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>
29 #include <linux/lockdep.h>
33 #include "print-tree.h"
37 #include "free-space-cache.h"
38 #include "free-space-tree.h"
42 #include "ref-verify.h"
44 #undef SCRAMBLE_DELAYED_REFS
47 * control flags for do_chunk_alloc's force field
48 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
49 * if we really need one.
51 * CHUNK_ALLOC_LIMITED means to only try and allocate one
52 * if we have very few chunks already allocated. This is
53 * used as part of the clustering code to help make sure
54 * we have a good pool of storage to cluster in, without
55 * filling the FS with empty chunks
57 * CHUNK_ALLOC_FORCE means it must try to allocate one
61 CHUNK_ALLOC_NO_FORCE
= 0,
62 CHUNK_ALLOC_LIMITED
= 1,
63 CHUNK_ALLOC_FORCE
= 2,
66 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
67 struct btrfs_fs_info
*fs_info
,
68 struct btrfs_delayed_ref_node
*node
, u64 parent
,
69 u64 root_objectid
, u64 owner_objectid
,
70 u64 owner_offset
, int refs_to_drop
,
71 struct btrfs_delayed_extent_op
*extra_op
);
72 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
73 struct extent_buffer
*leaf
,
74 struct btrfs_extent_item
*ei
);
75 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
76 struct btrfs_fs_info
*fs_info
,
77 u64 parent
, u64 root_objectid
,
78 u64 flags
, u64 owner
, u64 offset
,
79 struct btrfs_key
*ins
, int ref_mod
);
80 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
81 struct btrfs_fs_info
*fs_info
,
82 u64 parent
, u64 root_objectid
,
83 u64 flags
, struct btrfs_disk_key
*key
,
84 int level
, struct btrfs_key
*ins
);
85 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
86 struct btrfs_fs_info
*fs_info
, u64 flags
,
88 static int find_next_key(struct btrfs_path
*path
, int level
,
89 struct btrfs_key
*key
);
90 static void dump_space_info(struct btrfs_fs_info
*fs_info
,
91 struct btrfs_space_info
*info
, u64 bytes
,
92 int dump_block_groups
);
93 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
95 static void space_info_add_new_bytes(struct btrfs_fs_info
*fs_info
,
96 struct btrfs_space_info
*space_info
,
98 static void space_info_add_old_bytes(struct btrfs_fs_info
*fs_info
,
99 struct btrfs_space_info
*space_info
,
103 block_group_cache_done(struct btrfs_block_group_cache
*cache
)
106 return cache
->cached
== BTRFS_CACHE_FINISHED
||
107 cache
->cached
== BTRFS_CACHE_ERROR
;
110 static int block_group_bits(struct btrfs_block_group_cache
*cache
, u64 bits
)
112 return (cache
->flags
& bits
) == bits
;
115 void btrfs_get_block_group(struct btrfs_block_group_cache
*cache
)
117 atomic_inc(&cache
->count
);
120 void btrfs_put_block_group(struct btrfs_block_group_cache
*cache
)
122 if (atomic_dec_and_test(&cache
->count
)) {
123 WARN_ON(cache
->pinned
> 0);
124 WARN_ON(cache
->reserved
> 0);
127 * If not empty, someone is still holding mutex of
128 * full_stripe_lock, which can only be released by caller.
129 * And it will definitely cause use-after-free when caller
130 * tries to release full stripe lock.
132 * No better way to resolve, but only to warn.
134 WARN_ON(!RB_EMPTY_ROOT(&cache
->full_stripe_locks_root
.root
));
135 kfree(cache
->free_space_ctl
);
141 * this adds the block group to the fs_info rb tree for the block group
144 static int btrfs_add_block_group_cache(struct btrfs_fs_info
*info
,
145 struct btrfs_block_group_cache
*block_group
)
148 struct rb_node
*parent
= NULL
;
149 struct btrfs_block_group_cache
*cache
;
151 spin_lock(&info
->block_group_cache_lock
);
152 p
= &info
->block_group_cache_tree
.rb_node
;
156 cache
= rb_entry(parent
, struct btrfs_block_group_cache
,
158 if (block_group
->key
.objectid
< cache
->key
.objectid
) {
160 } else if (block_group
->key
.objectid
> cache
->key
.objectid
) {
163 spin_unlock(&info
->block_group_cache_lock
);
168 rb_link_node(&block_group
->cache_node
, parent
, p
);
169 rb_insert_color(&block_group
->cache_node
,
170 &info
->block_group_cache_tree
);
172 if (info
->first_logical_byte
> block_group
->key
.objectid
)
173 info
->first_logical_byte
= block_group
->key
.objectid
;
175 spin_unlock(&info
->block_group_cache_lock
);
181 * This will return the block group at or after bytenr if contains is 0, else
182 * it will return the block group that contains the bytenr
184 static struct btrfs_block_group_cache
*
185 block_group_cache_tree_search(struct btrfs_fs_info
*info
, u64 bytenr
,
188 struct btrfs_block_group_cache
*cache
, *ret
= NULL
;
192 spin_lock(&info
->block_group_cache_lock
);
193 n
= info
->block_group_cache_tree
.rb_node
;
196 cache
= rb_entry(n
, struct btrfs_block_group_cache
,
198 end
= cache
->key
.objectid
+ cache
->key
.offset
- 1;
199 start
= cache
->key
.objectid
;
201 if (bytenr
< start
) {
202 if (!contains
&& (!ret
|| start
< ret
->key
.objectid
))
205 } else if (bytenr
> start
) {
206 if (contains
&& bytenr
<= end
) {
217 btrfs_get_block_group(ret
);
218 if (bytenr
== 0 && info
->first_logical_byte
> ret
->key
.objectid
)
219 info
->first_logical_byte
= ret
->key
.objectid
;
221 spin_unlock(&info
->block_group_cache_lock
);
226 static int add_excluded_extent(struct btrfs_fs_info
*fs_info
,
227 u64 start
, u64 num_bytes
)
229 u64 end
= start
+ num_bytes
- 1;
230 set_extent_bits(&fs_info
->freed_extents
[0],
231 start
, end
, EXTENT_UPTODATE
);
232 set_extent_bits(&fs_info
->freed_extents
[1],
233 start
, end
, EXTENT_UPTODATE
);
237 static void free_excluded_extents(struct btrfs_fs_info
*fs_info
,
238 struct btrfs_block_group_cache
*cache
)
242 start
= cache
->key
.objectid
;
243 end
= start
+ cache
->key
.offset
- 1;
245 clear_extent_bits(&fs_info
->freed_extents
[0],
246 start
, end
, EXTENT_UPTODATE
);
247 clear_extent_bits(&fs_info
->freed_extents
[1],
248 start
, end
, EXTENT_UPTODATE
);
251 static int exclude_super_stripes(struct btrfs_fs_info
*fs_info
,
252 struct btrfs_block_group_cache
*cache
)
259 if (cache
->key
.objectid
< BTRFS_SUPER_INFO_OFFSET
) {
260 stripe_len
= BTRFS_SUPER_INFO_OFFSET
- cache
->key
.objectid
;
261 cache
->bytes_super
+= stripe_len
;
262 ret
= add_excluded_extent(fs_info
, cache
->key
.objectid
,
268 for (i
= 0; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
269 bytenr
= btrfs_sb_offset(i
);
270 ret
= btrfs_rmap_block(fs_info
, cache
->key
.objectid
,
271 bytenr
, 0, &logical
, &nr
, &stripe_len
);
278 if (logical
[nr
] > cache
->key
.objectid
+
282 if (logical
[nr
] + stripe_len
<= cache
->key
.objectid
)
286 if (start
< cache
->key
.objectid
) {
287 start
= cache
->key
.objectid
;
288 len
= (logical
[nr
] + stripe_len
) - start
;
290 len
= min_t(u64
, stripe_len
,
291 cache
->key
.objectid
+
292 cache
->key
.offset
- start
);
295 cache
->bytes_super
+= len
;
296 ret
= add_excluded_extent(fs_info
, start
, len
);
308 static struct btrfs_caching_control
*
309 get_caching_control(struct btrfs_block_group_cache
*cache
)
311 struct btrfs_caching_control
*ctl
;
313 spin_lock(&cache
->lock
);
314 if (!cache
->caching_ctl
) {
315 spin_unlock(&cache
->lock
);
319 ctl
= cache
->caching_ctl
;
320 refcount_inc(&ctl
->count
);
321 spin_unlock(&cache
->lock
);
325 static void put_caching_control(struct btrfs_caching_control
*ctl
)
327 if (refcount_dec_and_test(&ctl
->count
))
331 #ifdef CONFIG_BTRFS_DEBUG
332 static void fragment_free_space(struct btrfs_block_group_cache
*block_group
)
334 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
335 u64 start
= block_group
->key
.objectid
;
336 u64 len
= block_group
->key
.offset
;
337 u64 chunk
= block_group
->flags
& BTRFS_BLOCK_GROUP_METADATA
?
338 fs_info
->nodesize
: fs_info
->sectorsize
;
339 u64 step
= chunk
<< 1;
341 while (len
> chunk
) {
342 btrfs_remove_free_space(block_group
, start
, chunk
);
353 * this is only called by cache_block_group, since we could have freed extents
354 * we need to check the pinned_extents for any extents that can't be used yet
355 * since their free space will be released as soon as the transaction commits.
357 u64
add_new_free_space(struct btrfs_block_group_cache
*block_group
,
358 struct btrfs_fs_info
*info
, u64 start
, u64 end
)
360 u64 extent_start
, extent_end
, size
, total_added
= 0;
363 while (start
< end
) {
364 ret
= find_first_extent_bit(info
->pinned_extents
, start
,
365 &extent_start
, &extent_end
,
366 EXTENT_DIRTY
| EXTENT_UPTODATE
,
371 if (extent_start
<= start
) {
372 start
= extent_end
+ 1;
373 } else if (extent_start
> start
&& extent_start
< end
) {
374 size
= extent_start
- start
;
376 ret
= btrfs_add_free_space(block_group
, start
,
378 BUG_ON(ret
); /* -ENOMEM or logic error */
379 start
= extent_end
+ 1;
388 ret
= btrfs_add_free_space(block_group
, start
, size
);
389 BUG_ON(ret
); /* -ENOMEM or logic error */
395 static int load_extent_tree_free(struct btrfs_caching_control
*caching_ctl
)
397 struct btrfs_block_group_cache
*block_group
= caching_ctl
->block_group
;
398 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
399 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
400 struct btrfs_path
*path
;
401 struct extent_buffer
*leaf
;
402 struct btrfs_key key
;
409 path
= btrfs_alloc_path();
413 last
= max_t(u64
, block_group
->key
.objectid
, BTRFS_SUPER_INFO_OFFSET
);
415 #ifdef CONFIG_BTRFS_DEBUG
417 * If we're fragmenting we don't want to make anybody think we can
418 * allocate from this block group until we've had a chance to fragment
421 if (btrfs_should_fragment_free_space(block_group
))
425 * We don't want to deadlock with somebody trying to allocate a new
426 * extent for the extent root while also trying to search the extent
427 * root to add free space. So we skip locking and search the commit
428 * root, since its read-only
430 path
->skip_locking
= 1;
431 path
->search_commit_root
= 1;
432 path
->reada
= READA_FORWARD
;
436 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
439 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
443 leaf
= path
->nodes
[0];
444 nritems
= btrfs_header_nritems(leaf
);
447 if (btrfs_fs_closing(fs_info
) > 1) {
452 if (path
->slots
[0] < nritems
) {
453 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
455 ret
= find_next_key(path
, 0, &key
);
459 if (need_resched() ||
460 rwsem_is_contended(&fs_info
->commit_root_sem
)) {
462 caching_ctl
->progress
= last
;
463 btrfs_release_path(path
);
464 up_read(&fs_info
->commit_root_sem
);
465 mutex_unlock(&caching_ctl
->mutex
);
467 mutex_lock(&caching_ctl
->mutex
);
468 down_read(&fs_info
->commit_root_sem
);
472 ret
= btrfs_next_leaf(extent_root
, path
);
477 leaf
= path
->nodes
[0];
478 nritems
= btrfs_header_nritems(leaf
);
482 if (key
.objectid
< last
) {
485 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
488 caching_ctl
->progress
= last
;
489 btrfs_release_path(path
);
493 if (key
.objectid
< block_group
->key
.objectid
) {
498 if (key
.objectid
>= block_group
->key
.objectid
+
499 block_group
->key
.offset
)
502 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
||
503 key
.type
== BTRFS_METADATA_ITEM_KEY
) {
504 total_found
+= add_new_free_space(block_group
,
507 if (key
.type
== BTRFS_METADATA_ITEM_KEY
)
508 last
= key
.objectid
+
511 last
= key
.objectid
+ key
.offset
;
513 if (total_found
> CACHING_CTL_WAKE_UP
) {
516 wake_up(&caching_ctl
->wait
);
523 total_found
+= add_new_free_space(block_group
, fs_info
, last
,
524 block_group
->key
.objectid
+
525 block_group
->key
.offset
);
526 caching_ctl
->progress
= (u64
)-1;
529 btrfs_free_path(path
);
533 static noinline
void caching_thread(struct btrfs_work
*work
)
535 struct btrfs_block_group_cache
*block_group
;
536 struct btrfs_fs_info
*fs_info
;
537 struct btrfs_caching_control
*caching_ctl
;
538 struct btrfs_root
*extent_root
;
541 caching_ctl
= container_of(work
, struct btrfs_caching_control
, work
);
542 block_group
= caching_ctl
->block_group
;
543 fs_info
= block_group
->fs_info
;
544 extent_root
= fs_info
->extent_root
;
546 mutex_lock(&caching_ctl
->mutex
);
547 down_read(&fs_info
->commit_root_sem
);
549 if (btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
))
550 ret
= load_free_space_tree(caching_ctl
);
552 ret
= load_extent_tree_free(caching_ctl
);
554 spin_lock(&block_group
->lock
);
555 block_group
->caching_ctl
= NULL
;
556 block_group
->cached
= ret
? BTRFS_CACHE_ERROR
: BTRFS_CACHE_FINISHED
;
557 spin_unlock(&block_group
->lock
);
559 #ifdef CONFIG_BTRFS_DEBUG
560 if (btrfs_should_fragment_free_space(block_group
)) {
563 spin_lock(&block_group
->space_info
->lock
);
564 spin_lock(&block_group
->lock
);
565 bytes_used
= block_group
->key
.offset
-
566 btrfs_block_group_used(&block_group
->item
);
567 block_group
->space_info
->bytes_used
+= bytes_used
>> 1;
568 spin_unlock(&block_group
->lock
);
569 spin_unlock(&block_group
->space_info
->lock
);
570 fragment_free_space(block_group
);
574 caching_ctl
->progress
= (u64
)-1;
576 up_read(&fs_info
->commit_root_sem
);
577 free_excluded_extents(fs_info
, block_group
);
578 mutex_unlock(&caching_ctl
->mutex
);
580 wake_up(&caching_ctl
->wait
);
582 put_caching_control(caching_ctl
);
583 btrfs_put_block_group(block_group
);
586 static int cache_block_group(struct btrfs_block_group_cache
*cache
,
590 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
591 struct btrfs_caching_control
*caching_ctl
;
594 caching_ctl
= kzalloc(sizeof(*caching_ctl
), GFP_NOFS
);
598 INIT_LIST_HEAD(&caching_ctl
->list
);
599 mutex_init(&caching_ctl
->mutex
);
600 init_waitqueue_head(&caching_ctl
->wait
);
601 caching_ctl
->block_group
= cache
;
602 caching_ctl
->progress
= cache
->key
.objectid
;
603 refcount_set(&caching_ctl
->count
, 1);
604 btrfs_init_work(&caching_ctl
->work
, btrfs_cache_helper
,
605 caching_thread
, NULL
, NULL
);
607 spin_lock(&cache
->lock
);
609 * This should be a rare occasion, but this could happen I think in the
610 * case where one thread starts to load the space cache info, and then
611 * some other thread starts a transaction commit which tries to do an
612 * allocation while the other thread is still loading the space cache
613 * info. The previous loop should have kept us from choosing this block
614 * group, but if we've moved to the state where we will wait on caching
615 * block groups we need to first check if we're doing a fast load here,
616 * so we can wait for it to finish, otherwise we could end up allocating
617 * from a block group who's cache gets evicted for one reason or
620 while (cache
->cached
== BTRFS_CACHE_FAST
) {
621 struct btrfs_caching_control
*ctl
;
623 ctl
= cache
->caching_ctl
;
624 refcount_inc(&ctl
->count
);
625 prepare_to_wait(&ctl
->wait
, &wait
, TASK_UNINTERRUPTIBLE
);
626 spin_unlock(&cache
->lock
);
630 finish_wait(&ctl
->wait
, &wait
);
631 put_caching_control(ctl
);
632 spin_lock(&cache
->lock
);
635 if (cache
->cached
!= BTRFS_CACHE_NO
) {
636 spin_unlock(&cache
->lock
);
640 WARN_ON(cache
->caching_ctl
);
641 cache
->caching_ctl
= caching_ctl
;
642 cache
->cached
= BTRFS_CACHE_FAST
;
643 spin_unlock(&cache
->lock
);
645 if (btrfs_test_opt(fs_info
, SPACE_CACHE
)) {
646 mutex_lock(&caching_ctl
->mutex
);
647 ret
= load_free_space_cache(fs_info
, cache
);
649 spin_lock(&cache
->lock
);
651 cache
->caching_ctl
= NULL
;
652 cache
->cached
= BTRFS_CACHE_FINISHED
;
653 cache
->last_byte_to_unpin
= (u64
)-1;
654 caching_ctl
->progress
= (u64
)-1;
656 if (load_cache_only
) {
657 cache
->caching_ctl
= NULL
;
658 cache
->cached
= BTRFS_CACHE_NO
;
660 cache
->cached
= BTRFS_CACHE_STARTED
;
661 cache
->has_caching_ctl
= 1;
664 spin_unlock(&cache
->lock
);
665 #ifdef CONFIG_BTRFS_DEBUG
667 btrfs_should_fragment_free_space(cache
)) {
670 spin_lock(&cache
->space_info
->lock
);
671 spin_lock(&cache
->lock
);
672 bytes_used
= cache
->key
.offset
-
673 btrfs_block_group_used(&cache
->item
);
674 cache
->space_info
->bytes_used
+= bytes_used
>> 1;
675 spin_unlock(&cache
->lock
);
676 spin_unlock(&cache
->space_info
->lock
);
677 fragment_free_space(cache
);
680 mutex_unlock(&caching_ctl
->mutex
);
682 wake_up(&caching_ctl
->wait
);
684 put_caching_control(caching_ctl
);
685 free_excluded_extents(fs_info
, cache
);
690 * We're either using the free space tree or no caching at all.
691 * Set cached to the appropriate value and wakeup any waiters.
693 spin_lock(&cache
->lock
);
694 if (load_cache_only
) {
695 cache
->caching_ctl
= NULL
;
696 cache
->cached
= BTRFS_CACHE_NO
;
698 cache
->cached
= BTRFS_CACHE_STARTED
;
699 cache
->has_caching_ctl
= 1;
701 spin_unlock(&cache
->lock
);
702 wake_up(&caching_ctl
->wait
);
705 if (load_cache_only
) {
706 put_caching_control(caching_ctl
);
710 down_write(&fs_info
->commit_root_sem
);
711 refcount_inc(&caching_ctl
->count
);
712 list_add_tail(&caching_ctl
->list
, &fs_info
->caching_block_groups
);
713 up_write(&fs_info
->commit_root_sem
);
715 btrfs_get_block_group(cache
);
717 btrfs_queue_work(fs_info
->caching_workers
, &caching_ctl
->work
);
723 * return the block group that starts at or after bytenr
725 static struct btrfs_block_group_cache
*
726 btrfs_lookup_first_block_group(struct btrfs_fs_info
*info
, u64 bytenr
)
728 return block_group_cache_tree_search(info
, bytenr
, 0);
732 * return the block group that contains the given bytenr
734 struct btrfs_block_group_cache
*btrfs_lookup_block_group(
735 struct btrfs_fs_info
*info
,
738 return block_group_cache_tree_search(info
, bytenr
, 1);
741 static struct btrfs_space_info
*__find_space_info(struct btrfs_fs_info
*info
,
744 struct list_head
*head
= &info
->space_info
;
745 struct btrfs_space_info
*found
;
747 flags
&= BTRFS_BLOCK_GROUP_TYPE_MASK
;
750 list_for_each_entry_rcu(found
, head
, list
) {
751 if (found
->flags
& flags
) {
760 static void add_pinned_bytes(struct btrfs_fs_info
*fs_info
, s64 num_bytes
,
761 u64 owner
, u64 root_objectid
)
763 struct btrfs_space_info
*space_info
;
766 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
767 if (root_objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
768 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
770 flags
= BTRFS_BLOCK_GROUP_METADATA
;
772 flags
= BTRFS_BLOCK_GROUP_DATA
;
775 space_info
= __find_space_info(fs_info
, flags
);
777 percpu_counter_add(&space_info
->total_bytes_pinned
, num_bytes
);
781 * after adding space to the filesystem, we need to clear the full flags
782 * on all the space infos.
784 void btrfs_clear_space_info_full(struct btrfs_fs_info
*info
)
786 struct list_head
*head
= &info
->space_info
;
787 struct btrfs_space_info
*found
;
790 list_for_each_entry_rcu(found
, head
, list
)
795 /* simple helper to search for an existing data extent at a given offset */
796 int btrfs_lookup_data_extent(struct btrfs_fs_info
*fs_info
, u64 start
, u64 len
)
799 struct btrfs_key key
;
800 struct btrfs_path
*path
;
802 path
= btrfs_alloc_path();
806 key
.objectid
= start
;
808 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
809 ret
= btrfs_search_slot(NULL
, fs_info
->extent_root
, &key
, path
, 0, 0);
810 btrfs_free_path(path
);
815 * helper function to lookup reference count and flags of a tree block.
817 * the head node for delayed ref is used to store the sum of all the
818 * reference count modifications queued up in the rbtree. the head
819 * node may also store the extent flags to set. This way you can check
820 * to see what the reference count and extent flags would be if all of
821 * the delayed refs are not processed.
823 int btrfs_lookup_extent_info(struct btrfs_trans_handle
*trans
,
824 struct btrfs_fs_info
*fs_info
, u64 bytenr
,
825 u64 offset
, int metadata
, u64
*refs
, u64
*flags
)
827 struct btrfs_delayed_ref_head
*head
;
828 struct btrfs_delayed_ref_root
*delayed_refs
;
829 struct btrfs_path
*path
;
830 struct btrfs_extent_item
*ei
;
831 struct extent_buffer
*leaf
;
832 struct btrfs_key key
;
839 * If we don't have skinny metadata, don't bother doing anything
842 if (metadata
&& !btrfs_fs_incompat(fs_info
, SKINNY_METADATA
)) {
843 offset
= fs_info
->nodesize
;
847 path
= btrfs_alloc_path();
852 path
->skip_locking
= 1;
853 path
->search_commit_root
= 1;
857 key
.objectid
= bytenr
;
860 key
.type
= BTRFS_METADATA_ITEM_KEY
;
862 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
864 ret
= btrfs_search_slot(trans
, fs_info
->extent_root
, &key
, path
, 0, 0);
868 if (ret
> 0 && metadata
&& key
.type
== BTRFS_METADATA_ITEM_KEY
) {
869 if (path
->slots
[0]) {
871 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
873 if (key
.objectid
== bytenr
&&
874 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
875 key
.offset
== fs_info
->nodesize
)
881 leaf
= path
->nodes
[0];
882 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
883 if (item_size
>= sizeof(*ei
)) {
884 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
885 struct btrfs_extent_item
);
886 num_refs
= btrfs_extent_refs(leaf
, ei
);
887 extent_flags
= btrfs_extent_flags(leaf
, ei
);
889 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
890 struct btrfs_extent_item_v0
*ei0
;
891 BUG_ON(item_size
!= sizeof(*ei0
));
892 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
893 struct btrfs_extent_item_v0
);
894 num_refs
= btrfs_extent_refs_v0(leaf
, ei0
);
895 /* FIXME: this isn't correct for data */
896 extent_flags
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
901 BUG_ON(num_refs
== 0);
911 delayed_refs
= &trans
->transaction
->delayed_refs
;
912 spin_lock(&delayed_refs
->lock
);
913 head
= btrfs_find_delayed_ref_head(delayed_refs
, bytenr
);
915 if (!mutex_trylock(&head
->mutex
)) {
916 refcount_inc(&head
->refs
);
917 spin_unlock(&delayed_refs
->lock
);
919 btrfs_release_path(path
);
922 * Mutex was contended, block until it's released and try
925 mutex_lock(&head
->mutex
);
926 mutex_unlock(&head
->mutex
);
927 btrfs_put_delayed_ref_head(head
);
930 spin_lock(&head
->lock
);
931 if (head
->extent_op
&& head
->extent_op
->update_flags
)
932 extent_flags
|= head
->extent_op
->flags_to_set
;
934 BUG_ON(num_refs
== 0);
936 num_refs
+= head
->ref_mod
;
937 spin_unlock(&head
->lock
);
938 mutex_unlock(&head
->mutex
);
940 spin_unlock(&delayed_refs
->lock
);
942 WARN_ON(num_refs
== 0);
946 *flags
= extent_flags
;
948 btrfs_free_path(path
);
953 * Back reference rules. Back refs have three main goals:
955 * 1) differentiate between all holders of references to an extent so that
956 * when a reference is dropped we can make sure it was a valid reference
957 * before freeing the extent.
959 * 2) Provide enough information to quickly find the holders of an extent
960 * if we notice a given block is corrupted or bad.
962 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
963 * maintenance. This is actually the same as #2, but with a slightly
964 * different use case.
966 * There are two kinds of back refs. The implicit back refs is optimized
967 * for pointers in non-shared tree blocks. For a given pointer in a block,
968 * back refs of this kind provide information about the block's owner tree
969 * and the pointer's key. These information allow us to find the block by
970 * b-tree searching. The full back refs is for pointers in tree blocks not
971 * referenced by their owner trees. The location of tree block is recorded
972 * in the back refs. Actually the full back refs is generic, and can be
973 * used in all cases the implicit back refs is used. The major shortcoming
974 * of the full back refs is its overhead. Every time a tree block gets
975 * COWed, we have to update back refs entry for all pointers in it.
977 * For a newly allocated tree block, we use implicit back refs for
978 * pointers in it. This means most tree related operations only involve
979 * implicit back refs. For a tree block created in old transaction, the
980 * only way to drop a reference to it is COW it. So we can detect the
981 * event that tree block loses its owner tree's reference and do the
982 * back refs conversion.
984 * When a tree block is COWed through a tree, there are four cases:
986 * The reference count of the block is one and the tree is the block's
987 * owner tree. Nothing to do in this case.
989 * The reference count of the block is one and the tree is not the
990 * block's owner tree. In this case, full back refs is used for pointers
991 * in the block. Remove these full back refs, add implicit back refs for
992 * every pointers in the new block.
994 * The reference count of the block is greater than one and the tree is
995 * the block's owner tree. In this case, implicit back refs is used for
996 * pointers in the block. Add full back refs for every pointers in the
997 * block, increase lower level extents' reference counts. The original
998 * implicit back refs are entailed to the new block.
1000 * The reference count of the block is greater than one and the tree is
1001 * not the block's owner tree. Add implicit back refs for every pointer in
1002 * the new block, increase lower level extents' reference count.
1004 * Back Reference Key composing:
1006 * The key objectid corresponds to the first byte in the extent,
1007 * The key type is used to differentiate between types of back refs.
1008 * There are different meanings of the key offset for different types
1011 * File extents can be referenced by:
1013 * - multiple snapshots, subvolumes, or different generations in one subvol
1014 * - different files inside a single subvolume
1015 * - different offsets inside a file (bookend extents in file.c)
1017 * The extent ref structure for the implicit back refs has fields for:
1019 * - Objectid of the subvolume root
1020 * - objectid of the file holding the reference
1021 * - original offset in the file
1022 * - how many bookend extents
1024 * The key offset for the implicit back refs is hash of the first
1027 * The extent ref structure for the full back refs has field for:
1029 * - number of pointers in the tree leaf
1031 * The key offset for the implicit back refs is the first byte of
1034 * When a file extent is allocated, The implicit back refs is used.
1035 * the fields are filled in:
1037 * (root_key.objectid, inode objectid, offset in file, 1)
1039 * When a file extent is removed file truncation, we find the
1040 * corresponding implicit back refs and check the following fields:
1042 * (btrfs_header_owner(leaf), inode objectid, offset in file)
1044 * Btree extents can be referenced by:
1046 * - Different subvolumes
1048 * Both the implicit back refs and the full back refs for tree blocks
1049 * only consist of key. The key offset for the implicit back refs is
1050 * objectid of block's owner tree. The key offset for the full back refs
1051 * is the first byte of parent block.
1053 * When implicit back refs is used, information about the lowest key and
1054 * level of the tree block are required. These information are stored in
1055 * tree block info structure.
1058 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1059 static int convert_extent_item_v0(struct btrfs_trans_handle
*trans
,
1060 struct btrfs_fs_info
*fs_info
,
1061 struct btrfs_path
*path
,
1062 u64 owner
, u32 extra_size
)
1064 struct btrfs_root
*root
= fs_info
->extent_root
;
1065 struct btrfs_extent_item
*item
;
1066 struct btrfs_extent_item_v0
*ei0
;
1067 struct btrfs_extent_ref_v0
*ref0
;
1068 struct btrfs_tree_block_info
*bi
;
1069 struct extent_buffer
*leaf
;
1070 struct btrfs_key key
;
1071 struct btrfs_key found_key
;
1072 u32 new_size
= sizeof(*item
);
1076 leaf
= path
->nodes
[0];
1077 BUG_ON(btrfs_item_size_nr(leaf
, path
->slots
[0]) != sizeof(*ei0
));
1079 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1080 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1081 struct btrfs_extent_item_v0
);
1082 refs
= btrfs_extent_refs_v0(leaf
, ei0
);
1084 if (owner
== (u64
)-1) {
1086 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1087 ret
= btrfs_next_leaf(root
, path
);
1090 BUG_ON(ret
> 0); /* Corruption */
1091 leaf
= path
->nodes
[0];
1093 btrfs_item_key_to_cpu(leaf
, &found_key
,
1095 BUG_ON(key
.objectid
!= found_key
.objectid
);
1096 if (found_key
.type
!= BTRFS_EXTENT_REF_V0_KEY
) {
1100 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1101 struct btrfs_extent_ref_v0
);
1102 owner
= btrfs_ref_objectid_v0(leaf
, ref0
);
1106 btrfs_release_path(path
);
1108 if (owner
< BTRFS_FIRST_FREE_OBJECTID
)
1109 new_size
+= sizeof(*bi
);
1111 new_size
-= sizeof(*ei0
);
1112 ret
= btrfs_search_slot(trans
, root
, &key
, path
,
1113 new_size
+ extra_size
, 1);
1116 BUG_ON(ret
); /* Corruption */
1118 btrfs_extend_item(fs_info
, path
, new_size
);
1120 leaf
= path
->nodes
[0];
1121 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1122 btrfs_set_extent_refs(leaf
, item
, refs
);
1123 /* FIXME: get real generation */
1124 btrfs_set_extent_generation(leaf
, item
, 0);
1125 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1126 btrfs_set_extent_flags(leaf
, item
,
1127 BTRFS_EXTENT_FLAG_TREE_BLOCK
|
1128 BTRFS_BLOCK_FLAG_FULL_BACKREF
);
1129 bi
= (struct btrfs_tree_block_info
*)(item
+ 1);
1130 /* FIXME: get first key of the block */
1131 memzero_extent_buffer(leaf
, (unsigned long)bi
, sizeof(*bi
));
1132 btrfs_set_tree_block_level(leaf
, bi
, (int)owner
);
1134 btrfs_set_extent_flags(leaf
, item
, BTRFS_EXTENT_FLAG_DATA
);
1136 btrfs_mark_buffer_dirty(leaf
);
1142 * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required,
1143 * is_data == BTRFS_REF_TYPE_DATA, data type is requried,
1144 * is_data == BTRFS_REF_TYPE_ANY, either type is OK.
1146 int btrfs_get_extent_inline_ref_type(const struct extent_buffer
*eb
,
1147 struct btrfs_extent_inline_ref
*iref
,
1148 enum btrfs_inline_ref_type is_data
)
1150 int type
= btrfs_extent_inline_ref_type(eb
, iref
);
1151 u64 offset
= btrfs_extent_inline_ref_offset(eb
, iref
);
1153 if (type
== BTRFS_TREE_BLOCK_REF_KEY
||
1154 type
== BTRFS_SHARED_BLOCK_REF_KEY
||
1155 type
== BTRFS_SHARED_DATA_REF_KEY
||
1156 type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1157 if (is_data
== BTRFS_REF_TYPE_BLOCK
) {
1158 if (type
== BTRFS_TREE_BLOCK_REF_KEY
)
1160 if (type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
1161 ASSERT(eb
->fs_info
);
1163 * Every shared one has parent tree
1164 * block, which must be aligned to
1168 IS_ALIGNED(offset
, eb
->fs_info
->nodesize
))
1171 } else if (is_data
== BTRFS_REF_TYPE_DATA
) {
1172 if (type
== BTRFS_EXTENT_DATA_REF_KEY
)
1174 if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1175 ASSERT(eb
->fs_info
);
1177 * Every shared one has parent tree
1178 * block, which must be aligned to
1182 IS_ALIGNED(offset
, eb
->fs_info
->nodesize
))
1186 ASSERT(is_data
== BTRFS_REF_TYPE_ANY
);
1191 btrfs_print_leaf((struct extent_buffer
*)eb
);
1192 btrfs_err(eb
->fs_info
, "eb %llu invalid extent inline ref type %d",
1196 return BTRFS_REF_TYPE_INVALID
;
1199 static u64
hash_extent_data_ref(u64 root_objectid
, u64 owner
, u64 offset
)
1201 u32 high_crc
= ~(u32
)0;
1202 u32 low_crc
= ~(u32
)0;
1205 lenum
= cpu_to_le64(root_objectid
);
1206 high_crc
= btrfs_crc32c(high_crc
, &lenum
, sizeof(lenum
));
1207 lenum
= cpu_to_le64(owner
);
1208 low_crc
= btrfs_crc32c(low_crc
, &lenum
, sizeof(lenum
));
1209 lenum
= cpu_to_le64(offset
);
1210 low_crc
= btrfs_crc32c(low_crc
, &lenum
, sizeof(lenum
));
1212 return ((u64
)high_crc
<< 31) ^ (u64
)low_crc
;
1215 static u64
hash_extent_data_ref_item(struct extent_buffer
*leaf
,
1216 struct btrfs_extent_data_ref
*ref
)
1218 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf
, ref
),
1219 btrfs_extent_data_ref_objectid(leaf
, ref
),
1220 btrfs_extent_data_ref_offset(leaf
, ref
));
1223 static int match_extent_data_ref(struct extent_buffer
*leaf
,
1224 struct btrfs_extent_data_ref
*ref
,
1225 u64 root_objectid
, u64 owner
, u64 offset
)
1227 if (btrfs_extent_data_ref_root(leaf
, ref
) != root_objectid
||
1228 btrfs_extent_data_ref_objectid(leaf
, ref
) != owner
||
1229 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
1234 static noinline
int lookup_extent_data_ref(struct btrfs_trans_handle
*trans
,
1235 struct btrfs_fs_info
*fs_info
,
1236 struct btrfs_path
*path
,
1237 u64 bytenr
, u64 parent
,
1239 u64 owner
, u64 offset
)
1241 struct btrfs_root
*root
= fs_info
->extent_root
;
1242 struct btrfs_key key
;
1243 struct btrfs_extent_data_ref
*ref
;
1244 struct extent_buffer
*leaf
;
1250 key
.objectid
= bytenr
;
1252 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1253 key
.offset
= parent
;
1255 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1256 key
.offset
= hash_extent_data_ref(root_objectid
,
1261 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1270 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1271 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1272 btrfs_release_path(path
);
1273 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1284 leaf
= path
->nodes
[0];
1285 nritems
= btrfs_header_nritems(leaf
);
1287 if (path
->slots
[0] >= nritems
) {
1288 ret
= btrfs_next_leaf(root
, path
);
1294 leaf
= path
->nodes
[0];
1295 nritems
= btrfs_header_nritems(leaf
);
1299 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1300 if (key
.objectid
!= bytenr
||
1301 key
.type
!= BTRFS_EXTENT_DATA_REF_KEY
)
1304 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1305 struct btrfs_extent_data_ref
);
1307 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1310 btrfs_release_path(path
);
1322 static noinline
int insert_extent_data_ref(struct btrfs_trans_handle
*trans
,
1323 struct btrfs_fs_info
*fs_info
,
1324 struct btrfs_path
*path
,
1325 u64 bytenr
, u64 parent
,
1326 u64 root_objectid
, u64 owner
,
1327 u64 offset
, int refs_to_add
)
1329 struct btrfs_root
*root
= fs_info
->extent_root
;
1330 struct btrfs_key key
;
1331 struct extent_buffer
*leaf
;
1336 key
.objectid
= bytenr
;
1338 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1339 key
.offset
= parent
;
1340 size
= sizeof(struct btrfs_shared_data_ref
);
1342 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1343 key
.offset
= hash_extent_data_ref(root_objectid
,
1345 size
= sizeof(struct btrfs_extent_data_ref
);
1348 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, size
);
1349 if (ret
&& ret
!= -EEXIST
)
1352 leaf
= path
->nodes
[0];
1354 struct btrfs_shared_data_ref
*ref
;
1355 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1356 struct btrfs_shared_data_ref
);
1358 btrfs_set_shared_data_ref_count(leaf
, ref
, refs_to_add
);
1360 num_refs
= btrfs_shared_data_ref_count(leaf
, ref
);
1361 num_refs
+= refs_to_add
;
1362 btrfs_set_shared_data_ref_count(leaf
, ref
, num_refs
);
1365 struct btrfs_extent_data_ref
*ref
;
1366 while (ret
== -EEXIST
) {
1367 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1368 struct btrfs_extent_data_ref
);
1369 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1372 btrfs_release_path(path
);
1374 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1376 if (ret
&& ret
!= -EEXIST
)
1379 leaf
= path
->nodes
[0];
1381 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1382 struct btrfs_extent_data_ref
);
1384 btrfs_set_extent_data_ref_root(leaf
, ref
,
1386 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
1387 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
1388 btrfs_set_extent_data_ref_count(leaf
, ref
, refs_to_add
);
1390 num_refs
= btrfs_extent_data_ref_count(leaf
, ref
);
1391 num_refs
+= refs_to_add
;
1392 btrfs_set_extent_data_ref_count(leaf
, ref
, num_refs
);
1395 btrfs_mark_buffer_dirty(leaf
);
1398 btrfs_release_path(path
);
1402 static noinline
int remove_extent_data_ref(struct btrfs_trans_handle
*trans
,
1403 struct btrfs_fs_info
*fs_info
,
1404 struct btrfs_path
*path
,
1405 int refs_to_drop
, int *last_ref
)
1407 struct btrfs_key key
;
1408 struct btrfs_extent_data_ref
*ref1
= NULL
;
1409 struct btrfs_shared_data_ref
*ref2
= NULL
;
1410 struct extent_buffer
*leaf
;
1414 leaf
= path
->nodes
[0];
1415 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1417 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1418 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1419 struct btrfs_extent_data_ref
);
1420 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1421 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1422 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1423 struct btrfs_shared_data_ref
);
1424 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1425 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1426 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1427 struct btrfs_extent_ref_v0
*ref0
;
1428 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1429 struct btrfs_extent_ref_v0
);
1430 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1436 BUG_ON(num_refs
< refs_to_drop
);
1437 num_refs
-= refs_to_drop
;
1439 if (num_refs
== 0) {
1440 ret
= btrfs_del_item(trans
, fs_info
->extent_root
, path
);
1443 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
)
1444 btrfs_set_extent_data_ref_count(leaf
, ref1
, num_refs
);
1445 else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
)
1446 btrfs_set_shared_data_ref_count(leaf
, ref2
, num_refs
);
1447 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1449 struct btrfs_extent_ref_v0
*ref0
;
1450 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1451 struct btrfs_extent_ref_v0
);
1452 btrfs_set_ref_count_v0(leaf
, ref0
, num_refs
);
1455 btrfs_mark_buffer_dirty(leaf
);
1460 static noinline u32
extent_data_ref_count(struct btrfs_path
*path
,
1461 struct btrfs_extent_inline_ref
*iref
)
1463 struct btrfs_key key
;
1464 struct extent_buffer
*leaf
;
1465 struct btrfs_extent_data_ref
*ref1
;
1466 struct btrfs_shared_data_ref
*ref2
;
1470 leaf
= path
->nodes
[0];
1471 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1474 * If type is invalid, we should have bailed out earlier than
1477 type
= btrfs_get_extent_inline_ref_type(leaf
, iref
, BTRFS_REF_TYPE_DATA
);
1478 ASSERT(type
!= BTRFS_REF_TYPE_INVALID
);
1479 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1480 ref1
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1481 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1483 ref2
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1484 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1486 } else if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1487 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1488 struct btrfs_extent_data_ref
);
1489 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1490 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1491 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1492 struct btrfs_shared_data_ref
);
1493 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1494 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1495 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1496 struct btrfs_extent_ref_v0
*ref0
;
1497 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1498 struct btrfs_extent_ref_v0
);
1499 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1507 static noinline
int lookup_tree_block_ref(struct btrfs_trans_handle
*trans
,
1508 struct btrfs_fs_info
*fs_info
,
1509 struct btrfs_path
*path
,
1510 u64 bytenr
, u64 parent
,
1513 struct btrfs_root
*root
= fs_info
->extent_root
;
1514 struct btrfs_key key
;
1517 key
.objectid
= bytenr
;
1519 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1520 key
.offset
= parent
;
1522 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1523 key
.offset
= root_objectid
;
1526 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1529 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1530 if (ret
== -ENOENT
&& parent
) {
1531 btrfs_release_path(path
);
1532 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1533 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1541 static noinline
int insert_tree_block_ref(struct btrfs_trans_handle
*trans
,
1542 struct btrfs_fs_info
*fs_info
,
1543 struct btrfs_path
*path
,
1544 u64 bytenr
, u64 parent
,
1547 struct btrfs_key key
;
1550 key
.objectid
= bytenr
;
1552 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1553 key
.offset
= parent
;
1555 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1556 key
.offset
= root_objectid
;
1559 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
,
1561 btrfs_release_path(path
);
1565 static inline int extent_ref_type(u64 parent
, u64 owner
)
1568 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1570 type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1572 type
= BTRFS_TREE_BLOCK_REF_KEY
;
1575 type
= BTRFS_SHARED_DATA_REF_KEY
;
1577 type
= BTRFS_EXTENT_DATA_REF_KEY
;
1582 static int find_next_key(struct btrfs_path
*path
, int level
,
1583 struct btrfs_key
*key
)
1586 for (; level
< BTRFS_MAX_LEVEL
; level
++) {
1587 if (!path
->nodes
[level
])
1589 if (path
->slots
[level
] + 1 >=
1590 btrfs_header_nritems(path
->nodes
[level
]))
1593 btrfs_item_key_to_cpu(path
->nodes
[level
], key
,
1594 path
->slots
[level
] + 1);
1596 btrfs_node_key_to_cpu(path
->nodes
[level
], key
,
1597 path
->slots
[level
] + 1);
1604 * look for inline back ref. if back ref is found, *ref_ret is set
1605 * to the address of inline back ref, and 0 is returned.
1607 * if back ref isn't found, *ref_ret is set to the address where it
1608 * should be inserted, and -ENOENT is returned.
1610 * if insert is true and there are too many inline back refs, the path
1611 * points to the extent item, and -EAGAIN is returned.
1613 * NOTE: inline back refs are ordered in the same way that back ref
1614 * items in the tree are ordered.
1616 static noinline_for_stack
1617 int lookup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1618 struct btrfs_fs_info
*fs_info
,
1619 struct btrfs_path
*path
,
1620 struct btrfs_extent_inline_ref
**ref_ret
,
1621 u64 bytenr
, u64 num_bytes
,
1622 u64 parent
, u64 root_objectid
,
1623 u64 owner
, u64 offset
, int insert
)
1625 struct btrfs_root
*root
= fs_info
->extent_root
;
1626 struct btrfs_key key
;
1627 struct extent_buffer
*leaf
;
1628 struct btrfs_extent_item
*ei
;
1629 struct btrfs_extent_inline_ref
*iref
;
1639 bool skinny_metadata
= btrfs_fs_incompat(fs_info
, SKINNY_METADATA
);
1642 key
.objectid
= bytenr
;
1643 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1644 key
.offset
= num_bytes
;
1646 want
= extent_ref_type(parent
, owner
);
1648 extra_size
= btrfs_extent_inline_ref_size(want
);
1649 path
->keep_locks
= 1;
1654 * Owner is our parent level, so we can just add one to get the level
1655 * for the block we are interested in.
1657 if (skinny_metadata
&& owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1658 key
.type
= BTRFS_METADATA_ITEM_KEY
;
1663 ret
= btrfs_search_slot(trans
, root
, &key
, path
, extra_size
, 1);
1670 * We may be a newly converted file system which still has the old fat
1671 * extent entries for metadata, so try and see if we have one of those.
1673 if (ret
> 0 && skinny_metadata
) {
1674 skinny_metadata
= false;
1675 if (path
->slots
[0]) {
1677 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
1679 if (key
.objectid
== bytenr
&&
1680 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
1681 key
.offset
== num_bytes
)
1685 key
.objectid
= bytenr
;
1686 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1687 key
.offset
= num_bytes
;
1688 btrfs_release_path(path
);
1693 if (ret
&& !insert
) {
1696 } else if (WARN_ON(ret
)) {
1701 leaf
= path
->nodes
[0];
1702 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1703 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1704 if (item_size
< sizeof(*ei
)) {
1709 ret
= convert_extent_item_v0(trans
, fs_info
, path
, owner
,
1715 leaf
= path
->nodes
[0];
1716 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1719 BUG_ON(item_size
< sizeof(*ei
));
1721 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1722 flags
= btrfs_extent_flags(leaf
, ei
);
1724 ptr
= (unsigned long)(ei
+ 1);
1725 end
= (unsigned long)ei
+ item_size
;
1727 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
&& !skinny_metadata
) {
1728 ptr
+= sizeof(struct btrfs_tree_block_info
);
1732 if (owner
>= BTRFS_FIRST_FREE_OBJECTID
)
1733 needed
= BTRFS_REF_TYPE_DATA
;
1735 needed
= BTRFS_REF_TYPE_BLOCK
;
1743 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1744 type
= btrfs_get_extent_inline_ref_type(leaf
, iref
, needed
);
1745 if (type
== BTRFS_REF_TYPE_INVALID
) {
1753 ptr
+= btrfs_extent_inline_ref_size(type
);
1757 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1758 struct btrfs_extent_data_ref
*dref
;
1759 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1760 if (match_extent_data_ref(leaf
, dref
, root_objectid
,
1765 if (hash_extent_data_ref_item(leaf
, dref
) <
1766 hash_extent_data_ref(root_objectid
, owner
, offset
))
1770 ref_offset
= btrfs_extent_inline_ref_offset(leaf
, iref
);
1772 if (parent
== ref_offset
) {
1776 if (ref_offset
< parent
)
1779 if (root_objectid
== ref_offset
) {
1783 if (ref_offset
< root_objectid
)
1787 ptr
+= btrfs_extent_inline_ref_size(type
);
1789 if (err
== -ENOENT
&& insert
) {
1790 if (item_size
+ extra_size
>=
1791 BTRFS_MAX_EXTENT_ITEM_SIZE(root
)) {
1796 * To add new inline back ref, we have to make sure
1797 * there is no corresponding back ref item.
1798 * For simplicity, we just do not add new inline back
1799 * ref if there is any kind of item for this block
1801 if (find_next_key(path
, 0, &key
) == 0 &&
1802 key
.objectid
== bytenr
&&
1803 key
.type
< BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1808 *ref_ret
= (struct btrfs_extent_inline_ref
*)ptr
;
1811 path
->keep_locks
= 0;
1812 btrfs_unlock_up_safe(path
, 1);
1818 * helper to add new inline back ref
1820 static noinline_for_stack
1821 void setup_inline_extent_backref(struct btrfs_fs_info
*fs_info
,
1822 struct btrfs_path
*path
,
1823 struct btrfs_extent_inline_ref
*iref
,
1824 u64 parent
, u64 root_objectid
,
1825 u64 owner
, u64 offset
, int refs_to_add
,
1826 struct btrfs_delayed_extent_op
*extent_op
)
1828 struct extent_buffer
*leaf
;
1829 struct btrfs_extent_item
*ei
;
1832 unsigned long item_offset
;
1837 leaf
= path
->nodes
[0];
1838 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1839 item_offset
= (unsigned long)iref
- (unsigned long)ei
;
1841 type
= extent_ref_type(parent
, owner
);
1842 size
= btrfs_extent_inline_ref_size(type
);
1844 btrfs_extend_item(fs_info
, path
, size
);
1846 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1847 refs
= btrfs_extent_refs(leaf
, ei
);
1848 refs
+= refs_to_add
;
1849 btrfs_set_extent_refs(leaf
, ei
, refs
);
1851 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1853 ptr
= (unsigned long)ei
+ item_offset
;
1854 end
= (unsigned long)ei
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
1855 if (ptr
< end
- size
)
1856 memmove_extent_buffer(leaf
, ptr
+ size
, ptr
,
1859 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1860 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
1861 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1862 struct btrfs_extent_data_ref
*dref
;
1863 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1864 btrfs_set_extent_data_ref_root(leaf
, dref
, root_objectid
);
1865 btrfs_set_extent_data_ref_objectid(leaf
, dref
, owner
);
1866 btrfs_set_extent_data_ref_offset(leaf
, dref
, offset
);
1867 btrfs_set_extent_data_ref_count(leaf
, dref
, refs_to_add
);
1868 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1869 struct btrfs_shared_data_ref
*sref
;
1870 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1871 btrfs_set_shared_data_ref_count(leaf
, sref
, refs_to_add
);
1872 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1873 } else if (type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
1874 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1876 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
1878 btrfs_mark_buffer_dirty(leaf
);
1881 static int lookup_extent_backref(struct btrfs_trans_handle
*trans
,
1882 struct btrfs_fs_info
*fs_info
,
1883 struct btrfs_path
*path
,
1884 struct btrfs_extent_inline_ref
**ref_ret
,
1885 u64 bytenr
, u64 num_bytes
, u64 parent
,
1886 u64 root_objectid
, u64 owner
, u64 offset
)
1890 ret
= lookup_inline_extent_backref(trans
, fs_info
, path
, ref_ret
,
1891 bytenr
, num_bytes
, parent
,
1892 root_objectid
, owner
, offset
, 0);
1896 btrfs_release_path(path
);
1899 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1900 ret
= lookup_tree_block_ref(trans
, fs_info
, path
, bytenr
,
1901 parent
, root_objectid
);
1903 ret
= lookup_extent_data_ref(trans
, fs_info
, path
, bytenr
,
1904 parent
, root_objectid
, owner
,
1911 * helper to update/remove inline back ref
1913 static noinline_for_stack
1914 void update_inline_extent_backref(struct btrfs_fs_info
*fs_info
,
1915 struct btrfs_path
*path
,
1916 struct btrfs_extent_inline_ref
*iref
,
1918 struct btrfs_delayed_extent_op
*extent_op
,
1921 struct extent_buffer
*leaf
;
1922 struct btrfs_extent_item
*ei
;
1923 struct btrfs_extent_data_ref
*dref
= NULL
;
1924 struct btrfs_shared_data_ref
*sref
= NULL
;
1932 leaf
= path
->nodes
[0];
1933 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1934 refs
= btrfs_extent_refs(leaf
, ei
);
1935 WARN_ON(refs_to_mod
< 0 && refs
+ refs_to_mod
<= 0);
1936 refs
+= refs_to_mod
;
1937 btrfs_set_extent_refs(leaf
, ei
, refs
);
1939 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1942 * If type is invalid, we should have bailed out after
1943 * lookup_inline_extent_backref().
1945 type
= btrfs_get_extent_inline_ref_type(leaf
, iref
, BTRFS_REF_TYPE_ANY
);
1946 ASSERT(type
!= BTRFS_REF_TYPE_INVALID
);
1948 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1949 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1950 refs
= btrfs_extent_data_ref_count(leaf
, dref
);
1951 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1952 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1953 refs
= btrfs_shared_data_ref_count(leaf
, sref
);
1956 BUG_ON(refs_to_mod
!= -1);
1959 BUG_ON(refs_to_mod
< 0 && refs
< -refs_to_mod
);
1960 refs
+= refs_to_mod
;
1963 if (type
== BTRFS_EXTENT_DATA_REF_KEY
)
1964 btrfs_set_extent_data_ref_count(leaf
, dref
, refs
);
1966 btrfs_set_shared_data_ref_count(leaf
, sref
, refs
);
1969 size
= btrfs_extent_inline_ref_size(type
);
1970 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1971 ptr
= (unsigned long)iref
;
1972 end
= (unsigned long)ei
+ item_size
;
1973 if (ptr
+ size
< end
)
1974 memmove_extent_buffer(leaf
, ptr
, ptr
+ size
,
1977 btrfs_truncate_item(fs_info
, path
, item_size
, 1);
1979 btrfs_mark_buffer_dirty(leaf
);
1982 static noinline_for_stack
1983 int insert_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1984 struct btrfs_fs_info
*fs_info
,
1985 struct btrfs_path
*path
,
1986 u64 bytenr
, u64 num_bytes
, u64 parent
,
1987 u64 root_objectid
, u64 owner
,
1988 u64 offset
, int refs_to_add
,
1989 struct btrfs_delayed_extent_op
*extent_op
)
1991 struct btrfs_extent_inline_ref
*iref
;
1994 ret
= lookup_inline_extent_backref(trans
, fs_info
, path
, &iref
,
1995 bytenr
, num_bytes
, parent
,
1996 root_objectid
, owner
, offset
, 1);
1998 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
);
1999 update_inline_extent_backref(fs_info
, path
, iref
,
2000 refs_to_add
, extent_op
, NULL
);
2001 } else if (ret
== -ENOENT
) {
2002 setup_inline_extent_backref(fs_info
, path
, iref
, parent
,
2003 root_objectid
, owner
, offset
,
2004 refs_to_add
, extent_op
);
2010 static int insert_extent_backref(struct btrfs_trans_handle
*trans
,
2011 struct btrfs_fs_info
*fs_info
,
2012 struct btrfs_path
*path
,
2013 u64 bytenr
, u64 parent
, u64 root_objectid
,
2014 u64 owner
, u64 offset
, int refs_to_add
)
2017 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
2018 BUG_ON(refs_to_add
!= 1);
2019 ret
= insert_tree_block_ref(trans
, fs_info
, path
, bytenr
,
2020 parent
, root_objectid
);
2022 ret
= insert_extent_data_ref(trans
, fs_info
, path
, bytenr
,
2023 parent
, root_objectid
,
2024 owner
, offset
, refs_to_add
);
2029 static int remove_extent_backref(struct btrfs_trans_handle
*trans
,
2030 struct btrfs_fs_info
*fs_info
,
2031 struct btrfs_path
*path
,
2032 struct btrfs_extent_inline_ref
*iref
,
2033 int refs_to_drop
, int is_data
, int *last_ref
)
2037 BUG_ON(!is_data
&& refs_to_drop
!= 1);
2039 update_inline_extent_backref(fs_info
, path
, iref
,
2040 -refs_to_drop
, NULL
, last_ref
);
2041 } else if (is_data
) {
2042 ret
= remove_extent_data_ref(trans
, fs_info
, path
, refs_to_drop
,
2046 ret
= btrfs_del_item(trans
, fs_info
->extent_root
, path
);
2051 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
2052 static int btrfs_issue_discard(struct block_device
*bdev
, u64 start
, u64 len
,
2053 u64
*discarded_bytes
)
2056 u64 bytes_left
, end
;
2057 u64 aligned_start
= ALIGN(start
, 1 << 9);
2059 if (WARN_ON(start
!= aligned_start
)) {
2060 len
-= aligned_start
- start
;
2061 len
= round_down(len
, 1 << 9);
2062 start
= aligned_start
;
2065 *discarded_bytes
= 0;
2073 /* Skip any superblocks on this device. */
2074 for (j
= 0; j
< BTRFS_SUPER_MIRROR_MAX
; j
++) {
2075 u64 sb_start
= btrfs_sb_offset(j
);
2076 u64 sb_end
= sb_start
+ BTRFS_SUPER_INFO_SIZE
;
2077 u64 size
= sb_start
- start
;
2079 if (!in_range(sb_start
, start
, bytes_left
) &&
2080 !in_range(sb_end
, start
, bytes_left
) &&
2081 !in_range(start
, sb_start
, BTRFS_SUPER_INFO_SIZE
))
2085 * Superblock spans beginning of range. Adjust start and
2088 if (sb_start
<= start
) {
2089 start
+= sb_end
- start
;
2094 bytes_left
= end
- start
;
2099 ret
= blkdev_issue_discard(bdev
, start
>> 9, size
>> 9,
2102 *discarded_bytes
+= size
;
2103 else if (ret
!= -EOPNOTSUPP
)
2112 bytes_left
= end
- start
;
2116 ret
= blkdev_issue_discard(bdev
, start
>> 9, bytes_left
>> 9,
2119 *discarded_bytes
+= bytes_left
;
2124 int btrfs_discard_extent(struct btrfs_fs_info
*fs_info
, u64 bytenr
,
2125 u64 num_bytes
, u64
*actual_bytes
)
2128 u64 discarded_bytes
= 0;
2129 struct btrfs_bio
*bbio
= NULL
;
2133 * Avoid races with device replace and make sure our bbio has devices
2134 * associated to its stripes that don't go away while we are discarding.
2136 btrfs_bio_counter_inc_blocked(fs_info
);
2137 /* Tell the block device(s) that the sectors can be discarded */
2138 ret
= btrfs_map_block(fs_info
, BTRFS_MAP_DISCARD
, bytenr
, &num_bytes
,
2140 /* Error condition is -ENOMEM */
2142 struct btrfs_bio_stripe
*stripe
= bbio
->stripes
;
2146 for (i
= 0; i
< bbio
->num_stripes
; i
++, stripe
++) {
2148 if (!stripe
->dev
->can_discard
)
2151 ret
= btrfs_issue_discard(stripe
->dev
->bdev
,
2156 discarded_bytes
+= bytes
;
2157 else if (ret
!= -EOPNOTSUPP
)
2158 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2161 * Just in case we get back EOPNOTSUPP for some reason,
2162 * just ignore the return value so we don't screw up
2163 * people calling discard_extent.
2167 btrfs_put_bbio(bbio
);
2169 btrfs_bio_counter_dec(fs_info
);
2172 *actual_bytes
= discarded_bytes
;
2175 if (ret
== -EOPNOTSUPP
)
2180 /* Can return -ENOMEM */
2181 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
2182 struct btrfs_root
*root
,
2183 u64 bytenr
, u64 num_bytes
, u64 parent
,
2184 u64 root_objectid
, u64 owner
, u64 offset
)
2186 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2187 int old_ref_mod
, new_ref_mod
;
2190 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
&&
2191 root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
2193 btrfs_ref_tree_mod(root
, bytenr
, num_bytes
, parent
, root_objectid
,
2194 owner
, offset
, BTRFS_ADD_DELAYED_REF
);
2196 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
2197 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
2199 root_objectid
, (int)owner
,
2200 BTRFS_ADD_DELAYED_REF
, NULL
,
2201 &old_ref_mod
, &new_ref_mod
);
2203 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
2205 root_objectid
, owner
, offset
,
2206 0, BTRFS_ADD_DELAYED_REF
,
2207 &old_ref_mod
, &new_ref_mod
);
2210 if (ret
== 0 && old_ref_mod
< 0 && new_ref_mod
>= 0)
2211 add_pinned_bytes(fs_info
, -num_bytes
, owner
, root_objectid
);
2216 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
2217 struct btrfs_fs_info
*fs_info
,
2218 struct btrfs_delayed_ref_node
*node
,
2219 u64 parent
, u64 root_objectid
,
2220 u64 owner
, u64 offset
, int refs_to_add
,
2221 struct btrfs_delayed_extent_op
*extent_op
)
2223 struct btrfs_path
*path
;
2224 struct extent_buffer
*leaf
;
2225 struct btrfs_extent_item
*item
;
2226 struct btrfs_key key
;
2227 u64 bytenr
= node
->bytenr
;
2228 u64 num_bytes
= node
->num_bytes
;
2232 path
= btrfs_alloc_path();
2236 path
->reada
= READA_FORWARD
;
2237 path
->leave_spinning
= 1;
2238 /* this will setup the path even if it fails to insert the back ref */
2239 ret
= insert_inline_extent_backref(trans
, fs_info
, path
, bytenr
,
2240 num_bytes
, parent
, root_objectid
,
2242 refs_to_add
, extent_op
);
2243 if ((ret
< 0 && ret
!= -EAGAIN
) || !ret
)
2247 * Ok we had -EAGAIN which means we didn't have space to insert and
2248 * inline extent ref, so just update the reference count and add a
2251 leaf
= path
->nodes
[0];
2252 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2253 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2254 refs
= btrfs_extent_refs(leaf
, item
);
2255 btrfs_set_extent_refs(leaf
, item
, refs
+ refs_to_add
);
2257 __run_delayed_extent_op(extent_op
, leaf
, item
);
2259 btrfs_mark_buffer_dirty(leaf
);
2260 btrfs_release_path(path
);
2262 path
->reada
= READA_FORWARD
;
2263 path
->leave_spinning
= 1;
2264 /* now insert the actual backref */
2265 ret
= insert_extent_backref(trans
, fs_info
, path
, bytenr
, parent
,
2266 root_objectid
, owner
, offset
, refs_to_add
);
2268 btrfs_abort_transaction(trans
, ret
);
2270 btrfs_free_path(path
);
2274 static int run_delayed_data_ref(struct btrfs_trans_handle
*trans
,
2275 struct btrfs_fs_info
*fs_info
,
2276 struct btrfs_delayed_ref_node
*node
,
2277 struct btrfs_delayed_extent_op
*extent_op
,
2278 int insert_reserved
)
2281 struct btrfs_delayed_data_ref
*ref
;
2282 struct btrfs_key ins
;
2287 ins
.objectid
= node
->bytenr
;
2288 ins
.offset
= node
->num_bytes
;
2289 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2291 ref
= btrfs_delayed_node_to_data_ref(node
);
2292 trace_run_delayed_data_ref(fs_info
, node
, ref
, node
->action
);
2294 if (node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2295 parent
= ref
->parent
;
2296 ref_root
= ref
->root
;
2298 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2300 flags
|= extent_op
->flags_to_set
;
2301 ret
= alloc_reserved_file_extent(trans
, fs_info
,
2302 parent
, ref_root
, flags
,
2303 ref
->objectid
, ref
->offset
,
2304 &ins
, node
->ref_mod
);
2305 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2306 ret
= __btrfs_inc_extent_ref(trans
, fs_info
, node
, parent
,
2307 ref_root
, ref
->objectid
,
2308 ref
->offset
, node
->ref_mod
,
2310 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2311 ret
= __btrfs_free_extent(trans
, fs_info
, node
, parent
,
2312 ref_root
, ref
->objectid
,
2313 ref
->offset
, node
->ref_mod
,
2321 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
2322 struct extent_buffer
*leaf
,
2323 struct btrfs_extent_item
*ei
)
2325 u64 flags
= btrfs_extent_flags(leaf
, ei
);
2326 if (extent_op
->update_flags
) {
2327 flags
|= extent_op
->flags_to_set
;
2328 btrfs_set_extent_flags(leaf
, ei
, flags
);
2331 if (extent_op
->update_key
) {
2332 struct btrfs_tree_block_info
*bi
;
2333 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
));
2334 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
2335 btrfs_set_tree_block_key(leaf
, bi
, &extent_op
->key
);
2339 static int run_delayed_extent_op(struct btrfs_trans_handle
*trans
,
2340 struct btrfs_fs_info
*fs_info
,
2341 struct btrfs_delayed_ref_head
*head
,
2342 struct btrfs_delayed_extent_op
*extent_op
)
2344 struct btrfs_key key
;
2345 struct btrfs_path
*path
;
2346 struct btrfs_extent_item
*ei
;
2347 struct extent_buffer
*leaf
;
2351 int metadata
= !extent_op
->is_data
;
2356 if (metadata
&& !btrfs_fs_incompat(fs_info
, SKINNY_METADATA
))
2359 path
= btrfs_alloc_path();
2363 key
.objectid
= head
->bytenr
;
2366 key
.type
= BTRFS_METADATA_ITEM_KEY
;
2367 key
.offset
= extent_op
->level
;
2369 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2370 key
.offset
= head
->num_bytes
;
2374 path
->reada
= READA_FORWARD
;
2375 path
->leave_spinning
= 1;
2376 ret
= btrfs_search_slot(trans
, fs_info
->extent_root
, &key
, path
, 0, 1);
2383 if (path
->slots
[0] > 0) {
2385 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
2387 if (key
.objectid
== head
->bytenr
&&
2388 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
2389 key
.offset
== head
->num_bytes
)
2393 btrfs_release_path(path
);
2396 key
.objectid
= head
->bytenr
;
2397 key
.offset
= head
->num_bytes
;
2398 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2407 leaf
= path
->nodes
[0];
2408 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2409 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2410 if (item_size
< sizeof(*ei
)) {
2411 ret
= convert_extent_item_v0(trans
, fs_info
, path
, (u64
)-1, 0);
2416 leaf
= path
->nodes
[0];
2417 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2420 BUG_ON(item_size
< sizeof(*ei
));
2421 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2422 __run_delayed_extent_op(extent_op
, leaf
, ei
);
2424 btrfs_mark_buffer_dirty(leaf
);
2426 btrfs_free_path(path
);
2430 static int run_delayed_tree_ref(struct btrfs_trans_handle
*trans
,
2431 struct btrfs_fs_info
*fs_info
,
2432 struct btrfs_delayed_ref_node
*node
,
2433 struct btrfs_delayed_extent_op
*extent_op
,
2434 int insert_reserved
)
2437 struct btrfs_delayed_tree_ref
*ref
;
2438 struct btrfs_key ins
;
2441 bool skinny_metadata
= btrfs_fs_incompat(fs_info
, SKINNY_METADATA
);
2443 ref
= btrfs_delayed_node_to_tree_ref(node
);
2444 trace_run_delayed_tree_ref(fs_info
, node
, ref
, node
->action
);
2446 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2447 parent
= ref
->parent
;
2448 ref_root
= ref
->root
;
2450 ins
.objectid
= node
->bytenr
;
2451 if (skinny_metadata
) {
2452 ins
.offset
= ref
->level
;
2453 ins
.type
= BTRFS_METADATA_ITEM_KEY
;
2455 ins
.offset
= node
->num_bytes
;
2456 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2459 if (node
->ref_mod
!= 1) {
2461 "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu",
2462 node
->bytenr
, node
->ref_mod
, node
->action
, ref_root
,
2466 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2467 BUG_ON(!extent_op
|| !extent_op
->update_flags
);
2468 ret
= alloc_reserved_tree_block(trans
, fs_info
,
2470 extent_op
->flags_to_set
,
2473 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2474 ret
= __btrfs_inc_extent_ref(trans
, fs_info
, node
,
2478 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2479 ret
= __btrfs_free_extent(trans
, fs_info
, node
,
2481 ref
->level
, 0, 1, extent_op
);
2488 /* helper function to actually process a single delayed ref entry */
2489 static int run_one_delayed_ref(struct btrfs_trans_handle
*trans
,
2490 struct btrfs_fs_info
*fs_info
,
2491 struct btrfs_delayed_ref_node
*node
,
2492 struct btrfs_delayed_extent_op
*extent_op
,
2493 int insert_reserved
)
2497 if (trans
->aborted
) {
2498 if (insert_reserved
)
2499 btrfs_pin_extent(fs_info
, node
->bytenr
,
2500 node
->num_bytes
, 1);
2504 if (node
->type
== BTRFS_TREE_BLOCK_REF_KEY
||
2505 node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2506 ret
= run_delayed_tree_ref(trans
, fs_info
, node
, extent_op
,
2508 else if (node
->type
== BTRFS_EXTENT_DATA_REF_KEY
||
2509 node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2510 ret
= run_delayed_data_ref(trans
, fs_info
, node
, extent_op
,
2517 static inline struct btrfs_delayed_ref_node
*
2518 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
2520 struct btrfs_delayed_ref_node
*ref
;
2522 if (RB_EMPTY_ROOT(&head
->ref_tree
))
2526 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2527 * This is to prevent a ref count from going down to zero, which deletes
2528 * the extent item from the extent tree, when there still are references
2529 * to add, which would fail because they would not find the extent item.
2531 if (!list_empty(&head
->ref_add_list
))
2532 return list_first_entry(&head
->ref_add_list
,
2533 struct btrfs_delayed_ref_node
, add_list
);
2535 ref
= rb_entry(rb_first(&head
->ref_tree
),
2536 struct btrfs_delayed_ref_node
, ref_node
);
2537 ASSERT(list_empty(&ref
->add_list
));
2541 static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root
*delayed_refs
,
2542 struct btrfs_delayed_ref_head
*head
)
2544 spin_lock(&delayed_refs
->lock
);
2545 head
->processing
= 0;
2546 delayed_refs
->num_heads_ready
++;
2547 spin_unlock(&delayed_refs
->lock
);
2548 btrfs_delayed_ref_unlock(head
);
2551 static int cleanup_extent_op(struct btrfs_trans_handle
*trans
,
2552 struct btrfs_fs_info
*fs_info
,
2553 struct btrfs_delayed_ref_head
*head
)
2555 struct btrfs_delayed_extent_op
*extent_op
= head
->extent_op
;
2560 head
->extent_op
= NULL
;
2561 if (head
->must_insert_reserved
) {
2562 btrfs_free_delayed_extent_op(extent_op
);
2565 spin_unlock(&head
->lock
);
2566 ret
= run_delayed_extent_op(trans
, fs_info
, head
, extent_op
);
2567 btrfs_free_delayed_extent_op(extent_op
);
2568 return ret
? ret
: 1;
2571 static int cleanup_ref_head(struct btrfs_trans_handle
*trans
,
2572 struct btrfs_fs_info
*fs_info
,
2573 struct btrfs_delayed_ref_head
*head
)
2575 struct btrfs_delayed_ref_root
*delayed_refs
;
2578 delayed_refs
= &trans
->transaction
->delayed_refs
;
2580 ret
= cleanup_extent_op(trans
, fs_info
, head
);
2582 unselect_delayed_ref_head(delayed_refs
, head
);
2583 btrfs_debug(fs_info
, "run_delayed_extent_op returned %d", ret
);
2590 * Need to drop our head ref lock and re-acquire the delayed ref lock
2591 * and then re-check to make sure nobody got added.
2593 spin_unlock(&head
->lock
);
2594 spin_lock(&delayed_refs
->lock
);
2595 spin_lock(&head
->lock
);
2596 if (!RB_EMPTY_ROOT(&head
->ref_tree
) || head
->extent_op
) {
2597 spin_unlock(&head
->lock
);
2598 spin_unlock(&delayed_refs
->lock
);
2601 delayed_refs
->num_heads
--;
2602 rb_erase(&head
->href_node
, &delayed_refs
->href_root
);
2603 RB_CLEAR_NODE(&head
->href_node
);
2604 spin_unlock(&delayed_refs
->lock
);
2605 spin_unlock(&head
->lock
);
2606 atomic_dec(&delayed_refs
->num_entries
);
2608 trace_run_delayed_ref_head(fs_info
, head
, 0);
2610 if (head
->total_ref_mod
< 0) {
2611 struct btrfs_block_group_cache
*cache
;
2613 cache
= btrfs_lookup_block_group(fs_info
, head
->bytenr
);
2615 percpu_counter_add(&cache
->space_info
->total_bytes_pinned
,
2617 btrfs_put_block_group(cache
);
2619 if (head
->is_data
) {
2620 spin_lock(&delayed_refs
->lock
);
2621 delayed_refs
->pending_csums
-= head
->num_bytes
;
2622 spin_unlock(&delayed_refs
->lock
);
2626 if (head
->must_insert_reserved
) {
2627 btrfs_pin_extent(fs_info
, head
->bytenr
,
2628 head
->num_bytes
, 1);
2629 if (head
->is_data
) {
2630 ret
= btrfs_del_csums(trans
, fs_info
, head
->bytenr
,
2635 /* Also free its reserved qgroup space */
2636 btrfs_qgroup_free_delayed_ref(fs_info
, head
->qgroup_ref_root
,
2637 head
->qgroup_reserved
);
2638 btrfs_delayed_ref_unlock(head
);
2639 btrfs_put_delayed_ref_head(head
);
2644 * Returns 0 on success or if called with an already aborted transaction.
2645 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2647 static noinline
int __btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2648 struct btrfs_fs_info
*fs_info
,
2651 struct btrfs_delayed_ref_root
*delayed_refs
;
2652 struct btrfs_delayed_ref_node
*ref
;
2653 struct btrfs_delayed_ref_head
*locked_ref
= NULL
;
2654 struct btrfs_delayed_extent_op
*extent_op
;
2655 ktime_t start
= ktime_get();
2657 unsigned long count
= 0;
2658 unsigned long actual_count
= 0;
2659 int must_insert_reserved
= 0;
2661 delayed_refs
= &trans
->transaction
->delayed_refs
;
2667 spin_lock(&delayed_refs
->lock
);
2668 locked_ref
= btrfs_select_ref_head(trans
);
2670 spin_unlock(&delayed_refs
->lock
);
2674 /* grab the lock that says we are going to process
2675 * all the refs for this head */
2676 ret
= btrfs_delayed_ref_lock(trans
, locked_ref
);
2677 spin_unlock(&delayed_refs
->lock
);
2679 * we may have dropped the spin lock to get the head
2680 * mutex lock, and that might have given someone else
2681 * time to free the head. If that's true, it has been
2682 * removed from our list and we can move on.
2684 if (ret
== -EAGAIN
) {
2692 * We need to try and merge add/drops of the same ref since we
2693 * can run into issues with relocate dropping the implicit ref
2694 * and then it being added back again before the drop can
2695 * finish. If we merged anything we need to re-loop so we can
2697 * Or we can get node references of the same type that weren't
2698 * merged when created due to bumps in the tree mod seq, and
2699 * we need to merge them to prevent adding an inline extent
2700 * backref before dropping it (triggering a BUG_ON at
2701 * insert_inline_extent_backref()).
2703 spin_lock(&locked_ref
->lock
);
2704 btrfs_merge_delayed_refs(trans
, fs_info
, delayed_refs
,
2708 * locked_ref is the head node, so we have to go one
2709 * node back for any delayed ref updates
2711 ref
= select_delayed_ref(locked_ref
);
2713 if (ref
&& ref
->seq
&&
2714 btrfs_check_delayed_seq(fs_info
, delayed_refs
, ref
->seq
)) {
2715 spin_unlock(&locked_ref
->lock
);
2716 unselect_delayed_ref_head(delayed_refs
, locked_ref
);
2724 * We're done processing refs in this ref_head, clean everything
2725 * up and move on to the next ref_head.
2728 ret
= cleanup_ref_head(trans
, fs_info
, locked_ref
);
2730 /* We dropped our lock, we need to loop. */
2743 rb_erase(&ref
->ref_node
, &locked_ref
->ref_tree
);
2744 RB_CLEAR_NODE(&ref
->ref_node
);
2745 if (!list_empty(&ref
->add_list
))
2746 list_del(&ref
->add_list
);
2748 * When we play the delayed ref, also correct the ref_mod on
2751 switch (ref
->action
) {
2752 case BTRFS_ADD_DELAYED_REF
:
2753 case BTRFS_ADD_DELAYED_EXTENT
:
2754 locked_ref
->ref_mod
-= ref
->ref_mod
;
2756 case BTRFS_DROP_DELAYED_REF
:
2757 locked_ref
->ref_mod
+= ref
->ref_mod
;
2762 atomic_dec(&delayed_refs
->num_entries
);
2765 * Record the must-insert_reserved flag before we drop the spin
2768 must_insert_reserved
= locked_ref
->must_insert_reserved
;
2769 locked_ref
->must_insert_reserved
= 0;
2771 extent_op
= locked_ref
->extent_op
;
2772 locked_ref
->extent_op
= NULL
;
2773 spin_unlock(&locked_ref
->lock
);
2775 ret
= run_one_delayed_ref(trans
, fs_info
, ref
, extent_op
,
2776 must_insert_reserved
);
2778 btrfs_free_delayed_extent_op(extent_op
);
2780 unselect_delayed_ref_head(delayed_refs
, locked_ref
);
2781 btrfs_put_delayed_ref(ref
);
2782 btrfs_debug(fs_info
, "run_one_delayed_ref returned %d",
2787 btrfs_put_delayed_ref(ref
);
2793 * We don't want to include ref heads since we can have empty ref heads
2794 * and those will drastically skew our runtime down since we just do
2795 * accounting, no actual extent tree updates.
2797 if (actual_count
> 0) {
2798 u64 runtime
= ktime_to_ns(ktime_sub(ktime_get(), start
));
2802 * We weigh the current average higher than our current runtime
2803 * to avoid large swings in the average.
2805 spin_lock(&delayed_refs
->lock
);
2806 avg
= fs_info
->avg_delayed_ref_runtime
* 3 + runtime
;
2807 fs_info
->avg_delayed_ref_runtime
= avg
>> 2; /* div by 4 */
2808 spin_unlock(&delayed_refs
->lock
);
2813 #ifdef SCRAMBLE_DELAYED_REFS
2815 * Normally delayed refs get processed in ascending bytenr order. This
2816 * correlates in most cases to the order added. To expose dependencies on this
2817 * order, we start to process the tree in the middle instead of the beginning
2819 static u64
find_middle(struct rb_root
*root
)
2821 struct rb_node
*n
= root
->rb_node
;
2822 struct btrfs_delayed_ref_node
*entry
;
2825 u64 first
= 0, last
= 0;
2829 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2830 first
= entry
->bytenr
;
2834 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2835 last
= entry
->bytenr
;
2840 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2841 WARN_ON(!entry
->in_tree
);
2843 middle
= entry
->bytenr
;
2856 static inline u64
heads_to_leaves(struct btrfs_fs_info
*fs_info
, u64 heads
)
2860 num_bytes
= heads
* (sizeof(struct btrfs_extent_item
) +
2861 sizeof(struct btrfs_extent_inline_ref
));
2862 if (!btrfs_fs_incompat(fs_info
, SKINNY_METADATA
))
2863 num_bytes
+= heads
* sizeof(struct btrfs_tree_block_info
);
2866 * We don't ever fill up leaves all the way so multiply by 2 just to be
2867 * closer to what we're really going to want to use.
2869 return div_u64(num_bytes
, BTRFS_LEAF_DATA_SIZE(fs_info
));
2873 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2874 * would require to store the csums for that many bytes.
2876 u64
btrfs_csum_bytes_to_leaves(struct btrfs_fs_info
*fs_info
, u64 csum_bytes
)
2879 u64 num_csums_per_leaf
;
2882 csum_size
= BTRFS_MAX_ITEM_SIZE(fs_info
);
2883 num_csums_per_leaf
= div64_u64(csum_size
,
2884 (u64
)btrfs_super_csum_size(fs_info
->super_copy
));
2885 num_csums
= div64_u64(csum_bytes
, fs_info
->sectorsize
);
2886 num_csums
+= num_csums_per_leaf
- 1;
2887 num_csums
= div64_u64(num_csums
, num_csums_per_leaf
);
2891 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle
*trans
,
2892 struct btrfs_fs_info
*fs_info
)
2894 struct btrfs_block_rsv
*global_rsv
;
2895 u64 num_heads
= trans
->transaction
->delayed_refs
.num_heads_ready
;
2896 u64 csum_bytes
= trans
->transaction
->delayed_refs
.pending_csums
;
2897 u64 num_dirty_bgs
= trans
->transaction
->num_dirty_bgs
;
2898 u64 num_bytes
, num_dirty_bgs_bytes
;
2901 num_bytes
= btrfs_calc_trans_metadata_size(fs_info
, 1);
2902 num_heads
= heads_to_leaves(fs_info
, num_heads
);
2904 num_bytes
+= (num_heads
- 1) * fs_info
->nodesize
;
2906 num_bytes
+= btrfs_csum_bytes_to_leaves(fs_info
, csum_bytes
) *
2908 num_dirty_bgs_bytes
= btrfs_calc_trans_metadata_size(fs_info
,
2910 global_rsv
= &fs_info
->global_block_rsv
;
2913 * If we can't allocate any more chunks lets make sure we have _lots_ of
2914 * wiggle room since running delayed refs can create more delayed refs.
2916 if (global_rsv
->space_info
->full
) {
2917 num_dirty_bgs_bytes
<<= 1;
2921 spin_lock(&global_rsv
->lock
);
2922 if (global_rsv
->reserved
<= num_bytes
+ num_dirty_bgs_bytes
)
2924 spin_unlock(&global_rsv
->lock
);
2928 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle
*trans
,
2929 struct btrfs_fs_info
*fs_info
)
2932 atomic_read(&trans
->transaction
->delayed_refs
.num_entries
);
2937 avg_runtime
= fs_info
->avg_delayed_ref_runtime
;
2938 val
= num_entries
* avg_runtime
;
2939 if (val
>= NSEC_PER_SEC
)
2941 if (val
>= NSEC_PER_SEC
/ 2)
2944 return btrfs_check_space_for_delayed_refs(trans
, fs_info
);
2947 struct async_delayed_refs
{
2948 struct btrfs_root
*root
;
2953 struct completion wait
;
2954 struct btrfs_work work
;
2957 static inline struct async_delayed_refs
*
2958 to_async_delayed_refs(struct btrfs_work
*work
)
2960 return container_of(work
, struct async_delayed_refs
, work
);
2963 static void delayed_ref_async_start(struct btrfs_work
*work
)
2965 struct async_delayed_refs
*async
= to_async_delayed_refs(work
);
2966 struct btrfs_trans_handle
*trans
;
2967 struct btrfs_fs_info
*fs_info
= async
->root
->fs_info
;
2970 /* if the commit is already started, we don't need to wait here */
2971 if (btrfs_transaction_blocked(fs_info
))
2974 trans
= btrfs_join_transaction(async
->root
);
2975 if (IS_ERR(trans
)) {
2976 async
->error
= PTR_ERR(trans
);
2981 * trans->sync means that when we call end_transaction, we won't
2982 * wait on delayed refs
2986 /* Don't bother flushing if we got into a different transaction */
2987 if (trans
->transid
> async
->transid
)
2990 ret
= btrfs_run_delayed_refs(trans
, fs_info
, async
->count
);
2994 ret
= btrfs_end_transaction(trans
);
2995 if (ret
&& !async
->error
)
2999 complete(&async
->wait
);
3004 int btrfs_async_run_delayed_refs(struct btrfs_fs_info
*fs_info
,
3005 unsigned long count
, u64 transid
, int wait
)
3007 struct async_delayed_refs
*async
;
3010 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
3014 async
->root
= fs_info
->tree_root
;
3015 async
->count
= count
;
3017 async
->transid
= transid
;
3022 init_completion(&async
->wait
);
3024 btrfs_init_work(&async
->work
, btrfs_extent_refs_helper
,
3025 delayed_ref_async_start
, NULL
, NULL
);
3027 btrfs_queue_work(fs_info
->extent_workers
, &async
->work
);
3030 wait_for_completion(&async
->wait
);
3039 * this starts processing the delayed reference count updates and
3040 * extent insertions we have queued up so far. count can be
3041 * 0, which means to process everything in the tree at the start
3042 * of the run (but not newly added entries), or it can be some target
3043 * number you'd like to process.
3045 * Returns 0 on success or if called with an aborted transaction
3046 * Returns <0 on error and aborts the transaction
3048 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
3049 struct btrfs_fs_info
*fs_info
, unsigned long count
)
3051 struct rb_node
*node
;
3052 struct btrfs_delayed_ref_root
*delayed_refs
;
3053 struct btrfs_delayed_ref_head
*head
;
3055 int run_all
= count
== (unsigned long)-1;
3056 bool can_flush_pending_bgs
= trans
->can_flush_pending_bgs
;
3058 /* We'll clean this up in btrfs_cleanup_transaction */
3062 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE
, &fs_info
->flags
))
3065 delayed_refs
= &trans
->transaction
->delayed_refs
;
3067 count
= atomic_read(&delayed_refs
->num_entries
) * 2;
3070 #ifdef SCRAMBLE_DELAYED_REFS
3071 delayed_refs
->run_delayed_start
= find_middle(&delayed_refs
->root
);
3073 trans
->can_flush_pending_bgs
= false;
3074 ret
= __btrfs_run_delayed_refs(trans
, fs_info
, count
);
3076 btrfs_abort_transaction(trans
, ret
);
3081 if (!list_empty(&trans
->new_bgs
))
3082 btrfs_create_pending_block_groups(trans
, fs_info
);
3084 spin_lock(&delayed_refs
->lock
);
3085 node
= rb_first(&delayed_refs
->href_root
);
3087 spin_unlock(&delayed_refs
->lock
);
3090 head
= rb_entry(node
, struct btrfs_delayed_ref_head
,
3092 refcount_inc(&head
->refs
);
3093 spin_unlock(&delayed_refs
->lock
);
3095 /* Mutex was contended, block until it's released and retry. */
3096 mutex_lock(&head
->mutex
);
3097 mutex_unlock(&head
->mutex
);
3099 btrfs_put_delayed_ref_head(head
);
3104 trans
->can_flush_pending_bgs
= can_flush_pending_bgs
;
3108 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
3109 struct btrfs_fs_info
*fs_info
,
3110 u64 bytenr
, u64 num_bytes
, u64 flags
,
3111 int level
, int is_data
)
3113 struct btrfs_delayed_extent_op
*extent_op
;
3116 extent_op
= btrfs_alloc_delayed_extent_op();
3120 extent_op
->flags_to_set
= flags
;
3121 extent_op
->update_flags
= true;
3122 extent_op
->update_key
= false;
3123 extent_op
->is_data
= is_data
? true : false;
3124 extent_op
->level
= level
;
3126 ret
= btrfs_add_delayed_extent_op(fs_info
, trans
, bytenr
,
3127 num_bytes
, extent_op
);
3129 btrfs_free_delayed_extent_op(extent_op
);
3133 static noinline
int check_delayed_ref(struct btrfs_root
*root
,
3134 struct btrfs_path
*path
,
3135 u64 objectid
, u64 offset
, u64 bytenr
)
3137 struct btrfs_delayed_ref_head
*head
;
3138 struct btrfs_delayed_ref_node
*ref
;
3139 struct btrfs_delayed_data_ref
*data_ref
;
3140 struct btrfs_delayed_ref_root
*delayed_refs
;
3141 struct btrfs_transaction
*cur_trans
;
3142 struct rb_node
*node
;
3145 cur_trans
= root
->fs_info
->running_transaction
;
3149 delayed_refs
= &cur_trans
->delayed_refs
;
3150 spin_lock(&delayed_refs
->lock
);
3151 head
= btrfs_find_delayed_ref_head(delayed_refs
, bytenr
);
3153 spin_unlock(&delayed_refs
->lock
);
3157 if (!mutex_trylock(&head
->mutex
)) {
3158 refcount_inc(&head
->refs
);
3159 spin_unlock(&delayed_refs
->lock
);
3161 btrfs_release_path(path
);
3164 * Mutex was contended, block until it's released and let
3167 mutex_lock(&head
->mutex
);
3168 mutex_unlock(&head
->mutex
);
3169 btrfs_put_delayed_ref_head(head
);
3172 spin_unlock(&delayed_refs
->lock
);
3174 spin_lock(&head
->lock
);
3176 * XXX: We should replace this with a proper search function in the
3179 for (node
= rb_first(&head
->ref_tree
); node
; node
= rb_next(node
)) {
3180 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, ref_node
);
3181 /* If it's a shared ref we know a cross reference exists */
3182 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
) {
3187 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
3190 * If our ref doesn't match the one we're currently looking at
3191 * then we have a cross reference.
3193 if (data_ref
->root
!= root
->root_key
.objectid
||
3194 data_ref
->objectid
!= objectid
||
3195 data_ref
->offset
!= offset
) {
3200 spin_unlock(&head
->lock
);
3201 mutex_unlock(&head
->mutex
);
3205 static noinline
int check_committed_ref(struct btrfs_root
*root
,
3206 struct btrfs_path
*path
,
3207 u64 objectid
, u64 offset
, u64 bytenr
)
3209 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3210 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
3211 struct extent_buffer
*leaf
;
3212 struct btrfs_extent_data_ref
*ref
;
3213 struct btrfs_extent_inline_ref
*iref
;
3214 struct btrfs_extent_item
*ei
;
3215 struct btrfs_key key
;
3220 key
.objectid
= bytenr
;
3221 key
.offset
= (u64
)-1;
3222 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
3224 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
3227 BUG_ON(ret
== 0); /* Corruption */
3230 if (path
->slots
[0] == 0)
3234 leaf
= path
->nodes
[0];
3235 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
3237 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
3241 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
3242 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3243 if (item_size
< sizeof(*ei
)) {
3244 WARN_ON(item_size
!= sizeof(struct btrfs_extent_item_v0
));
3248 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
3250 if (item_size
!= sizeof(*ei
) +
3251 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
3254 if (btrfs_extent_generation(leaf
, ei
) <=
3255 btrfs_root_last_snapshot(&root
->root_item
))
3258 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
3260 type
= btrfs_get_extent_inline_ref_type(leaf
, iref
, BTRFS_REF_TYPE_DATA
);
3261 if (type
!= BTRFS_EXTENT_DATA_REF_KEY
)
3264 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
3265 if (btrfs_extent_refs(leaf
, ei
) !=
3266 btrfs_extent_data_ref_count(leaf
, ref
) ||
3267 btrfs_extent_data_ref_root(leaf
, ref
) !=
3268 root
->root_key
.objectid
||
3269 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
3270 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
3278 int btrfs_cross_ref_exist(struct btrfs_root
*root
, u64 objectid
, u64 offset
,
3281 struct btrfs_path
*path
;
3285 path
= btrfs_alloc_path();
3290 ret
= check_committed_ref(root
, path
, objectid
,
3292 if (ret
&& ret
!= -ENOENT
)
3295 ret2
= check_delayed_ref(root
, path
, objectid
,
3297 } while (ret2
== -EAGAIN
);
3299 if (ret2
&& ret2
!= -ENOENT
) {
3304 if (ret
!= -ENOENT
|| ret2
!= -ENOENT
)
3307 btrfs_free_path(path
);
3308 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
3313 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
3314 struct btrfs_root
*root
,
3315 struct extent_buffer
*buf
,
3316 int full_backref
, int inc
)
3318 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3324 struct btrfs_key key
;
3325 struct btrfs_file_extent_item
*fi
;
3329 int (*process_func
)(struct btrfs_trans_handle
*,
3330 struct btrfs_root
*,
3331 u64
, u64
, u64
, u64
, u64
, u64
);
3334 if (btrfs_is_testing(fs_info
))
3337 ref_root
= btrfs_header_owner(buf
);
3338 nritems
= btrfs_header_nritems(buf
);
3339 level
= btrfs_header_level(buf
);
3341 if (!test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) && level
== 0)
3345 process_func
= btrfs_inc_extent_ref
;
3347 process_func
= btrfs_free_extent
;
3350 parent
= buf
->start
;
3354 for (i
= 0; i
< nritems
; i
++) {
3356 btrfs_item_key_to_cpu(buf
, &key
, i
);
3357 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
3359 fi
= btrfs_item_ptr(buf
, i
,
3360 struct btrfs_file_extent_item
);
3361 if (btrfs_file_extent_type(buf
, fi
) ==
3362 BTRFS_FILE_EXTENT_INLINE
)
3364 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
3368 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
3369 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
3370 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
3371 parent
, ref_root
, key
.objectid
,
3376 bytenr
= btrfs_node_blockptr(buf
, i
);
3377 num_bytes
= fs_info
->nodesize
;
3378 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
3379 parent
, ref_root
, level
- 1, 0);
3389 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3390 struct extent_buffer
*buf
, int full_backref
)
3392 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1);
3395 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3396 struct extent_buffer
*buf
, int full_backref
)
3398 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0);
3401 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
3402 struct btrfs_fs_info
*fs_info
,
3403 struct btrfs_path
*path
,
3404 struct btrfs_block_group_cache
*cache
)
3407 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
3409 struct extent_buffer
*leaf
;
3411 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
3418 leaf
= path
->nodes
[0];
3419 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
3420 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
3421 btrfs_mark_buffer_dirty(leaf
);
3423 btrfs_release_path(path
);
3428 static struct btrfs_block_group_cache
*
3429 next_block_group(struct btrfs_fs_info
*fs_info
,
3430 struct btrfs_block_group_cache
*cache
)
3432 struct rb_node
*node
;
3434 spin_lock(&fs_info
->block_group_cache_lock
);
3436 /* If our block group was removed, we need a full search. */
3437 if (RB_EMPTY_NODE(&cache
->cache_node
)) {
3438 const u64 next_bytenr
= cache
->key
.objectid
+ cache
->key
.offset
;
3440 spin_unlock(&fs_info
->block_group_cache_lock
);
3441 btrfs_put_block_group(cache
);
3442 cache
= btrfs_lookup_first_block_group(fs_info
, next_bytenr
); return cache
;
3444 node
= rb_next(&cache
->cache_node
);
3445 btrfs_put_block_group(cache
);
3447 cache
= rb_entry(node
, struct btrfs_block_group_cache
,
3449 btrfs_get_block_group(cache
);
3452 spin_unlock(&fs_info
->block_group_cache_lock
);
3456 static int cache_save_setup(struct btrfs_block_group_cache
*block_group
,
3457 struct btrfs_trans_handle
*trans
,
3458 struct btrfs_path
*path
)
3460 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
3461 struct btrfs_root
*root
= fs_info
->tree_root
;
3462 struct inode
*inode
= NULL
;
3463 struct extent_changeset
*data_reserved
= NULL
;
3465 int dcs
= BTRFS_DC_ERROR
;
3471 * If this block group is smaller than 100 megs don't bother caching the
3474 if (block_group
->key
.offset
< (100 * SZ_1M
)) {
3475 spin_lock(&block_group
->lock
);
3476 block_group
->disk_cache_state
= BTRFS_DC_WRITTEN
;
3477 spin_unlock(&block_group
->lock
);
3484 inode
= lookup_free_space_inode(fs_info
, block_group
, path
);
3485 if (IS_ERR(inode
) && PTR_ERR(inode
) != -ENOENT
) {
3486 ret
= PTR_ERR(inode
);
3487 btrfs_release_path(path
);
3491 if (IS_ERR(inode
)) {
3495 if (block_group
->ro
)
3498 ret
= create_free_space_inode(fs_info
, trans
, block_group
,
3505 /* We've already setup this transaction, go ahead and exit */
3506 if (block_group
->cache_generation
== trans
->transid
&&
3507 i_size_read(inode
)) {
3508 dcs
= BTRFS_DC_SETUP
;
3513 * We want to set the generation to 0, that way if anything goes wrong
3514 * from here on out we know not to trust this cache when we load up next
3517 BTRFS_I(inode
)->generation
= 0;
3518 ret
= btrfs_update_inode(trans
, root
, inode
);
3521 * So theoretically we could recover from this, simply set the
3522 * super cache generation to 0 so we know to invalidate the
3523 * cache, but then we'd have to keep track of the block groups
3524 * that fail this way so we know we _have_ to reset this cache
3525 * before the next commit or risk reading stale cache. So to
3526 * limit our exposure to horrible edge cases lets just abort the
3527 * transaction, this only happens in really bad situations
3530 btrfs_abort_transaction(trans
, ret
);
3535 if (i_size_read(inode
) > 0) {
3536 ret
= btrfs_check_trunc_cache_free_space(fs_info
,
3537 &fs_info
->global_block_rsv
);
3541 ret
= btrfs_truncate_free_space_cache(trans
, NULL
, inode
);
3546 spin_lock(&block_group
->lock
);
3547 if (block_group
->cached
!= BTRFS_CACHE_FINISHED
||
3548 !btrfs_test_opt(fs_info
, SPACE_CACHE
)) {
3550 * don't bother trying to write stuff out _if_
3551 * a) we're not cached,
3552 * b) we're with nospace_cache mount option,
3553 * c) we're with v2 space_cache (FREE_SPACE_TREE).
3555 dcs
= BTRFS_DC_WRITTEN
;
3556 spin_unlock(&block_group
->lock
);
3559 spin_unlock(&block_group
->lock
);
3562 * We hit an ENOSPC when setting up the cache in this transaction, just
3563 * skip doing the setup, we've already cleared the cache so we're safe.
3565 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC
, &trans
->transaction
->flags
)) {
3571 * Try to preallocate enough space based on how big the block group is.
3572 * Keep in mind this has to include any pinned space which could end up
3573 * taking up quite a bit since it's not folded into the other space
3576 num_pages
= div_u64(block_group
->key
.offset
, SZ_256M
);
3581 num_pages
*= PAGE_SIZE
;
3583 ret
= btrfs_check_data_free_space(inode
, &data_reserved
, 0, num_pages
);
3587 ret
= btrfs_prealloc_file_range_trans(inode
, trans
, 0, 0, num_pages
,
3588 num_pages
, num_pages
,
3591 * Our cache requires contiguous chunks so that we don't modify a bunch
3592 * of metadata or split extents when writing the cache out, which means
3593 * we can enospc if we are heavily fragmented in addition to just normal
3594 * out of space conditions. So if we hit this just skip setting up any
3595 * other block groups for this transaction, maybe we'll unpin enough
3596 * space the next time around.
3599 dcs
= BTRFS_DC_SETUP
;
3600 else if (ret
== -ENOSPC
)
3601 set_bit(BTRFS_TRANS_CACHE_ENOSPC
, &trans
->transaction
->flags
);
3606 btrfs_release_path(path
);
3608 spin_lock(&block_group
->lock
);
3609 if (!ret
&& dcs
== BTRFS_DC_SETUP
)
3610 block_group
->cache_generation
= trans
->transid
;
3611 block_group
->disk_cache_state
= dcs
;
3612 spin_unlock(&block_group
->lock
);
3614 extent_changeset_free(data_reserved
);
3618 int btrfs_setup_space_cache(struct btrfs_trans_handle
*trans
,
3619 struct btrfs_fs_info
*fs_info
)
3621 struct btrfs_block_group_cache
*cache
, *tmp
;
3622 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3623 struct btrfs_path
*path
;
3625 if (list_empty(&cur_trans
->dirty_bgs
) ||
3626 !btrfs_test_opt(fs_info
, SPACE_CACHE
))
3629 path
= btrfs_alloc_path();
3633 /* Could add new block groups, use _safe just in case */
3634 list_for_each_entry_safe(cache
, tmp
, &cur_trans
->dirty_bgs
,
3636 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
)
3637 cache_save_setup(cache
, trans
, path
);
3640 btrfs_free_path(path
);
3645 * transaction commit does final block group cache writeback during a
3646 * critical section where nothing is allowed to change the FS. This is
3647 * required in order for the cache to actually match the block group,
3648 * but can introduce a lot of latency into the commit.
3650 * So, btrfs_start_dirty_block_groups is here to kick off block group
3651 * cache IO. There's a chance we'll have to redo some of it if the
3652 * block group changes again during the commit, but it greatly reduces
3653 * the commit latency by getting rid of the easy block groups while
3654 * we're still allowing others to join the commit.
3656 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle
*trans
,
3657 struct btrfs_fs_info
*fs_info
)
3659 struct btrfs_block_group_cache
*cache
;
3660 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3663 struct btrfs_path
*path
= NULL
;
3665 struct list_head
*io
= &cur_trans
->io_bgs
;
3666 int num_started
= 0;
3669 spin_lock(&cur_trans
->dirty_bgs_lock
);
3670 if (list_empty(&cur_trans
->dirty_bgs
)) {
3671 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3674 list_splice_init(&cur_trans
->dirty_bgs
, &dirty
);
3675 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3679 * make sure all the block groups on our dirty list actually
3682 btrfs_create_pending_block_groups(trans
, fs_info
);
3685 path
= btrfs_alloc_path();
3691 * cache_write_mutex is here only to save us from balance or automatic
3692 * removal of empty block groups deleting this block group while we are
3693 * writing out the cache
3695 mutex_lock(&trans
->transaction
->cache_write_mutex
);
3696 while (!list_empty(&dirty
)) {
3697 cache
= list_first_entry(&dirty
,
3698 struct btrfs_block_group_cache
,
3701 * this can happen if something re-dirties a block
3702 * group that is already under IO. Just wait for it to
3703 * finish and then do it all again
3705 if (!list_empty(&cache
->io_list
)) {
3706 list_del_init(&cache
->io_list
);
3707 btrfs_wait_cache_io(trans
, cache
, path
);
3708 btrfs_put_block_group(cache
);
3713 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3714 * if it should update the cache_state. Don't delete
3715 * until after we wait.
3717 * Since we're not running in the commit critical section
3718 * we need the dirty_bgs_lock to protect from update_block_group
3720 spin_lock(&cur_trans
->dirty_bgs_lock
);
3721 list_del_init(&cache
->dirty_list
);
3722 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3726 cache_save_setup(cache
, trans
, path
);
3728 if (cache
->disk_cache_state
== BTRFS_DC_SETUP
) {
3729 cache
->io_ctl
.inode
= NULL
;
3730 ret
= btrfs_write_out_cache(fs_info
, trans
,
3732 if (ret
== 0 && cache
->io_ctl
.inode
) {
3737 * the cache_write_mutex is protecting
3740 list_add_tail(&cache
->io_list
, io
);
3743 * if we failed to write the cache, the
3744 * generation will be bad and life goes on
3750 ret
= write_one_cache_group(trans
, fs_info
,
3753 * Our block group might still be attached to the list
3754 * of new block groups in the transaction handle of some
3755 * other task (struct btrfs_trans_handle->new_bgs). This
3756 * means its block group item isn't yet in the extent
3757 * tree. If this happens ignore the error, as we will
3758 * try again later in the critical section of the
3759 * transaction commit.
3761 if (ret
== -ENOENT
) {
3763 spin_lock(&cur_trans
->dirty_bgs_lock
);
3764 if (list_empty(&cache
->dirty_list
)) {
3765 list_add_tail(&cache
->dirty_list
,
3766 &cur_trans
->dirty_bgs
);
3767 btrfs_get_block_group(cache
);
3769 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3771 btrfs_abort_transaction(trans
, ret
);
3775 /* if its not on the io list, we need to put the block group */
3777 btrfs_put_block_group(cache
);
3783 * Avoid blocking other tasks for too long. It might even save
3784 * us from writing caches for block groups that are going to be
3787 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
3788 mutex_lock(&trans
->transaction
->cache_write_mutex
);
3790 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
3793 * go through delayed refs for all the stuff we've just kicked off
3794 * and then loop back (just once)
3796 ret
= btrfs_run_delayed_refs(trans
, fs_info
, 0);
3797 if (!ret
&& loops
== 0) {
3799 spin_lock(&cur_trans
->dirty_bgs_lock
);
3800 list_splice_init(&cur_trans
->dirty_bgs
, &dirty
);
3802 * dirty_bgs_lock protects us from concurrent block group
3803 * deletes too (not just cache_write_mutex).
3805 if (!list_empty(&dirty
)) {
3806 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3809 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3810 } else if (ret
< 0) {
3811 btrfs_cleanup_dirty_bgs(cur_trans
, fs_info
);
3814 btrfs_free_path(path
);
3818 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
3819 struct btrfs_fs_info
*fs_info
)
3821 struct btrfs_block_group_cache
*cache
;
3822 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3825 struct btrfs_path
*path
;
3826 struct list_head
*io
= &cur_trans
->io_bgs
;
3827 int num_started
= 0;
3829 path
= btrfs_alloc_path();
3834 * Even though we are in the critical section of the transaction commit,
3835 * we can still have concurrent tasks adding elements to this
3836 * transaction's list of dirty block groups. These tasks correspond to
3837 * endio free space workers started when writeback finishes for a
3838 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3839 * allocate new block groups as a result of COWing nodes of the root
3840 * tree when updating the free space inode. The writeback for the space
3841 * caches is triggered by an earlier call to
3842 * btrfs_start_dirty_block_groups() and iterations of the following
3844 * Also we want to do the cache_save_setup first and then run the
3845 * delayed refs to make sure we have the best chance at doing this all
3848 spin_lock(&cur_trans
->dirty_bgs_lock
);
3849 while (!list_empty(&cur_trans
->dirty_bgs
)) {
3850 cache
= list_first_entry(&cur_trans
->dirty_bgs
,
3851 struct btrfs_block_group_cache
,
3855 * this can happen if cache_save_setup re-dirties a block
3856 * group that is already under IO. Just wait for it to
3857 * finish and then do it all again
3859 if (!list_empty(&cache
->io_list
)) {
3860 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3861 list_del_init(&cache
->io_list
);
3862 btrfs_wait_cache_io(trans
, cache
, path
);
3863 btrfs_put_block_group(cache
);
3864 spin_lock(&cur_trans
->dirty_bgs_lock
);
3868 * don't remove from the dirty list until after we've waited
3871 list_del_init(&cache
->dirty_list
);
3872 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3875 cache_save_setup(cache
, trans
, path
);
3878 ret
= btrfs_run_delayed_refs(trans
, fs_info
,
3879 (unsigned long) -1);
3881 if (!ret
&& cache
->disk_cache_state
== BTRFS_DC_SETUP
) {
3882 cache
->io_ctl
.inode
= NULL
;
3883 ret
= btrfs_write_out_cache(fs_info
, trans
,
3885 if (ret
== 0 && cache
->io_ctl
.inode
) {
3888 list_add_tail(&cache
->io_list
, io
);
3891 * if we failed to write the cache, the
3892 * generation will be bad and life goes on
3898 ret
= write_one_cache_group(trans
, fs_info
,
3901 * One of the free space endio workers might have
3902 * created a new block group while updating a free space
3903 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3904 * and hasn't released its transaction handle yet, in
3905 * which case the new block group is still attached to
3906 * its transaction handle and its creation has not
3907 * finished yet (no block group item in the extent tree
3908 * yet, etc). If this is the case, wait for all free
3909 * space endio workers to finish and retry. This is a
3910 * a very rare case so no need for a more efficient and
3913 if (ret
== -ENOENT
) {
3914 wait_event(cur_trans
->writer_wait
,
3915 atomic_read(&cur_trans
->num_writers
) == 1);
3916 ret
= write_one_cache_group(trans
, fs_info
,
3920 btrfs_abort_transaction(trans
, ret
);
3923 /* if its not on the io list, we need to put the block group */
3925 btrfs_put_block_group(cache
);
3926 spin_lock(&cur_trans
->dirty_bgs_lock
);
3928 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3930 while (!list_empty(io
)) {
3931 cache
= list_first_entry(io
, struct btrfs_block_group_cache
,
3933 list_del_init(&cache
->io_list
);
3934 btrfs_wait_cache_io(trans
, cache
, path
);
3935 btrfs_put_block_group(cache
);
3938 btrfs_free_path(path
);
3942 int btrfs_extent_readonly(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
3944 struct btrfs_block_group_cache
*block_group
;
3947 block_group
= btrfs_lookup_block_group(fs_info
, bytenr
);
3948 if (!block_group
|| block_group
->ro
)
3951 btrfs_put_block_group(block_group
);
3955 bool btrfs_inc_nocow_writers(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
3957 struct btrfs_block_group_cache
*bg
;
3960 bg
= btrfs_lookup_block_group(fs_info
, bytenr
);
3964 spin_lock(&bg
->lock
);
3968 atomic_inc(&bg
->nocow_writers
);
3969 spin_unlock(&bg
->lock
);
3971 /* no put on block group, done by btrfs_dec_nocow_writers */
3973 btrfs_put_block_group(bg
);
3979 void btrfs_dec_nocow_writers(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
3981 struct btrfs_block_group_cache
*bg
;
3983 bg
= btrfs_lookup_block_group(fs_info
, bytenr
);
3985 if (atomic_dec_and_test(&bg
->nocow_writers
))
3986 wake_up_atomic_t(&bg
->nocow_writers
);
3988 * Once for our lookup and once for the lookup done by a previous call
3989 * to btrfs_inc_nocow_writers()
3991 btrfs_put_block_group(bg
);
3992 btrfs_put_block_group(bg
);
3995 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache
*bg
)
3997 wait_on_atomic_t(&bg
->nocow_writers
, atomic_t_wait
,
3998 TASK_UNINTERRUPTIBLE
);
4001 static const char *alloc_name(u64 flags
)
4004 case BTRFS_BLOCK_GROUP_METADATA
|BTRFS_BLOCK_GROUP_DATA
:
4006 case BTRFS_BLOCK_GROUP_METADATA
:
4008 case BTRFS_BLOCK_GROUP_DATA
:
4010 case BTRFS_BLOCK_GROUP_SYSTEM
:
4014 return "invalid-combination";
4018 static int create_space_info(struct btrfs_fs_info
*info
, u64 flags
,
4019 struct btrfs_space_info
**new)
4022 struct btrfs_space_info
*space_info
;
4026 space_info
= kzalloc(sizeof(*space_info
), GFP_NOFS
);
4030 ret
= percpu_counter_init(&space_info
->total_bytes_pinned
, 0,
4037 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
4038 INIT_LIST_HEAD(&space_info
->block_groups
[i
]);
4039 init_rwsem(&space_info
->groups_sem
);
4040 spin_lock_init(&space_info
->lock
);
4041 space_info
->flags
= flags
& BTRFS_BLOCK_GROUP_TYPE_MASK
;
4042 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
4043 init_waitqueue_head(&space_info
->wait
);
4044 INIT_LIST_HEAD(&space_info
->ro_bgs
);
4045 INIT_LIST_HEAD(&space_info
->tickets
);
4046 INIT_LIST_HEAD(&space_info
->priority_tickets
);
4048 ret
= kobject_init_and_add(&space_info
->kobj
, &space_info_ktype
,
4049 info
->space_info_kobj
, "%s",
4050 alloc_name(space_info
->flags
));
4052 percpu_counter_destroy(&space_info
->total_bytes_pinned
);
4058 list_add_rcu(&space_info
->list
, &info
->space_info
);
4059 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
4060 info
->data_sinfo
= space_info
;
4065 static void update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
4066 u64 total_bytes
, u64 bytes_used
,
4068 struct btrfs_space_info
**space_info
)
4070 struct btrfs_space_info
*found
;
4073 if (flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
4074 BTRFS_BLOCK_GROUP_RAID10
))
4079 found
= __find_space_info(info
, flags
);
4081 spin_lock(&found
->lock
);
4082 found
->total_bytes
+= total_bytes
;
4083 found
->disk_total
+= total_bytes
* factor
;
4084 found
->bytes_used
+= bytes_used
;
4085 found
->disk_used
+= bytes_used
* factor
;
4086 found
->bytes_readonly
+= bytes_readonly
;
4087 if (total_bytes
> 0)
4089 space_info_add_new_bytes(info
, found
, total_bytes
-
4090 bytes_used
- bytes_readonly
);
4091 spin_unlock(&found
->lock
);
4092 *space_info
= found
;
4095 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
4097 u64 extra_flags
= chunk_to_extended(flags
) &
4098 BTRFS_EXTENDED_PROFILE_MASK
;
4100 write_seqlock(&fs_info
->profiles_lock
);
4101 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
4102 fs_info
->avail_data_alloc_bits
|= extra_flags
;
4103 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
4104 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
4105 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
4106 fs_info
->avail_system_alloc_bits
|= extra_flags
;
4107 write_sequnlock(&fs_info
->profiles_lock
);
4111 * returns target flags in extended format or 0 if restripe for this
4112 * chunk_type is not in progress
4114 * should be called with either volume_mutex or balance_lock held
4116 static u64
get_restripe_target(struct btrfs_fs_info
*fs_info
, u64 flags
)
4118 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
4124 if (flags
& BTRFS_BLOCK_GROUP_DATA
&&
4125 bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
4126 target
= BTRFS_BLOCK_GROUP_DATA
| bctl
->data
.target
;
4127 } else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
&&
4128 bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
4129 target
= BTRFS_BLOCK_GROUP_SYSTEM
| bctl
->sys
.target
;
4130 } else if (flags
& BTRFS_BLOCK_GROUP_METADATA
&&
4131 bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
4132 target
= BTRFS_BLOCK_GROUP_METADATA
| bctl
->meta
.target
;
4139 * @flags: available profiles in extended format (see ctree.h)
4141 * Returns reduced profile in chunk format. If profile changing is in
4142 * progress (either running or paused) picks the target profile (if it's
4143 * already available), otherwise falls back to plain reducing.
4145 static u64
btrfs_reduce_alloc_profile(struct btrfs_fs_info
*fs_info
, u64 flags
)
4147 u64 num_devices
= fs_info
->fs_devices
->rw_devices
;
4153 * see if restripe for this chunk_type is in progress, if so
4154 * try to reduce to the target profile
4156 spin_lock(&fs_info
->balance_lock
);
4157 target
= get_restripe_target(fs_info
, flags
);
4159 /* pick target profile only if it's already available */
4160 if ((flags
& target
) & BTRFS_EXTENDED_PROFILE_MASK
) {
4161 spin_unlock(&fs_info
->balance_lock
);
4162 return extended_to_chunk(target
);
4165 spin_unlock(&fs_info
->balance_lock
);
4167 /* First, mask out the RAID levels which aren't possible */
4168 for (raid_type
= 0; raid_type
< BTRFS_NR_RAID_TYPES
; raid_type
++) {
4169 if (num_devices
>= btrfs_raid_array
[raid_type
].devs_min
)
4170 allowed
|= btrfs_raid_group
[raid_type
];
4174 if (allowed
& BTRFS_BLOCK_GROUP_RAID6
)
4175 allowed
= BTRFS_BLOCK_GROUP_RAID6
;
4176 else if (allowed
& BTRFS_BLOCK_GROUP_RAID5
)
4177 allowed
= BTRFS_BLOCK_GROUP_RAID5
;
4178 else if (allowed
& BTRFS_BLOCK_GROUP_RAID10
)
4179 allowed
= BTRFS_BLOCK_GROUP_RAID10
;
4180 else if (allowed
& BTRFS_BLOCK_GROUP_RAID1
)
4181 allowed
= BTRFS_BLOCK_GROUP_RAID1
;
4182 else if (allowed
& BTRFS_BLOCK_GROUP_RAID0
)
4183 allowed
= BTRFS_BLOCK_GROUP_RAID0
;
4185 flags
&= ~BTRFS_BLOCK_GROUP_PROFILE_MASK
;
4187 return extended_to_chunk(flags
| allowed
);
4190 static u64
get_alloc_profile(struct btrfs_fs_info
*fs_info
, u64 orig_flags
)
4197 seq
= read_seqbegin(&fs_info
->profiles_lock
);
4199 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
4200 flags
|= fs_info
->avail_data_alloc_bits
;
4201 else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
4202 flags
|= fs_info
->avail_system_alloc_bits
;
4203 else if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
4204 flags
|= fs_info
->avail_metadata_alloc_bits
;
4205 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
4207 return btrfs_reduce_alloc_profile(fs_info
, flags
);
4210 static u64
get_alloc_profile_by_root(struct btrfs_root
*root
, int data
)
4212 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4217 flags
= BTRFS_BLOCK_GROUP_DATA
;
4218 else if (root
== fs_info
->chunk_root
)
4219 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
4221 flags
= BTRFS_BLOCK_GROUP_METADATA
;
4223 ret
= get_alloc_profile(fs_info
, flags
);
4227 u64
btrfs_data_alloc_profile(struct btrfs_fs_info
*fs_info
)
4229 return get_alloc_profile(fs_info
, BTRFS_BLOCK_GROUP_DATA
);
4232 u64
btrfs_metadata_alloc_profile(struct btrfs_fs_info
*fs_info
)
4234 return get_alloc_profile(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4237 u64
btrfs_system_alloc_profile(struct btrfs_fs_info
*fs_info
)
4239 return get_alloc_profile(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
4242 static u64
btrfs_space_info_used(struct btrfs_space_info
*s_info
,
4243 bool may_use_included
)
4246 return s_info
->bytes_used
+ s_info
->bytes_reserved
+
4247 s_info
->bytes_pinned
+ s_info
->bytes_readonly
+
4248 (may_use_included
? s_info
->bytes_may_use
: 0);
4251 int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode
*inode
, u64 bytes
)
4253 struct btrfs_root
*root
= inode
->root
;
4254 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4255 struct btrfs_space_info
*data_sinfo
= fs_info
->data_sinfo
;
4258 int need_commit
= 2;
4259 int have_pinned_space
;
4261 /* make sure bytes are sectorsize aligned */
4262 bytes
= ALIGN(bytes
, fs_info
->sectorsize
);
4264 if (btrfs_is_free_space_inode(inode
)) {
4266 ASSERT(current
->journal_info
);
4270 /* make sure we have enough space to handle the data first */
4271 spin_lock(&data_sinfo
->lock
);
4272 used
= btrfs_space_info_used(data_sinfo
, true);
4274 if (used
+ bytes
> data_sinfo
->total_bytes
) {
4275 struct btrfs_trans_handle
*trans
;
4278 * if we don't have enough free bytes in this space then we need
4279 * to alloc a new chunk.
4281 if (!data_sinfo
->full
) {
4284 data_sinfo
->force_alloc
= CHUNK_ALLOC_FORCE
;
4285 spin_unlock(&data_sinfo
->lock
);
4287 alloc_target
= btrfs_data_alloc_profile(fs_info
);
4289 * It is ugly that we don't call nolock join
4290 * transaction for the free space inode case here.
4291 * But it is safe because we only do the data space
4292 * reservation for the free space cache in the
4293 * transaction context, the common join transaction
4294 * just increase the counter of the current transaction
4295 * handler, doesn't try to acquire the trans_lock of
4298 trans
= btrfs_join_transaction(root
);
4300 return PTR_ERR(trans
);
4302 ret
= do_chunk_alloc(trans
, fs_info
, alloc_target
,
4303 CHUNK_ALLOC_NO_FORCE
);
4304 btrfs_end_transaction(trans
);
4309 have_pinned_space
= 1;
4318 * If we don't have enough pinned space to deal with this
4319 * allocation, and no removed chunk in current transaction,
4320 * don't bother committing the transaction.
4322 have_pinned_space
= percpu_counter_compare(
4323 &data_sinfo
->total_bytes_pinned
,
4324 used
+ bytes
- data_sinfo
->total_bytes
);
4325 spin_unlock(&data_sinfo
->lock
);
4327 /* commit the current transaction and try again */
4330 !atomic_read(&fs_info
->open_ioctl_trans
)) {
4333 if (need_commit
> 0) {
4334 btrfs_start_delalloc_roots(fs_info
, 0, -1);
4335 btrfs_wait_ordered_roots(fs_info
, U64_MAX
, 0,
4339 trans
= btrfs_join_transaction(root
);
4341 return PTR_ERR(trans
);
4342 if (have_pinned_space
>= 0 ||
4343 test_bit(BTRFS_TRANS_HAVE_FREE_BGS
,
4344 &trans
->transaction
->flags
) ||
4346 ret
= btrfs_commit_transaction(trans
);
4350 * The cleaner kthread might still be doing iput
4351 * operations. Wait for it to finish so that
4352 * more space is released.
4354 mutex_lock(&fs_info
->cleaner_delayed_iput_mutex
);
4355 mutex_unlock(&fs_info
->cleaner_delayed_iput_mutex
);
4358 btrfs_end_transaction(trans
);
4362 trace_btrfs_space_reservation(fs_info
,
4363 "space_info:enospc",
4364 data_sinfo
->flags
, bytes
, 1);
4367 data_sinfo
->bytes_may_use
+= bytes
;
4368 trace_btrfs_space_reservation(fs_info
, "space_info",
4369 data_sinfo
->flags
, bytes
, 1);
4370 spin_unlock(&data_sinfo
->lock
);
4375 int btrfs_check_data_free_space(struct inode
*inode
,
4376 struct extent_changeset
**reserved
, u64 start
, u64 len
)
4378 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
4381 /* align the range */
4382 len
= round_up(start
+ len
, fs_info
->sectorsize
) -
4383 round_down(start
, fs_info
->sectorsize
);
4384 start
= round_down(start
, fs_info
->sectorsize
);
4386 ret
= btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode
), len
);
4390 /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */
4391 ret
= btrfs_qgroup_reserve_data(inode
, reserved
, start
, len
);
4393 btrfs_free_reserved_data_space_noquota(inode
, start
, len
);
4400 * Called if we need to clear a data reservation for this inode
4401 * Normally in a error case.
4403 * This one will *NOT* use accurate qgroup reserved space API, just for case
4404 * which we can't sleep and is sure it won't affect qgroup reserved space.
4405 * Like clear_bit_hook().
4407 void btrfs_free_reserved_data_space_noquota(struct inode
*inode
, u64 start
,
4410 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
4411 struct btrfs_space_info
*data_sinfo
;
4413 /* Make sure the range is aligned to sectorsize */
4414 len
= round_up(start
+ len
, fs_info
->sectorsize
) -
4415 round_down(start
, fs_info
->sectorsize
);
4416 start
= round_down(start
, fs_info
->sectorsize
);
4418 data_sinfo
= fs_info
->data_sinfo
;
4419 spin_lock(&data_sinfo
->lock
);
4420 if (WARN_ON(data_sinfo
->bytes_may_use
< len
))
4421 data_sinfo
->bytes_may_use
= 0;
4423 data_sinfo
->bytes_may_use
-= len
;
4424 trace_btrfs_space_reservation(fs_info
, "space_info",
4425 data_sinfo
->flags
, len
, 0);
4426 spin_unlock(&data_sinfo
->lock
);
4430 * Called if we need to clear a data reservation for this inode
4431 * Normally in a error case.
4433 * This one will handle the per-inode data rsv map for accurate reserved
4436 void btrfs_free_reserved_data_space(struct inode
*inode
,
4437 struct extent_changeset
*reserved
, u64 start
, u64 len
)
4439 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4441 /* Make sure the range is aligned to sectorsize */
4442 len
= round_up(start
+ len
, root
->fs_info
->sectorsize
) -
4443 round_down(start
, root
->fs_info
->sectorsize
);
4444 start
= round_down(start
, root
->fs_info
->sectorsize
);
4446 btrfs_free_reserved_data_space_noquota(inode
, start
, len
);
4447 btrfs_qgroup_free_data(inode
, reserved
, start
, len
);
4450 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
4452 struct list_head
*head
= &info
->space_info
;
4453 struct btrfs_space_info
*found
;
4456 list_for_each_entry_rcu(found
, head
, list
) {
4457 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
4458 found
->force_alloc
= CHUNK_ALLOC_FORCE
;
4463 static inline u64
calc_global_rsv_need_space(struct btrfs_block_rsv
*global
)
4465 return (global
->size
<< 1);
4468 static int should_alloc_chunk(struct btrfs_fs_info
*fs_info
,
4469 struct btrfs_space_info
*sinfo
, int force
)
4471 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
4472 u64 bytes_used
= btrfs_space_info_used(sinfo
, false);
4475 if (force
== CHUNK_ALLOC_FORCE
)
4479 * We need to take into account the global rsv because for all intents
4480 * and purposes it's used space. Don't worry about locking the
4481 * global_rsv, it doesn't change except when the transaction commits.
4483 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
4484 bytes_used
+= calc_global_rsv_need_space(global_rsv
);
4487 * in limited mode, we want to have some free space up to
4488 * about 1% of the FS size.
4490 if (force
== CHUNK_ALLOC_LIMITED
) {
4491 thresh
= btrfs_super_total_bytes(fs_info
->super_copy
);
4492 thresh
= max_t(u64
, SZ_64M
, div_factor_fine(thresh
, 1));
4494 if (sinfo
->total_bytes
- bytes_used
< thresh
)
4498 if (bytes_used
+ SZ_2M
< div_factor(sinfo
->total_bytes
, 8))
4503 static u64
get_profile_num_devs(struct btrfs_fs_info
*fs_info
, u64 type
)
4507 if (type
& (BTRFS_BLOCK_GROUP_RAID10
|
4508 BTRFS_BLOCK_GROUP_RAID0
|
4509 BTRFS_BLOCK_GROUP_RAID5
|
4510 BTRFS_BLOCK_GROUP_RAID6
))
4511 num_dev
= fs_info
->fs_devices
->rw_devices
;
4512 else if (type
& BTRFS_BLOCK_GROUP_RAID1
)
4515 num_dev
= 1; /* DUP or single */
4521 * If @is_allocation is true, reserve space in the system space info necessary
4522 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4525 void check_system_chunk(struct btrfs_trans_handle
*trans
,
4526 struct btrfs_fs_info
*fs_info
, u64 type
)
4528 struct btrfs_space_info
*info
;
4535 * Needed because we can end up allocating a system chunk and for an
4536 * atomic and race free space reservation in the chunk block reserve.
4538 ASSERT(mutex_is_locked(&fs_info
->chunk_mutex
));
4540 info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
4541 spin_lock(&info
->lock
);
4542 left
= info
->total_bytes
- btrfs_space_info_used(info
, true);
4543 spin_unlock(&info
->lock
);
4545 num_devs
= get_profile_num_devs(fs_info
, type
);
4547 /* num_devs device items to update and 1 chunk item to add or remove */
4548 thresh
= btrfs_calc_trunc_metadata_size(fs_info
, num_devs
) +
4549 btrfs_calc_trans_metadata_size(fs_info
, 1);
4551 if (left
< thresh
&& btrfs_test_opt(fs_info
, ENOSPC_DEBUG
)) {
4552 btrfs_info(fs_info
, "left=%llu, need=%llu, flags=%llu",
4553 left
, thresh
, type
);
4554 dump_space_info(fs_info
, info
, 0, 0);
4557 if (left
< thresh
) {
4558 u64 flags
= btrfs_system_alloc_profile(fs_info
);
4561 * Ignore failure to create system chunk. We might end up not
4562 * needing it, as we might not need to COW all nodes/leafs from
4563 * the paths we visit in the chunk tree (they were already COWed
4564 * or created in the current transaction for example).
4566 ret
= btrfs_alloc_chunk(trans
, fs_info
, flags
);
4570 ret
= btrfs_block_rsv_add(fs_info
->chunk_root
,
4571 &fs_info
->chunk_block_rsv
,
4572 thresh
, BTRFS_RESERVE_NO_FLUSH
);
4574 trans
->chunk_bytes_reserved
+= thresh
;
4579 * If force is CHUNK_ALLOC_FORCE:
4580 * - return 1 if it successfully allocates a chunk,
4581 * - return errors including -ENOSPC otherwise.
4582 * If force is NOT CHUNK_ALLOC_FORCE:
4583 * - return 0 if it doesn't need to allocate a new chunk,
4584 * - return 1 if it successfully allocates a chunk,
4585 * - return errors including -ENOSPC otherwise.
4587 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
4588 struct btrfs_fs_info
*fs_info
, u64 flags
, int force
)
4590 struct btrfs_space_info
*space_info
;
4591 int wait_for_alloc
= 0;
4594 /* Don't re-enter if we're already allocating a chunk */
4595 if (trans
->allocating_chunk
)
4598 space_info
= __find_space_info(fs_info
, flags
);
4600 ret
= create_space_info(fs_info
, flags
, &space_info
);
4606 spin_lock(&space_info
->lock
);
4607 if (force
< space_info
->force_alloc
)
4608 force
= space_info
->force_alloc
;
4609 if (space_info
->full
) {
4610 if (should_alloc_chunk(fs_info
, space_info
, force
))
4614 spin_unlock(&space_info
->lock
);
4618 if (!should_alloc_chunk(fs_info
, space_info
, force
)) {
4619 spin_unlock(&space_info
->lock
);
4621 } else if (space_info
->chunk_alloc
) {
4624 space_info
->chunk_alloc
= 1;
4627 spin_unlock(&space_info
->lock
);
4629 mutex_lock(&fs_info
->chunk_mutex
);
4632 * The chunk_mutex is held throughout the entirety of a chunk
4633 * allocation, so once we've acquired the chunk_mutex we know that the
4634 * other guy is done and we need to recheck and see if we should
4637 if (wait_for_alloc
) {
4638 mutex_unlock(&fs_info
->chunk_mutex
);
4643 trans
->allocating_chunk
= true;
4646 * If we have mixed data/metadata chunks we want to make sure we keep
4647 * allocating mixed chunks instead of individual chunks.
4649 if (btrfs_mixed_space_info(space_info
))
4650 flags
|= (BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
);
4653 * if we're doing a data chunk, go ahead and make sure that
4654 * we keep a reasonable number of metadata chunks allocated in the
4657 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
4658 fs_info
->data_chunk_allocations
++;
4659 if (!(fs_info
->data_chunk_allocations
%
4660 fs_info
->metadata_ratio
))
4661 force_metadata_allocation(fs_info
);
4665 * Check if we have enough space in SYSTEM chunk because we may need
4666 * to update devices.
4668 check_system_chunk(trans
, fs_info
, flags
);
4670 ret
= btrfs_alloc_chunk(trans
, fs_info
, flags
);
4671 trans
->allocating_chunk
= false;
4673 spin_lock(&space_info
->lock
);
4674 if (ret
< 0 && ret
!= -ENOSPC
)
4677 space_info
->full
= 1;
4681 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
4683 space_info
->chunk_alloc
= 0;
4684 spin_unlock(&space_info
->lock
);
4685 mutex_unlock(&fs_info
->chunk_mutex
);
4687 * When we allocate a new chunk we reserve space in the chunk block
4688 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4689 * add new nodes/leafs to it if we end up needing to do it when
4690 * inserting the chunk item and updating device items as part of the
4691 * second phase of chunk allocation, performed by
4692 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4693 * large number of new block groups to create in our transaction
4694 * handle's new_bgs list to avoid exhausting the chunk block reserve
4695 * in extreme cases - like having a single transaction create many new
4696 * block groups when starting to write out the free space caches of all
4697 * the block groups that were made dirty during the lifetime of the
4700 if (trans
->can_flush_pending_bgs
&&
4701 trans
->chunk_bytes_reserved
>= (u64
)SZ_2M
) {
4702 btrfs_create_pending_block_groups(trans
, fs_info
);
4703 btrfs_trans_release_chunk_metadata(trans
);
4708 static int can_overcommit(struct btrfs_fs_info
*fs_info
,
4709 struct btrfs_space_info
*space_info
, u64 bytes
,
4710 enum btrfs_reserve_flush_enum flush
,
4713 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
4719 /* Don't overcommit when in mixed mode. */
4720 if (space_info
->flags
& BTRFS_BLOCK_GROUP_DATA
)
4724 profile
= btrfs_system_alloc_profile(fs_info
);
4726 profile
= btrfs_metadata_alloc_profile(fs_info
);
4728 used
= btrfs_space_info_used(space_info
, false);
4731 * We only want to allow over committing if we have lots of actual space
4732 * free, but if we don't have enough space to handle the global reserve
4733 * space then we could end up having a real enospc problem when trying
4734 * to allocate a chunk or some other such important allocation.
4736 spin_lock(&global_rsv
->lock
);
4737 space_size
= calc_global_rsv_need_space(global_rsv
);
4738 spin_unlock(&global_rsv
->lock
);
4739 if (used
+ space_size
>= space_info
->total_bytes
)
4742 used
+= space_info
->bytes_may_use
;
4744 avail
= atomic64_read(&fs_info
->free_chunk_space
);
4747 * If we have dup, raid1 or raid10 then only half of the free
4748 * space is actually useable. For raid56, the space info used
4749 * doesn't include the parity drive, so we don't have to
4752 if (profile
& (BTRFS_BLOCK_GROUP_DUP
|
4753 BTRFS_BLOCK_GROUP_RAID1
|
4754 BTRFS_BLOCK_GROUP_RAID10
))
4758 * If we aren't flushing all things, let us overcommit up to
4759 * 1/2th of the space. If we can flush, don't let us overcommit
4760 * too much, let it overcommit up to 1/8 of the space.
4762 if (flush
== BTRFS_RESERVE_FLUSH_ALL
)
4767 if (used
+ bytes
< space_info
->total_bytes
+ avail
)
4772 static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info
*fs_info
,
4773 unsigned long nr_pages
, int nr_items
)
4775 struct super_block
*sb
= fs_info
->sb
;
4777 if (down_read_trylock(&sb
->s_umount
)) {
4778 writeback_inodes_sb_nr(sb
, nr_pages
, WB_REASON_FS_FREE_SPACE
);
4779 up_read(&sb
->s_umount
);
4782 * We needn't worry the filesystem going from r/w to r/o though
4783 * we don't acquire ->s_umount mutex, because the filesystem
4784 * should guarantee the delalloc inodes list be empty after
4785 * the filesystem is readonly(all dirty pages are written to
4788 btrfs_start_delalloc_roots(fs_info
, 0, nr_items
);
4789 if (!current
->journal_info
)
4790 btrfs_wait_ordered_roots(fs_info
, nr_items
, 0, (u64
)-1);
4794 static inline u64
calc_reclaim_items_nr(struct btrfs_fs_info
*fs_info
,
4800 bytes
= btrfs_calc_trans_metadata_size(fs_info
, 1);
4801 nr
= div64_u64(to_reclaim
, bytes
);
4807 #define EXTENT_SIZE_PER_ITEM SZ_256K
4810 * shrink metadata reservation for delalloc
4812 static void shrink_delalloc(struct btrfs_fs_info
*fs_info
, u64 to_reclaim
,
4813 u64 orig
, bool wait_ordered
)
4815 struct btrfs_space_info
*space_info
;
4816 struct btrfs_trans_handle
*trans
;
4821 unsigned long nr_pages
;
4823 enum btrfs_reserve_flush_enum flush
;
4825 /* Calc the number of the pages we need flush for space reservation */
4826 items
= calc_reclaim_items_nr(fs_info
, to_reclaim
);
4827 to_reclaim
= items
* EXTENT_SIZE_PER_ITEM
;
4829 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4830 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4832 delalloc_bytes
= percpu_counter_sum_positive(
4833 &fs_info
->delalloc_bytes
);
4834 if (delalloc_bytes
== 0) {
4838 btrfs_wait_ordered_roots(fs_info
, items
, 0, (u64
)-1);
4843 while (delalloc_bytes
&& loops
< 3) {
4844 max_reclaim
= min(delalloc_bytes
, to_reclaim
);
4845 nr_pages
= max_reclaim
>> PAGE_SHIFT
;
4846 btrfs_writeback_inodes_sb_nr(fs_info
, nr_pages
, items
);
4848 * We need to wait for the async pages to actually start before
4851 max_reclaim
= atomic_read(&fs_info
->async_delalloc_pages
);
4855 if (max_reclaim
<= nr_pages
)
4858 max_reclaim
-= nr_pages
;
4860 wait_event(fs_info
->async_submit_wait
,
4861 atomic_read(&fs_info
->async_delalloc_pages
) <=
4865 flush
= BTRFS_RESERVE_FLUSH_ALL
;
4867 flush
= BTRFS_RESERVE_NO_FLUSH
;
4868 spin_lock(&space_info
->lock
);
4869 if (list_empty(&space_info
->tickets
) &&
4870 list_empty(&space_info
->priority_tickets
)) {
4871 spin_unlock(&space_info
->lock
);
4874 spin_unlock(&space_info
->lock
);
4877 if (wait_ordered
&& !trans
) {
4878 btrfs_wait_ordered_roots(fs_info
, items
, 0, (u64
)-1);
4880 time_left
= schedule_timeout_killable(1);
4884 delalloc_bytes
= percpu_counter_sum_positive(
4885 &fs_info
->delalloc_bytes
);
4889 struct reserve_ticket
{
4892 struct list_head list
;
4893 wait_queue_head_t wait
;
4897 * maybe_commit_transaction - possibly commit the transaction if its ok to
4898 * @root - the root we're allocating for
4899 * @bytes - the number of bytes we want to reserve
4900 * @force - force the commit
4902 * This will check to make sure that committing the transaction will actually
4903 * get us somewhere and then commit the transaction if it does. Otherwise it
4904 * will return -ENOSPC.
4906 static int may_commit_transaction(struct btrfs_fs_info
*fs_info
,
4907 struct btrfs_space_info
*space_info
)
4909 struct reserve_ticket
*ticket
= NULL
;
4910 struct btrfs_block_rsv
*delayed_rsv
= &fs_info
->delayed_block_rsv
;
4911 struct btrfs_trans_handle
*trans
;
4914 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4918 spin_lock(&space_info
->lock
);
4919 if (!list_empty(&space_info
->priority_tickets
))
4920 ticket
= list_first_entry(&space_info
->priority_tickets
,
4921 struct reserve_ticket
, list
);
4922 else if (!list_empty(&space_info
->tickets
))
4923 ticket
= list_first_entry(&space_info
->tickets
,
4924 struct reserve_ticket
, list
);
4925 bytes
= (ticket
) ? ticket
->bytes
: 0;
4926 spin_unlock(&space_info
->lock
);
4931 /* See if there is enough pinned space to make this reservation */
4932 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4937 * See if there is some space in the delayed insertion reservation for
4940 if (space_info
!= delayed_rsv
->space_info
)
4943 spin_lock(&delayed_rsv
->lock
);
4944 if (delayed_rsv
->size
> bytes
)
4947 bytes
-= delayed_rsv
->size
;
4948 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4950 spin_unlock(&delayed_rsv
->lock
);
4953 spin_unlock(&delayed_rsv
->lock
);
4956 trans
= btrfs_join_transaction(fs_info
->extent_root
);
4960 return btrfs_commit_transaction(trans
);
4964 * Try to flush some data based on policy set by @state. This is only advisory
4965 * and may fail for various reasons. The caller is supposed to examine the
4966 * state of @space_info to detect the outcome.
4968 static void flush_space(struct btrfs_fs_info
*fs_info
,
4969 struct btrfs_space_info
*space_info
, u64 num_bytes
,
4972 struct btrfs_root
*root
= fs_info
->extent_root
;
4973 struct btrfs_trans_handle
*trans
;
4978 case FLUSH_DELAYED_ITEMS_NR
:
4979 case FLUSH_DELAYED_ITEMS
:
4980 if (state
== FLUSH_DELAYED_ITEMS_NR
)
4981 nr
= calc_reclaim_items_nr(fs_info
, num_bytes
) * 2;
4985 trans
= btrfs_join_transaction(root
);
4986 if (IS_ERR(trans
)) {
4987 ret
= PTR_ERR(trans
);
4990 ret
= btrfs_run_delayed_items_nr(trans
, fs_info
, nr
);
4991 btrfs_end_transaction(trans
);
4993 case FLUSH_DELALLOC
:
4994 case FLUSH_DELALLOC_WAIT
:
4995 shrink_delalloc(fs_info
, num_bytes
* 2, num_bytes
,
4996 state
== FLUSH_DELALLOC_WAIT
);
4999 trans
= btrfs_join_transaction(root
);
5000 if (IS_ERR(trans
)) {
5001 ret
= PTR_ERR(trans
);
5004 ret
= do_chunk_alloc(trans
, fs_info
,
5005 btrfs_metadata_alloc_profile(fs_info
),
5006 CHUNK_ALLOC_NO_FORCE
);
5007 btrfs_end_transaction(trans
);
5008 if (ret
> 0 || ret
== -ENOSPC
)
5012 ret
= may_commit_transaction(fs_info
, space_info
);
5019 trace_btrfs_flush_space(fs_info
, space_info
->flags
, num_bytes
, state
,
5025 btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info
*fs_info
,
5026 struct btrfs_space_info
*space_info
,
5029 struct reserve_ticket
*ticket
;
5034 list_for_each_entry(ticket
, &space_info
->tickets
, list
)
5035 to_reclaim
+= ticket
->bytes
;
5036 list_for_each_entry(ticket
, &space_info
->priority_tickets
, list
)
5037 to_reclaim
+= ticket
->bytes
;
5041 to_reclaim
= min_t(u64
, num_online_cpus() * SZ_1M
, SZ_16M
);
5042 if (can_overcommit(fs_info
, space_info
, to_reclaim
,
5043 BTRFS_RESERVE_FLUSH_ALL
, system_chunk
))
5046 used
= btrfs_space_info_used(space_info
, true);
5048 if (can_overcommit(fs_info
, space_info
, SZ_1M
,
5049 BTRFS_RESERVE_FLUSH_ALL
, system_chunk
))
5050 expected
= div_factor_fine(space_info
->total_bytes
, 95);
5052 expected
= div_factor_fine(space_info
->total_bytes
, 90);
5054 if (used
> expected
)
5055 to_reclaim
= used
- expected
;
5058 to_reclaim
= min(to_reclaim
, space_info
->bytes_may_use
+
5059 space_info
->bytes_reserved
);
5063 static inline int need_do_async_reclaim(struct btrfs_fs_info
*fs_info
,
5064 struct btrfs_space_info
*space_info
,
5065 u64 used
, bool system_chunk
)
5067 u64 thresh
= div_factor_fine(space_info
->total_bytes
, 98);
5069 /* If we're just plain full then async reclaim just slows us down. */
5070 if ((space_info
->bytes_used
+ space_info
->bytes_reserved
) >= thresh
)
5073 if (!btrfs_calc_reclaim_metadata_size(fs_info
, space_info
,
5077 return (used
>= thresh
&& !btrfs_fs_closing(fs_info
) &&
5078 !test_bit(BTRFS_FS_STATE_REMOUNTING
, &fs_info
->fs_state
));
5081 static void wake_all_tickets(struct list_head
*head
)
5083 struct reserve_ticket
*ticket
;
5085 while (!list_empty(head
)) {
5086 ticket
= list_first_entry(head
, struct reserve_ticket
, list
);
5087 list_del_init(&ticket
->list
);
5088 ticket
->error
= -ENOSPC
;
5089 wake_up(&ticket
->wait
);
5094 * This is for normal flushers, we can wait all goddamned day if we want to. We
5095 * will loop and continuously try to flush as long as we are making progress.
5096 * We count progress as clearing off tickets each time we have to loop.
5098 static void btrfs_async_reclaim_metadata_space(struct work_struct
*work
)
5100 struct btrfs_fs_info
*fs_info
;
5101 struct btrfs_space_info
*space_info
;
5104 int commit_cycles
= 0;
5105 u64 last_tickets_id
;
5107 fs_info
= container_of(work
, struct btrfs_fs_info
, async_reclaim_work
);
5108 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
5110 spin_lock(&space_info
->lock
);
5111 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
, space_info
,
5114 space_info
->flush
= 0;
5115 spin_unlock(&space_info
->lock
);
5118 last_tickets_id
= space_info
->tickets_id
;
5119 spin_unlock(&space_info
->lock
);
5121 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
5123 flush_space(fs_info
, space_info
, to_reclaim
, flush_state
);
5124 spin_lock(&space_info
->lock
);
5125 if (list_empty(&space_info
->tickets
)) {
5126 space_info
->flush
= 0;
5127 spin_unlock(&space_info
->lock
);
5130 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
,
5133 if (last_tickets_id
== space_info
->tickets_id
) {
5136 last_tickets_id
= space_info
->tickets_id
;
5137 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
5142 if (flush_state
> COMMIT_TRANS
) {
5144 if (commit_cycles
> 2) {
5145 wake_all_tickets(&space_info
->tickets
);
5146 space_info
->flush
= 0;
5148 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
5151 spin_unlock(&space_info
->lock
);
5152 } while (flush_state
<= COMMIT_TRANS
);
5155 void btrfs_init_async_reclaim_work(struct work_struct
*work
)
5157 INIT_WORK(work
, btrfs_async_reclaim_metadata_space
);
5160 static void priority_reclaim_metadata_space(struct btrfs_fs_info
*fs_info
,
5161 struct btrfs_space_info
*space_info
,
5162 struct reserve_ticket
*ticket
)
5165 int flush_state
= FLUSH_DELAYED_ITEMS_NR
;
5167 spin_lock(&space_info
->lock
);
5168 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
, space_info
,
5171 spin_unlock(&space_info
->lock
);
5174 spin_unlock(&space_info
->lock
);
5177 flush_space(fs_info
, space_info
, to_reclaim
, flush_state
);
5179 spin_lock(&space_info
->lock
);
5180 if (ticket
->bytes
== 0) {
5181 spin_unlock(&space_info
->lock
);
5184 spin_unlock(&space_info
->lock
);
5187 * Priority flushers can't wait on delalloc without
5190 if (flush_state
== FLUSH_DELALLOC
||
5191 flush_state
== FLUSH_DELALLOC_WAIT
)
5192 flush_state
= ALLOC_CHUNK
;
5193 } while (flush_state
< COMMIT_TRANS
);
5196 static int wait_reserve_ticket(struct btrfs_fs_info
*fs_info
,
5197 struct btrfs_space_info
*space_info
,
5198 struct reserve_ticket
*ticket
, u64 orig_bytes
)
5204 spin_lock(&space_info
->lock
);
5205 while (ticket
->bytes
> 0 && ticket
->error
== 0) {
5206 ret
= prepare_to_wait_event(&ticket
->wait
, &wait
, TASK_KILLABLE
);
5211 spin_unlock(&space_info
->lock
);
5215 finish_wait(&ticket
->wait
, &wait
);
5216 spin_lock(&space_info
->lock
);
5219 ret
= ticket
->error
;
5220 if (!list_empty(&ticket
->list
))
5221 list_del_init(&ticket
->list
);
5222 if (ticket
->bytes
&& ticket
->bytes
< orig_bytes
) {
5223 u64 num_bytes
= orig_bytes
- ticket
->bytes
;
5224 space_info
->bytes_may_use
-= num_bytes
;
5225 trace_btrfs_space_reservation(fs_info
, "space_info",
5226 space_info
->flags
, num_bytes
, 0);
5228 spin_unlock(&space_info
->lock
);
5234 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5235 * @root - the root we're allocating for
5236 * @space_info - the space info we want to allocate from
5237 * @orig_bytes - the number of bytes we want
5238 * @flush - whether or not we can flush to make our reservation
5240 * This will reserve orig_bytes number of bytes from the space info associated
5241 * with the block_rsv. If there is not enough space it will make an attempt to
5242 * flush out space to make room. It will do this by flushing delalloc if
5243 * possible or committing the transaction. If flush is 0 then no attempts to
5244 * regain reservations will be made and this will fail if there is not enough
5247 static int __reserve_metadata_bytes(struct btrfs_fs_info
*fs_info
,
5248 struct btrfs_space_info
*space_info
,
5250 enum btrfs_reserve_flush_enum flush
,
5253 struct reserve_ticket ticket
;
5258 ASSERT(!current
->journal_info
|| flush
!= BTRFS_RESERVE_FLUSH_ALL
);
5260 spin_lock(&space_info
->lock
);
5262 used
= btrfs_space_info_used(space_info
, true);
5265 * If we have enough space then hooray, make our reservation and carry
5266 * on. If not see if we can overcommit, and if we can, hooray carry on.
5267 * If not things get more complicated.
5269 if (used
+ orig_bytes
<= space_info
->total_bytes
) {
5270 space_info
->bytes_may_use
+= orig_bytes
;
5271 trace_btrfs_space_reservation(fs_info
, "space_info",
5272 space_info
->flags
, orig_bytes
, 1);
5274 } else if (can_overcommit(fs_info
, space_info
, orig_bytes
, flush
,
5276 space_info
->bytes_may_use
+= orig_bytes
;
5277 trace_btrfs_space_reservation(fs_info
, "space_info",
5278 space_info
->flags
, orig_bytes
, 1);
5283 * If we couldn't make a reservation then setup our reservation ticket
5284 * and kick the async worker if it's not already running.
5286 * If we are a priority flusher then we just need to add our ticket to
5287 * the list and we will do our own flushing further down.
5289 if (ret
&& flush
!= BTRFS_RESERVE_NO_FLUSH
) {
5290 ticket
.bytes
= orig_bytes
;
5292 init_waitqueue_head(&ticket
.wait
);
5293 if (flush
== BTRFS_RESERVE_FLUSH_ALL
) {
5294 list_add_tail(&ticket
.list
, &space_info
->tickets
);
5295 if (!space_info
->flush
) {
5296 space_info
->flush
= 1;
5297 trace_btrfs_trigger_flush(fs_info
,
5301 queue_work(system_unbound_wq
,
5302 &fs_info
->async_reclaim_work
);
5305 list_add_tail(&ticket
.list
,
5306 &space_info
->priority_tickets
);
5308 } else if (!ret
&& space_info
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
5311 * We will do the space reservation dance during log replay,
5312 * which means we won't have fs_info->fs_root set, so don't do
5313 * the async reclaim as we will panic.
5315 if (!test_bit(BTRFS_FS_LOG_RECOVERING
, &fs_info
->flags
) &&
5316 need_do_async_reclaim(fs_info
, space_info
,
5317 used
, system_chunk
) &&
5318 !work_busy(&fs_info
->async_reclaim_work
)) {
5319 trace_btrfs_trigger_flush(fs_info
, space_info
->flags
,
5320 orig_bytes
, flush
, "preempt");
5321 queue_work(system_unbound_wq
,
5322 &fs_info
->async_reclaim_work
);
5325 spin_unlock(&space_info
->lock
);
5326 if (!ret
|| flush
== BTRFS_RESERVE_NO_FLUSH
)
5329 if (flush
== BTRFS_RESERVE_FLUSH_ALL
)
5330 return wait_reserve_ticket(fs_info
, space_info
, &ticket
,
5334 priority_reclaim_metadata_space(fs_info
, space_info
, &ticket
);
5335 spin_lock(&space_info
->lock
);
5337 if (ticket
.bytes
< orig_bytes
) {
5338 u64 num_bytes
= orig_bytes
- ticket
.bytes
;
5339 space_info
->bytes_may_use
-= num_bytes
;
5340 trace_btrfs_space_reservation(fs_info
, "space_info",
5345 list_del_init(&ticket
.list
);
5348 spin_unlock(&space_info
->lock
);
5349 ASSERT(list_empty(&ticket
.list
));
5354 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5355 * @root - the root we're allocating for
5356 * @block_rsv - the block_rsv we're allocating for
5357 * @orig_bytes - the number of bytes we want
5358 * @flush - whether or not we can flush to make our reservation
5360 * This will reserve orgi_bytes number of bytes from the space info associated
5361 * with the block_rsv. If there is not enough space it will make an attempt to
5362 * flush out space to make room. It will do this by flushing delalloc if
5363 * possible or committing the transaction. If flush is 0 then no attempts to
5364 * regain reservations will be made and this will fail if there is not enough
5367 static int reserve_metadata_bytes(struct btrfs_root
*root
,
5368 struct btrfs_block_rsv
*block_rsv
,
5370 enum btrfs_reserve_flush_enum flush
)
5372 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5373 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5375 bool system_chunk
= (root
== fs_info
->chunk_root
);
5377 ret
= __reserve_metadata_bytes(fs_info
, block_rsv
->space_info
,
5378 orig_bytes
, flush
, system_chunk
);
5379 if (ret
== -ENOSPC
&&
5380 unlikely(root
->orphan_cleanup_state
== ORPHAN_CLEANUP_STARTED
)) {
5381 if (block_rsv
!= global_rsv
&&
5382 !block_rsv_use_bytes(global_rsv
, orig_bytes
))
5386 trace_btrfs_space_reservation(fs_info
, "space_info:enospc",
5387 block_rsv
->space_info
->flags
,
5392 static struct btrfs_block_rsv
*get_block_rsv(
5393 const struct btrfs_trans_handle
*trans
,
5394 const struct btrfs_root
*root
)
5396 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5397 struct btrfs_block_rsv
*block_rsv
= NULL
;
5399 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
5400 (root
== fs_info
->csum_root
&& trans
->adding_csums
) ||
5401 (root
== fs_info
->uuid_root
))
5402 block_rsv
= trans
->block_rsv
;
5405 block_rsv
= root
->block_rsv
;
5408 block_rsv
= &fs_info
->empty_block_rsv
;
5413 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
5417 spin_lock(&block_rsv
->lock
);
5418 if (block_rsv
->reserved
>= num_bytes
) {
5419 block_rsv
->reserved
-= num_bytes
;
5420 if (block_rsv
->reserved
< block_rsv
->size
)
5421 block_rsv
->full
= 0;
5424 spin_unlock(&block_rsv
->lock
);
5428 static void block_rsv_add_bytes(struct btrfs_block_rsv
*block_rsv
,
5429 u64 num_bytes
, int update_size
)
5431 spin_lock(&block_rsv
->lock
);
5432 block_rsv
->reserved
+= num_bytes
;
5434 block_rsv
->size
+= num_bytes
;
5435 else if (block_rsv
->reserved
>= block_rsv
->size
)
5436 block_rsv
->full
= 1;
5437 spin_unlock(&block_rsv
->lock
);
5440 int btrfs_cond_migrate_bytes(struct btrfs_fs_info
*fs_info
,
5441 struct btrfs_block_rsv
*dest
, u64 num_bytes
,
5444 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5447 if (global_rsv
->space_info
!= dest
->space_info
)
5450 spin_lock(&global_rsv
->lock
);
5451 min_bytes
= div_factor(global_rsv
->size
, min_factor
);
5452 if (global_rsv
->reserved
< min_bytes
+ num_bytes
) {
5453 spin_unlock(&global_rsv
->lock
);
5456 global_rsv
->reserved
-= num_bytes
;
5457 if (global_rsv
->reserved
< global_rsv
->size
)
5458 global_rsv
->full
= 0;
5459 spin_unlock(&global_rsv
->lock
);
5461 block_rsv_add_bytes(dest
, num_bytes
, 1);
5466 * This is for space we already have accounted in space_info->bytes_may_use, so
5467 * basically when we're returning space from block_rsv's.
5469 static void space_info_add_old_bytes(struct btrfs_fs_info
*fs_info
,
5470 struct btrfs_space_info
*space_info
,
5473 struct reserve_ticket
*ticket
;
5474 struct list_head
*head
;
5476 enum btrfs_reserve_flush_enum flush
= BTRFS_RESERVE_NO_FLUSH
;
5477 bool check_overcommit
= false;
5479 spin_lock(&space_info
->lock
);
5480 head
= &space_info
->priority_tickets
;
5483 * If we are over our limit then we need to check and see if we can
5484 * overcommit, and if we can't then we just need to free up our space
5485 * and not satisfy any requests.
5487 used
= btrfs_space_info_used(space_info
, true);
5488 if (used
- num_bytes
>= space_info
->total_bytes
)
5489 check_overcommit
= true;
5491 while (!list_empty(head
) && num_bytes
) {
5492 ticket
= list_first_entry(head
, struct reserve_ticket
,
5495 * We use 0 bytes because this space is already reserved, so
5496 * adding the ticket space would be a double count.
5498 if (check_overcommit
&&
5499 !can_overcommit(fs_info
, space_info
, 0, flush
, false))
5501 if (num_bytes
>= ticket
->bytes
) {
5502 list_del_init(&ticket
->list
);
5503 num_bytes
-= ticket
->bytes
;
5505 space_info
->tickets_id
++;
5506 wake_up(&ticket
->wait
);
5508 ticket
->bytes
-= num_bytes
;
5513 if (num_bytes
&& head
== &space_info
->priority_tickets
) {
5514 head
= &space_info
->tickets
;
5515 flush
= BTRFS_RESERVE_FLUSH_ALL
;
5518 space_info
->bytes_may_use
-= num_bytes
;
5519 trace_btrfs_space_reservation(fs_info
, "space_info",
5520 space_info
->flags
, num_bytes
, 0);
5521 spin_unlock(&space_info
->lock
);
5525 * This is for newly allocated space that isn't accounted in
5526 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5527 * we use this helper.
5529 static void space_info_add_new_bytes(struct btrfs_fs_info
*fs_info
,
5530 struct btrfs_space_info
*space_info
,
5533 struct reserve_ticket
*ticket
;
5534 struct list_head
*head
= &space_info
->priority_tickets
;
5537 while (!list_empty(head
) && num_bytes
) {
5538 ticket
= list_first_entry(head
, struct reserve_ticket
,
5540 if (num_bytes
>= ticket
->bytes
) {
5541 trace_btrfs_space_reservation(fs_info
, "space_info",
5544 list_del_init(&ticket
->list
);
5545 num_bytes
-= ticket
->bytes
;
5546 space_info
->bytes_may_use
+= ticket
->bytes
;
5548 space_info
->tickets_id
++;
5549 wake_up(&ticket
->wait
);
5551 trace_btrfs_space_reservation(fs_info
, "space_info",
5554 space_info
->bytes_may_use
+= num_bytes
;
5555 ticket
->bytes
-= num_bytes
;
5560 if (num_bytes
&& head
== &space_info
->priority_tickets
) {
5561 head
= &space_info
->tickets
;
5566 static u64
block_rsv_release_bytes(struct btrfs_fs_info
*fs_info
,
5567 struct btrfs_block_rsv
*block_rsv
,
5568 struct btrfs_block_rsv
*dest
, u64 num_bytes
)
5570 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
5573 spin_lock(&block_rsv
->lock
);
5574 if (num_bytes
== (u64
)-1)
5575 num_bytes
= block_rsv
->size
;
5576 block_rsv
->size
-= num_bytes
;
5577 if (block_rsv
->reserved
>= block_rsv
->size
) {
5578 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
5579 block_rsv
->reserved
= block_rsv
->size
;
5580 block_rsv
->full
= 1;
5584 spin_unlock(&block_rsv
->lock
);
5587 if (num_bytes
> 0) {
5589 spin_lock(&dest
->lock
);
5593 bytes_to_add
= dest
->size
- dest
->reserved
;
5594 bytes_to_add
= min(num_bytes
, bytes_to_add
);
5595 dest
->reserved
+= bytes_to_add
;
5596 if (dest
->reserved
>= dest
->size
)
5598 num_bytes
-= bytes_to_add
;
5600 spin_unlock(&dest
->lock
);
5603 space_info_add_old_bytes(fs_info
, space_info
,
5609 int btrfs_block_rsv_migrate(struct btrfs_block_rsv
*src
,
5610 struct btrfs_block_rsv
*dst
, u64 num_bytes
,
5615 ret
= block_rsv_use_bytes(src
, num_bytes
);
5619 block_rsv_add_bytes(dst
, num_bytes
, update_size
);
5623 void btrfs_init_block_rsv(struct btrfs_block_rsv
*rsv
, unsigned short type
)
5625 memset(rsv
, 0, sizeof(*rsv
));
5626 spin_lock_init(&rsv
->lock
);
5630 void btrfs_init_metadata_block_rsv(struct btrfs_fs_info
*fs_info
,
5631 struct btrfs_block_rsv
*rsv
,
5632 unsigned short type
)
5634 btrfs_init_block_rsv(rsv
, type
);
5635 rsv
->space_info
= __find_space_info(fs_info
,
5636 BTRFS_BLOCK_GROUP_METADATA
);
5639 struct btrfs_block_rsv
*btrfs_alloc_block_rsv(struct btrfs_fs_info
*fs_info
,
5640 unsigned short type
)
5642 struct btrfs_block_rsv
*block_rsv
;
5644 block_rsv
= kmalloc(sizeof(*block_rsv
), GFP_NOFS
);
5648 btrfs_init_metadata_block_rsv(fs_info
, block_rsv
, type
);
5652 void btrfs_free_block_rsv(struct btrfs_fs_info
*fs_info
,
5653 struct btrfs_block_rsv
*rsv
)
5657 btrfs_block_rsv_release(fs_info
, rsv
, (u64
)-1);
5661 void __btrfs_free_block_rsv(struct btrfs_block_rsv
*rsv
)
5666 int btrfs_block_rsv_add(struct btrfs_root
*root
,
5667 struct btrfs_block_rsv
*block_rsv
, u64 num_bytes
,
5668 enum btrfs_reserve_flush_enum flush
)
5675 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
5677 block_rsv_add_bytes(block_rsv
, num_bytes
, 1);
5684 int btrfs_block_rsv_check(struct btrfs_block_rsv
*block_rsv
, int min_factor
)
5692 spin_lock(&block_rsv
->lock
);
5693 num_bytes
= div_factor(block_rsv
->size
, min_factor
);
5694 if (block_rsv
->reserved
>= num_bytes
)
5696 spin_unlock(&block_rsv
->lock
);
5701 int btrfs_block_rsv_refill(struct btrfs_root
*root
,
5702 struct btrfs_block_rsv
*block_rsv
, u64 min_reserved
,
5703 enum btrfs_reserve_flush_enum flush
)
5711 spin_lock(&block_rsv
->lock
);
5712 num_bytes
= min_reserved
;
5713 if (block_rsv
->reserved
>= num_bytes
)
5716 num_bytes
-= block_rsv
->reserved
;
5717 spin_unlock(&block_rsv
->lock
);
5722 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
5724 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
5732 * btrfs_inode_rsv_refill - refill the inode block rsv.
5733 * @inode - the inode we are refilling.
5734 * @flush - the flusing restriction.
5736 * Essentially the same as btrfs_block_rsv_refill, except it uses the
5737 * block_rsv->size as the minimum size. We'll either refill the missing amount
5738 * or return if we already have enough space. This will also handle the resreve
5739 * tracepoint for the reserved amount.
5741 int btrfs_inode_rsv_refill(struct btrfs_inode
*inode
,
5742 enum btrfs_reserve_flush_enum flush
)
5744 struct btrfs_root
*root
= inode
->root
;
5745 struct btrfs_block_rsv
*block_rsv
= &inode
->block_rsv
;
5749 spin_lock(&block_rsv
->lock
);
5750 if (block_rsv
->reserved
< block_rsv
->size
)
5751 num_bytes
= block_rsv
->size
- block_rsv
->reserved
;
5752 spin_unlock(&block_rsv
->lock
);
5757 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
5759 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
5760 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5761 btrfs_ino(inode
), num_bytes
, 1);
5767 * btrfs_inode_rsv_release - release any excessive reservation.
5768 * @inode - the inode we need to release from.
5770 * This is the same as btrfs_block_rsv_release, except that it handles the
5771 * tracepoint for the reservation.
5773 void btrfs_inode_rsv_release(struct btrfs_inode
*inode
)
5775 struct btrfs_fs_info
*fs_info
= inode
->root
->fs_info
;
5776 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5777 struct btrfs_block_rsv
*block_rsv
= &inode
->block_rsv
;
5781 * Since we statically set the block_rsv->size we just want to say we
5782 * are releasing 0 bytes, and then we'll just get the reservation over
5785 released
= block_rsv_release_bytes(fs_info
, block_rsv
, global_rsv
, 0);
5787 trace_btrfs_space_reservation(fs_info
, "delalloc",
5788 btrfs_ino(inode
), released
, 0);
5791 void btrfs_block_rsv_release(struct btrfs_fs_info
*fs_info
,
5792 struct btrfs_block_rsv
*block_rsv
,
5795 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5797 if (global_rsv
== block_rsv
||
5798 block_rsv
->space_info
!= global_rsv
->space_info
)
5800 block_rsv_release_bytes(fs_info
, block_rsv
, global_rsv
, num_bytes
);
5803 static void update_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5805 struct btrfs_block_rsv
*block_rsv
= &fs_info
->global_block_rsv
;
5806 struct btrfs_space_info
*sinfo
= block_rsv
->space_info
;
5810 * The global block rsv is based on the size of the extent tree, the
5811 * checksum tree and the root tree. If the fs is empty we want to set
5812 * it to a minimal amount for safety.
5814 num_bytes
= btrfs_root_used(&fs_info
->extent_root
->root_item
) +
5815 btrfs_root_used(&fs_info
->csum_root
->root_item
) +
5816 btrfs_root_used(&fs_info
->tree_root
->root_item
);
5817 num_bytes
= max_t(u64
, num_bytes
, SZ_16M
);
5819 spin_lock(&sinfo
->lock
);
5820 spin_lock(&block_rsv
->lock
);
5822 block_rsv
->size
= min_t(u64
, num_bytes
, SZ_512M
);
5824 if (block_rsv
->reserved
< block_rsv
->size
) {
5825 num_bytes
= btrfs_space_info_used(sinfo
, true);
5826 if (sinfo
->total_bytes
> num_bytes
) {
5827 num_bytes
= sinfo
->total_bytes
- num_bytes
;
5828 num_bytes
= min(num_bytes
,
5829 block_rsv
->size
- block_rsv
->reserved
);
5830 block_rsv
->reserved
+= num_bytes
;
5831 sinfo
->bytes_may_use
+= num_bytes
;
5832 trace_btrfs_space_reservation(fs_info
, "space_info",
5833 sinfo
->flags
, num_bytes
,
5836 } else if (block_rsv
->reserved
> block_rsv
->size
) {
5837 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
5838 sinfo
->bytes_may_use
-= num_bytes
;
5839 trace_btrfs_space_reservation(fs_info
, "space_info",
5840 sinfo
->flags
, num_bytes
, 0);
5841 block_rsv
->reserved
= block_rsv
->size
;
5844 if (block_rsv
->reserved
== block_rsv
->size
)
5845 block_rsv
->full
= 1;
5847 block_rsv
->full
= 0;
5849 spin_unlock(&block_rsv
->lock
);
5850 spin_unlock(&sinfo
->lock
);
5853 static void init_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5855 struct btrfs_space_info
*space_info
;
5857 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
5858 fs_info
->chunk_block_rsv
.space_info
= space_info
;
5860 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
5861 fs_info
->global_block_rsv
.space_info
= space_info
;
5862 fs_info
->trans_block_rsv
.space_info
= space_info
;
5863 fs_info
->empty_block_rsv
.space_info
= space_info
;
5864 fs_info
->delayed_block_rsv
.space_info
= space_info
;
5866 fs_info
->extent_root
->block_rsv
= &fs_info
->global_block_rsv
;
5867 fs_info
->csum_root
->block_rsv
= &fs_info
->global_block_rsv
;
5868 fs_info
->dev_root
->block_rsv
= &fs_info
->global_block_rsv
;
5869 fs_info
->tree_root
->block_rsv
= &fs_info
->global_block_rsv
;
5870 if (fs_info
->quota_root
)
5871 fs_info
->quota_root
->block_rsv
= &fs_info
->global_block_rsv
;
5872 fs_info
->chunk_root
->block_rsv
= &fs_info
->chunk_block_rsv
;
5874 update_global_block_rsv(fs_info
);
5877 static void release_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5879 block_rsv_release_bytes(fs_info
, &fs_info
->global_block_rsv
, NULL
,
5881 WARN_ON(fs_info
->trans_block_rsv
.size
> 0);
5882 WARN_ON(fs_info
->trans_block_rsv
.reserved
> 0);
5883 WARN_ON(fs_info
->chunk_block_rsv
.size
> 0);
5884 WARN_ON(fs_info
->chunk_block_rsv
.reserved
> 0);
5885 WARN_ON(fs_info
->delayed_block_rsv
.size
> 0);
5886 WARN_ON(fs_info
->delayed_block_rsv
.reserved
> 0);
5889 void btrfs_trans_release_metadata(struct btrfs_trans_handle
*trans
,
5890 struct btrfs_fs_info
*fs_info
)
5892 if (!trans
->block_rsv
) {
5893 ASSERT(!trans
->bytes_reserved
);
5897 if (!trans
->bytes_reserved
)
5900 ASSERT(trans
->block_rsv
== &fs_info
->trans_block_rsv
);
5901 trace_btrfs_space_reservation(fs_info
, "transaction",
5902 trans
->transid
, trans
->bytes_reserved
, 0);
5903 btrfs_block_rsv_release(fs_info
, trans
->block_rsv
,
5904 trans
->bytes_reserved
);
5905 trans
->bytes_reserved
= 0;
5909 * To be called after all the new block groups attached to the transaction
5910 * handle have been created (btrfs_create_pending_block_groups()).
5912 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle
*trans
)
5914 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
5916 if (!trans
->chunk_bytes_reserved
)
5919 WARN_ON_ONCE(!list_empty(&trans
->new_bgs
));
5921 block_rsv_release_bytes(fs_info
, &fs_info
->chunk_block_rsv
, NULL
,
5922 trans
->chunk_bytes_reserved
);
5923 trans
->chunk_bytes_reserved
= 0;
5926 /* Can only return 0 or -ENOSPC */
5927 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle
*trans
,
5928 struct btrfs_inode
*inode
)
5930 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
5931 struct btrfs_root
*root
= inode
->root
;
5933 * We always use trans->block_rsv here as we will have reserved space
5934 * for our orphan when starting the transaction, using get_block_rsv()
5935 * here will sometimes make us choose the wrong block rsv as we could be
5936 * doing a reloc inode for a non refcounted root.
5938 struct btrfs_block_rsv
*src_rsv
= trans
->block_rsv
;
5939 struct btrfs_block_rsv
*dst_rsv
= root
->orphan_block_rsv
;
5942 * We need to hold space in order to delete our orphan item once we've
5943 * added it, so this takes the reservation so we can release it later
5944 * when we are truly done with the orphan item.
5946 u64 num_bytes
= btrfs_calc_trans_metadata_size(fs_info
, 1);
5948 trace_btrfs_space_reservation(fs_info
, "orphan", btrfs_ino(inode
),
5950 return btrfs_block_rsv_migrate(src_rsv
, dst_rsv
, num_bytes
, 1);
5953 void btrfs_orphan_release_metadata(struct btrfs_inode
*inode
)
5955 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
5956 struct btrfs_root
*root
= inode
->root
;
5957 u64 num_bytes
= btrfs_calc_trans_metadata_size(fs_info
, 1);
5959 trace_btrfs_space_reservation(fs_info
, "orphan", btrfs_ino(inode
),
5961 btrfs_block_rsv_release(fs_info
, root
->orphan_block_rsv
, num_bytes
);
5965 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5966 * root: the root of the parent directory
5967 * rsv: block reservation
5968 * items: the number of items that we need do reservation
5969 * qgroup_reserved: used to return the reserved size in qgroup
5971 * This function is used to reserve the space for snapshot/subvolume
5972 * creation and deletion. Those operations are different with the
5973 * common file/directory operations, they change two fs/file trees
5974 * and root tree, the number of items that the qgroup reserves is
5975 * different with the free space reservation. So we can not use
5976 * the space reservation mechanism in start_transaction().
5978 int btrfs_subvolume_reserve_metadata(struct btrfs_root
*root
,
5979 struct btrfs_block_rsv
*rsv
,
5981 u64
*qgroup_reserved
,
5982 bool use_global_rsv
)
5986 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5987 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5989 if (test_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
)) {
5990 /* One for parent inode, two for dir entries */
5991 num_bytes
= 3 * fs_info
->nodesize
;
5992 ret
= btrfs_qgroup_reserve_meta(root
, num_bytes
, true);
5999 *qgroup_reserved
= num_bytes
;
6001 num_bytes
= btrfs_calc_trans_metadata_size(fs_info
, items
);
6002 rsv
->space_info
= __find_space_info(fs_info
,
6003 BTRFS_BLOCK_GROUP_METADATA
);
6004 ret
= btrfs_block_rsv_add(root
, rsv
, num_bytes
,
6005 BTRFS_RESERVE_FLUSH_ALL
);
6007 if (ret
== -ENOSPC
&& use_global_rsv
)
6008 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
, num_bytes
, 1);
6010 if (ret
&& *qgroup_reserved
)
6011 btrfs_qgroup_free_meta(root
, *qgroup_reserved
);
6016 void btrfs_subvolume_release_metadata(struct btrfs_fs_info
*fs_info
,
6017 struct btrfs_block_rsv
*rsv
)
6019 btrfs_block_rsv_release(fs_info
, rsv
, (u64
)-1);
6022 static void btrfs_calculate_inode_block_rsv_size(struct btrfs_fs_info
*fs_info
,
6023 struct btrfs_inode
*inode
)
6025 struct btrfs_block_rsv
*block_rsv
= &inode
->block_rsv
;
6026 u64 reserve_size
= 0;
6028 unsigned outstanding_extents
;
6030 lockdep_assert_held(&inode
->lock
);
6031 outstanding_extents
= inode
->outstanding_extents
;
6032 if (outstanding_extents
)
6033 reserve_size
= btrfs_calc_trans_metadata_size(fs_info
,
6034 outstanding_extents
+ 1);
6035 csum_leaves
= btrfs_csum_bytes_to_leaves(fs_info
,
6037 reserve_size
+= btrfs_calc_trans_metadata_size(fs_info
,
6040 spin_lock(&block_rsv
->lock
);
6041 block_rsv
->size
= reserve_size
;
6042 spin_unlock(&block_rsv
->lock
);
6045 int btrfs_delalloc_reserve_metadata(struct btrfs_inode
*inode
, u64 num_bytes
)
6047 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
6048 struct btrfs_root
*root
= inode
->root
;
6049 unsigned nr_extents
;
6050 enum btrfs_reserve_flush_enum flush
= BTRFS_RESERVE_FLUSH_ALL
;
6052 bool delalloc_lock
= true;
6054 /* If we are a free space inode we need to not flush since we will be in
6055 * the middle of a transaction commit. We also don't need the delalloc
6056 * mutex since we won't race with anybody. We need this mostly to make
6057 * lockdep shut its filthy mouth.
6059 * If we have a transaction open (can happen if we call truncate_block
6060 * from truncate), then we need FLUSH_LIMIT so we don't deadlock.
6062 if (btrfs_is_free_space_inode(inode
)) {
6063 flush
= BTRFS_RESERVE_NO_FLUSH
;
6064 delalloc_lock
= false;
6065 } else if (current
->journal_info
) {
6066 flush
= BTRFS_RESERVE_FLUSH_LIMIT
;
6069 if (flush
!= BTRFS_RESERVE_NO_FLUSH
&&
6070 btrfs_transaction_in_commit(fs_info
))
6071 schedule_timeout(1);
6074 mutex_lock(&inode
->delalloc_mutex
);
6076 num_bytes
= ALIGN(num_bytes
, fs_info
->sectorsize
);
6078 /* Add our new extents and calculate the new rsv size. */
6079 spin_lock(&inode
->lock
);
6080 nr_extents
= count_max_extents(num_bytes
);
6081 btrfs_mod_outstanding_extents(inode
, nr_extents
);
6082 inode
->csum_bytes
+= num_bytes
;
6083 btrfs_calculate_inode_block_rsv_size(fs_info
, inode
);
6084 spin_unlock(&inode
->lock
);
6086 if (test_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
)) {
6087 ret
= btrfs_qgroup_reserve_meta(root
,
6088 nr_extents
* fs_info
->nodesize
, true);
6093 ret
= btrfs_inode_rsv_refill(inode
, flush
);
6094 if (unlikely(ret
)) {
6095 btrfs_qgroup_free_meta(root
,
6096 nr_extents
* fs_info
->nodesize
);
6101 mutex_unlock(&inode
->delalloc_mutex
);
6105 spin_lock(&inode
->lock
);
6106 nr_extents
= count_max_extents(num_bytes
);
6107 btrfs_mod_outstanding_extents(inode
, -nr_extents
);
6108 inode
->csum_bytes
-= num_bytes
;
6109 btrfs_calculate_inode_block_rsv_size(fs_info
, inode
);
6110 spin_unlock(&inode
->lock
);
6112 btrfs_inode_rsv_release(inode
);
6114 mutex_unlock(&inode
->delalloc_mutex
);
6119 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
6120 * @inode: the inode to release the reservation for.
6121 * @num_bytes: the number of bytes we are releasing.
6123 * This will release the metadata reservation for an inode. This can be called
6124 * once we complete IO for a given set of bytes to release their metadata
6125 * reservations, or on error for the same reason.
6127 void btrfs_delalloc_release_metadata(struct btrfs_inode
*inode
, u64 num_bytes
)
6129 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
6131 num_bytes
= ALIGN(num_bytes
, fs_info
->sectorsize
);
6132 spin_lock(&inode
->lock
);
6133 inode
->csum_bytes
-= num_bytes
;
6134 btrfs_calculate_inode_block_rsv_size(fs_info
, inode
);
6135 spin_unlock(&inode
->lock
);
6137 if (btrfs_is_testing(fs_info
))
6140 btrfs_inode_rsv_release(inode
);
6144 * btrfs_delalloc_release_extents - release our outstanding_extents
6145 * @inode: the inode to balance the reservation for.
6146 * @num_bytes: the number of bytes we originally reserved with
6148 * When we reserve space we increase outstanding_extents for the extents we may
6149 * add. Once we've set the range as delalloc or created our ordered extents we
6150 * have outstanding_extents to track the real usage, so we use this to free our
6151 * temporarily tracked outstanding_extents. This _must_ be used in conjunction
6152 * with btrfs_delalloc_reserve_metadata.
6154 void btrfs_delalloc_release_extents(struct btrfs_inode
*inode
, u64 num_bytes
)
6156 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
6157 unsigned num_extents
;
6159 spin_lock(&inode
->lock
);
6160 num_extents
= count_max_extents(num_bytes
);
6161 btrfs_mod_outstanding_extents(inode
, -num_extents
);
6162 btrfs_calculate_inode_block_rsv_size(fs_info
, inode
);
6163 spin_unlock(&inode
->lock
);
6165 if (btrfs_is_testing(fs_info
))
6168 btrfs_inode_rsv_release(inode
);
6172 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6174 * @inode: inode we're writing to
6175 * @start: start range we are writing to
6176 * @len: how long the range we are writing to
6177 * @reserved: mandatory parameter, record actually reserved qgroup ranges of
6178 * current reservation.
6180 * This will do the following things
6182 * o reserve space in data space info for num bytes
6183 * and reserve precious corresponding qgroup space
6184 * (Done in check_data_free_space)
6186 * o reserve space for metadata space, based on the number of outstanding
6187 * extents and how much csums will be needed
6188 * also reserve metadata space in a per root over-reserve method.
6189 * o add to the inodes->delalloc_bytes
6190 * o add it to the fs_info's delalloc inodes list.
6191 * (Above 3 all done in delalloc_reserve_metadata)
6193 * Return 0 for success
6194 * Return <0 for error(-ENOSPC or -EQUOT)
6196 int btrfs_delalloc_reserve_space(struct inode
*inode
,
6197 struct extent_changeset
**reserved
, u64 start
, u64 len
)
6201 ret
= btrfs_check_data_free_space(inode
, reserved
, start
, len
);
6204 ret
= btrfs_delalloc_reserve_metadata(BTRFS_I(inode
), len
);
6206 btrfs_free_reserved_data_space(inode
, *reserved
, start
, len
);
6211 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6212 * @inode: inode we're releasing space for
6213 * @start: start position of the space already reserved
6214 * @len: the len of the space already reserved
6215 * @release_bytes: the len of the space we consumed or didn't use
6217 * This function will release the metadata space that was not used and will
6218 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
6219 * list if there are no delalloc bytes left.
6220 * Also it will handle the qgroup reserved space.
6222 void btrfs_delalloc_release_space(struct inode
*inode
,
6223 struct extent_changeset
*reserved
,
6226 btrfs_delalloc_release_metadata(BTRFS_I(inode
), len
);
6227 btrfs_free_reserved_data_space(inode
, reserved
, start
, len
);
6230 static int update_block_group(struct btrfs_trans_handle
*trans
,
6231 struct btrfs_fs_info
*info
, u64 bytenr
,
6232 u64 num_bytes
, int alloc
)
6234 struct btrfs_block_group_cache
*cache
= NULL
;
6235 u64 total
= num_bytes
;
6240 /* block accounting for super block */
6241 spin_lock(&info
->delalloc_root_lock
);
6242 old_val
= btrfs_super_bytes_used(info
->super_copy
);
6244 old_val
+= num_bytes
;
6246 old_val
-= num_bytes
;
6247 btrfs_set_super_bytes_used(info
->super_copy
, old_val
);
6248 spin_unlock(&info
->delalloc_root_lock
);
6251 cache
= btrfs_lookup_block_group(info
, bytenr
);
6254 if (cache
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
6255 BTRFS_BLOCK_GROUP_RAID1
|
6256 BTRFS_BLOCK_GROUP_RAID10
))
6261 * If this block group has free space cache written out, we
6262 * need to make sure to load it if we are removing space. This
6263 * is because we need the unpinning stage to actually add the
6264 * space back to the block group, otherwise we will leak space.
6266 if (!alloc
&& cache
->cached
== BTRFS_CACHE_NO
)
6267 cache_block_group(cache
, 1);
6269 byte_in_group
= bytenr
- cache
->key
.objectid
;
6270 WARN_ON(byte_in_group
> cache
->key
.offset
);
6272 spin_lock(&cache
->space_info
->lock
);
6273 spin_lock(&cache
->lock
);
6275 if (btrfs_test_opt(info
, SPACE_CACHE
) &&
6276 cache
->disk_cache_state
< BTRFS_DC_CLEAR
)
6277 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
6279 old_val
= btrfs_block_group_used(&cache
->item
);
6280 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
6282 old_val
+= num_bytes
;
6283 btrfs_set_block_group_used(&cache
->item
, old_val
);
6284 cache
->reserved
-= num_bytes
;
6285 cache
->space_info
->bytes_reserved
-= num_bytes
;
6286 cache
->space_info
->bytes_used
+= num_bytes
;
6287 cache
->space_info
->disk_used
+= num_bytes
* factor
;
6288 spin_unlock(&cache
->lock
);
6289 spin_unlock(&cache
->space_info
->lock
);
6291 old_val
-= num_bytes
;
6292 btrfs_set_block_group_used(&cache
->item
, old_val
);
6293 cache
->pinned
+= num_bytes
;
6294 cache
->space_info
->bytes_pinned
+= num_bytes
;
6295 cache
->space_info
->bytes_used
-= num_bytes
;
6296 cache
->space_info
->disk_used
-= num_bytes
* factor
;
6297 spin_unlock(&cache
->lock
);
6298 spin_unlock(&cache
->space_info
->lock
);
6300 trace_btrfs_space_reservation(info
, "pinned",
6301 cache
->space_info
->flags
,
6303 percpu_counter_add(&cache
->space_info
->total_bytes_pinned
,
6305 set_extent_dirty(info
->pinned_extents
,
6306 bytenr
, bytenr
+ num_bytes
- 1,
6307 GFP_NOFS
| __GFP_NOFAIL
);
6310 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
6311 if (list_empty(&cache
->dirty_list
)) {
6312 list_add_tail(&cache
->dirty_list
,
6313 &trans
->transaction
->dirty_bgs
);
6314 trans
->transaction
->num_dirty_bgs
++;
6315 btrfs_get_block_group(cache
);
6317 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
6320 * No longer have used bytes in this block group, queue it for
6321 * deletion. We do this after adding the block group to the
6322 * dirty list to avoid races between cleaner kthread and space
6325 if (!alloc
&& old_val
== 0) {
6326 spin_lock(&info
->unused_bgs_lock
);
6327 if (list_empty(&cache
->bg_list
)) {
6328 btrfs_get_block_group(cache
);
6329 list_add_tail(&cache
->bg_list
,
6332 spin_unlock(&info
->unused_bgs_lock
);
6335 btrfs_put_block_group(cache
);
6337 bytenr
+= num_bytes
;
6342 static u64
first_logical_byte(struct btrfs_fs_info
*fs_info
, u64 search_start
)
6344 struct btrfs_block_group_cache
*cache
;
6347 spin_lock(&fs_info
->block_group_cache_lock
);
6348 bytenr
= fs_info
->first_logical_byte
;
6349 spin_unlock(&fs_info
->block_group_cache_lock
);
6351 if (bytenr
< (u64
)-1)
6354 cache
= btrfs_lookup_first_block_group(fs_info
, search_start
);
6358 bytenr
= cache
->key
.objectid
;
6359 btrfs_put_block_group(cache
);
6364 static int pin_down_extent(struct btrfs_fs_info
*fs_info
,
6365 struct btrfs_block_group_cache
*cache
,
6366 u64 bytenr
, u64 num_bytes
, int reserved
)
6368 spin_lock(&cache
->space_info
->lock
);
6369 spin_lock(&cache
->lock
);
6370 cache
->pinned
+= num_bytes
;
6371 cache
->space_info
->bytes_pinned
+= num_bytes
;
6373 cache
->reserved
-= num_bytes
;
6374 cache
->space_info
->bytes_reserved
-= num_bytes
;
6376 spin_unlock(&cache
->lock
);
6377 spin_unlock(&cache
->space_info
->lock
);
6379 trace_btrfs_space_reservation(fs_info
, "pinned",
6380 cache
->space_info
->flags
, num_bytes
, 1);
6381 percpu_counter_add(&cache
->space_info
->total_bytes_pinned
, num_bytes
);
6382 set_extent_dirty(fs_info
->pinned_extents
, bytenr
,
6383 bytenr
+ num_bytes
- 1, GFP_NOFS
| __GFP_NOFAIL
);
6388 * this function must be called within transaction
6390 int btrfs_pin_extent(struct btrfs_fs_info
*fs_info
,
6391 u64 bytenr
, u64 num_bytes
, int reserved
)
6393 struct btrfs_block_group_cache
*cache
;
6395 cache
= btrfs_lookup_block_group(fs_info
, bytenr
);
6396 BUG_ON(!cache
); /* Logic error */
6398 pin_down_extent(fs_info
, cache
, bytenr
, num_bytes
, reserved
);
6400 btrfs_put_block_group(cache
);
6405 * this function must be called within transaction
6407 int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info
*fs_info
,
6408 u64 bytenr
, u64 num_bytes
)
6410 struct btrfs_block_group_cache
*cache
;
6413 cache
= btrfs_lookup_block_group(fs_info
, bytenr
);
6418 * pull in the free space cache (if any) so that our pin
6419 * removes the free space from the cache. We have load_only set
6420 * to one because the slow code to read in the free extents does check
6421 * the pinned extents.
6423 cache_block_group(cache
, 1);
6425 pin_down_extent(fs_info
, cache
, bytenr
, num_bytes
, 0);
6427 /* remove us from the free space cache (if we're there at all) */
6428 ret
= btrfs_remove_free_space(cache
, bytenr
, num_bytes
);
6429 btrfs_put_block_group(cache
);
6433 static int __exclude_logged_extent(struct btrfs_fs_info
*fs_info
,
6434 u64 start
, u64 num_bytes
)
6437 struct btrfs_block_group_cache
*block_group
;
6438 struct btrfs_caching_control
*caching_ctl
;
6440 block_group
= btrfs_lookup_block_group(fs_info
, start
);
6444 cache_block_group(block_group
, 0);
6445 caching_ctl
= get_caching_control(block_group
);
6449 BUG_ON(!block_group_cache_done(block_group
));
6450 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
6452 mutex_lock(&caching_ctl
->mutex
);
6454 if (start
>= caching_ctl
->progress
) {
6455 ret
= add_excluded_extent(fs_info
, start
, num_bytes
);
6456 } else if (start
+ num_bytes
<= caching_ctl
->progress
) {
6457 ret
= btrfs_remove_free_space(block_group
,
6460 num_bytes
= caching_ctl
->progress
- start
;
6461 ret
= btrfs_remove_free_space(block_group
,
6466 num_bytes
= (start
+ num_bytes
) -
6467 caching_ctl
->progress
;
6468 start
= caching_ctl
->progress
;
6469 ret
= add_excluded_extent(fs_info
, start
, num_bytes
);
6472 mutex_unlock(&caching_ctl
->mutex
);
6473 put_caching_control(caching_ctl
);
6475 btrfs_put_block_group(block_group
);
6479 int btrfs_exclude_logged_extents(struct btrfs_fs_info
*fs_info
,
6480 struct extent_buffer
*eb
)
6482 struct btrfs_file_extent_item
*item
;
6483 struct btrfs_key key
;
6487 if (!btrfs_fs_incompat(fs_info
, MIXED_GROUPS
))
6490 for (i
= 0; i
< btrfs_header_nritems(eb
); i
++) {
6491 btrfs_item_key_to_cpu(eb
, &key
, i
);
6492 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
6494 item
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
6495 found_type
= btrfs_file_extent_type(eb
, item
);
6496 if (found_type
== BTRFS_FILE_EXTENT_INLINE
)
6498 if (btrfs_file_extent_disk_bytenr(eb
, item
) == 0)
6500 key
.objectid
= btrfs_file_extent_disk_bytenr(eb
, item
);
6501 key
.offset
= btrfs_file_extent_disk_num_bytes(eb
, item
);
6502 __exclude_logged_extent(fs_info
, key
.objectid
, key
.offset
);
6509 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache
*bg
)
6511 atomic_inc(&bg
->reservations
);
6514 void btrfs_dec_block_group_reservations(struct btrfs_fs_info
*fs_info
,
6517 struct btrfs_block_group_cache
*bg
;
6519 bg
= btrfs_lookup_block_group(fs_info
, start
);
6521 if (atomic_dec_and_test(&bg
->reservations
))
6522 wake_up_atomic_t(&bg
->reservations
);
6523 btrfs_put_block_group(bg
);
6526 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache
*bg
)
6528 struct btrfs_space_info
*space_info
= bg
->space_info
;
6532 if (!(bg
->flags
& BTRFS_BLOCK_GROUP_DATA
))
6536 * Our block group is read only but before we set it to read only,
6537 * some task might have had allocated an extent from it already, but it
6538 * has not yet created a respective ordered extent (and added it to a
6539 * root's list of ordered extents).
6540 * Therefore wait for any task currently allocating extents, since the
6541 * block group's reservations counter is incremented while a read lock
6542 * on the groups' semaphore is held and decremented after releasing
6543 * the read access on that semaphore and creating the ordered extent.
6545 down_write(&space_info
->groups_sem
);
6546 up_write(&space_info
->groups_sem
);
6548 wait_on_atomic_t(&bg
->reservations
, atomic_t_wait
,
6549 TASK_UNINTERRUPTIBLE
);
6553 * btrfs_add_reserved_bytes - update the block_group and space info counters
6554 * @cache: The cache we are manipulating
6555 * @ram_bytes: The number of bytes of file content, and will be same to
6556 * @num_bytes except for the compress path.
6557 * @num_bytes: The number of bytes in question
6558 * @delalloc: The blocks are allocated for the delalloc write
6560 * This is called by the allocator when it reserves space. If this is a
6561 * reservation and the block group has become read only we cannot make the
6562 * reservation and return -EAGAIN, otherwise this function always succeeds.
6564 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache
*cache
,
6565 u64 ram_bytes
, u64 num_bytes
, int delalloc
)
6567 struct btrfs_space_info
*space_info
= cache
->space_info
;
6570 spin_lock(&space_info
->lock
);
6571 spin_lock(&cache
->lock
);
6575 cache
->reserved
+= num_bytes
;
6576 space_info
->bytes_reserved
+= num_bytes
;
6578 trace_btrfs_space_reservation(cache
->fs_info
,
6579 "space_info", space_info
->flags
,
6581 space_info
->bytes_may_use
-= ram_bytes
;
6583 cache
->delalloc_bytes
+= num_bytes
;
6585 spin_unlock(&cache
->lock
);
6586 spin_unlock(&space_info
->lock
);
6591 * btrfs_free_reserved_bytes - update the block_group and space info counters
6592 * @cache: The cache we are manipulating
6593 * @num_bytes: The number of bytes in question
6594 * @delalloc: The blocks are allocated for the delalloc write
6596 * This is called by somebody who is freeing space that was never actually used
6597 * on disk. For example if you reserve some space for a new leaf in transaction
6598 * A and before transaction A commits you free that leaf, you call this with
6599 * reserve set to 0 in order to clear the reservation.
6602 static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache
*cache
,
6603 u64 num_bytes
, int delalloc
)
6605 struct btrfs_space_info
*space_info
= cache
->space_info
;
6608 spin_lock(&space_info
->lock
);
6609 spin_lock(&cache
->lock
);
6611 space_info
->bytes_readonly
+= num_bytes
;
6612 cache
->reserved
-= num_bytes
;
6613 space_info
->bytes_reserved
-= num_bytes
;
6616 cache
->delalloc_bytes
-= num_bytes
;
6617 spin_unlock(&cache
->lock
);
6618 spin_unlock(&space_info
->lock
);
6621 void btrfs_prepare_extent_commit(struct btrfs_fs_info
*fs_info
)
6623 struct btrfs_caching_control
*next
;
6624 struct btrfs_caching_control
*caching_ctl
;
6625 struct btrfs_block_group_cache
*cache
;
6627 down_write(&fs_info
->commit_root_sem
);
6629 list_for_each_entry_safe(caching_ctl
, next
,
6630 &fs_info
->caching_block_groups
, list
) {
6631 cache
= caching_ctl
->block_group
;
6632 if (block_group_cache_done(cache
)) {
6633 cache
->last_byte_to_unpin
= (u64
)-1;
6634 list_del_init(&caching_ctl
->list
);
6635 put_caching_control(caching_ctl
);
6637 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
6641 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
6642 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
6644 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
6646 up_write(&fs_info
->commit_root_sem
);
6648 update_global_block_rsv(fs_info
);
6652 * Returns the free cluster for the given space info and sets empty_cluster to
6653 * what it should be based on the mount options.
6655 static struct btrfs_free_cluster
*
6656 fetch_cluster_info(struct btrfs_fs_info
*fs_info
,
6657 struct btrfs_space_info
*space_info
, u64
*empty_cluster
)
6659 struct btrfs_free_cluster
*ret
= NULL
;
6662 if (btrfs_mixed_space_info(space_info
))
6665 if (space_info
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
6666 ret
= &fs_info
->meta_alloc_cluster
;
6667 if (btrfs_test_opt(fs_info
, SSD
))
6668 *empty_cluster
= SZ_2M
;
6670 *empty_cluster
= SZ_64K
;
6671 } else if ((space_info
->flags
& BTRFS_BLOCK_GROUP_DATA
) &&
6672 btrfs_test_opt(fs_info
, SSD_SPREAD
)) {
6673 *empty_cluster
= SZ_2M
;
6674 ret
= &fs_info
->data_alloc_cluster
;
6680 static int unpin_extent_range(struct btrfs_fs_info
*fs_info
,
6682 const bool return_free_space
)
6684 struct btrfs_block_group_cache
*cache
= NULL
;
6685 struct btrfs_space_info
*space_info
;
6686 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
6687 struct btrfs_free_cluster
*cluster
= NULL
;
6689 u64 total_unpinned
= 0;
6690 u64 empty_cluster
= 0;
6693 while (start
<= end
) {
6696 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
6698 btrfs_put_block_group(cache
);
6700 cache
= btrfs_lookup_block_group(fs_info
, start
);
6701 BUG_ON(!cache
); /* Logic error */
6703 cluster
= fetch_cluster_info(fs_info
,
6706 empty_cluster
<<= 1;
6709 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
6710 len
= min(len
, end
+ 1 - start
);
6712 if (start
< cache
->last_byte_to_unpin
) {
6713 len
= min(len
, cache
->last_byte_to_unpin
- start
);
6714 if (return_free_space
)
6715 btrfs_add_free_space(cache
, start
, len
);
6719 total_unpinned
+= len
;
6720 space_info
= cache
->space_info
;
6723 * If this space cluster has been marked as fragmented and we've
6724 * unpinned enough in this block group to potentially allow a
6725 * cluster to be created inside of it go ahead and clear the
6728 if (cluster
&& cluster
->fragmented
&&
6729 total_unpinned
> empty_cluster
) {
6730 spin_lock(&cluster
->lock
);
6731 cluster
->fragmented
= 0;
6732 spin_unlock(&cluster
->lock
);
6735 spin_lock(&space_info
->lock
);
6736 spin_lock(&cache
->lock
);
6737 cache
->pinned
-= len
;
6738 space_info
->bytes_pinned
-= len
;
6740 trace_btrfs_space_reservation(fs_info
, "pinned",
6741 space_info
->flags
, len
, 0);
6742 space_info
->max_extent_size
= 0;
6743 percpu_counter_add(&space_info
->total_bytes_pinned
, -len
);
6745 space_info
->bytes_readonly
+= len
;
6748 spin_unlock(&cache
->lock
);
6749 if (!readonly
&& return_free_space
&&
6750 global_rsv
->space_info
== space_info
) {
6753 spin_lock(&global_rsv
->lock
);
6754 if (!global_rsv
->full
) {
6755 to_add
= min(len
, global_rsv
->size
-
6756 global_rsv
->reserved
);
6757 global_rsv
->reserved
+= to_add
;
6758 space_info
->bytes_may_use
+= to_add
;
6759 if (global_rsv
->reserved
>= global_rsv
->size
)
6760 global_rsv
->full
= 1;
6761 trace_btrfs_space_reservation(fs_info
,
6767 spin_unlock(&global_rsv
->lock
);
6768 /* Add to any tickets we may have */
6770 space_info_add_new_bytes(fs_info
, space_info
,
6773 spin_unlock(&space_info
->lock
);
6777 btrfs_put_block_group(cache
);
6781 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
6782 struct btrfs_fs_info
*fs_info
)
6784 struct btrfs_block_group_cache
*block_group
, *tmp
;
6785 struct list_head
*deleted_bgs
;
6786 struct extent_io_tree
*unpin
;
6791 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
6792 unpin
= &fs_info
->freed_extents
[1];
6794 unpin
= &fs_info
->freed_extents
[0];
6796 while (!trans
->aborted
) {
6797 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
6798 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
6799 EXTENT_DIRTY
, NULL
);
6801 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
6805 if (btrfs_test_opt(fs_info
, DISCARD
))
6806 ret
= btrfs_discard_extent(fs_info
, start
,
6807 end
+ 1 - start
, NULL
);
6809 clear_extent_dirty(unpin
, start
, end
);
6810 unpin_extent_range(fs_info
, start
, end
, true);
6811 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
6816 * Transaction is finished. We don't need the lock anymore. We
6817 * do need to clean up the block groups in case of a transaction
6820 deleted_bgs
= &trans
->transaction
->deleted_bgs
;
6821 list_for_each_entry_safe(block_group
, tmp
, deleted_bgs
, bg_list
) {
6825 if (!trans
->aborted
)
6826 ret
= btrfs_discard_extent(fs_info
,
6827 block_group
->key
.objectid
,
6828 block_group
->key
.offset
,
6831 list_del_init(&block_group
->bg_list
);
6832 btrfs_put_block_group_trimming(block_group
);
6833 btrfs_put_block_group(block_group
);
6836 const char *errstr
= btrfs_decode_error(ret
);
6838 "discard failed while removing blockgroup: errno=%d %s",
6846 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
6847 struct btrfs_fs_info
*info
,
6848 struct btrfs_delayed_ref_node
*node
, u64 parent
,
6849 u64 root_objectid
, u64 owner_objectid
,
6850 u64 owner_offset
, int refs_to_drop
,
6851 struct btrfs_delayed_extent_op
*extent_op
)
6853 struct btrfs_key key
;
6854 struct btrfs_path
*path
;
6855 struct btrfs_root
*extent_root
= info
->extent_root
;
6856 struct extent_buffer
*leaf
;
6857 struct btrfs_extent_item
*ei
;
6858 struct btrfs_extent_inline_ref
*iref
;
6861 int extent_slot
= 0;
6862 int found_extent
= 0;
6866 u64 bytenr
= node
->bytenr
;
6867 u64 num_bytes
= node
->num_bytes
;
6869 bool skinny_metadata
= btrfs_fs_incompat(info
, SKINNY_METADATA
);
6871 path
= btrfs_alloc_path();
6875 path
->reada
= READA_FORWARD
;
6876 path
->leave_spinning
= 1;
6878 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
6879 BUG_ON(!is_data
&& refs_to_drop
!= 1);
6882 skinny_metadata
= false;
6884 ret
= lookup_extent_backref(trans
, info
, path
, &iref
,
6885 bytenr
, num_bytes
, parent
,
6886 root_objectid
, owner_objectid
,
6889 extent_slot
= path
->slots
[0];
6890 while (extent_slot
>= 0) {
6891 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
6893 if (key
.objectid
!= bytenr
)
6895 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
6896 key
.offset
== num_bytes
) {
6900 if (key
.type
== BTRFS_METADATA_ITEM_KEY
&&
6901 key
.offset
== owner_objectid
) {
6905 if (path
->slots
[0] - extent_slot
> 5)
6909 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6910 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
6911 if (found_extent
&& item_size
< sizeof(*ei
))
6914 if (!found_extent
) {
6916 ret
= remove_extent_backref(trans
, info
, path
, NULL
,
6918 is_data
, &last_ref
);
6920 btrfs_abort_transaction(trans
, ret
);
6923 btrfs_release_path(path
);
6924 path
->leave_spinning
= 1;
6926 key
.objectid
= bytenr
;
6927 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6928 key
.offset
= num_bytes
;
6930 if (!is_data
&& skinny_metadata
) {
6931 key
.type
= BTRFS_METADATA_ITEM_KEY
;
6932 key
.offset
= owner_objectid
;
6935 ret
= btrfs_search_slot(trans
, extent_root
,
6937 if (ret
> 0 && skinny_metadata
&& path
->slots
[0]) {
6939 * Couldn't find our skinny metadata item,
6940 * see if we have ye olde extent item.
6943 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
6945 if (key
.objectid
== bytenr
&&
6946 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
6947 key
.offset
== num_bytes
)
6951 if (ret
> 0 && skinny_metadata
) {
6952 skinny_metadata
= false;
6953 key
.objectid
= bytenr
;
6954 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6955 key
.offset
= num_bytes
;
6956 btrfs_release_path(path
);
6957 ret
= btrfs_search_slot(trans
, extent_root
,
6963 "umm, got %d back from search, was looking for %llu",
6966 btrfs_print_leaf(path
->nodes
[0]);
6969 btrfs_abort_transaction(trans
, ret
);
6972 extent_slot
= path
->slots
[0];
6974 } else if (WARN_ON(ret
== -ENOENT
)) {
6975 btrfs_print_leaf(path
->nodes
[0]);
6977 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6978 bytenr
, parent
, root_objectid
, owner_objectid
,
6980 btrfs_abort_transaction(trans
, ret
);
6983 btrfs_abort_transaction(trans
, ret
);
6987 leaf
= path
->nodes
[0];
6988 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
6989 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6990 if (item_size
< sizeof(*ei
)) {
6991 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
6992 ret
= convert_extent_item_v0(trans
, info
, path
, owner_objectid
,
6995 btrfs_abort_transaction(trans
, ret
);
6999 btrfs_release_path(path
);
7000 path
->leave_spinning
= 1;
7002 key
.objectid
= bytenr
;
7003 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
7004 key
.offset
= num_bytes
;
7006 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
7010 "umm, got %d back from search, was looking for %llu",
7012 btrfs_print_leaf(path
->nodes
[0]);
7015 btrfs_abort_transaction(trans
, ret
);
7019 extent_slot
= path
->slots
[0];
7020 leaf
= path
->nodes
[0];
7021 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
7024 BUG_ON(item_size
< sizeof(*ei
));
7025 ei
= btrfs_item_ptr(leaf
, extent_slot
,
7026 struct btrfs_extent_item
);
7027 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
7028 key
.type
== BTRFS_EXTENT_ITEM_KEY
) {
7029 struct btrfs_tree_block_info
*bi
;
7030 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
7031 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
7032 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
7035 refs
= btrfs_extent_refs(leaf
, ei
);
7036 if (refs
< refs_to_drop
) {
7038 "trying to drop %d refs but we only have %Lu for bytenr %Lu",
7039 refs_to_drop
, refs
, bytenr
);
7041 btrfs_abort_transaction(trans
, ret
);
7044 refs
-= refs_to_drop
;
7048 __run_delayed_extent_op(extent_op
, leaf
, ei
);
7050 * In the case of inline back ref, reference count will
7051 * be updated by remove_extent_backref
7054 BUG_ON(!found_extent
);
7056 btrfs_set_extent_refs(leaf
, ei
, refs
);
7057 btrfs_mark_buffer_dirty(leaf
);
7060 ret
= remove_extent_backref(trans
, info
, path
,
7062 is_data
, &last_ref
);
7064 btrfs_abort_transaction(trans
, ret
);
7070 BUG_ON(is_data
&& refs_to_drop
!=
7071 extent_data_ref_count(path
, iref
));
7073 BUG_ON(path
->slots
[0] != extent_slot
);
7075 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
7076 path
->slots
[0] = extent_slot
;
7082 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
7085 btrfs_abort_transaction(trans
, ret
);
7088 btrfs_release_path(path
);
7091 ret
= btrfs_del_csums(trans
, info
, bytenr
, num_bytes
);
7093 btrfs_abort_transaction(trans
, ret
);
7098 ret
= add_to_free_space_tree(trans
, info
, bytenr
, num_bytes
);
7100 btrfs_abort_transaction(trans
, ret
);
7104 ret
= update_block_group(trans
, info
, bytenr
, num_bytes
, 0);
7106 btrfs_abort_transaction(trans
, ret
);
7110 btrfs_release_path(path
);
7113 btrfs_free_path(path
);
7118 * when we free an block, it is possible (and likely) that we free the last
7119 * delayed ref for that extent as well. This searches the delayed ref tree for
7120 * a given extent, and if there are no other delayed refs to be processed, it
7121 * removes it from the tree.
7123 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
7126 struct btrfs_delayed_ref_head
*head
;
7127 struct btrfs_delayed_ref_root
*delayed_refs
;
7130 delayed_refs
= &trans
->transaction
->delayed_refs
;
7131 spin_lock(&delayed_refs
->lock
);
7132 head
= btrfs_find_delayed_ref_head(delayed_refs
, bytenr
);
7134 goto out_delayed_unlock
;
7136 spin_lock(&head
->lock
);
7137 if (!RB_EMPTY_ROOT(&head
->ref_tree
))
7140 if (head
->extent_op
) {
7141 if (!head
->must_insert_reserved
)
7143 btrfs_free_delayed_extent_op(head
->extent_op
);
7144 head
->extent_op
= NULL
;
7148 * waiting for the lock here would deadlock. If someone else has it
7149 * locked they are already in the process of dropping it anyway
7151 if (!mutex_trylock(&head
->mutex
))
7155 * at this point we have a head with no other entries. Go
7156 * ahead and process it.
7158 rb_erase(&head
->href_node
, &delayed_refs
->href_root
);
7159 RB_CLEAR_NODE(&head
->href_node
);
7160 atomic_dec(&delayed_refs
->num_entries
);
7163 * we don't take a ref on the node because we're removing it from the
7164 * tree, so we just steal the ref the tree was holding.
7166 delayed_refs
->num_heads
--;
7167 if (head
->processing
== 0)
7168 delayed_refs
->num_heads_ready
--;
7169 head
->processing
= 0;
7170 spin_unlock(&head
->lock
);
7171 spin_unlock(&delayed_refs
->lock
);
7173 BUG_ON(head
->extent_op
);
7174 if (head
->must_insert_reserved
)
7177 mutex_unlock(&head
->mutex
);
7178 btrfs_put_delayed_ref_head(head
);
7181 spin_unlock(&head
->lock
);
7184 spin_unlock(&delayed_refs
->lock
);
7188 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
7189 struct btrfs_root
*root
,
7190 struct extent_buffer
*buf
,
7191 u64 parent
, int last_ref
)
7193 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7197 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
7198 int old_ref_mod
, new_ref_mod
;
7200 btrfs_ref_tree_mod(root
, buf
->start
, buf
->len
, parent
,
7201 root
->root_key
.objectid
,
7202 btrfs_header_level(buf
), 0,
7203 BTRFS_DROP_DELAYED_REF
);
7204 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, buf
->start
,
7206 root
->root_key
.objectid
,
7207 btrfs_header_level(buf
),
7208 BTRFS_DROP_DELAYED_REF
, NULL
,
7209 &old_ref_mod
, &new_ref_mod
);
7210 BUG_ON(ret
); /* -ENOMEM */
7211 pin
= old_ref_mod
>= 0 && new_ref_mod
< 0;
7214 if (last_ref
&& btrfs_header_generation(buf
) == trans
->transid
) {
7215 struct btrfs_block_group_cache
*cache
;
7217 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
7218 ret
= check_ref_cleanup(trans
, buf
->start
);
7224 cache
= btrfs_lookup_block_group(fs_info
, buf
->start
);
7226 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
7227 pin_down_extent(fs_info
, cache
, buf
->start
,
7229 btrfs_put_block_group(cache
);
7233 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
7235 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
7236 btrfs_free_reserved_bytes(cache
, buf
->len
, 0);
7237 btrfs_put_block_group(cache
);
7238 trace_btrfs_reserved_extent_free(fs_info
, buf
->start
, buf
->len
);
7242 add_pinned_bytes(fs_info
, buf
->len
, btrfs_header_level(buf
),
7243 root
->root_key
.objectid
);
7247 * Deleting the buffer, clear the corrupt flag since it doesn't
7250 clear_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
);
7254 /* Can return -ENOMEM */
7255 int btrfs_free_extent(struct btrfs_trans_handle
*trans
,
7256 struct btrfs_root
*root
,
7257 u64 bytenr
, u64 num_bytes
, u64 parent
, u64 root_objectid
,
7258 u64 owner
, u64 offset
)
7260 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7261 int old_ref_mod
, new_ref_mod
;
7264 if (btrfs_is_testing(fs_info
))
7267 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
)
7268 btrfs_ref_tree_mod(root
, bytenr
, num_bytes
, parent
,
7269 root_objectid
, owner
, offset
,
7270 BTRFS_DROP_DELAYED_REF
);
7273 * tree log blocks never actually go into the extent allocation
7274 * tree, just update pinning info and exit early.
7276 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
7277 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
7278 /* unlocks the pinned mutex */
7279 btrfs_pin_extent(fs_info
, bytenr
, num_bytes
, 1);
7280 old_ref_mod
= new_ref_mod
= 0;
7282 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
7283 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
7285 root_objectid
, (int)owner
,
7286 BTRFS_DROP_DELAYED_REF
, NULL
,
7287 &old_ref_mod
, &new_ref_mod
);
7289 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
7291 root_objectid
, owner
, offset
,
7292 0, BTRFS_DROP_DELAYED_REF
,
7293 &old_ref_mod
, &new_ref_mod
);
7296 if (ret
== 0 && old_ref_mod
>= 0 && new_ref_mod
< 0)
7297 add_pinned_bytes(fs_info
, num_bytes
, owner
, root_objectid
);
7303 * when we wait for progress in the block group caching, its because
7304 * our allocation attempt failed at least once. So, we must sleep
7305 * and let some progress happen before we try again.
7307 * This function will sleep at least once waiting for new free space to
7308 * show up, and then it will check the block group free space numbers
7309 * for our min num_bytes. Another option is to have it go ahead
7310 * and look in the rbtree for a free extent of a given size, but this
7313 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7314 * any of the information in this block group.
7316 static noinline
void
7317 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
7320 struct btrfs_caching_control
*caching_ctl
;
7322 caching_ctl
= get_caching_control(cache
);
7326 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
7327 (cache
->free_space_ctl
->free_space
>= num_bytes
));
7329 put_caching_control(caching_ctl
);
7333 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
7335 struct btrfs_caching_control
*caching_ctl
;
7338 caching_ctl
= get_caching_control(cache
);
7340 return (cache
->cached
== BTRFS_CACHE_ERROR
) ? -EIO
: 0;
7342 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
7343 if (cache
->cached
== BTRFS_CACHE_ERROR
)
7345 put_caching_control(caching_ctl
);
7349 int __get_raid_index(u64 flags
)
7351 if (flags
& BTRFS_BLOCK_GROUP_RAID10
)
7352 return BTRFS_RAID_RAID10
;
7353 else if (flags
& BTRFS_BLOCK_GROUP_RAID1
)
7354 return BTRFS_RAID_RAID1
;
7355 else if (flags
& BTRFS_BLOCK_GROUP_DUP
)
7356 return BTRFS_RAID_DUP
;
7357 else if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
7358 return BTRFS_RAID_RAID0
;
7359 else if (flags
& BTRFS_BLOCK_GROUP_RAID5
)
7360 return BTRFS_RAID_RAID5
;
7361 else if (flags
& BTRFS_BLOCK_GROUP_RAID6
)
7362 return BTRFS_RAID_RAID6
;
7364 return BTRFS_RAID_SINGLE
; /* BTRFS_BLOCK_GROUP_SINGLE */
7367 int get_block_group_index(struct btrfs_block_group_cache
*cache
)
7369 return __get_raid_index(cache
->flags
);
7372 static const char *btrfs_raid_type_names
[BTRFS_NR_RAID_TYPES
] = {
7373 [BTRFS_RAID_RAID10
] = "raid10",
7374 [BTRFS_RAID_RAID1
] = "raid1",
7375 [BTRFS_RAID_DUP
] = "dup",
7376 [BTRFS_RAID_RAID0
] = "raid0",
7377 [BTRFS_RAID_SINGLE
] = "single",
7378 [BTRFS_RAID_RAID5
] = "raid5",
7379 [BTRFS_RAID_RAID6
] = "raid6",
7382 static const char *get_raid_name(enum btrfs_raid_types type
)
7384 if (type
>= BTRFS_NR_RAID_TYPES
)
7387 return btrfs_raid_type_names
[type
];
7390 enum btrfs_loop_type
{
7391 LOOP_CACHING_NOWAIT
= 0,
7392 LOOP_CACHING_WAIT
= 1,
7393 LOOP_ALLOC_CHUNK
= 2,
7394 LOOP_NO_EMPTY_SIZE
= 3,
7398 btrfs_lock_block_group(struct btrfs_block_group_cache
*cache
,
7402 down_read(&cache
->data_rwsem
);
7406 btrfs_grab_block_group(struct btrfs_block_group_cache
*cache
,
7409 btrfs_get_block_group(cache
);
7411 down_read(&cache
->data_rwsem
);
7414 static struct btrfs_block_group_cache
*
7415 btrfs_lock_cluster(struct btrfs_block_group_cache
*block_group
,
7416 struct btrfs_free_cluster
*cluster
,
7419 struct btrfs_block_group_cache
*used_bg
= NULL
;
7421 spin_lock(&cluster
->refill_lock
);
7423 used_bg
= cluster
->block_group
;
7427 if (used_bg
== block_group
)
7430 btrfs_get_block_group(used_bg
);
7435 if (down_read_trylock(&used_bg
->data_rwsem
))
7438 spin_unlock(&cluster
->refill_lock
);
7440 /* We should only have one-level nested. */
7441 down_read_nested(&used_bg
->data_rwsem
, SINGLE_DEPTH_NESTING
);
7443 spin_lock(&cluster
->refill_lock
);
7444 if (used_bg
== cluster
->block_group
)
7447 up_read(&used_bg
->data_rwsem
);
7448 btrfs_put_block_group(used_bg
);
7453 btrfs_release_block_group(struct btrfs_block_group_cache
*cache
,
7457 up_read(&cache
->data_rwsem
);
7458 btrfs_put_block_group(cache
);
7462 * walks the btree of allocated extents and find a hole of a given size.
7463 * The key ins is changed to record the hole:
7464 * ins->objectid == start position
7465 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7466 * ins->offset == the size of the hole.
7467 * Any available blocks before search_start are skipped.
7469 * If there is no suitable free space, we will record the max size of
7470 * the free space extent currently.
7472 static noinline
int find_free_extent(struct btrfs_fs_info
*fs_info
,
7473 u64 ram_bytes
, u64 num_bytes
, u64 empty_size
,
7474 u64 hint_byte
, struct btrfs_key
*ins
,
7475 u64 flags
, int delalloc
)
7478 struct btrfs_root
*root
= fs_info
->extent_root
;
7479 struct btrfs_free_cluster
*last_ptr
= NULL
;
7480 struct btrfs_block_group_cache
*block_group
= NULL
;
7481 u64 search_start
= 0;
7482 u64 max_extent_size
= 0;
7483 u64 empty_cluster
= 0;
7484 struct btrfs_space_info
*space_info
;
7486 int index
= __get_raid_index(flags
);
7487 bool failed_cluster_refill
= false;
7488 bool failed_alloc
= false;
7489 bool use_cluster
= true;
7490 bool have_caching_bg
= false;
7491 bool orig_have_caching_bg
= false;
7492 bool full_search
= false;
7494 WARN_ON(num_bytes
< fs_info
->sectorsize
);
7495 ins
->type
= BTRFS_EXTENT_ITEM_KEY
;
7499 trace_find_free_extent(fs_info
, num_bytes
, empty_size
, flags
);
7501 space_info
= __find_space_info(fs_info
, flags
);
7503 btrfs_err(fs_info
, "No space info for %llu", flags
);
7508 * If our free space is heavily fragmented we may not be able to make
7509 * big contiguous allocations, so instead of doing the expensive search
7510 * for free space, simply return ENOSPC with our max_extent_size so we
7511 * can go ahead and search for a more manageable chunk.
7513 * If our max_extent_size is large enough for our allocation simply
7514 * disable clustering since we will likely not be able to find enough
7515 * space to create a cluster and induce latency trying.
7517 if (unlikely(space_info
->max_extent_size
)) {
7518 spin_lock(&space_info
->lock
);
7519 if (space_info
->max_extent_size
&&
7520 num_bytes
> space_info
->max_extent_size
) {
7521 ins
->offset
= space_info
->max_extent_size
;
7522 spin_unlock(&space_info
->lock
);
7524 } else if (space_info
->max_extent_size
) {
7525 use_cluster
= false;
7527 spin_unlock(&space_info
->lock
);
7530 last_ptr
= fetch_cluster_info(fs_info
, space_info
, &empty_cluster
);
7532 spin_lock(&last_ptr
->lock
);
7533 if (last_ptr
->block_group
)
7534 hint_byte
= last_ptr
->window_start
;
7535 if (last_ptr
->fragmented
) {
7537 * We still set window_start so we can keep track of the
7538 * last place we found an allocation to try and save
7541 hint_byte
= last_ptr
->window_start
;
7542 use_cluster
= false;
7544 spin_unlock(&last_ptr
->lock
);
7547 search_start
= max(search_start
, first_logical_byte(fs_info
, 0));
7548 search_start
= max(search_start
, hint_byte
);
7549 if (search_start
== hint_byte
) {
7550 block_group
= btrfs_lookup_block_group(fs_info
, search_start
);
7552 * we don't want to use the block group if it doesn't match our
7553 * allocation bits, or if its not cached.
7555 * However if we are re-searching with an ideal block group
7556 * picked out then we don't care that the block group is cached.
7558 if (block_group
&& block_group_bits(block_group
, flags
) &&
7559 block_group
->cached
!= BTRFS_CACHE_NO
) {
7560 down_read(&space_info
->groups_sem
);
7561 if (list_empty(&block_group
->list
) ||
7564 * someone is removing this block group,
7565 * we can't jump into the have_block_group
7566 * target because our list pointers are not
7569 btrfs_put_block_group(block_group
);
7570 up_read(&space_info
->groups_sem
);
7572 index
= get_block_group_index(block_group
);
7573 btrfs_lock_block_group(block_group
, delalloc
);
7574 goto have_block_group
;
7576 } else if (block_group
) {
7577 btrfs_put_block_group(block_group
);
7581 have_caching_bg
= false;
7582 if (index
== 0 || index
== __get_raid_index(flags
))
7584 down_read(&space_info
->groups_sem
);
7585 list_for_each_entry(block_group
, &space_info
->block_groups
[index
],
7590 /* If the block group is read-only, we can skip it entirely. */
7591 if (unlikely(block_group
->ro
))
7594 btrfs_grab_block_group(block_group
, delalloc
);
7595 search_start
= block_group
->key
.objectid
;
7598 * this can happen if we end up cycling through all the
7599 * raid types, but we want to make sure we only allocate
7600 * for the proper type.
7602 if (!block_group_bits(block_group
, flags
)) {
7603 u64 extra
= BTRFS_BLOCK_GROUP_DUP
|
7604 BTRFS_BLOCK_GROUP_RAID1
|
7605 BTRFS_BLOCK_GROUP_RAID5
|
7606 BTRFS_BLOCK_GROUP_RAID6
|
7607 BTRFS_BLOCK_GROUP_RAID10
;
7610 * if they asked for extra copies and this block group
7611 * doesn't provide them, bail. This does allow us to
7612 * fill raid0 from raid1.
7614 if ((flags
& extra
) && !(block_group
->flags
& extra
))
7619 cached
= block_group_cache_done(block_group
);
7620 if (unlikely(!cached
)) {
7621 have_caching_bg
= true;
7622 ret
= cache_block_group(block_group
, 0);
7627 if (unlikely(block_group
->cached
== BTRFS_CACHE_ERROR
))
7631 * Ok we want to try and use the cluster allocator, so
7634 if (last_ptr
&& use_cluster
) {
7635 struct btrfs_block_group_cache
*used_block_group
;
7636 unsigned long aligned_cluster
;
7638 * the refill lock keeps out other
7639 * people trying to start a new cluster
7641 used_block_group
= btrfs_lock_cluster(block_group
,
7644 if (!used_block_group
)
7645 goto refill_cluster
;
7647 if (used_block_group
!= block_group
&&
7648 (used_block_group
->ro
||
7649 !block_group_bits(used_block_group
, flags
)))
7650 goto release_cluster
;
7652 offset
= btrfs_alloc_from_cluster(used_block_group
,
7655 used_block_group
->key
.objectid
,
7658 /* we have a block, we're done */
7659 spin_unlock(&last_ptr
->refill_lock
);
7660 trace_btrfs_reserve_extent_cluster(fs_info
,
7662 search_start
, num_bytes
);
7663 if (used_block_group
!= block_group
) {
7664 btrfs_release_block_group(block_group
,
7666 block_group
= used_block_group
;
7671 WARN_ON(last_ptr
->block_group
!= used_block_group
);
7673 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7674 * set up a new clusters, so lets just skip it
7675 * and let the allocator find whatever block
7676 * it can find. If we reach this point, we
7677 * will have tried the cluster allocator
7678 * plenty of times and not have found
7679 * anything, so we are likely way too
7680 * fragmented for the clustering stuff to find
7683 * However, if the cluster is taken from the
7684 * current block group, release the cluster
7685 * first, so that we stand a better chance of
7686 * succeeding in the unclustered
7688 if (loop
>= LOOP_NO_EMPTY_SIZE
&&
7689 used_block_group
!= block_group
) {
7690 spin_unlock(&last_ptr
->refill_lock
);
7691 btrfs_release_block_group(used_block_group
,
7693 goto unclustered_alloc
;
7697 * this cluster didn't work out, free it and
7700 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
7702 if (used_block_group
!= block_group
)
7703 btrfs_release_block_group(used_block_group
,
7706 if (loop
>= LOOP_NO_EMPTY_SIZE
) {
7707 spin_unlock(&last_ptr
->refill_lock
);
7708 goto unclustered_alloc
;
7711 aligned_cluster
= max_t(unsigned long,
7712 empty_cluster
+ empty_size
,
7713 block_group
->full_stripe_len
);
7715 /* allocate a cluster in this block group */
7716 ret
= btrfs_find_space_cluster(fs_info
, block_group
,
7717 last_ptr
, search_start
,
7722 * now pull our allocation out of this
7725 offset
= btrfs_alloc_from_cluster(block_group
,
7731 /* we found one, proceed */
7732 spin_unlock(&last_ptr
->refill_lock
);
7733 trace_btrfs_reserve_extent_cluster(fs_info
,
7734 block_group
, search_start
,
7738 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
7739 && !failed_cluster_refill
) {
7740 spin_unlock(&last_ptr
->refill_lock
);
7742 failed_cluster_refill
= true;
7743 wait_block_group_cache_progress(block_group
,
7744 num_bytes
+ empty_cluster
+ empty_size
);
7745 goto have_block_group
;
7749 * at this point we either didn't find a cluster
7750 * or we weren't able to allocate a block from our
7751 * cluster. Free the cluster we've been trying
7752 * to use, and go to the next block group
7754 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
7755 spin_unlock(&last_ptr
->refill_lock
);
7761 * We are doing an unclustered alloc, set the fragmented flag so
7762 * we don't bother trying to setup a cluster again until we get
7765 if (unlikely(last_ptr
)) {
7766 spin_lock(&last_ptr
->lock
);
7767 last_ptr
->fragmented
= 1;
7768 spin_unlock(&last_ptr
->lock
);
7771 struct btrfs_free_space_ctl
*ctl
=
7772 block_group
->free_space_ctl
;
7774 spin_lock(&ctl
->tree_lock
);
7775 if (ctl
->free_space
<
7776 num_bytes
+ empty_cluster
+ empty_size
) {
7777 if (ctl
->free_space
> max_extent_size
)
7778 max_extent_size
= ctl
->free_space
;
7779 spin_unlock(&ctl
->tree_lock
);
7782 spin_unlock(&ctl
->tree_lock
);
7785 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
7786 num_bytes
, empty_size
,
7789 * If we didn't find a chunk, and we haven't failed on this
7790 * block group before, and this block group is in the middle of
7791 * caching and we are ok with waiting, then go ahead and wait
7792 * for progress to be made, and set failed_alloc to true.
7794 * If failed_alloc is true then we've already waited on this
7795 * block group once and should move on to the next block group.
7797 if (!offset
&& !failed_alloc
&& !cached
&&
7798 loop
> LOOP_CACHING_NOWAIT
) {
7799 wait_block_group_cache_progress(block_group
,
7800 num_bytes
+ empty_size
);
7801 failed_alloc
= true;
7802 goto have_block_group
;
7803 } else if (!offset
) {
7807 search_start
= ALIGN(offset
, fs_info
->stripesize
);
7809 /* move on to the next group */
7810 if (search_start
+ num_bytes
>
7811 block_group
->key
.objectid
+ block_group
->key
.offset
) {
7812 btrfs_add_free_space(block_group
, offset
, num_bytes
);
7816 if (offset
< search_start
)
7817 btrfs_add_free_space(block_group
, offset
,
7818 search_start
- offset
);
7819 BUG_ON(offset
> search_start
);
7821 ret
= btrfs_add_reserved_bytes(block_group
, ram_bytes
,
7822 num_bytes
, delalloc
);
7823 if (ret
== -EAGAIN
) {
7824 btrfs_add_free_space(block_group
, offset
, num_bytes
);
7827 btrfs_inc_block_group_reservations(block_group
);
7829 /* we are all good, lets return */
7830 ins
->objectid
= search_start
;
7831 ins
->offset
= num_bytes
;
7833 trace_btrfs_reserve_extent(fs_info
, block_group
,
7834 search_start
, num_bytes
);
7835 btrfs_release_block_group(block_group
, delalloc
);
7838 failed_cluster_refill
= false;
7839 failed_alloc
= false;
7840 BUG_ON(index
!= get_block_group_index(block_group
));
7841 btrfs_release_block_group(block_group
, delalloc
);
7844 up_read(&space_info
->groups_sem
);
7846 if ((loop
== LOOP_CACHING_NOWAIT
) && have_caching_bg
7847 && !orig_have_caching_bg
)
7848 orig_have_caching_bg
= true;
7850 if (!ins
->objectid
&& loop
>= LOOP_CACHING_WAIT
&& have_caching_bg
)
7853 if (!ins
->objectid
&& ++index
< BTRFS_NR_RAID_TYPES
)
7857 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7858 * caching kthreads as we move along
7859 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7860 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7861 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7864 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
) {
7866 if (loop
== LOOP_CACHING_NOWAIT
) {
7868 * We want to skip the LOOP_CACHING_WAIT step if we
7869 * don't have any uncached bgs and we've already done a
7870 * full search through.
7872 if (orig_have_caching_bg
|| !full_search
)
7873 loop
= LOOP_CACHING_WAIT
;
7875 loop
= LOOP_ALLOC_CHUNK
;
7880 if (loop
== LOOP_ALLOC_CHUNK
) {
7881 struct btrfs_trans_handle
*trans
;
7884 trans
= current
->journal_info
;
7888 trans
= btrfs_join_transaction(root
);
7890 if (IS_ERR(trans
)) {
7891 ret
= PTR_ERR(trans
);
7895 ret
= do_chunk_alloc(trans
, fs_info
, flags
,
7899 * If we can't allocate a new chunk we've already looped
7900 * through at least once, move on to the NO_EMPTY_SIZE
7904 loop
= LOOP_NO_EMPTY_SIZE
;
7907 * Do not bail out on ENOSPC since we
7908 * can do more things.
7910 if (ret
< 0 && ret
!= -ENOSPC
)
7911 btrfs_abort_transaction(trans
, ret
);
7915 btrfs_end_transaction(trans
);
7920 if (loop
== LOOP_NO_EMPTY_SIZE
) {
7922 * Don't loop again if we already have no empty_size and
7925 if (empty_size
== 0 &&
7926 empty_cluster
== 0) {
7935 } else if (!ins
->objectid
) {
7937 } else if (ins
->objectid
) {
7938 if (!use_cluster
&& last_ptr
) {
7939 spin_lock(&last_ptr
->lock
);
7940 last_ptr
->window_start
= ins
->objectid
;
7941 spin_unlock(&last_ptr
->lock
);
7946 if (ret
== -ENOSPC
) {
7947 spin_lock(&space_info
->lock
);
7948 space_info
->max_extent_size
= max_extent_size
;
7949 spin_unlock(&space_info
->lock
);
7950 ins
->offset
= max_extent_size
;
7955 static void dump_space_info(struct btrfs_fs_info
*fs_info
,
7956 struct btrfs_space_info
*info
, u64 bytes
,
7957 int dump_block_groups
)
7959 struct btrfs_block_group_cache
*cache
;
7962 spin_lock(&info
->lock
);
7963 btrfs_info(fs_info
, "space_info %llu has %llu free, is %sfull",
7965 info
->total_bytes
- btrfs_space_info_used(info
, true),
7966 info
->full
? "" : "not ");
7968 "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu",
7969 info
->total_bytes
, info
->bytes_used
, info
->bytes_pinned
,
7970 info
->bytes_reserved
, info
->bytes_may_use
,
7971 info
->bytes_readonly
);
7972 spin_unlock(&info
->lock
);
7974 if (!dump_block_groups
)
7977 down_read(&info
->groups_sem
);
7979 list_for_each_entry(cache
, &info
->block_groups
[index
], list
) {
7980 spin_lock(&cache
->lock
);
7982 "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s",
7983 cache
->key
.objectid
, cache
->key
.offset
,
7984 btrfs_block_group_used(&cache
->item
), cache
->pinned
,
7985 cache
->reserved
, cache
->ro
? "[readonly]" : "");
7986 btrfs_dump_free_space(cache
, bytes
);
7987 spin_unlock(&cache
->lock
);
7989 if (++index
< BTRFS_NR_RAID_TYPES
)
7991 up_read(&info
->groups_sem
);
7994 int btrfs_reserve_extent(struct btrfs_root
*root
, u64 ram_bytes
,
7995 u64 num_bytes
, u64 min_alloc_size
,
7996 u64 empty_size
, u64 hint_byte
,
7997 struct btrfs_key
*ins
, int is_data
, int delalloc
)
7999 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8000 bool final_tried
= num_bytes
== min_alloc_size
;
8004 flags
= get_alloc_profile_by_root(root
, is_data
);
8006 WARN_ON(num_bytes
< fs_info
->sectorsize
);
8007 ret
= find_free_extent(fs_info
, ram_bytes
, num_bytes
, empty_size
,
8008 hint_byte
, ins
, flags
, delalloc
);
8009 if (!ret
&& !is_data
) {
8010 btrfs_dec_block_group_reservations(fs_info
, ins
->objectid
);
8011 } else if (ret
== -ENOSPC
) {
8012 if (!final_tried
&& ins
->offset
) {
8013 num_bytes
= min(num_bytes
>> 1, ins
->offset
);
8014 num_bytes
= round_down(num_bytes
,
8015 fs_info
->sectorsize
);
8016 num_bytes
= max(num_bytes
, min_alloc_size
);
8017 ram_bytes
= num_bytes
;
8018 if (num_bytes
== min_alloc_size
)
8021 } else if (btrfs_test_opt(fs_info
, ENOSPC_DEBUG
)) {
8022 struct btrfs_space_info
*sinfo
;
8024 sinfo
= __find_space_info(fs_info
, flags
);
8026 "allocation failed flags %llu, wanted %llu",
8029 dump_space_info(fs_info
, sinfo
, num_bytes
, 1);
8036 static int __btrfs_free_reserved_extent(struct btrfs_fs_info
*fs_info
,
8038 int pin
, int delalloc
)
8040 struct btrfs_block_group_cache
*cache
;
8043 cache
= btrfs_lookup_block_group(fs_info
, start
);
8045 btrfs_err(fs_info
, "Unable to find block group for %llu",
8051 pin_down_extent(fs_info
, cache
, start
, len
, 1);
8053 if (btrfs_test_opt(fs_info
, DISCARD
))
8054 ret
= btrfs_discard_extent(fs_info
, start
, len
, NULL
);
8055 btrfs_add_free_space(cache
, start
, len
);
8056 btrfs_free_reserved_bytes(cache
, len
, delalloc
);
8057 trace_btrfs_reserved_extent_free(fs_info
, start
, len
);
8060 btrfs_put_block_group(cache
);
8064 int btrfs_free_reserved_extent(struct btrfs_fs_info
*fs_info
,
8065 u64 start
, u64 len
, int delalloc
)
8067 return __btrfs_free_reserved_extent(fs_info
, start
, len
, 0, delalloc
);
8070 int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info
*fs_info
,
8073 return __btrfs_free_reserved_extent(fs_info
, start
, len
, 1, 0);
8076 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
8077 struct btrfs_fs_info
*fs_info
,
8078 u64 parent
, u64 root_objectid
,
8079 u64 flags
, u64 owner
, u64 offset
,
8080 struct btrfs_key
*ins
, int ref_mod
)
8083 struct btrfs_extent_item
*extent_item
;
8084 struct btrfs_extent_inline_ref
*iref
;
8085 struct btrfs_path
*path
;
8086 struct extent_buffer
*leaf
;
8091 type
= BTRFS_SHARED_DATA_REF_KEY
;
8093 type
= BTRFS_EXTENT_DATA_REF_KEY
;
8095 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
8097 path
= btrfs_alloc_path();
8101 path
->leave_spinning
= 1;
8102 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
8105 btrfs_free_path(path
);
8109 leaf
= path
->nodes
[0];
8110 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
8111 struct btrfs_extent_item
);
8112 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
8113 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
8114 btrfs_set_extent_flags(leaf
, extent_item
,
8115 flags
| BTRFS_EXTENT_FLAG_DATA
);
8117 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
8118 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
8120 struct btrfs_shared_data_ref
*ref
;
8121 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
8122 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
8123 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
8125 struct btrfs_extent_data_ref
*ref
;
8126 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
8127 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
8128 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
8129 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
8130 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
8133 btrfs_mark_buffer_dirty(path
->nodes
[0]);
8134 btrfs_free_path(path
);
8136 ret
= remove_from_free_space_tree(trans
, fs_info
, ins
->objectid
,
8141 ret
= update_block_group(trans
, fs_info
, ins
->objectid
, ins
->offset
, 1);
8142 if (ret
) { /* -ENOENT, logic error */
8143 btrfs_err(fs_info
, "update block group failed for %llu %llu",
8144 ins
->objectid
, ins
->offset
);
8147 trace_btrfs_reserved_extent_alloc(fs_info
, ins
->objectid
, ins
->offset
);
8151 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
8152 struct btrfs_fs_info
*fs_info
,
8153 u64 parent
, u64 root_objectid
,
8154 u64 flags
, struct btrfs_disk_key
*key
,
8155 int level
, struct btrfs_key
*ins
)
8158 struct btrfs_extent_item
*extent_item
;
8159 struct btrfs_tree_block_info
*block_info
;
8160 struct btrfs_extent_inline_ref
*iref
;
8161 struct btrfs_path
*path
;
8162 struct extent_buffer
*leaf
;
8163 u32 size
= sizeof(*extent_item
) + sizeof(*iref
);
8164 u64 num_bytes
= ins
->offset
;
8165 bool skinny_metadata
= btrfs_fs_incompat(fs_info
, SKINNY_METADATA
);
8167 if (!skinny_metadata
)
8168 size
+= sizeof(*block_info
);
8170 path
= btrfs_alloc_path();
8172 btrfs_free_and_pin_reserved_extent(fs_info
, ins
->objectid
,
8177 path
->leave_spinning
= 1;
8178 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
8181 btrfs_free_path(path
);
8182 btrfs_free_and_pin_reserved_extent(fs_info
, ins
->objectid
,
8187 leaf
= path
->nodes
[0];
8188 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
8189 struct btrfs_extent_item
);
8190 btrfs_set_extent_refs(leaf
, extent_item
, 1);
8191 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
8192 btrfs_set_extent_flags(leaf
, extent_item
,
8193 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
8195 if (skinny_metadata
) {
8196 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
8197 num_bytes
= fs_info
->nodesize
;
8199 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
8200 btrfs_set_tree_block_key(leaf
, block_info
, key
);
8201 btrfs_set_tree_block_level(leaf
, block_info
, level
);
8202 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
8206 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
8207 btrfs_set_extent_inline_ref_type(leaf
, iref
,
8208 BTRFS_SHARED_BLOCK_REF_KEY
);
8209 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
8211 btrfs_set_extent_inline_ref_type(leaf
, iref
,
8212 BTRFS_TREE_BLOCK_REF_KEY
);
8213 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
8216 btrfs_mark_buffer_dirty(leaf
);
8217 btrfs_free_path(path
);
8219 ret
= remove_from_free_space_tree(trans
, fs_info
, ins
->objectid
,
8224 ret
= update_block_group(trans
, fs_info
, ins
->objectid
,
8225 fs_info
->nodesize
, 1);
8226 if (ret
) { /* -ENOENT, logic error */
8227 btrfs_err(fs_info
, "update block group failed for %llu %llu",
8228 ins
->objectid
, ins
->offset
);
8232 trace_btrfs_reserved_extent_alloc(fs_info
, ins
->objectid
,
8237 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
8238 struct btrfs_root
*root
, u64 owner
,
8239 u64 offset
, u64 ram_bytes
,
8240 struct btrfs_key
*ins
)
8242 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8245 BUG_ON(root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
);
8247 btrfs_ref_tree_mod(root
, ins
->objectid
, ins
->offset
, 0,
8248 root
->root_key
.objectid
, owner
, offset
,
8249 BTRFS_ADD_DELAYED_EXTENT
);
8251 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, ins
->objectid
,
8253 root
->root_key
.objectid
, owner
,
8255 BTRFS_ADD_DELAYED_EXTENT
, NULL
, NULL
);
8260 * this is used by the tree logging recovery code. It records that
8261 * an extent has been allocated and makes sure to clear the free
8262 * space cache bits as well
8264 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
8265 struct btrfs_fs_info
*fs_info
,
8266 u64 root_objectid
, u64 owner
, u64 offset
,
8267 struct btrfs_key
*ins
)
8270 struct btrfs_block_group_cache
*block_group
;
8271 struct btrfs_space_info
*space_info
;
8274 * Mixed block groups will exclude before processing the log so we only
8275 * need to do the exclude dance if this fs isn't mixed.
8277 if (!btrfs_fs_incompat(fs_info
, MIXED_GROUPS
)) {
8278 ret
= __exclude_logged_extent(fs_info
, ins
->objectid
,
8284 block_group
= btrfs_lookup_block_group(fs_info
, ins
->objectid
);
8288 space_info
= block_group
->space_info
;
8289 spin_lock(&space_info
->lock
);
8290 spin_lock(&block_group
->lock
);
8291 space_info
->bytes_reserved
+= ins
->offset
;
8292 block_group
->reserved
+= ins
->offset
;
8293 spin_unlock(&block_group
->lock
);
8294 spin_unlock(&space_info
->lock
);
8296 ret
= alloc_reserved_file_extent(trans
, fs_info
, 0, root_objectid
,
8297 0, owner
, offset
, ins
, 1);
8298 btrfs_put_block_group(block_group
);
8302 static struct extent_buffer
*
8303 btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
8304 u64 bytenr
, int level
)
8306 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8307 struct extent_buffer
*buf
;
8309 buf
= btrfs_find_create_tree_block(fs_info
, bytenr
);
8313 btrfs_set_header_generation(buf
, trans
->transid
);
8314 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, buf
, level
);
8315 btrfs_tree_lock(buf
);
8316 clean_tree_block(fs_info
, buf
);
8317 clear_bit(EXTENT_BUFFER_STALE
, &buf
->bflags
);
8319 btrfs_set_lock_blocking(buf
);
8320 set_extent_buffer_uptodate(buf
);
8322 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
8323 buf
->log_index
= root
->log_transid
% 2;
8325 * we allow two log transactions at a time, use different
8326 * EXENT bit to differentiate dirty pages.
8328 if (buf
->log_index
== 0)
8329 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
8330 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
8332 set_extent_new(&root
->dirty_log_pages
, buf
->start
,
8333 buf
->start
+ buf
->len
- 1);
8335 buf
->log_index
= -1;
8336 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
8337 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
8339 trans
->dirty
= true;
8340 /* this returns a buffer locked for blocking */
8344 static struct btrfs_block_rsv
*
8345 use_block_rsv(struct btrfs_trans_handle
*trans
,
8346 struct btrfs_root
*root
, u32 blocksize
)
8348 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8349 struct btrfs_block_rsv
*block_rsv
;
8350 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
8352 bool global_updated
= false;
8354 block_rsv
= get_block_rsv(trans
, root
);
8356 if (unlikely(block_rsv
->size
== 0))
8359 ret
= block_rsv_use_bytes(block_rsv
, blocksize
);
8363 if (block_rsv
->failfast
)
8364 return ERR_PTR(ret
);
8366 if (block_rsv
->type
== BTRFS_BLOCK_RSV_GLOBAL
&& !global_updated
) {
8367 global_updated
= true;
8368 update_global_block_rsv(fs_info
);
8372 if (btrfs_test_opt(fs_info
, ENOSPC_DEBUG
)) {
8373 static DEFINE_RATELIMIT_STATE(_rs
,
8374 DEFAULT_RATELIMIT_INTERVAL
* 10,
8375 /*DEFAULT_RATELIMIT_BURST*/ 1);
8376 if (__ratelimit(&_rs
))
8378 "BTRFS: block rsv returned %d\n", ret
);
8381 ret
= reserve_metadata_bytes(root
, block_rsv
, blocksize
,
8382 BTRFS_RESERVE_NO_FLUSH
);
8386 * If we couldn't reserve metadata bytes try and use some from
8387 * the global reserve if its space type is the same as the global
8390 if (block_rsv
->type
!= BTRFS_BLOCK_RSV_GLOBAL
&&
8391 block_rsv
->space_info
== global_rsv
->space_info
) {
8392 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
8396 return ERR_PTR(ret
);
8399 static void unuse_block_rsv(struct btrfs_fs_info
*fs_info
,
8400 struct btrfs_block_rsv
*block_rsv
, u32 blocksize
)
8402 block_rsv_add_bytes(block_rsv
, blocksize
, 0);
8403 block_rsv_release_bytes(fs_info
, block_rsv
, NULL
, 0);
8407 * finds a free extent and does all the dirty work required for allocation
8408 * returns the tree buffer or an ERR_PTR on error.
8410 struct extent_buffer
*btrfs_alloc_tree_block(struct btrfs_trans_handle
*trans
,
8411 struct btrfs_root
*root
,
8412 u64 parent
, u64 root_objectid
,
8413 const struct btrfs_disk_key
*key
,
8414 int level
, u64 hint
,
8417 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8418 struct btrfs_key ins
;
8419 struct btrfs_block_rsv
*block_rsv
;
8420 struct extent_buffer
*buf
;
8421 struct btrfs_delayed_extent_op
*extent_op
;
8424 u32 blocksize
= fs_info
->nodesize
;
8425 bool skinny_metadata
= btrfs_fs_incompat(fs_info
, SKINNY_METADATA
);
8427 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8428 if (btrfs_is_testing(fs_info
)) {
8429 buf
= btrfs_init_new_buffer(trans
, root
, root
->alloc_bytenr
,
8432 root
->alloc_bytenr
+= blocksize
;
8437 block_rsv
= use_block_rsv(trans
, root
, blocksize
);
8438 if (IS_ERR(block_rsv
))
8439 return ERR_CAST(block_rsv
);
8441 ret
= btrfs_reserve_extent(root
, blocksize
, blocksize
, blocksize
,
8442 empty_size
, hint
, &ins
, 0, 0);
8446 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
, level
);
8449 goto out_free_reserved
;
8452 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
8454 parent
= ins
.objectid
;
8455 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
8459 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
8460 extent_op
= btrfs_alloc_delayed_extent_op();
8466 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
8468 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
8469 extent_op
->flags_to_set
= flags
;
8470 extent_op
->update_key
= skinny_metadata
? false : true;
8471 extent_op
->update_flags
= true;
8472 extent_op
->is_data
= false;
8473 extent_op
->level
= level
;
8475 btrfs_ref_tree_mod(root
, ins
.objectid
, ins
.offset
, parent
,
8476 root_objectid
, level
, 0,
8477 BTRFS_ADD_DELAYED_EXTENT
);
8478 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, ins
.objectid
,
8480 root_objectid
, level
,
8481 BTRFS_ADD_DELAYED_EXTENT
,
8482 extent_op
, NULL
, NULL
);
8484 goto out_free_delayed
;
8489 btrfs_free_delayed_extent_op(extent_op
);
8491 free_extent_buffer(buf
);
8493 btrfs_free_reserved_extent(fs_info
, ins
.objectid
, ins
.offset
, 0);
8495 unuse_block_rsv(fs_info
, block_rsv
, blocksize
);
8496 return ERR_PTR(ret
);
8499 struct walk_control
{
8500 u64 refs
[BTRFS_MAX_LEVEL
];
8501 u64 flags
[BTRFS_MAX_LEVEL
];
8502 struct btrfs_key update_progress
;
8513 #define DROP_REFERENCE 1
8514 #define UPDATE_BACKREF 2
8516 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
8517 struct btrfs_root
*root
,
8518 struct walk_control
*wc
,
8519 struct btrfs_path
*path
)
8521 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8527 struct btrfs_key key
;
8528 struct extent_buffer
*eb
;
8533 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
8534 wc
->reada_count
= wc
->reada_count
* 2 / 3;
8535 wc
->reada_count
= max(wc
->reada_count
, 2);
8537 wc
->reada_count
= wc
->reada_count
* 3 / 2;
8538 wc
->reada_count
= min_t(int, wc
->reada_count
,
8539 BTRFS_NODEPTRS_PER_BLOCK(fs_info
));
8542 eb
= path
->nodes
[wc
->level
];
8543 nritems
= btrfs_header_nritems(eb
);
8545 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
8546 if (nread
>= wc
->reada_count
)
8550 bytenr
= btrfs_node_blockptr(eb
, slot
);
8551 generation
= btrfs_node_ptr_generation(eb
, slot
);
8553 if (slot
== path
->slots
[wc
->level
])
8556 if (wc
->stage
== UPDATE_BACKREF
&&
8557 generation
<= root
->root_key
.offset
)
8560 /* We don't lock the tree block, it's OK to be racy here */
8561 ret
= btrfs_lookup_extent_info(trans
, fs_info
, bytenr
,
8562 wc
->level
- 1, 1, &refs
,
8564 /* We don't care about errors in readahead. */
8569 if (wc
->stage
== DROP_REFERENCE
) {
8573 if (wc
->level
== 1 &&
8574 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8576 if (!wc
->update_ref
||
8577 generation
<= root
->root_key
.offset
)
8579 btrfs_node_key_to_cpu(eb
, &key
, slot
);
8580 ret
= btrfs_comp_cpu_keys(&key
,
8581 &wc
->update_progress
);
8585 if (wc
->level
== 1 &&
8586 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8590 readahead_tree_block(fs_info
, bytenr
);
8593 wc
->reada_slot
= slot
;
8597 * helper to process tree block while walking down the tree.
8599 * when wc->stage == UPDATE_BACKREF, this function updates
8600 * back refs for pointers in the block.
8602 * NOTE: return value 1 means we should stop walking down.
8604 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
8605 struct btrfs_root
*root
,
8606 struct btrfs_path
*path
,
8607 struct walk_control
*wc
, int lookup_info
)
8609 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8610 int level
= wc
->level
;
8611 struct extent_buffer
*eb
= path
->nodes
[level
];
8612 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
8615 if (wc
->stage
== UPDATE_BACKREF
&&
8616 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
8620 * when reference count of tree block is 1, it won't increase
8621 * again. once full backref flag is set, we never clear it.
8624 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
8625 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
8626 BUG_ON(!path
->locks
[level
]);
8627 ret
= btrfs_lookup_extent_info(trans
, fs_info
,
8628 eb
->start
, level
, 1,
8631 BUG_ON(ret
== -ENOMEM
);
8634 BUG_ON(wc
->refs
[level
] == 0);
8637 if (wc
->stage
== DROP_REFERENCE
) {
8638 if (wc
->refs
[level
] > 1)
8641 if (path
->locks
[level
] && !wc
->keep_locks
) {
8642 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8643 path
->locks
[level
] = 0;
8648 /* wc->stage == UPDATE_BACKREF */
8649 if (!(wc
->flags
[level
] & flag
)) {
8650 BUG_ON(!path
->locks
[level
]);
8651 ret
= btrfs_inc_ref(trans
, root
, eb
, 1);
8652 BUG_ON(ret
); /* -ENOMEM */
8653 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
8654 BUG_ON(ret
); /* -ENOMEM */
8655 ret
= btrfs_set_disk_extent_flags(trans
, fs_info
, eb
->start
,
8657 btrfs_header_level(eb
), 0);
8658 BUG_ON(ret
); /* -ENOMEM */
8659 wc
->flags
[level
] |= flag
;
8663 * the block is shared by multiple trees, so it's not good to
8664 * keep the tree lock
8666 if (path
->locks
[level
] && level
> 0) {
8667 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8668 path
->locks
[level
] = 0;
8674 * helper to process tree block pointer.
8676 * when wc->stage == DROP_REFERENCE, this function checks
8677 * reference count of the block pointed to. if the block
8678 * is shared and we need update back refs for the subtree
8679 * rooted at the block, this function changes wc->stage to
8680 * UPDATE_BACKREF. if the block is shared and there is no
8681 * need to update back, this function drops the reference
8684 * NOTE: return value 1 means we should stop walking down.
8686 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
8687 struct btrfs_root
*root
,
8688 struct btrfs_path
*path
,
8689 struct walk_control
*wc
, int *lookup_info
)
8691 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8696 struct btrfs_key key
;
8697 struct extent_buffer
*next
;
8698 int level
= wc
->level
;
8701 bool need_account
= false;
8703 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
8704 path
->slots
[level
]);
8706 * if the lower level block was created before the snapshot
8707 * was created, we know there is no need to update back refs
8710 if (wc
->stage
== UPDATE_BACKREF
&&
8711 generation
<= root
->root_key
.offset
) {
8716 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
8717 blocksize
= fs_info
->nodesize
;
8719 next
= find_extent_buffer(fs_info
, bytenr
);
8721 next
= btrfs_find_create_tree_block(fs_info
, bytenr
);
8723 return PTR_ERR(next
);
8725 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, next
,
8729 btrfs_tree_lock(next
);
8730 btrfs_set_lock_blocking(next
);
8732 ret
= btrfs_lookup_extent_info(trans
, fs_info
, bytenr
, level
- 1, 1,
8733 &wc
->refs
[level
- 1],
8734 &wc
->flags
[level
- 1]);
8738 if (unlikely(wc
->refs
[level
- 1] == 0)) {
8739 btrfs_err(fs_info
, "Missing references.");
8745 if (wc
->stage
== DROP_REFERENCE
) {
8746 if (wc
->refs
[level
- 1] > 1) {
8747 need_account
= true;
8749 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8752 if (!wc
->update_ref
||
8753 generation
<= root
->root_key
.offset
)
8756 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
8757 path
->slots
[level
]);
8758 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
8762 wc
->stage
= UPDATE_BACKREF
;
8763 wc
->shared_level
= level
- 1;
8767 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8771 if (!btrfs_buffer_uptodate(next
, generation
, 0)) {
8772 btrfs_tree_unlock(next
);
8773 free_extent_buffer(next
);
8779 if (reada
&& level
== 1)
8780 reada_walk_down(trans
, root
, wc
, path
);
8781 next
= read_tree_block(fs_info
, bytenr
, generation
);
8783 return PTR_ERR(next
);
8784 } else if (!extent_buffer_uptodate(next
)) {
8785 free_extent_buffer(next
);
8788 btrfs_tree_lock(next
);
8789 btrfs_set_lock_blocking(next
);
8793 ASSERT(level
== btrfs_header_level(next
));
8794 if (level
!= btrfs_header_level(next
)) {
8795 btrfs_err(root
->fs_info
, "mismatched level");
8799 path
->nodes
[level
] = next
;
8800 path
->slots
[level
] = 0;
8801 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8807 wc
->refs
[level
- 1] = 0;
8808 wc
->flags
[level
- 1] = 0;
8809 if (wc
->stage
== DROP_REFERENCE
) {
8810 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
8811 parent
= path
->nodes
[level
]->start
;
8813 ASSERT(root
->root_key
.objectid
==
8814 btrfs_header_owner(path
->nodes
[level
]));
8815 if (root
->root_key
.objectid
!=
8816 btrfs_header_owner(path
->nodes
[level
])) {
8817 btrfs_err(root
->fs_info
,
8818 "mismatched block owner");
8826 ret
= btrfs_qgroup_trace_subtree(trans
, root
, next
,
8827 generation
, level
- 1);
8829 btrfs_err_rl(fs_info
,
8830 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
8834 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
,
8835 parent
, root
->root_key
.objectid
,
8845 btrfs_tree_unlock(next
);
8846 free_extent_buffer(next
);
8852 * helper to process tree block while walking up the tree.
8854 * when wc->stage == DROP_REFERENCE, this function drops
8855 * reference count on the block.
8857 * when wc->stage == UPDATE_BACKREF, this function changes
8858 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8859 * to UPDATE_BACKREF previously while processing the block.
8861 * NOTE: return value 1 means we should stop walking up.
8863 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
8864 struct btrfs_root
*root
,
8865 struct btrfs_path
*path
,
8866 struct walk_control
*wc
)
8868 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8870 int level
= wc
->level
;
8871 struct extent_buffer
*eb
= path
->nodes
[level
];
8874 if (wc
->stage
== UPDATE_BACKREF
) {
8875 BUG_ON(wc
->shared_level
< level
);
8876 if (level
< wc
->shared_level
)
8879 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
8883 wc
->stage
= DROP_REFERENCE
;
8884 wc
->shared_level
= -1;
8885 path
->slots
[level
] = 0;
8888 * check reference count again if the block isn't locked.
8889 * we should start walking down the tree again if reference
8892 if (!path
->locks
[level
]) {
8894 btrfs_tree_lock(eb
);
8895 btrfs_set_lock_blocking(eb
);
8896 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8898 ret
= btrfs_lookup_extent_info(trans
, fs_info
,
8899 eb
->start
, level
, 1,
8903 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8904 path
->locks
[level
] = 0;
8907 BUG_ON(wc
->refs
[level
] == 0);
8908 if (wc
->refs
[level
] == 1) {
8909 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8910 path
->locks
[level
] = 0;
8916 /* wc->stage == DROP_REFERENCE */
8917 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
8919 if (wc
->refs
[level
] == 1) {
8921 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8922 ret
= btrfs_dec_ref(trans
, root
, eb
, 1);
8924 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
8925 BUG_ON(ret
); /* -ENOMEM */
8926 ret
= btrfs_qgroup_trace_leaf_items(trans
, fs_info
, eb
);
8928 btrfs_err_rl(fs_info
,
8929 "error %d accounting leaf items. Quota is out of sync, rescan required.",
8933 /* make block locked assertion in clean_tree_block happy */
8934 if (!path
->locks
[level
] &&
8935 btrfs_header_generation(eb
) == trans
->transid
) {
8936 btrfs_tree_lock(eb
);
8937 btrfs_set_lock_blocking(eb
);
8938 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8940 clean_tree_block(fs_info
, eb
);
8943 if (eb
== root
->node
) {
8944 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8947 BUG_ON(root
->root_key
.objectid
!=
8948 btrfs_header_owner(eb
));
8950 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8951 parent
= path
->nodes
[level
+ 1]->start
;
8953 BUG_ON(root
->root_key
.objectid
!=
8954 btrfs_header_owner(path
->nodes
[level
+ 1]));
8957 btrfs_free_tree_block(trans
, root
, eb
, parent
, wc
->refs
[level
] == 1);
8959 wc
->refs
[level
] = 0;
8960 wc
->flags
[level
] = 0;
8964 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
8965 struct btrfs_root
*root
,
8966 struct btrfs_path
*path
,
8967 struct walk_control
*wc
)
8969 int level
= wc
->level
;
8970 int lookup_info
= 1;
8973 while (level
>= 0) {
8974 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
8981 if (path
->slots
[level
] >=
8982 btrfs_header_nritems(path
->nodes
[level
]))
8985 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
8987 path
->slots
[level
]++;
8996 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
8997 struct btrfs_root
*root
,
8998 struct btrfs_path
*path
,
8999 struct walk_control
*wc
, int max_level
)
9001 int level
= wc
->level
;
9004 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
9005 while (level
< max_level
&& path
->nodes
[level
]) {
9007 if (path
->slots
[level
] + 1 <
9008 btrfs_header_nritems(path
->nodes
[level
])) {
9009 path
->slots
[level
]++;
9012 ret
= walk_up_proc(trans
, root
, path
, wc
);
9016 if (path
->locks
[level
]) {
9017 btrfs_tree_unlock_rw(path
->nodes
[level
],
9018 path
->locks
[level
]);
9019 path
->locks
[level
] = 0;
9021 free_extent_buffer(path
->nodes
[level
]);
9022 path
->nodes
[level
] = NULL
;
9030 * drop a subvolume tree.
9032 * this function traverses the tree freeing any blocks that only
9033 * referenced by the tree.
9035 * when a shared tree block is found. this function decreases its
9036 * reference count by one. if update_ref is true, this function
9037 * also make sure backrefs for the shared block and all lower level
9038 * blocks are properly updated.
9040 * If called with for_reloc == 0, may exit early with -EAGAIN
9042 int btrfs_drop_snapshot(struct btrfs_root
*root
,
9043 struct btrfs_block_rsv
*block_rsv
, int update_ref
,
9046 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
9047 struct btrfs_path
*path
;
9048 struct btrfs_trans_handle
*trans
;
9049 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
9050 struct btrfs_root_item
*root_item
= &root
->root_item
;
9051 struct walk_control
*wc
;
9052 struct btrfs_key key
;
9056 bool root_dropped
= false;
9058 btrfs_debug(fs_info
, "Drop subvolume %llu", root
->objectid
);
9060 path
= btrfs_alloc_path();
9066 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
9068 btrfs_free_path(path
);
9073 trans
= btrfs_start_transaction(tree_root
, 0);
9074 if (IS_ERR(trans
)) {
9075 err
= PTR_ERR(trans
);
9080 trans
->block_rsv
= block_rsv
;
9082 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
9083 level
= btrfs_header_level(root
->node
);
9084 path
->nodes
[level
] = btrfs_lock_root_node(root
);
9085 btrfs_set_lock_blocking(path
->nodes
[level
]);
9086 path
->slots
[level
] = 0;
9087 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
9088 memset(&wc
->update_progress
, 0,
9089 sizeof(wc
->update_progress
));
9091 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
9092 memcpy(&wc
->update_progress
, &key
,
9093 sizeof(wc
->update_progress
));
9095 level
= root_item
->drop_level
;
9097 path
->lowest_level
= level
;
9098 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
9099 path
->lowest_level
= 0;
9107 * unlock our path, this is safe because only this
9108 * function is allowed to delete this snapshot
9110 btrfs_unlock_up_safe(path
, 0);
9112 level
= btrfs_header_level(root
->node
);
9114 btrfs_tree_lock(path
->nodes
[level
]);
9115 btrfs_set_lock_blocking(path
->nodes
[level
]);
9116 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
9118 ret
= btrfs_lookup_extent_info(trans
, fs_info
,
9119 path
->nodes
[level
]->start
,
9120 level
, 1, &wc
->refs
[level
],
9126 BUG_ON(wc
->refs
[level
] == 0);
9128 if (level
== root_item
->drop_level
)
9131 btrfs_tree_unlock(path
->nodes
[level
]);
9132 path
->locks
[level
] = 0;
9133 WARN_ON(wc
->refs
[level
] != 1);
9139 wc
->shared_level
= -1;
9140 wc
->stage
= DROP_REFERENCE
;
9141 wc
->update_ref
= update_ref
;
9143 wc
->for_reloc
= for_reloc
;
9144 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(fs_info
);
9148 ret
= walk_down_tree(trans
, root
, path
, wc
);
9154 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
9161 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
9165 if (wc
->stage
== DROP_REFERENCE
) {
9167 btrfs_node_key(path
->nodes
[level
],
9168 &root_item
->drop_progress
,
9169 path
->slots
[level
]);
9170 root_item
->drop_level
= level
;
9173 BUG_ON(wc
->level
== 0);
9174 if (btrfs_should_end_transaction(trans
) ||
9175 (!for_reloc
&& btrfs_need_cleaner_sleep(fs_info
))) {
9176 ret
= btrfs_update_root(trans
, tree_root
,
9180 btrfs_abort_transaction(trans
, ret
);
9185 btrfs_end_transaction_throttle(trans
);
9186 if (!for_reloc
&& btrfs_need_cleaner_sleep(fs_info
)) {
9187 btrfs_debug(fs_info
,
9188 "drop snapshot early exit");
9193 trans
= btrfs_start_transaction(tree_root
, 0);
9194 if (IS_ERR(trans
)) {
9195 err
= PTR_ERR(trans
);
9199 trans
->block_rsv
= block_rsv
;
9202 btrfs_release_path(path
);
9206 ret
= btrfs_del_root(trans
, fs_info
, &root
->root_key
);
9208 btrfs_abort_transaction(trans
, ret
);
9212 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
9213 ret
= btrfs_find_root(tree_root
, &root
->root_key
, path
,
9216 btrfs_abort_transaction(trans
, ret
);
9219 } else if (ret
> 0) {
9220 /* if we fail to delete the orphan item this time
9221 * around, it'll get picked up the next time.
9223 * The most common failure here is just -ENOENT.
9225 btrfs_del_orphan_item(trans
, tree_root
,
9226 root
->root_key
.objectid
);
9230 if (test_bit(BTRFS_ROOT_IN_RADIX
, &root
->state
)) {
9231 btrfs_add_dropped_root(trans
, root
);
9233 free_extent_buffer(root
->node
);
9234 free_extent_buffer(root
->commit_root
);
9235 btrfs_put_fs_root(root
);
9237 root_dropped
= true;
9239 btrfs_end_transaction_throttle(trans
);
9242 btrfs_free_path(path
);
9245 * So if we need to stop dropping the snapshot for whatever reason we
9246 * need to make sure to add it back to the dead root list so that we
9247 * keep trying to do the work later. This also cleans up roots if we
9248 * don't have it in the radix (like when we recover after a power fail
9249 * or unmount) so we don't leak memory.
9251 if (!for_reloc
&& !root_dropped
)
9252 btrfs_add_dead_root(root
);
9253 if (err
&& err
!= -EAGAIN
)
9254 btrfs_handle_fs_error(fs_info
, err
, NULL
);
9259 * drop subtree rooted at tree block 'node'.
9261 * NOTE: this function will unlock and release tree block 'node'
9262 * only used by relocation code
9264 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
9265 struct btrfs_root
*root
,
9266 struct extent_buffer
*node
,
9267 struct extent_buffer
*parent
)
9269 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
9270 struct btrfs_path
*path
;
9271 struct walk_control
*wc
;
9277 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
9279 path
= btrfs_alloc_path();
9283 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
9285 btrfs_free_path(path
);
9289 btrfs_assert_tree_locked(parent
);
9290 parent_level
= btrfs_header_level(parent
);
9291 extent_buffer_get(parent
);
9292 path
->nodes
[parent_level
] = parent
;
9293 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
9295 btrfs_assert_tree_locked(node
);
9296 level
= btrfs_header_level(node
);
9297 path
->nodes
[level
] = node
;
9298 path
->slots
[level
] = 0;
9299 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
9301 wc
->refs
[parent_level
] = 1;
9302 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
9304 wc
->shared_level
= -1;
9305 wc
->stage
= DROP_REFERENCE
;
9309 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(fs_info
);
9312 wret
= walk_down_tree(trans
, root
, path
, wc
);
9318 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
9326 btrfs_free_path(path
);
9330 static u64
update_block_group_flags(struct btrfs_fs_info
*fs_info
, u64 flags
)
9336 * if restripe for this chunk_type is on pick target profile and
9337 * return, otherwise do the usual balance
9339 stripped
= get_restripe_target(fs_info
, flags
);
9341 return extended_to_chunk(stripped
);
9343 num_devices
= fs_info
->fs_devices
->rw_devices
;
9345 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
9346 BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
|
9347 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
9349 if (num_devices
== 1) {
9350 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
9351 stripped
= flags
& ~stripped
;
9353 /* turn raid0 into single device chunks */
9354 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
9357 /* turn mirroring into duplication */
9358 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
9359 BTRFS_BLOCK_GROUP_RAID10
))
9360 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
9362 /* they already had raid on here, just return */
9363 if (flags
& stripped
)
9366 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
9367 stripped
= flags
& ~stripped
;
9369 /* switch duplicated blocks with raid1 */
9370 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
9371 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
9373 /* this is drive concat, leave it alone */
9379 static int inc_block_group_ro(struct btrfs_block_group_cache
*cache
, int force
)
9381 struct btrfs_space_info
*sinfo
= cache
->space_info
;
9383 u64 min_allocable_bytes
;
9387 * We need some metadata space and system metadata space for
9388 * allocating chunks in some corner cases until we force to set
9389 * it to be readonly.
9392 (BTRFS_BLOCK_GROUP_SYSTEM
| BTRFS_BLOCK_GROUP_METADATA
)) &&
9394 min_allocable_bytes
= SZ_1M
;
9396 min_allocable_bytes
= 0;
9398 spin_lock(&sinfo
->lock
);
9399 spin_lock(&cache
->lock
);
9407 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
9408 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
9410 if (btrfs_space_info_used(sinfo
, true) + num_bytes
+
9411 min_allocable_bytes
<= sinfo
->total_bytes
) {
9412 sinfo
->bytes_readonly
+= num_bytes
;
9414 list_add_tail(&cache
->ro_list
, &sinfo
->ro_bgs
);
9418 spin_unlock(&cache
->lock
);
9419 spin_unlock(&sinfo
->lock
);
9423 int btrfs_inc_block_group_ro(struct btrfs_fs_info
*fs_info
,
9424 struct btrfs_block_group_cache
*cache
)
9427 struct btrfs_trans_handle
*trans
;
9432 trans
= btrfs_join_transaction(fs_info
->extent_root
);
9434 return PTR_ERR(trans
);
9437 * we're not allowed to set block groups readonly after the dirty
9438 * block groups cache has started writing. If it already started,
9439 * back off and let this transaction commit
9441 mutex_lock(&fs_info
->ro_block_group_mutex
);
9442 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN
, &trans
->transaction
->flags
)) {
9443 u64 transid
= trans
->transid
;
9445 mutex_unlock(&fs_info
->ro_block_group_mutex
);
9446 btrfs_end_transaction(trans
);
9448 ret
= btrfs_wait_for_commit(fs_info
, transid
);
9455 * if we are changing raid levels, try to allocate a corresponding
9456 * block group with the new raid level.
9458 alloc_flags
= update_block_group_flags(fs_info
, cache
->flags
);
9459 if (alloc_flags
!= cache
->flags
) {
9460 ret
= do_chunk_alloc(trans
, fs_info
, alloc_flags
,
9463 * ENOSPC is allowed here, we may have enough space
9464 * already allocated at the new raid level to
9473 ret
= inc_block_group_ro(cache
, 0);
9476 alloc_flags
= get_alloc_profile(fs_info
, cache
->space_info
->flags
);
9477 ret
= do_chunk_alloc(trans
, fs_info
, alloc_flags
,
9481 ret
= inc_block_group_ro(cache
, 0);
9483 if (cache
->flags
& BTRFS_BLOCK_GROUP_SYSTEM
) {
9484 alloc_flags
= update_block_group_flags(fs_info
, cache
->flags
);
9485 mutex_lock(&fs_info
->chunk_mutex
);
9486 check_system_chunk(trans
, fs_info
, alloc_flags
);
9487 mutex_unlock(&fs_info
->chunk_mutex
);
9489 mutex_unlock(&fs_info
->ro_block_group_mutex
);
9491 btrfs_end_transaction(trans
);
9495 int btrfs_force_chunk_alloc(struct btrfs_trans_handle
*trans
,
9496 struct btrfs_fs_info
*fs_info
, u64 type
)
9498 u64 alloc_flags
= get_alloc_profile(fs_info
, type
);
9500 return do_chunk_alloc(trans
, fs_info
, alloc_flags
, CHUNK_ALLOC_FORCE
);
9504 * helper to account the unused space of all the readonly block group in the
9505 * space_info. takes mirrors into account.
9507 u64
btrfs_account_ro_block_groups_free_space(struct btrfs_space_info
*sinfo
)
9509 struct btrfs_block_group_cache
*block_group
;
9513 /* It's df, we don't care if it's racy */
9514 if (list_empty(&sinfo
->ro_bgs
))
9517 spin_lock(&sinfo
->lock
);
9518 list_for_each_entry(block_group
, &sinfo
->ro_bgs
, ro_list
) {
9519 spin_lock(&block_group
->lock
);
9521 if (!block_group
->ro
) {
9522 spin_unlock(&block_group
->lock
);
9526 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_RAID1
|
9527 BTRFS_BLOCK_GROUP_RAID10
|
9528 BTRFS_BLOCK_GROUP_DUP
))
9533 free_bytes
+= (block_group
->key
.offset
-
9534 btrfs_block_group_used(&block_group
->item
)) *
9537 spin_unlock(&block_group
->lock
);
9539 spin_unlock(&sinfo
->lock
);
9544 void btrfs_dec_block_group_ro(struct btrfs_block_group_cache
*cache
)
9546 struct btrfs_space_info
*sinfo
= cache
->space_info
;
9551 spin_lock(&sinfo
->lock
);
9552 spin_lock(&cache
->lock
);
9554 num_bytes
= cache
->key
.offset
- cache
->reserved
-
9555 cache
->pinned
- cache
->bytes_super
-
9556 btrfs_block_group_used(&cache
->item
);
9557 sinfo
->bytes_readonly
-= num_bytes
;
9558 list_del_init(&cache
->ro_list
);
9560 spin_unlock(&cache
->lock
);
9561 spin_unlock(&sinfo
->lock
);
9565 * checks to see if its even possible to relocate this block group.
9567 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9568 * ok to go ahead and try.
9570 int btrfs_can_relocate(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
9572 struct btrfs_root
*root
= fs_info
->extent_root
;
9573 struct btrfs_block_group_cache
*block_group
;
9574 struct btrfs_space_info
*space_info
;
9575 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
9576 struct btrfs_device
*device
;
9577 struct btrfs_trans_handle
*trans
;
9587 debug
= btrfs_test_opt(fs_info
, ENOSPC_DEBUG
);
9589 block_group
= btrfs_lookup_block_group(fs_info
, bytenr
);
9591 /* odd, couldn't find the block group, leave it alone */
9595 "can't find block group for bytenr %llu",
9600 min_free
= btrfs_block_group_used(&block_group
->item
);
9602 /* no bytes used, we're good */
9606 space_info
= block_group
->space_info
;
9607 spin_lock(&space_info
->lock
);
9609 full
= space_info
->full
;
9612 * if this is the last block group we have in this space, we can't
9613 * relocate it unless we're able to allocate a new chunk below.
9615 * Otherwise, we need to make sure we have room in the space to handle
9616 * all of the extents from this block group. If we can, we're good
9618 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
9619 (btrfs_space_info_used(space_info
, false) + min_free
<
9620 space_info
->total_bytes
)) {
9621 spin_unlock(&space_info
->lock
);
9624 spin_unlock(&space_info
->lock
);
9627 * ok we don't have enough space, but maybe we have free space on our
9628 * devices to allocate new chunks for relocation, so loop through our
9629 * alloc devices and guess if we have enough space. if this block
9630 * group is going to be restriped, run checks against the target
9631 * profile instead of the current one.
9643 target
= get_restripe_target(fs_info
, block_group
->flags
);
9645 index
= __get_raid_index(extended_to_chunk(target
));
9648 * this is just a balance, so if we were marked as full
9649 * we know there is no space for a new chunk
9654 "no space to alloc new chunk for block group %llu",
9655 block_group
->key
.objectid
);
9659 index
= get_block_group_index(block_group
);
9662 if (index
== BTRFS_RAID_RAID10
) {
9666 } else if (index
== BTRFS_RAID_RAID1
) {
9668 } else if (index
== BTRFS_RAID_DUP
) {
9671 } else if (index
== BTRFS_RAID_RAID0
) {
9672 dev_min
= fs_devices
->rw_devices
;
9673 min_free
= div64_u64(min_free
, dev_min
);
9676 /* We need to do this so that we can look at pending chunks */
9677 trans
= btrfs_join_transaction(root
);
9678 if (IS_ERR(trans
)) {
9679 ret
= PTR_ERR(trans
);
9683 mutex_lock(&fs_info
->chunk_mutex
);
9684 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
9688 * check to make sure we can actually find a chunk with enough
9689 * space to fit our block group in.
9691 if (device
->total_bytes
> device
->bytes_used
+ min_free
&&
9692 !device
->is_tgtdev_for_dev_replace
) {
9693 ret
= find_free_dev_extent(trans
, device
, min_free
,
9698 if (dev_nr
>= dev_min
)
9704 if (debug
&& ret
== -1)
9706 "no space to allocate a new chunk for block group %llu",
9707 block_group
->key
.objectid
);
9708 mutex_unlock(&fs_info
->chunk_mutex
);
9709 btrfs_end_transaction(trans
);
9711 btrfs_put_block_group(block_group
);
9715 static int find_first_block_group(struct btrfs_fs_info
*fs_info
,
9716 struct btrfs_path
*path
,
9717 struct btrfs_key
*key
)
9719 struct btrfs_root
*root
= fs_info
->extent_root
;
9721 struct btrfs_key found_key
;
9722 struct extent_buffer
*leaf
;
9725 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
9730 slot
= path
->slots
[0];
9731 leaf
= path
->nodes
[0];
9732 if (slot
>= btrfs_header_nritems(leaf
)) {
9733 ret
= btrfs_next_leaf(root
, path
);
9740 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
9742 if (found_key
.objectid
>= key
->objectid
&&
9743 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
9744 struct extent_map_tree
*em_tree
;
9745 struct extent_map
*em
;
9747 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
9748 read_lock(&em_tree
->lock
);
9749 em
= lookup_extent_mapping(em_tree
, found_key
.objectid
,
9751 read_unlock(&em_tree
->lock
);
9754 "logical %llu len %llu found bg but no related chunk",
9755 found_key
.objectid
, found_key
.offset
);
9760 free_extent_map(em
);
9769 void btrfs_put_block_group_cache(struct btrfs_fs_info
*info
)
9771 struct btrfs_block_group_cache
*block_group
;
9775 struct inode
*inode
;
9777 block_group
= btrfs_lookup_first_block_group(info
, last
);
9778 while (block_group
) {
9779 spin_lock(&block_group
->lock
);
9780 if (block_group
->iref
)
9782 spin_unlock(&block_group
->lock
);
9783 block_group
= next_block_group(info
, block_group
);
9792 inode
= block_group
->inode
;
9793 block_group
->iref
= 0;
9794 block_group
->inode
= NULL
;
9795 spin_unlock(&block_group
->lock
);
9796 ASSERT(block_group
->io_ctl
.inode
== NULL
);
9798 last
= block_group
->key
.objectid
+ block_group
->key
.offset
;
9799 btrfs_put_block_group(block_group
);
9804 * Must be called only after stopping all workers, since we could have block
9805 * group caching kthreads running, and therefore they could race with us if we
9806 * freed the block groups before stopping them.
9808 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
9810 struct btrfs_block_group_cache
*block_group
;
9811 struct btrfs_space_info
*space_info
;
9812 struct btrfs_caching_control
*caching_ctl
;
9815 down_write(&info
->commit_root_sem
);
9816 while (!list_empty(&info
->caching_block_groups
)) {
9817 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
9818 struct btrfs_caching_control
, list
);
9819 list_del(&caching_ctl
->list
);
9820 put_caching_control(caching_ctl
);
9822 up_write(&info
->commit_root_sem
);
9824 spin_lock(&info
->unused_bgs_lock
);
9825 while (!list_empty(&info
->unused_bgs
)) {
9826 block_group
= list_first_entry(&info
->unused_bgs
,
9827 struct btrfs_block_group_cache
,
9829 list_del_init(&block_group
->bg_list
);
9830 btrfs_put_block_group(block_group
);
9832 spin_unlock(&info
->unused_bgs_lock
);
9834 spin_lock(&info
->block_group_cache_lock
);
9835 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
9836 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
9838 rb_erase(&block_group
->cache_node
,
9839 &info
->block_group_cache_tree
);
9840 RB_CLEAR_NODE(&block_group
->cache_node
);
9841 spin_unlock(&info
->block_group_cache_lock
);
9843 down_write(&block_group
->space_info
->groups_sem
);
9844 list_del(&block_group
->list
);
9845 up_write(&block_group
->space_info
->groups_sem
);
9848 * We haven't cached this block group, which means we could
9849 * possibly have excluded extents on this block group.
9851 if (block_group
->cached
== BTRFS_CACHE_NO
||
9852 block_group
->cached
== BTRFS_CACHE_ERROR
)
9853 free_excluded_extents(info
, block_group
);
9855 btrfs_remove_free_space_cache(block_group
);
9856 ASSERT(block_group
->cached
!= BTRFS_CACHE_STARTED
);
9857 ASSERT(list_empty(&block_group
->dirty_list
));
9858 ASSERT(list_empty(&block_group
->io_list
));
9859 ASSERT(list_empty(&block_group
->bg_list
));
9860 ASSERT(atomic_read(&block_group
->count
) == 1);
9861 btrfs_put_block_group(block_group
);
9863 spin_lock(&info
->block_group_cache_lock
);
9865 spin_unlock(&info
->block_group_cache_lock
);
9867 /* now that all the block groups are freed, go through and
9868 * free all the space_info structs. This is only called during
9869 * the final stages of unmount, and so we know nobody is
9870 * using them. We call synchronize_rcu() once before we start,
9871 * just to be on the safe side.
9875 release_global_block_rsv(info
);
9877 while (!list_empty(&info
->space_info
)) {
9880 space_info
= list_entry(info
->space_info
.next
,
9881 struct btrfs_space_info
,
9885 * Do not hide this behind enospc_debug, this is actually
9886 * important and indicates a real bug if this happens.
9888 if (WARN_ON(space_info
->bytes_pinned
> 0 ||
9889 space_info
->bytes_reserved
> 0 ||
9890 space_info
->bytes_may_use
> 0))
9891 dump_space_info(info
, space_info
, 0, 0);
9892 list_del(&space_info
->list
);
9893 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++) {
9894 struct kobject
*kobj
;
9895 kobj
= space_info
->block_group_kobjs
[i
];
9896 space_info
->block_group_kobjs
[i
] = NULL
;
9902 kobject_del(&space_info
->kobj
);
9903 kobject_put(&space_info
->kobj
);
9908 static void link_block_group(struct btrfs_block_group_cache
*cache
)
9910 struct btrfs_space_info
*space_info
= cache
->space_info
;
9911 int index
= get_block_group_index(cache
);
9914 down_write(&space_info
->groups_sem
);
9915 if (list_empty(&space_info
->block_groups
[index
]))
9917 list_add_tail(&cache
->list
, &space_info
->block_groups
[index
]);
9918 up_write(&space_info
->groups_sem
);
9921 struct raid_kobject
*rkobj
;
9924 rkobj
= kzalloc(sizeof(*rkobj
), GFP_NOFS
);
9927 rkobj
->raid_type
= index
;
9928 kobject_init(&rkobj
->kobj
, &btrfs_raid_ktype
);
9929 ret
= kobject_add(&rkobj
->kobj
, &space_info
->kobj
,
9930 "%s", get_raid_name(index
));
9932 kobject_put(&rkobj
->kobj
);
9935 space_info
->block_group_kobjs
[index
] = &rkobj
->kobj
;
9940 btrfs_warn(cache
->fs_info
,
9941 "failed to add kobject for block cache, ignoring");
9944 static struct btrfs_block_group_cache
*
9945 btrfs_create_block_group_cache(struct btrfs_fs_info
*fs_info
,
9946 u64 start
, u64 size
)
9948 struct btrfs_block_group_cache
*cache
;
9950 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
9954 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
9956 if (!cache
->free_space_ctl
) {
9961 cache
->key
.objectid
= start
;
9962 cache
->key
.offset
= size
;
9963 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
9965 cache
->fs_info
= fs_info
;
9966 cache
->full_stripe_len
= btrfs_full_stripe_len(fs_info
, start
);
9967 set_free_space_tree_thresholds(cache
);
9969 atomic_set(&cache
->count
, 1);
9970 spin_lock_init(&cache
->lock
);
9971 init_rwsem(&cache
->data_rwsem
);
9972 INIT_LIST_HEAD(&cache
->list
);
9973 INIT_LIST_HEAD(&cache
->cluster_list
);
9974 INIT_LIST_HEAD(&cache
->bg_list
);
9975 INIT_LIST_HEAD(&cache
->ro_list
);
9976 INIT_LIST_HEAD(&cache
->dirty_list
);
9977 INIT_LIST_HEAD(&cache
->io_list
);
9978 btrfs_init_free_space_ctl(cache
);
9979 atomic_set(&cache
->trimming
, 0);
9980 mutex_init(&cache
->free_space_lock
);
9981 btrfs_init_full_stripe_locks_tree(&cache
->full_stripe_locks_root
);
9986 int btrfs_read_block_groups(struct btrfs_fs_info
*info
)
9988 struct btrfs_path
*path
;
9990 struct btrfs_block_group_cache
*cache
;
9991 struct btrfs_space_info
*space_info
;
9992 struct btrfs_key key
;
9993 struct btrfs_key found_key
;
9994 struct extent_buffer
*leaf
;
10000 feature
= btrfs_super_incompat_flags(info
->super_copy
);
10001 mixed
= !!(feature
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
);
10005 key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
10006 path
= btrfs_alloc_path();
10009 path
->reada
= READA_FORWARD
;
10011 cache_gen
= btrfs_super_cache_generation(info
->super_copy
);
10012 if (btrfs_test_opt(info
, SPACE_CACHE
) &&
10013 btrfs_super_generation(info
->super_copy
) != cache_gen
)
10015 if (btrfs_test_opt(info
, CLEAR_CACHE
))
10019 ret
= find_first_block_group(info
, path
, &key
);
10025 leaf
= path
->nodes
[0];
10026 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
10028 cache
= btrfs_create_block_group_cache(info
, found_key
.objectid
,
10037 * When we mount with old space cache, we need to
10038 * set BTRFS_DC_CLEAR and set dirty flag.
10040 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
10041 * truncate the old free space cache inode and
10043 * b) Setting 'dirty flag' makes sure that we flush
10044 * the new space cache info onto disk.
10046 if (btrfs_test_opt(info
, SPACE_CACHE
))
10047 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
10050 read_extent_buffer(leaf
, &cache
->item
,
10051 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
10052 sizeof(cache
->item
));
10053 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
10055 ((cache
->flags
& BTRFS_BLOCK_GROUP_METADATA
) &&
10056 (cache
->flags
& BTRFS_BLOCK_GROUP_DATA
))) {
10058 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
10059 cache
->key
.objectid
);
10064 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
10065 btrfs_release_path(path
);
10068 * We need to exclude the super stripes now so that the space
10069 * info has super bytes accounted for, otherwise we'll think
10070 * we have more space than we actually do.
10072 ret
= exclude_super_stripes(info
, cache
);
10075 * We may have excluded something, so call this just in
10078 free_excluded_extents(info
, cache
);
10079 btrfs_put_block_group(cache
);
10084 * check for two cases, either we are full, and therefore
10085 * don't need to bother with the caching work since we won't
10086 * find any space, or we are empty, and we can just add all
10087 * the space in and be done with it. This saves us _alot_ of
10088 * time, particularly in the full case.
10090 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
10091 cache
->last_byte_to_unpin
= (u64
)-1;
10092 cache
->cached
= BTRFS_CACHE_FINISHED
;
10093 free_excluded_extents(info
, cache
);
10094 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
10095 cache
->last_byte_to_unpin
= (u64
)-1;
10096 cache
->cached
= BTRFS_CACHE_FINISHED
;
10097 add_new_free_space(cache
, info
,
10098 found_key
.objectid
,
10099 found_key
.objectid
+
10101 free_excluded_extents(info
, cache
);
10104 ret
= btrfs_add_block_group_cache(info
, cache
);
10106 btrfs_remove_free_space_cache(cache
);
10107 btrfs_put_block_group(cache
);
10111 trace_btrfs_add_block_group(info
, cache
, 0);
10112 update_space_info(info
, cache
->flags
, found_key
.offset
,
10113 btrfs_block_group_used(&cache
->item
),
10114 cache
->bytes_super
, &space_info
);
10116 cache
->space_info
= space_info
;
10118 link_block_group(cache
);
10120 set_avail_alloc_bits(info
, cache
->flags
);
10121 if (btrfs_chunk_readonly(info
, cache
->key
.objectid
)) {
10122 inc_block_group_ro(cache
, 1);
10123 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
10124 spin_lock(&info
->unused_bgs_lock
);
10125 /* Should always be true but just in case. */
10126 if (list_empty(&cache
->bg_list
)) {
10127 btrfs_get_block_group(cache
);
10128 list_add_tail(&cache
->bg_list
,
10129 &info
->unused_bgs
);
10131 spin_unlock(&info
->unused_bgs_lock
);
10135 list_for_each_entry_rcu(space_info
, &info
->space_info
, list
) {
10136 if (!(get_alloc_profile(info
, space_info
->flags
) &
10137 (BTRFS_BLOCK_GROUP_RAID10
|
10138 BTRFS_BLOCK_GROUP_RAID1
|
10139 BTRFS_BLOCK_GROUP_RAID5
|
10140 BTRFS_BLOCK_GROUP_RAID6
|
10141 BTRFS_BLOCK_GROUP_DUP
)))
10144 * avoid allocating from un-mirrored block group if there are
10145 * mirrored block groups.
10147 list_for_each_entry(cache
,
10148 &space_info
->block_groups
[BTRFS_RAID_RAID0
],
10150 inc_block_group_ro(cache
, 1);
10151 list_for_each_entry(cache
,
10152 &space_info
->block_groups
[BTRFS_RAID_SINGLE
],
10154 inc_block_group_ro(cache
, 1);
10157 init_global_block_rsv(info
);
10160 btrfs_free_path(path
);
10164 void btrfs_create_pending_block_groups(struct btrfs_trans_handle
*trans
,
10165 struct btrfs_fs_info
*fs_info
)
10167 struct btrfs_block_group_cache
*block_group
, *tmp
;
10168 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
10169 struct btrfs_block_group_item item
;
10170 struct btrfs_key key
;
10172 bool can_flush_pending_bgs
= trans
->can_flush_pending_bgs
;
10174 trans
->can_flush_pending_bgs
= false;
10175 list_for_each_entry_safe(block_group
, tmp
, &trans
->new_bgs
, bg_list
) {
10179 spin_lock(&block_group
->lock
);
10180 memcpy(&item
, &block_group
->item
, sizeof(item
));
10181 memcpy(&key
, &block_group
->key
, sizeof(key
));
10182 spin_unlock(&block_group
->lock
);
10184 ret
= btrfs_insert_item(trans
, extent_root
, &key
, &item
,
10187 btrfs_abort_transaction(trans
, ret
);
10188 ret
= btrfs_finish_chunk_alloc(trans
, fs_info
, key
.objectid
,
10191 btrfs_abort_transaction(trans
, ret
);
10192 add_block_group_free_space(trans
, fs_info
, block_group
);
10193 /* already aborted the transaction if it failed. */
10195 list_del_init(&block_group
->bg_list
);
10197 trans
->can_flush_pending_bgs
= can_flush_pending_bgs
;
10200 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
10201 struct btrfs_fs_info
*fs_info
, u64 bytes_used
,
10202 u64 type
, u64 chunk_offset
, u64 size
)
10204 struct btrfs_block_group_cache
*cache
;
10207 btrfs_set_log_full_commit(fs_info
, trans
);
10209 cache
= btrfs_create_block_group_cache(fs_info
, chunk_offset
, size
);
10213 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
10214 btrfs_set_block_group_chunk_objectid(&cache
->item
,
10215 BTRFS_FIRST_CHUNK_TREE_OBJECTID
);
10216 btrfs_set_block_group_flags(&cache
->item
, type
);
10218 cache
->flags
= type
;
10219 cache
->last_byte_to_unpin
= (u64
)-1;
10220 cache
->cached
= BTRFS_CACHE_FINISHED
;
10221 cache
->needs_free_space
= 1;
10222 ret
= exclude_super_stripes(fs_info
, cache
);
10225 * We may have excluded something, so call this just in
10228 free_excluded_extents(fs_info
, cache
);
10229 btrfs_put_block_group(cache
);
10233 add_new_free_space(cache
, fs_info
, chunk_offset
, chunk_offset
+ size
);
10235 free_excluded_extents(fs_info
, cache
);
10237 #ifdef CONFIG_BTRFS_DEBUG
10238 if (btrfs_should_fragment_free_space(cache
)) {
10239 u64 new_bytes_used
= size
- bytes_used
;
10241 bytes_used
+= new_bytes_used
>> 1;
10242 fragment_free_space(cache
);
10246 * Ensure the corresponding space_info object is created and
10247 * assigned to our block group. We want our bg to be added to the rbtree
10248 * with its ->space_info set.
10250 cache
->space_info
= __find_space_info(fs_info
, cache
->flags
);
10251 if (!cache
->space_info
) {
10252 ret
= create_space_info(fs_info
, cache
->flags
,
10253 &cache
->space_info
);
10255 btrfs_remove_free_space_cache(cache
);
10256 btrfs_put_block_group(cache
);
10261 ret
= btrfs_add_block_group_cache(fs_info
, cache
);
10263 btrfs_remove_free_space_cache(cache
);
10264 btrfs_put_block_group(cache
);
10269 * Now that our block group has its ->space_info set and is inserted in
10270 * the rbtree, update the space info's counters.
10272 trace_btrfs_add_block_group(fs_info
, cache
, 1);
10273 update_space_info(fs_info
, cache
->flags
, size
, bytes_used
,
10274 cache
->bytes_super
, &cache
->space_info
);
10275 update_global_block_rsv(fs_info
);
10277 link_block_group(cache
);
10279 list_add_tail(&cache
->bg_list
, &trans
->new_bgs
);
10281 set_avail_alloc_bits(fs_info
, type
);
10285 static void clear_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
10287 u64 extra_flags
= chunk_to_extended(flags
) &
10288 BTRFS_EXTENDED_PROFILE_MASK
;
10290 write_seqlock(&fs_info
->profiles_lock
);
10291 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
10292 fs_info
->avail_data_alloc_bits
&= ~extra_flags
;
10293 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
10294 fs_info
->avail_metadata_alloc_bits
&= ~extra_flags
;
10295 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
10296 fs_info
->avail_system_alloc_bits
&= ~extra_flags
;
10297 write_sequnlock(&fs_info
->profiles_lock
);
10300 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
10301 struct btrfs_fs_info
*fs_info
, u64 group_start
,
10302 struct extent_map
*em
)
10304 struct btrfs_root
*root
= fs_info
->extent_root
;
10305 struct btrfs_path
*path
;
10306 struct btrfs_block_group_cache
*block_group
;
10307 struct btrfs_free_cluster
*cluster
;
10308 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
10309 struct btrfs_key key
;
10310 struct inode
*inode
;
10311 struct kobject
*kobj
= NULL
;
10315 struct btrfs_caching_control
*caching_ctl
= NULL
;
10318 block_group
= btrfs_lookup_block_group(fs_info
, group_start
);
10319 BUG_ON(!block_group
);
10320 BUG_ON(!block_group
->ro
);
10323 * Free the reserved super bytes from this block group before
10326 free_excluded_extents(fs_info
, block_group
);
10327 btrfs_free_ref_tree_range(fs_info
, block_group
->key
.objectid
,
10328 block_group
->key
.offset
);
10330 memcpy(&key
, &block_group
->key
, sizeof(key
));
10331 index
= get_block_group_index(block_group
);
10332 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
10333 BTRFS_BLOCK_GROUP_RAID1
|
10334 BTRFS_BLOCK_GROUP_RAID10
))
10339 /* make sure this block group isn't part of an allocation cluster */
10340 cluster
= &fs_info
->data_alloc_cluster
;
10341 spin_lock(&cluster
->refill_lock
);
10342 btrfs_return_cluster_to_free_space(block_group
, cluster
);
10343 spin_unlock(&cluster
->refill_lock
);
10346 * make sure this block group isn't part of a metadata
10347 * allocation cluster
10349 cluster
= &fs_info
->meta_alloc_cluster
;
10350 spin_lock(&cluster
->refill_lock
);
10351 btrfs_return_cluster_to_free_space(block_group
, cluster
);
10352 spin_unlock(&cluster
->refill_lock
);
10354 path
= btrfs_alloc_path();
10361 * get the inode first so any iput calls done for the io_list
10362 * aren't the final iput (no unlinks allowed now)
10364 inode
= lookup_free_space_inode(fs_info
, block_group
, path
);
10366 mutex_lock(&trans
->transaction
->cache_write_mutex
);
10368 * make sure our free spache cache IO is done before remove the
10371 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
10372 if (!list_empty(&block_group
->io_list
)) {
10373 list_del_init(&block_group
->io_list
);
10375 WARN_ON(!IS_ERR(inode
) && inode
!= block_group
->io_ctl
.inode
);
10377 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
10378 btrfs_wait_cache_io(trans
, block_group
, path
);
10379 btrfs_put_block_group(block_group
);
10380 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
10383 if (!list_empty(&block_group
->dirty_list
)) {
10384 list_del_init(&block_group
->dirty_list
);
10385 btrfs_put_block_group(block_group
);
10387 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
10388 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
10390 if (!IS_ERR(inode
)) {
10391 ret
= btrfs_orphan_add(trans
, BTRFS_I(inode
));
10393 btrfs_add_delayed_iput(inode
);
10396 clear_nlink(inode
);
10397 /* One for the block groups ref */
10398 spin_lock(&block_group
->lock
);
10399 if (block_group
->iref
) {
10400 block_group
->iref
= 0;
10401 block_group
->inode
= NULL
;
10402 spin_unlock(&block_group
->lock
);
10405 spin_unlock(&block_group
->lock
);
10407 /* One for our lookup ref */
10408 btrfs_add_delayed_iput(inode
);
10411 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
10412 key
.offset
= block_group
->key
.objectid
;
10415 ret
= btrfs_search_slot(trans
, tree_root
, &key
, path
, -1, 1);
10419 btrfs_release_path(path
);
10421 ret
= btrfs_del_item(trans
, tree_root
, path
);
10424 btrfs_release_path(path
);
10427 spin_lock(&fs_info
->block_group_cache_lock
);
10428 rb_erase(&block_group
->cache_node
,
10429 &fs_info
->block_group_cache_tree
);
10430 RB_CLEAR_NODE(&block_group
->cache_node
);
10432 if (fs_info
->first_logical_byte
== block_group
->key
.objectid
)
10433 fs_info
->first_logical_byte
= (u64
)-1;
10434 spin_unlock(&fs_info
->block_group_cache_lock
);
10436 down_write(&block_group
->space_info
->groups_sem
);
10438 * we must use list_del_init so people can check to see if they
10439 * are still on the list after taking the semaphore
10441 list_del_init(&block_group
->list
);
10442 if (list_empty(&block_group
->space_info
->block_groups
[index
])) {
10443 kobj
= block_group
->space_info
->block_group_kobjs
[index
];
10444 block_group
->space_info
->block_group_kobjs
[index
] = NULL
;
10445 clear_avail_alloc_bits(fs_info
, block_group
->flags
);
10447 up_write(&block_group
->space_info
->groups_sem
);
10453 if (block_group
->has_caching_ctl
)
10454 caching_ctl
= get_caching_control(block_group
);
10455 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
10456 wait_block_group_cache_done(block_group
);
10457 if (block_group
->has_caching_ctl
) {
10458 down_write(&fs_info
->commit_root_sem
);
10459 if (!caching_ctl
) {
10460 struct btrfs_caching_control
*ctl
;
10462 list_for_each_entry(ctl
,
10463 &fs_info
->caching_block_groups
, list
)
10464 if (ctl
->block_group
== block_group
) {
10466 refcount_inc(&caching_ctl
->count
);
10471 list_del_init(&caching_ctl
->list
);
10472 up_write(&fs_info
->commit_root_sem
);
10474 /* Once for the caching bgs list and once for us. */
10475 put_caching_control(caching_ctl
);
10476 put_caching_control(caching_ctl
);
10480 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
10481 if (!list_empty(&block_group
->dirty_list
)) {
10484 if (!list_empty(&block_group
->io_list
)) {
10487 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
10488 btrfs_remove_free_space_cache(block_group
);
10490 spin_lock(&block_group
->space_info
->lock
);
10491 list_del_init(&block_group
->ro_list
);
10493 if (btrfs_test_opt(fs_info
, ENOSPC_DEBUG
)) {
10494 WARN_ON(block_group
->space_info
->total_bytes
10495 < block_group
->key
.offset
);
10496 WARN_ON(block_group
->space_info
->bytes_readonly
10497 < block_group
->key
.offset
);
10498 WARN_ON(block_group
->space_info
->disk_total
10499 < block_group
->key
.offset
* factor
);
10501 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
10502 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
10503 block_group
->space_info
->disk_total
-= block_group
->key
.offset
* factor
;
10505 spin_unlock(&block_group
->space_info
->lock
);
10507 memcpy(&key
, &block_group
->key
, sizeof(key
));
10509 mutex_lock(&fs_info
->chunk_mutex
);
10510 if (!list_empty(&em
->list
)) {
10511 /* We're in the transaction->pending_chunks list. */
10512 free_extent_map(em
);
10514 spin_lock(&block_group
->lock
);
10515 block_group
->removed
= 1;
10517 * At this point trimming can't start on this block group, because we
10518 * removed the block group from the tree fs_info->block_group_cache_tree
10519 * so no one can't find it anymore and even if someone already got this
10520 * block group before we removed it from the rbtree, they have already
10521 * incremented block_group->trimming - if they didn't, they won't find
10522 * any free space entries because we already removed them all when we
10523 * called btrfs_remove_free_space_cache().
10525 * And we must not remove the extent map from the fs_info->mapping_tree
10526 * to prevent the same logical address range and physical device space
10527 * ranges from being reused for a new block group. This is because our
10528 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10529 * completely transactionless, so while it is trimming a range the
10530 * currently running transaction might finish and a new one start,
10531 * allowing for new block groups to be created that can reuse the same
10532 * physical device locations unless we take this special care.
10534 * There may also be an implicit trim operation if the file system
10535 * is mounted with -odiscard. The same protections must remain
10536 * in place until the extents have been discarded completely when
10537 * the transaction commit has completed.
10539 remove_em
= (atomic_read(&block_group
->trimming
) == 0);
10541 * Make sure a trimmer task always sees the em in the pinned_chunks list
10542 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10543 * before checking block_group->removed).
10547 * Our em might be in trans->transaction->pending_chunks which
10548 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10549 * and so is the fs_info->pinned_chunks list.
10551 * So at this point we must be holding the chunk_mutex to avoid
10552 * any races with chunk allocation (more specifically at
10553 * volumes.c:contains_pending_extent()), to ensure it always
10554 * sees the em, either in the pending_chunks list or in the
10555 * pinned_chunks list.
10557 list_move_tail(&em
->list
, &fs_info
->pinned_chunks
);
10559 spin_unlock(&block_group
->lock
);
10562 struct extent_map_tree
*em_tree
;
10564 em_tree
= &fs_info
->mapping_tree
.map_tree
;
10565 write_lock(&em_tree
->lock
);
10567 * The em might be in the pending_chunks list, so make sure the
10568 * chunk mutex is locked, since remove_extent_mapping() will
10569 * delete us from that list.
10571 remove_extent_mapping(em_tree
, em
);
10572 write_unlock(&em_tree
->lock
);
10573 /* once for the tree */
10574 free_extent_map(em
);
10577 mutex_unlock(&fs_info
->chunk_mutex
);
10579 ret
= remove_block_group_free_space(trans
, fs_info
, block_group
);
10583 btrfs_put_block_group(block_group
);
10584 btrfs_put_block_group(block_group
);
10586 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
10592 ret
= btrfs_del_item(trans
, root
, path
);
10594 btrfs_free_path(path
);
10598 struct btrfs_trans_handle
*
10599 btrfs_start_trans_remove_block_group(struct btrfs_fs_info
*fs_info
,
10600 const u64 chunk_offset
)
10602 struct extent_map_tree
*em_tree
= &fs_info
->mapping_tree
.map_tree
;
10603 struct extent_map
*em
;
10604 struct map_lookup
*map
;
10605 unsigned int num_items
;
10607 read_lock(&em_tree
->lock
);
10608 em
= lookup_extent_mapping(em_tree
, chunk_offset
, 1);
10609 read_unlock(&em_tree
->lock
);
10610 ASSERT(em
&& em
->start
== chunk_offset
);
10613 * We need to reserve 3 + N units from the metadata space info in order
10614 * to remove a block group (done at btrfs_remove_chunk() and at
10615 * btrfs_remove_block_group()), which are used for:
10617 * 1 unit for adding the free space inode's orphan (located in the tree
10619 * 1 unit for deleting the block group item (located in the extent
10621 * 1 unit for deleting the free space item (located in tree of tree
10623 * N units for deleting N device extent items corresponding to each
10624 * stripe (located in the device tree).
10626 * In order to remove a block group we also need to reserve units in the
10627 * system space info in order to update the chunk tree (update one or
10628 * more device items and remove one chunk item), but this is done at
10629 * btrfs_remove_chunk() through a call to check_system_chunk().
10631 map
= em
->map_lookup
;
10632 num_items
= 3 + map
->num_stripes
;
10633 free_extent_map(em
);
10635 return btrfs_start_transaction_fallback_global_rsv(fs_info
->extent_root
,
10640 * Process the unused_bgs list and remove any that don't have any allocated
10641 * space inside of them.
10643 void btrfs_delete_unused_bgs(struct btrfs_fs_info
*fs_info
)
10645 struct btrfs_block_group_cache
*block_group
;
10646 struct btrfs_space_info
*space_info
;
10647 struct btrfs_trans_handle
*trans
;
10650 if (!test_bit(BTRFS_FS_OPEN
, &fs_info
->flags
))
10653 spin_lock(&fs_info
->unused_bgs_lock
);
10654 while (!list_empty(&fs_info
->unused_bgs
)) {
10658 block_group
= list_first_entry(&fs_info
->unused_bgs
,
10659 struct btrfs_block_group_cache
,
10661 list_del_init(&block_group
->bg_list
);
10663 space_info
= block_group
->space_info
;
10665 if (ret
|| btrfs_mixed_space_info(space_info
)) {
10666 btrfs_put_block_group(block_group
);
10669 spin_unlock(&fs_info
->unused_bgs_lock
);
10671 mutex_lock(&fs_info
->delete_unused_bgs_mutex
);
10673 /* Don't want to race with allocators so take the groups_sem */
10674 down_write(&space_info
->groups_sem
);
10675 spin_lock(&block_group
->lock
);
10676 if (block_group
->reserved
||
10677 btrfs_block_group_used(&block_group
->item
) ||
10679 list_is_singular(&block_group
->list
)) {
10681 * We want to bail if we made new allocations or have
10682 * outstanding allocations in this block group. We do
10683 * the ro check in case balance is currently acting on
10684 * this block group.
10686 spin_unlock(&block_group
->lock
);
10687 up_write(&space_info
->groups_sem
);
10690 spin_unlock(&block_group
->lock
);
10692 /* We don't want to force the issue, only flip if it's ok. */
10693 ret
= inc_block_group_ro(block_group
, 0);
10694 up_write(&space_info
->groups_sem
);
10701 * Want to do this before we do anything else so we can recover
10702 * properly if we fail to join the transaction.
10704 trans
= btrfs_start_trans_remove_block_group(fs_info
,
10705 block_group
->key
.objectid
);
10706 if (IS_ERR(trans
)) {
10707 btrfs_dec_block_group_ro(block_group
);
10708 ret
= PTR_ERR(trans
);
10713 * We could have pending pinned extents for this block group,
10714 * just delete them, we don't care about them anymore.
10716 start
= block_group
->key
.objectid
;
10717 end
= start
+ block_group
->key
.offset
- 1;
10719 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10720 * btrfs_finish_extent_commit(). If we are at transaction N,
10721 * another task might be running finish_extent_commit() for the
10722 * previous transaction N - 1, and have seen a range belonging
10723 * to the block group in freed_extents[] before we were able to
10724 * clear the whole block group range from freed_extents[]. This
10725 * means that task can lookup for the block group after we
10726 * unpinned it from freed_extents[] and removed it, leading to
10727 * a BUG_ON() at btrfs_unpin_extent_range().
10729 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
10730 ret
= clear_extent_bits(&fs_info
->freed_extents
[0], start
, end
,
10733 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10734 btrfs_dec_block_group_ro(block_group
);
10737 ret
= clear_extent_bits(&fs_info
->freed_extents
[1], start
, end
,
10740 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10741 btrfs_dec_block_group_ro(block_group
);
10744 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10746 /* Reset pinned so btrfs_put_block_group doesn't complain */
10747 spin_lock(&space_info
->lock
);
10748 spin_lock(&block_group
->lock
);
10750 space_info
->bytes_pinned
-= block_group
->pinned
;
10751 space_info
->bytes_readonly
+= block_group
->pinned
;
10752 percpu_counter_add(&space_info
->total_bytes_pinned
,
10753 -block_group
->pinned
);
10754 block_group
->pinned
= 0;
10756 spin_unlock(&block_group
->lock
);
10757 spin_unlock(&space_info
->lock
);
10759 /* DISCARD can flip during remount */
10760 trimming
= btrfs_test_opt(fs_info
, DISCARD
);
10762 /* Implicit trim during transaction commit. */
10764 btrfs_get_block_group_trimming(block_group
);
10767 * Btrfs_remove_chunk will abort the transaction if things go
10770 ret
= btrfs_remove_chunk(trans
, fs_info
,
10771 block_group
->key
.objectid
);
10775 btrfs_put_block_group_trimming(block_group
);
10780 * If we're not mounted with -odiscard, we can just forget
10781 * about this block group. Otherwise we'll need to wait
10782 * until transaction commit to do the actual discard.
10785 spin_lock(&fs_info
->unused_bgs_lock
);
10787 * A concurrent scrub might have added us to the list
10788 * fs_info->unused_bgs, so use a list_move operation
10789 * to add the block group to the deleted_bgs list.
10791 list_move(&block_group
->bg_list
,
10792 &trans
->transaction
->deleted_bgs
);
10793 spin_unlock(&fs_info
->unused_bgs_lock
);
10794 btrfs_get_block_group(block_group
);
10797 btrfs_end_transaction(trans
);
10799 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
10800 btrfs_put_block_group(block_group
);
10801 spin_lock(&fs_info
->unused_bgs_lock
);
10803 spin_unlock(&fs_info
->unused_bgs_lock
);
10806 int btrfs_init_space_info(struct btrfs_fs_info
*fs_info
)
10808 struct btrfs_space_info
*space_info
;
10809 struct btrfs_super_block
*disk_super
;
10815 disk_super
= fs_info
->super_copy
;
10816 if (!btrfs_super_root(disk_super
))
10819 features
= btrfs_super_incompat_flags(disk_super
);
10820 if (features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
10823 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
10824 ret
= create_space_info(fs_info
, flags
, &space_info
);
10829 flags
= BTRFS_BLOCK_GROUP_METADATA
| BTRFS_BLOCK_GROUP_DATA
;
10830 ret
= create_space_info(fs_info
, flags
, &space_info
);
10832 flags
= BTRFS_BLOCK_GROUP_METADATA
;
10833 ret
= create_space_info(fs_info
, flags
, &space_info
);
10837 flags
= BTRFS_BLOCK_GROUP_DATA
;
10838 ret
= create_space_info(fs_info
, flags
, &space_info
);
10844 int btrfs_error_unpin_extent_range(struct btrfs_fs_info
*fs_info
,
10845 u64 start
, u64 end
)
10847 return unpin_extent_range(fs_info
, start
, end
, false);
10851 * It used to be that old block groups would be left around forever.
10852 * Iterating over them would be enough to trim unused space. Since we
10853 * now automatically remove them, we also need to iterate over unallocated
10856 * We don't want a transaction for this since the discard may take a
10857 * substantial amount of time. We don't require that a transaction be
10858 * running, but we do need to take a running transaction into account
10859 * to ensure that we're not discarding chunks that were released in
10860 * the current transaction.
10862 * Holding the chunks lock will prevent other threads from allocating
10863 * or releasing chunks, but it won't prevent a running transaction
10864 * from committing and releasing the memory that the pending chunks
10865 * list head uses. For that, we need to take a reference to the
10868 static int btrfs_trim_free_extents(struct btrfs_device
*device
,
10869 u64 minlen
, u64
*trimmed
)
10871 u64 start
= 0, len
= 0;
10876 /* Not writeable = nothing to do. */
10877 if (!device
->writeable
)
10880 /* No free space = nothing to do. */
10881 if (device
->total_bytes
<= device
->bytes_used
)
10887 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
10888 struct btrfs_transaction
*trans
;
10891 ret
= mutex_lock_interruptible(&fs_info
->chunk_mutex
);
10895 down_read(&fs_info
->commit_root_sem
);
10897 spin_lock(&fs_info
->trans_lock
);
10898 trans
= fs_info
->running_transaction
;
10900 refcount_inc(&trans
->use_count
);
10901 spin_unlock(&fs_info
->trans_lock
);
10903 ret
= find_free_dev_extent_start(trans
, device
, minlen
, start
,
10906 btrfs_put_transaction(trans
);
10909 up_read(&fs_info
->commit_root_sem
);
10910 mutex_unlock(&fs_info
->chunk_mutex
);
10911 if (ret
== -ENOSPC
)
10916 ret
= btrfs_issue_discard(device
->bdev
, start
, len
, &bytes
);
10917 up_read(&fs_info
->commit_root_sem
);
10918 mutex_unlock(&fs_info
->chunk_mutex
);
10926 if (fatal_signal_pending(current
)) {
10927 ret
= -ERESTARTSYS
;
10937 int btrfs_trim_fs(struct btrfs_fs_info
*fs_info
, struct fstrim_range
*range
)
10939 struct btrfs_block_group_cache
*cache
= NULL
;
10940 struct btrfs_device
*device
;
10941 struct list_head
*devices
;
10946 u64 total_bytes
= btrfs_super_total_bytes(fs_info
->super_copy
);
10950 * try to trim all FS space, our block group may start from non-zero.
10952 if (range
->len
== total_bytes
)
10953 cache
= btrfs_lookup_first_block_group(fs_info
, range
->start
);
10955 cache
= btrfs_lookup_block_group(fs_info
, range
->start
);
10958 if (cache
->key
.objectid
>= (range
->start
+ range
->len
)) {
10959 btrfs_put_block_group(cache
);
10963 start
= max(range
->start
, cache
->key
.objectid
);
10964 end
= min(range
->start
+ range
->len
,
10965 cache
->key
.objectid
+ cache
->key
.offset
);
10967 if (end
- start
>= range
->minlen
) {
10968 if (!block_group_cache_done(cache
)) {
10969 ret
= cache_block_group(cache
, 0);
10971 btrfs_put_block_group(cache
);
10974 ret
= wait_block_group_cache_done(cache
);
10976 btrfs_put_block_group(cache
);
10980 ret
= btrfs_trim_block_group(cache
,
10986 trimmed
+= group_trimmed
;
10988 btrfs_put_block_group(cache
);
10993 cache
= next_block_group(fs_info
, cache
);
10996 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
10997 devices
= &fs_info
->fs_devices
->alloc_list
;
10998 list_for_each_entry(device
, devices
, dev_alloc_list
) {
10999 ret
= btrfs_trim_free_extents(device
, range
->minlen
,
11004 trimmed
+= group_trimmed
;
11006 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
11008 range
->len
= trimmed
;
11013 * btrfs_{start,end}_write_no_snapshotting() are similar to
11014 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
11015 * data into the page cache through nocow before the subvolume is snapshoted,
11016 * but flush the data into disk after the snapshot creation, or to prevent
11017 * operations while snapshotting is ongoing and that cause the snapshot to be
11018 * inconsistent (writes followed by expanding truncates for example).
11020 void btrfs_end_write_no_snapshotting(struct btrfs_root
*root
)
11022 percpu_counter_dec(&root
->subv_writers
->counter
);
11024 * Make sure counter is updated before we wake up waiters.
11027 if (waitqueue_active(&root
->subv_writers
->wait
))
11028 wake_up(&root
->subv_writers
->wait
);
11031 int btrfs_start_write_no_snapshotting(struct btrfs_root
*root
)
11033 if (atomic_read(&root
->will_be_snapshotted
))
11036 percpu_counter_inc(&root
->subv_writers
->counter
);
11038 * Make sure counter is updated before we check for snapshot creation.
11041 if (atomic_read(&root
->will_be_snapshotted
)) {
11042 btrfs_end_write_no_snapshotting(root
);
11048 void btrfs_wait_for_snapshot_creation(struct btrfs_root
*root
)
11053 ret
= btrfs_start_write_no_snapshotting(root
);
11056 wait_on_atomic_t(&root
->will_be_snapshotted
, atomic_t_wait
,
11057 TASK_UNINTERRUPTIBLE
);