 struct btrfs_delayed_extent_op
*extent_op
,
2392 int insert_reserved
)
2396 if (trans
->aborted
) {
2397 if (insert_reserved
)
2398 btrfs_pin_extent(fs_info
, node
->bytenr
,
2399 node
->num_bytes
, 1);
2403 if (btrfs_delayed_ref_is_head(node
)) {
2404 struct btrfs_delayed_ref_head
*head
;
2406 * we've hit the end of the chain and we were supposed
2407 * to insert this extent into the tree. But, it got
2408 * deleted before we ever needed to insert it, so all
2409 * we have to do is clean up the accounting
2412 head
= btrfs_delayed_node_to_head(node
);
2413 trace_run_delayed_ref_head(fs_info
, node
, head
, node
->action
);
2415 if (insert_reserved
) {
2416 btrfs_pin_extent(fs_info
, node
->bytenr
,
2417 node
->num_bytes
, 1);
2418 if (head
->is_data
) {
2419 ret
= btrfs_del_csums(trans
, fs_info
,
2425 /* Also free its reserved qgroup space */
2426 btrfs_qgroup_free_delayed_ref(fs_info
, head
->qgroup_ref_root
,
2427 head
->qgroup_reserved
);
2431 if (node
->type
== BTRFS_TREE_BLOCK_REF_KEY
||
2432 node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2433 ret
= run_delayed_tree_ref(trans
, fs_info
, node
, extent_op
,
2435 else if (node
->type
== BTRFS_EXTENT_DATA_REF_KEY
||
2436 node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2437 ret
= run_delayed_data_ref(trans
, fs_info
, node
, extent_op
,
2444 static inline struct btrfs_delayed_ref_node
*
2445 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
2447 struct btrfs_delayed_ref_node
*ref
;
2449 if (list_empty(&head
->ref_list
))
2453 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2454 * This is to prevent a ref count from going down to zero, which deletes
2455 * the extent item from the extent tree, when there still are references
2456 * to add, which would fail because they would not find the extent item.
2458 if (!list_empty(&head
->ref_add_list
))
2459 return list_first_entry(&head
->ref_add_list
,
2460 struct btrfs_delayed_ref_node
, add_list
);
2462 ref
= list_first_entry(&head
->ref_list
, struct btrfs_delayed_ref_node
,
2464 ASSERT(list_empty(&ref
->add_list
));
2469 * Returns 0 on success or if called with an already aborted transaction.
2470 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2472 static noinline
int __btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2473 struct btrfs_fs_info
*fs_info
,
2476 struct btrfs_delayed_ref_root
*delayed_refs
;
2477 struct btrfs_delayed_ref_node
*ref
;
2478 struct btrfs_delayed_ref_head
*locked_ref
= NULL
;
2479 struct btrfs_delayed_extent_op
*extent_op
;
2480 ktime_t start
= ktime_get();
2482 unsigned long count
= 0;
2483 unsigned long actual_count
= 0;
2484 int must_insert_reserved
= 0;
2486 delayed_refs
= &trans
->transaction
->delayed_refs
;
2492 spin_lock(&delayed_refs
->lock
);
2493 locked_ref
= btrfs_select_ref_head(trans
);
2495 spin_unlock(&delayed_refs
->lock
);
2499 /* grab the lock that says we are going to process
2500 * all the refs for this head */
2501 ret
= btrfs_delayed_ref_lock(trans
, locked_ref
);
2502 spin_unlock(&delayed_refs
->lock
);
2504 * we may have dropped the spin lock to get the head
2505 * mutex lock, and that might have given someone else
2506 * time to free the head. If that's true, it has been
2507 * removed from our list and we can move on.
2509 if (ret
== -EAGAIN
) {
2517 * We need to try and merge add/drops of the same ref since we
2518 * can run into issues with relocate dropping the implicit ref
2519 * and then it being added back again before the drop can
2520 * finish. If we merged anything we need to re-loop so we can
2522 * Or we can get node references of the same type that weren't
2523 * merged when created due to bumps in the tree mod seq, and
2524 * we need to merge them to prevent adding an inline extent
2525 * backref before dropping it (triggering a BUG_ON at
2526 * insert_inline_extent_backref()).
2528 spin_lock(&locked_ref
->lock
);
2529 btrfs_merge_delayed_refs(trans
, fs_info
, delayed_refs
,
2533 * locked_ref is the head node, so we have to go one
2534 * node back for any delayed ref updates
2536 ref
= select_delayed_ref(locked_ref
);
2538 if (ref
&& ref
->seq
&&
2539 btrfs_check_delayed_seq(fs_info
, delayed_refs
, ref
->seq
)) {
2540 spin_unlock(&locked_ref
->lock
);
2541 spin_lock(&delayed_refs
->lock
);
2542 locked_ref
->processing
= 0;
2543 delayed_refs
->num_heads_ready
++;
2544 spin_unlock(&delayed_refs
->lock
);
2545 btrfs_delayed_ref_unlock(locked_ref
);
2553 * record the must insert reserved flag before we
2554 * drop the spin lock.
2556 must_insert_reserved
= locked_ref
->must_insert_reserved
;
2557 locked_ref
->must_insert_reserved
= 0;
2559 extent_op
= locked_ref
->extent_op
;
2560 locked_ref
->extent_op
= NULL
;
2565 /* All delayed refs have been processed, Go ahead
2566 * and send the head node to run_one_delayed_ref,
2567 * so that any accounting fixes can happen
2569 ref
= &locked_ref
->node
;
2571 if (extent_op
&& must_insert_reserved
) {
2572 btrfs_free_delayed_extent_op(extent_op
);
2577 spin_unlock(&locked_ref
->lock
);
2578 ret
= run_delayed_extent_op(trans
, fs_info
,
2580 btrfs_free_delayed_extent_op(extent_op
);
2584 * Need to reset must_insert_reserved if
2585 * there was an error so the abort stuff
2586 * can cleanup the reserved space
2589 if (must_insert_reserved
)
2590 locked_ref
->must_insert_reserved
= 1;
2591 spin_lock(&delayed_refs
->lock
);
2592 locked_ref
->processing
= 0;
2593 delayed_refs
->num_heads_ready
++;
2594 spin_unlock(&delayed_refs
->lock
);
2595 btrfs_debug(fs_info
,
2596 "run_delayed_extent_op returned %d",
2598 btrfs_delayed_ref_unlock(locked_ref
);
2605 * Need to drop our head ref lock and re-acquire the
2606 * delayed ref lock and then re-check to make sure
2609 spin_unlock(&locked_ref
->lock
);
2610 spin_lock(&delayed_refs
->lock
);
2611 spin_lock(&locked_ref
->lock
);
2612 if (!list_empty(&locked_ref
->ref_list
) ||
2613 locked_ref
->extent_op
) {
2614 spin_unlock(&locked_ref
->lock
);
2615 spin_unlock(&delayed_refs
->lock
);
2619 delayed_refs
->num_heads
--;
2620 rb_erase(&locked_ref
->href_node
,
2621 &delayed_refs
->href_root
);
2622 spin_unlock(&delayed_refs
->lock
);
2626 list_del(&ref
->list
);
2627 if (!list_empty(&ref
->add_list
))
2628 list_del(&ref
->add_list
);
2630 atomic_dec(&delayed_refs
->num_entries
);
2632 if (!btrfs_delayed_ref_is_head(ref
)) {
2634 * when we play the delayed ref, also correct the
2637 switch (ref
->action
) {
2638 case BTRFS_ADD_DELAYED_REF
:
2639 case BTRFS_ADD_DELAYED_EXTENT
:
2640 locked_ref
->node
.ref_mod
-= ref
->ref_mod
;
2642 case BTRFS_DROP_DELAYED_REF
:
2643 locked_ref
->node
.ref_mod
+= ref
->ref_mod
;
2649 spin_unlock(&locked_ref
->lock
);
2651 ret
= run_one_delayed_ref(trans
, fs_info
, ref
, extent_op
,
2652 must_insert_reserved
);
2654 btrfs_free_delayed_extent_op(extent_op
);
2656 spin_lock(&delayed_refs
->lock
);
2657 locked_ref
->processing
= 0;
2658 delayed_refs
->num_heads_ready
++;
2659 spin_unlock(&delayed_refs
->lock
);
2660 btrfs_delayed_ref_unlock(locked_ref
);
2661 btrfs_put_delayed_ref(ref
);
2662 btrfs_debug(fs_info
, "run_one_delayed_ref returned %d",
2668 * If this node is a head, that means all the refs in this head
2669 * have been dealt with, and we will pick the next head to deal
2670 * with, so we must unlock the head and drop it from the cluster
2671 * list before we release it.
2673 if (btrfs_delayed_ref_is_head(ref
)) {
2674 if (locked_ref
->is_data
&&
2675 locked_ref
->total_ref_mod
< 0) {
2676 spin_lock(&delayed_refs
->lock
);
2677 delayed_refs
->pending_csums
-= ref
->num_bytes
;
2678 spin_unlock(&delayed_refs
->lock
);
2680 btrfs_delayed_ref_unlock(locked_ref
);
2683 btrfs_put_delayed_ref(ref
);
2689 * We don't want to include ref heads since we can have empty ref heads
2690 * and those will drastically skew our runtime down since we just do
2691 * accounting, no actual extent tree updates.
2693 if (actual_count
> 0) {
2694 u64 runtime
= ktime_to_ns(ktime_sub(ktime_get(), start
));
2698 * We weigh the current average higher than our current runtime
2699 * to avoid large swings in the average.
2701 spin_lock(&delayed_refs
->lock
);
2702 avg
= fs_info
->avg_delayed_ref_runtime
* 3 + runtime
;
2703 fs_info
->avg_delayed_ref_runtime
= avg
>> 2; /* div by 4 */
2704 spin_unlock(&delayed_refs
->lock
);
2709 #ifdef SCRAMBLE_DELAYED_REFS
2711 * Normally delayed refs get processed in ascending bytenr order. This
2712 * correlates in most cases to the order added. To expose dependencies on this
2713 * order, we start to process the tree in the middle instead of the beginning
2715 static u64
find_middle(struct rb_root
*root
)
2717 struct rb_node
*n
= root
->rb_node
;
2718 struct btrfs_delayed_ref_node
*entry
;
2721 u64 first
= 0, last
= 0;
2725 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2726 first
= entry
->bytenr
;
2730 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2731 last
= entry
->bytenr
;
2736 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2737 WARN_ON(!entry
->in_tree
);
2739 middle
= entry
->bytenr
;
2752 static inline u64
heads_to_leaves(struct btrfs_fs_info
*fs_info
, u64 heads
)
2756 num_bytes
= heads
* (sizeof(struct btrfs_extent_item
) +
2757 sizeof(struct btrfs_extent_inline_ref
));
2758 if (!btrfs_fs_incompat(fs_info
, SKINNY_METADATA
))
2759 num_bytes
+= heads
* sizeof(struct btrfs_tree_block_info
);
2762 * We don't ever fill up leaves all the way so multiply by 2 just to be
2763 * closer to what we're really going to want to use.
2765 return div_u64(num_bytes
, BTRFS_LEAF_DATA_SIZE(fs_info
));
2769 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2770 * would require to store the csums for that many bytes.
2772 u64
btrfs_csum_bytes_to_leaves(struct btrfs_fs_info
*fs_info
, u64 csum_bytes
)
2775 u64 num_csums_per_leaf
;
2778 csum_size
= BTRFS_MAX_ITEM_SIZE(fs_info
);
2779 num_csums_per_leaf
= div64_u64(csum_size
,
2780 (u64
)btrfs_super_csum_size(fs_info
->super_copy
));
2781 num_csums
= div64_u64(csum_bytes
, fs_info
->sectorsize
);
2782 num_csums
+= num_csums_per_leaf
- 1;
2783 num_csums
= div64_u64(num_csums
, num_csums_per_leaf
);
2787 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle
*trans
,
2788 struct btrfs_fs_info
*fs_info
)
2790 struct btrfs_block_rsv
*global_rsv
;
2791 u64 num_heads
= trans
->transaction
->delayed_refs
.num_heads_ready
;
2792 u64 csum_bytes
= trans
->transaction
->delayed_refs
.pending_csums
;
2793 u64 num_dirty_bgs
= trans
->transaction
->num_dirty_bgs
;
2794 u64 num_bytes
, num_dirty_bgs_bytes
;
2797 num_bytes
= btrfs_calc_trans_metadata_size(fs_info
, 1);
2798 num_heads
= heads_to_leaves(fs_info
, num_heads
);
2800 num_bytes
+= (num_heads
- 1) * fs_info
->nodesize
;
2802 num_bytes
+= btrfs_csum_bytes_to_leaves(fs_info
, csum_bytes
) *
2804 num_dirty_bgs_bytes
= btrfs_calc_trans_metadata_size(fs_info
,
2806 global_rsv
= &fs_info
->global_block_rsv
;
2809 * If we can't allocate any more chunks lets make sure we have _lots_ of
2810 * wiggle room since running delayed refs can create more delayed refs.
2812 if (global_rsv
->space_info
->full
) {
2813 num_dirty_bgs_bytes
<<= 1;
2817 spin_lock(&global_rsv
->lock
);
2818 if (global_rsv
->reserved
<= num_bytes
+ num_dirty_bgs_bytes
)
2820 spin_unlock(&global_rsv
->lock
);
2824 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle
*trans
,
2825 struct btrfs_fs_info
*fs_info
)
2828 atomic_read(&trans
->transaction
->delayed_refs
.num_entries
);
2833 avg_runtime
= fs_info
->avg_delayed_ref_runtime
;
2834 val
= num_entries
* avg_runtime
;
2835 if (val
>= NSEC_PER_SEC
)
2837 if (val
>= NSEC_PER_SEC
/ 2)
2840 return btrfs_check_space_for_delayed_refs(trans
, fs_info
);
2843 struct async_delayed_refs
{
2844 struct btrfs_root
*root
;
2849 struct completion wait
;
2850 struct btrfs_work work
;
2853 static inline struct async_delayed_refs
*
2854 to_async_delayed_refs(struct btrfs_work
*work
)
2856 return container_of(work
, struct async_delayed_refs
, work
);
2859 static void delayed_ref_async_start(struct btrfs_work
*work
)
2861 struct async_delayed_refs
*async
= to_async_delayed_refs(work
);
2862 struct btrfs_trans_handle
*trans
;
2863 struct btrfs_fs_info
*fs_info
= async
->root
->fs_info
;
2866 /* if the commit is already started, we don't need to wait here */
2867 if (btrfs_transaction_blocked(fs_info
))
2870 trans
= btrfs_join_transaction(async
->root
);
2871 if (IS_ERR(trans
)) {
2872 async
->error
= PTR_ERR(trans
);
2877 * trans->sync means that when we call end_transaction, we won't
2878 * wait on delayed refs
2882 /* Don't bother flushing if we got into a different transaction */
2883 if (trans
->transid
> async
->transid
)
2886 ret
= btrfs_run_delayed_refs(trans
, fs_info
, async
->count
);
2890 ret
= btrfs_end_transaction(trans
);
2891 if (ret
&& !async
->error
)
2895 complete(&async
->wait
);
2900 int btrfs_async_run_delayed_refs(struct btrfs_fs_info
*fs_info
,
2901 unsigned long count
, u64 transid
, int wait
)
2903 struct async_delayed_refs
*async
;
2906 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
2910 async
->root
= fs_info
->tree_root
;
2911 async
->count
= count
;
2913 async
->transid
= transid
;
2918 init_completion(&async
->wait
);
2920 btrfs_init_work(&async
->work
, btrfs_extent_refs_helper
,
2921 delayed_ref_async_start
, NULL
, NULL
);
2923 btrfs_queue_work(fs_info
->extent_workers
, &async
->work
);
2926 wait_for_completion(&async
->wait
);
2935 * this starts processing the delayed reference count updates and
2936 * extent insertions we have queued up so far. count can be
2937 * 0, which means to process everything in the tree at the start
2938 * of the run (but not newly added entries), or it can be some target
2939 * number you'd like to process.
2941 * Returns 0 on success or if called with an aborted transaction
2942 * Returns <0 on error and aborts the transaction
2944 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2945 struct btrfs_fs_info
*fs_info
, unsigned long count
)
2947 struct rb_node
*node
;
2948 struct btrfs_delayed_ref_root
*delayed_refs
;
2949 struct btrfs_delayed_ref_head
*head
;
2951 int run_all
= count
== (unsigned long)-1;
2952 bool can_flush_pending_bgs
= trans
->can_flush_pending_bgs
;
2954 /* We'll clean this up in btrfs_cleanup_transaction */
2958 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE
, &fs_info
->flags
))
2961 delayed_refs
= &trans
->transaction
->delayed_refs
;
2963 count
= atomic_read(&delayed_refs
->num_entries
) * 2;
2966 #ifdef SCRAMBLE_DELAYED_REFS
2967 delayed_refs
->run_delayed_start
= find_middle(&delayed_refs
->root
);
2969 trans
->can_flush_pending_bgs
= false;
2970 ret
= __btrfs_run_delayed_refs(trans
, fs_info
, count
);
2972 btrfs_abort_transaction(trans
, ret
);
2977 if (!list_empty(&trans
->new_bgs
))
2978 btrfs_create_pending_block_groups(trans
, fs_info
);
2980 spin_lock(&delayed_refs
->lock
);
2981 node
= rb_first(&delayed_refs
->href_root
);
2983 spin_unlock(&delayed_refs
->lock
);
2988 head
= rb_entry(node
, struct btrfs_delayed_ref_head
,
2990 if (btrfs_delayed_ref_is_head(&head
->node
)) {
2991 struct btrfs_delayed_ref_node
*ref
;
2994 refcount_inc(&ref
->refs
);
2996 spin_unlock(&delayed_refs
->lock
);
2998 * Mutex was contended, block until it's
2999 * released and try again
3001 mutex_lock(&head
->mutex
);
3002 mutex_unlock(&head
->mutex
);
3004 btrfs_put_delayed_ref(ref
);
3010 node
= rb_next(node
);
3012 spin_unlock(&delayed_refs
->lock
);
3017 trans
->can_flush_pending_bgs
= can_flush_pending_bgs
;
3021 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
3022 struct btrfs_fs_info
*fs_info
,
3023 u64 bytenr
, u64 num_bytes
, u64 flags
,
3024 int level
, int is_data
)
3026 struct btrfs_delayed_extent_op
*extent_op
;
3029 extent_op
= btrfs_alloc_delayed_extent_op();
3033 extent_op
->flags_to_set
= flags
;
3034 extent_op
->update_flags
= true;
3035 extent_op
->update_key
= false;
3036 extent_op
->is_data
= is_data
? true : false;
3037 extent_op
->level
= level
;
3039 ret
= btrfs_add_delayed_extent_op(fs_info
, trans
, bytenr
,
3040 num_bytes
, extent_op
);
3042 btrfs_free_delayed_extent_op(extent_op
);
3046 static noinline
int check_delayed_ref(struct btrfs_root
*root
,
3047 struct btrfs_path
*path
,
3048 u64 objectid
, u64 offset
, u64 bytenr
)
3050 struct btrfs_delayed_ref_head
*head
;
3051 struct btrfs_delayed_ref_node
*ref
;
3052 struct btrfs_delayed_data_ref
*data_ref
;
3053 struct btrfs_delayed_ref_root
*delayed_refs
;
3054 struct btrfs_transaction
*cur_trans
;
3057 cur_trans
= root
->fs_info
->running_transaction
;
3061 delayed_refs
= &cur_trans
->delayed_refs
;
3062 spin_lock(&delayed_refs
->lock
);
3063 head
= btrfs_find_delayed_ref_head(delayed_refs
, bytenr
);
3065 spin_unlock(&delayed_refs
->lock
);
3069 if (!mutex_trylock(&head
->mutex
)) {
3070 refcount_inc(&head
->node
.refs
);
3071 spin_unlock(&delayed_refs
->lock
);
3073 btrfs_release_path(path
);
3076 * Mutex was contended, block until it's released and let
3079 mutex_lock(&head
->mutex
);
3080 mutex_unlock(&head
->mutex
);
3081 btrfs_put_delayed_ref(&head
->node
);
3084 spin_unlock(&delayed_refs
->lock
);
3086 spin_lock(&head
->lock
);
3087 list_for_each_entry(ref
, &head
->ref_list
, list
) {
3088 /* If it's a shared ref we know a cross reference exists */
3089 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
) {
3094 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
3097 * If our ref doesn't match the one we're currently looking at
3098 * then we have a cross reference.
3100 if (data_ref
->root
!= root
->root_key
.objectid
||
3101 data_ref
->objectid
!= objectid
||
3102 data_ref
->offset
!= offset
) {
3107 spin_unlock(&head
->lock
);
3108 mutex_unlock(&head
->mutex
);
3112 static noinline
int check_committed_ref(struct btrfs_root
*root
,
3113 struct btrfs_path
*path
,
3114 u64 objectid
, u64 offset
, u64 bytenr
)
3116 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3117 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
3118 struct extent_buffer
*leaf
;
3119 struct btrfs_extent_data_ref
*ref
;
3120 struct btrfs_extent_inline_ref
*iref
;
3121 struct btrfs_extent_item
*ei
;
3122 struct btrfs_key key
;
3126 key
.objectid
= bytenr
;
3127 key
.offset
= (u64
)-1;
3128 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
3130 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
3133 BUG_ON(ret
== 0); /* Corruption */
3136 if (path
->slots
[0] == 0)
3140 leaf
= path
->nodes
[0];
3141 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
3143 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
3147 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
3148 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3149 if (item_size
< sizeof(*ei
)) {
3150 WARN_ON(item_size
!= sizeof(struct btrfs_extent_item_v0
));
3154 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
3156 if (item_size
!= sizeof(*ei
) +
3157 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
3160 if (btrfs_extent_generation(leaf
, ei
) <=
3161 btrfs_root_last_snapshot(&root
->root_item
))
3164 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
3165 if (btrfs_extent_inline_ref_type(leaf
, iref
) !=
3166 BTRFS_EXTENT_DATA_REF_KEY
)
3169 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
3170 if (btrfs_extent_refs(leaf
, ei
) !=
3171 btrfs_extent_data_ref_count(leaf
, ref
) ||
3172 btrfs_extent_data_ref_root(leaf
, ref
) !=
3173 root
->root_key
.objectid
||
3174 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
3175 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
3183 int btrfs_cross_ref_exist(struct btrfs_root
*root
, u64 objectid
, u64 offset
,
3186 struct btrfs_path
*path
;
3190 path
= btrfs_alloc_path();
3195 ret
= check_committed_ref(root
, path
, objectid
,
3197 if (ret
&& ret
!= -ENOENT
)
3200 ret2
= check_delayed_ref(root
, path
, objectid
,
3202 } while (ret2
== -EAGAIN
);
3204 if (ret2
&& ret2
!= -ENOENT
) {
3209 if (ret
!= -ENOENT
|| ret2
!= -ENOENT
)
3212 btrfs_free_path(path
);
3213 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
3218 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
3219 struct btrfs_root
*root
,
3220 struct extent_buffer
*buf
,
3221 int full_backref
, int inc
)
3223 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3229 struct btrfs_key key
;
3230 struct btrfs_file_extent_item
*fi
;
3234 int (*process_func
)(struct btrfs_trans_handle
*,
3235 struct btrfs_fs_info
*,
3236 u64
, u64
, u64
, u64
, u64
, u64
);
3239 if (btrfs_is_testing(fs_info
))
3242 ref_root
= btrfs_header_owner(buf
);
3243 nritems
= btrfs_header_nritems(buf
);
3244 level
= btrfs_header_level(buf
);
3246 if (!test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) && level
== 0)
3250 process_func
= btrfs_inc_extent_ref
;
3252 process_func
= btrfs_free_extent
;
3255 parent
= buf
->start
;
3259 for (i
= 0; i
< nritems
; i
++) {
3261 btrfs_item_key_to_cpu(buf
, &key
, i
);
3262 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
3264 fi
= btrfs_item_ptr(buf
, i
,
3265 struct btrfs_file_extent_item
);
3266 if (btrfs_file_extent_type(buf
, fi
) ==
3267 BTRFS_FILE_EXTENT_INLINE
)
3269 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
3273 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
3274 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
3275 ret
= process_func(trans
, fs_info
, bytenr
, num_bytes
,
3276 parent
, ref_root
, key
.objectid
,
3281 bytenr
= btrfs_node_blockptr(buf
, i
);
3282 num_bytes
= fs_info
->nodesize
;
3283 ret
= process_func(trans
, fs_info
, bytenr
, num_bytes
,
3284 parent
, ref_root
, level
- 1, 0);
3294 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3295 struct extent_buffer
*buf
, int full_backref
)
3297 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1);
3300 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3301 struct extent_buffer
*buf
, int full_backref
)
3303 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0);
3306 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
3307 struct btrfs_fs_info
*fs_info
,
3308 struct btrfs_path
*path
,
3309 struct btrfs_block_group_cache
*cache
)
3312 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
3314 struct extent_buffer
*leaf
;
3316 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
3323 leaf
= path
->nodes
[0];
3324 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
3325 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
3326 btrfs_mark_buffer_dirty(leaf
);
3328 btrfs_release_path(path
);
3333 static struct btrfs_block_group_cache
*
3334 next_block_group(struct btrfs_fs_info
*fs_info
,
3335 struct btrfs_block_group_cache
*cache
)
3337 struct rb_node
*node
;
3339 spin_lock(&fs_info
->block_group_cache_lock
);
3341 /* If our block group was removed, we need a full search. */
3342 if (RB_EMPTY_NODE(&cache
->cache_node
)) {
3343 const u64 next_bytenr
= cache
->key
.objectid
+ cache
->key
.offset
;
3345 spin_unlock(&fs_info
->block_group_cache_lock
);
3346 btrfs_put_block_group(cache
);
3347 cache
= btrfs_lookup_first_block_group(fs_info
, next_bytenr
); return cache
;
3349 node
= rb_next(&cache
->cache_node
);
3350 btrfs_put_block_group(cache
);
3352 cache
= rb_entry(node
, struct btrfs_block_group_cache
,
3354 btrfs_get_block_group(cache
);
3357 spin_unlock(&fs_info
->block_group_cache_lock
);
3361 static int cache_save_setup(struct btrfs_block_group_cache
*block_group
,
3362 struct btrfs_trans_handle
*trans
,
3363 struct btrfs_path
*path
)
3365 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
3366 struct btrfs_root
*root
= fs_info
->tree_root
;
3367 struct inode
*inode
= NULL
;
3369 int dcs
= BTRFS_DC_ERROR
;
3375 * If this block group is smaller than 100 megs don't bother caching the
3378 if (block_group
->key
.offset
< (100 * SZ_1M
)) {
3379 spin_lock(&block_group
->lock
);
3380 block_group
->disk_cache_state
= BTRFS_DC_WRITTEN
;
3381 spin_unlock(&block_group
->lock
);
3388 inode
= lookup_free_space_inode(fs_info
, block_group
, path
);
3389 if (IS_ERR(inode
) && PTR_ERR(inode
) != -ENOENT
) {
3390 ret
= PTR_ERR(inode
);
3391 btrfs_release_path(path
);
3395 if (IS_ERR(inode
)) {
3399 if (block_group
->ro
)
3402 ret
= create_free_space_inode(fs_info
, trans
, block_group
,
3409 /* We've already setup this transaction, go ahead and exit */
3410 if (block_group
->cache_generation
== trans
->transid
&&
3411 i_size_read(inode
)) {
3412 dcs
= BTRFS_DC_SETUP
;
3417 * We want to set the generation to 0, that way if anything goes wrong
3418 * from here on out we know not to trust this cache when we load up next
3421 BTRFS_I(inode
)->generation
= 0;
3422 ret
= btrfs_update_inode(trans
, root
, inode
);
3425 * So theoretically we could recover from this, simply set the
3426 * super cache generation to 0 so we know to invalidate the
3427 * cache, but then we'd have to keep track of the block groups
3428 * that fail this way so we know we _have_ to reset this cache
3429 * before the next commit or risk reading stale cache. So to
3430 * limit our exposure to horrible edge cases lets just abort the
3431 * transaction, this only happens in really bad situations
3434 btrfs_abort_transaction(trans
, ret
);
3439 if (i_size_read(inode
) > 0) {
3440 ret
= btrfs_check_trunc_cache_free_space(fs_info
,
3441 &fs_info
->global_block_rsv
);
3445 ret
= btrfs_truncate_free_space_cache(trans
, NULL
, inode
);
3450 spin_lock(&block_group
->lock
);
3451 if (block_group
->cached
!= BTRFS_CACHE_FINISHED
||
3452 !btrfs_test_opt(fs_info
, SPACE_CACHE
)) {
3454 * don't bother trying to write stuff out _if_
3455 * a) we're not cached,
3456 * b) we're with nospace_cache mount option,
3457 * c) we're with v2 space_cache (FREE_SPACE_TREE).
3459 dcs
= BTRFS_DC_WRITTEN
;
3460 spin_unlock(&block_group
->lock
);
3463 spin_unlock(&block_group
->lock
);
3466 * We hit an ENOSPC when setting up the cache in this transaction, just
3467 * skip doing the setup, we've already cleared the cache so we're safe.
3469 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC
, &trans
->transaction
->flags
)) {
3475 * Try to preallocate enough space based on how big the block group is.
3476 * Keep in mind this has to include any pinned space which could end up
3477 * taking up quite a bit since it's not folded into the other space
3480 num_pages
= div_u64(block_group
->key
.offset
, SZ_256M
);
3485 num_pages
*= PAGE_SIZE
;
3487 ret
= btrfs_check_data_free_space(inode
, 0, num_pages
);
3491 ret
= btrfs_prealloc_file_range_trans(inode
, trans
, 0, 0, num_pages
,
3492 num_pages
, num_pages
,
3495 * Our cache requires contiguous chunks so that we don't modify a bunch
3496 * of metadata or split extents when writing the cache out, which means
3497 * we can enospc if we are heavily fragmented in addition to just normal
3498 * out of space conditions. So if we hit this just skip setting up any
3499 * other block groups for this transaction, maybe we'll unpin enough
3500 * space the next time around.
3503 dcs
= BTRFS_DC_SETUP
;
3504 else if (ret
== -ENOSPC
)
3505 set_bit(BTRFS_TRANS_CACHE_ENOSPC
, &trans
->transaction
->flags
);
3510 btrfs_release_path(path
);
3512 spin_lock(&block_group
->lock
);
3513 if (!ret
&& dcs
== BTRFS_DC_SETUP
)
3514 block_group
->cache_generation
= trans
->transid
;
3515 block_group
->disk_cache_state
= dcs
;
3516 spin_unlock(&block_group
->lock
);
3521 int btrfs_setup_space_cache(struct btrfs_trans_handle
*trans
,
3522 struct btrfs_fs_info
*fs_info
)
3524 struct btrfs_block_group_cache
*cache
, *tmp
;
3525 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3526 struct btrfs_path
*path
;
3528 if (list_empty(&cur_trans
->dirty_bgs
) ||
3529 !btrfs_test_opt(fs_info
, SPACE_CACHE
))
3532 path
= btrfs_alloc_path();
3536 /* Could add new block groups, use _safe just in case */
3537 list_for_each_entry_safe(cache
, tmp
, &cur_trans
->dirty_bgs
,
3539 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
)
3540 cache_save_setup(cache
, trans
, path
);
3543 btrfs_free_path(path
);
3548 * transaction commit does final block group cache writeback during a
3549 * critical section where nothing is allowed to change the FS. This is
3550 * required in order for the cache to actually match the block group,
3551 * but can introduce a lot of latency into the commit.
3553 * So, btrfs_start_dirty_block_groups is here to kick off block group
3554 * cache IO. There's a chance we'll have to redo some of it if the
3555 * block group changes again during the commit, but it greatly reduces
3556 * the commit latency by getting rid of the easy block groups while
3557 * we're still allowing others to join the commit.
3559 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle
*trans
,
3560 struct btrfs_fs_info
*fs_info
)
3562 struct btrfs_block_group_cache
*cache
;
3563 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3566 struct btrfs_path
*path
= NULL
;
3568 struct list_head
*io
= &cur_trans
->io_bgs
;
3569 int num_started
= 0;
3572 spin_lock(&cur_trans
->dirty_bgs_lock
);
3573 if (list_empty(&cur_trans
->dirty_bgs
)) {
3574 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3577 list_splice_init(&cur_trans
->dirty_bgs
, &dirty
);
3578 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3582 * make sure all the block groups on our dirty list actually
3585 btrfs_create_pending_block_groups(trans
, fs_info
);
3588 path
= btrfs_alloc_path();
3594 * cache_write_mutex is here only to save us from balance or automatic
3595 * removal of empty block groups deleting this block group while we are
3596 * writing out the cache
3598 mutex_lock(&trans
->transaction
->cache_write_mutex
);
3599 while (!list_empty(&dirty
)) {
3600 cache
= list_first_entry(&dirty
,
3601 struct btrfs_block_group_cache
,
3604 * this can happen if something re-dirties a block
3605 * group that is already under IO. Just wait for it to
3606 * finish and then do it all again
3608 if (!list_empty(&cache
->io_list
)) {
3609 list_del_init(&cache
->io_list
);
3610 btrfs_wait_cache_io(trans
, cache
, path
);
3611 btrfs_put_block_group(cache
);
3616 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3617 * if it should update the cache_state. Don't delete
3618 * until after we wait.
3620 * Since we're not running in the commit critical section
3621 * we need the dirty_bgs_lock to protect from update_block_group
3623 spin_lock(&cur_trans
->dirty_bgs_lock
);
3624 list_del_init(&cache
->dirty_list
);
3625 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3629 cache_save_setup(cache
, trans
, path
);
3631 if (cache
->disk_cache_state
== BTRFS_DC_SETUP
) {
3632 cache
->io_ctl
.inode
= NULL
;
3633 ret
= btrfs_write_out_cache(fs_info
, trans
,
3635 if (ret
== 0 && cache
->io_ctl
.inode
) {
3640 * the cache_write_mutex is protecting
3643 list_add_tail(&cache
->io_list
, io
);
3646 * if we failed to write the cache, the
3647 * generation will be bad and life goes on
3653 ret
= write_one_cache_group(trans
, fs_info
,
3656 * Our block group might still be attached to the list
3657 * of new block groups in the transaction handle of some
3658 * other task (struct btrfs_trans_handle->new_bgs). This
3659 * means its block group item isn't yet in the extent
3660 * tree. If this happens ignore the error, as we will
3661 * try again later in the critical section of the
3662 * transaction commit.
3664 if (ret
== -ENOENT
) {
3666 spin_lock(&cur_trans
->dirty_bgs_lock
);
3667 if (list_empty(&cache
->dirty_list
)) {
3668 list_add_tail(&cache
->dirty_list
,
3669 &cur_trans
->dirty_bgs
);
3670 btrfs_get_block_group(cache
);
3672 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3674 btrfs_abort_transaction(trans
, ret
);
3678 /* if its not on the io list, we need to put the block group */
3680 btrfs_put_block_group(cache
);
3686 * Avoid blocking other tasks for too long. It might even save
3687 * us from writing caches for block groups that are going to be
3690 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
3691 mutex_lock(&trans
->transaction
->cache_write_mutex
);
3693 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
3696 * go through delayed refs for all the stuff we've just kicked off
3697 * and then loop back (just once)
3699 ret
= btrfs_run_delayed_refs(trans
, fs_info
, 0);
3700 if (!ret
&& loops
== 0) {
3702 spin_lock(&cur_trans
->dirty_bgs_lock
);
3703 list_splice_init(&cur_trans
->dirty_bgs
, &dirty
);
3705 * dirty_bgs_lock protects us from concurrent block group
3706 * deletes too (not just cache_write_mutex).
3708 if (!list_empty(&dirty
)) {
3709 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3712 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3713 } else if (ret
< 0) {
3714 btrfs_cleanup_dirty_bgs(cur_trans
, fs_info
);
3717 btrfs_free_path(path
);
3721 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
3722 struct btrfs_fs_info
*fs_info
)
3724 struct btrfs_block_group_cache
*cache
;
3725 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3728 struct btrfs_path
*path
;
3729 struct list_head
*io
= &cur_trans
->io_bgs
;
3730 int num_started
= 0;
3732 path
= btrfs_alloc_path();
3737 * Even though we are in the critical section of the transaction commit,
3738 * we can still have concurrent tasks adding elements to this
3739 * transaction's list of dirty block groups. These tasks correspond to
3740 * endio free space workers started when writeback finishes for a
3741 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3742 * allocate new block groups as a result of COWing nodes of the root
3743 * tree when updating the free space inode. The writeback for the space
3744 * caches is triggered by an earlier call to
3745 * btrfs_start_dirty_block_groups() and iterations of the following
3747 * Also we want to do the cache_save_setup first and then run the
3748 * delayed refs to make sure we have the best chance at doing this all
3751 spin_lock(&cur_trans
->dirty_bgs_lock
);
3752 while (!list_empty(&cur_trans
->dirty_bgs
)) {
3753 cache
= list_first_entry(&cur_trans
->dirty_bgs
,
3754 struct btrfs_block_group_cache
,
3758 * this can happen if cache_save_setup re-dirties a block
3759 * group that is already under IO. Just wait for it to
3760 * finish and then do it all again
3762 if (!list_empty(&cache
->io_list
)) {
3763 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3764 list_del_init(&cache
->io_list
);
3765 btrfs_wait_cache_io(trans
, cache
, path
);
3766 btrfs_put_block_group(cache
);
3767 spin_lock(&cur_trans
->dirty_bgs_lock
);
3771 * don't remove from the dirty list until after we've waited
3774 list_del_init(&cache
->dirty_list
);
3775 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3778 cache_save_setup(cache
, trans
, path
);
3781 ret
= btrfs_run_delayed_refs(trans
, fs_info
,
3782 (unsigned long) -1);
3784 if (!ret
&& cache
->disk_cache_state
== BTRFS_DC_SETUP
) {
3785 cache
->io_ctl
.inode
= NULL
;
3786 ret
= btrfs_write_out_cache(fs_info
, trans
,
3788 if (ret
== 0 && cache
->io_ctl
.inode
) {
3791 list_add_tail(&cache
->io_list
, io
);
3794 * if we failed to write the cache, the
3795 * generation will be bad and life goes on
3801 ret
= write_one_cache_group(trans
, fs_info
,
3804 * One of the free space endio workers might have
3805 * created a new block group while updating a free space
3806 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3807 * and hasn't released its transaction handle yet, in
3808 * which case the new block group is still attached to
3809 * its transaction handle and its creation has not
3810 * finished yet (no block group item in the extent tree
3811 * yet, etc). If this is the case, wait for all free
3812 * space endio workers to finish and retry. This is a
3813 * a very rare case so no need for a more efficient and
3816 if (ret
== -ENOENT
) {
3817 wait_event(cur_trans
->writer_wait
,
3818 atomic_read(&cur_trans
->num_writers
) == 1);
3819 ret
= write_one_cache_group(trans
, fs_info
,
3823 btrfs_abort_transaction(trans
, ret
);
3826 /* if its not on the io list, we need to put the block group */
3828 btrfs_put_block_group(cache
);
3829 spin_lock(&cur_trans
->dirty_bgs_lock
);
3831 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3833 while (!list_empty(io
)) {
3834 cache
= list_first_entry(io
, struct btrfs_block_group_cache
,
3836 list_del_init(&cache
->io_list
);
3837 btrfs_wait_cache_io(trans
, cache
, path
);
3838 btrfs_put_block_group(cache
);
3841 btrfs_free_path(path
);
3845 int btrfs_extent_readonly(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
3847 struct btrfs_block_group_cache
*block_group
;
3850 block_group
= btrfs_lookup_block_group(fs_info
, bytenr
);
3851 if (!block_group
|| block_group
->ro
)
3854 btrfs_put_block_group(block_group
);
3858 bool btrfs_inc_nocow_writers(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
3860 struct btrfs_block_group_cache
*bg
;
3863 bg
= btrfs_lookup_block_group(fs_info
, bytenr
);
3867 spin_lock(&bg
->lock
);
3871 atomic_inc(&bg
->nocow_writers
);
3872 spin_unlock(&bg
->lock
);
3874 /* no put on block group, done by btrfs_dec_nocow_writers */
3876 btrfs_put_block_group(bg
);
3882 void btrfs_dec_nocow_writers(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
3884 struct btrfs_block_group_cache
*bg
;
3886 bg
= btrfs_lookup_block_group(fs_info
, bytenr
);
3888 if (atomic_dec_and_test(&bg
->nocow_writers
))
3889 wake_up_atomic_t(&bg
->nocow_writers
);
3891 * Once for our lookup and once for the lookup done by a previous call
3892 * to btrfs_inc_nocow_writers()
3894 btrfs_put_block_group(bg
);
3895 btrfs_put_block_group(bg
);
3898 static int btrfs_wait_nocow_writers_atomic_t(atomic_t
*a
)
3904 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache
*bg
)
3906 wait_on_atomic_t(&bg
->nocow_writers
,
3907 btrfs_wait_nocow_writers_atomic_t
,
3908 TASK_UNINTERRUPTIBLE
);
3911 static const char *alloc_name(u64 flags
)
3914 case BTRFS_BLOCK_GROUP_METADATA
|BTRFS_BLOCK_GROUP_DATA
:
3916 case BTRFS_BLOCK_GROUP_METADATA
:
3918 case BTRFS_BLOCK_GROUP_DATA
:
3920 case BTRFS_BLOCK_GROUP_SYSTEM
:
3924 return "invalid-combination";
3928 static int create_space_info(struct btrfs_fs_info
*info
, u64 flags
,
3929 struct btrfs_space_info
**new)
3932 struct btrfs_space_info
*space_info
;
3936 space_info
= kzalloc(sizeof(*space_info
), GFP_NOFS
);
3940 ret
= percpu_counter_init(&space_info
->total_bytes_pinned
, 0,
3947 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
3948 INIT_LIST_HEAD(&space_info
->block_groups
[i
]);
3949 init_rwsem(&space_info
->groups_sem
);
3950 spin_lock_init(&space_info
->lock
);
3951 space_info
->flags
= flags
& BTRFS_BLOCK_GROUP_TYPE_MASK
;
3952 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3953 init_waitqueue_head(&space_info
->wait
);
3954 INIT_LIST_HEAD(&space_info
->ro_bgs
);
3955 INIT_LIST_HEAD(&space_info
->tickets
);
3956 INIT_LIST_HEAD(&space_info
->priority_tickets
);
3958 ret
= kobject_init_and_add(&space_info
->kobj
, &space_info_ktype
,
3959 info
->space_info_kobj
, "%s",
3960 alloc_name(space_info
->flags
));
3962 percpu_counter_destroy(&space_info
->total_bytes_pinned
);
3968 list_add_rcu(&space_info
->list
, &info
->space_info
);
3969 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3970 info
->data_sinfo
= space_info
;
3975 static void update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
3976 u64 total_bytes
, u64 bytes_used
,
3978 struct btrfs_space_info
**space_info
)
3980 struct btrfs_space_info
*found
;
3983 if (flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3984 BTRFS_BLOCK_GROUP_RAID10
))
3989 found
= __find_space_info(info
, flags
);
3991 spin_lock(&found
->lock
);
3992 found
->total_bytes
+= total_bytes
;
3993 found
->disk_total
+= total_bytes
* factor
;
3994 found
->bytes_used
+= bytes_used
;
3995 found
->disk_used
+= bytes_used
* factor
;
3996 found
->bytes_readonly
+= bytes_readonly
;
3997 if (total_bytes
> 0)
3999 space_info_add_new_bytes(info
, found
, total_bytes
-
4000 bytes_used
- bytes_readonly
);
4001 spin_unlock(&found
->lock
);
4002 *space_info
= found
;
4005 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
4007 u64 extra_flags
= chunk_to_extended(flags
) &
4008 BTRFS_EXTENDED_PROFILE_MASK
;
4010 write_seqlock(&fs_info
->profiles_lock
);
4011 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
4012 fs_info
->avail_data_alloc_bits
|= extra_flags
;
4013 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
4014 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
4015 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
4016 fs_info
->avail_system_alloc_bits
|= extra_flags
;
4017 write_sequnlock(&fs_info
->profiles_lock
);
4021 * returns target flags in extended format or 0 if restripe for this
4022 * chunk_type is not in progress
4024 * should be called with either volume_mutex or balance_lock held
4026 static u64
get_restripe_target(struct btrfs_fs_info
*fs_info
, u64 flags
)
4028 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
4034 if (flags
& BTRFS_BLOCK_GROUP_DATA
&&
4035 bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
4036 target
= BTRFS_BLOCK_GROUP_DATA
| bctl
->data
.target
;
4037 } else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
&&
4038 bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
4039 target
= BTRFS_BLOCK_GROUP_SYSTEM
| bctl
->sys
.target
;
4040 } else if (flags
& BTRFS_BLOCK_GROUP_METADATA
&&
4041 bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
4042 target
= BTRFS_BLOCK_GROUP_METADATA
| bctl
->meta
.target
;
4049 * @flags: available profiles in extended format (see ctree.h)
4051 * Returns reduced profile in chunk format. If profile changing is in
4052 * progress (either running or paused) picks the target profile (if it's
4053 * already available), otherwise falls back to plain reducing.
4055 static u64
btrfs_reduce_alloc_profile(struct btrfs_fs_info
*fs_info
, u64 flags
)
4057 u64 num_devices
= fs_info
->fs_devices
->rw_devices
;
4063 * see if restripe for this chunk_type is in progress, if so
4064 * try to reduce to the target profile
4066 spin_lock(&fs_info
->balance_lock
);
4067 target
= get_restripe_target(fs_info
, flags
);
4069 /* pick target profile only if it's already available */
4070 if ((flags
& target
) & BTRFS_EXTENDED_PROFILE_MASK
) {
4071 spin_unlock(&fs_info
->balance_lock
);
4072 return extended_to_chunk(target
);
4075 spin_unlock(&fs_info
->balance_lock
);
4077 /* First, mask out the RAID levels which aren't possible */
4078 for (raid_type
= 0; raid_type
< BTRFS_NR_RAID_TYPES
; raid_type
++) {
4079 if (num_devices
>= btrfs_raid_array
[raid_type
].devs_min
)
4080 allowed
|= btrfs_raid_group
[raid_type
];
4084 if (allowed
& BTRFS_BLOCK_GROUP_RAID6
)
4085 allowed
= BTRFS_BLOCK_GROUP_RAID6
;
4086 else if (allowed
& BTRFS_BLOCK_GROUP_RAID5
)
4087 allowed
= BTRFS_BLOCK_GROUP_RAID5
;
4088 else if (allowed
& BTRFS_BLOCK_GROUP_RAID10
)
4089 allowed
= BTRFS_BLOCK_GROUP_RAID10
;
4090 else if (allowed
& BTRFS_BLOCK_GROUP_RAID1
)
4091 allowed
= BTRFS_BLOCK_GROUP_RAID1
;
4092 else if (allowed
& BTRFS_BLOCK_GROUP_RAID0
)
4093 allowed
= BTRFS_BLOCK_GROUP_RAID0
;
4095 flags
&= ~BTRFS_BLOCK_GROUP_PROFILE_MASK
;
4097 return extended_to_chunk(flags
| allowed
);
4100 static u64
get_alloc_profile(struct btrfs_fs_info
*fs_info
, u64 orig_flags
)
4107 seq
= read_seqbegin(&fs_info
->profiles_lock
);
4109 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
4110 flags
|= fs_info
->avail_data_alloc_bits
;
4111 else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
4112 flags
|= fs_info
->avail_system_alloc_bits
;
4113 else if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
4114 flags
|= fs_info
->avail_metadata_alloc_bits
;
4115 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
4117 return btrfs_reduce_alloc_profile(fs_info
, flags
);
4120 static u64
get_alloc_profile_by_root(struct btrfs_root
*root
, int data
)
4122 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4127 flags
= BTRFS_BLOCK_GROUP_DATA
;
4128 else if (root
== fs_info
->chunk_root
)
4129 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
4131 flags
= BTRFS_BLOCK_GROUP_METADATA
;
4133 ret
= get_alloc_profile(fs_info
, flags
);
4137 u64
btrfs_data_alloc_profile(struct btrfs_fs_info
*fs_info
)
4139 return get_alloc_profile(fs_info
, BTRFS_BLOCK_GROUP_DATA
);
4142 u64
btrfs_metadata_alloc_profile(struct btrfs_fs_info
*fs_info
)
4144 return get_alloc_profile(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4147 u64
btrfs_system_alloc_profile(struct btrfs_fs_info
*fs_info
)
4149 return get_alloc_profile(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
4152 static u64
btrfs_space_info_used(struct btrfs_space_info
*s_info
,
4153 bool may_use_included
)
4156 return s_info
->bytes_used
+ s_info
->bytes_reserved
+
4157 s_info
->bytes_pinned
+ s_info
->bytes_readonly
+
4158 (may_use_included
? s_info
->bytes_may_use
: 0);
4161 int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode
*inode
, u64 bytes
)
4163 struct btrfs_space_info
*data_sinfo
;
4164 struct btrfs_root
*root
= inode
->root
;
4165 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4168 int need_commit
= 2;
4169 int have_pinned_space
;
4171 /* make sure bytes are sectorsize aligned */
4172 bytes
= ALIGN(bytes
, fs_info
->sectorsize
);
4174 if (btrfs_is_free_space_inode(inode
)) {
4176 ASSERT(current
->journal_info
);
4179 data_sinfo
= fs_info
->data_sinfo
;
4184 /* make sure we have enough space to handle the data first */
4185 spin_lock(&data_sinfo
->lock
);
4186 used
= btrfs_space_info_used(data_sinfo
, true);
4188 if (used
+ bytes
> data_sinfo
->total_bytes
) {
4189 struct btrfs_trans_handle
*trans
;
4192 * if we don't have enough free bytes in this space then we need
4193 * to alloc a new chunk.
4195 if (!data_sinfo
->full
) {
4198 data_sinfo
->force_alloc
= CHUNK_ALLOC_FORCE
;
4199 spin_unlock(&data_sinfo
->lock
);
4201 alloc_target
= btrfs_data_alloc_profile(fs_info
);
4203 * It is ugly that we don't call nolock join
4204 * transaction for the free space inode case here.
4205 * But it is safe because we only do the data space
4206 * reservation for the free space cache in the
4207 * transaction context, the common join transaction
4208 * just increase the counter of the current transaction
4209 * handler, doesn't try to acquire the trans_lock of
4212 trans
= btrfs_join_transaction(root
);
4214 return PTR_ERR(trans
);
4216 ret
= do_chunk_alloc(trans
, fs_info
, alloc_target
,
4217 CHUNK_ALLOC_NO_FORCE
);
4218 btrfs_end_transaction(trans
);
4223 have_pinned_space
= 1;
4229 data_sinfo
= fs_info
->data_sinfo
;
4235 * If we don't have enough pinned space to deal with this
4236 * allocation, and no removed chunk in current transaction,
4237 * don't bother committing the transaction.
4239 have_pinned_space
= percpu_counter_compare(
4240 &data_sinfo
->total_bytes_pinned
,
4241 used
+ bytes
- data_sinfo
->total_bytes
);
4242 spin_unlock(&data_sinfo
->lock
);
4244 /* commit the current transaction and try again */
4247 !atomic_read(&fs_info
->open_ioctl_trans
)) {
4250 if (need_commit
> 0) {
4251 btrfs_start_delalloc_roots(fs_info
, 0, -1);
4252 btrfs_wait_ordered_roots(fs_info
, -1, 0,
4256 trans
= btrfs_join_transaction(root
);
4258 return PTR_ERR(trans
);
4259 if (have_pinned_space
>= 0 ||
4260 test_bit(BTRFS_TRANS_HAVE_FREE_BGS
,
4261 &trans
->transaction
->flags
) ||
4263 ret
= btrfs_commit_transaction(trans
);
4267 * The cleaner kthread might still be doing iput
4268 * operations. Wait for it to finish so that
4269 * more space is released.
4271 mutex_lock(&fs_info
->cleaner_delayed_iput_mutex
);
4272 mutex_unlock(&fs_info
->cleaner_delayed_iput_mutex
);
4275 btrfs_end_transaction(trans
);
4279 trace_btrfs_space_reservation(fs_info
,
4280 "space_info:enospc",
4281 data_sinfo
->flags
, bytes
, 1);
4284 data_sinfo
->bytes_may_use
+= bytes
;
4285 trace_btrfs_space_reservation(fs_info
, "space_info",
4286 data_sinfo
->flags
, bytes
, 1);
4287 spin_unlock(&data_sinfo
->lock
);
4293 * New check_data_free_space() with ability for precious data reservation
4294 * Will replace old btrfs_check_data_free_space(), but for patch split,
4295 * add a new function first and then replace it.
4297 int btrfs_check_data_free_space(struct inode
*inode
, u64 start
, u64 len
)
4299 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
4302 /* align the range */
4303 len
= round_up(start
+ len
, fs_info
->sectorsize
) -
4304 round_down(start
, fs_info
->sectorsize
);
4305 start
= round_down(start
, fs_info
->sectorsize
);
4307 ret
= btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode
), len
);
4311 /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */
4312 ret
= btrfs_qgroup_reserve_data(inode
, start
, len
);
4314 btrfs_free_reserved_data_space_noquota(inode
, start
, len
);
4319 * Called if we need to clear a data reservation for this inode
4320 * Normally in a error case.
4322 * This one will *NOT* use accurate qgroup reserved space API, just for case
4323 * which we can't sleep and is sure it won't affect qgroup reserved space.
4324 * Like clear_bit_hook().
4326 void btrfs_free_reserved_data_space_noquota(struct inode
*inode
, u64 start
,
4329 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
4330 struct btrfs_space_info
*data_sinfo
;
4332 /* Make sure the range is aligned to sectorsize */
4333 len
= round_up(start
+ len
, fs_info
->sectorsize
) -
4334 round_down(start
, fs_info
->sectorsize
);
4335 start
= round_down(start
, fs_info
->sectorsize
);
4337 data_sinfo
= fs_info
->data_sinfo
;
4338 spin_lock(&data_sinfo
->lock
);
4339 if (WARN_ON(data_sinfo
->bytes_may_use
< len
))
4340 data_sinfo
->bytes_may_use
= 0;
4342 data_sinfo
->bytes_may_use
-= len
;
4343 trace_btrfs_space_reservation(fs_info
, "space_info",
4344 data_sinfo
->flags
, len
, 0);
4345 spin_unlock(&data_sinfo
->lock
);
4349 * Called if we need to clear a data reservation for this inode
4350 * Normally in a error case.
4352 * This one will handle the per-inode data rsv map for accurate reserved
4355 void btrfs_free_reserved_data_space(struct inode
*inode
, u64 start
, u64 len
)
4357 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4359 /* Make sure the range is aligned to sectorsize */
4360 len
= round_up(start
+ len
, root
->fs_info
->sectorsize
) -
4361 round_down(start
, root
->fs_info
->sectorsize
);
4362 start
= round_down(start
, root
->fs_info
->sectorsize
);
4364 btrfs_free_reserved_data_space_noquota(inode
, start
, len
);
4365 btrfs_qgroup_free_data(inode
, start
, len
);
4368 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
4370 struct list_head
*head
= &info
->space_info
;
4371 struct btrfs_space_info
*found
;
4374 list_for_each_entry_rcu(found
, head
, list
) {
4375 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
4376 found
->force_alloc
= CHUNK_ALLOC_FORCE
;
4381 static inline u64
calc_global_rsv_need_space(struct btrfs_block_rsv
*global
)
4383 return (global
->size
<< 1);
4386 static int should_alloc_chunk(struct btrfs_fs_info
*fs_info
,
4387 struct btrfs_space_info
*sinfo
, int force
)
4389 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
4390 u64 num_bytes
= sinfo
->total_bytes
- sinfo
->bytes_readonly
;
4391 u64 num_allocated
= sinfo
->bytes_used
+ sinfo
->bytes_reserved
;
4394 if (force
== CHUNK_ALLOC_FORCE
)
4398 * We need to take into account the global rsv because for all intents
4399 * and purposes it's used space. Don't worry about locking the
4400 * global_rsv, it doesn't change except when the transaction commits.
4402 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
4403 num_allocated
+= calc_global_rsv_need_space(global_rsv
);
4406 * in limited mode, we want to have some free space up to
4407 * about 1% of the FS size.
4409 if (force
== CHUNK_ALLOC_LIMITED
) {
4410 thresh
= btrfs_super_total_bytes(fs_info
->super_copy
);
4411 thresh
= max_t(u64
, SZ_64M
, div_factor_fine(thresh
, 1));
4413 if (num_bytes
- num_allocated
< thresh
)
4417 if (num_allocated
+ SZ_2M
< div_factor(num_bytes
, 8))
4422 static u64
get_profile_num_devs(struct btrfs_fs_info
*fs_info
, u64 type
)
4426 if (type
& (BTRFS_BLOCK_GROUP_RAID10
|
4427 BTRFS_BLOCK_GROUP_RAID0
|
4428 BTRFS_BLOCK_GROUP_RAID5
|
4429 BTRFS_BLOCK_GROUP_RAID6
))
4430 num_dev
= fs_info
->fs_devices
->rw_devices
;
4431 else if (type
& BTRFS_BLOCK_GROUP_RAID1
)
4434 num_dev
= 1; /* DUP or single */
4440 * If @is_allocation is true, reserve space in the system space info necessary
4441 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4444 void check_system_chunk(struct btrfs_trans_handle
*trans
,
4445 struct btrfs_fs_info
*fs_info
, u64 type
)
4447 struct btrfs_space_info
*info
;
4454 * Needed because we can end up allocating a system chunk and for an
4455 * atomic and race free space reservation in the chunk block reserve.
4457 ASSERT(mutex_is_locked(&fs_info
->chunk_mutex
));
4459 info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
4460 spin_lock(&info
->lock
);
4461 left
= info
->total_bytes
- btrfs_space_info_used(info
, true);
4462 spin_unlock(&info
->lock
);
4464 num_devs
= get_profile_num_devs(fs_info
, type
);
4466 /* num_devs device items to update and 1 chunk item to add or remove */
4467 thresh
= btrfs_calc_trunc_metadata_size(fs_info
, num_devs
) +
4468 btrfs_calc_trans_metadata_size(fs_info
, 1);
4470 if (left
< thresh
&& btrfs_test_opt(fs_info
, ENOSPC_DEBUG
)) {
4471 btrfs_info(fs_info
, "left=%llu, need=%llu, flags=%llu",
4472 left
, thresh
, type
);
4473 dump_space_info(fs_info
, info
, 0, 0);
4476 if (left
< thresh
) {
4477 u64 flags
= btrfs_system_alloc_profile(fs_info
);
4480 * Ignore failure to create system chunk. We might end up not
4481 * needing it, as we might not need to COW all nodes/leafs from
4482 * the paths we visit in the chunk tree (they were already COWed
4483 * or created in the current transaction for example).
4485 ret
= btrfs_alloc_chunk(trans
, fs_info
, flags
);
4489 ret
= btrfs_block_rsv_add(fs_info
->chunk_root
,
4490 &fs_info
->chunk_block_rsv
,
4491 thresh
, BTRFS_RESERVE_NO_FLUSH
);
4493 trans
->chunk_bytes_reserved
+= thresh
;
4498 * If force is CHUNK_ALLOC_FORCE:
4499 * - return 1 if it successfully allocates a chunk,
4500 * - return errors including -ENOSPC otherwise.
4501 * If force is NOT CHUNK_ALLOC_FORCE:
4502 * - return 0 if it doesn't need to allocate a new chunk,
4503 * - return 1 if it successfully allocates a chunk,
4504 * - return errors including -ENOSPC otherwise.
4506 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
4507 struct btrfs_fs_info
*fs_info
, u64 flags
, int force
)
4509 struct btrfs_space_info
*space_info
;
4510 int wait_for_alloc
= 0;
4513 /* Don't re-enter if we're already allocating a chunk */
4514 if (trans
->allocating_chunk
)
4517 space_info
= __find_space_info(fs_info
, flags
);
4519 ret
= create_space_info(fs_info
, flags
, &space_info
);
4525 spin_lock(&space_info
->lock
);
4526 if (force
< space_info
->force_alloc
)
4527 force
= space_info
->force_alloc
;
4528 if (space_info
->full
) {
4529 if (should_alloc_chunk(fs_info
, space_info
, force
))
4533 spin_unlock(&space_info
->lock
);
4537 if (!should_alloc_chunk(fs_info
, space_info
, force
)) {
4538 spin_unlock(&space_info
->lock
);
4540 } else if (space_info
->chunk_alloc
) {
4543 space_info
->chunk_alloc
= 1;
4546 spin_unlock(&space_info
->lock
);
4548 mutex_lock(&fs_info
->chunk_mutex
);
4551 * The chunk_mutex is held throughout the entirety of a chunk
4552 * allocation, so once we've acquired the chunk_mutex we know that the
4553 * other guy is done and we need to recheck and see if we should
4556 if (wait_for_alloc
) {
4557 mutex_unlock(&fs_info
->chunk_mutex
);
4562 trans
->allocating_chunk
= true;
4565 * If we have mixed data/metadata chunks we want to make sure we keep
4566 * allocating mixed chunks instead of individual chunks.
4568 if (btrfs_mixed_space_info(space_info
))
4569 flags
|= (BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
);
4572 * if we're doing a data chunk, go ahead and make sure that
4573 * we keep a reasonable number of metadata chunks allocated in the
4576 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
4577 fs_info
->data_chunk_allocations
++;
4578 if (!(fs_info
->data_chunk_allocations
%
4579 fs_info
->metadata_ratio
))
4580 force_metadata_allocation(fs_info
);
4584 * Check if we have enough space in SYSTEM chunk because we may need
4585 * to update devices.
4587 check_system_chunk(trans
, fs_info
, flags
);
4589 ret
= btrfs_alloc_chunk(trans
, fs_info
, flags
);
4590 trans
->allocating_chunk
= false;
4592 spin_lock(&space_info
->lock
);
4593 if (ret
< 0 && ret
!= -ENOSPC
)
4596 space_info
->full
= 1;
4600 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
4602 space_info
->chunk_alloc
= 0;
4603 spin_unlock(&space_info
->lock
);
4604 mutex_unlock(&fs_info
->chunk_mutex
);
4606 * When we allocate a new chunk we reserve space in the chunk block
4607 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4608 * add new nodes/leafs to it if we end up needing to do it when
4609 * inserting the chunk item and updating device items as part of the
4610 * second phase of chunk allocation, performed by
4611 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4612 * large number of new block groups to create in our transaction
4613 * handle's new_bgs list to avoid exhausting the chunk block reserve
4614 * in extreme cases - like having a single transaction create many new
4615 * block groups when starting to write out the free space caches of all
4616 * the block groups that were made dirty during the lifetime of the
4619 if (trans
->can_flush_pending_bgs
&&
4620 trans
->chunk_bytes_reserved
>= (u64
)SZ_2M
) {
4621 btrfs_create_pending_block_groups(trans
, fs_info
);
4622 btrfs_trans_release_chunk_metadata(trans
);
4627 static int can_overcommit(struct btrfs_fs_info
*fs_info
,
4628 struct btrfs_space_info
*space_info
, u64 bytes
,
4629 enum btrfs_reserve_flush_enum flush
,
4632 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
4638 /* Don't overcommit when in mixed mode. */
4639 if (space_info
->flags
& BTRFS_BLOCK_GROUP_DATA
)
4643 profile
= btrfs_system_alloc_profile(fs_info
);
4645 profile
= btrfs_metadata_alloc_profile(fs_info
);
4647 used
= btrfs_space_info_used(space_info
, false);
4650 * We only want to allow over committing if we have lots of actual space
4651 * free, but if we don't have enough space to handle the global reserve
4652 * space then we could end up having a real enospc problem when trying
4653 * to allocate a chunk or some other such important allocation.
4655 spin_lock(&global_rsv
->lock
);
4656 space_size
= calc_global_rsv_need_space(global_rsv
);
4657 spin_unlock(&global_rsv
->lock
);
4658 if (used
+ space_size
>= space_info
->total_bytes
)
4661 used
+= space_info
->bytes_may_use
;
4663 avail
= atomic64_read(&fs_info
->free_chunk_space
);
4666 * If we have dup, raid1 or raid10 then only half of the free
4667 * space is actually useable. For raid56, the space info used
4668 * doesn't include the parity drive, so we don't have to
4671 if (profile
& (BTRFS_BLOCK_GROUP_DUP
|
4672 BTRFS_BLOCK_GROUP_RAID1
|
4673 BTRFS_BLOCK_GROUP_RAID10
))
4677 * If we aren't flushing all things, let us overcommit up to
4678 * 1/2th of the space. If we can flush, don't let us overcommit
4679 * too much, let it overcommit up to 1/8 of the space.
4681 if (flush
== BTRFS_RESERVE_FLUSH_ALL
)
4686 if (used
+ bytes
< space_info
->total_bytes
+ avail
)
4691 static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info
*fs_info
,
4692 unsigned long nr_pages
, int nr_items
)
4694 struct super_block
*sb
= fs_info
->sb
;
4696 if (down_read_trylock(&sb
->s_umount
)) {
4697 writeback_inodes_sb_nr(sb
, nr_pages
, WB_REASON_FS_FREE_SPACE
);
4698 up_read(&sb
->s_umount
);
4701 * We needn't worry the filesystem going from r/w to r/o though
4702 * we don't acquire ->s_umount mutex, because the filesystem
4703 * should guarantee the delalloc inodes list be empty after
4704 * the filesystem is readonly(all dirty pages are written to
4707 btrfs_start_delalloc_roots(fs_info
, 0, nr_items
);
4708 if (!current
->journal_info
)
4709 btrfs_wait_ordered_roots(fs_info
, nr_items
, 0, (u64
)-1);
4713 static inline int calc_reclaim_items_nr(struct btrfs_fs_info
*fs_info
,
4719 bytes
= btrfs_calc_trans_metadata_size(fs_info
, 1);
4720 nr
= (int)div64_u64(to_reclaim
, bytes
);
4726 #define EXTENT_SIZE_PER_ITEM SZ_256K
4729 * shrink metadata reservation for delalloc
4731 static void shrink_delalloc(struct btrfs_fs_info
*fs_info
, u64 to_reclaim
,
4732 u64 orig
, bool wait_ordered
)
4734 struct btrfs_block_rsv
*block_rsv
;
4735 struct btrfs_space_info
*space_info
;
4736 struct btrfs_trans_handle
*trans
;
4740 unsigned long nr_pages
;
4743 enum btrfs_reserve_flush_enum flush
;
4745 /* Calc the number of the pages we need flush for space reservation */
4746 items
= calc_reclaim_items_nr(fs_info
, to_reclaim
);
4747 to_reclaim
= (u64
)items
* EXTENT_SIZE_PER_ITEM
;
4749 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4750 block_rsv
= &fs_info
->delalloc_block_rsv
;
4751 space_info
= block_rsv
->space_info
;
4753 delalloc_bytes
= percpu_counter_sum_positive(
4754 &fs_info
->delalloc_bytes
);
4755 if (delalloc_bytes
== 0) {
4759 btrfs_wait_ordered_roots(fs_info
, items
, 0, (u64
)-1);
4764 while (delalloc_bytes
&& loops
< 3) {
4765 max_reclaim
= min(delalloc_bytes
, to_reclaim
);
4766 nr_pages
= max_reclaim
>> PAGE_SHIFT
;
4767 btrfs_writeback_inodes_sb_nr(fs_info
, nr_pages
, items
);
4769 * We need to wait for the async pages to actually start before
4772 max_reclaim
= atomic_read(&fs_info
->async_delalloc_pages
);
4776 if (max_reclaim
<= nr_pages
)
4779 max_reclaim
-= nr_pages
;
4781 wait_event(fs_info
->async_submit_wait
,
4782 atomic_read(&fs_info
->async_delalloc_pages
) <=
4786 flush
= BTRFS_RESERVE_FLUSH_ALL
;
4788 flush
= BTRFS_RESERVE_NO_FLUSH
;
4789 spin_lock(&space_info
->lock
);
4790 if (can_overcommit(fs_info
, space_info
, orig
, flush
, false)) {
4791 spin_unlock(&space_info
->lock
);
4794 if (list_empty(&space_info
->tickets
) &&
4795 list_empty(&space_info
->priority_tickets
)) {
4796 spin_unlock(&space_info
->lock
);
4799 spin_unlock(&space_info
->lock
);
4802 if (wait_ordered
&& !trans
) {
4803 btrfs_wait_ordered_roots(fs_info
, items
, 0, (u64
)-1);
4805 time_left
= schedule_timeout_killable(1);
4809 delalloc_bytes
= percpu_counter_sum_positive(
4810 &fs_info
->delalloc_bytes
);
4815 * maybe_commit_transaction - possibly commit the transaction if its ok to
4816 * @root - the root we're allocating for
4817 * @bytes - the number of bytes we want to reserve
4818 * @force - force the commit
4820 * This will check to make sure that committing the transaction will actually
4821 * get us somewhere and then commit the transaction if it does. Otherwise it
4822 * will return -ENOSPC.
4824 static int may_commit_transaction(struct btrfs_fs_info
*fs_info
,
4825 struct btrfs_space_info
*space_info
,
4826 u64 bytes
, int force
)
4828 struct btrfs_block_rsv
*delayed_rsv
= &fs_info
->delayed_block_rsv
;
4829 struct btrfs_trans_handle
*trans
;
4831 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4838 /* See if there is enough pinned space to make this reservation */
4839 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4844 * See if there is some space in the delayed insertion reservation for
4847 if (space_info
!= delayed_rsv
->space_info
)
4850 spin_lock(&delayed_rsv
->lock
);
4851 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4852 bytes
- delayed_rsv
->size
) < 0) {
4853 spin_unlock(&delayed_rsv
->lock
);
4856 spin_unlock(&delayed_rsv
->lock
);
4859 trans
= btrfs_join_transaction(fs_info
->extent_root
);
4863 return btrfs_commit_transaction(trans
);
4866 struct reserve_ticket
{
4869 struct list_head list
;
4870 wait_queue_head_t wait
;
4873 static int flush_space(struct btrfs_fs_info
*fs_info
,
4874 struct btrfs_space_info
*space_info
, u64 num_bytes
,
4875 u64 orig_bytes
, int state
)
4877 struct btrfs_root
*root
= fs_info
->extent_root
;
4878 struct btrfs_trans_handle
*trans
;
4883 case FLUSH_DELAYED_ITEMS_NR
:
4884 case FLUSH_DELAYED_ITEMS
:
4885 if (state
== FLUSH_DELAYED_ITEMS_NR
)
4886 nr
= calc_reclaim_items_nr(fs_info
, num_bytes
) * 2;
4890 trans
= btrfs_join_transaction(root
);
4891 if (IS_ERR(trans
)) {
4892 ret
= PTR_ERR(trans
);
4895 ret
= btrfs_run_delayed_items_nr(trans
, fs_info
, nr
);
4896 btrfs_end_transaction(trans
);
4898 case FLUSH_DELALLOC
:
4899 case FLUSH_DELALLOC_WAIT
:
4900 shrink_delalloc(fs_info
, num_bytes
* 2, orig_bytes
,
4901 state
== FLUSH_DELALLOC_WAIT
);
4904 trans
= btrfs_join_transaction(root
);
4905 if (IS_ERR(trans
)) {
4906 ret
= PTR_ERR(trans
);
4909 ret
= do_chunk_alloc(trans
, fs_info
,
4910 btrfs_metadata_alloc_profile(fs_info
),
4911 CHUNK_ALLOC_NO_FORCE
);
4912 btrfs_end_transaction(trans
);
4913 if (ret
> 0 || ret
== -ENOSPC
)
4917 ret
= may_commit_transaction(fs_info
, space_info
,
4925 trace_btrfs_flush_space(fs_info
, space_info
->flags
, num_bytes
,
4926 orig_bytes
, state
, ret
);
4931 btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info
*fs_info
,
4932 struct btrfs_space_info
*space_info
,
4935 struct reserve_ticket
*ticket
;
4940 list_for_each_entry(ticket
, &space_info
->tickets
, list
)
4941 to_reclaim
+= ticket
->bytes
;
4942 list_for_each_entry(ticket
, &space_info
->priority_tickets
, list
)
4943 to_reclaim
+= ticket
->bytes
;
4947 to_reclaim
= min_t(u64
, num_online_cpus() * SZ_1M
, SZ_16M
);
4948 if (can_overcommit(fs_info
, space_info
, to_reclaim
,
4949 BTRFS_RESERVE_FLUSH_ALL
, system_chunk
))
4952 used
= btrfs_space_info_used(space_info
, true);
4954 if (can_overcommit(fs_info
, space_info
, SZ_1M
,
4955 BTRFS_RESERVE_FLUSH_ALL
, system_chunk
))
4956 expected
= div_factor_fine(space_info
->total_bytes
, 95);
4958 expected
= div_factor_fine(space_info
->total_bytes
, 90);
4960 if (used
> expected
)
4961 to_reclaim
= used
- expected
;
4964 to_reclaim
= min(to_reclaim
, space_info
->bytes_may_use
+
4965 space_info
->bytes_reserved
);
4969 static inline int need_do_async_reclaim(struct btrfs_fs_info
*fs_info
,
4970 struct btrfs_space_info
*space_info
,
4971 u64 used
, bool system_chunk
)
4973 u64 thresh
= div_factor_fine(space_info
->total_bytes
, 98);
4975 /* If we're just plain full then async reclaim just slows us down. */
4976 if ((space_info
->bytes_used
+ space_info
->bytes_reserved
) >= thresh
)
4979 if (!btrfs_calc_reclaim_metadata_size(fs_info
, space_info
,
4983 return (used
>= thresh
&& !btrfs_fs_closing(fs_info
) &&
4984 !test_bit(BTRFS_FS_STATE_REMOUNTING
, &fs_info
->fs_state
));
4987 static void wake_all_tickets(struct list_head
*head
)
4989 struct reserve_ticket
*ticket
;
4991 while (!list_empty(head
)) {
4992 ticket
= list_first_entry(head
, struct reserve_ticket
, list
);
4993 list_del_init(&ticket
->list
);
4994 ticket
->error
= -ENOSPC
;
4995 wake_up(&ticket
->wait
);
5000 * This is for normal flushers, we can wait all goddamned day if we want to. We
5001 * will loop and continuously try to flush as long as we are making progress.
5002 * We count progress as clearing off tickets each time we have to loop.
5004 static void btrfs_async_reclaim_metadata_space(struct work_struct
*work
)
5006 struct btrfs_fs_info
*fs_info
;
5007 struct btrfs_space_info
*space_info
;
5010 int commit_cycles
= 0;
5011 u64 last_tickets_id
;
5013 fs_info
= container_of(work
, struct btrfs_fs_info
, async_reclaim_work
);
5014 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
5016 spin_lock(&space_info
->lock
);
5017 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
, space_info
,
5020 space_info
->flush
= 0;
5021 spin_unlock(&space_info
->lock
);
5024 last_tickets_id
= space_info
->tickets_id
;
5025 spin_unlock(&space_info
->lock
);
5027 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
5029 struct reserve_ticket
*ticket
;
5032 ret
= flush_space(fs_info
, space_info
, to_reclaim
, to_reclaim
,
5034 spin_lock(&space_info
->lock
);
5035 if (list_empty(&space_info
->tickets
)) {
5036 space_info
->flush
= 0;
5037 spin_unlock(&space_info
->lock
);
5040 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
,
5043 ticket
= list_first_entry(&space_info
->tickets
,
5044 struct reserve_ticket
, list
);
5045 if (last_tickets_id
== space_info
->tickets_id
) {
5048 last_tickets_id
= space_info
->tickets_id
;
5049 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
5054 if (flush_state
> COMMIT_TRANS
) {
5056 if (commit_cycles
> 2) {
5057 wake_all_tickets(&space_info
->tickets
);
5058 space_info
->flush
= 0;
5060 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
5063 spin_unlock(&space_info
->lock
);
5064 } while (flush_state
<= COMMIT_TRANS
);
5067 void btrfs_init_async_reclaim_work(struct work_struct
*work
)
5069 INIT_WORK(work
, btrfs_async_reclaim_metadata_space
);
5072 static void priority_reclaim_metadata_space(struct btrfs_fs_info
*fs_info
,
5073 struct btrfs_space_info
*space_info
,
5074 struct reserve_ticket
*ticket
)
5077 int flush_state
= FLUSH_DELAYED_ITEMS_NR
;
5079 spin_lock(&space_info
->lock
);
5080 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
, space_info
,
5083 spin_unlock(&space_info
->lock
);
5086 spin_unlock(&space_info
->lock
);
5089 flush_space(fs_info
, space_info
, to_reclaim
, to_reclaim
,
5092 spin_lock(&space_info
->lock
);
5093 if (ticket
->bytes
== 0) {
5094 spin_unlock(&space_info
->lock
);
5097 spin_unlock(&space_info
->lock
);
5100 * Priority flushers can't wait on delalloc without
5103 if (flush_state
== FLUSH_DELALLOC
||
5104 flush_state
== FLUSH_DELALLOC_WAIT
)
5105 flush_state
= ALLOC_CHUNK
;
5106 } while (flush_state
< COMMIT_TRANS
);
5109 static int wait_reserve_ticket(struct btrfs_fs_info
*fs_info
,
5110 struct btrfs_space_info
*space_info
,
5111 struct reserve_ticket
*ticket
, u64 orig_bytes
)
5117 spin_lock(&space_info
->lock
);
5118 while (ticket
->bytes
> 0 && ticket
->error
== 0) {
5119 ret
= prepare_to_wait_event(&ticket
->wait
, &wait
, TASK_KILLABLE
);
5124 spin_unlock(&space_info
->lock
);
5128 finish_wait(&ticket
->wait
, &wait
);
5129 spin_lock(&space_info
->lock
);
5132 ret
= ticket
->error
;
5133 if (!list_empty(&ticket
->list
))
5134 list_del_init(&ticket
->list
);
5135 if (ticket
->bytes
&& ticket
->bytes
< orig_bytes
) {
5136 u64 num_bytes
= orig_bytes
- ticket
->bytes
;
5137 space_info
->bytes_may_use
-= num_bytes
;
5138 trace_btrfs_space_reservation(fs_info
, "space_info",
5139 space_info
->flags
, num_bytes
, 0);
5141 spin_unlock(&space_info
->lock
);
5147 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5148 * @root - the root we're allocating for
5149 * @space_info - the space info we want to allocate from
5150 * @orig_bytes - the number of bytes we want
5151 * @flush - whether or not we can flush to make our reservation
5153 * This will reserve orig_bytes number of bytes from the space info associated
5154 * with the block_rsv. If there is not enough space it will make an attempt to
5155 * flush out space to make room. It will do this by flushing delalloc if
5156 * possible or committing the transaction. If flush is 0 then no attempts to
5157 * regain reservations will be made and this will fail if there is not enough
5160 static int __reserve_metadata_bytes(struct btrfs_fs_info
*fs_info
,
5161 struct btrfs_space_info
*space_info
,
5163 enum btrfs_reserve_flush_enum flush
,
5166 struct reserve_ticket ticket
;
5171 ASSERT(!current
->journal_info
|| flush
!= BTRFS_RESERVE_FLUSH_ALL
);
5173 spin_lock(&space_info
->lock
);
5175 used
= btrfs_space_info_used(space_info
, true);
5178 * If we have enough space then hooray, make our reservation and carry
5179 * on. If not see if we can overcommit, and if we can, hooray carry on.
5180 * If not things get more complicated.
5182 if (used
+ orig_bytes
<= space_info
->total_bytes
) {
5183 space_info
->bytes_may_use
+= orig_bytes
;
5184 trace_btrfs_space_reservation(fs_info
, "space_info",
5185 space_info
->flags
, orig_bytes
, 1);
5187 } else if (can_overcommit(fs_info
, space_info
, orig_bytes
, flush
,
5189 space_info
->bytes_may_use
+= orig_bytes
;
5190 trace_btrfs_space_reservation(fs_info
, "space_info",
5191 space_info
->flags
, orig_bytes
, 1);
5196 * If we couldn't make a reservation then setup our reservation ticket
5197 * and kick the async worker if it's not already running.
5199 * If we are a priority flusher then we just need to add our ticket to
5200 * the list and we will do our own flushing further down.
5202 if (ret
&& flush
!= BTRFS_RESERVE_NO_FLUSH
) {
5203 ticket
.bytes
= orig_bytes
;
5205 init_waitqueue_head(&ticket
.wait
);
5206 if (flush
== BTRFS_RESERVE_FLUSH_ALL
) {
5207 list_add_tail(&ticket
.list
, &space_info
->tickets
);
5208 if (!space_info
->flush
) {
5209 space_info
->flush
= 1;
5210 trace_btrfs_trigger_flush(fs_info
,
5214 queue_work(system_unbound_wq
,
5215 &fs_info
->async_reclaim_work
);
5218 list_add_tail(&ticket
.list
,
5219 &space_info
->priority_tickets
);
5221 } else if (!ret
&& space_info
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
5224 * We will do the space reservation dance during log replay,
5225 * which means we won't have fs_info->fs_root set, so don't do
5226 * the async reclaim as we will panic.
5228 if (!test_bit(BTRFS_FS_LOG_RECOVERING
, &fs_info
->flags
) &&
5229 need_do_async_reclaim(fs_info
, space_info
,
5230 used
, system_chunk
) &&
5231 !work_busy(&fs_info
->async_reclaim_work
)) {
5232 trace_btrfs_trigger_flush(fs_info
, space_info
->flags
,
5233 orig_bytes
, flush
, "preempt");
5234 queue_work(system_unbound_wq
,
5235 &fs_info
->async_reclaim_work
);
5238 spin_unlock(&space_info
->lock
);
5239 if (!ret
|| flush
== BTRFS_RESERVE_NO_FLUSH
)
5242 if (flush
== BTRFS_RESERVE_FLUSH_ALL
)
5243 return wait_reserve_ticket(fs_info
, space_info
, &ticket
,
5247 priority_reclaim_metadata_space(fs_info
, space_info
, &ticket
);
5248 spin_lock(&space_info
->lock
);
5250 if (ticket
.bytes
< orig_bytes
) {
5251 u64 num_bytes
= orig_bytes
- ticket
.bytes
;
5252 space_info
->bytes_may_use
-= num_bytes
;
5253 trace_btrfs_space_reservation(fs_info
, "space_info",
5258 list_del_init(&ticket
.list
);
5261 spin_unlock(&space_info
->lock
);
5262 ASSERT(list_empty(&ticket
.list
));
5267 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5268 * @root - the root we're allocating for
5269 * @block_rsv - the block_rsv we're allocating for
5270 * @orig_bytes - the number of bytes we want
5271 * @flush - whether or not we can flush to make our reservation
5273 * This will reserve orgi_bytes number of bytes from the space info associated
5274 * with the block_rsv. If there is not enough space it will make an attempt to
5275 * flush out space to make room. It will do this by flushing delalloc if
5276 * possible or committing the transaction. If flush is 0 then no attempts to
5277 * regain reservations will be made and this will fail if there is not enough
5280 static int reserve_metadata_bytes(struct btrfs_root
*root
,
5281 struct btrfs_block_rsv
*block_rsv
,
5283 enum btrfs_reserve_flush_enum flush
)
5285 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5286 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5288 bool system_chunk
= (root
== fs_info
->chunk_root
);
5290 ret
= __reserve_metadata_bytes(fs_info
, block_rsv
->space_info
,
5291 orig_bytes
, flush
, system_chunk
);
5292 if (ret
== -ENOSPC
&&
5293 unlikely(root
->orphan_cleanup_state
== ORPHAN_CLEANUP_STARTED
)) {
5294 if (block_rsv
!= global_rsv
&&
5295 !block_rsv_use_bytes(global_rsv
, orig_bytes
))
5299 trace_btrfs_space_reservation(fs_info
, "space_info:enospc",
5300 block_rsv
->space_info
->flags
,
5305 static struct btrfs_block_rsv
*get_block_rsv(
5306 const struct btrfs_trans_handle
*trans
,
5307 const struct btrfs_root
*root
)
5309 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5310 struct btrfs_block_rsv
*block_rsv
= NULL
;
5312 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
5313 (root
== fs_info
->csum_root
&& trans
->adding_csums
) ||
5314 (root
== fs_info
->uuid_root
))
5315 block_rsv
= trans
->block_rsv
;
5318 block_rsv
= root
->block_rsv
;
5321 block_rsv
= &fs_info
->empty_block_rsv
;
5326 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
5330 spin_lock(&block_rsv
->lock
);
5331 if (block_rsv
->reserved
>= num_bytes
) {
5332 block_rsv
->reserved
-= num_bytes
;
5333 if (block_rsv
->reserved
< block_rsv
->size
)
5334 block_rsv
->full
= 0;
5337 spin_unlock(&block_rsv
->lock
);
5341 static void block_rsv_add_bytes(struct btrfs_block_rsv
*block_rsv
,
5342 u64 num_bytes
, int update_size
)
5344 spin_lock(&block_rsv
->lock
);
5345 block_rsv
->reserved
+= num_bytes
;
5347 block_rsv
->size
+= num_bytes
;
5348 else if (block_rsv
->reserved
>= block_rsv
->size
)
5349 block_rsv
->full
= 1;
5350 spin_unlock(&block_rsv
->lock
);
5353 int btrfs_cond_migrate_bytes(struct btrfs_fs_info
*fs_info
,
5354 struct btrfs_block_rsv
*dest
, u64 num_bytes
,
5357 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5360 if (global_rsv
->space_info
!= dest
->space_info
)
5363 spin_lock(&global_rsv
->lock
);
5364 min_bytes
= div_factor(global_rsv
->size
, min_factor
);
5365 if (global_rsv
->reserved
< min_bytes
+ num_bytes
) {
5366 spin_unlock(&global_rsv
->lock
);
5369 global_rsv
->reserved
-= num_bytes
;
5370 if (global_rsv
->reserved
< global_rsv
->size
)
5371 global_rsv
->full
= 0;
5372 spin_unlock(&global_rsv
->lock
);
5374 block_rsv_add_bytes(dest
, num_bytes
, 1);
5379 * This is for space we already have accounted in space_info->bytes_may_use, so
5380 * basically when we're returning space from block_rsv's.
5382 static void space_info_add_old_bytes(struct btrfs_fs_info
*fs_info
,
5383 struct btrfs_space_info
*space_info
,
5386 struct reserve_ticket
*ticket
;
5387 struct list_head
*head
;
5389 enum btrfs_reserve_flush_enum flush
= BTRFS_RESERVE_NO_FLUSH
;
5390 bool check_overcommit
= false;
5392 spin_lock(&space_info
->lock
);
5393 head
= &space_info
->priority_tickets
;
5396 * If we are over our limit then we need to check and see if we can
5397 * overcommit, and if we can't then we just need to free up our space
5398 * and not satisfy any requests.
5400 used
= btrfs_space_info_used(space_info
, true);
5401 if (used
- num_bytes
>= space_info
->total_bytes
)
5402 check_overcommit
= true;
5404 while (!list_empty(head
) && num_bytes
) {
5405 ticket
= list_first_entry(head
, struct reserve_ticket
,
5408 * We use 0 bytes because this space is already reserved, so
5409 * adding the ticket space would be a double count.
5411 if (check_overcommit
&&
5412 !can_overcommit(fs_info
, space_info
, 0, flush
, false))
5414 if (num_bytes
>= ticket
->bytes
) {
5415 list_del_init(&ticket
->list
);
5416 num_bytes
-= ticket
->bytes
;
5418 space_info
->tickets_id
++;
5419 wake_up(&ticket
->wait
);
5421 ticket
->bytes
-= num_bytes
;
5426 if (num_bytes
&& head
== &space_info
->priority_tickets
) {
5427 head
= &space_info
->tickets
;
5428 flush
= BTRFS_RESERVE_FLUSH_ALL
;
5431 space_info
->bytes_may_use
-= num_bytes
;
5432 trace_btrfs_space_reservation(fs_info
, "space_info",
5433 space_info
->flags
, num_bytes
, 0);
5434 spin_unlock(&space_info
->lock
);
5438 * This is for newly allocated space that isn't accounted in
5439 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5440 * we use this helper.
5442 static void space_info_add_new_bytes(struct btrfs_fs_info
*fs_info
,
5443 struct btrfs_space_info
*space_info
,
5446 struct reserve_ticket
*ticket
;
5447 struct list_head
*head
= &space_info
->priority_tickets
;
5450 while (!list_empty(head
) && num_bytes
) {
5451 ticket
= list_first_entry(head
, struct reserve_ticket
,
5453 if (num_bytes
>= ticket
->bytes
) {
5454 trace_btrfs_space_reservation(fs_info
, "space_info",
5457 list_del_init(&ticket
->list
);
5458 num_bytes
-= ticket
->bytes
;
5459 space_info
->bytes_may_use
+= ticket
->bytes
;
5461 space_info
->tickets_id
++;
5462 wake_up(&ticket
->wait
);
5464 trace_btrfs_space_reservation(fs_info
, "space_info",
5467 space_info
->bytes_may_use
+= num_bytes
;
5468 ticket
->bytes
-= num_bytes
;
5473 if (num_bytes
&& head
== &space_info
->priority_tickets
) {
5474 head
= &space_info
->tickets
;
5479 static void block_rsv_release_bytes(struct btrfs_fs_info
*fs_info
,
5480 struct btrfs_block_rsv
*block_rsv
,
5481 struct btrfs_block_rsv
*dest
, u64 num_bytes
)
5483 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
5485 spin_lock(&block_rsv
->lock
);
5486 if (num_bytes
== (u64
)-1)
5487 num_bytes
= block_rsv
->size
;
5488 block_rsv
->size
-= num_bytes
;
5489 if (block_rsv
->reserved
>= block_rsv
->size
) {
5490 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
5491 block_rsv
->reserved
= block_rsv
->size
;
5492 block_rsv
->full
= 1;
5496 spin_unlock(&block_rsv
->lock
);
5498 if (num_bytes
> 0) {
5500 spin_lock(&dest
->lock
);
5504 bytes_to_add
= dest
->size
- dest
->reserved
;
5505 bytes_to_add
= min(num_bytes
, bytes_to_add
);
5506 dest
->reserved
+= bytes_to_add
;
5507 if (dest
->reserved
>= dest
->size
)
5509 num_bytes
-= bytes_to_add
;
5511 spin_unlock(&dest
->lock
);
5514 space_info_add_old_bytes(fs_info
, space_info
,
5519 int btrfs_block_rsv_migrate(struct btrfs_block_rsv
*src
,
5520 struct btrfs_block_rsv
*dst
, u64 num_bytes
,
5525 ret
= block_rsv_use_bytes(src
, num_bytes
);
5529 block_rsv_add_bytes(dst
, num_bytes
, update_size
);
5533 void btrfs_init_block_rsv(struct btrfs_block_rsv
*rsv
, unsigned short type
)
5535 memset(rsv
, 0, sizeof(*rsv
));
5536 spin_lock_init(&rsv
->lock
);
5540 struct btrfs_block_rsv
*btrfs_alloc_block_rsv(struct btrfs_fs_info
*fs_info
,
5541 unsigned short type
)
5543 struct btrfs_block_rsv
*block_rsv
;
5545 block_rsv
= kmalloc(sizeof(*block_rsv
), GFP_NOFS
);
5549 btrfs_init_block_rsv(block_rsv
, type
);
5550 block_rsv
->space_info
= __find_space_info(fs_info
,
5551 BTRFS_BLOCK_GROUP_METADATA
);
5555 void btrfs_free_block_rsv(struct btrfs_fs_info
*fs_info
,
5556 struct btrfs_block_rsv
*rsv
)
5560 btrfs_block_rsv_release(fs_info
, rsv
, (u64
)-1);
5564 void __btrfs_free_block_rsv(struct btrfs_block_rsv
*rsv
)
5569 int btrfs_block_rsv_add(struct btrfs_root
*root
,
5570 struct btrfs_block_rsv
*block_rsv
, u64 num_bytes
,
5571 enum btrfs_reserve_flush_enum flush
)
5578 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
5580 block_rsv_add_bytes(block_rsv
, num_bytes
, 1);
5587 int btrfs_block_rsv_check(struct btrfs_block_rsv
*block_rsv
, int min_factor
)
5595 spin_lock(&block_rsv
->lock
);
5596 num_bytes
= div_factor(block_rsv
->size
, min_factor
);
5597 if (block_rsv
->reserved
>= num_bytes
)
5599 spin_unlock(&block_rsv
->lock
);
5604 int btrfs_block_rsv_refill(struct btrfs_root
*root
,
5605 struct btrfs_block_rsv
*block_rsv
, u64 min_reserved
,
5606 enum btrfs_reserve_flush_enum flush
)
5614 spin_lock(&block_rsv
->lock
);
5615 num_bytes
= min_reserved
;
5616 if (block_rsv
->reserved
>= num_bytes
)
5619 num_bytes
-= block_rsv
->reserved
;
5620 spin_unlock(&block_rsv
->lock
);
5625 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
5627 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
5634 void btrfs_block_rsv_release(struct btrfs_fs_info
*fs_info
,
5635 struct btrfs_block_rsv
*block_rsv
,
5638 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5640 if (global_rsv
== block_rsv
||
5641 block_rsv
->space_info
!= global_rsv
->space_info
)
5643 block_rsv_release_bytes(fs_info
, block_rsv
, global_rsv
, num_bytes
);
5646 static void update_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5648 struct btrfs_block_rsv
*block_rsv
= &fs_info
->global_block_rsv
;
5649 struct btrfs_space_info
*sinfo
= block_rsv
->space_info
;
5653 * The global block rsv is based on the size of the extent tree, the
5654 * checksum tree and the root tree. If the fs is empty we want to set
5655 * it to a minimal amount for safety.
5657 num_bytes
= btrfs_root_used(&fs_info
->extent_root
->root_item
) +
5658 btrfs_root_used(&fs_info
->csum_root
->root_item
) +
5659 btrfs_root_used(&fs_info
->tree_root
->root_item
);
5660 num_bytes
= max_t(u64
, num_bytes
, SZ_16M
);
5662 spin_lock(&sinfo
->lock
);
5663 spin_lock(&block_rsv
->lock
);
5665 block_rsv
->size
= min_t(u64
, num_bytes
, SZ_512M
);
5667 if (block_rsv
->reserved
< block_rsv
->size
) {
5668 num_bytes
= btrfs_space_info_used(sinfo
, true);
5669 if (sinfo
->total_bytes
> num_bytes
) {
5670 num_bytes
= sinfo
->total_bytes
- num_bytes
;
5671 num_bytes
= min(num_bytes
,
5672 block_rsv
->size
- block_rsv
->reserved
);
5673 block_rsv
->reserved
+= num_bytes
;
5674 sinfo
->bytes_may_use
+= num_bytes
;
5675 trace_btrfs_space_reservation(fs_info
, "space_info",
5676 sinfo
->flags
, num_bytes
,
5679 } else if (block_rsv
->reserved
> block_rsv
->size
) {
5680 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
5681 sinfo
->bytes_may_use
-= num_bytes
;
5682 trace_btrfs_space_reservation(fs_info
, "space_info",
5683 sinfo
->flags
, num_bytes
, 0);
5684 block_rsv
->reserved
= block_rsv
->size
;
5687 if (block_rsv
->reserved
== block_rsv
->size
)
5688 block_rsv
->full
= 1;
5690 block_rsv
->full
= 0;
5692 spin_unlock(&block_rsv
->lock
);
5693 spin_unlock(&sinfo
->lock
);
5696 static void init_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5698 struct btrfs_space_info
*space_info
;
5700 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
5701 fs_info
->chunk_block_rsv
.space_info
= space_info
;
5703 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
5704 fs_info
->global_block_rsv
.space_info
= space_info
;
5705 fs_info
->delalloc_block_rsv
.space_info
= space_info
;
5706 fs_info
->trans_block_rsv
.space_info
= space_info
;
5707 fs_info
->empty_block_rsv
.space_info
= space_info
;
5708 fs_info
->delayed_block_rsv
.space_info
= space_info
;
5710 fs_info
->extent_root
->block_rsv
= &fs_info
->global_block_rsv
;
5711 fs_info
->csum_root
->block_rsv
= &fs_info
->global_block_rsv
;
5712 fs_info
->dev_root
->block_rsv
= &fs_info
->global_block_rsv
;
5713 fs_info
->tree_root
->block_rsv
= &fs_info
->global_block_rsv
;
5714 if (fs_info
->quota_root
)
5715 fs_info
->quota_root
->block_rsv
= &fs_info
->global_block_rsv
;
5716 fs_info
->chunk_root
->block_rsv
= &fs_info
->chunk_block_rsv
;
5718 update_global_block_rsv(fs_info
);
5721 static void release_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5723 block_rsv_release_bytes(fs_info
, &fs_info
->global_block_rsv
, NULL
,
5725 WARN_ON(fs_info
->delalloc_block_rsv
.size
> 0);
5726 WARN_ON(fs_info
->delalloc_block_rsv
.reserved
> 0);
5727 WARN_ON(fs_info
->trans_block_rsv
.size
> 0);
5728 WARN_ON(fs_info
->trans_block_rsv
.reserved
> 0);
5729 WARN_ON(fs_info
->chunk_block_rsv
.size
> 0);
5730 WARN_ON(fs_info
->chunk_block_rsv
.reserved
> 0);
5731 WARN_ON(fs_info
->delayed_block_rsv
.size
> 0);
5732 WARN_ON(fs_info
->delayed_block_rsv
.reserved
> 0);
5735 void btrfs_trans_release_metadata(struct btrfs_trans_handle
*trans
,
5736 struct btrfs_fs_info
*fs_info
)
5738 if (!trans
->block_rsv
)
5741 if (!trans
->bytes_reserved
)
5744 trace_btrfs_space_reservation(fs_info
, "transaction",
5745 trans
->transid
, trans
->bytes_reserved
, 0);
5746 btrfs_block_rsv_release(fs_info
, trans
->block_rsv
,
5747 trans
->bytes_reserved
);
5748 trans
->bytes_reserved
= 0;
5752 * To be called after all the new block groups attached to the transaction
5753 * handle have been created (btrfs_create_pending_block_groups()).
5755 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle
*trans
)
5757 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
5759 if (!trans
->chunk_bytes_reserved
)
5762 WARN_ON_ONCE(!list_empty(&trans
->new_bgs
));
5764 block_rsv_release_bytes(fs_info
, &fs_info
->chunk_block_rsv
, NULL
,
5765 trans
->chunk_bytes_reserved
);
5766 trans
->chunk_bytes_reserved
= 0;
5769 /* Can only return 0 or -ENOSPC */
5770 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle
*trans
,
5771 struct btrfs_inode
*inode
)
5773 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
5774 struct btrfs_root
*root
= inode
->root
;
5776 * We always use trans->block_rsv here as we will have reserved space
5777 * for our orphan when starting the transaction, using get_block_rsv()
5778 * here will sometimes make us choose the wrong block rsv as we could be
5779 * doing a reloc inode for a non refcounted root.
5781 struct btrfs_block_rsv
*src_rsv
= trans
->block_rsv
;
5782 struct btrfs_block_rsv
*dst_rsv
= root
->orphan_block_rsv
;
5785 * We need to hold space in order to delete our orphan item once we've
5786 * added it, so this takes the reservation so we can release it later
5787 * when we are truly done with the orphan item.
5789 u64 num_bytes
= btrfs_calc_trans_metadata_size(fs_info
, 1);
5791 trace_btrfs_space_reservation(fs_info
, "orphan", btrfs_ino(inode
),
5793 return btrfs_block_rsv_migrate(src_rsv
, dst_rsv
, num_bytes
, 1);
5796 void btrfs_orphan_release_metadata(struct btrfs_inode
*inode
)
5798 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
5799 struct btrfs_root
*root
= inode
->root
;
5800 u64 num_bytes
= btrfs_calc_trans_metadata_size(fs_info
, 1);
5802 trace_btrfs_space_reservation(fs_info
, "orphan", btrfs_ino(inode
),
5804 btrfs_block_rsv_release(fs_info
, root
->orphan_block_rsv
, num_bytes
);
5808 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5809 * root: the root of the parent directory
5810 * rsv: block reservation
5811 * items: the number of items that we need do reservation
5812 * qgroup_reserved: used to return the reserved size in qgroup
5814 * This function is used to reserve the space for snapshot/subvolume
5815 * creation and deletion. Those operations are different with the
5816 * common file/directory operations, they change two fs/file trees
5817 * and root tree, the number of items that the qgroup reserves is
5818 * different with the free space reservation. So we can not use
5819 * the space reservation mechanism in start_transaction().
5821 int btrfs_subvolume_reserve_metadata(struct btrfs_root
*root
,
5822 struct btrfs_block_rsv
*rsv
,
5824 u64
*qgroup_reserved
,
5825 bool use_global_rsv
)
5829 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5830 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5832 if (test_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
)) {
5833 /* One for parent inode, two for dir entries */
5834 num_bytes
= 3 * fs_info
->nodesize
;
5835 ret
= btrfs_qgroup_reserve_meta(root
, num_bytes
, true);
5842 *qgroup_reserved
= num_bytes
;
5844 num_bytes
= btrfs_calc_trans_metadata_size(fs_info
, items
);
5845 rsv
->space_info
= __find_space_info(fs_info
,
5846 BTRFS_BLOCK_GROUP_METADATA
);
5847 ret
= btrfs_block_rsv_add(root
, rsv
, num_bytes
,
5848 BTRFS_RESERVE_FLUSH_ALL
);
5850 if (ret
== -ENOSPC
&& use_global_rsv
)
5851 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
, num_bytes
, 1);
5853 if (ret
&& *qgroup_reserved
)
5854 btrfs_qgroup_free_meta(root
, *qgroup_reserved
);
5859 void btrfs_subvolume_release_metadata(struct btrfs_fs_info
*fs_info
,
5860 struct btrfs_block_rsv
*rsv
)
5862 btrfs_block_rsv_release(fs_info
, rsv
, (u64
)-1);
5866 * drop_outstanding_extent - drop an outstanding extent
5867 * @inode: the inode we're dropping the extent for
5868 * @num_bytes: the number of bytes we're releasing.
5870 * This is called when we are freeing up an outstanding extent, either called
5871 * after an error or after an extent is written. This will return the number of
5872 * reserved extents that need to be freed. This must be called with
5873 * BTRFS_I(inode)->lock held.
5875 static unsigned drop_outstanding_extent(struct btrfs_inode
*inode
,
5878 unsigned drop_inode_space
= 0;
5879 unsigned dropped_extents
= 0;
5880 unsigned num_extents
;
5882 num_extents
= count_max_extents(num_bytes
);
5883 ASSERT(num_extents
);
5884 ASSERT(inode
->outstanding_extents
>= num_extents
);
5885 inode
->outstanding_extents
-= num_extents
;
5887 if (inode
->outstanding_extents
== 0 &&
5888 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5889 &inode
->runtime_flags
))
5890 drop_inode_space
= 1;
5893 * If we have more or the same amount of outstanding extents than we have
5894 * reserved then we need to leave the reserved extents count alone.
5896 if (inode
->outstanding_extents
>= inode
->reserved_extents
)
5897 return drop_inode_space
;
5899 dropped_extents
= inode
->reserved_extents
- inode
->outstanding_extents
;
5900 inode
->reserved_extents
-= dropped_extents
;
5901 return dropped_extents
+ drop_inode_space
;
5905 * calc_csum_metadata_size - return the amount of metadata space that must be
5906 * reserved/freed for the given bytes.
5907 * @inode: the inode we're manipulating
5908 * @num_bytes: the number of bytes in question
5909 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5911 * This adjusts the number of csum_bytes in the inode and then returns the
5912 * correct amount of metadata that must either be reserved or freed. We
5913 * calculate how many checksums we can fit into one leaf and then divide the
5914 * number of bytes that will need to be checksumed by this value to figure out
5915 * how many checksums will be required. If we are adding bytes then the number
5916 * may go up and we will return the number of additional bytes that must be
5917 * reserved. If it is going down we will return the number of bytes that must
5920 * This must be called with BTRFS_I(inode)->lock held.
5922 static u64
calc_csum_metadata_size(struct btrfs_inode
*inode
, u64 num_bytes
,
5925 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
5926 u64 old_csums
, num_csums
;
5928 if (inode
->flags
& BTRFS_INODE_NODATASUM
&& inode
->csum_bytes
== 0)
5931 old_csums
= btrfs_csum_bytes_to_leaves(fs_info
, inode
->csum_bytes
);
5933 inode
->csum_bytes
+= num_bytes
;
5935 inode
->csum_bytes
-= num_bytes
;
5936 num_csums
= btrfs_csum_bytes_to_leaves(fs_info
, inode
->csum_bytes
);
5938 /* No change, no need to reserve more */
5939 if (old_csums
== num_csums
)
5943 return btrfs_calc_trans_metadata_size(fs_info
,
5944 num_csums
- old_csums
);
5946 return btrfs_calc_trans_metadata_size(fs_info
, old_csums
- num_csums
);
5949 int btrfs_delalloc_reserve_metadata(struct btrfs_inode
*inode
, u64 num_bytes
)
5951 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
5952 struct btrfs_root
*root
= inode
->root
;
5953 struct btrfs_block_rsv
*block_rsv
= &fs_info
->delalloc_block_rsv
;
5956 unsigned nr_extents
;
5957 enum btrfs_reserve_flush_enum flush
= BTRFS_RESERVE_FLUSH_ALL
;
5959 bool delalloc_lock
= true;
5962 bool release_extra
= false;
5964 /* If we are a free space inode we need to not flush since we will be in
5965 * the middle of a transaction commit. We also don't need the delalloc
5966 * mutex since we won't race with anybody. We need this mostly to make
5967 * lockdep shut its filthy mouth.
5969 * If we have a transaction open (can happen if we call truncate_block
5970 * from truncate), then we need FLUSH_LIMIT so we don't deadlock.
5972 if (btrfs_is_free_space_inode(inode
)) {
5973 flush
= BTRFS_RESERVE_NO_FLUSH
;
5974 delalloc_lock
= false;
5975 } else if (current
->journal_info
) {
5976 flush
= BTRFS_RESERVE_FLUSH_LIMIT
;
5979 if (flush
!= BTRFS_RESERVE_NO_FLUSH
&&
5980 btrfs_transaction_in_commit(fs_info
))
5981 schedule_timeout(1);
5984 mutex_lock(&inode
->delalloc_mutex
);
5986 num_bytes
= ALIGN(num_bytes
, fs_info
->sectorsize
);
5988 spin_lock(&inode
->lock
);
5989 nr_extents
= count_max_extents(num_bytes
);
5990 inode
->outstanding_extents
+= nr_extents
;
5993 if (inode
->outstanding_extents
> inode
->reserved_extents
)
5994 nr_extents
+= inode
->outstanding_extents
-
5995 inode
->reserved_extents
;
5997 /* We always want to reserve a slot for updating the inode. */
5998 to_reserve
= btrfs_calc_trans_metadata_size(fs_info
, nr_extents
+ 1);
5999 to_reserve
+= calc_csum_metadata_size(inode
, num_bytes
, 1);
6000 csum_bytes
= inode
->csum_bytes
;
6001 spin_unlock(&inode
->lock
);
6003 if (test_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
)) {
6004 ret
= btrfs_qgroup_reserve_meta(root
,
6005 nr_extents
* fs_info
->nodesize
, true);
6010 ret
= btrfs_block_rsv_add(root
, block_rsv
, to_reserve
, flush
);
6011 if (unlikely(ret
)) {
6012 btrfs_qgroup_free_meta(root
,
6013 nr_extents
* fs_info
->nodesize
);
6017 spin_lock(&inode
->lock
);
6018 if (test_and_set_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
6019 &inode
->runtime_flags
)) {
6020 to_reserve
-= btrfs_calc_trans_metadata_size(fs_info
, 1);
6021 release_extra
= true;
6023 inode
->reserved_extents
+= nr_extents
;
6024 spin_unlock(&inode
->lock
);
6027 mutex_unlock(&inode
->delalloc_mutex
);
6030 trace_btrfs_space_reservation(fs_info
, "delalloc",
6031 btrfs_ino(inode
), to_reserve
, 1);
6033 btrfs_block_rsv_release(fs_info
, block_rsv
,
6034 btrfs_calc_trans_metadata_size(fs_info
, 1));
6038 spin_lock(&inode
->lock
);
6039 dropped
= drop_outstanding_extent(inode
, num_bytes
);
6041 * If the inodes csum_bytes is the same as the original
6042 * csum_bytes then we know we haven't raced with any free()ers
6043 * so we can just reduce our inodes csum bytes and carry on.
6045 if (inode
->csum_bytes
== csum_bytes
) {
6046 calc_csum_metadata_size(inode
, num_bytes
, 0);
6048 u64 orig_csum_bytes
= inode
->csum_bytes
;
6052 * This is tricky, but first we need to figure out how much we
6053 * freed from any free-ers that occurred during this
6054 * reservation, so we reset ->csum_bytes to the csum_bytes
6055 * before we dropped our lock, and then call the free for the
6056 * number of bytes that were freed while we were trying our
6059 bytes
= csum_bytes
- inode
->csum_bytes
;
6060 inode
->csum_bytes
= csum_bytes
;
6061 to_free
= calc_csum_metadata_size(inode
, bytes
, 0);
6065 * Now we need to see how much we would have freed had we not
6066 * been making this reservation and our ->csum_bytes were not
6067 * artificially inflated.
6069 inode
->csum_bytes
= csum_bytes
- num_bytes
;
6070 bytes
= csum_bytes
- orig_csum_bytes
;
6071 bytes
= calc_csum_metadata_size(inode
, bytes
, 0);
6074 * Now reset ->csum_bytes to what it should be. If bytes is
6075 * more than to_free then we would have freed more space had we
6076 * not had an artificially high ->csum_bytes, so we need to free
6077 * the remainder. If bytes is the same or less then we don't
6078 * need to do anything, the other free-ers did the correct
6081 inode
->csum_bytes
= orig_csum_bytes
- num_bytes
;
6082 if (bytes
> to_free
)
6083 to_free
= bytes
- to_free
;
6087 spin_unlock(&inode
->lock
);
6089 to_free
+= btrfs_calc_trans_metadata_size(fs_info
, dropped
);
6092 btrfs_block_rsv_release(fs_info
, block_rsv
, to_free
);
6093 trace_btrfs_space_reservation(fs_info
, "delalloc",
6094 btrfs_ino(inode
), to_free
, 0);
6097 mutex_unlock(&inode
->delalloc_mutex
);
6102 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
6103 * @inode: the inode to release the reservation for
6104 * @num_bytes: the number of bytes we're releasing
6106 * This will release the metadata reservation for an inode. This can be called
6107 * once we complete IO for a given set of bytes to release their metadata
6110 void btrfs_delalloc_release_metadata(struct btrfs_inode
*inode
, u64 num_bytes
)
6112 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
6116 num_bytes
= ALIGN(num_bytes
, fs_info
->sectorsize
);
6117 spin_lock(&inode
->lock
);
6118 dropped
= drop_outstanding_extent(inode
, num_bytes
);
6121 to_free
= calc_csum_metadata_size(inode
, num_bytes
, 0);
6122 spin_unlock(&inode
->lock
);
6124 to_free
+= btrfs_calc_trans_metadata_size(fs_info
, dropped
);
6126 if (btrfs_is_testing(fs_info
))
6129 trace_btrfs_space_reservation(fs_info
, "delalloc", btrfs_ino(inode
),
6132 btrfs_block_rsv_release(fs_info
, &fs_info
->delalloc_block_rsv
, to_free
);
6136 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6138 * @inode: inode we're writing to
6139 * @start: start range we are writing to
6140 * @len: how long the range we are writing to
6142 * This will do the following things
6144 * o reserve space in data space info for num bytes
6145 * and reserve precious corresponding qgroup space
6146 * (Done in check_data_free_space)
6148 * o reserve space for metadata space, based on the number of outstanding
6149 * extents and how much csums will be needed
6150 * also reserve metadata space in a per root over-reserve method.
6151 * o add to the inodes->delalloc_bytes
6152 * o add it to the fs_info's delalloc inodes list.
6153 * (Above 3 all done in delalloc_reserve_metadata)
6155 * Return 0 for success
6156 * Return <0 for error(-ENOSPC or -EQUOT)
6158 int btrfs_delalloc_reserve_space(struct inode
*inode
, u64 start
, u64 len
)
6162 ret
= btrfs_check_data_free_space(inode
, start
, len
);
6165 ret
= btrfs_delalloc_reserve_metadata(BTRFS_I(inode
), len
);
6167 btrfs_free_reserved_data_space(inode
, start
, len
);
6172 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6173 * @inode: inode we're releasing space for
6174 * @start: start position of the space already reserved
6175 * @len: the len of the space already reserved
6177 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
6178 * called in the case that we don't need the metadata AND data reservations
6179 * anymore. So if there is an error or we insert an inline extent.
6181 * This function will release the metadata space that was not used and will
6182 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
6183 * list if there are no delalloc bytes left.
6184 * Also it will handle the qgroup reserved space.
6186 void btrfs_delalloc_release_space(struct inode
*inode
, u64 start
, u64 len
)
6188 btrfs_delalloc_release_metadata(BTRFS_I(inode
), len
);
6189 btrfs_free_reserved_data_space(inode
, start
, len
);
6192 static int update_block_group(struct btrfs_trans_handle
*trans
,
6193 struct btrfs_fs_info
*info
, u64 bytenr
,
6194 u64 num_bytes
, int alloc
)
6196 struct btrfs_block_group_cache
*cache
= NULL
;
6197 u64 total
= num_bytes
;
6202 /* block accounting for super block */
6203 spin_lock(&info
->delalloc_root_lock
);
6204 old_val
= btrfs_super_bytes_used(info
->super_copy
);
6206 old_val
+= num_bytes
;
6208 old_val
-= num_bytes
;
6209 btrfs_set_super_bytes_used(info
->super_copy
, old_val
);
6210 spin_unlock(&info
->delalloc_root_lock
);
6213 cache
= btrfs_lookup_block_group(info
, bytenr
);
6216 if (cache
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
6217 BTRFS_BLOCK_GROUP_RAID1
|
6218 BTRFS_BLOCK_GROUP_RAID10
))
6223 * If this block group has free space cache written out, we
6224 * need to make sure to load it if we are removing space. This
6225 * is because we need the unpinning stage to actually add the
6226 * space back to the block group, otherwise we will leak space.
6228 if (!alloc
&& cache
->cached
== BTRFS_CACHE_NO
)
6229 cache_block_group(cache
, 1);
6231 byte_in_group
= bytenr
- cache
->key
.objectid
;
6232 WARN_ON(byte_in_group
> cache
->key
.offset
);
6234 spin_lock(&cache
->space_info
->lock
);
6235 spin_lock(&cache
->lock
);
6237 if (btrfs_test_opt(info
, SPACE_CACHE
) &&
6238 cache
->disk_cache_state
< BTRFS_DC_CLEAR
)
6239 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
6241 old_val
= btrfs_block_group_used(&cache
->item
);
6242 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
6244 old_val
+= num_bytes
;
6245 btrfs_set_block_group_used(&cache
->item
, old_val
);
6246 cache
->reserved
-= num_bytes
;
6247 cache
->space_info
->bytes_reserved
-= num_bytes
;
6248 cache
->space_info
->bytes_used
+= num_bytes
;
6249 cache
->space_info
->disk_used
+= num_bytes
* factor
;
6250 spin_unlock(&cache
->lock
);
6251 spin_unlock(&cache
->space_info
->lock
);
6253 old_val
-= num_bytes
;
6254 btrfs_set_block_group_used(&cache
->item
, old_val
);
6255 cache
->pinned
+= num_bytes
;
6256 cache
->space_info
->bytes_pinned
+= num_bytes
;
6257 cache
->space_info
->bytes_used
-= num_bytes
;
6258 cache
->space_info
->disk_used
-= num_bytes
* factor
;
6259 spin_unlock(&cache
->lock
);
6260 spin_unlock(&cache
->space_info
->lock
);
6262 trace_btrfs_space_reservation(info
, "pinned",
6263 cache
->space_info
->flags
,
6265 set_extent_dirty(info
->pinned_extents
,
6266 bytenr
, bytenr
+ num_bytes
- 1,
6267 GFP_NOFS
| __GFP_NOFAIL
);
6270 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
6271 if (list_empty(&cache
->dirty_list
)) {
6272 list_add_tail(&cache
->dirty_list
,
6273 &trans
->transaction
->dirty_bgs
);
6274 trans
->transaction
->num_dirty_bgs
++;
6275 btrfs_get_block_group(cache
);
6277 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
6280 * No longer have used bytes in this block group, queue it for
6281 * deletion. We do this after adding the block group to the
6282 * dirty list to avoid races between cleaner kthread and space
6285 if (!alloc
&& old_val
== 0) {
6286 spin_lock(&info
->unused_bgs_lock
);
6287 if (list_empty(&cache
->bg_list
)) {
6288 btrfs_get_block_group(cache
);
6289 list_add_tail(&cache
->bg_list
,
6292 spin_unlock(&info
->unused_bgs_lock
);
6295 btrfs_put_block_group(cache
);
6297 bytenr
+= num_bytes
;
6302 static u64
first_logical_byte(struct btrfs_fs_info
*fs_info
, u64 search_start
)
6304 struct btrfs_block_group_cache
*cache
;
6307 spin_lock(&fs_info
->block_group_cache_lock
);
6308 bytenr
= fs_info
->first_logical_byte
;
6309 spin_unlock(&fs_info
->block_group_cache_lock
);
6311 if (bytenr
< (u64
)-1)
6314 cache
= btrfs_lookup_first_block_group(fs_info
, search_start
);
6318 bytenr
= cache
->key
.objectid
;
6319 btrfs_put_block_group(cache
);
6324 static int pin_down_extent(struct btrfs_fs_info
*fs_info
,
6325 struct btrfs_block_group_cache
*cache
,
6326 u64 bytenr
, u64 num_bytes
, int reserved
)
6328 spin_lock(&cache
->space_info
->lock
);
6329 spin_lock(&cache
->lock
);
6330 cache
->pinned
+= num_bytes
;
6331 cache
->space_info
->bytes_pinned
+= num_bytes
;
6333 cache
->reserved
-= num_bytes
;
6334 cache
->space_info
->bytes_reserved
-= num_bytes
;
6336 spin_unlock(&cache
->lock
);
6337 spin_unlock(&cache
->space_info
->lock
);
6339 trace_btrfs_space_reservation(fs_info
, "pinned",
6340 cache
->space_info
->flags
, num_bytes
, 1);
6341 set_extent_dirty(fs_info
->pinned_extents
, bytenr
,
6342 bytenr
+ num_bytes
- 1, GFP_NOFS
| __GFP_NOFAIL
);
6347 * this function must be called within transaction
6349 int btrfs_pin_extent(struct btrfs_fs_info
*fs_info
,
6350 u64 bytenr
, u64 num_bytes
, int reserved
)
6352 struct btrfs_block_group_cache
*cache
;
6354 cache
= btrfs_lookup_block_group(fs_info
, bytenr
);
6355 BUG_ON(!cache
); /* Logic error */
6357 pin_down_extent(fs_info
, cache
, bytenr
, num_bytes
, reserved
);
6359 btrfs_put_block_group(cache
);
6364 * this function must be called within transaction
6366 int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info
*fs_info
,
6367 u64 bytenr
, u64 num_bytes
)
6369 struct btrfs_block_group_cache
*cache
;
6372 cache
= btrfs_lookup_block_group(fs_info
, bytenr
);
6377 * pull in the free space cache (if any) so that our pin
6378 * removes the free space from the cache. We have load_only set
6379 * to one because the slow code to read in the free extents does check
6380 * the pinned extents.
6382 cache_block_group(cache
, 1);
6384 pin_down_extent(fs_info
, cache
, bytenr
, num_bytes
, 0);
6386 /* remove us from the free space cache (if we're there at all) */
6387 ret
= btrfs_remove_free_space(cache
, bytenr
, num_bytes
);
6388 btrfs_put_block_group(cache
);
6392 static int __exclude_logged_extent(struct btrfs_fs_info
*fs_info
,
6393 u64 start
, u64 num_bytes
)
6396 struct btrfs_block_group_cache
*block_group
;
6397 struct btrfs_caching_control
*caching_ctl
;
6399 block_group
= btrfs_lookup_block_group(fs_info
, start
);
6403 cache_block_group(block_group
, 0);
6404 caching_ctl
= get_caching_control(block_group
);
6408 BUG_ON(!block_group_cache_done(block_group
));
6409 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
6411 mutex_lock(&caching_ctl
->mutex
);
6413 if (start
>= caching_ctl
->progress
) {
6414 ret
= add_excluded_extent(fs_info
, start
, num_bytes
);
6415 } else if (start
+ num_bytes
<= caching_ctl
->progress
) {
6416 ret
= btrfs_remove_free_space(block_group
,
6419 num_bytes
= caching_ctl
->progress
- start
;
6420 ret
= btrfs_remove_free_space(block_group
,
6425 num_bytes
= (start
+ num_bytes
) -
6426 caching_ctl
->progress
;
6427 start
= caching_ctl
->progress
;
6428 ret
= add_excluded_extent(fs_info
, start
, num_bytes
);
6431 mutex_unlock(&caching_ctl
->mutex
);
6432 put_caching_control(caching_ctl
);
6434 btrfs_put_block_group(block_group
);
6438 int btrfs_exclude_logged_extents(struct btrfs_fs_info
*fs_info
,
6439 struct extent_buffer
*eb
)
6441 struct btrfs_file_extent_item
*item
;
6442 struct btrfs_key key
;
6446 if (!btrfs_fs_incompat(fs_info
, MIXED_GROUPS
))
6449 for (i
= 0; i
< btrfs_header_nritems(eb
); i
++) {
6450 btrfs_item_key_to_cpu(eb
, &key
, i
);
6451 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
6453 item
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
6454 found_type
= btrfs_file_extent_type(eb
, item
);
6455 if (found_type
== BTRFS_FILE_EXTENT_INLINE
)
6457 if (btrfs_file_extent_disk_bytenr(eb
, item
) == 0)
6459 key
.objectid
= btrfs_file_extent_disk_bytenr(eb
, item
);
6460 key
.offset
= btrfs_file_extent_disk_num_bytes(eb
, item
);
6461 __exclude_logged_extent(fs_info
, key
.objectid
, key
.offset
);
6468 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache
*bg
)
6470 atomic_inc(&bg
->reservations
);
6473 void btrfs_dec_block_group_reservations(struct btrfs_fs_info
*fs_info
,
6476 struct btrfs_block_group_cache
*bg
;
6478 bg
= btrfs_lookup_block_group(fs_info
, start
);
6480 if (atomic_dec_and_test(&bg
->reservations
))
6481 wake_up_atomic_t(&bg
->reservations
);
6482 btrfs_put_block_group(bg
);
6485 static int btrfs_wait_bg_reservations_atomic_t(atomic_t
*a
)
6491 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache
*bg
)
6493 struct btrfs_space_info
*space_info
= bg
->space_info
;
6497 if (!(bg
->flags
& BTRFS_BLOCK_GROUP_DATA
))
6501 * Our block group is read only but before we set it to read only,
6502 * some task might have had allocated an extent from it already, but it
6503 * has not yet created a respective ordered extent (and added it to a
6504 * root's list of ordered extents).
6505 * Therefore wait for any task currently allocating extents, since the
6506 * block group's reservations counter is incremented while a read lock
6507 * on the groups' semaphore is held and decremented after releasing
6508 * the read access on that semaphore and creating the ordered extent.
6510 down_write(&space_info
->groups_sem
);
6511 up_write(&space_info
->groups_sem
);
6513 wait_on_atomic_t(&bg
->reservations
,
6514 btrfs_wait_bg_reservations_atomic_t
,
6515 TASK_UNINTERRUPTIBLE
);
6519 * btrfs_add_reserved_bytes - update the block_group and space info counters
6520 * @cache: The cache we are manipulating
6521 * @ram_bytes: The number of bytes of file content, and will be same to
6522 * @num_bytes except for the compress path.
6523 * @num_bytes: The number of bytes in question
6524 * @delalloc: The blocks are allocated for the delalloc write
6526 * This is called by the allocator when it reserves space. If this is a
6527 * reservation and the block group has become read only we cannot make the
6528 * reservation and return -EAGAIN, otherwise this function always succeeds.
6530 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache
*cache
,
6531 u64 ram_bytes
, u64 num_bytes
, int delalloc
)
6533 struct btrfs_space_info
*space_info
= cache
->space_info
;
6536 spin_lock(&space_info
->lock
);
6537 spin_lock(&cache
->lock
);
6541 cache
->reserved
+= num_bytes
;
6542 space_info
->bytes_reserved
+= num_bytes
;
6544 trace_btrfs_space_reservation(cache
->fs_info
,
6545 "space_info", space_info
->flags
,
6547 space_info
->bytes_may_use
-= ram_bytes
;
6549 cache
->delalloc_bytes
+= num_bytes
;
6551 spin_unlock(&cache
->lock
);
6552 spin_unlock(&space_info
->lock
);
6557 * btrfs_free_reserved_bytes - update the block_group and space info counters
6558 * @cache: The cache we are manipulating
6559 * @num_bytes: The number of bytes in question
6560 * @delalloc: The blocks are allocated for the delalloc write
6562 * This is called by somebody who is freeing space that was never actually used
6563 * on disk. For example if you reserve some space for a new leaf in transaction
6564 * A and before transaction A commits you free that leaf, you call this with
6565 * reserve set to 0 in order to clear the reservation.
6568 static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache
*cache
,
6569 u64 num_bytes
, int delalloc
)
6571 struct btrfs_space_info
*space_info
= cache
->space_info
;
6574 spin_lock(&space_info
->lock
);
6575 spin_lock(&cache
->lock
);
6577 space_info
->bytes_readonly
+= num_bytes
;
6578 cache
->reserved
-= num_bytes
;
6579 space_info
->bytes_reserved
-= num_bytes
;
6582 cache
->delalloc_bytes
-= num_bytes
;
6583 spin_unlock(&cache
->lock
);
6584 spin_unlock(&space_info
->lock
);
6587 void btrfs_prepare_extent_commit(struct btrfs_fs_info
*fs_info
)
6589 struct btrfs_caching_control
*next
;
6590 struct btrfs_caching_control
*caching_ctl
;
6591 struct btrfs_block_group_cache
*cache
;
6593 down_write(&fs_info
->commit_root_sem
);
6595 list_for_each_entry_safe(caching_ctl
, next
,
6596 &fs_info
->caching_block_groups
, list
) {
6597 cache
= caching_ctl
->block_group
;
6598 if (block_group_cache_done(cache
)) {
6599 cache
->last_byte_to_unpin
= (u64
)-1;
6600 list_del_init(&caching_ctl
->list
);
6601 put_caching_control(caching_ctl
);
6603 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
6607 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
6608 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
6610 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
6612 up_write(&fs_info
->commit_root_sem
);
6614 update_global_block_rsv(fs_info
);
6618 * Returns the free cluster for the given space info and sets empty_cluster to
6619 * what it should be based on the mount options.
6621 static struct btrfs_free_cluster
*
6622 fetch_cluster_info(struct btrfs_fs_info
*fs_info
,
6623 struct btrfs_space_info
*space_info
, u64
*empty_cluster
)
6625 struct btrfs_free_cluster
*ret
= NULL
;
6626 bool ssd
= btrfs_test_opt(fs_info
, SSD
);
6629 if (btrfs_mixed_space_info(space_info
))
6633 *empty_cluster
= SZ_2M
;
6634 if (space_info
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
6635 ret
= &fs_info
->meta_alloc_cluster
;
6637 *empty_cluster
= SZ_64K
;
6638 } else if ((space_info
->flags
& BTRFS_BLOCK_GROUP_DATA
) && ssd
) {
6639 ret
= &fs_info
->data_alloc_cluster
;
6645 static int unpin_extent_range(struct btrfs_fs_info
*fs_info
,
6647 const bool return_free_space
)
6649 struct btrfs_block_group_cache
*cache
= NULL
;
6650 struct btrfs_space_info
*space_info
;
6651 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
6652 struct btrfs_free_cluster
*cluster
= NULL
;
6654 u64 total_unpinned
= 0;
6655 u64 empty_cluster
= 0;
6658 while (start
<= end
) {
6661 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
6663 btrfs_put_block_group(cache
);
6665 cache
= btrfs_lookup_block_group(fs_info
, start
);
6666 BUG_ON(!cache
); /* Logic error */
6668 cluster
= fetch_cluster_info(fs_info
,
6671 empty_cluster
<<= 1;
6674 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
6675 len
= min(len
, end
+ 1 - start
);
6677 if (start
< cache
->last_byte_to_unpin
) {
6678 len
= min(len
, cache
->last_byte_to_unpin
- start
);
6679 if (return_free_space
)
6680 btrfs_add_free_space(cache
, start
, len
);
6684 total_unpinned
+= len
;
6685 space_info
= cache
->space_info
;
6688 * If this space cluster has been marked as fragmented and we've
6689 * unpinned enough in this block group to potentially allow a
6690 * cluster to be created inside of it go ahead and clear the
6693 if (cluster
&& cluster
->fragmented
&&
6694 total_unpinned
> empty_cluster
) {
6695 spin_lock(&cluster
->lock
);
6696 cluster
->fragmented
= 0;
6697 spin_unlock(&cluster
->lock
);
6700 spin_lock(&space_info
->lock
);
6701 spin_lock(&cache
->lock
);
6702 cache
->pinned
-= len
;
6703 space_info
->bytes_pinned
-= len
;
6705 trace_btrfs_space_reservation(fs_info
, "pinned",
6706 space_info
->flags
, len
, 0);
6707 space_info
->max_extent_size
= 0;
6708 percpu_counter_add(&space_info
->total_bytes_pinned
, -len
);
6710 space_info
->bytes_readonly
+= len
;
6713 spin_unlock(&cache
->lock
);
6714 if (!readonly
&& return_free_space
&&
6715 global_rsv
->space_info
== space_info
) {
6717 WARN_ON(!return_free_space
);
6718 spin_lock(&global_rsv
->lock
);
6719 if (!global_rsv
->full
) {
6720 to_add
= min(len
, global_rsv
->size
-
6721 global_rsv
->reserved
);
6722 global_rsv
->reserved
+= to_add
;
6723 space_info
->bytes_may_use
+= to_add
;
6724 if (global_rsv
->reserved
>= global_rsv
->size
)
6725 global_rsv
->full
= 1;
6726 trace_btrfs_space_reservation(fs_info
,
6732 spin_unlock(&global_rsv
->lock
);
6733 /* Add to any tickets we may have */
6735 space_info_add_new_bytes(fs_info
, space_info
,
6738 spin_unlock(&space_info
->lock
);
6742 btrfs_put_block_group(cache
);
6746 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
6747 struct btrfs_fs_info
*fs_info
)
6749 struct btrfs_block_group_cache
*block_group
, *tmp
;
6750 struct list_head
*deleted_bgs
;
6751 struct extent_io_tree
*unpin
;
6756 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
6757 unpin
= &fs_info
->freed_extents
[1];
6759 unpin
= &fs_info
->freed_extents
[0];
6761 while (!trans
->aborted
) {
6762 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
6763 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
6764 EXTENT_DIRTY
, NULL
);
6766 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
6770 if (btrfs_test_opt(fs_info
, DISCARD
))
6771 ret
= btrfs_discard_extent(fs_info
, start
,
6772 end
+ 1 - start
, NULL
);
6774 clear_extent_dirty(unpin
, start
, end
);
6775 unpin_extent_range(fs_info
, start
, end
, true);
6776 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
6781 * Transaction is finished. We don't need the lock anymore. We
6782 * do need to clean up the block groups in case of a transaction
6785 deleted_bgs
= &trans
->transaction
->deleted_bgs
;
6786 list_for_each_entry_safe(block_group
, tmp
, deleted_bgs
, bg_list
) {
6790 if (!trans
->aborted
)
6791 ret
= btrfs_discard_extent(fs_info
,
6792 block_group
->key
.objectid
,
6793 block_group
->key
.offset
,
6796 list_del_init(&block_group
->bg_list
);
6797 btrfs_put_block_group_trimming(block_group
);
6798 btrfs_put_block_group(block_group
);
6801 const char *errstr
= btrfs_decode_error(ret
);
6803 "Discard failed while removing blockgroup: errno=%d %s\n",
6811 static void add_pinned_bytes(struct btrfs_fs_info
*fs_info
, u64 num_bytes
,
6812 u64 owner
, u64 root_objectid
)
6814 struct btrfs_space_info
*space_info
;
6817 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
6818 if (root_objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
6819 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
6821 flags
= BTRFS_BLOCK_GROUP_METADATA
;
6823 flags
= BTRFS_BLOCK_GROUP_DATA
;
6826 space_info
= __find_space_info(fs_info
, flags
);
6827 BUG_ON(!space_info
); /* Logic bug */
6828 percpu_counter_add(&space_info
->total_bytes_pinned
, num_bytes
);
6832 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
6833 struct btrfs_fs_info
*info
,
6834 struct btrfs_delayed_ref_node
*node
, u64 parent
,
6835 u64 root_objectid
, u64 owner_objectid
,
6836 u64 owner_offset
, int refs_to_drop
,
6837 struct btrfs_delayed_extent_op
*extent_op
)
6839 struct btrfs_key key
;
6840 struct btrfs_path
*path
;
6841 struct btrfs_root
*extent_root
= info
->extent_root
;
6842 struct extent_buffer
*leaf
;
6843 struct btrfs_extent_item
*ei
;
6844 struct btrfs_extent_inline_ref
*iref
;
6847 int extent_slot
= 0;
6848 int found_extent
= 0;
6852 u64 bytenr
= node
->bytenr
;
6853 u64 num_bytes
= node
->num_bytes
;
6855 bool skinny_metadata
= btrfs_fs_incompat(info
, SKINNY_METADATA
);
6857 path
= btrfs_alloc_path();
6861 path
->reada
= READA_FORWARD
;
6862 path
->leave_spinning
= 1;
6864 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
6865 BUG_ON(!is_data
&& refs_to_drop
!= 1);
6868 skinny_metadata
= 0;
6870 ret
= lookup_extent_backref(trans
, info
, path
, &iref
,
6871 bytenr
, num_bytes
, parent
,
6872 root_objectid
, owner_objectid
,
6875 extent_slot
= path
->slots
[0];
6876 while (extent_slot
>= 0) {
6877 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
6879 if (key
.objectid
!= bytenr
)
6881 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
6882 key
.offset
== num_bytes
) {
6886 if (key
.type
== BTRFS_METADATA_ITEM_KEY
&&
6887 key
.offset
== owner_objectid
) {
6891 if (path
->slots
[0] - extent_slot
> 5)
6895 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6896 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
6897 if (found_extent
&& item_size
< sizeof(*ei
))
6900 if (!found_extent
) {
6902 ret
= remove_extent_backref(trans
, info
, path
, NULL
,
6904 is_data
, &last_ref
);
6906 btrfs_abort_transaction(trans
, ret
);
6909 btrfs_release_path(path
);
6910 path
->leave_spinning
= 1;
6912 key
.objectid
= bytenr
;
6913 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6914 key
.offset
= num_bytes
;
6916 if (!is_data
&& skinny_metadata
) {
6917 key
.type
= BTRFS_METADATA_ITEM_KEY
;
6918 key
.offset
= owner_objectid
;
6921 ret
= btrfs_search_slot(trans
, extent_root
,
6923 if (ret
> 0 && skinny_metadata
&& path
->slots
[0]) {
6925 * Couldn't find our skinny metadata item,
6926 * see if we have ye olde extent item.
6929 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
6931 if (key
.objectid
== bytenr
&&
6932 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
6933 key
.offset
== num_bytes
)
6937 if (ret
> 0 && skinny_metadata
) {
6938 skinny_metadata
= false;
6939 key
.objectid
= bytenr
;
6940 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6941 key
.offset
= num_bytes
;
6942 btrfs_release_path(path
);
6943 ret
= btrfs_search_slot(trans
, extent_root
,
6949 "umm, got %d back from search, was looking for %llu",
6952 btrfs_print_leaf(info
, path
->nodes
[0]);
6955 btrfs_abort_transaction(trans
, ret
);
6958 extent_slot
= path
->slots
[0];
6960 } else if (WARN_ON(ret
== -ENOENT
)) {
6961 btrfs_print_leaf(info
, path
->nodes
[0]);
6963 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6964 bytenr
, parent
, root_objectid
, owner_objectid
,
6966 btrfs_abort_transaction(trans
, ret
);
6969 btrfs_abort_transaction(trans
, ret
);
6973 leaf
= path
->nodes
[0];
6974 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
6975 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6976 if (item_size
< sizeof(*ei
)) {
6977 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
6978 ret
= convert_extent_item_v0(trans
, info
, path
, owner_objectid
,
6981 btrfs_abort_transaction(trans
, ret
);
6985 btrfs_release_path(path
);
6986 path
->leave_spinning
= 1;
6988 key
.objectid
= bytenr
;
6989 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6990 key
.offset
= num_bytes
;
6992 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
6996 "umm, got %d back from search, was looking for %llu",
6998 btrfs_print_leaf(info
, path
->nodes
[0]);
7001 btrfs_abort_transaction(trans
, ret
);
7005 extent_slot
= path
->slots
[0];
7006 leaf
= path
->nodes
[0];
7007 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
7010 BUG_ON(item_size
< sizeof(*ei
));
7011 ei
= btrfs_item_ptr(leaf
, extent_slot
,
7012 struct btrfs_extent_item
);
7013 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
7014 key
.type
== BTRFS_EXTENT_ITEM_KEY
) {
7015 struct btrfs_tree_block_info
*bi
;
7016 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
7017 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
7018 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
7021 refs
= btrfs_extent_refs(leaf
, ei
);
7022 if (refs
< refs_to_drop
) {
7024 "trying to drop %d refs but we only have %Lu for bytenr %Lu",
7025 refs_to_drop
, refs
, bytenr
);
7027 btrfs_abort_transaction(trans
, ret
);
7030 refs
-= refs_to_drop
;
7034 __run_delayed_extent_op(extent_op
, leaf
, ei
);
7036 * In the case of inline back ref, reference count will
7037 * be updated by remove_extent_backref
7040 BUG_ON(!found_extent
);
7042 btrfs_set_extent_refs(leaf
, ei
, refs
);
7043 btrfs_mark_buffer_dirty(leaf
);
7046 ret
= remove_extent_backref(trans
, info
, path
,
7048 is_data
, &last_ref
);
7050 btrfs_abort_transaction(trans
, ret
);
7054 add_pinned_bytes(info
, -num_bytes
, owner_objectid
,
7058 BUG_ON(is_data
&& refs_to_drop
!=
7059 extent_data_ref_count(path
, iref
));
7061 BUG_ON(path
->slots
[0] != extent_slot
);
7063 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
7064 path
->slots
[0] = extent_slot
;
7070 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
7073 btrfs_abort_transaction(trans
, ret
);
7076 btrfs_release_path(path
);
7079 ret
= btrfs_del_csums(trans
, info
, bytenr
, num_bytes
);
7081 btrfs_abort_transaction(trans
, ret
);
7086 ret
= add_to_free_space_tree(trans
, info
, bytenr
, num_bytes
);
7088 btrfs_abort_transaction(trans
, ret
);
7092 ret
= update_block_group(trans
, info
, bytenr
, num_bytes
, 0);
7094 btrfs_abort_transaction(trans
, ret
);
7098 btrfs_release_path(path
);
7101 btrfs_free_path(path
);
7106 * when we free an block, it is possible (and likely) that we free the last
7107 * delayed ref for that extent as well. This searches the delayed ref tree for
7108 * a given extent, and if there are no other delayed refs to be processed, it
7109 * removes it from the tree.
7111 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
7114 struct btrfs_delayed_ref_head
*head
;
7115 struct btrfs_delayed_ref_root
*delayed_refs
;
7118 delayed_refs
= &trans
->transaction
->delayed_refs
;
7119 spin_lock(&delayed_refs
->lock
);
7120 head
= btrfs_find_delayed_ref_head(delayed_refs
, bytenr
);
7122 goto out_delayed_unlock
;
7124 spin_lock(&head
->lock
);
7125 if (!list_empty(&head
->ref_list
))
7128 if (head
->extent_op
) {
7129 if (!head
->must_insert_reserved
)
7131 btrfs_free_delayed_extent_op(head
->extent_op
);
7132 head
->extent_op
= NULL
;
7136 * waiting for the lock here would deadlock. If someone else has it
7137 * locked they are already in the process of dropping it anyway
7139 if (!mutex_trylock(&head
->mutex
))
7143 * at this point we have a head with no other entries. Go
7144 * ahead and process it.
7146 head
->node
.in_tree
= 0;
7147 rb_erase(&head
->href_node
, &delayed_refs
->href_root
);
7149 atomic_dec(&delayed_refs
->num_entries
);
7152 * we don't take a ref on the node because we're removing it from the
7153 * tree, so we just steal the ref the tree was holding.
7155 delayed_refs
->num_heads
--;
7156 if (head
->processing
== 0)
7157 delayed_refs
->num_heads_ready
--;
7158 head
->processing
= 0;
7159 spin_unlock(&head
->lock
);
7160 spin_unlock(&delayed_refs
->lock
);
7162 BUG_ON(head
->extent_op
);
7163 if (head
->must_insert_reserved
)
7166 mutex_unlock(&head
->mutex
);
7167 btrfs_put_delayed_ref(&head
->node
);
7170 spin_unlock(&head
->lock
);
7173 spin_unlock(&delayed_refs
->lock
);
7177 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
7178 struct btrfs_root
*root
,
7179 struct extent_buffer
*buf
,
7180 u64 parent
, int last_ref
)
7182 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7186 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
7187 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
,
7188 buf
->start
, buf
->len
,
7190 root
->root_key
.objectid
,
7191 btrfs_header_level(buf
),
7192 BTRFS_DROP_DELAYED_REF
, NULL
);
7193 BUG_ON(ret
); /* -ENOMEM */
7199 if (btrfs_header_generation(buf
) == trans
->transid
) {
7200 struct btrfs_block_group_cache
*cache
;
7202 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
7203 ret
= check_ref_cleanup(trans
, buf
->start
);
7208 cache
= btrfs_lookup_block_group(fs_info
, buf
->start
);
7210 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
7211 pin_down_extent(fs_info
, cache
, buf
->start
,
7213 btrfs_put_block_group(cache
);
7217 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
7219 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
7220 btrfs_free_reserved_bytes(cache
, buf
->len
, 0);
7221 btrfs_put_block_group(cache
);
7222 trace_btrfs_reserved_extent_free(fs_info
, buf
->start
, buf
->len
);
7227 add_pinned_bytes(fs_info
, buf
->len
, btrfs_header_level(buf
),
7228 root
->root_key
.objectid
);
7231 * Deleting the buffer, clear the corrupt flag since it doesn't matter
7234 clear_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
);
7237 /* Can return -ENOMEM */
7238 int btrfs_free_extent(struct btrfs_trans_handle
*trans
,
7239 struct btrfs_fs_info
*fs_info
,
7240 u64 bytenr
, u64 num_bytes
, u64 parent
, u64 root_objectid
,
7241 u64 owner
, u64 offset
)
7245 if (btrfs_is_testing(fs_info
))
7248 add_pinned_bytes(fs_info
, num_bytes
, owner
, root_objectid
);
7251 * tree log blocks never actually go into the extent allocation
7252 * tree, just update pinning info and exit early.
7254 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
7255 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
7256 /* unlocks the pinned mutex */
7257 btrfs_pin_extent(fs_info
, bytenr
, num_bytes
, 1);
7259 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
7260 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
7262 parent
, root_objectid
, (int)owner
,
7263 BTRFS_DROP_DELAYED_REF
, NULL
);
7265 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
7267 parent
, root_objectid
, owner
,
7269 BTRFS_DROP_DELAYED_REF
);
7275 * when we wait for progress in the block group caching, its because
7276 * our allocation attempt failed at least once. So, we must sleep
7277 * and let some progress happen before we try again.
7279 * This function will sleep at least once waiting for new free space to
7280 * show up, and then it will check the block group free space numbers
7281 * for our min num_bytes. Another option is to have it go ahead
7282 * and look in the rbtree for a free extent of a given size, but this
7285 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7286 * any of the information in this block group.
7288 static noinline
void
7289 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
7292 struct btrfs_caching_control
*caching_ctl
;
7294 caching_ctl
= get_caching_control(cache
);
7298 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
7299 (cache
->free_space_ctl
->free_space
>= num_bytes
));
7301 put_caching_control(caching_ctl
);
7305 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
7307 struct btrfs_caching_control
*caching_ctl
;
7310 caching_ctl
= get_caching_control(cache
);
7312 return (cache
->cached
== BTRFS_CACHE_ERROR
) ? -EIO
: 0;
7314 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
7315 if (cache
->cached
== BTRFS_CACHE_ERROR
)
7317 put_caching_control(caching_ctl
);
7321 int __get_raid_index(u64 flags
)
7323 if (flags
& BTRFS_BLOCK_GROUP_RAID10
)
7324 return BTRFS_RAID_RAID10
;
7325 else if (flags
& BTRFS_BLOCK_GROUP_RAID1
)
7326 return BTRFS_RAID_RAID1
;
7327 else if (flags
& BTRFS_BLOCK_GROUP_DUP
)
7328 return BTRFS_RAID_DUP
;
7329 else if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
7330 return BTRFS_RAID_RAID0
;
7331 else if (flags
& BTRFS_BLOCK_GROUP_RAID5
)
7332 return BTRFS_RAID_RAID5
;
7333 else if (flags
& BTRFS_BLOCK_GROUP_RAID6
)
7334 return BTRFS_RAID_RAID6
;
7336 return BTRFS_RAID_SINGLE
; /* BTRFS_BLOCK_GROUP_SINGLE */
7339 int get_block_group_index(struct btrfs_block_group_cache
*cache
)
7341 return __get_raid_index(cache
->flags
);
7344 static const char *btrfs_raid_type_names
[BTRFS_NR_RAID_TYPES
] = {
7345 [BTRFS_RAID_RAID10
] = "raid10",
7346 [BTRFS_RAID_RAID1
] = "raid1",
7347 [BTRFS_RAID_DUP
] = "dup",
7348 [BTRFS_RAID_RAID0
] = "raid0",
7349 [BTRFS_RAID_SINGLE
] = "single",
7350 [BTRFS_RAID_RAID5
] = "raid5",
7351 [BTRFS_RAID_RAID6
] = "raid6",
7354 static const char *get_raid_name(enum btrfs_raid_types type
)
7356 if (type
>= BTRFS_NR_RAID_TYPES
)
7359 return btrfs_raid_type_names
[type
];
7362 enum btrfs_loop_type
{
7363 LOOP_CACHING_NOWAIT
= 0,
7364 LOOP_CACHING_WAIT
= 1,
7365 LOOP_ALLOC_CHUNK
= 2,
7366 LOOP_NO_EMPTY_SIZE
= 3,
7370 btrfs_lock_block_group(struct btrfs_block_group_cache
*cache
,
7374 down_read(&cache
->data_rwsem
);
7378 btrfs_grab_block_group(struct btrfs_block_group_cache
*cache
,
7381 btrfs_get_block_group(cache
);
7383 down_read(&cache
->data_rwsem
);
7386 static struct btrfs_block_group_cache
*
7387 btrfs_lock_cluster(struct btrfs_block_group_cache
*block_group
,
7388 struct btrfs_free_cluster
*cluster
,
7391 struct btrfs_block_group_cache
*used_bg
= NULL
;
7393 spin_lock(&cluster
->refill_lock
);
7395 used_bg
= cluster
->block_group
;
7399 if (used_bg
== block_group
)
7402 btrfs_get_block_group(used_bg
);
7407 if (down_read_trylock(&used_bg
->data_rwsem
))
7410 spin_unlock(&cluster
->refill_lock
);
7412 /* We should only have one-level nested. */
7413 down_read_nested(&used_bg
->data_rwsem
, SINGLE_DEPTH_NESTING
);
7415 spin_lock(&cluster
->refill_lock
);
7416 if (used_bg
== cluster
->block_group
)
7419 up_read(&used_bg
->data_rwsem
);
7420 btrfs_put_block_group(used_bg
);
7425 btrfs_release_block_group(struct btrfs_block_group_cache
*cache
,
7429 up_read(&cache
->data_rwsem
);
7430 btrfs_put_block_group(cache
);
7434 * walks the btree of allocated extents and find a hole of a given size.
7435 * The key ins is changed to record the hole:
7436 * ins->objectid == start position
7437 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7438 * ins->offset == the size of the hole.
7439 * Any available blocks before search_start are skipped.
7441 * If there is no suitable free space, we will record the max size of
7442 * the free space extent currently.
7444 static noinline
int find_free_extent(struct btrfs_fs_info
*fs_info
,
7445 u64 ram_bytes
, u64 num_bytes
, u64 empty_size
,
7446 u64 hint_byte
, struct btrfs_key
*ins
,
7447 u64 flags
, int delalloc
)
7450 struct btrfs_root
*root
= fs_info
->extent_root
;
7451 struct btrfs_free_cluster
*last_ptr
= NULL
;
7452 struct btrfs_block_group_cache
*block_group
= NULL
;
7453 u64 search_start
= 0;
7454 u64 max_extent_size
= 0;
7455 u64 empty_cluster
= 0;
7456 struct btrfs_space_info
*space_info
;
7458 int index
= __get_raid_index(flags
);
7459 bool failed_cluster_refill
= false;
7460 bool failed_alloc
= false;
7461 bool use_cluster
= true;
7462 bool have_caching_bg
= false;
7463 bool orig_have_caching_bg
= false;
7464 bool full_search
= false;
7466 WARN_ON(num_bytes
< fs_info
->sectorsize
);
7467 ins
->type
= BTRFS_EXTENT_ITEM_KEY
;
7471 trace_find_free_extent(fs_info
, num_bytes
, empty_size
, flags
);
7473 space_info
= __find_space_info(fs_info
, flags
);
7475 btrfs_err(fs_info
, "No space info for %llu", flags
);
7480 * If our free space is heavily fragmented we may not be able to make
7481 * big contiguous allocations, so instead of doing the expensive search
7482 * for free space, simply return ENOSPC with our max_extent_size so we
7483 * can go ahead and search for a more manageable chunk.
7485 * If our max_extent_size is large enough for our allocation simply
7486 * disable clustering since we will likely not be able to find enough
7487 * space to create a cluster and induce latency trying.
7489 if (unlikely(space_info
->max_extent_size
)) {
7490 spin_lock(&space_info
->lock
);
7491 if (space_info
->max_extent_size
&&
7492 num_bytes
> space_info
->max_extent_size
) {
7493 ins
->offset
= space_info
->max_extent_size
;
7494 spin_unlock(&space_info
->lock
);
7496 } else if (space_info
->max_extent_size
) {
7497 use_cluster
= false;
7499 spin_unlock(&space_info
->lock
);
7502 last_ptr
= fetch_cluster_info(fs_info
, space_info
, &empty_cluster
);
7504 spin_lock(&last_ptr
->lock
);
7505 if (last_ptr
->block_group
)
7506 hint_byte
= last_ptr
->window_start
;
7507 if (last_ptr
->fragmented
) {
7509 * We still set window_start so we can keep track of the
7510 * last place we found an allocation to try and save
7513 hint_byte
= last_ptr
->window_start
;
7514 use_cluster
= false;
7516 spin_unlock(&last_ptr
->lock
);
7519 search_start
= max(search_start
, first_logical_byte(fs_info
, 0));
7520 search_start
= max(search_start
, hint_byte
);
7521 if (search_start
== hint_byte
) {
7522 block_group
= btrfs_lookup_block_group(fs_info
, search_start
);
7524 * we don't want to use the block group if it doesn't match our
7525 * allocation bits, or if its not cached.
7527 * However if we are re-searching with an ideal block group
7528 * picked out then we don't care that the block group is cached.
7530 if (block_group
&& block_group_bits(block_group
, flags
) &&
7531 block_group
->cached
!= BTRFS_CACHE_NO
) {
7532 down_read(&space_info
->groups_sem
);
7533 if (list_empty(&block_group
->list
) ||
7536 * someone is removing this block group,
7537 * we can't jump into the have_block_group
7538 * target because our list pointers are not
7541 btrfs_put_block_group(block_group
);
7542 up_read(&space_info
->groups_sem
);
7544 index
= get_block_group_index(block_group
);
7545 btrfs_lock_block_group(block_group
, delalloc
);
7546 goto have_block_group
;
7548 } else if (block_group
) {
7549 btrfs_put_block_group(block_group
);
7553 have_caching_bg
= false;
7554 if (index
== 0 || index
== __get_raid_index(flags
))
7556 down_read(&space_info
->groups_sem
);
7557 list_for_each_entry(block_group
, &space_info
->block_groups
[index
],
7562 btrfs_grab_block_group(block_group
, delalloc
);
7563 search_start
= block_group
->key
.objectid
;
7566 * this can happen if we end up cycling through all the
7567 * raid types, but we want to make sure we only allocate
7568 * for the proper type.
7570 if (!block_group_bits(block_group
, flags
)) {
7571 u64 extra
= BTRFS_BLOCK_GROUP_DUP
|
7572 BTRFS_BLOCK_GROUP_RAID1
|
7573 BTRFS_BLOCK_GROUP_RAID5
|
7574 BTRFS_BLOCK_GROUP_RAID6
|
7575 BTRFS_BLOCK_GROUP_RAID10
;
7578 * if they asked for extra copies and this block group
7579 * doesn't provide them, bail. This does allow us to
7580 * fill raid0 from raid1.
7582 if ((flags
& extra
) && !(block_group
->flags
& extra
))
7587 cached
= block_group_cache_done(block_group
);
7588 if (unlikely(!cached
)) {
7589 have_caching_bg
= true;
7590 ret
= cache_block_group(block_group
, 0);
7595 if (unlikely(block_group
->cached
== BTRFS_CACHE_ERROR
))
7597 if (unlikely(block_group
->ro
))
7601 * Ok we want to try and use the cluster allocator, so
7604 if (last_ptr
&& use_cluster
) {
7605 struct btrfs_block_group_cache
*used_block_group
;
7606 unsigned long aligned_cluster
;
7608 * the refill lock keeps out other
7609 * people trying to start a new cluster
7611 used_block_group
= btrfs_lock_cluster(block_group
,
7614 if (!used_block_group
)
7615 goto refill_cluster
;
7617 if (used_block_group
!= block_group
&&
7618 (used_block_group
->ro
||
7619 !block_group_bits(used_block_group
, flags
)))
7620 goto release_cluster
;
7622 offset
= btrfs_alloc_from_cluster(used_block_group
,
7625 used_block_group
->key
.objectid
,
7628 /* we have a block, we're done */
7629 spin_unlock(&last_ptr
->refill_lock
);
7630 trace_btrfs_reserve_extent_cluster(fs_info
,
7632 search_start
, num_bytes
);
7633 if (used_block_group
!= block_group
) {
7634 btrfs_release_block_group(block_group
,
7636 block_group
= used_block_group
;
7641 WARN_ON(last_ptr
->block_group
!= used_block_group
);
7643 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7644 * set up a new clusters, so lets just skip it
7645 * and let the allocator find whatever block
7646 * it can find. If we reach this point, we
7647 * will have tried the cluster allocator
7648 * plenty of times and not have found
7649 * anything, so we are likely way too
7650 * fragmented for the clustering stuff to find
7653 * However, if the cluster is taken from the
7654 * current block group, release the cluster
7655 * first, so that we stand a better chance of
7656 * succeeding in the unclustered
7658 if (loop
>= LOOP_NO_EMPTY_SIZE
&&
7659 used_block_group
!= block_group
) {
7660 spin_unlock(&last_ptr
->refill_lock
);
7661 btrfs_release_block_group(used_block_group
,
7663 goto unclustered_alloc
;
7667 * this cluster didn't work out, free it and
7670 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
7672 if (used_block_group
!= block_group
)
7673 btrfs_release_block_group(used_block_group
,
7676 if (loop
>= LOOP_NO_EMPTY_SIZE
) {
7677 spin_unlock(&last_ptr
->refill_lock
);
7678 goto unclustered_alloc
;
7681 aligned_cluster
= max_t(unsigned long,
7682 empty_cluster
+ empty_size
,
7683 block_group
->full_stripe_len
);
7685 /* allocate a cluster in this block group */
7686 ret
= btrfs_find_space_cluster(fs_info
, block_group
,
7687 last_ptr
, search_start
,
7692 * now pull our allocation out of this
7695 offset
= btrfs_alloc_from_cluster(block_group
,
7701 /* we found one, proceed */
7702 spin_unlock(&last_ptr
->refill_lock
);
7703 trace_btrfs_reserve_extent_cluster(fs_info
,
7704 block_group
, search_start
,
7708 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
7709 && !failed_cluster_refill
) {
7710 spin_unlock(&last_ptr
->refill_lock
);
7712 failed_cluster_refill
= true;
7713 wait_block_group_cache_progress(block_group
,
7714 num_bytes
+ empty_cluster
+ empty_size
);
7715 goto have_block_group
;
7719 * at this point we either didn't find a cluster
7720 * or we weren't able to allocate a block from our
7721 * cluster. Free the cluster we've been trying
7722 * to use, and go to the next block group
7724 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
7725 spin_unlock(&last_ptr
->refill_lock
);
7731 * We are doing an unclustered alloc, set the fragmented flag so
7732 * we don't bother trying to setup a cluster again until we get
7735 if (unlikely(last_ptr
)) {
7736 spin_lock(&last_ptr
->lock
);
7737 last_ptr
->fragmented
= 1;
7738 spin_unlock(&last_ptr
->lock
);
7741 struct btrfs_free_space_ctl
*ctl
=
7742 block_group
->free_space_ctl
;
7744 spin_lock(&ctl
->tree_lock
);
7745 if (ctl
->free_space
<
7746 num_bytes
+ empty_cluster
+ empty_size
) {
7747 if (ctl
->free_space
> max_extent_size
)
7748 max_extent_size
= ctl
->free_space
;
7749 spin_unlock(&ctl
->tree_lock
);
7752 spin_unlock(&ctl
->tree_lock
);
7755 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
7756 num_bytes
, empty_size
,
7759 * If we didn't find a chunk, and we haven't failed on this
7760 * block group before, and this block group is in the middle of
7761 * caching and we are ok with waiting, then go ahead and wait
7762 * for progress to be made, and set failed_alloc to true.
7764 * If failed_alloc is true then we've already waited on this
7765 * block group once and should move on to the next block group.
7767 if (!offset
&& !failed_alloc
&& !cached
&&
7768 loop
> LOOP_CACHING_NOWAIT
) {
7769 wait_block_group_cache_progress(block_group
,
7770 num_bytes
+ empty_size
);
7771 failed_alloc
= true;
7772 goto have_block_group
;
7773 } else if (!offset
) {
7777 search_start
= ALIGN(offset
, fs_info
->stripesize
);
7779 /* move on to the next group */
7780 if (search_start
+ num_bytes
>
7781 block_group
->key
.objectid
+ block_group
->key
.offset
) {
7782 btrfs_add_free_space(block_group
, offset
, num_bytes
);
7786 if (offset
< search_start
)
7787 btrfs_add_free_space(block_group
, offset
,
7788 search_start
- offset
);
7789 BUG_ON(offset
> search_start
);
7791 ret
= btrfs_add_reserved_bytes(block_group
, ram_bytes
,
7792 num_bytes
, delalloc
);
7793 if (ret
== -EAGAIN
) {
7794 btrfs_add_free_space(block_group
, offset
, num_bytes
);
7797 btrfs_inc_block_group_reservations(block_group
);
7799 /* we are all good, lets return */
7800 ins
->objectid
= search_start
;
7801 ins
->offset
= num_bytes
;
7803 trace_btrfs_reserve_extent(fs_info
, block_group
,
7804 search_start
, num_bytes
);
7805 btrfs_release_block_group(block_group
, delalloc
);
7808 failed_cluster_refill
= false;
7809 failed_alloc
= false;
7810 BUG_ON(index
!= get_block_group_index(block_group
));
7811 btrfs_release_block_group(block_group
, delalloc
);
7813 up_read(&space_info
->groups_sem
);
7815 if ((loop
== LOOP_CACHING_NOWAIT
) && have_caching_bg
7816 && !orig_have_caching_bg
)
7817 orig_have_caching_bg
= true;
7819 if (!ins
->objectid
&& loop
>= LOOP_CACHING_WAIT
&& have_caching_bg
)
7822 if (!ins
->objectid
&& ++index
< BTRFS_NR_RAID_TYPES
)
7826 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7827 * caching kthreads as we move along
7828 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7829 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7830 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7833 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
) {
7835 if (loop
== LOOP_CACHING_NOWAIT
) {
7837 * We want to skip the LOOP_CACHING_WAIT step if we
7838 * don't have any uncached bgs and we've already done a
7839 * full search through.
7841 if (orig_have_caching_bg
|| !full_search
)
7842 loop
= LOOP_CACHING_WAIT
;
7844 loop
= LOOP_ALLOC_CHUNK
;
7849 if (loop
== LOOP_ALLOC_CHUNK
) {
7850 struct btrfs_trans_handle
*trans
;
7853 trans
= current
->journal_info
;
7857 trans
= btrfs_join_transaction(root
);
7859 if (IS_ERR(trans
)) {
7860 ret
= PTR_ERR(trans
);
7864 ret
= do_chunk_alloc(trans
, fs_info
, flags
,
7868 * If we can't allocate a new chunk we've already looped
7869 * through at least once, move on to the NO_EMPTY_SIZE
7873 loop
= LOOP_NO_EMPTY_SIZE
;
7876 * Do not bail out on ENOSPC since we
7877 * can do more things.
7879 if (ret
< 0 && ret
!= -ENOSPC
)
7880 btrfs_abort_transaction(trans
, ret
);
7884 btrfs_end_transaction(trans
);
7889 if (loop
== LOOP_NO_EMPTY_SIZE
) {
7891 * Don't loop again if we already have no empty_size and
7894 if (empty_size
== 0 &&
7895 empty_cluster
== 0) {
7904 } else if (!ins
->objectid
) {
7906 } else if (ins
->objectid
) {
7907 if (!use_cluster
&& last_ptr
) {
7908 spin_lock(&last_ptr
->lock
);
7909 last_ptr
->window_start
= ins
->objectid
;
7910 spin_unlock(&last_ptr
->lock
);
7915 if (ret
== -ENOSPC
) {
7916 spin_lock(&space_info
->lock
);
7917 space_info
->max_extent_size
= max_extent_size
;
7918 spin_unlock(&space_info
->lock
);
7919 ins
->offset
= max_extent_size
;
7924 static void dump_space_info(struct btrfs_fs_info
*fs_info
,
7925 struct btrfs_space_info
*info
, u64 bytes
,
7926 int dump_block_groups
)
7928 struct btrfs_block_group_cache
*cache
;
7931 spin_lock(&info
->lock
);
7932 btrfs_info(fs_info
, "space_info %llu has %llu free, is %sfull",
7934 info
->total_bytes
- btrfs_space_info_used(info
, true),
7935 info
->full
? "" : "not ");
7937 "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu",
7938 info
->total_bytes
, info
->bytes_used
, info
->bytes_pinned
,
7939 info
->bytes_reserved
, info
->bytes_may_use
,
7940 info
->bytes_readonly
);
7941 spin_unlock(&info
->lock
);
7943 if (!dump_block_groups
)
7946 down_read(&info
->groups_sem
);
7948 list_for_each_entry(cache
, &info
->block_groups
[index
], list
) {
7949 spin_lock(&cache
->lock
);
7951 "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s",
7952 cache
->key
.objectid
, cache
->key
.offset
,
7953 btrfs_block_group_used(&cache
->item
), cache
->pinned
,
7954 cache
->reserved
, cache
->ro
? "[readonly]" : "");
7955 btrfs_dump_free_space(cache
, bytes
);
7956 spin_unlock(&cache
->lock
);
7958 if (++index
< BTRFS_NR_RAID_TYPES
)
7960 up_read(&info
->groups_sem
);
7963 int btrfs_reserve_extent(struct btrfs_root
*root
, u64 ram_bytes
,
7964 u64 num_bytes
, u64 min_alloc_size
,
7965 u64 empty_size
, u64 hint_byte
,
7966 struct btrfs_key
*ins
, int is_data
, int delalloc
)
7968 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7969 bool final_tried
= num_bytes
== min_alloc_size
;
7973 flags
= get_alloc_profile_by_root(root
, is_data
);
7975 WARN_ON(num_bytes
< fs_info
->sectorsize
);
7976 ret
= find_free_extent(fs_info
, ram_bytes
, num_bytes
, empty_size
,
7977 hint_byte
, ins
, flags
, delalloc
);
7978 if (!ret
&& !is_data
) {
7979 btrfs_dec_block_group_reservations(fs_info
, ins
->objectid
);
7980 } else if (ret
== -ENOSPC
) {
7981 if (!final_tried
&& ins
->offset
) {
7982 num_bytes
= min(num_bytes
>> 1, ins
->offset
);
7983 num_bytes
= round_down(num_bytes
,
7984 fs_info
->sectorsize
);
7985 num_bytes
= max(num_bytes
, min_alloc_size
);
7986 ram_bytes
= num_bytes
;
7987 if (num_bytes
== min_alloc_size
)
7990 } else if (btrfs_test_opt(fs_info
, ENOSPC_DEBUG
)) {
7991 struct btrfs_space_info
*sinfo
;
7993 sinfo
= __find_space_info(fs_info
, flags
);
7995 "allocation failed flags %llu, wanted %llu",
7998 dump_space_info(fs_info
, sinfo
, num_bytes
, 1);
8005 static int __btrfs_free_reserved_extent(struct btrfs_fs_info
*fs_info
,
8007 int pin
, int delalloc
)
8009 struct btrfs_block_group_cache
*cache
;
8012 cache
= btrfs_lookup_block_group(fs_info
, start
);
8014 btrfs_err(fs_info
, "Unable to find block group for %llu",
8020 pin_down_extent(fs_info
, cache
, start
, len
, 1);
8022 if (btrfs_test_opt(fs_info
, DISCARD
))
8023 ret
= btrfs_discard_extent(fs_info
, start
, len
, NULL
);
8024 btrfs_add_free_space(cache
, start
, len
);
8025 btrfs_free_reserved_bytes(cache
, len
, delalloc
);
8026 trace_btrfs_reserved_extent_free(fs_info
, start
, len
);
8029 btrfs_put_block_group(cache
);
8033 int btrfs_free_reserved_extent(struct btrfs_fs_info
*fs_info
,
8034 u64 start
, u64 len
, int delalloc
)
8036 return __btrfs_free_reserved_extent(fs_info
, start
, len
, 0, delalloc
);
8039 int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info
*fs_info
,
8042 return __btrfs_free_reserved_extent(fs_info
, start
, len
, 1, 0);
8045 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
8046 struct btrfs_fs_info
*fs_info
,
8047 u64 parent
, u64 root_objectid
,
8048 u64 flags
, u64 owner
, u64 offset
,
8049 struct btrfs_key
*ins
, int ref_mod
)
8052 struct btrfs_extent_item
*extent_item
;
8053 struct btrfs_extent_inline_ref
*iref
;
8054 struct btrfs_path
*path
;
8055 struct extent_buffer
*leaf
;
8060 type
= BTRFS_SHARED_DATA_REF_KEY
;
8062 type
= BTRFS_EXTENT_DATA_REF_KEY
;
8064 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
8066 path
= btrfs_alloc_path();
8070 path
->leave_spinning
= 1;
8071 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
8074 btrfs_free_path(path
);
8078 leaf
= path
->nodes
[0];
8079 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
8080 struct btrfs_extent_item
);
8081 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
8082 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
8083 btrfs_set_extent_flags(leaf
, extent_item
,
8084 flags
| BTRFS_EXTENT_FLAG_DATA
);
8086 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
8087 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
8089 struct btrfs_shared_data_ref
*ref
;
8090 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
8091 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
8092 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
8094 struct btrfs_extent_data_ref
*ref
;
8095 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
8096 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
8097 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
8098 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
8099 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
8102 btrfs_mark_buffer_dirty(path
->nodes
[0]);
8103 btrfs_free_path(path
);
8105 ret
= remove_from_free_space_tree(trans
, fs_info
, ins
->objectid
,
8110 ret
= update_block_group(trans
, fs_info
, ins
->objectid
, ins
->offset
, 1);
8111 if (ret
) { /* -ENOENT, logic error */
8112 btrfs_err(fs_info
, "update block group failed for %llu %llu",
8113 ins
->objectid
, ins
->offset
);
8116 trace_btrfs_reserved_extent_alloc(fs_info
, ins
->objectid
, ins
->offset
);
8120 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
8121 struct btrfs_fs_info
*fs_info
,
8122 u64 parent
, u64 root_objectid
,
8123 u64 flags
, struct btrfs_disk_key
*key
,
8124 int level
, struct btrfs_key
*ins
)
8127 struct btrfs_extent_item
*extent_item
;
8128 struct btrfs_tree_block_info
*block_info
;
8129 struct btrfs_extent_inline_ref
*iref
;
8130 struct btrfs_path
*path
;
8131 struct extent_buffer
*leaf
;
8132 u32 size
= sizeof(*extent_item
) + sizeof(*iref
);
8133 u64 num_bytes
= ins
->offset
;
8134 bool skinny_metadata
= btrfs_fs_incompat(fs_info
, SKINNY_METADATA
);
8136 if (!skinny_metadata
)
8137 size
+= sizeof(*block_info
);
8139 path
= btrfs_alloc_path();
8141 btrfs_free_and_pin_reserved_extent(fs_info
, ins
->objectid
,
8146 path
->leave_spinning
= 1;
8147 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
8150 btrfs_free_path(path
);
8151 btrfs_free_and_pin_reserved_extent(fs_info
, ins
->objectid
,
8156 leaf
= path
->nodes
[0];
8157 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
8158 struct btrfs_extent_item
);
8159 btrfs_set_extent_refs(leaf
, extent_item
, 1);
8160 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
8161 btrfs_set_extent_flags(leaf
, extent_item
,
8162 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
8164 if (skinny_metadata
) {
8165 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
8166 num_bytes
= fs_info
->nodesize
;
8168 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
8169 btrfs_set_tree_block_key(leaf
, block_info
, key
);
8170 btrfs_set_tree_block_level(leaf
, block_info
, level
);
8171 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
8175 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
8176 btrfs_set_extent_inline_ref_type(leaf
, iref
,
8177 BTRFS_SHARED_BLOCK_REF_KEY
);
8178 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
8180 btrfs_set_extent_inline_ref_type(leaf
, iref
,
8181 BTRFS_TREE_BLOCK_REF_KEY
);
8182 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
8185 btrfs_mark_buffer_dirty(leaf
);
8186 btrfs_free_path(path
);
8188 ret
= remove_from_free_space_tree(trans
, fs_info
, ins
->objectid
,
8193 ret
= update_block_group(trans
, fs_info
, ins
->objectid
,
8194 fs_info
->nodesize
, 1);
8195 if (ret
) { /* -ENOENT, logic error */
8196 btrfs_err(fs_info
, "update block group failed for %llu %llu",
8197 ins
->objectid
, ins
->offset
);
8201 trace_btrfs_reserved_extent_alloc(fs_info
, ins
->objectid
,
8206 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
8207 u64 root_objectid
, u64 owner
,
8208 u64 offset
, u64 ram_bytes
,
8209 struct btrfs_key
*ins
)
8211 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
8214 BUG_ON(root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
8216 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, ins
->objectid
,
8218 root_objectid
, owner
, offset
,
8219 ram_bytes
, BTRFS_ADD_DELAYED_EXTENT
);
8224 * this is used by the tree logging recovery code. It records that
8225 * an extent has been allocated and makes sure to clear the free
8226 * space cache bits as well
8228 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
8229 struct btrfs_fs_info
*fs_info
,
8230 u64 root_objectid
, u64 owner
, u64 offset
,
8231 struct btrfs_key
*ins
)
8234 struct btrfs_block_group_cache
*block_group
;
8235 struct btrfs_space_info
*space_info
;
8238 * Mixed block groups will exclude before processing the log so we only
8239 * need to do the exclude dance if this fs isn't mixed.
8241 if (!btrfs_fs_incompat(fs_info
, MIXED_GROUPS
)) {
8242 ret
= __exclude_logged_extent(fs_info
, ins
->objectid
,
8248 block_group
= btrfs_lookup_block_group(fs_info
, ins
->objectid
);
8252 space_info
= block_group
->space_info
;
8253 spin_lock(&space_info
->lock
);
8254 spin_lock(&block_group
->lock
);
8255 space_info
->bytes_reserved
+= ins
->offset
;
8256 block_group
->reserved
+= ins
->offset
;
8257 spin_unlock(&block_group
->lock
);
8258 spin_unlock(&space_info
->lock
);
8260 ret
= alloc_reserved_file_extent(trans
, fs_info
, 0, root_objectid
,
8261 0, owner
, offset
, ins
, 1);
8262 btrfs_put_block_group(block_group
);
8266 static struct extent_buffer
*
8267 btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
8268 u64 bytenr
, int level
)
8270 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8271 struct extent_buffer
*buf
;
8273 buf
= btrfs_find_create_tree_block(fs_info
, bytenr
);
8277 btrfs_set_header_generation(buf
, trans
->transid
);
8278 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, buf
, level
);
8279 btrfs_tree_lock(buf
);
8280 clean_tree_block(fs_info
, buf
);
8281 clear_bit(EXTENT_BUFFER_STALE
, &buf
->bflags
);
8283 btrfs_set_lock_blocking(buf
);
8284 set_extent_buffer_uptodate(buf
);
8286 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
8287 buf
->log_index
= root
->log_transid
% 2;
8289 * we allow two log transactions at a time, use different
8290 * EXENT bit to differentiate dirty pages.
8292 if (buf
->log_index
== 0)
8293 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
8294 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
8296 set_extent_new(&root
->dirty_log_pages
, buf
->start
,
8297 buf
->start
+ buf
->len
- 1);
8299 buf
->log_index
= -1;
8300 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
8301 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
8303 trans
->dirty
= true;
8304 /* this returns a buffer locked for blocking */
8308 static struct btrfs_block_rsv
*
8309 use_block_rsv(struct btrfs_trans_handle
*trans
,
8310 struct btrfs_root
*root
, u32 blocksize
)
8312 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8313 struct btrfs_block_rsv
*block_rsv
;
8314 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
8316 bool global_updated
= false;
8318 block_rsv
= get_block_rsv(trans
, root
);
8320 if (unlikely(block_rsv
->size
== 0))
8323 ret
= block_rsv_use_bytes(block_rsv
, blocksize
);
8327 if (block_rsv
->failfast
)
8328 return ERR_PTR(ret
);
8330 if (block_rsv
->type
== BTRFS_BLOCK_RSV_GLOBAL
&& !global_updated
) {
8331 global_updated
= true;
8332 update_global_block_rsv(fs_info
);
8336 if (btrfs_test_opt(fs_info
, ENOSPC_DEBUG
)) {
8337 static DEFINE_RATELIMIT_STATE(_rs
,
8338 DEFAULT_RATELIMIT_INTERVAL
* 10,
8339 /*DEFAULT_RATELIMIT_BURST*/ 1);
8340 if (__ratelimit(&_rs
))
8342 "BTRFS: block rsv returned %d\n", ret
);
8345 ret
= reserve_metadata_bytes(root
, block_rsv
, blocksize
,
8346 BTRFS_RESERVE_NO_FLUSH
);
8350 * If we couldn't reserve metadata bytes try and use some from
8351 * the global reserve if its space type is the same as the global
8354 if (block_rsv
->type
!= BTRFS_BLOCK_RSV_GLOBAL
&&
8355 block_rsv
->space_info
== global_rsv
->space_info
) {
8356 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
8360 return ERR_PTR(ret
);
8363 static void unuse_block_rsv(struct btrfs_fs_info
*fs_info
,
8364 struct btrfs_block_rsv
*block_rsv
, u32 blocksize
)
8366 block_rsv_add_bytes(block_rsv
, blocksize
, 0);
8367 block_rsv_release_bytes(fs_info
, block_rsv
, NULL
, 0);
8371 * finds a free extent and does all the dirty work required for allocation
8372 * returns the tree buffer or an ERR_PTR on error.
8374 struct extent_buffer
*btrfs_alloc_tree_block(struct btrfs_trans_handle
*trans
,
8375 struct btrfs_root
*root
,
8376 u64 parent
, u64 root_objectid
,
8377 const struct btrfs_disk_key
*key
,
8378 int level
, u64 hint
,
8381 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8382 struct btrfs_key ins
;
8383 struct btrfs_block_rsv
*block_rsv
;
8384 struct extent_buffer
*buf
;
8385 struct btrfs_delayed_extent_op
*extent_op
;
8388 u32 blocksize
= fs_info
->nodesize
;
8389 bool skinny_metadata
= btrfs_fs_incompat(fs_info
, SKINNY_METADATA
);
8391 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8392 if (btrfs_is_testing(fs_info
)) {
8393 buf
= btrfs_init_new_buffer(trans
, root
, root
->alloc_bytenr
,
8396 root
->alloc_bytenr
+= blocksize
;
8401 block_rsv
= use_block_rsv(trans
, root
, blocksize
);
8402 if (IS_ERR(block_rsv
))
8403 return ERR_CAST(block_rsv
);
8405 ret
= btrfs_reserve_extent(root
, blocksize
, blocksize
, blocksize
,
8406 empty_size
, hint
, &ins
, 0, 0);
8410 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
, level
);
8413 goto out_free_reserved
;
8416 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
8418 parent
= ins
.objectid
;
8419 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
8423 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
8424 extent_op
= btrfs_alloc_delayed_extent_op();
8430 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
8432 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
8433 extent_op
->flags_to_set
= flags
;
8434 extent_op
->update_key
= skinny_metadata
? false : true;
8435 extent_op
->update_flags
= true;
8436 extent_op
->is_data
= false;
8437 extent_op
->level
= level
;
8439 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
,
8440 ins
.objectid
, ins
.offset
,
8441 parent
, root_objectid
, level
,
8442 BTRFS_ADD_DELAYED_EXTENT
,
8445 goto out_free_delayed
;
8450 btrfs_free_delayed_extent_op(extent_op
);
8452 free_extent_buffer(buf
);
8454 btrfs_free_reserved_extent(fs_info
, ins
.objectid
, ins
.offset
, 0);
8456 unuse_block_rsv(fs_info
, block_rsv
, blocksize
);
8457 return ERR_PTR(ret
);
8460 struct walk_control
{
8461 u64 refs
[BTRFS_MAX_LEVEL
];
8462 u64 flags
[BTRFS_MAX_LEVEL
];
8463 struct btrfs_key update_progress
;
8474 #define DROP_REFERENCE 1
8475 #define UPDATE_BACKREF 2
8477 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
8478 struct btrfs_root
*root
,
8479 struct walk_control
*wc
,
8480 struct btrfs_path
*path
)
8482 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8488 struct btrfs_key key
;
8489 struct extent_buffer
*eb
;
8494 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
8495 wc
->reada_count
= wc
->reada_count
* 2 / 3;
8496 wc
->reada_count
= max(wc
->reada_count
, 2);
8498 wc
->reada_count
= wc
->reada_count
* 3 / 2;
8499 wc
->reada_count
= min_t(int, wc
->reada_count
,
8500 BTRFS_NODEPTRS_PER_BLOCK(fs_info
));
8503 eb
= path
->nodes
[wc
->level
];
8504 nritems
= btrfs_header_nritems(eb
);
8506 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
8507 if (nread
>= wc
->reada_count
)
8511 bytenr
= btrfs_node_blockptr(eb
, slot
);
8512 generation
= btrfs_node_ptr_generation(eb
, slot
);
8514 if (slot
== path
->slots
[wc
->level
])
8517 if (wc
->stage
== UPDATE_BACKREF
&&
8518 generation
<= root
->root_key
.offset
)
8521 /* We don't lock the tree block, it's OK to be racy here */
8522 ret
= btrfs_lookup_extent_info(trans
, fs_info
, bytenr
,
8523 wc
->level
- 1, 1, &refs
,
8525 /* We don't care about errors in readahead. */
8530 if (wc
->stage
== DROP_REFERENCE
) {
8534 if (wc
->level
== 1 &&
8535 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8537 if (!wc
->update_ref
||
8538 generation
<= root
->root_key
.offset
)
8540 btrfs_node_key_to_cpu(eb
, &key
, slot
);
8541 ret
= btrfs_comp_cpu_keys(&key
,
8542 &wc
->update_progress
);
8546 if (wc
->level
== 1 &&
8547 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8551 readahead_tree_block(fs_info
, bytenr
);
8554 wc
->reada_slot
= slot
;
8558 * helper to process tree block while walking down the tree.
8560 * when wc->stage == UPDATE_BACKREF, this function updates
8561 * back refs for pointers in the block.
8563 * NOTE: return value 1 means we should stop walking down.
8565 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
8566 struct btrfs_root
*root
,
8567 struct btrfs_path
*path
,
8568 struct walk_control
*wc
, int lookup_info
)
8570 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8571 int level
= wc
->level
;
8572 struct extent_buffer
*eb
= path
->nodes
[level
];
8573 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
8576 if (wc
->stage
== UPDATE_BACKREF
&&
8577 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
8581 * when reference count of tree block is 1, it won't increase
8582 * again. once full backref flag is set, we never clear it.
8585 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
8586 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
8587 BUG_ON(!path
->locks
[level
]);
8588 ret
= btrfs_lookup_extent_info(trans
, fs_info
,
8589 eb
->start
, level
, 1,
8592 BUG_ON(ret
== -ENOMEM
);
8595 BUG_ON(wc
->refs
[level
] == 0);
8598 if (wc
->stage
== DROP_REFERENCE
) {
8599 if (wc
->refs
[level
] > 1)
8602 if (path
->locks
[level
] && !wc
->keep_locks
) {
8603 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8604 path
->locks
[level
] = 0;
8609 /* wc->stage == UPDATE_BACKREF */
8610 if (!(wc
->flags
[level
] & flag
)) {
8611 BUG_ON(!path
->locks
[level
]);
8612 ret
= btrfs_inc_ref(trans
, root
, eb
, 1);
8613 BUG_ON(ret
); /* -ENOMEM */
8614 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
8615 BUG_ON(ret
); /* -ENOMEM */
8616 ret
= btrfs_set_disk_extent_flags(trans
, fs_info
, eb
->start
,
8618 btrfs_header_level(eb
), 0);
8619 BUG_ON(ret
); /* -ENOMEM */
8620 wc
->flags
[level
] |= flag
;
8624 * the block is shared by multiple trees, so it's not good to
8625 * keep the tree lock
8627 if (path
->locks
[level
] && level
> 0) {
8628 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8629 path
->locks
[level
] = 0;
8635 * helper to process tree block pointer.
8637 * when wc->stage == DROP_REFERENCE, this function checks
8638 * reference count of the block pointed to. if the block
8639 * is shared and we need update back refs for the subtree
8640 * rooted at the block, this function changes wc->stage to
8641 * UPDATE_BACKREF. if the block is shared and there is no
8642 * need to update back, this function drops the reference
8645 * NOTE: return value 1 means we should stop walking down.
8647 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
8648 struct btrfs_root
*root
,
8649 struct btrfs_path
*path
,
8650 struct walk_control
*wc
, int *lookup_info
)
8652 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8657 struct btrfs_key key
;
8658 struct extent_buffer
*next
;
8659 int level
= wc
->level
;
8662 bool need_account
= false;
8664 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
8665 path
->slots
[level
]);
8667 * if the lower level block was created before the snapshot
8668 * was created, we know there is no need to update back refs
8671 if (wc
->stage
== UPDATE_BACKREF
&&
8672 generation
<= root
->root_key
.offset
) {
8677 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
8678 blocksize
= fs_info
->nodesize
;
8680 next
= find_extent_buffer(fs_info
, bytenr
);
8682 next
= btrfs_find_create_tree_block(fs_info
, bytenr
);
8684 return PTR_ERR(next
);
8686 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, next
,
8690 btrfs_tree_lock(next
);
8691 btrfs_set_lock_blocking(next
);
8693 ret
= btrfs_lookup_extent_info(trans
, fs_info
, bytenr
, level
- 1, 1,
8694 &wc
->refs
[level
- 1],
8695 &wc
->flags
[level
- 1]);
8699 if (unlikely(wc
->refs
[level
- 1] == 0)) {
8700 btrfs_err(fs_info
, "Missing references.");
8706 if (wc
->stage
== DROP_REFERENCE
) {
8707 if (wc
->refs
[level
- 1] > 1) {
8708 need_account
= true;
8710 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8713 if (!wc
->update_ref
||
8714 generation
<= root
->root_key
.offset
)
8717 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
8718 path
->slots
[level
]);
8719 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
8723 wc
->stage
= UPDATE_BACKREF
;
8724 wc
->shared_level
= level
- 1;
8728 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8732 if (!btrfs_buffer_uptodate(next
, generation
, 0)) {
8733 btrfs_tree_unlock(next
);
8734 free_extent_buffer(next
);
8740 if (reada
&& level
== 1)
8741 reada_walk_down(trans
, root
, wc
, path
);
8742 next
= read_tree_block(fs_info
, bytenr
, generation
);
8744 return PTR_ERR(next
);
8745 } else if (!extent_buffer_uptodate(next
)) {
8746 free_extent_buffer(next
);
8749 btrfs_tree_lock(next
);
8750 btrfs_set_lock_blocking(next
);
8754 ASSERT(level
== btrfs_header_level(next
));
8755 if (level
!= btrfs_header_level(next
)) {
8756 btrfs_err(root
->fs_info
, "mismatched level");
8760 path
->nodes
[level
] = next
;
8761 path
->slots
[level
] = 0;
8762 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8768 wc
->refs
[level
- 1] = 0;
8769 wc
->flags
[level
- 1] = 0;
8770 if (wc
->stage
== DROP_REFERENCE
) {
8771 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
8772 parent
= path
->nodes
[level
]->start
;
8774 ASSERT(root
->root_key
.objectid
==
8775 btrfs_header_owner(path
->nodes
[level
]));
8776 if (root
->root_key
.objectid
!=
8777 btrfs_header_owner(path
->nodes
[level
])) {
8778 btrfs_err(root
->fs_info
,
8779 "mismatched block owner");
8787 ret
= btrfs_qgroup_trace_subtree(trans
, root
, next
,
8788 generation
, level
- 1);
8790 btrfs_err_rl(fs_info
,
8791 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
8795 ret
= btrfs_free_extent(trans
, fs_info
, bytenr
, blocksize
,
8796 parent
, root
->root_key
.objectid
,
8806 btrfs_tree_unlock(next
);
8807 free_extent_buffer(next
);
8813 * helper to process tree block while walking up the tree.
8815 * when wc->stage == DROP_REFERENCE, this function drops
8816 * reference count on the block.
8818 * when wc->stage == UPDATE_BACKREF, this function changes
8819 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8820 * to UPDATE_BACKREF previously while processing the block.
8822 * NOTE: return value 1 means we should stop walking up.
8824 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
8825 struct btrfs_root
*root
,
8826 struct btrfs_path
*path
,
8827 struct walk_control
*wc
)
8829 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8831 int level
= wc
->level
;
8832 struct extent_buffer
*eb
= path
->nodes
[level
];
8835 if (wc
->stage
== UPDATE_BACKREF
) {
8836 BUG_ON(wc
->shared_level
< level
);
8837 if (level
< wc
->shared_level
)
8840 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
8844 wc
->stage
= DROP_REFERENCE
;
8845 wc
->shared_level
= -1;
8846 path
->slots
[level
] = 0;
8849 * check reference count again if the block isn't locked.
8850 * we should start walking down the tree again if reference
8853 if (!path
->locks
[level
]) {
8855 btrfs_tree_lock(eb
);
8856 btrfs_set_lock_blocking(eb
);
8857 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8859 ret
= btrfs_lookup_extent_info(trans
, fs_info
,
8860 eb
->start
, level
, 1,
8864 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8865 path
->locks
[level
] = 0;
8868 BUG_ON(wc
->refs
[level
] == 0);
8869 if (wc
->refs
[level
] == 1) {
8870 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8871 path
->locks
[level
] = 0;
8877 /* wc->stage == DROP_REFERENCE */
8878 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
8880 if (wc
->refs
[level
] == 1) {
8882 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8883 ret
= btrfs_dec_ref(trans
, root
, eb
, 1);
8885 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
8886 BUG_ON(ret
); /* -ENOMEM */
8887 ret
= btrfs_qgroup_trace_leaf_items(trans
, fs_info
, eb
);
8889 btrfs_err_rl(fs_info
,
8890 "error %d accounting leaf items. Quota is out of sync, rescan required.",
8894 /* make block locked assertion in clean_tree_block happy */
8895 if (!path
->locks
[level
] &&
8896 btrfs_header_generation(eb
) == trans
->transid
) {
8897 btrfs_tree_lock(eb
);
8898 btrfs_set_lock_blocking(eb
);
8899 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8901 clean_tree_block(fs_info
, eb
);
8904 if (eb
== root
->node
) {
8905 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8908 BUG_ON(root
->root_key
.objectid
!=
8909 btrfs_header_owner(eb
));
8911 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8912 parent
= path
->nodes
[level
+ 1]->start
;
8914 BUG_ON(root
->root_key
.objectid
!=
8915 btrfs_header_owner(path
->nodes
[level
+ 1]));
8918 btrfs_free_tree_block(trans
, root
, eb
, parent
, wc
->refs
[level
] == 1);
8920 wc
->refs
[level
] = 0;
8921 wc
->flags
[level
] = 0;
8925 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
8926 struct btrfs_root
*root
,
8927 struct btrfs_path
*path
,
8928 struct walk_control
*wc
)
8930 int level
= wc
->level
;
8931 int lookup_info
= 1;
8934 while (level
>= 0) {
8935 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
8942 if (path
->slots
[level
] >=
8943 btrfs_header_nritems(path
->nodes
[level
]))
8946 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
8948 path
->slots
[level
]++;
8957 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
8958 struct btrfs_root
*root
,
8959 struct btrfs_path
*path
,
8960 struct walk_control
*wc
, int max_level
)
8962 int level
= wc
->level
;
8965 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
8966 while (level
< max_level
&& path
->nodes
[level
]) {
8968 if (path
->slots
[level
] + 1 <
8969 btrfs_header_nritems(path
->nodes
[level
])) {
8970 path
->slots
[level
]++;
8973 ret
= walk_up_proc(trans
, root
, path
, wc
);
8977 if (path
->locks
[level
]) {
8978 btrfs_tree_unlock_rw(path
->nodes
[level
],
8979 path
->locks
[level
]);
8980 path
->locks
[level
] = 0;
8982 free_extent_buffer(path
->nodes
[level
]);
8983 path
->nodes
[level
] = NULL
;
8991 * drop a subvolume tree.
8993 * this function traverses the tree freeing any blocks that only
8994 * referenced by the tree.
8996 * when a shared tree block is found. this function decreases its
8997 * reference count by one. if update_ref is true, this function
8998 * also make sure backrefs for the shared block and all lower level
8999 * blocks are properly updated.
9001 * If called with for_reloc == 0, may exit early with -EAGAIN
9003 int btrfs_drop_snapshot(struct btrfs_root
*root
,
9004 struct btrfs_block_rsv
*block_rsv
, int update_ref
,
9007 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
9008 struct btrfs_path
*path
;
9009 struct btrfs_trans_handle
*trans
;
9010 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
9011 struct btrfs_root_item
*root_item
= &root
->root_item
;
9012 struct walk_control
*wc
;
9013 struct btrfs_key key
;
9017 bool root_dropped
= false;
9019 btrfs_debug(fs_info
, "Drop subvolume %llu", root
->objectid
);
9021 path
= btrfs_alloc_path();
9027 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
9029 btrfs_free_path(path
);
9034 trans
= btrfs_start_transaction(tree_root
, 0);
9035 if (IS_ERR(trans
)) {
9036 err
= PTR_ERR(trans
);
9041 trans
->block_rsv
= block_rsv
;
9043 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
9044 level
= btrfs_header_level(root
->node
);
9045 path
->nodes
[level
] = btrfs_lock_root_node(root
);
9046 btrfs_set_lock_blocking(path
->nodes
[level
]);
9047 path
->slots
[level
] = 0;
9048 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
9049 memset(&wc
->update_progress
, 0,
9050 sizeof(wc
->update_progress
));
9052 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
9053 memcpy(&wc
->update_progress
, &key
,
9054 sizeof(wc
->update_progress
));
9056 level
= root_item
->drop_level
;
9058 path
->lowest_level
= level
;
9059 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
9060 path
->lowest_level
= 0;
9068 * unlock our path, this is safe because only this
9069 * function is allowed to delete this snapshot
9071 btrfs_unlock_up_safe(path
, 0);
9073 level
= btrfs_header_level(root
->node
);
9075 btrfs_tree_lock(path
->nodes
[level
]);
9076 btrfs_set_lock_blocking(path
->nodes
[level
]);
9077 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
9079 ret
= btrfs_lookup_extent_info(trans
, fs_info
,
9080 path
->nodes
[level
]->start
,
9081 level
, 1, &wc
->refs
[level
],
9087 BUG_ON(wc
->refs
[level
] == 0);
9089 if (level
== root_item
->drop_level
)
9092 btrfs_tree_unlock(path
->nodes
[level
]);
9093 path
->locks
[level
] = 0;
9094 WARN_ON(wc
->refs
[level
] != 1);
9100 wc
->shared_level
= -1;
9101 wc
->stage
= DROP_REFERENCE
;
9102 wc
->update_ref
= update_ref
;
9104 wc
->for_reloc
= for_reloc
;
9105 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(fs_info
);
9109 ret
= walk_down_tree(trans
, root
, path
, wc
);
9115 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
9122 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
9126 if (wc
->stage
== DROP_REFERENCE
) {
9128 btrfs_node_key(path
->nodes
[level
],
9129 &root_item
->drop_progress
,
9130 path
->slots
[level
]);
9131 root_item
->drop_level
= level
;
9134 BUG_ON(wc
->level
== 0);
9135 if (btrfs_should_end_transaction(trans
) ||
9136 (!for_reloc
&& btrfs_need_cleaner_sleep(fs_info
))) {
9137 ret
= btrfs_update_root(trans
, tree_root
,
9141 btrfs_abort_transaction(trans
, ret
);
9146 btrfs_end_transaction_throttle(trans
);
9147 if (!for_reloc
&& btrfs_need_cleaner_sleep(fs_info
)) {
9148 btrfs_debug(fs_info
,
9149 "drop snapshot early exit");
9154 trans
= btrfs_start_transaction(tree_root
, 0);
9155 if (IS_ERR(trans
)) {
9156 err
= PTR_ERR(trans
);
9160 trans
->block_rsv
= block_rsv
;
9163 btrfs_release_path(path
);
9167 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
9169 btrfs_abort_transaction(trans
, ret
);
9173 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
9174 ret
= btrfs_find_root(tree_root
, &root
->root_key
, path
,
9177 btrfs_abort_transaction(trans
, ret
);
9180 } else if (ret
> 0) {
9181 /* if we fail to delete the orphan item this time
9182 * around, it'll get picked up the next time.
9184 * The most common failure here is just -ENOENT.
9186 btrfs_del_orphan_item(trans
, tree_root
,
9187 root
->root_key
.objectid
);
9191 if (test_bit(BTRFS_ROOT_IN_RADIX
, &root
->state
)) {
9192 btrfs_add_dropped_root(trans
, root
);
9194 free_extent_buffer(root
->node
);
9195 free_extent_buffer(root
->commit_root
);
9196 btrfs_put_fs_root(root
);
9198 root_dropped
= true;
9200 btrfs_end_transaction_throttle(trans
);
9203 btrfs_free_path(path
);
9206 * So if we need to stop dropping the snapshot for whatever reason we
9207 * need to make sure to add it back to the dead root list so that we
9208 * keep trying to do the work later. This also cleans up roots if we
9209 * don't have it in the radix (like when we recover after a power fail
9210 * or unmount) so we don't leak memory.
9212 if (!for_reloc
&& root_dropped
== false)
9213 btrfs_add_dead_root(root
);
9214 if (err
&& err
!= -EAGAIN
)
9215 btrfs_handle_fs_error(fs_info
, err
, NULL
);
9220 * drop subtree rooted at tree block 'node'.
9222 * NOTE: this function will unlock and release tree block 'node'
9223 * only used by relocation code
9225 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
9226 struct btrfs_root
*root
,
9227 struct extent_buffer
*node
,
9228 struct extent_buffer
*parent
)
9230 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
9231 struct btrfs_path
*path
;
9232 struct walk_control
*wc
;
9238 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
9240 path
= btrfs_alloc_path();
9244 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
9246 btrfs_free_path(path
);
9250 btrfs_assert_tree_locked(parent
);
9251 parent_level
= btrfs_header_level(parent
);
9252 extent_buffer_get(parent
);
9253 path
->nodes
[parent_level
] = parent
;
9254 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
9256 btrfs_assert_tree_locked(node
);
9257 level
= btrfs_header_level(node
);
9258 path
->nodes
[level
] = node
;
9259 path
->slots
[level
] = 0;
9260 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
9262 wc
->refs
[parent_level
] = 1;
9263 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
9265 wc
->shared_level
= -1;
9266 wc
->stage
= DROP_REFERENCE
;
9270 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(fs_info
);
9273 wret
= walk_down_tree(trans
, root
, path
, wc
);
9279 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
9287 btrfs_free_path(path
);
9291 static u64
update_block_group_flags(struct btrfs_fs_info
*fs_info
, u64 flags
)
9297 * if restripe for this chunk_type is on pick target profile and
9298 * return, otherwise do the usual balance
9300 stripped
= get_restripe_target(fs_info
, flags
);
9302 return extended_to_chunk(stripped
);
9304 num_devices
= fs_info
->fs_devices
->rw_devices
;
9306 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
9307 BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
|
9308 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
9310 if (num_devices
== 1) {
9311 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
9312 stripped
= flags
& ~stripped
;
9314 /* turn raid0 into single device chunks */
9315 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
9318 /* turn mirroring into duplication */
9319 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
9320 BTRFS_BLOCK_GROUP_RAID10
))
9321 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
9323 /* they already had raid on here, just return */
9324 if (flags
& stripped
)
9327 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
9328 stripped
= flags
& ~stripped
;
9330 /* switch duplicated blocks with raid1 */
9331 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
9332 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
9334 /* this is drive concat, leave it alone */
9340 static int inc_block_group_ro(struct btrfs_block_group_cache
*cache
, int force
)
9342 struct btrfs_space_info
*sinfo
= cache
->space_info
;
9344 u64 min_allocable_bytes
;
9348 * We need some metadata space and system metadata space for
9349 * allocating chunks in some corner cases until we force to set
9350 * it to be readonly.
9353 (BTRFS_BLOCK_GROUP_SYSTEM
| BTRFS_BLOCK_GROUP_METADATA
)) &&
9355 min_allocable_bytes
= SZ_1M
;
9357 min_allocable_bytes
= 0;
9359 spin_lock(&sinfo
->lock
);
9360 spin_lock(&cache
->lock
);
9368 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
9369 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
9371 if (btrfs_space_info_used(sinfo
, true) + num_bytes
+
9372 min_allocable_bytes
<= sinfo
->total_bytes
) {
9373 sinfo
->bytes_readonly
+= num_bytes
;
9375 list_add_tail(&cache
->ro_list
, &sinfo
->ro_bgs
);
9379 spin_unlock(&cache
->lock
);
9380 spin_unlock(&sinfo
->lock
);
9384 int btrfs_inc_block_group_ro(struct btrfs_fs_info
*fs_info
,
9385 struct btrfs_block_group_cache
*cache
)
9388 struct btrfs_trans_handle
*trans
;
9393 trans
= btrfs_join_transaction(fs_info
->extent_root
);
9395 return PTR_ERR(trans
);
9398 * we're not allowed to set block groups readonly after the dirty
9399 * block groups cache has started writing. If it already started,
9400 * back off and let this transaction commit
9402 mutex_lock(&fs_info
->ro_block_group_mutex
);
9403 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN
, &trans
->transaction
->flags
)) {
9404 u64 transid
= trans
->transid
;
9406 mutex_unlock(&fs_info
->ro_block_group_mutex
);
9407 btrfs_end_transaction(trans
);
9409 ret
= btrfs_wait_for_commit(fs_info
, transid
);
9416 * if we are changing raid levels, try to allocate a corresponding
9417 * block group with the new raid level.
9419 alloc_flags
= update_block_group_flags(fs_info
, cache
->flags
);
9420 if (alloc_flags
!= cache
->flags
) {
9421 ret
= do_chunk_alloc(trans
, fs_info
, alloc_flags
,
9424 * ENOSPC is allowed here, we may have enough space
9425 * already allocated at the new raid level to
9434 ret
= inc_block_group_ro(cache
, 0);
9437 alloc_flags
= get_alloc_profile(fs_info
, cache
->space_info
->flags
);
9438 ret
= do_chunk_alloc(trans
, fs_info
, alloc_flags
,
9442 ret
= inc_block_group_ro(cache
, 0);
9444 if (cache
->flags
& BTRFS_BLOCK_GROUP_SYSTEM
) {
9445 alloc_flags
= update_block_group_flags(fs_info
, cache
->flags
);
9446 mutex_lock(&fs_info
->chunk_mutex
);
9447 check_system_chunk(trans
, fs_info
, alloc_flags
);
9448 mutex_unlock(&fs_info
->chunk_mutex
);
9450 mutex_unlock(&fs_info
->ro_block_group_mutex
);
9452 btrfs_end_transaction(trans
);
9456 int btrfs_force_chunk_alloc(struct btrfs_trans_handle
*trans
,
9457 struct btrfs_fs_info
*fs_info
, u64 type
)
9459 u64 alloc_flags
= get_alloc_profile(fs_info
, type
);
9461 return do_chunk_alloc(trans
, fs_info
, alloc_flags
, CHUNK_ALLOC_FORCE
);
9465 * helper to account the unused space of all the readonly block group in the
9466 * space_info. takes mirrors into account.
9468 u64
btrfs_account_ro_block_groups_free_space(struct btrfs_space_info
*sinfo
)
9470 struct btrfs_block_group_cache
*block_group
;
9474 /* It's df, we don't care if it's racy */
9475 if (list_empty(&sinfo
->ro_bgs
))
9478 spin_lock(&sinfo
->lock
);
9479 list_for_each_entry(block_group
, &sinfo
->ro_bgs
, ro_list
) {
9480 spin_lock(&block_group
->lock
);
9482 if (!block_group
->ro
) {
9483 spin_unlock(&block_group
->lock
);
9487 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_RAID1
|
9488 BTRFS_BLOCK_GROUP_RAID10
|
9489 BTRFS_BLOCK_GROUP_DUP
))
9494 free_bytes
+= (block_group
->key
.offset
-
9495 btrfs_block_group_used(&block_group
->item
)) *
9498 spin_unlock(&block_group
->lock
);
9500 spin_unlock(&sinfo
->lock
);
9505 void btrfs_dec_block_group_ro(struct btrfs_block_group_cache
*cache
)
9507 struct btrfs_space_info
*sinfo
= cache
->space_info
;
9512 spin_lock(&sinfo
->lock
);
9513 spin_lock(&cache
->lock
);
9515 num_bytes
= cache
->key
.offset
- cache
->reserved
-
9516 cache
->pinned
- cache
->bytes_super
-
9517 btrfs_block_group_used(&cache
->item
);
9518 sinfo
->bytes_readonly
-= num_bytes
;
9519 list_del_init(&cache
->ro_list
);
9521 spin_unlock(&cache
->lock
);
9522 spin_unlock(&sinfo
->lock
);
9526 * checks to see if its even possible to relocate this block group.
9528 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9529 * ok to go ahead and try.
9531 int btrfs_can_relocate(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
9533 struct btrfs_root
*root
= fs_info
->extent_root
;
9534 struct btrfs_block_group_cache
*block_group
;
9535 struct btrfs_space_info
*space_info
;
9536 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
9537 struct btrfs_device
*device
;
9538 struct btrfs_trans_handle
*trans
;
9548 debug
= btrfs_test_opt(fs_info
, ENOSPC_DEBUG
);
9550 block_group
= btrfs_lookup_block_group(fs_info
, bytenr
);
9552 /* odd, couldn't find the block group, leave it alone */
9556 "can't find block group for bytenr %llu",
9561 min_free
= btrfs_block_group_used(&block_group
->item
);
9563 /* no bytes used, we're good */
9567 space_info
= block_group
->space_info
;
9568 spin_lock(&space_info
->lock
);
9570 full
= space_info
->full
;
9573 * if this is the last block group we have in this space, we can't
9574 * relocate it unless we're able to allocate a new chunk below.
9576 * Otherwise, we need to make sure we have room in the space to handle
9577 * all of the extents from this block group. If we can, we're good
9579 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
9580 (btrfs_space_info_used(space_info
, false) + min_free
<
9581 space_info
->total_bytes
)) {
9582 spin_unlock(&space_info
->lock
);
9585 spin_unlock(&space_info
->lock
);
9588 * ok we don't have enough space, but maybe we have free space on our
9589 * devices to allocate new chunks for relocation, so loop through our
9590 * alloc devices and guess if we have enough space. if this block
9591 * group is going to be restriped, run checks against the target
9592 * profile instead of the current one.
9604 target
= get_restripe_target(fs_info
, block_group
->flags
);
9606 index
= __get_raid_index(extended_to_chunk(target
));
9609 * this is just a balance, so if we were marked as full
9610 * we know there is no space for a new chunk
9615 "no space to alloc new chunk for block group %llu",
9616 block_group
->key
.objectid
);
9620 index
= get_block_group_index(block_group
);
9623 if (index
== BTRFS_RAID_RAID10
) {
9627 } else if (index
== BTRFS_RAID_RAID1
) {
9629 } else if (index
== BTRFS_RAID_DUP
) {
9632 } else if (index
== BTRFS_RAID_RAID0
) {
9633 dev_min
= fs_devices
->rw_devices
;
9634 min_free
= div64_u64(min_free
, dev_min
);
9637 /* We need to do this so that we can look at pending chunks */
9638 trans
= btrfs_join_transaction(root
);
9639 if (IS_ERR(trans
)) {
9640 ret
= PTR_ERR(trans
);
9644 mutex_lock(&fs_info
->chunk_mutex
);
9645 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
9649 * check to make sure we can actually find a chunk with enough
9650 * space to fit our block group in.
9652 if (device
->total_bytes
> device
->bytes_used
+ min_free
&&
9653 !device
->is_tgtdev_for_dev_replace
) {
9654 ret
= find_free_dev_extent(trans
, device
, min_free
,
9659 if (dev_nr
>= dev_min
)
9665 if (debug
&& ret
== -1)
9667 "no space to allocate a new chunk for block group %llu",
9668 block_group
->key
.objectid
);
9669 mutex_unlock(&fs_info
->chunk_mutex
);
9670 btrfs_end_transaction(trans
);
9672 btrfs_put_block_group(block_group
);
9676 static int find_first_block_group(struct btrfs_fs_info
*fs_info
,
9677 struct btrfs_path
*path
,
9678 struct btrfs_key
*key
)
9680 struct btrfs_root
*root
= fs_info
->extent_root
;
9682 struct btrfs_key found_key
;
9683 struct extent_buffer
*leaf
;
9686 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
9691 slot
= path
->slots
[0];
9692 leaf
= path
->nodes
[0];
9693 if (slot
>= btrfs_header_nritems(leaf
)) {
9694 ret
= btrfs_next_leaf(root
, path
);
9701 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
9703 if (found_key
.objectid
>= key
->objectid
&&
9704 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
9705 struct extent_map_tree
*em_tree
;
9706 struct extent_map
*em
;
9708 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
9709 read_lock(&em_tree
->lock
);
9710 em
= lookup_extent_mapping(em_tree
, found_key
.objectid
,
9712 read_unlock(&em_tree
->lock
);
9715 "logical %llu len %llu found bg but no related chunk",
9716 found_key
.objectid
, found_key
.offset
);
9721 free_extent_map(em
);
9730 void btrfs_put_block_group_cache(struct btrfs_fs_info
*info
)
9732 struct btrfs_block_group_cache
*block_group
;
9736 struct inode
*inode
;
9738 block_group
= btrfs_lookup_first_block_group(info
, last
);
9739 while (block_group
) {
9740 spin_lock(&block_group
->lock
);
9741 if (block_group
->iref
)
9743 spin_unlock(&block_group
->lock
);
9744 block_group
= next_block_group(info
, block_group
);
9753 inode
= block_group
->inode
;
9754 block_group
->iref
= 0;
9755 block_group
->inode
= NULL
;
9756 spin_unlock(&block_group
->lock
);
9757 ASSERT(block_group
->io_ctl
.inode
== NULL
);
9759 last
= block_group
->key
.objectid
+ block_group
->key
.offset
;
9760 btrfs_put_block_group(block_group
);
9765 * Must be called only after stopping all workers, since we could have block
9766 * group caching kthreads running, and therefore they could race with us if we
9767 * freed the block groups before stopping them.
9769 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
9771 struct btrfs_block_group_cache
*block_group
;
9772 struct btrfs_space_info
*space_info
;
9773 struct btrfs_caching_control
*caching_ctl
;
9776 down_write(&info
->commit_root_sem
);
9777 while (!list_empty(&info
->caching_block_groups
)) {
9778 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
9779 struct btrfs_caching_control
, list
);
9780 list_del(&caching_ctl
->list
);
9781 put_caching_control(caching_ctl
);
9783 up_write(&info
->commit_root_sem
);
9785 spin_lock(&info
->unused_bgs_lock
);
9786 while (!list_empty(&info
->unused_bgs
)) {
9787 block_group
= list_first_entry(&info
->unused_bgs
,
9788 struct btrfs_block_group_cache
,
9790 list_del_init(&block_group
->bg_list
);
9791 btrfs_put_block_group(block_group
);
9793 spin_unlock(&info
->unused_bgs_lock
);
9795 spin_lock(&info
->block_group_cache_lock
);
9796 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
9797 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
9799 rb_erase(&block_group
->cache_node
,
9800 &info
->block_group_cache_tree
);
9801 RB_CLEAR_NODE(&block_group
->cache_node
);
9802 spin_unlock(&info
->block_group_cache_lock
);
9804 down_write(&block_group
->space_info
->groups_sem
);
9805 list_del(&block_group
->list
);
9806 up_write(&block_group
->space_info
->groups_sem
);
9809 * We haven't cached this block group, which means we could
9810 * possibly have excluded extents on this block group.
9812 if (block_group
->cached
== BTRFS_CACHE_NO
||
9813 block_group
->cached
== BTRFS_CACHE_ERROR
)
9814 free_excluded_extents(info
, block_group
);
9816 btrfs_remove_free_space_cache(block_group
);
9817 ASSERT(block_group
->cached
!= BTRFS_CACHE_STARTED
);
9818 ASSERT(list_empty(&block_group
->dirty_list
));
9819 ASSERT(list_empty(&block_group
->io_list
));
9820 ASSERT(list_empty(&block_group
->bg_list
));
9821 ASSERT(atomic_read(&block_group
->count
) == 1);
9822 btrfs_put_block_group(block_group
);
9824 spin_lock(&info
->block_group_cache_lock
);
9826 spin_unlock(&info
->block_group_cache_lock
);
9828 /* now that all the block groups are freed, go through and
9829 * free all the space_info structs. This is only called during
9830 * the final stages of unmount, and so we know nobody is
9831 * using them. We call synchronize_rcu() once before we start,
9832 * just to be on the safe side.
9836 release_global_block_rsv(info
);
9838 while (!list_empty(&info
->space_info
)) {
9841 space_info
= list_entry(info
->space_info
.next
,
9842 struct btrfs_space_info
,
9846 * Do not hide this behind enospc_debug, this is actually
9847 * important and indicates a real bug if this happens.
9849 if (WARN_ON(space_info
->bytes_pinned
> 0 ||
9850 space_info
->bytes_reserved
> 0 ||
9851 space_info
->bytes_may_use
> 0))
9852 dump_space_info(info
, space_info
, 0, 0);
9853 list_del(&space_info
->list
);
9854 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++) {
9855 struct kobject
*kobj
;
9856 kobj
= space_info
->block_group_kobjs
[i
];
9857 space_info
->block_group_kobjs
[i
] = NULL
;
9863 kobject_del(&space_info
->kobj
);
9864 kobject_put(&space_info
->kobj
);
9869 static void __link_block_group(struct btrfs_space_info
*space_info
,
9870 struct btrfs_block_group_cache
*cache
)
9872 int index
= get_block_group_index(cache
);
9875 down_write(&space_info
->groups_sem
);
9876 if (list_empty(&space_info
->block_groups
[index
]))
9878 list_add_tail(&cache
->list
, &space_info
->block_groups
[index
]);
9879 up_write(&space_info
->groups_sem
);
9882 struct raid_kobject
*rkobj
;
9885 rkobj
= kzalloc(sizeof(*rkobj
), GFP_NOFS
);
9888 rkobj
->raid_type
= index
;
9889 kobject_init(&rkobj
->kobj
, &btrfs_raid_ktype
);
9890 ret
= kobject_add(&rkobj
->kobj
, &space_info
->kobj
,
9891 "%s", get_raid_name(index
));
9893 kobject_put(&rkobj
->kobj
);
9896 space_info
->block_group_kobjs
[index
] = &rkobj
->kobj
;
9901 btrfs_warn(cache
->fs_info
,
9902 "failed to add kobject for block cache, ignoring");
9905 static struct btrfs_block_group_cache
*
9906 btrfs_create_block_group_cache(struct btrfs_fs_info
*fs_info
,
9907 u64 start
, u64 size
)
9909 struct btrfs_block_group_cache
*cache
;
9911 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
9915 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
9917 if (!cache
->free_space_ctl
) {
9922 cache
->key
.objectid
= start
;
9923 cache
->key
.offset
= size
;
9924 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
9926 cache
->sectorsize
= fs_info
->sectorsize
;
9927 cache
->fs_info
= fs_info
;
9928 cache
->full_stripe_len
= btrfs_full_stripe_len(fs_info
,
9929 &fs_info
->mapping_tree
,
9931 set_free_space_tree_thresholds(cache
);
9933 atomic_set(&cache
->count
, 1);
9934 spin_lock_init(&cache
->lock
);
9935 init_rwsem(&cache
->data_rwsem
);
9936 INIT_LIST_HEAD(&cache
->list
);
9937 INIT_LIST_HEAD(&cache
->cluster_list
);
9938 INIT_LIST_HEAD(&cache
->bg_list
);
9939 INIT_LIST_HEAD(&cache
->ro_list
);
9940 INIT_LIST_HEAD(&cache
->dirty_list
);
9941 INIT_LIST_HEAD(&cache
->io_list
);
9942 btrfs_init_free_space_ctl(cache
);
9943 atomic_set(&cache
->trimming
, 0);
9944 mutex_init(&cache
->free_space_lock
);
9945 btrfs_init_full_stripe_locks_tree(&cache
->full_stripe_locks_root
);
9950 int btrfs_read_block_groups(struct btrfs_fs_info
*info
)
9952 struct btrfs_path
*path
;
9954 struct btrfs_block_group_cache
*cache
;
9955 struct btrfs_space_info
*space_info
;
9956 struct btrfs_key key
;
9957 struct btrfs_key found_key
;
9958 struct extent_buffer
*leaf
;
9964 feature
= btrfs_super_incompat_flags(info
->super_copy
);
9965 mixed
= !!(feature
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
);
9969 key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
9970 path
= btrfs_alloc_path();
9973 path
->reada
= READA_FORWARD
;
9975 cache_gen
= btrfs_super_cache_generation(info
->super_copy
);
9976 if (btrfs_test_opt(info
, SPACE_CACHE
) &&
9977 btrfs_super_generation(info
->super_copy
) != cache_gen
)
9979 if (btrfs_test_opt(info
, CLEAR_CACHE
))
9983 ret
= find_first_block_group(info
, path
, &key
);
9989 leaf
= path
->nodes
[0];
9990 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
9992 cache
= btrfs_create_block_group_cache(info
, found_key
.objectid
,
10001 * When we mount with old space cache, we need to
10002 * set BTRFS_DC_CLEAR and set dirty flag.
10004 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
10005 * truncate the old free space cache inode and
10007 * b) Setting 'dirty flag' makes sure that we flush
10008 * the new space cache info onto disk.
10010 if (btrfs_test_opt(info
, SPACE_CACHE
))
10011 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
10014 read_extent_buffer(leaf
, &cache
->item
,
10015 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
10016 sizeof(cache
->item
));
10017 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
10019 ((cache
->flags
& BTRFS_BLOCK_GROUP_METADATA
) &&
10020 (cache
->flags
& BTRFS_BLOCK_GROUP_DATA
))) {
10022 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
10023 cache
->key
.objectid
);
10028 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
10029 btrfs_release_path(path
);
10032 * We need to exclude the super stripes now so that the space
10033 * info has super bytes accounted for, otherwise we'll think
10034 * we have more space than we actually do.
10036 ret
= exclude_super_stripes(info
, cache
);
10039 * We may have excluded something, so call this just in
10042 free_excluded_extents(info
, cache
);
10043 btrfs_put_block_group(cache
);
10048 * check for two cases, either we are full, and therefore
10049 * don't need to bother with the caching work since we won't
10050 * find any space, or we are empty, and we can just add all
10051 * the space in and be done with it. This saves us _alot_ of
10052 * time, particularly in the full case.
10054 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
10055 cache
->last_byte_to_unpin
= (u64
)-1;
10056 cache
->cached
= BTRFS_CACHE_FINISHED
;
10057 free_excluded_extents(info
, cache
);
10058 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
10059 cache
->last_byte_to_unpin
= (u64
)-1;
10060 cache
->cached
= BTRFS_CACHE_FINISHED
;
10061 add_new_free_space(cache
, info
,
10062 found_key
.objectid
,
10063 found_key
.objectid
+
10065 free_excluded_extents(info
, cache
);
10068 ret
= btrfs_add_block_group_cache(info
, cache
);
10070 btrfs_remove_free_space_cache(cache
);
10071 btrfs_put_block_group(cache
);
10075 trace_btrfs_add_block_group(info
, cache
, 0);
10076 update_space_info(info
, cache
->flags
, found_key
.offset
,
10077 btrfs_block_group_used(&cache
->item
),
10078 cache
->bytes_super
, &space_info
);
10080 cache
->space_info
= space_info
;
10082 __link_block_group(space_info
, cache
);
10084 set_avail_alloc_bits(info
, cache
->flags
);
10085 if (btrfs_chunk_readonly(info
, cache
->key
.objectid
)) {
10086 inc_block_group_ro(cache
, 1);
10087 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
10088 spin_lock(&info
->unused_bgs_lock
);
10089 /* Should always be true but just in case. */
10090 if (list_empty(&cache
->bg_list
)) {
10091 btrfs_get_block_group(cache
);
10092 list_add_tail(&cache
->bg_list
,
10093 &info
->unused_bgs
);
10095 spin_unlock(&info
->unused_bgs_lock
);
10099 list_for_each_entry_rcu(space_info
, &info
->space_info
, list
) {
10100 if (!(get_alloc_profile(info
, space_info
->flags
) &
10101 (BTRFS_BLOCK_GROUP_RAID10
|
10102 BTRFS_BLOCK_GROUP_RAID1
|
10103 BTRFS_BLOCK_GROUP_RAID5
|
10104 BTRFS_BLOCK_GROUP_RAID6
|
10105 BTRFS_BLOCK_GROUP_DUP
)))
10108 * avoid allocating from un-mirrored block group if there are
10109 * mirrored block groups.
10111 list_for_each_entry(cache
,
10112 &space_info
->block_groups
[BTRFS_RAID_RAID0
],
10114 inc_block_group_ro(cache
, 1);
10115 list_for_each_entry(cache
,
10116 &space_info
->block_groups
[BTRFS_RAID_SINGLE
],
10118 inc_block_group_ro(cache
, 1);
10121 init_global_block_rsv(info
);
10124 btrfs_free_path(path
);
10128 void btrfs_create_pending_block_groups(struct btrfs_trans_handle
*trans
,
10129 struct btrfs_fs_info
*fs_info
)
10131 struct btrfs_block_group_cache
*block_group
, *tmp
;
10132 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
10133 struct btrfs_block_group_item item
;
10134 struct btrfs_key key
;
10136 bool can_flush_pending_bgs
= trans
->can_flush_pending_bgs
;
10138 trans
->can_flush_pending_bgs
= false;
10139 list_for_each_entry_safe(block_group
, tmp
, &trans
->new_bgs
, bg_list
) {
10143 spin_lock(&block_group
->lock
);
10144 memcpy(&item
, &block_group
->item
, sizeof(item
));
10145 memcpy(&key
, &block_group
->key
, sizeof(key
));
10146 spin_unlock(&block_group
->lock
);
10148 ret
= btrfs_insert_item(trans
, extent_root
, &key
, &item
,
10151 btrfs_abort_transaction(trans
, ret
);
10152 ret
= btrfs_finish_chunk_alloc(trans
, fs_info
, key
.objectid
,
10155 btrfs_abort_transaction(trans
, ret
);
10156 add_block_group_free_space(trans
, fs_info
, block_group
);
10157 /* already aborted the transaction if it failed. */
10159 list_del_init(&block_group
->bg_list
);
10161 trans
->can_flush_pending_bgs
= can_flush_pending_bgs
;
10164 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
10165 struct btrfs_fs_info
*fs_info
, u64 bytes_used
,
10166 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
10169 struct btrfs_block_group_cache
*cache
;
10172 btrfs_set_log_full_commit(fs_info
, trans
);
10174 cache
= btrfs_create_block_group_cache(fs_info
, chunk_offset
, size
);
10178 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
10179 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
10180 btrfs_set_block_group_flags(&cache
->item
, type
);
10182 cache
->flags
= type
;
10183 cache
->last_byte_to_unpin
= (u64
)-1;
10184 cache
->cached
= BTRFS_CACHE_FINISHED
;
10185 cache
->needs_free_space
= 1;
10186 ret
= exclude_super_stripes(fs_info
, cache
);
10189 * We may have excluded something, so call this just in
10192 free_excluded_extents(fs_info
, cache
);
10193 btrfs_put_block_group(cache
);
10197 add_new_free_space(cache
, fs_info
, chunk_offset
, chunk_offset
+ size
);
10199 free_excluded_extents(fs_info
, cache
);
10201 #ifdef CONFIG_BTRFS_DEBUG
10202 if (btrfs_should_fragment_free_space(cache
)) {
10203 u64 new_bytes_used
= size
- bytes_used
;
10205 bytes_used
+= new_bytes_used
>> 1;
10206 fragment_free_space(cache
);
10210 * Ensure the corresponding space_info object is created and
10211 * assigned to our block group. We want our bg to be added to the rbtree
10212 * with its ->space_info set.
10214 cache
->space_info
= __find_space_info(fs_info
, cache
->flags
);
10215 if (!cache
->space_info
) {
10216 ret
= create_space_info(fs_info
, cache
->flags
,
10217 &cache
->space_info
);
10219 btrfs_remove_free_space_cache(cache
);
10220 btrfs_put_block_group(cache
);
10225 ret
= btrfs_add_block_group_cache(fs_info
, cache
);
10227 btrfs_remove_free_space_cache(cache
);
10228 btrfs_put_block_group(cache
);
10233 * Now that our block group has its ->space_info set and is inserted in
10234 * the rbtree, update the space info's counters.
10236 trace_btrfs_add_block_group(fs_info
, cache
, 1);
10237 update_space_info(fs_info
, cache
->flags
, size
, bytes_used
,
10238 cache
->bytes_super
, &cache
->space_info
);
10239 update_global_block_rsv(fs_info
);
10241 __link_block_group(cache
->space_info
, cache
);
10243 list_add_tail(&cache
->bg_list
, &trans
->new_bgs
);
10245 set_avail_alloc_bits(fs_info
, type
);
10249 static void clear_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
10251 u64 extra_flags
= chunk_to_extended(flags
) &
10252 BTRFS_EXTENDED_PROFILE_MASK
;
10254 write_seqlock(&fs_info
->profiles_lock
);
10255 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
10256 fs_info
->avail_data_alloc_bits
&= ~extra_flags
;
10257 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
10258 fs_info
->avail_metadata_alloc_bits
&= ~extra_flags
;
10259 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
10260 fs_info
->avail_system_alloc_bits
&= ~extra_flags
;
10261 write_sequnlock(&fs_info
->profiles_lock
);
10264 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
10265 struct btrfs_fs_info
*fs_info
, u64 group_start
,
10266 struct extent_map
*em
)
10268 struct btrfs_root
*root
= fs_info
->extent_root
;
10269 struct btrfs_path
*path
;
10270 struct btrfs_block_group_cache
*block_group
;
10271 struct btrfs_free_cluster
*cluster
;
10272 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
10273 struct btrfs_key key
;
10274 struct inode
*inode
;
10275 struct kobject
*kobj
= NULL
;
10279 struct btrfs_caching_control
*caching_ctl
= NULL
;
10282 block_group
= btrfs_lookup_block_group(fs_info
, group_start
);
10283 BUG_ON(!block_group
);
10284 BUG_ON(!block_group
->ro
);
10287 * Free the reserved super bytes from this block group before
10290 free_excluded_extents(fs_info
, block_group
);
10292 memcpy(&key
, &block_group
->key
, sizeof(key
));
10293 index
= get_block_group_index(block_group
);
10294 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
10295 BTRFS_BLOCK_GROUP_RAID1
|
10296 BTRFS_BLOCK_GROUP_RAID10
))
10301 /* make sure this block group isn't part of an allocation cluster */
10302 cluster
= &fs_info
->data_alloc_cluster
;
10303 spin_lock(&cluster
->refill_lock
);
10304 btrfs_return_cluster_to_free_space(block_group
, cluster
);
10305 spin_unlock(&cluster
->refill_lock
);
10308 * make sure this block group isn't part of a metadata
10309 * allocation cluster
10311 cluster
= &fs_info
->meta_alloc_cluster
;
10312 spin_lock(&cluster
->refill_lock
);
10313 btrfs_return_cluster_to_free_space(block_group
, cluster
);
10314 spin_unlock(&cluster
->refill_lock
);
10316 path
= btrfs_alloc_path();
10323 * get the inode first so any iput calls done for the io_list
10324 * aren't the final iput (no unlinks allowed now)
10326 inode
= lookup_free_space_inode(fs_info
, block_group
, path
);
10328 mutex_lock(&trans
->transaction
->cache_write_mutex
);
10330 * make sure our free spache cache IO is done before remove the
10333 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
10334 if (!list_empty(&block_group
->io_list
)) {
10335 list_del_init(&block_group
->io_list
);
10337 WARN_ON(!IS_ERR(inode
) && inode
!= block_group
->io_ctl
.inode
);
10339 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
10340 btrfs_wait_cache_io(trans
, block_group
, path
);
10341 btrfs_put_block_group(block_group
);
10342 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
10345 if (!list_empty(&block_group
->dirty_list
)) {
10346 list_del_init(&block_group
->dirty_list
);
10347 btrfs_put_block_group(block_group
);
10349 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
10350 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
10352 if (!IS_ERR(inode
)) {
10353 ret
= btrfs_orphan_add(trans
, BTRFS_I(inode
));
10355 btrfs_add_delayed_iput(inode
);
10358 clear_nlink(inode
);
10359 /* One for the block groups ref */
10360 spin_lock(&block_group
->lock
);
10361 if (block_group
->iref
) {
10362 block_group
->iref
= 0;
10363 block_group
->inode
= NULL
;
10364 spin_unlock(&block_group
->lock
);
10367 spin_unlock(&block_group
->lock
);
10369 /* One for our lookup ref */
10370 btrfs_add_delayed_iput(inode
);
10373 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
10374 key
.offset
= block_group
->key
.objectid
;
10377 ret
= btrfs_search_slot(trans
, tree_root
, &key
, path
, -1, 1);
10381 btrfs_release_path(path
);
10383 ret
= btrfs_del_item(trans
, tree_root
, path
);
10386 btrfs_release_path(path
);
10389 spin_lock(&fs_info
->block_group_cache_lock
);
10390 rb_erase(&block_group
->cache_node
,
10391 &fs_info
->block_group_cache_tree
);
10392 RB_CLEAR_NODE(&block_group
->cache_node
);
10394 if (fs_info
->first_logical_byte
== block_group
->key
.objectid
)
10395 fs_info
->first_logical_byte
= (u64
)-1;
10396 spin_unlock(&fs_info
->block_group_cache_lock
);
10398 down_write(&block_group
->space_info
->groups_sem
);
10400 * we must use list_del_init so people can check to see if they
10401 * are still on the list after taking the semaphore
10403 list_del_init(&block_group
->list
);
10404 if (list_empty(&block_group
->space_info
->block_groups
[index
])) {
10405 kobj
= block_group
->space_info
->block_group_kobjs
[index
];
10406 block_group
->space_info
->block_group_kobjs
[index
] = NULL
;
10407 clear_avail_alloc_bits(fs_info
, block_group
->flags
);
10409 up_write(&block_group
->space_info
->groups_sem
);
10415 if (block_group
->has_caching_ctl
)
10416 caching_ctl
= get_caching_control(block_group
);
10417 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
10418 wait_block_group_cache_done(block_group
);
10419 if (block_group
->has_caching_ctl
) {
10420 down_write(&fs_info
->commit_root_sem
);
10421 if (!caching_ctl
) {
10422 struct btrfs_caching_control
*ctl
;
10424 list_for_each_entry(ctl
,
10425 &fs_info
->caching_block_groups
, list
)
10426 if (ctl
->block_group
== block_group
) {
10428 refcount_inc(&caching_ctl
->count
);
10433 list_del_init(&caching_ctl
->list
);
10434 up_write(&fs_info
->commit_root_sem
);
10436 /* Once for the caching bgs list and once for us. */
10437 put_caching_control(caching_ctl
);
10438 put_caching_control(caching_ctl
);
10442 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
10443 if (!list_empty(&block_group
->dirty_list
)) {
10446 if (!list_empty(&block_group
->io_list
)) {
10449 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
10450 btrfs_remove_free_space_cache(block_group
);
10452 spin_lock(&block_group
->space_info
->lock
);
10453 list_del_init(&block_group
->ro_list
);
10455 if (btrfs_test_opt(fs_info
, ENOSPC_DEBUG
)) {
10456 WARN_ON(block_group
->space_info
->total_bytes
10457 < block_group
->key
.offset
);
10458 WARN_ON(block_group
->space_info
->bytes_readonly
10459 < block_group
->key
.offset
);
10460 WARN_ON(block_group
->space_info
->disk_total
10461 < block_group
->key
.offset
* factor
);
10463 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
10464 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
10465 block_group
->space_info
->disk_total
-= block_group
->key
.offset
* factor
;
10467 spin_unlock(&block_group
->space_info
->lock
);
10469 memcpy(&key
, &block_group
->key
, sizeof(key
));
10471 mutex_lock(&fs_info
->chunk_mutex
);
10472 if (!list_empty(&em
->list
)) {
10473 /* We're in the transaction->pending_chunks list. */
10474 free_extent_map(em
);
10476 spin_lock(&block_group
->lock
);
10477 block_group
->removed
= 1;
10479 * At this point trimming can't start on this block group, because we
10480 * removed the block group from the tree fs_info->block_group_cache_tree
10481 * so no one can't find it anymore and even if someone already got this
10482 * block group before we removed it from the rbtree, they have already
10483 * incremented block_group->trimming - if they didn't, they won't find
10484 * any free space entries because we already removed them all when we
10485 * called btrfs_remove_free_space_cache().
10487 * And we must not remove the extent map from the fs_info->mapping_tree
10488 * to prevent the same logical address range and physical device space
10489 * ranges from being reused for a new block group. This is because our
10490 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10491 * completely transactionless, so while it is trimming a range the
10492 * currently running transaction might finish and a new one start,
10493 * allowing for new block groups to be created that can reuse the same
10494 * physical device locations unless we take this special care.
10496 * There may also be an implicit trim operation if the file system
10497 * is mounted with -odiscard. The same protections must remain
10498 * in place until the extents have been discarded completely when
10499 * the transaction commit has completed.
10501 remove_em
= (atomic_read(&block_group
->trimming
) == 0);
10503 * Make sure a trimmer task always sees the em in the pinned_chunks list
10504 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10505 * before checking block_group->removed).
10509 * Our em might be in trans->transaction->pending_chunks which
10510 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10511 * and so is the fs_info->pinned_chunks list.
10513 * So at this point we must be holding the chunk_mutex to avoid
10514 * any races with chunk allocation (more specifically at
10515 * volumes.c:contains_pending_extent()), to ensure it always
10516 * sees the em, either in the pending_chunks list or in the
10517 * pinned_chunks list.
10519 list_move_tail(&em
->list
, &fs_info
->pinned_chunks
);
10521 spin_unlock(&block_group
->lock
);
10524 struct extent_map_tree
*em_tree
;
10526 em_tree
= &fs_info
->mapping_tree
.map_tree
;
10527 write_lock(&em_tree
->lock
);
10529 * The em might be in the pending_chunks list, so make sure the
10530 * chunk mutex is locked, since remove_extent_mapping() will
10531 * delete us from that list.
10533 remove_extent_mapping(em_tree
, em
);
10534 write_unlock(&em_tree
->lock
);
10535 /* once for the tree */
10536 free_extent_map(em
);
10539 mutex_unlock(&fs_info
->chunk_mutex
);
10541 ret
= remove_block_group_free_space(trans
, fs_info
, block_group
);
10545 btrfs_put_block_group(block_group
);
10546 btrfs_put_block_group(block_group
);
10548 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
10554 ret
= btrfs_del_item(trans
, root
, path
);
10556 btrfs_free_path(path
);
10560 struct btrfs_trans_handle
*
10561 btrfs_start_trans_remove_block_group(struct btrfs_fs_info
*fs_info
,
10562 const u64 chunk_offset
)
10564 struct extent_map_tree
*em_tree
= &fs_info
->mapping_tree
.map_tree
;
10565 struct extent_map
*em
;
10566 struct map_lookup
*map
;
10567 unsigned int num_items
;
10569 read_lock(&em_tree
->lock
);
10570 em
= lookup_extent_mapping(em_tree
, chunk_offset
, 1);
10571 read_unlock(&em_tree
->lock
);
10572 ASSERT(em
&& em
->start
== chunk_offset
);
10575 * We need to reserve 3 + N units from the metadata space info in order
10576 * to remove a block group (done at btrfs_remove_chunk() and at
10577 * btrfs_remove_block_group()), which are used for:
10579 * 1 unit for adding the free space inode's orphan (located in the tree
10581 * 1 unit for deleting the block group item (located in the extent
10583 * 1 unit for deleting the free space item (located in tree of tree
10585 * N units for deleting N device extent items corresponding to each
10586 * stripe (located in the device tree).
10588 * In order to remove a block group we also need to reserve units in the
10589 * system space info in order to update the chunk tree (update one or
10590 * more device items and remove one chunk item), but this is done at
10591 * btrfs_remove_chunk() through a call to check_system_chunk().
10593 map
= em
->map_lookup
;
10594 num_items
= 3 + map
->num_stripes
;
10595 free_extent_map(em
);
10597 return btrfs_start_transaction_fallback_global_rsv(fs_info
->extent_root
,
10602 * Process the unused_bgs list and remove any that don't have any allocated
10603 * space inside of them.
10605 void btrfs_delete_unused_bgs(struct btrfs_fs_info
*fs_info
)
10607 struct btrfs_block_group_cache
*block_group
;
10608 struct btrfs_space_info
*space_info
;
10609 struct btrfs_trans_handle
*trans
;
10612 if (!test_bit(BTRFS_FS_OPEN
, &fs_info
->flags
))
10615 spin_lock(&fs_info
->unused_bgs_lock
);
10616 while (!list_empty(&fs_info
->unused_bgs
)) {
10620 block_group
= list_first_entry(&fs_info
->unused_bgs
,
10621 struct btrfs_block_group_cache
,
10623 list_del_init(&block_group
->bg_list
);
10625 space_info
= block_group
->space_info
;
10627 if (ret
|| btrfs_mixed_space_info(space_info
)) {
10628 btrfs_put_block_group(block_group
);
10631 spin_unlock(&fs_info
->unused_bgs_lock
);
10633 mutex_lock(&fs_info
->delete_unused_bgs_mutex
);
10635 /* Don't want to race with allocators so take the groups_sem */
10636 down_write(&space_info
->groups_sem
);
10637 spin_lock(&block_group
->lock
);
10638 if (block_group
->reserved
||
10639 btrfs_block_group_used(&block_group
->item
) ||
10641 list_is_singular(&block_group
->list
)) {
10643 * We want to bail if we made new allocations or have
10644 * outstanding allocations in this block group. We do
10645 * the ro check in case balance is currently acting on
10646 * this block group.
10648 spin_unlock(&block_group
->lock
);
10649 up_write(&space_info
->groups_sem
);
10652 spin_unlock(&block_group
->lock
);
10654 /* We don't want to force the issue, only flip if it's ok. */
10655 ret
= inc_block_group_ro(block_group
, 0);
10656 up_write(&space_info
->groups_sem
);
10663 * Want to do this before we do anything else so we can recover
10664 * properly if we fail to join the transaction.
10666 trans
= btrfs_start_trans_remove_block_group(fs_info
,
10667 block_group
->key
.objectid
);
10668 if (IS_ERR(trans
)) {
10669 btrfs_dec_block_group_ro(block_group
);
10670 ret
= PTR_ERR(trans
);
10675 * We could have pending pinned extents for this block group,
10676 * just delete them, we don't care about them anymore.
10678 start
= block_group
->key
.objectid
;
10679 end
= start
+ block_group
->key
.offset
- 1;
10681 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10682 * btrfs_finish_extent_commit(). If we are at transaction N,
10683 * another task might be running finish_extent_commit() for the
10684 * previous transaction N - 1, and have seen a range belonging
10685 * to the block group in freed_extents[] before we were able to
10686 * clear the whole block group range from freed_extents[]. This
10687 * means that task can lookup for the block group after we
10688 * unpinned it from freed_extents[] and removed it, leading to
10689 * a BUG_ON() at btrfs_unpin_extent_range().
10691 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
10692 ret
= clear_extent_bits(&fs_info
->freed_extents
[0], start
, end
,
10695 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10696 btrfs_dec_block_group_ro(block_group
);
10699 ret
= clear_extent_bits(&fs_info
->freed_extents
[1], start
, end
,
10702 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10703 btrfs_dec_block_group_ro(block_group
);
10706 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10708 /* Reset pinned so btrfs_put_block_group doesn't complain */
10709 spin_lock(&space_info
->lock
);
10710 spin_lock(&block_group
->lock
);
10712 space_info
->bytes_pinned
-= block_group
->pinned
;
10713 space_info
->bytes_readonly
+= block_group
->pinned
;
10714 percpu_counter_add(&space_info
->total_bytes_pinned
,
10715 -block_group
->pinned
);
10716 block_group
->pinned
= 0;
10718 spin_unlock(&block_group
->lock
);
10719 spin_unlock(&space_info
->lock
);
10721 /* DISCARD can flip during remount */
10722 trimming
= btrfs_test_opt(fs_info
, DISCARD
);
10724 /* Implicit trim during transaction commit. */
10726 btrfs_get_block_group_trimming(block_group
);
10729 * Btrfs_remove_chunk will abort the transaction if things go
10732 ret
= btrfs_remove_chunk(trans
, fs_info
,
10733 block_group
->key
.objectid
);
10737 btrfs_put_block_group_trimming(block_group
);
10742 * If we're not mounted with -odiscard, we can just forget
10743 * about this block group. Otherwise we'll need to wait
10744 * until transaction commit to do the actual discard.
10747 spin_lock(&fs_info
->unused_bgs_lock
);
10749 * A concurrent scrub might have added us to the list
10750 * fs_info->unused_bgs, so use a list_move operation
10751 * to add the block group to the deleted_bgs list.
10753 list_move(&block_group
->bg_list
,
10754 &trans
->transaction
->deleted_bgs
);
10755 spin_unlock(&fs_info
->unused_bgs_lock
);
10756 btrfs_get_block_group(block_group
);
10759 btrfs_end_transaction(trans
);
10761 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
10762 btrfs_put_block_group(block_group
);
10763 spin_lock(&fs_info
->unused_bgs_lock
);
10765 spin_unlock(&fs_info
->unused_bgs_lock
);
10768 int btrfs_init_space_info(struct btrfs_fs_info
*fs_info
)
10770 struct btrfs_space_info
*space_info
;
10771 struct btrfs_super_block
*disk_super
;
10777 disk_super
= fs_info
->super_copy
;
10778 if (!btrfs_super_root(disk_super
))
10781 features
= btrfs_super_incompat_flags(disk_super
);
10782 if (features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
10785 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
10786 ret
= create_space_info(fs_info
, flags
, &space_info
);
10791 flags
= BTRFS_BLOCK_GROUP_METADATA
| BTRFS_BLOCK_GROUP_DATA
;
10792 ret
= create_space_info(fs_info
, flags
, &space_info
);
10794 flags
= BTRFS_BLOCK_GROUP_METADATA
;
10795 ret
= create_space_info(fs_info
, flags
, &space_info
);
10799 flags
= BTRFS_BLOCK_GROUP_DATA
;
10800 ret
= create_space_info(fs_info
, flags
, &space_info
);
10806 int btrfs_error_unpin_extent_range(struct btrfs_fs_info
*fs_info
,
10807 u64 start
, u64 end
)
10809 return unpin_extent_range(fs_info
, start
, end
, false);
10813 * It used to be that old block groups would be left around forever.
10814 * Iterating over them would be enough to trim unused space. Since we
10815 * now automatically remove them, we also need to iterate over unallocated
10818 * We don't want a transaction for this since the discard may take a
10819 * substantial amount of time. We don't require that a transaction be
10820 * running, but we do need to take a running transaction into account
10821 * to ensure that we're not discarding chunks that were released in
10822 * the current transaction.
10824 * Holding the chunks lock will prevent other threads from allocating
10825 * or releasing chunks, but it won't prevent a running transaction
10826 * from committing and releasing the memory that the pending chunks
10827 * list head uses. For that, we need to take a reference to the
10830 static int btrfs_trim_free_extents(struct btrfs_device
*device
,
10831 u64 minlen
, u64
*trimmed
)
10833 u64 start
= 0, len
= 0;
10838 /* Not writeable = nothing to do. */
10839 if (!device
->writeable
)
10842 /* No free space = nothing to do. */
10843 if (device
->total_bytes
<= device
->bytes_used
)
10849 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
10850 struct btrfs_transaction
*trans
;
10853 ret
= mutex_lock_interruptible(&fs_info
->chunk_mutex
);
10857 down_read(&fs_info
->commit_root_sem
);
10859 spin_lock(&fs_info
->trans_lock
);
10860 trans
= fs_info
->running_transaction
;
10862 refcount_inc(&trans
->use_count
);
10863 spin_unlock(&fs_info
->trans_lock
);
10865 ret
= find_free_dev_extent_start(trans
, device
, minlen
, start
,
10868 btrfs_put_transaction(trans
);
10871 up_read(&fs_info
->commit_root_sem
);
10872 mutex_unlock(&fs_info
->chunk_mutex
);
10873 if (ret
== -ENOSPC
)
10878 ret
= btrfs_issue_discard(device
->bdev
, start
, len
, &bytes
);
10879 up_read(&fs_info
->commit_root_sem
);
10880 mutex_unlock(&fs_info
->chunk_mutex
);
10888 if (fatal_signal_pending(current
)) {
10889 ret
= -ERESTARTSYS
;
10899 int btrfs_trim_fs(struct btrfs_fs_info
*fs_info
, struct fstrim_range
*range
)
10901 struct btrfs_block_group_cache
*cache
= NULL
;
10902 struct btrfs_device
*device
;
10903 struct list_head
*devices
;
10908 u64 total_bytes
= btrfs_super_total_bytes(fs_info
->super_copy
);
10912 * try to trim all FS space, our block group may start from non-zero.
10914 if (range
->len
== total_bytes
)
10915 cache
= btrfs_lookup_first_block_group(fs_info
, range
->start
);
10917 cache
= btrfs_lookup_block_group(fs_info
, range
->start
);
10920 if (cache
->key
.objectid
>= (range
->start
+ range
->len
)) {
10921 btrfs_put_block_group(cache
);
10925 start
= max(range
->start
, cache
->key
.objectid
);
10926 end
= min(range
->start
+ range
->len
,
10927 cache
->key
.objectid
+ cache
->key
.offset
);
10929 if (end
- start
>= range
->minlen
) {
10930 if (!block_group_cache_done(cache
)) {
10931 ret
= cache_block_group(cache
, 0);
10933 btrfs_put_block_group(cache
);
10936 ret
= wait_block_group_cache_done(cache
);
10938 btrfs_put_block_group(cache
);
10942 ret
= btrfs_trim_block_group(cache
,
10948 trimmed
+= group_trimmed
;
10950 btrfs_put_block_group(cache
);
10955 cache
= next_block_group(fs_info
, cache
);
10958 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
10959 devices
= &fs_info
->fs_devices
->alloc_list
;
10960 list_for_each_entry(device
, devices
, dev_alloc_list
) {
10961 ret
= btrfs_trim_free_extents(device
, range
->minlen
,
10966 trimmed
+= group_trimmed
;
10968 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
10970 range
->len
= trimmed
;
10975 * btrfs_{start,end}_write_no_snapshoting() are similar to
10976 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10977 * data into the page cache through nocow before the subvolume is snapshoted,
10978 * but flush the data into disk after the snapshot creation, or to prevent
10979 * operations while snapshoting is ongoing and that cause the snapshot to be
10980 * inconsistent (writes followed by expanding truncates for example).
10982 void btrfs_end_write_no_snapshoting(struct btrfs_root
*root
)
10984 percpu_counter_dec(&root
->subv_writers
->counter
);
10986 * Make sure counter is updated before we wake up waiters.
10989 if (waitqueue_active(&root
->subv_writers
->wait
))
10990 wake_up(&root
->subv_writers
->wait
);
10993 int btrfs_start_write_no_snapshoting(struct btrfs_root
*root
)
10995 if (atomic_read(&root
->will_be_snapshoted
))
10998 percpu_counter_inc(&root
->subv_writers
->counter
);
11000 * Make sure counter is updated before we check for snapshot creation.
11003 if (atomic_read(&root
->will_be_snapshoted
)) {
11004 btrfs_end_write_no_snapshoting(root
);
11010 static int wait_snapshoting_atomic_t(atomic_t
*a
)
11016 void btrfs_wait_for_snapshot_creation(struct btrfs_root
*root
)
11021 ret
= btrfs_start_write_no_snapshoting(root
);
11024 wait_on_atomic_t(&root
->will_be_snapshoted
,
11025 wait_snapshoting_atomic_t
,
11026 TASK_UNINTERRUPTIBLE
);