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
,
2514 if (ret
&& insert_reserved
)
2515 btrfs_pin_extent(trans
->fs_info
, node
->bytenr
,
2516 node
->num_bytes
, 1);
2520 static inline struct btrfs_delayed_ref_node
*
2521 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
2523 struct btrfs_delayed_ref_node
*ref
;
2525 if (RB_EMPTY_ROOT(&head
->ref_tree
))
2529 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2530 * This is to prevent a ref count from going down to zero, which deletes
2531 * the extent item from the extent tree, when there still are references
2532 * to add, which would fail because they would not find the extent item.
2534 if (!list_empty(&head
->ref_add_list
))
2535 return list_first_entry(&head
->ref_add_list
,
2536 struct btrfs_delayed_ref_node
, add_list
);
2538 ref
= rb_entry(rb_first(&head
->ref_tree
),
2539 struct btrfs_delayed_ref_node
, ref_node
);
2540 ASSERT(list_empty(&ref
->add_list
));
2544 static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root
*delayed_refs
,
2545 struct btrfs_delayed_ref_head
*head
)
2547 spin_lock(&delayed_refs
->lock
);
2548 head
->processing
= 0;
2549 delayed_refs
->num_heads_ready
++;
2550 spin_unlock(&delayed_refs
->lock
);
2551 btrfs_delayed_ref_unlock(head
);
2554 static int cleanup_extent_op(struct btrfs_trans_handle
*trans
,
2555 struct btrfs_fs_info
*fs_info
,
2556 struct btrfs_delayed_ref_head
*head
)
2558 struct btrfs_delayed_extent_op
*extent_op
= head
->extent_op
;
2563 head
->extent_op
= NULL
;
2564 if (head
->must_insert_reserved
) {
2565 btrfs_free_delayed_extent_op(extent_op
);
2568 spin_unlock(&head
->lock
);
2569 ret
= run_delayed_extent_op(trans
, fs_info
, head
, extent_op
);
2570 btrfs_free_delayed_extent_op(extent_op
);
2571 return ret
? ret
: 1;
2574 static int cleanup_ref_head(struct btrfs_trans_handle
*trans
,
2575 struct btrfs_fs_info
*fs_info
,
2576 struct btrfs_delayed_ref_head
*head
)
2578 struct btrfs_delayed_ref_root
*delayed_refs
;
2581 delayed_refs
= &trans
->transaction
->delayed_refs
;
2583 ret
= cleanup_extent_op(trans
, fs_info
, head
);
2585 unselect_delayed_ref_head(delayed_refs
, head
);
2586 btrfs_debug(fs_info
, "run_delayed_extent_op returned %d", ret
);
2593 * Need to drop our head ref lock and re-acquire the delayed ref lock
2594 * and then re-check to make sure nobody got added.
2596 spin_unlock(&head
->lock
);
2597 spin_lock(&delayed_refs
->lock
);
2598 spin_lock(&head
->lock
);
2599 if (!RB_EMPTY_ROOT(&head
->ref_tree
) || head
->extent_op
) {
2600 spin_unlock(&head
->lock
);
2601 spin_unlock(&delayed_refs
->lock
);
2604 delayed_refs
->num_heads
--;
2605 rb_erase(&head
->href_node
, &delayed_refs
->href_root
);
2606 RB_CLEAR_NODE(&head
->href_node
);
2607 spin_unlock(&delayed_refs
->lock
);
2608 spin_unlock(&head
->lock
);
2609 atomic_dec(&delayed_refs
->num_entries
);
2611 trace_run_delayed_ref_head(fs_info
, head
, 0);
2613 if (head
->total_ref_mod
< 0) {
2614 struct btrfs_space_info
*space_info
;
2618 flags
= BTRFS_BLOCK_GROUP_DATA
;
2619 else if (head
->is_system
)
2620 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
2622 flags
= BTRFS_BLOCK_GROUP_METADATA
;
2623 space_info
= __find_space_info(fs_info
, flags
);
2625 percpu_counter_add(&space_info
->total_bytes_pinned
,
2628 if (head
->is_data
) {
2629 spin_lock(&delayed_refs
->lock
);
2630 delayed_refs
->pending_csums
-= head
->num_bytes
;
2631 spin_unlock(&delayed_refs
->lock
);
2635 if (head
->must_insert_reserved
) {
2636 btrfs_pin_extent(fs_info
, head
->bytenr
,
2637 head
->num_bytes
, 1);
2638 if (head
->is_data
) {
2639 ret
= btrfs_del_csums(trans
, fs_info
, head
->bytenr
,
2644 /* Also free its reserved qgroup space */
2645 btrfs_qgroup_free_delayed_ref(fs_info
, head
->qgroup_ref_root
,
2646 head
->qgroup_reserved
);
2647 btrfs_delayed_ref_unlock(head
);
2648 btrfs_put_delayed_ref_head(head
);
2653 * Returns 0 on success or if called with an already aborted transaction.
2654 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2656 static noinline
int __btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2657 struct btrfs_fs_info
*fs_info
,
2660 struct btrfs_delayed_ref_root
*delayed_refs
;
2661 struct btrfs_delayed_ref_node
*ref
;
2662 struct btrfs_delayed_ref_head
*locked_ref
= NULL
;
2663 struct btrfs_delayed_extent_op
*extent_op
;
2664 ktime_t start
= ktime_get();
2666 unsigned long count
= 0;
2667 unsigned long actual_count
= 0;
2668 int must_insert_reserved
= 0;
2670 delayed_refs
= &trans
->transaction
->delayed_refs
;
2676 spin_lock(&delayed_refs
->lock
);
2677 locked_ref
= btrfs_select_ref_head(trans
);
2679 spin_unlock(&delayed_refs
->lock
);
2683 /* grab the lock that says we are going to process
2684 * all the refs for this head */
2685 ret
= btrfs_delayed_ref_lock(trans
, locked_ref
);
2686 spin_unlock(&delayed_refs
->lock
);
2688 * we may have dropped the spin lock to get the head
2689 * mutex lock, and that might have given someone else
2690 * time to free the head. If that's true, it has been
2691 * removed from our list and we can move on.
2693 if (ret
== -EAGAIN
) {
2701 * We need to try and merge add/drops of the same ref since we
2702 * can run into issues with relocate dropping the implicit ref
2703 * and then it being added back again before the drop can
2704 * finish. If we merged anything we need to re-loop so we can
2706 * Or we can get node references of the same type that weren't
2707 * merged when created due to bumps in the tree mod seq, and
2708 * we need to merge them to prevent adding an inline extent
2709 * backref before dropping it (triggering a BUG_ON at
2710 * insert_inline_extent_backref()).
2712 spin_lock(&locked_ref
->lock
);
2713 btrfs_merge_delayed_refs(trans
, fs_info
, delayed_refs
,
2717 * locked_ref is the head node, so we have to go one
2718 * node back for any delayed ref updates
2720 ref
= select_delayed_ref(locked_ref
);
2722 if (ref
&& ref
->seq
&&
2723 btrfs_check_delayed_seq(fs_info
, delayed_refs
, ref
->seq
)) {
2724 spin_unlock(&locked_ref
->lock
);
2725 unselect_delayed_ref_head(delayed_refs
, locked_ref
);
2733 * We're done processing refs in this ref_head, clean everything
2734 * up and move on to the next ref_head.
2737 ret
= cleanup_ref_head(trans
, fs_info
, locked_ref
);
2739 /* We dropped our lock, we need to loop. */
2752 rb_erase(&ref
->ref_node
, &locked_ref
->ref_tree
);
2753 RB_CLEAR_NODE(&ref
->ref_node
);
2754 if (!list_empty(&ref
->add_list
))
2755 list_del(&ref
->add_list
);
2757 * When we play the delayed ref, also correct the ref_mod on
2760 switch (ref
->action
) {
2761 case BTRFS_ADD_DELAYED_REF
:
2762 case BTRFS_ADD_DELAYED_EXTENT
:
2763 locked_ref
->ref_mod
-= ref
->ref_mod
;
2765 case BTRFS_DROP_DELAYED_REF
:
2766 locked_ref
->ref_mod
+= ref
->ref_mod
;
2771 atomic_dec(&delayed_refs
->num_entries
);
2774 * Record the must-insert_reserved flag before we drop the spin
2777 must_insert_reserved
= locked_ref
->must_insert_reserved
;
2778 locked_ref
->must_insert_reserved
= 0;
2780 extent_op
= locked_ref
->extent_op
;
2781 locked_ref
->extent_op
= NULL
;
2782 spin_unlock(&locked_ref
->lock
);
2784 ret
= run_one_delayed_ref(trans
, fs_info
, ref
, extent_op
,
2785 must_insert_reserved
);
2787 btrfs_free_delayed_extent_op(extent_op
);
2789 unselect_delayed_ref_head(delayed_refs
, locked_ref
);
2790 btrfs_put_delayed_ref(ref
);
2791 btrfs_debug(fs_info
, "run_one_delayed_ref returned %d",
2796 btrfs_put_delayed_ref(ref
);
2802 * We don't want to include ref heads since we can have empty ref heads
2803 * and those will drastically skew our runtime down since we just do
2804 * accounting, no actual extent tree updates.
2806 if (actual_count
> 0) {
2807 u64 runtime
= ktime_to_ns(ktime_sub(ktime_get(), start
));
2811 * We weigh the current average higher than our current runtime
2812 * to avoid large swings in the average.
2814 spin_lock(&delayed_refs
->lock
);
2815 avg
= fs_info
->avg_delayed_ref_runtime
* 3 + runtime
;
2816 fs_info
->avg_delayed_ref_runtime
= avg
>> 2; /* div by 4 */
2817 spin_unlock(&delayed_refs
->lock
);
2822 #ifdef SCRAMBLE_DELAYED_REFS
2824 * Normally delayed refs get processed in ascending bytenr order. This
2825 * correlates in most cases to the order added. To expose dependencies on this
2826 * order, we start to process the tree in the middle instead of the beginning
2828 static u64
find_middle(struct rb_root
*root
)
2830 struct rb_node
*n
= root
->rb_node
;
2831 struct btrfs_delayed_ref_node
*entry
;
2834 u64 first
= 0, last
= 0;
2838 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2839 first
= entry
->bytenr
;
2843 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2844 last
= entry
->bytenr
;
2849 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2850 WARN_ON(!entry
->in_tree
);
2852 middle
= entry
->bytenr
;
2865 static inline u64
heads_to_leaves(struct btrfs_fs_info
*fs_info
, u64 heads
)
2869 num_bytes
= heads
* (sizeof(struct btrfs_extent_item
) +
2870 sizeof(struct btrfs_extent_inline_ref
));
2871 if (!btrfs_fs_incompat(fs_info
, SKINNY_METADATA
))
2872 num_bytes
+= heads
* sizeof(struct btrfs_tree_block_info
);
2875 * We don't ever fill up leaves all the way so multiply by 2 just to be
2876 * closer to what we're really going to want to use.
2878 return div_u64(num_bytes
, BTRFS_LEAF_DATA_SIZE(fs_info
));
2882 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2883 * would require to store the csums for that many bytes.
2885 u64
btrfs_csum_bytes_to_leaves(struct btrfs_fs_info
*fs_info
, u64 csum_bytes
)
2888 u64 num_csums_per_leaf
;
2891 csum_size
= BTRFS_MAX_ITEM_SIZE(fs_info
);
2892 num_csums_per_leaf
= div64_u64(csum_size
,
2893 (u64
)btrfs_super_csum_size(fs_info
->super_copy
));
2894 num_csums
= div64_u64(csum_bytes
, fs_info
->sectorsize
);
2895 num_csums
+= num_csums_per_leaf
- 1;
2896 num_csums
= div64_u64(num_csums
, num_csums_per_leaf
);
2900 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle
*trans
,
2901 struct btrfs_fs_info
*fs_info
)
2903 struct btrfs_block_rsv
*global_rsv
;
2904 u64 num_heads
= trans
->transaction
->delayed_refs
.num_heads_ready
;
2905 u64 csum_bytes
= trans
->transaction
->delayed_refs
.pending_csums
;
2906 u64 num_dirty_bgs
= trans
->transaction
->num_dirty_bgs
;
2907 u64 num_bytes
, num_dirty_bgs_bytes
;
2910 num_bytes
= btrfs_calc_trans_metadata_size(fs_info
, 1);
2911 num_heads
= heads_to_leaves(fs_info
, num_heads
);
2913 num_bytes
+= (num_heads
- 1) * fs_info
->nodesize
;
2915 num_bytes
+= btrfs_csum_bytes_to_leaves(fs_info
, csum_bytes
) *
2917 num_dirty_bgs_bytes
= btrfs_calc_trans_metadata_size(fs_info
,
2919 global_rsv
= &fs_info
->global_block_rsv
;
2922 * If we can't allocate any more chunks lets make sure we have _lots_ of
2923 * wiggle room since running delayed refs can create more delayed refs.
2925 if (global_rsv
->space_info
->full
) {
2926 num_dirty_bgs_bytes
<<= 1;
2930 spin_lock(&global_rsv
->lock
);
2931 if (global_rsv
->reserved
<= num_bytes
+ num_dirty_bgs_bytes
)
2933 spin_unlock(&global_rsv
->lock
);
2937 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle
*trans
,
2938 struct btrfs_fs_info
*fs_info
)
2941 atomic_read(&trans
->transaction
->delayed_refs
.num_entries
);
2946 avg_runtime
= fs_info
->avg_delayed_ref_runtime
;
2947 val
= num_entries
* avg_runtime
;
2948 if (val
>= NSEC_PER_SEC
)
2950 if (val
>= NSEC_PER_SEC
/ 2)
2953 return btrfs_check_space_for_delayed_refs(trans
, fs_info
);
2956 struct async_delayed_refs
{
2957 struct btrfs_root
*root
;
2962 struct completion wait
;
2963 struct btrfs_work work
;
2966 static inline struct async_delayed_refs
*
2967 to_async_delayed_refs(struct btrfs_work
*work
)
2969 return container_of(work
, struct async_delayed_refs
, work
);
2972 static void delayed_ref_async_start(struct btrfs_work
*work
)
2974 struct async_delayed_refs
*async
= to_async_delayed_refs(work
);
2975 struct btrfs_trans_handle
*trans
;
2976 struct btrfs_fs_info
*fs_info
= async
->root
->fs_info
;
2979 /* if the commit is already started, we don't need to wait here */
2980 if (btrfs_transaction_blocked(fs_info
))
2983 trans
= btrfs_join_transaction(async
->root
);
2984 if (IS_ERR(trans
)) {
2985 async
->error
= PTR_ERR(trans
);
2990 * trans->sync means that when we call end_transaction, we won't
2991 * wait on delayed refs
2995 /* Don't bother flushing if we got into a different transaction */
2996 if (trans
->transid
> async
->transid
)
2999 ret
= btrfs_run_delayed_refs(trans
, fs_info
, async
->count
);
3003 ret
= btrfs_end_transaction(trans
);
3004 if (ret
&& !async
->error
)
3008 complete(&async
->wait
);
3013 int btrfs_async_run_delayed_refs(struct btrfs_fs_info
*fs_info
,
3014 unsigned long count
, u64 transid
, int wait
)
3016 struct async_delayed_refs
*async
;
3019 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
3023 async
->root
= fs_info
->tree_root
;
3024 async
->count
= count
;
3026 async
->transid
= transid
;
3031 init_completion(&async
->wait
);
3033 btrfs_init_work(&async
->work
, btrfs_extent_refs_helper
,
3034 delayed_ref_async_start
, NULL
, NULL
);
3036 btrfs_queue_work(fs_info
->extent_workers
, &async
->work
);
3039 wait_for_completion(&async
->wait
);
3048 * this starts processing the delayed reference count updates and
3049 * extent insertions we have queued up so far. count can be
3050 * 0, which means to process everything in the tree at the start
3051 * of the run (but not newly added entries), or it can be some target
3052 * number you'd like to process.
3054 * Returns 0 on success or if called with an aborted transaction
3055 * Returns <0 on error and aborts the transaction
3057 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
3058 struct btrfs_fs_info
*fs_info
, unsigned long count
)
3060 struct rb_node
*node
;
3061 struct btrfs_delayed_ref_root
*delayed_refs
;
3062 struct btrfs_delayed_ref_head
*head
;
3064 int run_all
= count
== (unsigned long)-1;
3066 /* We'll clean this up in btrfs_cleanup_transaction */
3070 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE
, &fs_info
->flags
))
3073 delayed_refs
= &trans
->transaction
->delayed_refs
;
3075 count
= atomic_read(&delayed_refs
->num_entries
) * 2;
3078 #ifdef SCRAMBLE_DELAYED_REFS
3079 delayed_refs
->run_delayed_start
= find_middle(&delayed_refs
->root
);
3081 ret
= __btrfs_run_delayed_refs(trans
, fs_info
, count
);
3083 btrfs_abort_transaction(trans
, ret
);
3088 if (!list_empty(&trans
->new_bgs
))
3089 btrfs_create_pending_block_groups(trans
, fs_info
);
3091 spin_lock(&delayed_refs
->lock
);
3092 node
= rb_first(&delayed_refs
->href_root
);
3094 spin_unlock(&delayed_refs
->lock
);
3097 head
= rb_entry(node
, struct btrfs_delayed_ref_head
,
3099 refcount_inc(&head
->refs
);
3100 spin_unlock(&delayed_refs
->lock
);
3102 /* Mutex was contended, block until it's released and retry. */
3103 mutex_lock(&head
->mutex
);
3104 mutex_unlock(&head
->mutex
);
3106 btrfs_put_delayed_ref_head(head
);
3114 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
3115 struct btrfs_fs_info
*fs_info
,
3116 u64 bytenr
, u64 num_bytes
, u64 flags
,
3117 int level
, int is_data
)
3119 struct btrfs_delayed_extent_op
*extent_op
;
3122 extent_op
= btrfs_alloc_delayed_extent_op();
3126 extent_op
->flags_to_set
= flags
;
3127 extent_op
->update_flags
= true;
3128 extent_op
->update_key
= false;
3129 extent_op
->is_data
= is_data
? true : false;
3130 extent_op
->level
= level
;
3132 ret
= btrfs_add_delayed_extent_op(fs_info
, trans
, bytenr
,
3133 num_bytes
, extent_op
);
3135 btrfs_free_delayed_extent_op(extent_op
);
3139 static noinline
int check_delayed_ref(struct btrfs_root
*root
,
3140 struct btrfs_path
*path
,
3141 u64 objectid
, u64 offset
, u64 bytenr
)
3143 struct btrfs_delayed_ref_head
*head
;
3144 struct btrfs_delayed_ref_node
*ref
;
3145 struct btrfs_delayed_data_ref
*data_ref
;
3146 struct btrfs_delayed_ref_root
*delayed_refs
;
3147 struct btrfs_transaction
*cur_trans
;
3148 struct rb_node
*node
;
3151 spin_lock(&root
->fs_info
->trans_lock
);
3152 cur_trans
= root
->fs_info
->running_transaction
;
3154 refcount_inc(&cur_trans
->use_count
);
3155 spin_unlock(&root
->fs_info
->trans_lock
);
3159 delayed_refs
= &cur_trans
->delayed_refs
;
3160 spin_lock(&delayed_refs
->lock
);
3161 head
= btrfs_find_delayed_ref_head(delayed_refs
, bytenr
);
3163 spin_unlock(&delayed_refs
->lock
);
3164 btrfs_put_transaction(cur_trans
);
3168 if (!mutex_trylock(&head
->mutex
)) {
3169 refcount_inc(&head
->refs
);
3170 spin_unlock(&delayed_refs
->lock
);
3172 btrfs_release_path(path
);
3175 * Mutex was contended, block until it's released and let
3178 mutex_lock(&head
->mutex
);
3179 mutex_unlock(&head
->mutex
);
3180 btrfs_put_delayed_ref_head(head
);
3181 btrfs_put_transaction(cur_trans
);
3184 spin_unlock(&delayed_refs
->lock
);
3186 spin_lock(&head
->lock
);
3188 * XXX: We should replace this with a proper search function in the
3191 for (node
= rb_first(&head
->ref_tree
); node
; node
= rb_next(node
)) {
3192 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, ref_node
);
3193 /* If it's a shared ref we know a cross reference exists */
3194 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
) {
3199 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
3202 * If our ref doesn't match the one we're currently looking at
3203 * then we have a cross reference.
3205 if (data_ref
->root
!= root
->root_key
.objectid
||
3206 data_ref
->objectid
!= objectid
||
3207 data_ref
->offset
!= offset
) {
3212 spin_unlock(&head
->lock
);
3213 mutex_unlock(&head
->mutex
);
3214 btrfs_put_transaction(cur_trans
);
3218 static noinline
int check_committed_ref(struct btrfs_root
*root
,
3219 struct btrfs_path
*path
,
3220 u64 objectid
, u64 offset
, u64 bytenr
)
3222 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3223 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
3224 struct extent_buffer
*leaf
;
3225 struct btrfs_extent_data_ref
*ref
;
3226 struct btrfs_extent_inline_ref
*iref
;
3227 struct btrfs_extent_item
*ei
;
3228 struct btrfs_key key
;
3233 key
.objectid
= bytenr
;
3234 key
.offset
= (u64
)-1;
3235 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
3237 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
3240 BUG_ON(ret
== 0); /* Corruption */
3243 if (path
->slots
[0] == 0)
3247 leaf
= path
->nodes
[0];
3248 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
3250 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
3254 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
3255 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3256 if (item_size
< sizeof(*ei
)) {
3257 WARN_ON(item_size
!= sizeof(struct btrfs_extent_item_v0
));
3261 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
3263 if (item_size
!= sizeof(*ei
) +
3264 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
3267 if (btrfs_extent_generation(leaf
, ei
) <=
3268 btrfs_root_last_snapshot(&root
->root_item
))
3271 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
3273 type
= btrfs_get_extent_inline_ref_type(leaf
, iref
, BTRFS_REF_TYPE_DATA
);
3274 if (type
!= BTRFS_EXTENT_DATA_REF_KEY
)
3277 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
3278 if (btrfs_extent_refs(leaf
, ei
) !=
3279 btrfs_extent_data_ref_count(leaf
, ref
) ||
3280 btrfs_extent_data_ref_root(leaf
, ref
) !=
3281 root
->root_key
.objectid
||
3282 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
3283 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
3291 int btrfs_cross_ref_exist(struct btrfs_root
*root
, u64 objectid
, u64 offset
,
3294 struct btrfs_path
*path
;
3298 path
= btrfs_alloc_path();
3303 ret
= check_committed_ref(root
, path
, objectid
,
3305 if (ret
&& ret
!= -ENOENT
)
3308 ret2
= check_delayed_ref(root
, path
, objectid
,
3310 } while (ret2
== -EAGAIN
);
3312 if (ret2
&& ret2
!= -ENOENT
) {
3317 if (ret
!= -ENOENT
|| ret2
!= -ENOENT
)
3320 btrfs_free_path(path
);
3321 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
3326 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
3327 struct btrfs_root
*root
,
3328 struct extent_buffer
*buf
,
3329 int full_backref
, int inc
)
3331 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3337 struct btrfs_key key
;
3338 struct btrfs_file_extent_item
*fi
;
3342 int (*process_func
)(struct btrfs_trans_handle
*,
3343 struct btrfs_root
*,
3344 u64
, u64
, u64
, u64
, u64
, u64
);
3347 if (btrfs_is_testing(fs_info
))
3350 ref_root
= btrfs_header_owner(buf
);
3351 nritems
= btrfs_header_nritems(buf
);
3352 level
= btrfs_header_level(buf
);
3354 if (!test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) && level
== 0)
3358 process_func
= btrfs_inc_extent_ref
;
3360 process_func
= btrfs_free_extent
;
3363 parent
= buf
->start
;
3367 for (i
= 0; i
< nritems
; i
++) {
3369 btrfs_item_key_to_cpu(buf
, &key
, i
);
3370 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
3372 fi
= btrfs_item_ptr(buf
, i
,
3373 struct btrfs_file_extent_item
);
3374 if (btrfs_file_extent_type(buf
, fi
) ==
3375 BTRFS_FILE_EXTENT_INLINE
)
3377 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
3381 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
3382 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
3383 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
3384 parent
, ref_root
, key
.objectid
,
3389 bytenr
= btrfs_node_blockptr(buf
, i
);
3390 num_bytes
= fs_info
->nodesize
;
3391 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
3392 parent
, ref_root
, level
- 1, 0);
3402 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3403 struct extent_buffer
*buf
, int full_backref
)
3405 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1);
3408 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3409 struct extent_buffer
*buf
, int full_backref
)
3411 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0);
3414 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
3415 struct btrfs_fs_info
*fs_info
,
3416 struct btrfs_path
*path
,
3417 struct btrfs_block_group_cache
*cache
)
3420 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
3422 struct extent_buffer
*leaf
;
3424 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
3431 leaf
= path
->nodes
[0];
3432 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
3433 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
3434 btrfs_mark_buffer_dirty(leaf
);
3436 btrfs_release_path(path
);
3441 static struct btrfs_block_group_cache
*
3442 next_block_group(struct btrfs_fs_info
*fs_info
,
3443 struct btrfs_block_group_cache
*cache
)
3445 struct rb_node
*node
;
3447 spin_lock(&fs_info
->block_group_cache_lock
);
3449 /* If our block group was removed, we need a full search. */
3450 if (RB_EMPTY_NODE(&cache
->cache_node
)) {
3451 const u64 next_bytenr
= cache
->key
.objectid
+ cache
->key
.offset
;
3453 spin_unlock(&fs_info
->block_group_cache_lock
);
3454 btrfs_put_block_group(cache
);
3455 cache
= btrfs_lookup_first_block_group(fs_info
, next_bytenr
); return cache
;
3457 node
= rb_next(&cache
->cache_node
);
3458 btrfs_put_block_group(cache
);
3460 cache
= rb_entry(node
, struct btrfs_block_group_cache
,
3462 btrfs_get_block_group(cache
);
3465 spin_unlock(&fs_info
->block_group_cache_lock
);
3469 static int cache_save_setup(struct btrfs_block_group_cache
*block_group
,
3470 struct btrfs_trans_handle
*trans
,
3471 struct btrfs_path
*path
)
3473 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
3474 struct btrfs_root
*root
= fs_info
->tree_root
;
3475 struct inode
*inode
= NULL
;
3476 struct extent_changeset
*data_reserved
= NULL
;
3478 int dcs
= BTRFS_DC_ERROR
;
3484 * If this block group is smaller than 100 megs don't bother caching the
3487 if (block_group
->key
.offset
< (100 * SZ_1M
)) {
3488 spin_lock(&block_group
->lock
);
3489 block_group
->disk_cache_state
= BTRFS_DC_WRITTEN
;
3490 spin_unlock(&block_group
->lock
);
3497 inode
= lookup_free_space_inode(fs_info
, block_group
, path
);
3498 if (IS_ERR(inode
) && PTR_ERR(inode
) != -ENOENT
) {
3499 ret
= PTR_ERR(inode
);
3500 btrfs_release_path(path
);
3504 if (IS_ERR(inode
)) {
3508 if (block_group
->ro
)
3511 ret
= create_free_space_inode(fs_info
, trans
, block_group
,
3519 * We want to set the generation to 0, that way if anything goes wrong
3520 * from here on out we know not to trust this cache when we load up next
3523 BTRFS_I(inode
)->generation
= 0;
3524 ret
= btrfs_update_inode(trans
, root
, inode
);
3527 * So theoretically we could recover from this, simply set the
3528 * super cache generation to 0 so we know to invalidate the
3529 * cache, but then we'd have to keep track of the block groups
3530 * that fail this way so we know we _have_ to reset this cache
3531 * before the next commit or risk reading stale cache. So to
3532 * limit our exposure to horrible edge cases lets just abort the
3533 * transaction, this only happens in really bad situations
3536 btrfs_abort_transaction(trans
, ret
);
3541 /* We've already setup this transaction, go ahead and exit */
3542 if (block_group
->cache_generation
== trans
->transid
&&
3543 i_size_read(inode
)) {
3544 dcs
= BTRFS_DC_SETUP
;
3548 if (i_size_read(inode
) > 0) {
3549 ret
= btrfs_check_trunc_cache_free_space(fs_info
,
3550 &fs_info
->global_block_rsv
);
3554 ret
= btrfs_truncate_free_space_cache(trans
, NULL
, inode
);
3559 spin_lock(&block_group
->lock
);
3560 if (block_group
->cached
!= BTRFS_CACHE_FINISHED
||
3561 !btrfs_test_opt(fs_info
, SPACE_CACHE
)) {
3563 * don't bother trying to write stuff out _if_
3564 * a) we're not cached,
3565 * b) we're with nospace_cache mount option,
3566 * c) we're with v2 space_cache (FREE_SPACE_TREE).
3568 dcs
= BTRFS_DC_WRITTEN
;
3569 spin_unlock(&block_group
->lock
);
3572 spin_unlock(&block_group
->lock
);
3575 * We hit an ENOSPC when setting up the cache in this transaction, just
3576 * skip doing the setup, we've already cleared the cache so we're safe.
3578 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC
, &trans
->transaction
->flags
)) {
3584 * Try to preallocate enough space based on how big the block group is.
3585 * Keep in mind this has to include any pinned space which could end up
3586 * taking up quite a bit since it's not folded into the other space
3589 num_pages
= div_u64(block_group
->key
.offset
, SZ_256M
);
3594 num_pages
*= PAGE_SIZE
;
3596 ret
= btrfs_check_data_free_space(inode
, &data_reserved
, 0, num_pages
);
3600 ret
= btrfs_prealloc_file_range_trans(inode
, trans
, 0, 0, num_pages
,
3601 num_pages
, num_pages
,
3604 * Our cache requires contiguous chunks so that we don't modify a bunch
3605 * of metadata or split extents when writing the cache out, which means
3606 * we can enospc if we are heavily fragmented in addition to just normal
3607 * out of space conditions. So if we hit this just skip setting up any
3608 * other block groups for this transaction, maybe we'll unpin enough
3609 * space the next time around.
3612 dcs
= BTRFS_DC_SETUP
;
3613 else if (ret
== -ENOSPC
)
3614 set_bit(BTRFS_TRANS_CACHE_ENOSPC
, &trans
->transaction
->flags
);
3619 btrfs_release_path(path
);
3621 spin_lock(&block_group
->lock
);
3622 if (!ret
&& dcs
== BTRFS_DC_SETUP
)
3623 block_group
->cache_generation
= trans
->transid
;
3624 block_group
->disk_cache_state
= dcs
;
3625 spin_unlock(&block_group
->lock
);
3627 extent_changeset_free(data_reserved
);
3631 int btrfs_setup_space_cache(struct btrfs_trans_handle
*trans
,
3632 struct btrfs_fs_info
*fs_info
)
3634 struct btrfs_block_group_cache
*cache
, *tmp
;
3635 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3636 struct btrfs_path
*path
;
3638 if (list_empty(&cur_trans
->dirty_bgs
) ||
3639 !btrfs_test_opt(fs_info
, SPACE_CACHE
))
3642 path
= btrfs_alloc_path();
3646 /* Could add new block groups, use _safe just in case */
3647 list_for_each_entry_safe(cache
, tmp
, &cur_trans
->dirty_bgs
,
3649 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
)
3650 cache_save_setup(cache
, trans
, path
);
3653 btrfs_free_path(path
);
3658 * transaction commit does final block group cache writeback during a
3659 * critical section where nothing is allowed to change the FS. This is
3660 * required in order for the cache to actually match the block group,
3661 * but can introduce a lot of latency into the commit.
3663 * So, btrfs_start_dirty_block_groups is here to kick off block group
3664 * cache IO. There's a chance we'll have to redo some of it if the
3665 * block group changes again during the commit, but it greatly reduces
3666 * the commit latency by getting rid of the easy block groups while
3667 * we're still allowing others to join the commit.
3669 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle
*trans
,
3670 struct btrfs_fs_info
*fs_info
)
3672 struct btrfs_block_group_cache
*cache
;
3673 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3676 struct btrfs_path
*path
= NULL
;
3678 struct list_head
*io
= &cur_trans
->io_bgs
;
3679 int num_started
= 0;
3682 spin_lock(&cur_trans
->dirty_bgs_lock
);
3683 if (list_empty(&cur_trans
->dirty_bgs
)) {
3684 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3687 list_splice_init(&cur_trans
->dirty_bgs
, &dirty
);
3688 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3692 * make sure all the block groups on our dirty list actually
3695 btrfs_create_pending_block_groups(trans
, fs_info
);
3698 path
= btrfs_alloc_path();
3704 * cache_write_mutex is here only to save us from balance or automatic
3705 * removal of empty block groups deleting this block group while we are
3706 * writing out the cache
3708 mutex_lock(&trans
->transaction
->cache_write_mutex
);
3709 while (!list_empty(&dirty
)) {
3710 cache
= list_first_entry(&dirty
,
3711 struct btrfs_block_group_cache
,
3714 * this can happen if something re-dirties a block
3715 * group that is already under IO. Just wait for it to
3716 * finish and then do it all again
3718 if (!list_empty(&cache
->io_list
)) {
3719 list_del_init(&cache
->io_list
);
3720 btrfs_wait_cache_io(trans
, cache
, path
);
3721 btrfs_put_block_group(cache
);
3726 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3727 * if it should update the cache_state. Don't delete
3728 * until after we wait.
3730 * Since we're not running in the commit critical section
3731 * we need the dirty_bgs_lock to protect from update_block_group
3733 spin_lock(&cur_trans
->dirty_bgs_lock
);
3734 list_del_init(&cache
->dirty_list
);
3735 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3739 cache_save_setup(cache
, trans
, path
);
3741 if (cache
->disk_cache_state
== BTRFS_DC_SETUP
) {
3742 cache
->io_ctl
.inode
= NULL
;
3743 ret
= btrfs_write_out_cache(fs_info
, trans
,
3745 if (ret
== 0 && cache
->io_ctl
.inode
) {
3750 * the cache_write_mutex is protecting
3753 list_add_tail(&cache
->io_list
, io
);
3756 * if we failed to write the cache, the
3757 * generation will be bad and life goes on
3763 ret
= write_one_cache_group(trans
, fs_info
,
3766 * Our block group might still be attached to the list
3767 * of new block groups in the transaction handle of some
3768 * other task (struct btrfs_trans_handle->new_bgs). This
3769 * means its block group item isn't yet in the extent
3770 * tree. If this happens ignore the error, as we will
3771 * try again later in the critical section of the
3772 * transaction commit.
3774 if (ret
== -ENOENT
) {
3776 spin_lock(&cur_trans
->dirty_bgs_lock
);
3777 if (list_empty(&cache
->dirty_list
)) {
3778 list_add_tail(&cache
->dirty_list
,
3779 &cur_trans
->dirty_bgs
);
3780 btrfs_get_block_group(cache
);
3782 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3784 btrfs_abort_transaction(trans
, ret
);
3788 /* if its not on the io list, we need to put the block group */
3790 btrfs_put_block_group(cache
);
3796 * Avoid blocking other tasks for too long. It might even save
3797 * us from writing caches for block groups that are going to be
3800 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
3801 mutex_lock(&trans
->transaction
->cache_write_mutex
);
3803 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
3806 * go through delayed refs for all the stuff we've just kicked off
3807 * and then loop back (just once)
3809 ret
= btrfs_run_delayed_refs(trans
, fs_info
, 0);
3810 if (!ret
&& loops
== 0) {
3812 spin_lock(&cur_trans
->dirty_bgs_lock
);
3813 list_splice_init(&cur_trans
->dirty_bgs
, &dirty
);
3815 * dirty_bgs_lock protects us from concurrent block group
3816 * deletes too (not just cache_write_mutex).
3818 if (!list_empty(&dirty
)) {
3819 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3822 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3823 } else if (ret
< 0) {
3824 btrfs_cleanup_dirty_bgs(cur_trans
, fs_info
);
3827 btrfs_free_path(path
);
3831 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
3832 struct btrfs_fs_info
*fs_info
)
3834 struct btrfs_block_group_cache
*cache
;
3835 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3838 struct btrfs_path
*path
;
3839 struct list_head
*io
= &cur_trans
->io_bgs
;
3840 int num_started
= 0;
3842 path
= btrfs_alloc_path();
3847 * Even though we are in the critical section of the transaction commit,
3848 * we can still have concurrent tasks adding elements to this
3849 * transaction's list of dirty block groups. These tasks correspond to
3850 * endio free space workers started when writeback finishes for a
3851 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3852 * allocate new block groups as a result of COWing nodes of the root
3853 * tree when updating the free space inode. The writeback for the space
3854 * caches is triggered by an earlier call to
3855 * btrfs_start_dirty_block_groups() and iterations of the following
3857 * Also we want to do the cache_save_setup first and then run the
3858 * delayed refs to make sure we have the best chance at doing this all
3861 spin_lock(&cur_trans
->dirty_bgs_lock
);
3862 while (!list_empty(&cur_trans
->dirty_bgs
)) {
3863 cache
= list_first_entry(&cur_trans
->dirty_bgs
,
3864 struct btrfs_block_group_cache
,
3868 * this can happen if cache_save_setup re-dirties a block
3869 * group that is already under IO. Just wait for it to
3870 * finish and then do it all again
3872 if (!list_empty(&cache
->io_list
)) {
3873 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3874 list_del_init(&cache
->io_list
);
3875 btrfs_wait_cache_io(trans
, cache
, path
);
3876 btrfs_put_block_group(cache
);
3877 spin_lock(&cur_trans
->dirty_bgs_lock
);
3881 * don't remove from the dirty list until after we've waited
3884 list_del_init(&cache
->dirty_list
);
3885 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3888 cache_save_setup(cache
, trans
, path
);
3891 ret
= btrfs_run_delayed_refs(trans
, fs_info
,
3892 (unsigned long) -1);
3894 if (!ret
&& cache
->disk_cache_state
== BTRFS_DC_SETUP
) {
3895 cache
->io_ctl
.inode
= NULL
;
3896 ret
= btrfs_write_out_cache(fs_info
, trans
,
3898 if (ret
== 0 && cache
->io_ctl
.inode
) {
3901 list_add_tail(&cache
->io_list
, io
);
3904 * if we failed to write the cache, the
3905 * generation will be bad and life goes on
3911 ret
= write_one_cache_group(trans
, fs_info
,
3914 * One of the free space endio workers might have
3915 * created a new block group while updating a free space
3916 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3917 * and hasn't released its transaction handle yet, in
3918 * which case the new block group is still attached to
3919 * its transaction handle and its creation has not
3920 * finished yet (no block group item in the extent tree
3921 * yet, etc). If this is the case, wait for all free
3922 * space endio workers to finish and retry. This is a
3923 * a very rare case so no need for a more efficient and
3926 if (ret
== -ENOENT
) {
3927 wait_event(cur_trans
->writer_wait
,
3928 atomic_read(&cur_trans
->num_writers
) == 1);
3929 ret
= write_one_cache_group(trans
, fs_info
,
3933 btrfs_abort_transaction(trans
, ret
);
3936 /* if its not on the io list, we need to put the block group */
3938 btrfs_put_block_group(cache
);
3939 spin_lock(&cur_trans
->dirty_bgs_lock
);
3941 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3943 while (!list_empty(io
)) {
3944 cache
= list_first_entry(io
, struct btrfs_block_group_cache
,
3946 list_del_init(&cache
->io_list
);
3947 btrfs_wait_cache_io(trans
, cache
, path
);
3948 btrfs_put_block_group(cache
);
3951 btrfs_free_path(path
);
3955 int btrfs_extent_readonly(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
3957 struct btrfs_block_group_cache
*block_group
;
3960 block_group
= btrfs_lookup_block_group(fs_info
, bytenr
);
3961 if (!block_group
|| block_group
->ro
)
3964 btrfs_put_block_group(block_group
);
3968 bool btrfs_inc_nocow_writers(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
3970 struct btrfs_block_group_cache
*bg
;
3973 bg
= btrfs_lookup_block_group(fs_info
, bytenr
);
3977 spin_lock(&bg
->lock
);
3981 atomic_inc(&bg
->nocow_writers
);
3982 spin_unlock(&bg
->lock
);
3984 /* no put on block group, done by btrfs_dec_nocow_writers */
3986 btrfs_put_block_group(bg
);
3992 void btrfs_dec_nocow_writers(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
3994 struct btrfs_block_group_cache
*bg
;
3996 bg
= btrfs_lookup_block_group(fs_info
, bytenr
);
3998 if (atomic_dec_and_test(&bg
->nocow_writers
))
3999 wake_up_atomic_t(&bg
->nocow_writers
);
4001 * Once for our lookup and once for the lookup done by a previous call
4002 * to btrfs_inc_nocow_writers()
4004 btrfs_put_block_group(bg
);
4005 btrfs_put_block_group(bg
);
4008 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache
*bg
)
4010 wait_on_atomic_t(&bg
->nocow_writers
, atomic_t_wait
,
4011 TASK_UNINTERRUPTIBLE
);
4014 static const char *alloc_name(u64 flags
)
4017 case BTRFS_BLOCK_GROUP_METADATA
|BTRFS_BLOCK_GROUP_DATA
:
4019 case BTRFS_BLOCK_GROUP_METADATA
:
4021 case BTRFS_BLOCK_GROUP_DATA
:
4023 case BTRFS_BLOCK_GROUP_SYSTEM
:
4027 return "invalid-combination";
4031 static int create_space_info(struct btrfs_fs_info
*info
, u64 flags
,
4032 struct btrfs_space_info
**new)
4035 struct btrfs_space_info
*space_info
;
4039 space_info
= kzalloc(sizeof(*space_info
), GFP_NOFS
);
4043 ret
= percpu_counter_init(&space_info
->total_bytes_pinned
, 0,
4050 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
4051 INIT_LIST_HEAD(&space_info
->block_groups
[i
]);
4052 init_rwsem(&space_info
->groups_sem
);
4053 spin_lock_init(&space_info
->lock
);
4054 space_info
->flags
= flags
& BTRFS_BLOCK_GROUP_TYPE_MASK
;
4055 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
4056 init_waitqueue_head(&space_info
->wait
);
4057 INIT_LIST_HEAD(&space_info
->ro_bgs
);
4058 INIT_LIST_HEAD(&space_info
->tickets
);
4059 INIT_LIST_HEAD(&space_info
->priority_tickets
);
4061 ret
= kobject_init_and_add(&space_info
->kobj
, &space_info_ktype
,
4062 info
->space_info_kobj
, "%s",
4063 alloc_name(space_info
->flags
));
4065 kobject_put(&space_info
->kobj
);
4070 list_add_rcu(&space_info
->list
, &info
->space_info
);
4071 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
4072 info
->data_sinfo
= space_info
;
4077 static void update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
4078 u64 total_bytes
, u64 bytes_used
,
4080 struct btrfs_space_info
**space_info
)
4082 struct btrfs_space_info
*found
;
4085 if (flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
4086 BTRFS_BLOCK_GROUP_RAID10
))
4091 found
= __find_space_info(info
, flags
);
4093 spin_lock(&found
->lock
);
4094 found
->total_bytes
+= total_bytes
;
4095 found
->disk_total
+= total_bytes
* factor
;
4096 found
->bytes_used
+= bytes_used
;
4097 found
->disk_used
+= bytes_used
* factor
;
4098 found
->bytes_readonly
+= bytes_readonly
;
4099 if (total_bytes
> 0)
4101 space_info_add_new_bytes(info
, found
, total_bytes
-
4102 bytes_used
- bytes_readonly
);
4103 spin_unlock(&found
->lock
);
4104 *space_info
= found
;
4107 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
4109 u64 extra_flags
= chunk_to_extended(flags
) &
4110 BTRFS_EXTENDED_PROFILE_MASK
;
4112 write_seqlock(&fs_info
->profiles_lock
);
4113 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
4114 fs_info
->avail_data_alloc_bits
|= extra_flags
;
4115 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
4116 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
4117 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
4118 fs_info
->avail_system_alloc_bits
|= extra_flags
;
4119 write_sequnlock(&fs_info
->profiles_lock
);
4123 * returns target flags in extended format or 0 if restripe for this
4124 * chunk_type is not in progress
4126 * should be called with either volume_mutex or balance_lock held
4128 static u64
get_restripe_target(struct btrfs_fs_info
*fs_info
, u64 flags
)
4130 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
4136 if (flags
& BTRFS_BLOCK_GROUP_DATA
&&
4137 bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
4138 target
= BTRFS_BLOCK_GROUP_DATA
| bctl
->data
.target
;
4139 } else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
&&
4140 bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
4141 target
= BTRFS_BLOCK_GROUP_SYSTEM
| bctl
->sys
.target
;
4142 } else if (flags
& BTRFS_BLOCK_GROUP_METADATA
&&
4143 bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
4144 target
= BTRFS_BLOCK_GROUP_METADATA
| bctl
->meta
.target
;
4151 * @flags: available profiles in extended format (see ctree.h)
4153 * Returns reduced profile in chunk format. If profile changing is in
4154 * progress (either running or paused) picks the target profile (if it's
4155 * already available), otherwise falls back to plain reducing.
4157 static u64
btrfs_reduce_alloc_profile(struct btrfs_fs_info
*fs_info
, u64 flags
)
4159 u64 num_devices
= fs_info
->fs_devices
->rw_devices
;
4165 * see if restripe for this chunk_type is in progress, if so
4166 * try to reduce to the target profile
4168 spin_lock(&fs_info
->balance_lock
);
4169 target
= get_restripe_target(fs_info
, flags
);
4171 /* pick target profile only if it's already available */
4172 if ((flags
& target
) & BTRFS_EXTENDED_PROFILE_MASK
) {
4173 spin_unlock(&fs_info
->balance_lock
);
4174 return extended_to_chunk(target
);
4177 spin_unlock(&fs_info
->balance_lock
);
4179 /* First, mask out the RAID levels which aren't possible */
4180 for (raid_type
= 0; raid_type
< BTRFS_NR_RAID_TYPES
; raid_type
++) {
4181 if (num_devices
>= btrfs_raid_array
[raid_type
].devs_min
)
4182 allowed
|= btrfs_raid_group
[raid_type
];
4186 if (allowed
& BTRFS_BLOCK_GROUP_RAID6
)
4187 allowed
= BTRFS_BLOCK_GROUP_RAID6
;
4188 else if (allowed
& BTRFS_BLOCK_GROUP_RAID5
)
4189 allowed
= BTRFS_BLOCK_GROUP_RAID5
;
4190 else if (allowed
& BTRFS_BLOCK_GROUP_RAID10
)
4191 allowed
= BTRFS_BLOCK_GROUP_RAID10
;
4192 else if (allowed
& BTRFS_BLOCK_GROUP_RAID1
)
4193 allowed
= BTRFS_BLOCK_GROUP_RAID1
;
4194 else if (allowed
& BTRFS_BLOCK_GROUP_RAID0
)
4195 allowed
= BTRFS_BLOCK_GROUP_RAID0
;
4197 flags
&= ~BTRFS_BLOCK_GROUP_PROFILE_MASK
;
4199 return extended_to_chunk(flags
| allowed
);
4202 static u64
get_alloc_profile(struct btrfs_fs_info
*fs_info
, u64 orig_flags
)
4209 seq
= read_seqbegin(&fs_info
->profiles_lock
);
4211 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
4212 flags
|= fs_info
->avail_data_alloc_bits
;
4213 else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
4214 flags
|= fs_info
->avail_system_alloc_bits
;
4215 else if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
4216 flags
|= fs_info
->avail_metadata_alloc_bits
;
4217 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
4219 return btrfs_reduce_alloc_profile(fs_info
, flags
);
4222 static u64
get_alloc_profile_by_root(struct btrfs_root
*root
, int data
)
4224 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4229 flags
= BTRFS_BLOCK_GROUP_DATA
;
4230 else if (root
== fs_info
->chunk_root
)
4231 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
4233 flags
= BTRFS_BLOCK_GROUP_METADATA
;
4235 ret
= get_alloc_profile(fs_info
, flags
);
4239 u64
btrfs_data_alloc_profile(struct btrfs_fs_info
*fs_info
)
4241 return get_alloc_profile(fs_info
, BTRFS_BLOCK_GROUP_DATA
);
4244 u64
btrfs_metadata_alloc_profile(struct btrfs_fs_info
*fs_info
)
4246 return get_alloc_profile(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4249 u64
btrfs_system_alloc_profile(struct btrfs_fs_info
*fs_info
)
4251 return get_alloc_profile(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
4254 static u64
btrfs_space_info_used(struct btrfs_space_info
*s_info
,
4255 bool may_use_included
)
4258 return s_info
->bytes_used
+ s_info
->bytes_reserved
+
4259 s_info
->bytes_pinned
+ s_info
->bytes_readonly
+
4260 (may_use_included
? s_info
->bytes_may_use
: 0);
4263 int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode
*inode
, u64 bytes
)
4265 struct btrfs_root
*root
= inode
->root
;
4266 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4267 struct btrfs_space_info
*data_sinfo
= fs_info
->data_sinfo
;
4270 int need_commit
= 2;
4271 int have_pinned_space
;
4273 /* make sure bytes are sectorsize aligned */
4274 bytes
= ALIGN(bytes
, fs_info
->sectorsize
);
4276 if (btrfs_is_free_space_inode(inode
)) {
4278 ASSERT(current
->journal_info
);
4282 /* make sure we have enough space to handle the data first */
4283 spin_lock(&data_sinfo
->lock
);
4284 used
= btrfs_space_info_used(data_sinfo
, true);
4286 if (used
+ bytes
> data_sinfo
->total_bytes
) {
4287 struct btrfs_trans_handle
*trans
;
4290 * if we don't have enough free bytes in this space then we need
4291 * to alloc a new chunk.
4293 if (!data_sinfo
->full
) {
4296 data_sinfo
->force_alloc
= CHUNK_ALLOC_FORCE
;
4297 spin_unlock(&data_sinfo
->lock
);
4299 alloc_target
= btrfs_data_alloc_profile(fs_info
);
4301 * It is ugly that we don't call nolock join
4302 * transaction for the free space inode case here.
4303 * But it is safe because we only do the data space
4304 * reservation for the free space cache in the
4305 * transaction context, the common join transaction
4306 * just increase the counter of the current transaction
4307 * handler, doesn't try to acquire the trans_lock of
4310 trans
= btrfs_join_transaction(root
);
4312 return PTR_ERR(trans
);
4314 ret
= do_chunk_alloc(trans
, fs_info
, alloc_target
,
4315 CHUNK_ALLOC_NO_FORCE
);
4316 btrfs_end_transaction(trans
);
4321 have_pinned_space
= 1;
4330 * If we don't have enough pinned space to deal with this
4331 * allocation, and no removed chunk in current transaction,
4332 * don't bother committing the transaction.
4334 have_pinned_space
= percpu_counter_compare(
4335 &data_sinfo
->total_bytes_pinned
,
4336 used
+ bytes
- data_sinfo
->total_bytes
);
4337 spin_unlock(&data_sinfo
->lock
);
4339 /* commit the current transaction and try again */
4342 !atomic_read(&fs_info
->open_ioctl_trans
)) {
4345 if (need_commit
> 0) {
4346 btrfs_start_delalloc_roots(fs_info
, 0, -1);
4347 btrfs_wait_ordered_roots(fs_info
, U64_MAX
, 0,
4351 trans
= btrfs_join_transaction(root
);
4353 return PTR_ERR(trans
);
4354 if (have_pinned_space
>= 0 ||
4355 test_bit(BTRFS_TRANS_HAVE_FREE_BGS
,
4356 &trans
->transaction
->flags
) ||
4358 ret
= btrfs_commit_transaction(trans
);
4362 * The cleaner kthread might still be doing iput
4363 * operations. Wait for it to finish so that
4364 * more space is released.
4366 mutex_lock(&fs_info
->cleaner_delayed_iput_mutex
);
4367 mutex_unlock(&fs_info
->cleaner_delayed_iput_mutex
);
4370 btrfs_end_transaction(trans
);
4374 trace_btrfs_space_reservation(fs_info
,
4375 "space_info:enospc",
4376 data_sinfo
->flags
, bytes
, 1);
4379 data_sinfo
->bytes_may_use
+= bytes
;
4380 trace_btrfs_space_reservation(fs_info
, "space_info",
4381 data_sinfo
->flags
, bytes
, 1);
4382 spin_unlock(&data_sinfo
->lock
);
4387 int btrfs_check_data_free_space(struct inode
*inode
,
4388 struct extent_changeset
**reserved
, u64 start
, u64 len
)
4390 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
4393 /* align the range */
4394 len
= round_up(start
+ len
, fs_info
->sectorsize
) -
4395 round_down(start
, fs_info
->sectorsize
);
4396 start
= round_down(start
, fs_info
->sectorsize
);
4398 ret
= btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode
), len
);
4402 /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */
4403 ret
= btrfs_qgroup_reserve_data(inode
, reserved
, start
, len
);
4405 btrfs_free_reserved_data_space_noquota(inode
, start
, len
);
4412 * Called if we need to clear a data reservation for this inode
4413 * Normally in a error case.
4415 * This one will *NOT* use accurate qgroup reserved space API, just for case
4416 * which we can't sleep and is sure it won't affect qgroup reserved space.
4417 * Like clear_bit_hook().
4419 void btrfs_free_reserved_data_space_noquota(struct inode
*inode
, u64 start
,
4422 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
4423 struct btrfs_space_info
*data_sinfo
;
4425 /* Make sure the range is aligned to sectorsize */
4426 len
= round_up(start
+ len
, fs_info
->sectorsize
) -
4427 round_down(start
, fs_info
->sectorsize
);
4428 start
= round_down(start
, fs_info
->sectorsize
);
4430 data_sinfo
= fs_info
->data_sinfo
;
4431 spin_lock(&data_sinfo
->lock
);
4432 if (WARN_ON(data_sinfo
->bytes_may_use
< len
))
4433 data_sinfo
->bytes_may_use
= 0;
4435 data_sinfo
->bytes_may_use
-= len
;
4436 trace_btrfs_space_reservation(fs_info
, "space_info",
4437 data_sinfo
->flags
, len
, 0);
4438 spin_unlock(&data_sinfo
->lock
);
4442 * Called if we need to clear a data reservation for this inode
4443 * Normally in a error case.
4445 * This one will handle the per-inode data rsv map for accurate reserved
4448 void btrfs_free_reserved_data_space(struct inode
*inode
,
4449 struct extent_changeset
*reserved
, u64 start
, u64 len
)
4451 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4453 /* Make sure the range is aligned to sectorsize */
4454 len
= round_up(start
+ len
, root
->fs_info
->sectorsize
) -
4455 round_down(start
, root
->fs_info
->sectorsize
);
4456 start
= round_down(start
, root
->fs_info
->sectorsize
);
4458 btrfs_free_reserved_data_space_noquota(inode
, start
, len
);
4459 btrfs_qgroup_free_data(inode
, reserved
, start
, len
);
4462 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
4464 struct list_head
*head
= &info
->space_info
;
4465 struct btrfs_space_info
*found
;
4468 list_for_each_entry_rcu(found
, head
, list
) {
4469 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
4470 found
->force_alloc
= CHUNK_ALLOC_FORCE
;
4475 static inline u64
calc_global_rsv_need_space(struct btrfs_block_rsv
*global
)
4477 return (global
->size
<< 1);
4480 static int should_alloc_chunk(struct btrfs_fs_info
*fs_info
,
4481 struct btrfs_space_info
*sinfo
, int force
)
4483 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
4484 u64 bytes_used
= btrfs_space_info_used(sinfo
, false);
4487 if (force
== CHUNK_ALLOC_FORCE
)
4491 * We need to take into account the global rsv because for all intents
4492 * and purposes it's used space. Don't worry about locking the
4493 * global_rsv, it doesn't change except when the transaction commits.
4495 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
4496 bytes_used
+= calc_global_rsv_need_space(global_rsv
);
4499 * in limited mode, we want to have some free space up to
4500 * about 1% of the FS size.
4502 if (force
== CHUNK_ALLOC_LIMITED
) {
4503 thresh
= btrfs_super_total_bytes(fs_info
->super_copy
);
4504 thresh
= max_t(u64
, SZ_64M
, div_factor_fine(thresh
, 1));
4506 if (sinfo
->total_bytes
- bytes_used
< thresh
)
4510 if (bytes_used
+ SZ_2M
< div_factor(sinfo
->total_bytes
, 8))
4515 static u64
get_profile_num_devs(struct btrfs_fs_info
*fs_info
, u64 type
)
4519 if (type
& (BTRFS_BLOCK_GROUP_RAID10
|
4520 BTRFS_BLOCK_GROUP_RAID0
|
4521 BTRFS_BLOCK_GROUP_RAID5
|
4522 BTRFS_BLOCK_GROUP_RAID6
))
4523 num_dev
= fs_info
->fs_devices
->rw_devices
;
4524 else if (type
& BTRFS_BLOCK_GROUP_RAID1
)
4527 num_dev
= 1; /* DUP or single */
4533 * If @is_allocation is true, reserve space in the system space info necessary
4534 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4537 void check_system_chunk(struct btrfs_trans_handle
*trans
,
4538 struct btrfs_fs_info
*fs_info
, u64 type
)
4540 struct btrfs_space_info
*info
;
4547 * Needed because we can end up allocating a system chunk and for an
4548 * atomic and race free space reservation in the chunk block reserve.
4550 ASSERT(mutex_is_locked(&fs_info
->chunk_mutex
));
4552 info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
4553 spin_lock(&info
->lock
);
4554 left
= info
->total_bytes
- btrfs_space_info_used(info
, true);
4555 spin_unlock(&info
->lock
);
4557 num_devs
= get_profile_num_devs(fs_info
, type
);
4559 /* num_devs device items to update and 1 chunk item to add or remove */
4560 thresh
= btrfs_calc_trunc_metadata_size(fs_info
, num_devs
) +
4561 btrfs_calc_trans_metadata_size(fs_info
, 1);
4563 if (left
< thresh
&& btrfs_test_opt(fs_info
, ENOSPC_DEBUG
)) {
4564 btrfs_info(fs_info
, "left=%llu, need=%llu, flags=%llu",
4565 left
, thresh
, type
);
4566 dump_space_info(fs_info
, info
, 0, 0);
4569 if (left
< thresh
) {
4570 u64 flags
= btrfs_system_alloc_profile(fs_info
);
4573 * Ignore failure to create system chunk. We might end up not
4574 * needing it, as we might not need to COW all nodes/leafs from
4575 * the paths we visit in the chunk tree (they were already COWed
4576 * or created in the current transaction for example).
4578 ret
= btrfs_alloc_chunk(trans
, fs_info
, flags
);
4582 ret
= btrfs_block_rsv_add(fs_info
->chunk_root
,
4583 &fs_info
->chunk_block_rsv
,
4584 thresh
, BTRFS_RESERVE_NO_FLUSH
);
4586 trans
->chunk_bytes_reserved
+= thresh
;
4591 * If force is CHUNK_ALLOC_FORCE:
4592 * - return 1 if it successfully allocates a chunk,
4593 * - return errors including -ENOSPC otherwise.
4594 * If force is NOT CHUNK_ALLOC_FORCE:
4595 * - return 0 if it doesn't need to allocate a new chunk,
4596 * - return 1 if it successfully allocates a chunk,
4597 * - return errors including -ENOSPC otherwise.
4599 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
4600 struct btrfs_fs_info
*fs_info
, u64 flags
, int force
)
4602 struct btrfs_space_info
*space_info
;
4603 int wait_for_alloc
= 0;
4606 /* Don't re-enter if we're already allocating a chunk */
4607 if (trans
->allocating_chunk
)
4610 space_info
= __find_space_info(fs_info
, flags
);
4612 ret
= create_space_info(fs_info
, flags
, &space_info
);
4618 spin_lock(&space_info
->lock
);
4619 if (force
< space_info
->force_alloc
)
4620 force
= space_info
->force_alloc
;
4621 if (space_info
->full
) {
4622 if (should_alloc_chunk(fs_info
, space_info
, force
))
4626 spin_unlock(&space_info
->lock
);
4630 if (!should_alloc_chunk(fs_info
, space_info
, force
)) {
4631 spin_unlock(&space_info
->lock
);
4633 } else if (space_info
->chunk_alloc
) {
4636 space_info
->chunk_alloc
= 1;
4639 spin_unlock(&space_info
->lock
);
4641 mutex_lock(&fs_info
->chunk_mutex
);
4644 * The chunk_mutex is held throughout the entirety of a chunk
4645 * allocation, so once we've acquired the chunk_mutex we know that the
4646 * other guy is done and we need to recheck and see if we should
4649 if (wait_for_alloc
) {
4650 mutex_unlock(&fs_info
->chunk_mutex
);
4656 trans
->allocating_chunk
= true;
4659 * If we have mixed data/metadata chunks we want to make sure we keep
4660 * allocating mixed chunks instead of individual chunks.
4662 if (btrfs_mixed_space_info(space_info
))
4663 flags
|= (BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
);
4666 * if we're doing a data chunk, go ahead and make sure that
4667 * we keep a reasonable number of metadata chunks allocated in the
4670 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
4671 fs_info
->data_chunk_allocations
++;
4672 if (!(fs_info
->data_chunk_allocations
%
4673 fs_info
->metadata_ratio
))
4674 force_metadata_allocation(fs_info
);
4678 * Check if we have enough space in SYSTEM chunk because we may need
4679 * to update devices.
4681 check_system_chunk(trans
, fs_info
, flags
);
4683 ret
= btrfs_alloc_chunk(trans
, fs_info
, flags
);
4684 trans
->allocating_chunk
= false;
4686 spin_lock(&space_info
->lock
);
4687 if (ret
< 0 && ret
!= -ENOSPC
)
4690 space_info
->full
= 1;
4693 space_info
->max_extent_size
= 0;
4696 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
4698 space_info
->chunk_alloc
= 0;
4699 spin_unlock(&space_info
->lock
);
4700 mutex_unlock(&fs_info
->chunk_mutex
);
4702 * When we allocate a new chunk we reserve space in the chunk block
4703 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4704 * add new nodes/leafs to it if we end up needing to do it when
4705 * inserting the chunk item and updating device items as part of the
4706 * second phase of chunk allocation, performed by
4707 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4708 * large number of new block groups to create in our transaction
4709 * handle's new_bgs list to avoid exhausting the chunk block reserve
4710 * in extreme cases - like having a single transaction create many new
4711 * block groups when starting to write out the free space caches of all
4712 * the block groups that were made dirty during the lifetime of the
4715 if (trans
->chunk_bytes_reserved
>= (u64
)SZ_2M
)
4716 btrfs_create_pending_block_groups(trans
, fs_info
);
4721 static int can_overcommit(struct btrfs_fs_info
*fs_info
,
4722 struct btrfs_space_info
*space_info
, u64 bytes
,
4723 enum btrfs_reserve_flush_enum flush
,
4726 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
4732 /* Don't overcommit when in mixed mode. */
4733 if (space_info
->flags
& BTRFS_BLOCK_GROUP_DATA
)
4737 profile
= btrfs_system_alloc_profile(fs_info
);
4739 profile
= btrfs_metadata_alloc_profile(fs_info
);
4741 used
= btrfs_space_info_used(space_info
, false);
4744 * We only want to allow over committing if we have lots of actual space
4745 * free, but if we don't have enough space to handle the global reserve
4746 * space then we could end up having a real enospc problem when trying
4747 * to allocate a chunk or some other such important allocation.
4749 spin_lock(&global_rsv
->lock
);
4750 space_size
= calc_global_rsv_need_space(global_rsv
);
4751 spin_unlock(&global_rsv
->lock
);
4752 if (used
+ space_size
>= space_info
->total_bytes
)
4755 used
+= space_info
->bytes_may_use
;
4757 avail
= atomic64_read(&fs_info
->free_chunk_space
);
4760 * If we have dup, raid1 or raid10 then only half of the free
4761 * space is actually useable. For raid56, the space info used
4762 * doesn't include the parity drive, so we don't have to
4765 if (profile
& (BTRFS_BLOCK_GROUP_DUP
|
4766 BTRFS_BLOCK_GROUP_RAID1
|
4767 BTRFS_BLOCK_GROUP_RAID10
))
4771 * If we aren't flushing all things, let us overcommit up to
4772 * 1/2th of the space. If we can flush, don't let us overcommit
4773 * too much, let it overcommit up to 1/8 of the space.
4775 if (flush
== BTRFS_RESERVE_FLUSH_ALL
)
4780 if (used
+ bytes
< space_info
->total_bytes
+ avail
)
4785 static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info
*fs_info
,
4786 unsigned long nr_pages
, int nr_items
)
4788 struct super_block
*sb
= fs_info
->sb
;
4790 if (down_read_trylock(&sb
->s_umount
)) {
4791 writeback_inodes_sb_nr(sb
, nr_pages
, WB_REASON_FS_FREE_SPACE
);
4792 up_read(&sb
->s_umount
);
4795 * We needn't worry the filesystem going from r/w to r/o though
4796 * we don't acquire ->s_umount mutex, because the filesystem
4797 * should guarantee the delalloc inodes list be empty after
4798 * the filesystem is readonly(all dirty pages are written to
4801 btrfs_start_delalloc_roots(fs_info
, 0, nr_items
);
4802 if (!current
->journal_info
)
4803 btrfs_wait_ordered_roots(fs_info
, nr_items
, 0, (u64
)-1);
4807 static inline u64
calc_reclaim_items_nr(struct btrfs_fs_info
*fs_info
,
4813 bytes
= btrfs_calc_trans_metadata_size(fs_info
, 1);
4814 nr
= div64_u64(to_reclaim
, bytes
);
4820 #define EXTENT_SIZE_PER_ITEM SZ_256K
4823 * shrink metadata reservation for delalloc
4825 static void shrink_delalloc(struct btrfs_fs_info
*fs_info
, u64 to_reclaim
,
4826 u64 orig
, bool wait_ordered
)
4828 struct btrfs_space_info
*space_info
;
4829 struct btrfs_trans_handle
*trans
;
4834 unsigned long nr_pages
;
4836 enum btrfs_reserve_flush_enum flush
;
4838 /* Calc the number of the pages we need flush for space reservation */
4839 items
= calc_reclaim_items_nr(fs_info
, to_reclaim
);
4840 to_reclaim
= items
* EXTENT_SIZE_PER_ITEM
;
4842 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4843 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4845 delalloc_bytes
= percpu_counter_sum_positive(
4846 &fs_info
->delalloc_bytes
);
4847 if (delalloc_bytes
== 0) {
4851 btrfs_wait_ordered_roots(fs_info
, items
, 0, (u64
)-1);
4856 while (delalloc_bytes
&& loops
< 3) {
4857 max_reclaim
= min(delalloc_bytes
, to_reclaim
);
4858 nr_pages
= max_reclaim
>> PAGE_SHIFT
;
4859 btrfs_writeback_inodes_sb_nr(fs_info
, nr_pages
, items
);
4861 * We need to wait for the async pages to actually start before
4864 max_reclaim
= atomic_read(&fs_info
->async_delalloc_pages
);
4868 if (max_reclaim
<= nr_pages
)
4871 max_reclaim
-= nr_pages
;
4873 wait_event(fs_info
->async_submit_wait
,
4874 atomic_read(&fs_info
->async_delalloc_pages
) <=
4878 flush
= BTRFS_RESERVE_FLUSH_ALL
;
4880 flush
= BTRFS_RESERVE_NO_FLUSH
;
4881 spin_lock(&space_info
->lock
);
4882 if (list_empty(&space_info
->tickets
) &&
4883 list_empty(&space_info
->priority_tickets
)) {
4884 spin_unlock(&space_info
->lock
);
4887 spin_unlock(&space_info
->lock
);
4890 if (wait_ordered
&& !trans
) {
4891 btrfs_wait_ordered_roots(fs_info
, items
, 0, (u64
)-1);
4893 time_left
= schedule_timeout_killable(1);
4897 delalloc_bytes
= percpu_counter_sum_positive(
4898 &fs_info
->delalloc_bytes
);
4902 struct reserve_ticket
{
4905 struct list_head list
;
4906 wait_queue_head_t wait
;
4910 * maybe_commit_transaction - possibly commit the transaction if its ok to
4911 * @root - the root we're allocating for
4912 * @bytes - the number of bytes we want to reserve
4913 * @force - force the commit
4915 * This will check to make sure that committing the transaction will actually
4916 * get us somewhere and then commit the transaction if it does. Otherwise it
4917 * will return -ENOSPC.
4919 static int may_commit_transaction(struct btrfs_fs_info
*fs_info
,
4920 struct btrfs_space_info
*space_info
)
4922 struct reserve_ticket
*ticket
= NULL
;
4923 struct btrfs_block_rsv
*delayed_rsv
= &fs_info
->delayed_block_rsv
;
4924 struct btrfs_trans_handle
*trans
;
4927 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4931 spin_lock(&space_info
->lock
);
4932 if (!list_empty(&space_info
->priority_tickets
))
4933 ticket
= list_first_entry(&space_info
->priority_tickets
,
4934 struct reserve_ticket
, list
);
4935 else if (!list_empty(&space_info
->tickets
))
4936 ticket
= list_first_entry(&space_info
->tickets
,
4937 struct reserve_ticket
, list
);
4938 bytes
= (ticket
) ? ticket
->bytes
: 0;
4939 spin_unlock(&space_info
->lock
);
4944 /* See if there is enough pinned space to make this reservation */
4945 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4950 * See if there is some space in the delayed insertion reservation for
4953 if (space_info
!= delayed_rsv
->space_info
)
4956 spin_lock(&delayed_rsv
->lock
);
4957 if (delayed_rsv
->size
> bytes
)
4960 bytes
-= delayed_rsv
->size
;
4961 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4963 spin_unlock(&delayed_rsv
->lock
);
4966 spin_unlock(&delayed_rsv
->lock
);
4969 trans
= btrfs_join_transaction(fs_info
->extent_root
);
4973 return btrfs_commit_transaction(trans
);
4977 * Try to flush some data based on policy set by @state. This is only advisory
4978 * and may fail for various reasons. The caller is supposed to examine the
4979 * state of @space_info to detect the outcome.
4981 static void flush_space(struct btrfs_fs_info
*fs_info
,
4982 struct btrfs_space_info
*space_info
, u64 num_bytes
,
4985 struct btrfs_root
*root
= fs_info
->extent_root
;
4986 struct btrfs_trans_handle
*trans
;
4991 case FLUSH_DELAYED_ITEMS_NR
:
4992 case FLUSH_DELAYED_ITEMS
:
4993 if (state
== FLUSH_DELAYED_ITEMS_NR
)
4994 nr
= calc_reclaim_items_nr(fs_info
, num_bytes
) * 2;
4998 trans
= btrfs_join_transaction(root
);
4999 if (IS_ERR(trans
)) {
5000 ret
= PTR_ERR(trans
);
5003 ret
= btrfs_run_delayed_items_nr(trans
, fs_info
, nr
);
5004 btrfs_end_transaction(trans
);
5006 case FLUSH_DELALLOC
:
5007 case FLUSH_DELALLOC_WAIT
:
5008 shrink_delalloc(fs_info
, num_bytes
* 2, num_bytes
,
5009 state
== FLUSH_DELALLOC_WAIT
);
5012 trans
= btrfs_join_transaction(root
);
5013 if (IS_ERR(trans
)) {
5014 ret
= PTR_ERR(trans
);
5017 ret
= do_chunk_alloc(trans
, fs_info
,
5018 btrfs_metadata_alloc_profile(fs_info
),
5019 CHUNK_ALLOC_NO_FORCE
);
5020 btrfs_end_transaction(trans
);
5021 if (ret
> 0 || ret
== -ENOSPC
)
5025 ret
= may_commit_transaction(fs_info
, space_info
);
5032 trace_btrfs_flush_space(fs_info
, space_info
->flags
, num_bytes
, state
,
5038 btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info
*fs_info
,
5039 struct btrfs_space_info
*space_info
,
5042 struct reserve_ticket
*ticket
;
5047 list_for_each_entry(ticket
, &space_info
->tickets
, list
)
5048 to_reclaim
+= ticket
->bytes
;
5049 list_for_each_entry(ticket
, &space_info
->priority_tickets
, list
)
5050 to_reclaim
+= ticket
->bytes
;
5054 to_reclaim
= min_t(u64
, num_online_cpus() * SZ_1M
, SZ_16M
);
5055 if (can_overcommit(fs_info
, space_info
, to_reclaim
,
5056 BTRFS_RESERVE_FLUSH_ALL
, system_chunk
))
5059 used
= btrfs_space_info_used(space_info
, true);
5061 if (can_overcommit(fs_info
, space_info
, SZ_1M
,
5062 BTRFS_RESERVE_FLUSH_ALL
, system_chunk
))
5063 expected
= div_factor_fine(space_info
->total_bytes
, 95);
5065 expected
= div_factor_fine(space_info
->total_bytes
, 90);
5067 if (used
> expected
)
5068 to_reclaim
= used
- expected
;
5071 to_reclaim
= min(to_reclaim
, space_info
->bytes_may_use
+
5072 space_info
->bytes_reserved
);
5076 static inline int need_do_async_reclaim(struct btrfs_fs_info
*fs_info
,
5077 struct btrfs_space_info
*space_info
,
5078 u64 used
, bool system_chunk
)
5080 u64 thresh
= div_factor_fine(space_info
->total_bytes
, 98);
5082 /* If we're just plain full then async reclaim just slows us down. */
5083 if ((space_info
->bytes_used
+ space_info
->bytes_reserved
) >= thresh
)
5086 if (!btrfs_calc_reclaim_metadata_size(fs_info
, space_info
,
5090 return (used
>= thresh
&& !btrfs_fs_closing(fs_info
) &&
5091 !test_bit(BTRFS_FS_STATE_REMOUNTING
, &fs_info
->fs_state
));
5094 static void wake_all_tickets(struct list_head
*head
)
5096 struct reserve_ticket
*ticket
;
5098 while (!list_empty(head
)) {
5099 ticket
= list_first_entry(head
, struct reserve_ticket
, list
);
5100 list_del_init(&ticket
->list
);
5101 ticket
->error
= -ENOSPC
;
5102 wake_up(&ticket
->wait
);
5107 * This is for normal flushers, we can wait all goddamned day if we want to. We
5108 * will loop and continuously try to flush as long as we are making progress.
5109 * We count progress as clearing off tickets each time we have to loop.
5111 static void btrfs_async_reclaim_metadata_space(struct work_struct
*work
)
5113 struct btrfs_fs_info
*fs_info
;
5114 struct btrfs_space_info
*space_info
;
5117 int commit_cycles
= 0;
5118 u64 last_tickets_id
;
5120 fs_info
= container_of(work
, struct btrfs_fs_info
, async_reclaim_work
);
5121 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
5123 spin_lock(&space_info
->lock
);
5124 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
, space_info
,
5127 space_info
->flush
= 0;
5128 spin_unlock(&space_info
->lock
);
5131 last_tickets_id
= space_info
->tickets_id
;
5132 spin_unlock(&space_info
->lock
);
5134 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
5136 flush_space(fs_info
, space_info
, to_reclaim
, flush_state
);
5137 spin_lock(&space_info
->lock
);
5138 if (list_empty(&space_info
->tickets
)) {
5139 space_info
->flush
= 0;
5140 spin_unlock(&space_info
->lock
);
5143 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
,
5146 if (last_tickets_id
== space_info
->tickets_id
) {
5149 last_tickets_id
= space_info
->tickets_id
;
5150 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
5155 if (flush_state
> COMMIT_TRANS
) {
5157 if (commit_cycles
> 2) {
5158 wake_all_tickets(&space_info
->tickets
);
5159 space_info
->flush
= 0;
5161 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
5164 spin_unlock(&space_info
->lock
);
5165 } while (flush_state
<= COMMIT_TRANS
);
5168 void btrfs_init_async_reclaim_work(struct work_struct
*work
)
5170 INIT_WORK(work
, btrfs_async_reclaim_metadata_space
);
5173 static void priority_reclaim_metadata_space(struct btrfs_fs_info
*fs_info
,
5174 struct btrfs_space_info
*space_info
,
5175 struct reserve_ticket
*ticket
)
5178 int flush_state
= FLUSH_DELAYED_ITEMS_NR
;
5180 spin_lock(&space_info
->lock
);
5181 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
, space_info
,
5184 spin_unlock(&space_info
->lock
);
5187 spin_unlock(&space_info
->lock
);
5190 flush_space(fs_info
, space_info
, to_reclaim
, flush_state
);
5192 spin_lock(&space_info
->lock
);
5193 if (ticket
->bytes
== 0) {
5194 spin_unlock(&space_info
->lock
);
5197 spin_unlock(&space_info
->lock
);
5200 * Priority flushers can't wait on delalloc without
5203 if (flush_state
== FLUSH_DELALLOC
||
5204 flush_state
== FLUSH_DELALLOC_WAIT
)
5205 flush_state
= ALLOC_CHUNK
;
5206 } while (flush_state
< COMMIT_TRANS
);
5209 static int wait_reserve_ticket(struct btrfs_fs_info
*fs_info
,
5210 struct btrfs_space_info
*space_info
,
5211 struct reserve_ticket
*ticket
, u64 orig_bytes
)
5217 spin_lock(&space_info
->lock
);
5218 while (ticket
->bytes
> 0 && ticket
->error
== 0) {
5219 ret
= prepare_to_wait_event(&ticket
->wait
, &wait
, TASK_KILLABLE
);
5224 spin_unlock(&space_info
->lock
);
5228 finish_wait(&ticket
->wait
, &wait
);
5229 spin_lock(&space_info
->lock
);
5232 ret
= ticket
->error
;
5233 if (!list_empty(&ticket
->list
))
5234 list_del_init(&ticket
->list
);
5235 if (ticket
->bytes
&& ticket
->bytes
< orig_bytes
) {
5236 u64 num_bytes
= orig_bytes
- ticket
->bytes
;
5237 space_info
->bytes_may_use
-= num_bytes
;
5238 trace_btrfs_space_reservation(fs_info
, "space_info",
5239 space_info
->flags
, num_bytes
, 0);
5241 spin_unlock(&space_info
->lock
);
5247 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5248 * @root - the root we're allocating for
5249 * @space_info - the space info we want to allocate from
5250 * @orig_bytes - the number of bytes we want
5251 * @flush - whether or not we can flush to make our reservation
5253 * This will reserve orig_bytes number of bytes from the space info associated
5254 * with the block_rsv. If there is not enough space it will make an attempt to
5255 * flush out space to make room. It will do this by flushing delalloc if
5256 * possible or committing the transaction. If flush is 0 then no attempts to
5257 * regain reservations will be made and this will fail if there is not enough
5260 static int __reserve_metadata_bytes(struct btrfs_fs_info
*fs_info
,
5261 struct btrfs_space_info
*space_info
,
5263 enum btrfs_reserve_flush_enum flush
,
5266 struct reserve_ticket ticket
;
5271 ASSERT(!current
->journal_info
|| flush
!= BTRFS_RESERVE_FLUSH_ALL
);
5273 spin_lock(&space_info
->lock
);
5275 used
= btrfs_space_info_used(space_info
, true);
5278 * If we have enough space then hooray, make our reservation and carry
5279 * on. If not see if we can overcommit, and if we can, hooray carry on.
5280 * If not things get more complicated.
5282 if (used
+ orig_bytes
<= space_info
->total_bytes
) {
5283 space_info
->bytes_may_use
+= orig_bytes
;
5284 trace_btrfs_space_reservation(fs_info
, "space_info",
5285 space_info
->flags
, orig_bytes
, 1);
5287 } else if (can_overcommit(fs_info
, space_info
, orig_bytes
, flush
,
5289 space_info
->bytes_may_use
+= orig_bytes
;
5290 trace_btrfs_space_reservation(fs_info
, "space_info",
5291 space_info
->flags
, orig_bytes
, 1);
5296 * If we couldn't make a reservation then setup our reservation ticket
5297 * and kick the async worker if it's not already running.
5299 * If we are a priority flusher then we just need to add our ticket to
5300 * the list and we will do our own flushing further down.
5302 if (ret
&& flush
!= BTRFS_RESERVE_NO_FLUSH
) {
5303 ticket
.bytes
= orig_bytes
;
5305 init_waitqueue_head(&ticket
.wait
);
5306 if (flush
== BTRFS_RESERVE_FLUSH_ALL
) {
5307 list_add_tail(&ticket
.list
, &space_info
->tickets
);
5308 if (!space_info
->flush
) {
5309 space_info
->flush
= 1;
5310 trace_btrfs_trigger_flush(fs_info
,
5314 queue_work(system_unbound_wq
,
5315 &fs_info
->async_reclaim_work
);
5318 list_add_tail(&ticket
.list
,
5319 &space_info
->priority_tickets
);
5321 } else if (!ret
&& space_info
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
5324 * We will do the space reservation dance during log replay,
5325 * which means we won't have fs_info->fs_root set, so don't do
5326 * the async reclaim as we will panic.
5328 if (!test_bit(BTRFS_FS_LOG_RECOVERING
, &fs_info
->flags
) &&
5329 need_do_async_reclaim(fs_info
, space_info
,
5330 used
, system_chunk
) &&
5331 !work_busy(&fs_info
->async_reclaim_work
)) {
5332 trace_btrfs_trigger_flush(fs_info
, space_info
->flags
,
5333 orig_bytes
, flush
, "preempt");
5334 queue_work(system_unbound_wq
,
5335 &fs_info
->async_reclaim_work
);
5338 spin_unlock(&space_info
->lock
);
5339 if (!ret
|| flush
== BTRFS_RESERVE_NO_FLUSH
)
5342 if (flush
== BTRFS_RESERVE_FLUSH_ALL
)
5343 return wait_reserve_ticket(fs_info
, space_info
, &ticket
,
5347 priority_reclaim_metadata_space(fs_info
, space_info
, &ticket
);
5348 spin_lock(&space_info
->lock
);
5350 if (ticket
.bytes
< orig_bytes
) {
5351 u64 num_bytes
= orig_bytes
- ticket
.bytes
;
5352 space_info
->bytes_may_use
-= num_bytes
;
5353 trace_btrfs_space_reservation(fs_info
, "space_info",
5358 list_del_init(&ticket
.list
);
5361 spin_unlock(&space_info
->lock
);
5362 ASSERT(list_empty(&ticket
.list
));
5367 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5368 * @root - the root we're allocating for
5369 * @block_rsv - the block_rsv we're allocating for
5370 * @orig_bytes - the number of bytes we want
5371 * @flush - whether or not we can flush to make our reservation
5373 * This will reserve orgi_bytes number of bytes from the space info associated
5374 * with the block_rsv. If there is not enough space it will make an attempt to
5375 * flush out space to make room. It will do this by flushing delalloc if
5376 * possible or committing the transaction. If flush is 0 then no attempts to
5377 * regain reservations will be made and this will fail if there is not enough
5380 static int reserve_metadata_bytes(struct btrfs_root
*root
,
5381 struct btrfs_block_rsv
*block_rsv
,
5383 enum btrfs_reserve_flush_enum flush
)
5385 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5386 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5388 bool system_chunk
= (root
== fs_info
->chunk_root
);
5390 ret
= __reserve_metadata_bytes(fs_info
, block_rsv
->space_info
,
5391 orig_bytes
, flush
, system_chunk
);
5392 if (ret
== -ENOSPC
&&
5393 unlikely(root
->orphan_cleanup_state
== ORPHAN_CLEANUP_STARTED
)) {
5394 if (block_rsv
!= global_rsv
&&
5395 !block_rsv_use_bytes(global_rsv
, orig_bytes
))
5399 trace_btrfs_space_reservation(fs_info
, "space_info:enospc",
5400 block_rsv
->space_info
->flags
,
5405 static struct btrfs_block_rsv
*get_block_rsv(
5406 const struct btrfs_trans_handle
*trans
,
5407 const struct btrfs_root
*root
)
5409 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5410 struct btrfs_block_rsv
*block_rsv
= NULL
;
5412 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
5413 (root
== fs_info
->csum_root
&& trans
->adding_csums
) ||
5414 (root
== fs_info
->uuid_root
))
5415 block_rsv
= trans
->block_rsv
;
5418 block_rsv
= root
->block_rsv
;
5421 block_rsv
= &fs_info
->empty_block_rsv
;
5426 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
5430 spin_lock(&block_rsv
->lock
);
5431 if (block_rsv
->reserved
>= num_bytes
) {
5432 block_rsv
->reserved
-= num_bytes
;
5433 if (block_rsv
->reserved
< block_rsv
->size
)
5434 block_rsv
->full
= 0;
5437 spin_unlock(&block_rsv
->lock
);
5441 static void block_rsv_add_bytes(struct btrfs_block_rsv
*block_rsv
,
5442 u64 num_bytes
, int update_size
)
5444 spin_lock(&block_rsv
->lock
);
5445 block_rsv
->reserved
+= num_bytes
;
5447 block_rsv
->size
+= num_bytes
;
5448 else if (block_rsv
->reserved
>= block_rsv
->size
)
5449 block_rsv
->full
= 1;
5450 spin_unlock(&block_rsv
->lock
);
5453 int btrfs_cond_migrate_bytes(struct btrfs_fs_info
*fs_info
,
5454 struct btrfs_block_rsv
*dest
, u64 num_bytes
,
5457 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5460 if (global_rsv
->space_info
!= dest
->space_info
)
5463 spin_lock(&global_rsv
->lock
);
5464 min_bytes
= div_factor(global_rsv
->size
, min_factor
);
5465 if (global_rsv
->reserved
< min_bytes
+ num_bytes
) {
5466 spin_unlock(&global_rsv
->lock
);
5469 global_rsv
->reserved
-= num_bytes
;
5470 if (global_rsv
->reserved
< global_rsv
->size
)
5471 global_rsv
->full
= 0;
5472 spin_unlock(&global_rsv
->lock
);
5474 block_rsv_add_bytes(dest
, num_bytes
, 1);
5479 * This is for space we already have accounted in space_info->bytes_may_use, so
5480 * basically when we're returning space from block_rsv's.
5482 static void space_info_add_old_bytes(struct btrfs_fs_info
*fs_info
,
5483 struct btrfs_space_info
*space_info
,
5486 struct reserve_ticket
*ticket
;
5487 struct list_head
*head
;
5489 enum btrfs_reserve_flush_enum flush
= BTRFS_RESERVE_NO_FLUSH
;
5490 bool check_overcommit
= false;
5492 spin_lock(&space_info
->lock
);
5493 head
= &space_info
->priority_tickets
;
5496 * If we are over our limit then we need to check and see if we can
5497 * overcommit, and if we can't then we just need to free up our space
5498 * and not satisfy any requests.
5500 used
= btrfs_space_info_used(space_info
, true);
5501 if (used
- num_bytes
>= space_info
->total_bytes
)
5502 check_overcommit
= true;
5504 while (!list_empty(head
) && num_bytes
) {
5505 ticket
= list_first_entry(head
, struct reserve_ticket
,
5508 * We use 0 bytes because this space is already reserved, so
5509 * adding the ticket space would be a double count.
5511 if (check_overcommit
&&
5512 !can_overcommit(fs_info
, space_info
, 0, flush
, false))
5514 if (num_bytes
>= ticket
->bytes
) {
5515 list_del_init(&ticket
->list
);
5516 num_bytes
-= ticket
->bytes
;
5518 space_info
->tickets_id
++;
5519 wake_up(&ticket
->wait
);
5521 ticket
->bytes
-= num_bytes
;
5526 if (num_bytes
&& head
== &space_info
->priority_tickets
) {
5527 head
= &space_info
->tickets
;
5528 flush
= BTRFS_RESERVE_FLUSH_ALL
;
5531 space_info
->bytes_may_use
-= num_bytes
;
5532 trace_btrfs_space_reservation(fs_info
, "space_info",
5533 space_info
->flags
, num_bytes
, 0);
5534 spin_unlock(&space_info
->lock
);
5538 * This is for newly allocated space that isn't accounted in
5539 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5540 * we use this helper.
5542 static void space_info_add_new_bytes(struct btrfs_fs_info
*fs_info
,
5543 struct btrfs_space_info
*space_info
,
5546 struct reserve_ticket
*ticket
;
5547 struct list_head
*head
= &space_info
->priority_tickets
;
5550 while (!list_empty(head
) && num_bytes
) {
5551 ticket
= list_first_entry(head
, struct reserve_ticket
,
5553 if (num_bytes
>= ticket
->bytes
) {
5554 trace_btrfs_space_reservation(fs_info
, "space_info",
5557 list_del_init(&ticket
->list
);
5558 num_bytes
-= ticket
->bytes
;
5559 space_info
->bytes_may_use
+= ticket
->bytes
;
5561 space_info
->tickets_id
++;
5562 wake_up(&ticket
->wait
);
5564 trace_btrfs_space_reservation(fs_info
, "space_info",
5567 space_info
->bytes_may_use
+= num_bytes
;
5568 ticket
->bytes
-= num_bytes
;
5573 if (num_bytes
&& head
== &space_info
->priority_tickets
) {
5574 head
= &space_info
->tickets
;
5579 static u64
block_rsv_release_bytes(struct btrfs_fs_info
*fs_info
,
5580 struct btrfs_block_rsv
*block_rsv
,
5581 struct btrfs_block_rsv
*dest
, u64 num_bytes
)
5583 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
5586 spin_lock(&block_rsv
->lock
);
5587 if (num_bytes
== (u64
)-1)
5588 num_bytes
= block_rsv
->size
;
5589 block_rsv
->size
-= num_bytes
;
5590 if (block_rsv
->reserved
>= block_rsv
->size
) {
5591 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
5592 block_rsv
->reserved
= block_rsv
->size
;
5593 block_rsv
->full
= 1;
5597 spin_unlock(&block_rsv
->lock
);
5600 if (num_bytes
> 0) {
5602 spin_lock(&dest
->lock
);
5606 bytes_to_add
= dest
->size
- dest
->reserved
;
5607 bytes_to_add
= min(num_bytes
, bytes_to_add
);
5608 dest
->reserved
+= bytes_to_add
;
5609 if (dest
->reserved
>= dest
->size
)
5611 num_bytes
-= bytes_to_add
;
5613 spin_unlock(&dest
->lock
);
5616 space_info_add_old_bytes(fs_info
, space_info
,
5622 int btrfs_block_rsv_migrate(struct btrfs_block_rsv
*src
,
5623 struct btrfs_block_rsv
*dst
, u64 num_bytes
,
5628 ret
= block_rsv_use_bytes(src
, num_bytes
);
5632 block_rsv_add_bytes(dst
, num_bytes
, update_size
);
5636 void btrfs_init_block_rsv(struct btrfs_block_rsv
*rsv
, unsigned short type
)
5638 memset(rsv
, 0, sizeof(*rsv
));
5639 spin_lock_init(&rsv
->lock
);
5643 void btrfs_init_metadata_block_rsv(struct btrfs_fs_info
*fs_info
,
5644 struct btrfs_block_rsv
*rsv
,
5645 unsigned short type
)
5647 btrfs_init_block_rsv(rsv
, type
);
5648 rsv
->space_info
= __find_space_info(fs_info
,
5649 BTRFS_BLOCK_GROUP_METADATA
);
5652 struct btrfs_block_rsv
*btrfs_alloc_block_rsv(struct btrfs_fs_info
*fs_info
,
5653 unsigned short type
)
5655 struct btrfs_block_rsv
*block_rsv
;
5657 block_rsv
= kmalloc(sizeof(*block_rsv
), GFP_NOFS
);
5661 btrfs_init_metadata_block_rsv(fs_info
, block_rsv
, type
);
5665 void btrfs_free_block_rsv(struct btrfs_fs_info
*fs_info
,
5666 struct btrfs_block_rsv
*rsv
)
5670 btrfs_block_rsv_release(fs_info
, rsv
, (u64
)-1);
5674 void __btrfs_free_block_rsv(struct btrfs_block_rsv
*rsv
)
5679 int btrfs_block_rsv_add(struct btrfs_root
*root
,
5680 struct btrfs_block_rsv
*block_rsv
, u64 num_bytes
,
5681 enum btrfs_reserve_flush_enum flush
)
5688 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
5690 block_rsv_add_bytes(block_rsv
, num_bytes
, 1);
5697 int btrfs_block_rsv_check(struct btrfs_block_rsv
*block_rsv
, int min_factor
)
5705 spin_lock(&block_rsv
->lock
);
5706 num_bytes
= div_factor(block_rsv
->size
, min_factor
);
5707 if (block_rsv
->reserved
>= num_bytes
)
5709 spin_unlock(&block_rsv
->lock
);
5714 int btrfs_block_rsv_refill(struct btrfs_root
*root
,
5715 struct btrfs_block_rsv
*block_rsv
, u64 min_reserved
,
5716 enum btrfs_reserve_flush_enum flush
)
5724 spin_lock(&block_rsv
->lock
);
5725 num_bytes
= min_reserved
;
5726 if (block_rsv
->reserved
>= num_bytes
)
5729 num_bytes
-= block_rsv
->reserved
;
5730 spin_unlock(&block_rsv
->lock
);
5735 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
5737 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
5745 * btrfs_inode_rsv_refill - refill the inode block rsv.
5746 * @inode - the inode we are refilling.
5747 * @flush - the flusing restriction.
5749 * Essentially the same as btrfs_block_rsv_refill, except it uses the
5750 * block_rsv->size as the minimum size. We'll either refill the missing amount
5751 * or return if we already have enough space. This will also handle the resreve
5752 * tracepoint for the reserved amount.
5754 int btrfs_inode_rsv_refill(struct btrfs_inode
*inode
,
5755 enum btrfs_reserve_flush_enum flush
)
5757 struct btrfs_root
*root
= inode
->root
;
5758 struct btrfs_block_rsv
*block_rsv
= &inode
->block_rsv
;
5762 spin_lock(&block_rsv
->lock
);
5763 if (block_rsv
->reserved
< block_rsv
->size
)
5764 num_bytes
= block_rsv
->size
- block_rsv
->reserved
;
5765 spin_unlock(&block_rsv
->lock
);
5770 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
5772 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
5773 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5774 btrfs_ino(inode
), num_bytes
, 1);
5780 * btrfs_inode_rsv_release - release any excessive reservation.
5781 * @inode - the inode we need to release from.
5783 * This is the same as btrfs_block_rsv_release, except that it handles the
5784 * tracepoint for the reservation.
5786 void btrfs_inode_rsv_release(struct btrfs_inode
*inode
)
5788 struct btrfs_fs_info
*fs_info
= inode
->root
->fs_info
;
5789 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5790 struct btrfs_block_rsv
*block_rsv
= &inode
->block_rsv
;
5794 * Since we statically set the block_rsv->size we just want to say we
5795 * are releasing 0 bytes, and then we'll just get the reservation over
5798 released
= block_rsv_release_bytes(fs_info
, block_rsv
, global_rsv
, 0);
5800 trace_btrfs_space_reservation(fs_info
, "delalloc",
5801 btrfs_ino(inode
), released
, 0);
5804 void btrfs_block_rsv_release(struct btrfs_fs_info
*fs_info
,
5805 struct btrfs_block_rsv
*block_rsv
,
5808 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5810 if (global_rsv
== block_rsv
||
5811 block_rsv
->space_info
!= global_rsv
->space_info
)
5813 block_rsv_release_bytes(fs_info
, block_rsv
, global_rsv
, num_bytes
);
5816 static void update_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5818 struct btrfs_block_rsv
*block_rsv
= &fs_info
->global_block_rsv
;
5819 struct btrfs_space_info
*sinfo
= block_rsv
->space_info
;
5823 * The global block rsv is based on the size of the extent tree, the
5824 * checksum tree and the root tree. If the fs is empty we want to set
5825 * it to a minimal amount for safety.
5827 num_bytes
= btrfs_root_used(&fs_info
->extent_root
->root_item
) +
5828 btrfs_root_used(&fs_info
->csum_root
->root_item
) +
5829 btrfs_root_used(&fs_info
->tree_root
->root_item
);
5830 num_bytes
= max_t(u64
, num_bytes
, SZ_16M
);
5832 spin_lock(&sinfo
->lock
);
5833 spin_lock(&block_rsv
->lock
);
5835 block_rsv
->size
= min_t(u64
, num_bytes
, SZ_512M
);
5837 if (block_rsv
->reserved
< block_rsv
->size
) {
5838 num_bytes
= btrfs_space_info_used(sinfo
, true);
5839 if (sinfo
->total_bytes
> num_bytes
) {
5840 num_bytes
= sinfo
->total_bytes
- num_bytes
;
5841 num_bytes
= min(num_bytes
,
5842 block_rsv
->size
- block_rsv
->reserved
);
5843 block_rsv
->reserved
+= num_bytes
;
5844 sinfo
->bytes_may_use
+= num_bytes
;
5845 trace_btrfs_space_reservation(fs_info
, "space_info",
5846 sinfo
->flags
, num_bytes
,
5849 } else if (block_rsv
->reserved
> block_rsv
->size
) {
5850 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
5851 sinfo
->bytes_may_use
-= num_bytes
;
5852 trace_btrfs_space_reservation(fs_info
, "space_info",
5853 sinfo
->flags
, num_bytes
, 0);
5854 block_rsv
->reserved
= block_rsv
->size
;
5857 if (block_rsv
->reserved
== block_rsv
->size
)
5858 block_rsv
->full
= 1;
5860 block_rsv
->full
= 0;
5862 spin_unlock(&block_rsv
->lock
);
5863 spin_unlock(&sinfo
->lock
);
5866 static void init_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5868 struct btrfs_space_info
*space_info
;
5870 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
5871 fs_info
->chunk_block_rsv
.space_info
= space_info
;
5873 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
5874 fs_info
->global_block_rsv
.space_info
= space_info
;
5875 fs_info
->trans_block_rsv
.space_info
= space_info
;
5876 fs_info
->empty_block_rsv
.space_info
= space_info
;
5877 fs_info
->delayed_block_rsv
.space_info
= space_info
;
5879 fs_info
->extent_root
->block_rsv
= &fs_info
->global_block_rsv
;
5880 fs_info
->csum_root
->block_rsv
= &fs_info
->global_block_rsv
;
5881 fs_info
->dev_root
->block_rsv
= &fs_info
->global_block_rsv
;
5882 fs_info
->tree_root
->block_rsv
= &fs_info
->global_block_rsv
;
5883 if (fs_info
->quota_root
)
5884 fs_info
->quota_root
->block_rsv
= &fs_info
->global_block_rsv
;
5885 fs_info
->chunk_root
->block_rsv
= &fs_info
->chunk_block_rsv
;
5887 update_global_block_rsv(fs_info
);
5890 static void release_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5892 block_rsv_release_bytes(fs_info
, &fs_info
->global_block_rsv
, NULL
,
5894 WARN_ON(fs_info
->trans_block_rsv
.size
> 0);
5895 WARN_ON(fs_info
->trans_block_rsv
.reserved
> 0);
5896 WARN_ON(fs_info
->chunk_block_rsv
.size
> 0);
5897 WARN_ON(fs_info
->chunk_block_rsv
.reserved
> 0);
5898 WARN_ON(fs_info
->delayed_block_rsv
.size
> 0);
5899 WARN_ON(fs_info
->delayed_block_rsv
.reserved
> 0);
5902 void btrfs_trans_release_metadata(struct btrfs_trans_handle
*trans
,
5903 struct btrfs_fs_info
*fs_info
)
5905 if (!trans
->block_rsv
) {
5906 ASSERT(!trans
->bytes_reserved
);
5910 if (!trans
->bytes_reserved
)
5913 ASSERT(trans
->block_rsv
== &fs_info
->trans_block_rsv
);
5914 trace_btrfs_space_reservation(fs_info
, "transaction",
5915 trans
->transid
, trans
->bytes_reserved
, 0);
5916 btrfs_block_rsv_release(fs_info
, trans
->block_rsv
,
5917 trans
->bytes_reserved
);
5918 trans
->bytes_reserved
= 0;
5922 * To be called after all the new block groups attached to the transaction
5923 * handle have been created (btrfs_create_pending_block_groups()).
5925 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle
*trans
)
5927 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
5929 if (!trans
->chunk_bytes_reserved
)
5932 WARN_ON_ONCE(!list_empty(&trans
->new_bgs
));
5934 block_rsv_release_bytes(fs_info
, &fs_info
->chunk_block_rsv
, NULL
,
5935 trans
->chunk_bytes_reserved
);
5936 trans
->chunk_bytes_reserved
= 0;
5939 /* Can only return 0 or -ENOSPC */
5940 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle
*trans
,
5941 struct btrfs_inode
*inode
)
5943 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
5944 struct btrfs_root
*root
= inode
->root
;
5946 * We always use trans->block_rsv here as we will have reserved space
5947 * for our orphan when starting the transaction, using get_block_rsv()
5948 * here will sometimes make us choose the wrong block rsv as we could be
5949 * doing a reloc inode for a non refcounted root.
5951 struct btrfs_block_rsv
*src_rsv
= trans
->block_rsv
;
5952 struct btrfs_block_rsv
*dst_rsv
= root
->orphan_block_rsv
;
5955 * We need to hold space in order to delete our orphan item once we've
5956 * added it, so this takes the reservation so we can release it later
5957 * when we are truly done with the orphan item.
5959 u64 num_bytes
= btrfs_calc_trans_metadata_size(fs_info
, 1);
5961 trace_btrfs_space_reservation(fs_info
, "orphan", btrfs_ino(inode
),
5963 return btrfs_block_rsv_migrate(src_rsv
, dst_rsv
, num_bytes
, 1);
5966 void btrfs_orphan_release_metadata(struct btrfs_inode
*inode
)
5968 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
5969 struct btrfs_root
*root
= inode
->root
;
5970 u64 num_bytes
= btrfs_calc_trans_metadata_size(fs_info
, 1);
5972 trace_btrfs_space_reservation(fs_info
, "orphan", btrfs_ino(inode
),
5974 btrfs_block_rsv_release(fs_info
, root
->orphan_block_rsv
, num_bytes
);
5978 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5979 * root: the root of the parent directory
5980 * rsv: block reservation
5981 * items: the number of items that we need do reservation
5982 * qgroup_reserved: used to return the reserved size in qgroup
5984 * This function is used to reserve the space for snapshot/subvolume
5985 * creation and deletion. Those operations are different with the
5986 * common file/directory operations, they change two fs/file trees
5987 * and root tree, the number of items that the qgroup reserves is
5988 * different with the free space reservation. So we can not use
5989 * the space reservation mechanism in start_transaction().
5991 int btrfs_subvolume_reserve_metadata(struct btrfs_root
*root
,
5992 struct btrfs_block_rsv
*rsv
,
5994 u64
*qgroup_reserved
,
5995 bool use_global_rsv
)
5999 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6000 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
6002 if (test_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
)) {
6003 /* One for parent inode, two for dir entries */
6004 num_bytes
= 3 * fs_info
->nodesize
;
6005 ret
= btrfs_qgroup_reserve_meta(root
, num_bytes
, true);
6012 *qgroup_reserved
= num_bytes
;
6014 num_bytes
= btrfs_calc_trans_metadata_size(fs_info
, items
);
6015 rsv
->space_info
= __find_space_info(fs_info
,
6016 BTRFS_BLOCK_GROUP_METADATA
);
6017 ret
= btrfs_block_rsv_add(root
, rsv
, num_bytes
,
6018 BTRFS_RESERVE_FLUSH_ALL
);
6020 if (ret
== -ENOSPC
&& use_global_rsv
)
6021 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
, num_bytes
, 1);
6023 if (ret
&& *qgroup_reserved
)
6024 btrfs_qgroup_free_meta(root
, *qgroup_reserved
);
6029 void btrfs_subvolume_release_metadata(struct btrfs_fs_info
*fs_info
,
6030 struct btrfs_block_rsv
*rsv
)
6032 btrfs_block_rsv_release(fs_info
, rsv
, (u64
)-1);
6035 static void btrfs_calculate_inode_block_rsv_size(struct btrfs_fs_info
*fs_info
,
6036 struct btrfs_inode
*inode
)
6038 struct btrfs_block_rsv
*block_rsv
= &inode
->block_rsv
;
6039 u64 reserve_size
= 0;
6041 unsigned outstanding_extents
;
6043 lockdep_assert_held(&inode
->lock
);
6044 outstanding_extents
= inode
->outstanding_extents
;
6045 if (outstanding_extents
)
6046 reserve_size
= btrfs_calc_trans_metadata_size(fs_info
,
6047 outstanding_extents
+ 1);
6048 csum_leaves
= btrfs_csum_bytes_to_leaves(fs_info
,
6050 reserve_size
+= btrfs_calc_trans_metadata_size(fs_info
,
6053 spin_lock(&block_rsv
->lock
);
6054 block_rsv
->size
= reserve_size
;
6055 spin_unlock(&block_rsv
->lock
);
6058 int btrfs_delalloc_reserve_metadata(struct btrfs_inode
*inode
, u64 num_bytes
)
6060 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
6061 struct btrfs_root
*root
= inode
->root
;
6062 unsigned nr_extents
;
6063 enum btrfs_reserve_flush_enum flush
= BTRFS_RESERVE_FLUSH_ALL
;
6065 bool delalloc_lock
= true;
6067 /* If we are a free space inode we need to not flush since we will be in
6068 * the middle of a transaction commit. We also don't need the delalloc
6069 * mutex since we won't race with anybody. We need this mostly to make
6070 * lockdep shut its filthy mouth.
6072 * If we have a transaction open (can happen if we call truncate_block
6073 * from truncate), then we need FLUSH_LIMIT so we don't deadlock.
6075 if (btrfs_is_free_space_inode(inode
)) {
6076 flush
= BTRFS_RESERVE_NO_FLUSH
;
6077 delalloc_lock
= false;
6078 } else if (current
->journal_info
) {
6079 flush
= BTRFS_RESERVE_FLUSH_LIMIT
;
6082 if (flush
!= BTRFS_RESERVE_NO_FLUSH
&&
6083 btrfs_transaction_in_commit(fs_info
))
6084 schedule_timeout(1);
6087 mutex_lock(&inode
->delalloc_mutex
);
6089 num_bytes
= ALIGN(num_bytes
, fs_info
->sectorsize
);
6091 /* Add our new extents and calculate the new rsv size. */
6092 spin_lock(&inode
->lock
);
6093 nr_extents
= count_max_extents(num_bytes
);
6094 btrfs_mod_outstanding_extents(inode
, nr_extents
);
6095 inode
->csum_bytes
+= num_bytes
;
6096 btrfs_calculate_inode_block_rsv_size(fs_info
, inode
);
6097 spin_unlock(&inode
->lock
);
6099 if (test_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
)) {
6100 ret
= btrfs_qgroup_reserve_meta(root
,
6101 nr_extents
* fs_info
->nodesize
, true);
6106 ret
= btrfs_inode_rsv_refill(inode
, flush
);
6107 if (unlikely(ret
)) {
6108 btrfs_qgroup_free_meta(root
,
6109 nr_extents
* fs_info
->nodesize
);
6114 mutex_unlock(&inode
->delalloc_mutex
);
6118 spin_lock(&inode
->lock
);
6119 nr_extents
= count_max_extents(num_bytes
);
6120 btrfs_mod_outstanding_extents(inode
, -nr_extents
);
6121 inode
->csum_bytes
-= num_bytes
;
6122 btrfs_calculate_inode_block_rsv_size(fs_info
, inode
);
6123 spin_unlock(&inode
->lock
);
6125 btrfs_inode_rsv_release(inode
);
6127 mutex_unlock(&inode
->delalloc_mutex
);
6132 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
6133 * @inode: the inode to release the reservation for.
6134 * @num_bytes: the number of bytes we are releasing.
6136 * This will release the metadata reservation for an inode. This can be called
6137 * once we complete IO for a given set of bytes to release their metadata
6138 * reservations, or on error for the same reason.
6140 void btrfs_delalloc_release_metadata(struct btrfs_inode
*inode
, u64 num_bytes
)
6142 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
6144 num_bytes
= ALIGN(num_bytes
, fs_info
->sectorsize
);
6145 spin_lock(&inode
->lock
);
6146 inode
->csum_bytes
-= num_bytes
;
6147 btrfs_calculate_inode_block_rsv_size(fs_info
, inode
);
6148 spin_unlock(&inode
->lock
);
6150 if (btrfs_is_testing(fs_info
))
6153 btrfs_inode_rsv_release(inode
);
6157 * btrfs_delalloc_release_extents - release our outstanding_extents
6158 * @inode: the inode to balance the reservation for.
6159 * @num_bytes: the number of bytes we originally reserved with
6161 * When we reserve space we increase outstanding_extents for the extents we may
6162 * add. Once we've set the range as delalloc or created our ordered extents we
6163 * have outstanding_extents to track the real usage, so we use this to free our
6164 * temporarily tracked outstanding_extents. This _must_ be used in conjunction
6165 * with btrfs_delalloc_reserve_metadata.
6167 void btrfs_delalloc_release_extents(struct btrfs_inode
*inode
, u64 num_bytes
)
6169 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
6170 unsigned num_extents
;
6172 spin_lock(&inode
->lock
);
6173 num_extents
= count_max_extents(num_bytes
);
6174 btrfs_mod_outstanding_extents(inode
, -num_extents
);
6175 btrfs_calculate_inode_block_rsv_size(fs_info
, inode
);
6176 spin_unlock(&inode
->lock
);
6178 if (btrfs_is_testing(fs_info
))
6181 btrfs_inode_rsv_release(inode
);
6185 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6187 * @inode: inode we're writing to
6188 * @start: start range we are writing to
6189 * @len: how long the range we are writing to
6190 * @reserved: mandatory parameter, record actually reserved qgroup ranges of
6191 * current reservation.
6193 * This will do the following things
6195 * o reserve space in data space info for num bytes
6196 * and reserve precious corresponding qgroup space
6197 * (Done in check_data_free_space)
6199 * o reserve space for metadata space, based on the number of outstanding
6200 * extents and how much csums will be needed
6201 * also reserve metadata space in a per root over-reserve method.
6202 * o add to the inodes->delalloc_bytes
6203 * o add it to the fs_info's delalloc inodes list.
6204 * (Above 3 all done in delalloc_reserve_metadata)
6206 * Return 0 for success
6207 * Return <0 for error(-ENOSPC or -EQUOT)
6209 int btrfs_delalloc_reserve_space(struct inode
*inode
,
6210 struct extent_changeset
**reserved
, u64 start
, u64 len
)
6214 ret
= btrfs_check_data_free_space(inode
, reserved
, start
, len
);
6217 ret
= btrfs_delalloc_reserve_metadata(BTRFS_I(inode
), len
);
6219 btrfs_free_reserved_data_space(inode
, *reserved
, start
, len
);
6224 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6225 * @inode: inode we're releasing space for
6226 * @start: start position of the space already reserved
6227 * @len: the len of the space already reserved
6228 * @release_bytes: the len of the space we consumed or didn't use
6230 * This function will release the metadata space that was not used and will
6231 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
6232 * list if there are no delalloc bytes left.
6233 * Also it will handle the qgroup reserved space.
6235 void btrfs_delalloc_release_space(struct inode
*inode
,
6236 struct extent_changeset
*reserved
,
6239 btrfs_delalloc_release_metadata(BTRFS_I(inode
), len
);
6240 btrfs_free_reserved_data_space(inode
, reserved
, start
, len
);
6243 static int update_block_group(struct btrfs_trans_handle
*trans
,
6244 struct btrfs_fs_info
*info
, u64 bytenr
,
6245 u64 num_bytes
, int alloc
)
6247 struct btrfs_block_group_cache
*cache
= NULL
;
6248 u64 total
= num_bytes
;
6253 /* block accounting for super block */
6254 spin_lock(&info
->delalloc_root_lock
);
6255 old_val
= btrfs_super_bytes_used(info
->super_copy
);
6257 old_val
+= num_bytes
;
6259 old_val
-= num_bytes
;
6260 btrfs_set_super_bytes_used(info
->super_copy
, old_val
);
6261 spin_unlock(&info
->delalloc_root_lock
);
6264 cache
= btrfs_lookup_block_group(info
, bytenr
);
6267 if (cache
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
6268 BTRFS_BLOCK_GROUP_RAID1
|
6269 BTRFS_BLOCK_GROUP_RAID10
))
6274 * If this block group has free space cache written out, we
6275 * need to make sure to load it if we are removing space. This
6276 * is because we need the unpinning stage to actually add the
6277 * space back to the block group, otherwise we will leak space.
6279 if (!alloc
&& cache
->cached
== BTRFS_CACHE_NO
)
6280 cache_block_group(cache
, 1);
6282 byte_in_group
= bytenr
- cache
->key
.objectid
;
6283 WARN_ON(byte_in_group
> cache
->key
.offset
);
6285 spin_lock(&cache
->space_info
->lock
);
6286 spin_lock(&cache
->lock
);
6288 if (btrfs_test_opt(info
, SPACE_CACHE
) &&
6289 cache
->disk_cache_state
< BTRFS_DC_CLEAR
)
6290 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
6292 old_val
= btrfs_block_group_used(&cache
->item
);
6293 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
6295 old_val
+= num_bytes
;
6296 btrfs_set_block_group_used(&cache
->item
, old_val
);
6297 cache
->reserved
-= num_bytes
;
6298 cache
->space_info
->bytes_reserved
-= num_bytes
;
6299 cache
->space_info
->bytes_used
+= num_bytes
;
6300 cache
->space_info
->disk_used
+= num_bytes
* factor
;
6301 spin_unlock(&cache
->lock
);
6302 spin_unlock(&cache
->space_info
->lock
);
6304 old_val
-= num_bytes
;
6305 btrfs_set_block_group_used(&cache
->item
, old_val
);
6306 cache
->pinned
+= num_bytes
;
6307 cache
->space_info
->bytes_pinned
+= num_bytes
;
6308 cache
->space_info
->bytes_used
-= num_bytes
;
6309 cache
->space_info
->disk_used
-= num_bytes
* factor
;
6310 spin_unlock(&cache
->lock
);
6311 spin_unlock(&cache
->space_info
->lock
);
6313 trace_btrfs_space_reservation(info
, "pinned",
6314 cache
->space_info
->flags
,
6316 percpu_counter_add(&cache
->space_info
->total_bytes_pinned
,
6318 set_extent_dirty(info
->pinned_extents
,
6319 bytenr
, bytenr
+ num_bytes
- 1,
6320 GFP_NOFS
| __GFP_NOFAIL
);
6323 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
6324 if (list_empty(&cache
->dirty_list
)) {
6325 list_add_tail(&cache
->dirty_list
,
6326 &trans
->transaction
->dirty_bgs
);
6327 trans
->transaction
->num_dirty_bgs
++;
6328 btrfs_get_block_group(cache
);
6330 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
6333 * No longer have used bytes in this block group, queue it for
6334 * deletion. We do this after adding the block group to the
6335 * dirty list to avoid races between cleaner kthread and space
6338 if (!alloc
&& old_val
== 0) {
6339 spin_lock(&info
->unused_bgs_lock
);
6340 if (list_empty(&cache
->bg_list
)) {
6341 btrfs_get_block_group(cache
);
6342 list_add_tail(&cache
->bg_list
,
6345 spin_unlock(&info
->unused_bgs_lock
);
6348 btrfs_put_block_group(cache
);
6350 bytenr
+= num_bytes
;
6355 static u64
first_logical_byte(struct btrfs_fs_info
*fs_info
, u64 search_start
)
6357 struct btrfs_block_group_cache
*cache
;
6360 spin_lock(&fs_info
->block_group_cache_lock
);
6361 bytenr
= fs_info
->first_logical_byte
;
6362 spin_unlock(&fs_info
->block_group_cache_lock
);
6364 if (bytenr
< (u64
)-1)
6367 cache
= btrfs_lookup_first_block_group(fs_info
, search_start
);
6371 bytenr
= cache
->key
.objectid
;
6372 btrfs_put_block_group(cache
);
6377 static int pin_down_extent(struct btrfs_fs_info
*fs_info
,
6378 struct btrfs_block_group_cache
*cache
,
6379 u64 bytenr
, u64 num_bytes
, int reserved
)
6381 spin_lock(&cache
->space_info
->lock
);
6382 spin_lock(&cache
->lock
);
6383 cache
->pinned
+= num_bytes
;
6384 cache
->space_info
->bytes_pinned
+= num_bytes
;
6386 cache
->reserved
-= num_bytes
;
6387 cache
->space_info
->bytes_reserved
-= num_bytes
;
6389 spin_unlock(&cache
->lock
);
6390 spin_unlock(&cache
->space_info
->lock
);
6392 trace_btrfs_space_reservation(fs_info
, "pinned",
6393 cache
->space_info
->flags
, num_bytes
, 1);
6394 percpu_counter_add(&cache
->space_info
->total_bytes_pinned
, num_bytes
);
6395 set_extent_dirty(fs_info
->pinned_extents
, bytenr
,
6396 bytenr
+ num_bytes
- 1, GFP_NOFS
| __GFP_NOFAIL
);
6401 * this function must be called within transaction
6403 int btrfs_pin_extent(struct btrfs_fs_info
*fs_info
,
6404 u64 bytenr
, u64 num_bytes
, int reserved
)
6406 struct btrfs_block_group_cache
*cache
;
6408 cache
= btrfs_lookup_block_group(fs_info
, bytenr
);
6409 BUG_ON(!cache
); /* Logic error */
6411 pin_down_extent(fs_info
, cache
, bytenr
, num_bytes
, reserved
);
6413 btrfs_put_block_group(cache
);
6418 * this function must be called within transaction
6420 int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info
*fs_info
,
6421 u64 bytenr
, u64 num_bytes
)
6423 struct btrfs_block_group_cache
*cache
;
6426 cache
= btrfs_lookup_block_group(fs_info
, bytenr
);
6431 * pull in the free space cache (if any) so that our pin
6432 * removes the free space from the cache. We have load_only set
6433 * to one because the slow code to read in the free extents does check
6434 * the pinned extents.
6436 cache_block_group(cache
, 1);
6438 pin_down_extent(fs_info
, cache
, bytenr
, num_bytes
, 0);
6440 /* remove us from the free space cache (if we're there at all) */
6441 ret
= btrfs_remove_free_space(cache
, bytenr
, num_bytes
);
6442 btrfs_put_block_group(cache
);
6446 static int __exclude_logged_extent(struct btrfs_fs_info
*fs_info
,
6447 u64 start
, u64 num_bytes
)
6450 struct btrfs_block_group_cache
*block_group
;
6451 struct btrfs_caching_control
*caching_ctl
;
6453 block_group
= btrfs_lookup_block_group(fs_info
, start
);
6457 cache_block_group(block_group
, 0);
6458 caching_ctl
= get_caching_control(block_group
);
6462 BUG_ON(!block_group_cache_done(block_group
));
6463 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
6465 mutex_lock(&caching_ctl
->mutex
);
6467 if (start
>= caching_ctl
->progress
) {
6468 ret
= add_excluded_extent(fs_info
, start
, num_bytes
);
6469 } else if (start
+ num_bytes
<= caching_ctl
->progress
) {
6470 ret
= btrfs_remove_free_space(block_group
,
6473 num_bytes
= caching_ctl
->progress
- start
;
6474 ret
= btrfs_remove_free_space(block_group
,
6479 num_bytes
= (start
+ num_bytes
) -
6480 caching_ctl
->progress
;
6481 start
= caching_ctl
->progress
;
6482 ret
= add_excluded_extent(fs_info
, start
, num_bytes
);
6485 mutex_unlock(&caching_ctl
->mutex
);
6486 put_caching_control(caching_ctl
);
6488 btrfs_put_block_group(block_group
);
6492 int btrfs_exclude_logged_extents(struct btrfs_fs_info
*fs_info
,
6493 struct extent_buffer
*eb
)
6495 struct btrfs_file_extent_item
*item
;
6496 struct btrfs_key key
;
6500 if (!btrfs_fs_incompat(fs_info
, MIXED_GROUPS
))
6503 for (i
= 0; i
< btrfs_header_nritems(eb
); i
++) {
6504 btrfs_item_key_to_cpu(eb
, &key
, i
);
6505 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
6507 item
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
6508 found_type
= btrfs_file_extent_type(eb
, item
);
6509 if (found_type
== BTRFS_FILE_EXTENT_INLINE
)
6511 if (btrfs_file_extent_disk_bytenr(eb
, item
) == 0)
6513 key
.objectid
= btrfs_file_extent_disk_bytenr(eb
, item
);
6514 key
.offset
= btrfs_file_extent_disk_num_bytes(eb
, item
);
6515 __exclude_logged_extent(fs_info
, key
.objectid
, key
.offset
);
6522 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache
*bg
)
6524 atomic_inc(&bg
->reservations
);
6527 void btrfs_dec_block_group_reservations(struct btrfs_fs_info
*fs_info
,
6530 struct btrfs_block_group_cache
*bg
;
6532 bg
= btrfs_lookup_block_group(fs_info
, start
);
6534 if (atomic_dec_and_test(&bg
->reservations
))
6535 wake_up_atomic_t(&bg
->reservations
);
6536 btrfs_put_block_group(bg
);
6539 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache
*bg
)
6541 struct btrfs_space_info
*space_info
= bg
->space_info
;
6545 if (!(bg
->flags
& BTRFS_BLOCK_GROUP_DATA
))
6549 * Our block group is read only but before we set it to read only,
6550 * some task might have had allocated an extent from it already, but it
6551 * has not yet created a respective ordered extent (and added it to a
6552 * root's list of ordered extents).
6553 * Therefore wait for any task currently allocating extents, since the
6554 * block group's reservations counter is incremented while a read lock
6555 * on the groups' semaphore is held and decremented after releasing
6556 * the read access on that semaphore and creating the ordered extent.
6558 down_write(&space_info
->groups_sem
);
6559 up_write(&space_info
->groups_sem
);
6561 wait_on_atomic_t(&bg
->reservations
, atomic_t_wait
,
6562 TASK_UNINTERRUPTIBLE
);
6566 * btrfs_add_reserved_bytes - update the block_group and space info counters
6567 * @cache: The cache we are manipulating
6568 * @ram_bytes: The number of bytes of file content, and will be same to
6569 * @num_bytes except for the compress path.
6570 * @num_bytes: The number of bytes in question
6571 * @delalloc: The blocks are allocated for the delalloc write
6573 * This is called by the allocator when it reserves space. If this is a
6574 * reservation and the block group has become read only we cannot make the
6575 * reservation and return -EAGAIN, otherwise this function always succeeds.
6577 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache
*cache
,
6578 u64 ram_bytes
, u64 num_bytes
, int delalloc
)
6580 struct btrfs_space_info
*space_info
= cache
->space_info
;
6583 spin_lock(&space_info
->lock
);
6584 spin_lock(&cache
->lock
);
6588 cache
->reserved
+= num_bytes
;
6589 space_info
->bytes_reserved
+= num_bytes
;
6591 trace_btrfs_space_reservation(cache
->fs_info
,
6592 "space_info", space_info
->flags
,
6594 space_info
->bytes_may_use
-= ram_bytes
;
6596 cache
->delalloc_bytes
+= num_bytes
;
6598 spin_unlock(&cache
->lock
);
6599 spin_unlock(&space_info
->lock
);
6604 * btrfs_free_reserved_bytes - update the block_group and space info counters
6605 * @cache: The cache we are manipulating
6606 * @num_bytes: The number of bytes in question
6607 * @delalloc: The blocks are allocated for the delalloc write
6609 * This is called by somebody who is freeing space that was never actually used
6610 * on disk. For example if you reserve some space for a new leaf in transaction
6611 * A and before transaction A commits you free that leaf, you call this with
6612 * reserve set to 0 in order to clear the reservation.
6615 static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache
*cache
,
6616 u64 num_bytes
, int delalloc
)
6618 struct btrfs_space_info
*space_info
= cache
->space_info
;
6621 spin_lock(&space_info
->lock
);
6622 spin_lock(&cache
->lock
);
6624 space_info
->bytes_readonly
+= num_bytes
;
6625 cache
->reserved
-= num_bytes
;
6626 space_info
->bytes_reserved
-= num_bytes
;
6627 space_info
->max_extent_size
= 0;
6630 cache
->delalloc_bytes
-= num_bytes
;
6631 spin_unlock(&cache
->lock
);
6632 spin_unlock(&space_info
->lock
);
6635 void btrfs_prepare_extent_commit(struct btrfs_fs_info
*fs_info
)
6637 struct btrfs_caching_control
*next
;
6638 struct btrfs_caching_control
*caching_ctl
;
6639 struct btrfs_block_group_cache
*cache
;
6641 down_write(&fs_info
->commit_root_sem
);
6643 list_for_each_entry_safe(caching_ctl
, next
,
6644 &fs_info
->caching_block_groups
, list
) {
6645 cache
= caching_ctl
->block_group
;
6646 if (block_group_cache_done(cache
)) {
6647 cache
->last_byte_to_unpin
= (u64
)-1;
6648 list_del_init(&caching_ctl
->list
);
6649 put_caching_control(caching_ctl
);
6651 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
6655 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
6656 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
6658 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
6660 up_write(&fs_info
->commit_root_sem
);
6662 update_global_block_rsv(fs_info
);
6666 * Returns the free cluster for the given space info and sets empty_cluster to
6667 * what it should be based on the mount options.
6669 static struct btrfs_free_cluster
*
6670 fetch_cluster_info(struct btrfs_fs_info
*fs_info
,
6671 struct btrfs_space_info
*space_info
, u64
*empty_cluster
)
6673 struct btrfs_free_cluster
*ret
= NULL
;
6676 if (btrfs_mixed_space_info(space_info
))
6679 if (space_info
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
6680 ret
= &fs_info
->meta_alloc_cluster
;
6681 if (btrfs_test_opt(fs_info
, SSD
))
6682 *empty_cluster
= SZ_2M
;
6684 *empty_cluster
= SZ_64K
;
6685 } else if ((space_info
->flags
& BTRFS_BLOCK_GROUP_DATA
) &&
6686 btrfs_test_opt(fs_info
, SSD_SPREAD
)) {
6687 *empty_cluster
= SZ_2M
;
6688 ret
= &fs_info
->data_alloc_cluster
;
6694 static int unpin_extent_range(struct btrfs_fs_info
*fs_info
,
6696 const bool return_free_space
)
6698 struct btrfs_block_group_cache
*cache
= NULL
;
6699 struct btrfs_space_info
*space_info
;
6700 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
6701 struct btrfs_free_cluster
*cluster
= NULL
;
6703 u64 total_unpinned
= 0;
6704 u64 empty_cluster
= 0;
6707 while (start
<= end
) {
6710 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
6712 btrfs_put_block_group(cache
);
6714 cache
= btrfs_lookup_block_group(fs_info
, start
);
6715 BUG_ON(!cache
); /* Logic error */
6717 cluster
= fetch_cluster_info(fs_info
,
6720 empty_cluster
<<= 1;
6723 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
6724 len
= min(len
, end
+ 1 - start
);
6726 if (start
< cache
->last_byte_to_unpin
) {
6727 len
= min(len
, cache
->last_byte_to_unpin
- start
);
6728 if (return_free_space
)
6729 btrfs_add_free_space(cache
, start
, len
);
6733 total_unpinned
+= len
;
6734 space_info
= cache
->space_info
;
6737 * If this space cluster has been marked as fragmented and we've
6738 * unpinned enough in this block group to potentially allow a
6739 * cluster to be created inside of it go ahead and clear the
6742 if (cluster
&& cluster
->fragmented
&&
6743 total_unpinned
> empty_cluster
) {
6744 spin_lock(&cluster
->lock
);
6745 cluster
->fragmented
= 0;
6746 spin_unlock(&cluster
->lock
);
6749 spin_lock(&space_info
->lock
);
6750 spin_lock(&cache
->lock
);
6751 cache
->pinned
-= len
;
6752 space_info
->bytes_pinned
-= len
;
6754 trace_btrfs_space_reservation(fs_info
, "pinned",
6755 space_info
->flags
, len
, 0);
6756 space_info
->max_extent_size
= 0;
6757 percpu_counter_add(&space_info
->total_bytes_pinned
, -len
);
6759 space_info
->bytes_readonly
+= len
;
6762 spin_unlock(&cache
->lock
);
6763 if (!readonly
&& return_free_space
&&
6764 global_rsv
->space_info
== space_info
) {
6767 spin_lock(&global_rsv
->lock
);
6768 if (!global_rsv
->full
) {
6769 to_add
= min(len
, global_rsv
->size
-
6770 global_rsv
->reserved
);
6771 global_rsv
->reserved
+= to_add
;
6772 space_info
->bytes_may_use
+= to_add
;
6773 if (global_rsv
->reserved
>= global_rsv
->size
)
6774 global_rsv
->full
= 1;
6775 trace_btrfs_space_reservation(fs_info
,
6781 spin_unlock(&global_rsv
->lock
);
6782 /* Add to any tickets we may have */
6784 space_info_add_new_bytes(fs_info
, space_info
,
6787 spin_unlock(&space_info
->lock
);
6791 btrfs_put_block_group(cache
);
6795 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
6796 struct btrfs_fs_info
*fs_info
)
6798 struct btrfs_block_group_cache
*block_group
, *tmp
;
6799 struct list_head
*deleted_bgs
;
6800 struct extent_io_tree
*unpin
;
6805 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
6806 unpin
= &fs_info
->freed_extents
[1];
6808 unpin
= &fs_info
->freed_extents
[0];
6810 while (!trans
->aborted
) {
6811 struct extent_state
*cached_state
= NULL
;
6813 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
6814 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
6815 EXTENT_DIRTY
, &cached_state
);
6817 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
6821 if (btrfs_test_opt(fs_info
, DISCARD
))
6822 ret
= btrfs_discard_extent(fs_info
, start
,
6823 end
+ 1 - start
, NULL
);
6825 clear_extent_dirty(unpin
, start
, end
, &cached_state
);
6826 unpin_extent_range(fs_info
, start
, end
, true);
6827 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
6828 free_extent_state(cached_state
);
6833 * Transaction is finished. We don't need the lock anymore. We
6834 * do need to clean up the block groups in case of a transaction
6837 deleted_bgs
= &trans
->transaction
->deleted_bgs
;
6838 list_for_each_entry_safe(block_group
, tmp
, deleted_bgs
, bg_list
) {
6842 if (!trans
->aborted
)
6843 ret
= btrfs_discard_extent(fs_info
,
6844 block_group
->key
.objectid
,
6845 block_group
->key
.offset
,
6848 list_del_init(&block_group
->bg_list
);
6849 btrfs_put_block_group_trimming(block_group
);
6850 btrfs_put_block_group(block_group
);
6853 const char *errstr
= btrfs_decode_error(ret
);
6855 "discard failed while removing blockgroup: errno=%d %s",
6863 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
6864 struct btrfs_fs_info
*info
,
6865 struct btrfs_delayed_ref_node
*node
, u64 parent
,
6866 u64 root_objectid
, u64 owner_objectid
,
6867 u64 owner_offset
, int refs_to_drop
,
6868 struct btrfs_delayed_extent_op
*extent_op
)
6870 struct btrfs_key key
;
6871 struct btrfs_path
*path
;
6872 struct btrfs_root
*extent_root
= info
->extent_root
;
6873 struct extent_buffer
*leaf
;
6874 struct btrfs_extent_item
*ei
;
6875 struct btrfs_extent_inline_ref
*iref
;
6878 int extent_slot
= 0;
6879 int found_extent
= 0;
6883 u64 bytenr
= node
->bytenr
;
6884 u64 num_bytes
= node
->num_bytes
;
6886 bool skinny_metadata
= btrfs_fs_incompat(info
, SKINNY_METADATA
);
6888 path
= btrfs_alloc_path();
6892 path
->reada
= READA_FORWARD
;
6893 path
->leave_spinning
= 1;
6895 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
6896 BUG_ON(!is_data
&& refs_to_drop
!= 1);
6899 skinny_metadata
= false;
6901 ret
= lookup_extent_backref(trans
, info
, path
, &iref
,
6902 bytenr
, num_bytes
, parent
,
6903 root_objectid
, owner_objectid
,
6906 extent_slot
= path
->slots
[0];
6907 while (extent_slot
>= 0) {
6908 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
6910 if (key
.objectid
!= bytenr
)
6912 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
6913 key
.offset
== num_bytes
) {
6917 if (key
.type
== BTRFS_METADATA_ITEM_KEY
&&
6918 key
.offset
== owner_objectid
) {
6922 if (path
->slots
[0] - extent_slot
> 5)
6926 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6927 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
6928 if (found_extent
&& item_size
< sizeof(*ei
))
6931 if (!found_extent
) {
6933 ret
= remove_extent_backref(trans
, info
, path
, NULL
,
6935 is_data
, &last_ref
);
6937 btrfs_abort_transaction(trans
, ret
);
6940 btrfs_release_path(path
);
6941 path
->leave_spinning
= 1;
6943 key
.objectid
= bytenr
;
6944 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6945 key
.offset
= num_bytes
;
6947 if (!is_data
&& skinny_metadata
) {
6948 key
.type
= BTRFS_METADATA_ITEM_KEY
;
6949 key
.offset
= owner_objectid
;
6952 ret
= btrfs_search_slot(trans
, extent_root
,
6954 if (ret
> 0 && skinny_metadata
&& path
->slots
[0]) {
6956 * Couldn't find our skinny metadata item,
6957 * see if we have ye olde extent item.
6960 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
6962 if (key
.objectid
== bytenr
&&
6963 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
6964 key
.offset
== num_bytes
)
6968 if (ret
> 0 && skinny_metadata
) {
6969 skinny_metadata
= false;
6970 key
.objectid
= bytenr
;
6971 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6972 key
.offset
= num_bytes
;
6973 btrfs_release_path(path
);
6974 ret
= btrfs_search_slot(trans
, extent_root
,
6980 "umm, got %d back from search, was looking for %llu",
6983 btrfs_print_leaf(path
->nodes
[0]);
6986 btrfs_abort_transaction(trans
, ret
);
6989 extent_slot
= path
->slots
[0];
6991 } else if (WARN_ON(ret
== -ENOENT
)) {
6992 btrfs_print_leaf(path
->nodes
[0]);
6994 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6995 bytenr
, parent
, root_objectid
, owner_objectid
,
6997 btrfs_abort_transaction(trans
, ret
);
7000 btrfs_abort_transaction(trans
, ret
);
7004 leaf
= path
->nodes
[0];
7005 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
7006 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
7007 if (item_size
< sizeof(*ei
)) {
7008 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
7009 ret
= convert_extent_item_v0(trans
, info
, path
, owner_objectid
,
7012 btrfs_abort_transaction(trans
, ret
);
7016 btrfs_release_path(path
);
7017 path
->leave_spinning
= 1;
7019 key
.objectid
= bytenr
;
7020 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
7021 key
.offset
= num_bytes
;
7023 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
7027 "umm, got %d back from search, was looking for %llu",
7029 btrfs_print_leaf(path
->nodes
[0]);
7032 btrfs_abort_transaction(trans
, ret
);
7036 extent_slot
= path
->slots
[0];
7037 leaf
= path
->nodes
[0];
7038 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
7041 BUG_ON(item_size
< sizeof(*ei
));
7042 ei
= btrfs_item_ptr(leaf
, extent_slot
,
7043 struct btrfs_extent_item
);
7044 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
7045 key
.type
== BTRFS_EXTENT_ITEM_KEY
) {
7046 struct btrfs_tree_block_info
*bi
;
7047 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
7048 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
7049 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
7052 refs
= btrfs_extent_refs(leaf
, ei
);
7053 if (refs
< refs_to_drop
) {
7055 "trying to drop %d refs but we only have %Lu for bytenr %Lu",
7056 refs_to_drop
, refs
, bytenr
);
7058 btrfs_abort_transaction(trans
, ret
);
7061 refs
-= refs_to_drop
;
7065 __run_delayed_extent_op(extent_op
, leaf
, ei
);
7067 * In the case of inline back ref, reference count will
7068 * be updated by remove_extent_backref
7071 BUG_ON(!found_extent
);
7073 btrfs_set_extent_refs(leaf
, ei
, refs
);
7074 btrfs_mark_buffer_dirty(leaf
);
7077 ret
= remove_extent_backref(trans
, info
, path
,
7079 is_data
, &last_ref
);
7081 btrfs_abort_transaction(trans
, ret
);
7087 BUG_ON(is_data
&& refs_to_drop
!=
7088 extent_data_ref_count(path
, iref
));
7090 BUG_ON(path
->slots
[0] != extent_slot
);
7092 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
7093 path
->slots
[0] = extent_slot
;
7099 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
7102 btrfs_abort_transaction(trans
, ret
);
7105 btrfs_release_path(path
);
7108 ret
= btrfs_del_csums(trans
, info
, bytenr
, num_bytes
);
7110 btrfs_abort_transaction(trans
, ret
);
7115 ret
= add_to_free_space_tree(trans
, info
, bytenr
, num_bytes
);
7117 btrfs_abort_transaction(trans
, ret
);
7121 ret
= update_block_group(trans
, info
, bytenr
, num_bytes
, 0);
7123 btrfs_abort_transaction(trans
, ret
);
7127 btrfs_release_path(path
);
7130 btrfs_free_path(path
);
7135 * when we free an block, it is possible (and likely) that we free the last
7136 * delayed ref for that extent as well. This searches the delayed ref tree for
7137 * a given extent, and if there are no other delayed refs to be processed, it
7138 * removes it from the tree.
7140 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
7143 struct btrfs_delayed_ref_head
*head
;
7144 struct btrfs_delayed_ref_root
*delayed_refs
;
7147 delayed_refs
= &trans
->transaction
->delayed_refs
;
7148 spin_lock(&delayed_refs
->lock
);
7149 head
= btrfs_find_delayed_ref_head(delayed_refs
, bytenr
);
7151 goto out_delayed_unlock
;
7153 spin_lock(&head
->lock
);
7154 if (!RB_EMPTY_ROOT(&head
->ref_tree
))
7157 if (head
->extent_op
) {
7158 if (!head
->must_insert_reserved
)
7160 btrfs_free_delayed_extent_op(head
->extent_op
);
7161 head
->extent_op
= NULL
;
7165 * waiting for the lock here would deadlock. If someone else has it
7166 * locked they are already in the process of dropping it anyway
7168 if (!mutex_trylock(&head
->mutex
))
7172 * at this point we have a head with no other entries. Go
7173 * ahead and process it.
7175 rb_erase(&head
->href_node
, &delayed_refs
->href_root
);
7176 RB_CLEAR_NODE(&head
->href_node
);
7177 atomic_dec(&delayed_refs
->num_entries
);
7180 * we don't take a ref on the node because we're removing it from the
7181 * tree, so we just steal the ref the tree was holding.
7183 delayed_refs
->num_heads
--;
7184 if (head
->processing
== 0)
7185 delayed_refs
->num_heads_ready
--;
7186 head
->processing
= 0;
7187 spin_unlock(&head
->lock
);
7188 spin_unlock(&delayed_refs
->lock
);
7190 BUG_ON(head
->extent_op
);
7191 if (head
->must_insert_reserved
)
7194 mutex_unlock(&head
->mutex
);
7195 btrfs_put_delayed_ref_head(head
);
7198 spin_unlock(&head
->lock
);
7201 spin_unlock(&delayed_refs
->lock
);
7205 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
7206 struct btrfs_root
*root
,
7207 struct extent_buffer
*buf
,
7208 u64 parent
, int last_ref
)
7210 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7214 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
7215 int old_ref_mod
, new_ref_mod
;
7217 btrfs_ref_tree_mod(root
, buf
->start
, buf
->len
, parent
,
7218 root
->root_key
.objectid
,
7219 btrfs_header_level(buf
), 0,
7220 BTRFS_DROP_DELAYED_REF
);
7221 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, buf
->start
,
7223 root
->root_key
.objectid
,
7224 btrfs_header_level(buf
),
7225 BTRFS_DROP_DELAYED_REF
, NULL
,
7226 &old_ref_mod
, &new_ref_mod
);
7227 BUG_ON(ret
); /* -ENOMEM */
7228 pin
= old_ref_mod
>= 0 && new_ref_mod
< 0;
7231 if (last_ref
&& btrfs_header_generation(buf
) == trans
->transid
) {
7232 struct btrfs_block_group_cache
*cache
;
7234 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
7235 ret
= check_ref_cleanup(trans
, buf
->start
);
7241 cache
= btrfs_lookup_block_group(fs_info
, buf
->start
);
7243 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
7244 pin_down_extent(fs_info
, cache
, buf
->start
,
7246 btrfs_put_block_group(cache
);
7250 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
7252 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
7253 btrfs_free_reserved_bytes(cache
, buf
->len
, 0);
7254 btrfs_put_block_group(cache
);
7255 trace_btrfs_reserved_extent_free(fs_info
, buf
->start
, buf
->len
);
7259 add_pinned_bytes(fs_info
, buf
->len
, btrfs_header_level(buf
),
7260 root
->root_key
.objectid
);
7264 * Deleting the buffer, clear the corrupt flag since it doesn't
7267 clear_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
);
7271 /* Can return -ENOMEM */
7272 int btrfs_free_extent(struct btrfs_trans_handle
*trans
,
7273 struct btrfs_root
*root
,
7274 u64 bytenr
, u64 num_bytes
, u64 parent
, u64 root_objectid
,
7275 u64 owner
, u64 offset
)
7277 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7278 int old_ref_mod
, new_ref_mod
;
7281 if (btrfs_is_testing(fs_info
))
7284 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
)
7285 btrfs_ref_tree_mod(root
, bytenr
, num_bytes
, parent
,
7286 root_objectid
, owner
, offset
,
7287 BTRFS_DROP_DELAYED_REF
);
7290 * tree log blocks never actually go into the extent allocation
7291 * tree, just update pinning info and exit early.
7293 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
7294 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
7295 /* unlocks the pinned mutex */
7296 btrfs_pin_extent(fs_info
, bytenr
, num_bytes
, 1);
7297 old_ref_mod
= new_ref_mod
= 0;
7299 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
7300 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
7302 root_objectid
, (int)owner
,
7303 BTRFS_DROP_DELAYED_REF
, NULL
,
7304 &old_ref_mod
, &new_ref_mod
);
7306 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
7308 root_objectid
, owner
, offset
,
7309 0, BTRFS_DROP_DELAYED_REF
,
7310 &old_ref_mod
, &new_ref_mod
);
7313 if (ret
== 0 && old_ref_mod
>= 0 && new_ref_mod
< 0)
7314 add_pinned_bytes(fs_info
, num_bytes
, owner
, root_objectid
);
7320 * when we wait for progress in the block group caching, its because
7321 * our allocation attempt failed at least once. So, we must sleep
7322 * and let some progress happen before we try again.
7324 * This function will sleep at least once waiting for new free space to
7325 * show up, and then it will check the block group free space numbers
7326 * for our min num_bytes. Another option is to have it go ahead
7327 * and look in the rbtree for a free extent of a given size, but this
7330 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7331 * any of the information in this block group.
7333 static noinline
void
7334 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
7337 struct btrfs_caching_control
*caching_ctl
;
7339 caching_ctl
= get_caching_control(cache
);
7343 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
7344 (cache
->free_space_ctl
->free_space
>= num_bytes
));
7346 put_caching_control(caching_ctl
);
7350 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
7352 struct btrfs_caching_control
*caching_ctl
;
7355 caching_ctl
= get_caching_control(cache
);
7357 return (cache
->cached
== BTRFS_CACHE_ERROR
) ? -EIO
: 0;
7359 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
7360 if (cache
->cached
== BTRFS_CACHE_ERROR
)
7362 put_caching_control(caching_ctl
);
7366 int __get_raid_index(u64 flags
)
7368 if (flags
& BTRFS_BLOCK_GROUP_RAID10
)
7369 return BTRFS_RAID_RAID10
;
7370 else if (flags
& BTRFS_BLOCK_GROUP_RAID1
)
7371 return BTRFS_RAID_RAID1
;
7372 else if (flags
& BTRFS_BLOCK_GROUP_DUP
)
7373 return BTRFS_RAID_DUP
;
7374 else if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
7375 return BTRFS_RAID_RAID0
;
7376 else if (flags
& BTRFS_BLOCK_GROUP_RAID5
)
7377 return BTRFS_RAID_RAID5
;
7378 else if (flags
& BTRFS_BLOCK_GROUP_RAID6
)
7379 return BTRFS_RAID_RAID6
;
7381 return BTRFS_RAID_SINGLE
; /* BTRFS_BLOCK_GROUP_SINGLE */
7384 int get_block_group_index(struct btrfs_block_group_cache
*cache
)
7386 return __get_raid_index(cache
->flags
);
7389 static const char *btrfs_raid_type_names
[BTRFS_NR_RAID_TYPES
] = {
7390 [BTRFS_RAID_RAID10
] = "raid10",
7391 [BTRFS_RAID_RAID1
] = "raid1",
7392 [BTRFS_RAID_DUP
] = "dup",
7393 [BTRFS_RAID_RAID0
] = "raid0",
7394 [BTRFS_RAID_SINGLE
] = "single",
7395 [BTRFS_RAID_RAID5
] = "raid5",
7396 [BTRFS_RAID_RAID6
] = "raid6",
7399 static const char *get_raid_name(enum btrfs_raid_types type
)
7401 if (type
>= BTRFS_NR_RAID_TYPES
)
7404 return btrfs_raid_type_names
[type
];
7407 enum btrfs_loop_type
{
7408 LOOP_CACHING_NOWAIT
= 0,
7409 LOOP_CACHING_WAIT
= 1,
7410 LOOP_ALLOC_CHUNK
= 2,
7411 LOOP_NO_EMPTY_SIZE
= 3,
7415 btrfs_lock_block_group(struct btrfs_block_group_cache
*cache
,
7419 down_read(&cache
->data_rwsem
);
7423 btrfs_grab_block_group(struct btrfs_block_group_cache
*cache
,
7426 btrfs_get_block_group(cache
);
7428 down_read(&cache
->data_rwsem
);
7431 static struct btrfs_block_group_cache
*
7432 btrfs_lock_cluster(struct btrfs_block_group_cache
*block_group
,
7433 struct btrfs_free_cluster
*cluster
,
7436 struct btrfs_block_group_cache
*used_bg
= NULL
;
7438 spin_lock(&cluster
->refill_lock
);
7440 used_bg
= cluster
->block_group
;
7444 if (used_bg
== block_group
)
7447 btrfs_get_block_group(used_bg
);
7452 if (down_read_trylock(&used_bg
->data_rwsem
))
7455 spin_unlock(&cluster
->refill_lock
);
7457 /* We should only have one-level nested. */
7458 down_read_nested(&used_bg
->data_rwsem
, SINGLE_DEPTH_NESTING
);
7460 spin_lock(&cluster
->refill_lock
);
7461 if (used_bg
== cluster
->block_group
)
7464 up_read(&used_bg
->data_rwsem
);
7465 btrfs_put_block_group(used_bg
);
7470 btrfs_release_block_group(struct btrfs_block_group_cache
*cache
,
7474 up_read(&cache
->data_rwsem
);
7475 btrfs_put_block_group(cache
);
7479 * walks the btree of allocated extents and find a hole of a given size.
7480 * The key ins is changed to record the hole:
7481 * ins->objectid == start position
7482 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7483 * ins->offset == the size of the hole.
7484 * Any available blocks before search_start are skipped.
7486 * If there is no suitable free space, we will record the max size of
7487 * the free space extent currently.
7489 static noinline
int find_free_extent(struct btrfs_fs_info
*fs_info
,
7490 u64 ram_bytes
, u64 num_bytes
, u64 empty_size
,
7491 u64 hint_byte
, struct btrfs_key
*ins
,
7492 u64 flags
, int delalloc
)
7495 struct btrfs_root
*root
= fs_info
->extent_root
;
7496 struct btrfs_free_cluster
*last_ptr
= NULL
;
7497 struct btrfs_block_group_cache
*block_group
= NULL
;
7498 u64 search_start
= 0;
7499 u64 max_extent_size
= 0;
7500 u64 max_free_space
= 0;
7501 u64 empty_cluster
= 0;
7502 struct btrfs_space_info
*space_info
;
7504 int index
= __get_raid_index(flags
);
7505 bool failed_cluster_refill
= false;
7506 bool failed_alloc
= false;
7507 bool use_cluster
= true;
7508 bool have_caching_bg
= false;
7509 bool orig_have_caching_bg
= false;
7510 bool full_search
= false;
7512 WARN_ON(num_bytes
< fs_info
->sectorsize
);
7513 ins
->type
= BTRFS_EXTENT_ITEM_KEY
;
7517 trace_find_free_extent(fs_info
, num_bytes
, empty_size
, flags
);
7519 space_info
= __find_space_info(fs_info
, flags
);
7521 btrfs_err(fs_info
, "No space info for %llu", flags
);
7526 * If our free space is heavily fragmented we may not be able to make
7527 * big contiguous allocations, so instead of doing the expensive search
7528 * for free space, simply return ENOSPC with our max_extent_size so we
7529 * can go ahead and search for a more manageable chunk.
7531 * If our max_extent_size is large enough for our allocation simply
7532 * disable clustering since we will likely not be able to find enough
7533 * space to create a cluster and induce latency trying.
7535 if (unlikely(space_info
->max_extent_size
)) {
7536 spin_lock(&space_info
->lock
);
7537 if (space_info
->max_extent_size
&&
7538 num_bytes
> space_info
->max_extent_size
) {
7539 ins
->offset
= space_info
->max_extent_size
;
7540 spin_unlock(&space_info
->lock
);
7542 } else if (space_info
->max_extent_size
) {
7543 use_cluster
= false;
7545 spin_unlock(&space_info
->lock
);
7548 last_ptr
= fetch_cluster_info(fs_info
, space_info
, &empty_cluster
);
7550 spin_lock(&last_ptr
->lock
);
7551 if (last_ptr
->block_group
)
7552 hint_byte
= last_ptr
->window_start
;
7553 if (last_ptr
->fragmented
) {
7555 * We still set window_start so we can keep track of the
7556 * last place we found an allocation to try and save
7559 hint_byte
= last_ptr
->window_start
;
7560 use_cluster
= false;
7562 spin_unlock(&last_ptr
->lock
);
7565 search_start
= max(search_start
, first_logical_byte(fs_info
, 0));
7566 search_start
= max(search_start
, hint_byte
);
7567 if (search_start
== hint_byte
) {
7568 block_group
= btrfs_lookup_block_group(fs_info
, search_start
);
7570 * we don't want to use the block group if it doesn't match our
7571 * allocation bits, or if its not cached.
7573 * However if we are re-searching with an ideal block group
7574 * picked out then we don't care that the block group is cached.
7576 if (block_group
&& block_group_bits(block_group
, flags
) &&
7577 block_group
->cached
!= BTRFS_CACHE_NO
) {
7578 down_read(&space_info
->groups_sem
);
7579 if (list_empty(&block_group
->list
) ||
7582 * someone is removing this block group,
7583 * we can't jump into the have_block_group
7584 * target because our list pointers are not
7587 btrfs_put_block_group(block_group
);
7588 up_read(&space_info
->groups_sem
);
7590 index
= get_block_group_index(block_group
);
7591 btrfs_lock_block_group(block_group
, delalloc
);
7592 goto have_block_group
;
7594 } else if (block_group
) {
7595 btrfs_put_block_group(block_group
);
7599 have_caching_bg
= false;
7600 if (index
== 0 || index
== __get_raid_index(flags
))
7602 down_read(&space_info
->groups_sem
);
7603 list_for_each_entry(block_group
, &space_info
->block_groups
[index
],
7608 /* If the block group is read-only, we can skip it entirely. */
7609 if (unlikely(block_group
->ro
))
7612 btrfs_grab_block_group(block_group
, delalloc
);
7613 search_start
= block_group
->key
.objectid
;
7616 * this can happen if we end up cycling through all the
7617 * raid types, but we want to make sure we only allocate
7618 * for the proper type.
7620 if (!block_group_bits(block_group
, flags
)) {
7621 u64 extra
= BTRFS_BLOCK_GROUP_DUP
|
7622 BTRFS_BLOCK_GROUP_RAID1
|
7623 BTRFS_BLOCK_GROUP_RAID5
|
7624 BTRFS_BLOCK_GROUP_RAID6
|
7625 BTRFS_BLOCK_GROUP_RAID10
;
7628 * if they asked for extra copies and this block group
7629 * doesn't provide them, bail. This does allow us to
7630 * fill raid0 from raid1.
7632 if ((flags
& extra
) && !(block_group
->flags
& extra
))
7636 * This block group has different flags than we want.
7637 * It's possible that we have MIXED_GROUP flag but no
7638 * block group is mixed. Just skip such block group.
7640 btrfs_release_block_group(block_group
, delalloc
);
7645 cached
= block_group_cache_done(block_group
);
7646 if (unlikely(!cached
)) {
7647 have_caching_bg
= true;
7648 ret
= cache_block_group(block_group
, 0);
7653 if (unlikely(block_group
->cached
== BTRFS_CACHE_ERROR
))
7657 * Ok we want to try and use the cluster allocator, so
7660 if (last_ptr
&& use_cluster
) {
7661 struct btrfs_block_group_cache
*used_block_group
;
7662 unsigned long aligned_cluster
;
7664 * the refill lock keeps out other
7665 * people trying to start a new cluster
7667 used_block_group
= btrfs_lock_cluster(block_group
,
7670 if (!used_block_group
)
7671 goto refill_cluster
;
7673 if (used_block_group
!= block_group
&&
7674 (used_block_group
->ro
||
7675 !block_group_bits(used_block_group
, flags
)))
7676 goto release_cluster
;
7678 offset
= btrfs_alloc_from_cluster(used_block_group
,
7681 used_block_group
->key
.objectid
,
7684 /* we have a block, we're done */
7685 spin_unlock(&last_ptr
->refill_lock
);
7686 trace_btrfs_reserve_extent_cluster(fs_info
,
7688 search_start
, num_bytes
);
7689 if (used_block_group
!= block_group
) {
7690 btrfs_release_block_group(block_group
,
7692 block_group
= used_block_group
;
7697 WARN_ON(last_ptr
->block_group
!= used_block_group
);
7699 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7700 * set up a new clusters, so lets just skip it
7701 * and let the allocator find whatever block
7702 * it can find. If we reach this point, we
7703 * will have tried the cluster allocator
7704 * plenty of times and not have found
7705 * anything, so we are likely way too
7706 * fragmented for the clustering stuff to find
7709 * However, if the cluster is taken from the
7710 * current block group, release the cluster
7711 * first, so that we stand a better chance of
7712 * succeeding in the unclustered
7714 if (loop
>= LOOP_NO_EMPTY_SIZE
&&
7715 used_block_group
!= block_group
) {
7716 spin_unlock(&last_ptr
->refill_lock
);
7717 btrfs_release_block_group(used_block_group
,
7719 goto unclustered_alloc
;
7723 * this cluster didn't work out, free it and
7726 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
7728 if (used_block_group
!= block_group
)
7729 btrfs_release_block_group(used_block_group
,
7732 if (loop
>= LOOP_NO_EMPTY_SIZE
) {
7733 spin_unlock(&last_ptr
->refill_lock
);
7734 goto unclustered_alloc
;
7737 aligned_cluster
= max_t(unsigned long,
7738 empty_cluster
+ empty_size
,
7739 block_group
->full_stripe_len
);
7741 /* allocate a cluster in this block group */
7742 ret
= btrfs_find_space_cluster(fs_info
, block_group
,
7743 last_ptr
, search_start
,
7748 * now pull our allocation out of this
7751 offset
= btrfs_alloc_from_cluster(block_group
,
7757 /* we found one, proceed */
7758 spin_unlock(&last_ptr
->refill_lock
);
7759 trace_btrfs_reserve_extent_cluster(fs_info
,
7760 block_group
, search_start
,
7764 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
7765 && !failed_cluster_refill
) {
7766 spin_unlock(&last_ptr
->refill_lock
);
7768 failed_cluster_refill
= true;
7769 wait_block_group_cache_progress(block_group
,
7770 num_bytes
+ empty_cluster
+ empty_size
);
7771 goto have_block_group
;
7775 * at this point we either didn't find a cluster
7776 * or we weren't able to allocate a block from our
7777 * cluster. Free the cluster we've been trying
7778 * to use, and go to the next block group
7780 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
7781 spin_unlock(&last_ptr
->refill_lock
);
7787 * We are doing an unclustered alloc, set the fragmented flag so
7788 * we don't bother trying to setup a cluster again until we get
7791 if (unlikely(last_ptr
)) {
7792 spin_lock(&last_ptr
->lock
);
7793 last_ptr
->fragmented
= 1;
7794 spin_unlock(&last_ptr
->lock
);
7797 struct btrfs_free_space_ctl
*ctl
=
7798 block_group
->free_space_ctl
;
7800 spin_lock(&ctl
->tree_lock
);
7801 if (ctl
->free_space
<
7802 num_bytes
+ empty_cluster
+ empty_size
) {
7803 max_free_space
= max(max_free_space
,
7805 spin_unlock(&ctl
->tree_lock
);
7808 spin_unlock(&ctl
->tree_lock
);
7811 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
7812 num_bytes
, empty_size
,
7815 * If we didn't find a chunk, and we haven't failed on this
7816 * block group before, and this block group is in the middle of
7817 * caching and we are ok with waiting, then go ahead and wait
7818 * for progress to be made, and set failed_alloc to true.
7820 * If failed_alloc is true then we've already waited on this
7821 * block group once and should move on to the next block group.
7823 if (!offset
&& !failed_alloc
&& !cached
&&
7824 loop
> LOOP_CACHING_NOWAIT
) {
7825 wait_block_group_cache_progress(block_group
,
7826 num_bytes
+ empty_size
);
7827 failed_alloc
= true;
7828 goto have_block_group
;
7829 } else if (!offset
) {
7833 search_start
= ALIGN(offset
, fs_info
->stripesize
);
7835 /* move on to the next group */
7836 if (search_start
+ num_bytes
>
7837 block_group
->key
.objectid
+ block_group
->key
.offset
) {
7838 btrfs_add_free_space(block_group
, offset
, num_bytes
);
7842 if (offset
< search_start
)
7843 btrfs_add_free_space(block_group
, offset
,
7844 search_start
- offset
);
7845 BUG_ON(offset
> search_start
);
7847 ret
= btrfs_add_reserved_bytes(block_group
, ram_bytes
,
7848 num_bytes
, delalloc
);
7849 if (ret
== -EAGAIN
) {
7850 btrfs_add_free_space(block_group
, offset
, num_bytes
);
7853 btrfs_inc_block_group_reservations(block_group
);
7855 /* we are all good, lets return */
7856 ins
->objectid
= search_start
;
7857 ins
->offset
= num_bytes
;
7859 trace_btrfs_reserve_extent(fs_info
, block_group
,
7860 search_start
, num_bytes
);
7861 btrfs_release_block_group(block_group
, delalloc
);
7864 failed_cluster_refill
= false;
7865 failed_alloc
= false;
7866 BUG_ON(index
!= get_block_group_index(block_group
));
7867 btrfs_release_block_group(block_group
, delalloc
);
7870 up_read(&space_info
->groups_sem
);
7872 if ((loop
== LOOP_CACHING_NOWAIT
) && have_caching_bg
7873 && !orig_have_caching_bg
)
7874 orig_have_caching_bg
= true;
7876 if (!ins
->objectid
&& loop
>= LOOP_CACHING_WAIT
&& have_caching_bg
)
7879 if (!ins
->objectid
&& ++index
< BTRFS_NR_RAID_TYPES
)
7883 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7884 * caching kthreads as we move along
7885 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7886 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7887 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7890 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
) {
7892 if (loop
== LOOP_CACHING_NOWAIT
) {
7894 * We want to skip the LOOP_CACHING_WAIT step if we
7895 * don't have any uncached bgs and we've already done a
7896 * full search through.
7898 if (orig_have_caching_bg
|| !full_search
)
7899 loop
= LOOP_CACHING_WAIT
;
7901 loop
= LOOP_ALLOC_CHUNK
;
7906 if (loop
== LOOP_ALLOC_CHUNK
) {
7907 struct btrfs_trans_handle
*trans
;
7910 trans
= current
->journal_info
;
7914 trans
= btrfs_join_transaction(root
);
7916 if (IS_ERR(trans
)) {
7917 ret
= PTR_ERR(trans
);
7921 ret
= do_chunk_alloc(trans
, fs_info
, flags
,
7925 * If we can't allocate a new chunk we've already looped
7926 * through at least once, move on to the NO_EMPTY_SIZE
7930 loop
= LOOP_NO_EMPTY_SIZE
;
7933 * Do not bail out on ENOSPC since we
7934 * can do more things.
7936 if (ret
< 0 && ret
!= -ENOSPC
)
7937 btrfs_abort_transaction(trans
, ret
);
7941 btrfs_end_transaction(trans
);
7946 if (loop
== LOOP_NO_EMPTY_SIZE
) {
7948 * Don't loop again if we already have no empty_size and
7951 if (empty_size
== 0 &&
7952 empty_cluster
== 0) {
7961 } else if (!ins
->objectid
) {
7963 } else if (ins
->objectid
) {
7964 if (!use_cluster
&& last_ptr
) {
7965 spin_lock(&last_ptr
->lock
);
7966 last_ptr
->window_start
= ins
->objectid
;
7967 spin_unlock(&last_ptr
->lock
);
7972 if (ret
== -ENOSPC
) {
7973 if (!max_extent_size
)
7974 max_extent_size
= max_free_space
;
7975 spin_lock(&space_info
->lock
);
7976 space_info
->max_extent_size
= max_extent_size
;
7977 spin_unlock(&space_info
->lock
);
7978 ins
->offset
= max_extent_size
;
7983 static void dump_space_info(struct btrfs_fs_info
*fs_info
,
7984 struct btrfs_space_info
*info
, u64 bytes
,
7985 int dump_block_groups
)
7987 struct btrfs_block_group_cache
*cache
;
7990 spin_lock(&info
->lock
);
7991 btrfs_info(fs_info
, "space_info %llu has %llu free, is %sfull",
7993 info
->total_bytes
- btrfs_space_info_used(info
, true),
7994 info
->full
? "" : "not ");
7996 "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu",
7997 info
->total_bytes
, info
->bytes_used
, info
->bytes_pinned
,
7998 info
->bytes_reserved
, info
->bytes_may_use
,
7999 info
->bytes_readonly
);
8000 spin_unlock(&info
->lock
);
8002 if (!dump_block_groups
)
8005 down_read(&info
->groups_sem
);
8007 list_for_each_entry(cache
, &info
->block_groups
[index
], list
) {
8008 spin_lock(&cache
->lock
);
8010 "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s",
8011 cache
->key
.objectid
, cache
->key
.offset
,
8012 btrfs_block_group_used(&cache
->item
), cache
->pinned
,
8013 cache
->reserved
, cache
->ro
? "[readonly]" : "");
8014 btrfs_dump_free_space(cache
, bytes
);
8015 spin_unlock(&cache
->lock
);
8017 if (++index
< BTRFS_NR_RAID_TYPES
)
8019 up_read(&info
->groups_sem
);
8022 int btrfs_reserve_extent(struct btrfs_root
*root
, u64 ram_bytes
,
8023 u64 num_bytes
, u64 min_alloc_size
,
8024 u64 empty_size
, u64 hint_byte
,
8025 struct btrfs_key
*ins
, int is_data
, int delalloc
)
8027 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8028 bool final_tried
= num_bytes
== min_alloc_size
;
8032 flags
= get_alloc_profile_by_root(root
, is_data
);
8034 WARN_ON(num_bytes
< fs_info
->sectorsize
);
8035 ret
= find_free_extent(fs_info
, ram_bytes
, num_bytes
, empty_size
,
8036 hint_byte
, ins
, flags
, delalloc
);
8037 if (!ret
&& !is_data
) {
8038 btrfs_dec_block_group_reservations(fs_info
, ins
->objectid
);
8039 } else if (ret
== -ENOSPC
) {
8040 if (!final_tried
&& ins
->offset
) {
8041 num_bytes
= min(num_bytes
>> 1, ins
->offset
);
8042 num_bytes
= round_down(num_bytes
,
8043 fs_info
->sectorsize
);
8044 num_bytes
= max(num_bytes
, min_alloc_size
);
8045 ram_bytes
= num_bytes
;
8046 if (num_bytes
== min_alloc_size
)
8049 } else if (btrfs_test_opt(fs_info
, ENOSPC_DEBUG
)) {
8050 struct btrfs_space_info
*sinfo
;
8052 sinfo
= __find_space_info(fs_info
, flags
);
8054 "allocation failed flags %llu, wanted %llu",
8057 dump_space_info(fs_info
, sinfo
, num_bytes
, 1);
8064 static int __btrfs_free_reserved_extent(struct btrfs_fs_info
*fs_info
,
8066 int pin
, int delalloc
)
8068 struct btrfs_block_group_cache
*cache
;
8071 cache
= btrfs_lookup_block_group(fs_info
, start
);
8073 btrfs_err(fs_info
, "Unable to find block group for %llu",
8079 pin_down_extent(fs_info
, cache
, start
, len
, 1);
8081 if (btrfs_test_opt(fs_info
, DISCARD
))
8082 ret
= btrfs_discard_extent(fs_info
, start
, len
, NULL
);
8083 btrfs_add_free_space(cache
, start
, len
);
8084 btrfs_free_reserved_bytes(cache
, len
, delalloc
);
8085 trace_btrfs_reserved_extent_free(fs_info
, start
, len
);
8088 btrfs_put_block_group(cache
);
8092 int btrfs_free_reserved_extent(struct btrfs_fs_info
*fs_info
,
8093 u64 start
, u64 len
, int delalloc
)
8095 return __btrfs_free_reserved_extent(fs_info
, start
, len
, 0, delalloc
);
8098 int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info
*fs_info
,
8101 return __btrfs_free_reserved_extent(fs_info
, start
, len
, 1, 0);
8104 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
8105 struct btrfs_fs_info
*fs_info
,
8106 u64 parent
, u64 root_objectid
,
8107 u64 flags
, u64 owner
, u64 offset
,
8108 struct btrfs_key
*ins
, int ref_mod
)
8111 struct btrfs_extent_item
*extent_item
;
8112 struct btrfs_extent_inline_ref
*iref
;
8113 struct btrfs_path
*path
;
8114 struct extent_buffer
*leaf
;
8119 type
= BTRFS_SHARED_DATA_REF_KEY
;
8121 type
= BTRFS_EXTENT_DATA_REF_KEY
;
8123 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
8125 path
= btrfs_alloc_path();
8129 path
->leave_spinning
= 1;
8130 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
8133 btrfs_free_path(path
);
8137 leaf
= path
->nodes
[0];
8138 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
8139 struct btrfs_extent_item
);
8140 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
8141 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
8142 btrfs_set_extent_flags(leaf
, extent_item
,
8143 flags
| BTRFS_EXTENT_FLAG_DATA
);
8145 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
8146 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
8148 struct btrfs_shared_data_ref
*ref
;
8149 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
8150 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
8151 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
8153 struct btrfs_extent_data_ref
*ref
;
8154 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
8155 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
8156 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
8157 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
8158 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
8161 btrfs_mark_buffer_dirty(path
->nodes
[0]);
8162 btrfs_free_path(path
);
8164 ret
= remove_from_free_space_tree(trans
, fs_info
, ins
->objectid
,
8169 ret
= update_block_group(trans
, fs_info
, ins
->objectid
, ins
->offset
, 1);
8170 if (ret
) { /* -ENOENT, logic error */
8171 btrfs_err(fs_info
, "update block group failed for %llu %llu",
8172 ins
->objectid
, ins
->offset
);
8175 trace_btrfs_reserved_extent_alloc(fs_info
, ins
->objectid
, ins
->offset
);
8179 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
8180 struct btrfs_fs_info
*fs_info
,
8181 u64 parent
, u64 root_objectid
,
8182 u64 flags
, struct btrfs_disk_key
*key
,
8183 int level
, struct btrfs_key
*ins
)
8186 struct btrfs_extent_item
*extent_item
;
8187 struct btrfs_tree_block_info
*block_info
;
8188 struct btrfs_extent_inline_ref
*iref
;
8189 struct btrfs_path
*path
;
8190 struct extent_buffer
*leaf
;
8191 u32 size
= sizeof(*extent_item
) + sizeof(*iref
);
8192 u64 num_bytes
= ins
->offset
;
8193 bool skinny_metadata
= btrfs_fs_incompat(fs_info
, SKINNY_METADATA
);
8195 if (!skinny_metadata
)
8196 size
+= sizeof(*block_info
);
8198 path
= btrfs_alloc_path();
8202 path
->leave_spinning
= 1;
8203 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
8206 btrfs_free_path(path
);
8210 leaf
= path
->nodes
[0];
8211 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
8212 struct btrfs_extent_item
);
8213 btrfs_set_extent_refs(leaf
, extent_item
, 1);
8214 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
8215 btrfs_set_extent_flags(leaf
, extent_item
,
8216 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
8218 if (skinny_metadata
) {
8219 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
8220 num_bytes
= fs_info
->nodesize
;
8222 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
8223 btrfs_set_tree_block_key(leaf
, block_info
, key
);
8224 btrfs_set_tree_block_level(leaf
, block_info
, level
);
8225 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
8229 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
8230 btrfs_set_extent_inline_ref_type(leaf
, iref
,
8231 BTRFS_SHARED_BLOCK_REF_KEY
);
8232 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
8234 btrfs_set_extent_inline_ref_type(leaf
, iref
,
8235 BTRFS_TREE_BLOCK_REF_KEY
);
8236 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
8239 btrfs_mark_buffer_dirty(leaf
);
8240 btrfs_free_path(path
);
8242 ret
= remove_from_free_space_tree(trans
, fs_info
, ins
->objectid
,
8247 ret
= update_block_group(trans
, fs_info
, ins
->objectid
,
8248 fs_info
->nodesize
, 1);
8249 if (ret
) { /* -ENOENT, logic error */
8250 btrfs_err(fs_info
, "update block group failed for %llu %llu",
8251 ins
->objectid
, ins
->offset
);
8255 trace_btrfs_reserved_extent_alloc(fs_info
, ins
->objectid
,
8260 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
8261 struct btrfs_root
*root
, u64 owner
,
8262 u64 offset
, u64 ram_bytes
,
8263 struct btrfs_key
*ins
)
8265 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8268 BUG_ON(root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
);
8270 btrfs_ref_tree_mod(root
, ins
->objectid
, ins
->offset
, 0,
8271 root
->root_key
.objectid
, owner
, offset
,
8272 BTRFS_ADD_DELAYED_EXTENT
);
8274 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, ins
->objectid
,
8276 root
->root_key
.objectid
, owner
,
8278 BTRFS_ADD_DELAYED_EXTENT
, NULL
, NULL
);
8283 * this is used by the tree logging recovery code. It records that
8284 * an extent has been allocated and makes sure to clear the free
8285 * space cache bits as well
8287 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
8288 struct btrfs_fs_info
*fs_info
,
8289 u64 root_objectid
, u64 owner
, u64 offset
,
8290 struct btrfs_key
*ins
)
8293 struct btrfs_block_group_cache
*block_group
;
8294 struct btrfs_space_info
*space_info
;
8297 * Mixed block groups will exclude before processing the log so we only
8298 * need to do the exclude dance if this fs isn't mixed.
8300 if (!btrfs_fs_incompat(fs_info
, MIXED_GROUPS
)) {
8301 ret
= __exclude_logged_extent(fs_info
, ins
->objectid
,
8307 block_group
= btrfs_lookup_block_group(fs_info
, ins
->objectid
);
8311 space_info
= block_group
->space_info
;
8312 spin_lock(&space_info
->lock
);
8313 spin_lock(&block_group
->lock
);
8314 space_info
->bytes_reserved
+= ins
->offset
;
8315 block_group
->reserved
+= ins
->offset
;
8316 spin_unlock(&block_group
->lock
);
8317 spin_unlock(&space_info
->lock
);
8319 ret
= alloc_reserved_file_extent(trans
, fs_info
, 0, root_objectid
,
8320 0, owner
, offset
, ins
, 1);
8321 btrfs_put_block_group(block_group
);
8325 static struct extent_buffer
*
8326 btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
8327 u64 bytenr
, int level
)
8329 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8330 struct extent_buffer
*buf
;
8332 buf
= btrfs_find_create_tree_block(fs_info
, bytenr
);
8337 * Extra safety check in case the extent tree is corrupted and extent
8338 * allocator chooses to use a tree block which is already used and
8341 if (buf
->lock_owner
== current
->pid
) {
8342 btrfs_err_rl(fs_info
,
8343 "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected",
8344 buf
->start
, btrfs_header_owner(buf
), current
->pid
);
8345 free_extent_buffer(buf
);
8346 return ERR_PTR(-EUCLEAN
);
8349 btrfs_set_header_generation(buf
, trans
->transid
);
8350 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, buf
, level
);
8351 btrfs_tree_lock(buf
);
8352 clean_tree_block(fs_info
, buf
);
8353 clear_bit(EXTENT_BUFFER_STALE
, &buf
->bflags
);
8355 btrfs_set_lock_blocking(buf
);
8356 set_extent_buffer_uptodate(buf
);
8358 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
8359 buf
->log_index
= root
->log_transid
% 2;
8361 * we allow two log transactions at a time, use different
8362 * EXENT bit to differentiate dirty pages.
8364 if (buf
->log_index
== 0)
8365 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
8366 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
8368 set_extent_new(&root
->dirty_log_pages
, buf
->start
,
8369 buf
->start
+ buf
->len
- 1);
8371 buf
->log_index
= -1;
8372 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
8373 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
8375 trans
->dirty
= true;
8376 /* this returns a buffer locked for blocking */
8380 static struct btrfs_block_rsv
*
8381 use_block_rsv(struct btrfs_trans_handle
*trans
,
8382 struct btrfs_root
*root
, u32 blocksize
)
8384 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8385 struct btrfs_block_rsv
*block_rsv
;
8386 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
8388 bool global_updated
= false;
8390 block_rsv
= get_block_rsv(trans
, root
);
8392 if (unlikely(block_rsv
->size
== 0))
8395 ret
= block_rsv_use_bytes(block_rsv
, blocksize
);
8399 if (block_rsv
->failfast
)
8400 return ERR_PTR(ret
);
8402 if (block_rsv
->type
== BTRFS_BLOCK_RSV_GLOBAL
&& !global_updated
) {
8403 global_updated
= true;
8404 update_global_block_rsv(fs_info
);
8408 if (btrfs_test_opt(fs_info
, ENOSPC_DEBUG
)) {
8409 static DEFINE_RATELIMIT_STATE(_rs
,
8410 DEFAULT_RATELIMIT_INTERVAL
* 10,
8411 /*DEFAULT_RATELIMIT_BURST*/ 1);
8412 if (__ratelimit(&_rs
))
8414 "BTRFS: block rsv returned %d\n", ret
);
8417 ret
= reserve_metadata_bytes(root
, block_rsv
, blocksize
,
8418 BTRFS_RESERVE_NO_FLUSH
);
8422 * If we couldn't reserve metadata bytes try and use some from
8423 * the global reserve if its space type is the same as the global
8426 if (block_rsv
->type
!= BTRFS_BLOCK_RSV_GLOBAL
&&
8427 block_rsv
->space_info
== global_rsv
->space_info
) {
8428 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
8432 return ERR_PTR(ret
);
8435 static void unuse_block_rsv(struct btrfs_fs_info
*fs_info
,
8436 struct btrfs_block_rsv
*block_rsv
, u32 blocksize
)
8438 block_rsv_add_bytes(block_rsv
, blocksize
, 0);
8439 block_rsv_release_bytes(fs_info
, block_rsv
, NULL
, 0);
8443 * finds a free extent and does all the dirty work required for allocation
8444 * returns the tree buffer or an ERR_PTR on error.
8446 struct extent_buffer
*btrfs_alloc_tree_block(struct btrfs_trans_handle
*trans
,
8447 struct btrfs_root
*root
,
8448 u64 parent
, u64 root_objectid
,
8449 const struct btrfs_disk_key
*key
,
8450 int level
, u64 hint
,
8453 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8454 struct btrfs_key ins
;
8455 struct btrfs_block_rsv
*block_rsv
;
8456 struct extent_buffer
*buf
;
8457 struct btrfs_delayed_extent_op
*extent_op
;
8460 u32 blocksize
= fs_info
->nodesize
;
8461 bool skinny_metadata
= btrfs_fs_incompat(fs_info
, SKINNY_METADATA
);
8463 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8464 if (btrfs_is_testing(fs_info
)) {
8465 buf
= btrfs_init_new_buffer(trans
, root
, root
->alloc_bytenr
,
8468 root
->alloc_bytenr
+= blocksize
;
8473 block_rsv
= use_block_rsv(trans
, root
, blocksize
);
8474 if (IS_ERR(block_rsv
))
8475 return ERR_CAST(block_rsv
);
8477 ret
= btrfs_reserve_extent(root
, blocksize
, blocksize
, blocksize
,
8478 empty_size
, hint
, &ins
, 0, 0);
8482 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
, level
);
8485 goto out_free_reserved
;
8488 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
8490 parent
= ins
.objectid
;
8491 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
8495 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
8496 extent_op
= btrfs_alloc_delayed_extent_op();
8502 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
8504 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
8505 extent_op
->flags_to_set
= flags
;
8506 extent_op
->update_key
= skinny_metadata
? false : true;
8507 extent_op
->update_flags
= true;
8508 extent_op
->is_data
= false;
8509 extent_op
->level
= level
;
8511 btrfs_ref_tree_mod(root
, ins
.objectid
, ins
.offset
, parent
,
8512 root_objectid
, level
, 0,
8513 BTRFS_ADD_DELAYED_EXTENT
);
8514 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, ins
.objectid
,
8516 root_objectid
, level
,
8517 BTRFS_ADD_DELAYED_EXTENT
,
8518 extent_op
, NULL
, NULL
);
8520 goto out_free_delayed
;
8525 btrfs_free_delayed_extent_op(extent_op
);
8527 free_extent_buffer(buf
);
8529 btrfs_free_reserved_extent(fs_info
, ins
.objectid
, ins
.offset
, 0);
8531 unuse_block_rsv(fs_info
, block_rsv
, blocksize
);
8532 return ERR_PTR(ret
);
8535 struct walk_control
{
8536 u64 refs
[BTRFS_MAX_LEVEL
];
8537 u64 flags
[BTRFS_MAX_LEVEL
];
8538 struct btrfs_key update_progress
;
8549 #define DROP_REFERENCE 1
8550 #define UPDATE_BACKREF 2
8552 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
8553 struct btrfs_root
*root
,
8554 struct walk_control
*wc
,
8555 struct btrfs_path
*path
)
8557 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8563 struct btrfs_key key
;
8564 struct extent_buffer
*eb
;
8569 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
8570 wc
->reada_count
= wc
->reada_count
* 2 / 3;
8571 wc
->reada_count
= max(wc
->reada_count
, 2);
8573 wc
->reada_count
= wc
->reada_count
* 3 / 2;
8574 wc
->reada_count
= min_t(int, wc
->reada_count
,
8575 BTRFS_NODEPTRS_PER_BLOCK(fs_info
));
8578 eb
= path
->nodes
[wc
->level
];
8579 nritems
= btrfs_header_nritems(eb
);
8581 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
8582 if (nread
>= wc
->reada_count
)
8586 bytenr
= btrfs_node_blockptr(eb
, slot
);
8587 generation
= btrfs_node_ptr_generation(eb
, slot
);
8589 if (slot
== path
->slots
[wc
->level
])
8592 if (wc
->stage
== UPDATE_BACKREF
&&
8593 generation
<= root
->root_key
.offset
)
8596 /* We don't lock the tree block, it's OK to be racy here */
8597 ret
= btrfs_lookup_extent_info(trans
, fs_info
, bytenr
,
8598 wc
->level
- 1, 1, &refs
,
8600 /* We don't care about errors in readahead. */
8605 if (wc
->stage
== DROP_REFERENCE
) {
8609 if (wc
->level
== 1 &&
8610 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8612 if (!wc
->update_ref
||
8613 generation
<= root
->root_key
.offset
)
8615 btrfs_node_key_to_cpu(eb
, &key
, slot
);
8616 ret
= btrfs_comp_cpu_keys(&key
,
8617 &wc
->update_progress
);
8621 if (wc
->level
== 1 &&
8622 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8626 readahead_tree_block(fs_info
, bytenr
);
8629 wc
->reada_slot
= slot
;
8633 * helper to process tree block while walking down the tree.
8635 * when wc->stage == UPDATE_BACKREF, this function updates
8636 * back refs for pointers in the block.
8638 * NOTE: return value 1 means we should stop walking down.
8640 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
8641 struct btrfs_root
*root
,
8642 struct btrfs_path
*path
,
8643 struct walk_control
*wc
, int lookup_info
)
8645 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8646 int level
= wc
->level
;
8647 struct extent_buffer
*eb
= path
->nodes
[level
];
8648 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
8651 if (wc
->stage
== UPDATE_BACKREF
&&
8652 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
8656 * when reference count of tree block is 1, it won't increase
8657 * again. once full backref flag is set, we never clear it.
8660 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
8661 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
8662 BUG_ON(!path
->locks
[level
]);
8663 ret
= btrfs_lookup_extent_info(trans
, fs_info
,
8664 eb
->start
, level
, 1,
8667 BUG_ON(ret
== -ENOMEM
);
8670 BUG_ON(wc
->refs
[level
] == 0);
8673 if (wc
->stage
== DROP_REFERENCE
) {
8674 if (wc
->refs
[level
] > 1)
8677 if (path
->locks
[level
] && !wc
->keep_locks
) {
8678 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8679 path
->locks
[level
] = 0;
8684 /* wc->stage == UPDATE_BACKREF */
8685 if (!(wc
->flags
[level
] & flag
)) {
8686 BUG_ON(!path
->locks
[level
]);
8687 ret
= btrfs_inc_ref(trans
, root
, eb
, 1);
8688 BUG_ON(ret
); /* -ENOMEM */
8689 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
8690 BUG_ON(ret
); /* -ENOMEM */
8691 ret
= btrfs_set_disk_extent_flags(trans
, fs_info
, eb
->start
,
8693 btrfs_header_level(eb
), 0);
8694 BUG_ON(ret
); /* -ENOMEM */
8695 wc
->flags
[level
] |= flag
;
8699 * the block is shared by multiple trees, so it's not good to
8700 * keep the tree lock
8702 if (path
->locks
[level
] && level
> 0) {
8703 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8704 path
->locks
[level
] = 0;
8710 * helper to process tree block pointer.
8712 * when wc->stage == DROP_REFERENCE, this function checks
8713 * reference count of the block pointed to. if the block
8714 * is shared and we need update back refs for the subtree
8715 * rooted at the block, this function changes wc->stage to
8716 * UPDATE_BACKREF. if the block is shared and there is no
8717 * need to update back, this function drops the reference
8720 * NOTE: return value 1 means we should stop walking down.
8722 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
8723 struct btrfs_root
*root
,
8724 struct btrfs_path
*path
,
8725 struct walk_control
*wc
, int *lookup_info
)
8727 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8732 struct btrfs_key key
;
8733 struct extent_buffer
*next
;
8734 int level
= wc
->level
;
8737 bool need_account
= false;
8739 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
8740 path
->slots
[level
]);
8742 * if the lower level block was created before the snapshot
8743 * was created, we know there is no need to update back refs
8746 if (wc
->stage
== UPDATE_BACKREF
&&
8747 generation
<= root
->root_key
.offset
) {
8752 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
8753 blocksize
= fs_info
->nodesize
;
8755 next
= find_extent_buffer(fs_info
, bytenr
);
8757 next
= btrfs_find_create_tree_block(fs_info
, bytenr
);
8759 return PTR_ERR(next
);
8761 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, next
,
8765 btrfs_tree_lock(next
);
8766 btrfs_set_lock_blocking(next
);
8768 ret
= btrfs_lookup_extent_info(trans
, fs_info
, bytenr
, level
- 1, 1,
8769 &wc
->refs
[level
- 1],
8770 &wc
->flags
[level
- 1]);
8774 if (unlikely(wc
->refs
[level
- 1] == 0)) {
8775 btrfs_err(fs_info
, "Missing references.");
8781 if (wc
->stage
== DROP_REFERENCE
) {
8782 if (wc
->refs
[level
- 1] > 1) {
8783 need_account
= true;
8785 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8788 if (!wc
->update_ref
||
8789 generation
<= root
->root_key
.offset
)
8792 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
8793 path
->slots
[level
]);
8794 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
8798 wc
->stage
= UPDATE_BACKREF
;
8799 wc
->shared_level
= level
- 1;
8803 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8807 if (!btrfs_buffer_uptodate(next
, generation
, 0)) {
8808 btrfs_tree_unlock(next
);
8809 free_extent_buffer(next
);
8815 if (reada
&& level
== 1)
8816 reada_walk_down(trans
, root
, wc
, path
);
8817 next
= read_tree_block(fs_info
, bytenr
, generation
);
8819 return PTR_ERR(next
);
8820 } else if (!extent_buffer_uptodate(next
)) {
8821 free_extent_buffer(next
);
8824 btrfs_tree_lock(next
);
8825 btrfs_set_lock_blocking(next
);
8829 ASSERT(level
== btrfs_header_level(next
));
8830 if (level
!= btrfs_header_level(next
)) {
8831 btrfs_err(root
->fs_info
, "mismatched level");
8835 path
->nodes
[level
] = next
;
8836 path
->slots
[level
] = 0;
8837 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8843 wc
->refs
[level
- 1] = 0;
8844 wc
->flags
[level
- 1] = 0;
8845 if (wc
->stage
== DROP_REFERENCE
) {
8846 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
8847 parent
= path
->nodes
[level
]->start
;
8849 ASSERT(root
->root_key
.objectid
==
8850 btrfs_header_owner(path
->nodes
[level
]));
8851 if (root
->root_key
.objectid
!=
8852 btrfs_header_owner(path
->nodes
[level
])) {
8853 btrfs_err(root
->fs_info
,
8854 "mismatched block owner");
8862 ret
= btrfs_qgroup_trace_subtree(trans
, root
, next
,
8863 generation
, level
- 1);
8865 btrfs_err_rl(fs_info
,
8866 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
8870 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
,
8871 parent
, root
->root_key
.objectid
,
8881 btrfs_tree_unlock(next
);
8882 free_extent_buffer(next
);
8888 * helper to process tree block while walking up the tree.
8890 * when wc->stage == DROP_REFERENCE, this function drops
8891 * reference count on the block.
8893 * when wc->stage == UPDATE_BACKREF, this function changes
8894 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8895 * to UPDATE_BACKREF previously while processing the block.
8897 * NOTE: return value 1 means we should stop walking up.
8899 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
8900 struct btrfs_root
*root
,
8901 struct btrfs_path
*path
,
8902 struct walk_control
*wc
)
8904 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8906 int level
= wc
->level
;
8907 struct extent_buffer
*eb
= path
->nodes
[level
];
8910 if (wc
->stage
== UPDATE_BACKREF
) {
8911 BUG_ON(wc
->shared_level
< level
);
8912 if (level
< wc
->shared_level
)
8915 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
8919 wc
->stage
= DROP_REFERENCE
;
8920 wc
->shared_level
= -1;
8921 path
->slots
[level
] = 0;
8924 * check reference count again if the block isn't locked.
8925 * we should start walking down the tree again if reference
8928 if (!path
->locks
[level
]) {
8930 btrfs_tree_lock(eb
);
8931 btrfs_set_lock_blocking(eb
);
8932 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8934 ret
= btrfs_lookup_extent_info(trans
, fs_info
,
8935 eb
->start
, level
, 1,
8939 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8940 path
->locks
[level
] = 0;
8943 BUG_ON(wc
->refs
[level
] == 0);
8944 if (wc
->refs
[level
] == 1) {
8945 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8946 path
->locks
[level
] = 0;
8952 /* wc->stage == DROP_REFERENCE */
8953 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
8955 if (wc
->refs
[level
] == 1) {
8957 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8958 ret
= btrfs_dec_ref(trans
, root
, eb
, 1);
8960 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
8961 BUG_ON(ret
); /* -ENOMEM */
8962 ret
= btrfs_qgroup_trace_leaf_items(trans
, fs_info
, eb
);
8964 btrfs_err_rl(fs_info
,
8965 "error %d accounting leaf items. Quota is out of sync, rescan required.",
8969 /* make block locked assertion in clean_tree_block happy */
8970 if (!path
->locks
[level
] &&
8971 btrfs_header_generation(eb
) == trans
->transid
) {
8972 btrfs_tree_lock(eb
);
8973 btrfs_set_lock_blocking(eb
);
8974 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8976 clean_tree_block(fs_info
, eb
);
8979 if (eb
== root
->node
) {
8980 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8982 else if (root
->root_key
.objectid
!= btrfs_header_owner(eb
))
8983 goto owner_mismatch
;
8985 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8986 parent
= path
->nodes
[level
+ 1]->start
;
8987 else if (root
->root_key
.objectid
!=
8988 btrfs_header_owner(path
->nodes
[level
+ 1]))
8989 goto owner_mismatch
;
8992 btrfs_free_tree_block(trans
, root
, eb
, parent
, wc
->refs
[level
] == 1);
8994 wc
->refs
[level
] = 0;
8995 wc
->flags
[level
] = 0;
8999 btrfs_err_rl(fs_info
, "unexpected tree owner, have %llu expect %llu",
9000 btrfs_header_owner(eb
), root
->root_key
.objectid
);
9004 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
9005 struct btrfs_root
*root
,
9006 struct btrfs_path
*path
,
9007 struct walk_control
*wc
)
9009 int level
= wc
->level
;
9010 int lookup_info
= 1;
9013 while (level
>= 0) {
9014 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
9021 if (path
->slots
[level
] >=
9022 btrfs_header_nritems(path
->nodes
[level
]))
9025 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
9027 path
->slots
[level
]++;
9036 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
9037 struct btrfs_root
*root
,
9038 struct btrfs_path
*path
,
9039 struct walk_control
*wc
, int max_level
)
9041 int level
= wc
->level
;
9044 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
9045 while (level
< max_level
&& path
->nodes
[level
]) {
9047 if (path
->slots
[level
] + 1 <
9048 btrfs_header_nritems(path
->nodes
[level
])) {
9049 path
->slots
[level
]++;
9052 ret
= walk_up_proc(trans
, root
, path
, wc
);
9058 if (path
->locks
[level
]) {
9059 btrfs_tree_unlock_rw(path
->nodes
[level
],
9060 path
->locks
[level
]);
9061 path
->locks
[level
] = 0;
9063 free_extent_buffer(path
->nodes
[level
]);
9064 path
->nodes
[level
] = NULL
;
9072 * drop a subvolume tree.
9074 * this function traverses the tree freeing any blocks that only
9075 * referenced by the tree.
9077 * when a shared tree block is found. this function decreases its
9078 * reference count by one. if update_ref is true, this function
9079 * also make sure backrefs for the shared block and all lower level
9080 * blocks are properly updated.
9082 * If called with for_reloc == 0, may exit early with -EAGAIN
9084 int btrfs_drop_snapshot(struct btrfs_root
*root
,
9085 struct btrfs_block_rsv
*block_rsv
, int update_ref
,
9088 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
9089 struct btrfs_path
*path
;
9090 struct btrfs_trans_handle
*trans
;
9091 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
9092 struct btrfs_root_item
*root_item
= &root
->root_item
;
9093 struct walk_control
*wc
;
9094 struct btrfs_key key
;
9098 bool root_dropped
= false;
9100 btrfs_debug(fs_info
, "Drop subvolume %llu", root
->objectid
);
9102 path
= btrfs_alloc_path();
9108 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
9110 btrfs_free_path(path
);
9115 trans
= btrfs_start_transaction(tree_root
, 0);
9116 if (IS_ERR(trans
)) {
9117 err
= PTR_ERR(trans
);
9121 err
= btrfs_run_delayed_items(trans
, fs_info
);
9126 trans
->block_rsv
= block_rsv
;
9128 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
9129 level
= btrfs_header_level(root
->node
);
9130 path
->nodes
[level
] = btrfs_lock_root_node(root
);
9131 btrfs_set_lock_blocking(path
->nodes
[level
]);
9132 path
->slots
[level
] = 0;
9133 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
9134 memset(&wc
->update_progress
, 0,
9135 sizeof(wc
->update_progress
));
9137 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
9138 memcpy(&wc
->update_progress
, &key
,
9139 sizeof(wc
->update_progress
));
9141 level
= root_item
->drop_level
;
9143 path
->lowest_level
= level
;
9144 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
9145 path
->lowest_level
= 0;
9153 * unlock our path, this is safe because only this
9154 * function is allowed to delete this snapshot
9156 btrfs_unlock_up_safe(path
, 0);
9158 level
= btrfs_header_level(root
->node
);
9160 btrfs_tree_lock(path
->nodes
[level
]);
9161 btrfs_set_lock_blocking(path
->nodes
[level
]);
9162 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
9164 ret
= btrfs_lookup_extent_info(trans
, fs_info
,
9165 path
->nodes
[level
]->start
,
9166 level
, 1, &wc
->refs
[level
],
9172 BUG_ON(wc
->refs
[level
] == 0);
9174 if (level
== root_item
->drop_level
)
9177 btrfs_tree_unlock(path
->nodes
[level
]);
9178 path
->locks
[level
] = 0;
9179 WARN_ON(wc
->refs
[level
] != 1);
9185 wc
->shared_level
= -1;
9186 wc
->stage
= DROP_REFERENCE
;
9187 wc
->update_ref
= update_ref
;
9189 wc
->for_reloc
= for_reloc
;
9190 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(fs_info
);
9194 ret
= walk_down_tree(trans
, root
, path
, wc
);
9200 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
9207 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
9211 if (wc
->stage
== DROP_REFERENCE
) {
9213 btrfs_node_key(path
->nodes
[level
],
9214 &root_item
->drop_progress
,
9215 path
->slots
[level
]);
9216 root_item
->drop_level
= level
;
9219 BUG_ON(wc
->level
== 0);
9220 if (btrfs_should_end_transaction(trans
) ||
9221 (!for_reloc
&& btrfs_need_cleaner_sleep(fs_info
))) {
9222 ret
= btrfs_update_root(trans
, tree_root
,
9226 btrfs_abort_transaction(trans
, ret
);
9231 btrfs_end_transaction_throttle(trans
);
9232 if (!for_reloc
&& btrfs_need_cleaner_sleep(fs_info
)) {
9233 btrfs_debug(fs_info
,
9234 "drop snapshot early exit");
9239 trans
= btrfs_start_transaction(tree_root
, 0);
9240 if (IS_ERR(trans
)) {
9241 err
= PTR_ERR(trans
);
9245 trans
->block_rsv
= block_rsv
;
9248 btrfs_release_path(path
);
9252 ret
= btrfs_del_root(trans
, fs_info
, &root
->root_key
);
9254 btrfs_abort_transaction(trans
, ret
);
9259 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
9260 ret
= btrfs_find_root(tree_root
, &root
->root_key
, path
,
9263 btrfs_abort_transaction(trans
, ret
);
9266 } else if (ret
> 0) {
9267 /* if we fail to delete the orphan item this time
9268 * around, it'll get picked up the next time.
9270 * The most common failure here is just -ENOENT.
9272 btrfs_del_orphan_item(trans
, tree_root
,
9273 root
->root_key
.objectid
);
9277 if (test_bit(BTRFS_ROOT_IN_RADIX
, &root
->state
)) {
9278 btrfs_add_dropped_root(trans
, root
);
9280 free_extent_buffer(root
->node
);
9281 free_extent_buffer(root
->commit_root
);
9282 btrfs_put_fs_root(root
);
9284 root_dropped
= true;
9286 btrfs_end_transaction_throttle(trans
);
9289 btrfs_free_path(path
);
9292 * So if we need to stop dropping the snapshot for whatever reason we
9293 * need to make sure to add it back to the dead root list so that we
9294 * keep trying to do the work later. This also cleans up roots if we
9295 * don't have it in the radix (like when we recover after a power fail
9296 * or unmount) so we don't leak memory.
9298 if (!for_reloc
&& !root_dropped
)
9299 btrfs_add_dead_root(root
);
9300 if (err
&& err
!= -EAGAIN
)
9301 btrfs_handle_fs_error(fs_info
, err
, NULL
);
9306 * drop subtree rooted at tree block 'node'.
9308 * NOTE: this function will unlock and release tree block 'node'
9309 * only used by relocation code
9311 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
9312 struct btrfs_root
*root
,
9313 struct extent_buffer
*node
,
9314 struct extent_buffer
*parent
)
9316 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
9317 struct btrfs_path
*path
;
9318 struct walk_control
*wc
;
9324 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
9326 path
= btrfs_alloc_path();
9330 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
9332 btrfs_free_path(path
);
9336 btrfs_assert_tree_locked(parent
);
9337 parent_level
= btrfs_header_level(parent
);
9338 extent_buffer_get(parent
);
9339 path
->nodes
[parent_level
] = parent
;
9340 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
9342 btrfs_assert_tree_locked(node
);
9343 level
= btrfs_header_level(node
);
9344 path
->nodes
[level
] = node
;
9345 path
->slots
[level
] = 0;
9346 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
9348 wc
->refs
[parent_level
] = 1;
9349 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
9351 wc
->shared_level
= -1;
9352 wc
->stage
= DROP_REFERENCE
;
9356 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(fs_info
);
9359 wret
= walk_down_tree(trans
, root
, path
, wc
);
9365 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
9373 btrfs_free_path(path
);
9377 static u64
update_block_group_flags(struct btrfs_fs_info
*fs_info
, u64 flags
)
9383 * if restripe for this chunk_type is on pick target profile and
9384 * return, otherwise do the usual balance
9386 stripped
= get_restripe_target(fs_info
, flags
);
9388 return extended_to_chunk(stripped
);
9390 num_devices
= fs_info
->fs_devices
->rw_devices
;
9392 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
9393 BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
|
9394 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
9396 if (num_devices
== 1) {
9397 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
9398 stripped
= flags
& ~stripped
;
9400 /* turn raid0 into single device chunks */
9401 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
9404 /* turn mirroring into duplication */
9405 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
9406 BTRFS_BLOCK_GROUP_RAID10
))
9407 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
9409 /* they already had raid on here, just return */
9410 if (flags
& stripped
)
9413 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
9414 stripped
= flags
& ~stripped
;
9416 /* switch duplicated blocks with raid1 */
9417 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
9418 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
9420 /* this is drive concat, leave it alone */
9426 static int inc_block_group_ro(struct btrfs_block_group_cache
*cache
, int force
)
9428 struct btrfs_space_info
*sinfo
= cache
->space_info
;
9430 u64 min_allocable_bytes
;
9434 * We need some metadata space and system metadata space for
9435 * allocating chunks in some corner cases until we force to set
9436 * it to be readonly.
9439 (BTRFS_BLOCK_GROUP_SYSTEM
| BTRFS_BLOCK_GROUP_METADATA
)) &&
9441 min_allocable_bytes
= SZ_1M
;
9443 min_allocable_bytes
= 0;
9445 spin_lock(&sinfo
->lock
);
9446 spin_lock(&cache
->lock
);
9454 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
9455 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
9457 if (btrfs_space_info_used(sinfo
, true) + num_bytes
+
9458 min_allocable_bytes
<= sinfo
->total_bytes
) {
9459 sinfo
->bytes_readonly
+= num_bytes
;
9461 list_add_tail(&cache
->ro_list
, &sinfo
->ro_bgs
);
9465 spin_unlock(&cache
->lock
);
9466 spin_unlock(&sinfo
->lock
);
9470 int btrfs_inc_block_group_ro(struct btrfs_fs_info
*fs_info
,
9471 struct btrfs_block_group_cache
*cache
)
9474 struct btrfs_trans_handle
*trans
;
9479 trans
= btrfs_join_transaction(fs_info
->extent_root
);
9481 return PTR_ERR(trans
);
9484 * we're not allowed to set block groups readonly after the dirty
9485 * block groups cache has started writing. If it already started,
9486 * back off and let this transaction commit
9488 mutex_lock(&fs_info
->ro_block_group_mutex
);
9489 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN
, &trans
->transaction
->flags
)) {
9490 u64 transid
= trans
->transid
;
9492 mutex_unlock(&fs_info
->ro_block_group_mutex
);
9493 btrfs_end_transaction(trans
);
9495 ret
= btrfs_wait_for_commit(fs_info
, transid
);
9502 * if we are changing raid levels, try to allocate a corresponding
9503 * block group with the new raid level.
9505 alloc_flags
= update_block_group_flags(fs_info
, cache
->flags
);
9506 if (alloc_flags
!= cache
->flags
) {
9507 ret
= do_chunk_alloc(trans
, fs_info
, alloc_flags
,
9510 * ENOSPC is allowed here, we may have enough space
9511 * already allocated at the new raid level to
9520 ret
= inc_block_group_ro(cache
, 0);
9523 alloc_flags
= get_alloc_profile(fs_info
, cache
->space_info
->flags
);
9524 ret
= do_chunk_alloc(trans
, fs_info
, alloc_flags
,
9528 ret
= inc_block_group_ro(cache
, 0);
9530 if (cache
->flags
& BTRFS_BLOCK_GROUP_SYSTEM
) {
9531 alloc_flags
= update_block_group_flags(fs_info
, cache
->flags
);
9532 mutex_lock(&fs_info
->chunk_mutex
);
9533 check_system_chunk(trans
, fs_info
, alloc_flags
);
9534 mutex_unlock(&fs_info
->chunk_mutex
);
9536 mutex_unlock(&fs_info
->ro_block_group_mutex
);
9538 btrfs_end_transaction(trans
);
9542 int btrfs_force_chunk_alloc(struct btrfs_trans_handle
*trans
,
9543 struct btrfs_fs_info
*fs_info
, u64 type
)
9545 u64 alloc_flags
= get_alloc_profile(fs_info
, type
);
9547 return do_chunk_alloc(trans
, fs_info
, alloc_flags
, CHUNK_ALLOC_FORCE
);
9551 * helper to account the unused space of all the readonly block group in the
9552 * space_info. takes mirrors into account.
9554 u64
btrfs_account_ro_block_groups_free_space(struct btrfs_space_info
*sinfo
)
9556 struct btrfs_block_group_cache
*block_group
;
9560 /* It's df, we don't care if it's racy */
9561 if (list_empty(&sinfo
->ro_bgs
))
9564 spin_lock(&sinfo
->lock
);
9565 list_for_each_entry(block_group
, &sinfo
->ro_bgs
, ro_list
) {
9566 spin_lock(&block_group
->lock
);
9568 if (!block_group
->ro
) {
9569 spin_unlock(&block_group
->lock
);
9573 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_RAID1
|
9574 BTRFS_BLOCK_GROUP_RAID10
|
9575 BTRFS_BLOCK_GROUP_DUP
))
9580 free_bytes
+= (block_group
->key
.offset
-
9581 btrfs_block_group_used(&block_group
->item
)) *
9584 spin_unlock(&block_group
->lock
);
9586 spin_unlock(&sinfo
->lock
);
9591 void btrfs_dec_block_group_ro(struct btrfs_block_group_cache
*cache
)
9593 struct btrfs_space_info
*sinfo
= cache
->space_info
;
9598 spin_lock(&sinfo
->lock
);
9599 spin_lock(&cache
->lock
);
9601 num_bytes
= cache
->key
.offset
- cache
->reserved
-
9602 cache
->pinned
- cache
->bytes_super
-
9603 btrfs_block_group_used(&cache
->item
);
9604 sinfo
->bytes_readonly
-= num_bytes
;
9605 list_del_init(&cache
->ro_list
);
9607 spin_unlock(&cache
->lock
);
9608 spin_unlock(&sinfo
->lock
);
9612 * checks to see if its even possible to relocate this block group.
9614 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9615 * ok to go ahead and try.
9617 int btrfs_can_relocate(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
9619 struct btrfs_root
*root
= fs_info
->extent_root
;
9620 struct btrfs_block_group_cache
*block_group
;
9621 struct btrfs_space_info
*space_info
;
9622 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
9623 struct btrfs_device
*device
;
9624 struct btrfs_trans_handle
*trans
;
9634 debug
= btrfs_test_opt(fs_info
, ENOSPC_DEBUG
);
9636 block_group
= btrfs_lookup_block_group(fs_info
, bytenr
);
9638 /* odd, couldn't find the block group, leave it alone */
9642 "can't find block group for bytenr %llu",
9647 min_free
= btrfs_block_group_used(&block_group
->item
);
9649 /* no bytes used, we're good */
9653 space_info
= block_group
->space_info
;
9654 spin_lock(&space_info
->lock
);
9656 full
= space_info
->full
;
9659 * if this is the last block group we have in this space, we can't
9660 * relocate it unless we're able to allocate a new chunk below.
9662 * Otherwise, we need to make sure we have room in the space to handle
9663 * all of the extents from this block group. If we can, we're good
9665 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
9666 (btrfs_space_info_used(space_info
, false) + min_free
<
9667 space_info
->total_bytes
)) {
9668 spin_unlock(&space_info
->lock
);
9671 spin_unlock(&space_info
->lock
);
9674 * ok we don't have enough space, but maybe we have free space on our
9675 * devices to allocate new chunks for relocation, so loop through our
9676 * alloc devices and guess if we have enough space. if this block
9677 * group is going to be restriped, run checks against the target
9678 * profile instead of the current one.
9690 target
= get_restripe_target(fs_info
, block_group
->flags
);
9692 index
= __get_raid_index(extended_to_chunk(target
));
9695 * this is just a balance, so if we were marked as full
9696 * we know there is no space for a new chunk
9701 "no space to alloc new chunk for block group %llu",
9702 block_group
->key
.objectid
);
9706 index
= get_block_group_index(block_group
);
9709 if (index
== BTRFS_RAID_RAID10
) {
9713 } else if (index
== BTRFS_RAID_RAID1
) {
9715 } else if (index
== BTRFS_RAID_DUP
) {
9718 } else if (index
== BTRFS_RAID_RAID0
) {
9719 dev_min
= fs_devices
->rw_devices
;
9720 min_free
= div64_u64(min_free
, dev_min
);
9723 /* We need to do this so that we can look at pending chunks */
9724 trans
= btrfs_join_transaction(root
);
9725 if (IS_ERR(trans
)) {
9726 ret
= PTR_ERR(trans
);
9730 mutex_lock(&fs_info
->chunk_mutex
);
9731 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
9735 * check to make sure we can actually find a chunk with enough
9736 * space to fit our block group in.
9738 if (device
->total_bytes
> device
->bytes_used
+ min_free
&&
9739 !device
->is_tgtdev_for_dev_replace
) {
9740 ret
= find_free_dev_extent(trans
, device
, min_free
,
9745 if (dev_nr
>= dev_min
)
9751 if (debug
&& ret
== -1)
9753 "no space to allocate a new chunk for block group %llu",
9754 block_group
->key
.objectid
);
9755 mutex_unlock(&fs_info
->chunk_mutex
);
9756 btrfs_end_transaction(trans
);
9758 btrfs_put_block_group(block_group
);
9762 static int find_first_block_group(struct btrfs_fs_info
*fs_info
,
9763 struct btrfs_path
*path
,
9764 struct btrfs_key
*key
)
9766 struct btrfs_root
*root
= fs_info
->extent_root
;
9768 struct btrfs_key found_key
;
9769 struct extent_buffer
*leaf
;
9770 struct btrfs_block_group_item bg
;
9774 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
9779 slot
= path
->slots
[0];
9780 leaf
= path
->nodes
[0];
9781 if (slot
>= btrfs_header_nritems(leaf
)) {
9782 ret
= btrfs_next_leaf(root
, path
);
9789 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
9791 if (found_key
.objectid
>= key
->objectid
&&
9792 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
9793 struct extent_map_tree
*em_tree
;
9794 struct extent_map
*em
;
9796 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
9797 read_lock(&em_tree
->lock
);
9798 em
= lookup_extent_mapping(em_tree
, found_key
.objectid
,
9800 read_unlock(&em_tree
->lock
);
9803 "logical %llu len %llu found bg but no related chunk",
9804 found_key
.objectid
, found_key
.offset
);
9806 } else if (em
->start
!= found_key
.objectid
||
9807 em
->len
!= found_key
.offset
) {
9809 "block group %llu len %llu mismatch with chunk %llu len %llu",
9810 found_key
.objectid
, found_key
.offset
,
9811 em
->start
, em
->len
);
9814 read_extent_buffer(leaf
, &bg
,
9815 btrfs_item_ptr_offset(leaf
, slot
),
9817 flags
= btrfs_block_group_flags(&bg
) &
9818 BTRFS_BLOCK_GROUP_TYPE_MASK
;
9820 if (flags
!= (em
->map_lookup
->type
&
9821 BTRFS_BLOCK_GROUP_TYPE_MASK
)) {
9823 "block group %llu len %llu type flags 0x%llx mismatch with chunk type flags 0x%llx",
9825 found_key
.offset
, flags
,
9826 (BTRFS_BLOCK_GROUP_TYPE_MASK
&
9827 em
->map_lookup
->type
));
9833 free_extent_map(em
);
9842 void btrfs_put_block_group_cache(struct btrfs_fs_info
*info
)
9844 struct btrfs_block_group_cache
*block_group
;
9848 struct inode
*inode
;
9850 block_group
= btrfs_lookup_first_block_group(info
, last
);
9851 while (block_group
) {
9852 wait_block_group_cache_done(block_group
);
9853 spin_lock(&block_group
->lock
);
9854 if (block_group
->iref
)
9856 spin_unlock(&block_group
->lock
);
9857 block_group
= next_block_group(info
, block_group
);
9866 inode
= block_group
->inode
;
9867 block_group
->iref
= 0;
9868 block_group
->inode
= NULL
;
9869 spin_unlock(&block_group
->lock
);
9870 ASSERT(block_group
->io_ctl
.inode
== NULL
);
9872 last
= block_group
->key
.objectid
+ block_group
->key
.offset
;
9873 btrfs_put_block_group(block_group
);
9878 * Must be called only after stopping all workers, since we could have block
9879 * group caching kthreads running, and therefore they could race with us if we
9880 * freed the block groups before stopping them.
9882 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
9884 struct btrfs_block_group_cache
*block_group
;
9885 struct btrfs_space_info
*space_info
;
9886 struct btrfs_caching_control
*caching_ctl
;
9889 down_write(&info
->commit_root_sem
);
9890 while (!list_empty(&info
->caching_block_groups
)) {
9891 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
9892 struct btrfs_caching_control
, list
);
9893 list_del(&caching_ctl
->list
);
9894 put_caching_control(caching_ctl
);
9896 up_write(&info
->commit_root_sem
);
9898 spin_lock(&info
->unused_bgs_lock
);
9899 while (!list_empty(&info
->unused_bgs
)) {
9900 block_group
= list_first_entry(&info
->unused_bgs
,
9901 struct btrfs_block_group_cache
,
9903 list_del_init(&block_group
->bg_list
);
9904 btrfs_put_block_group(block_group
);
9906 spin_unlock(&info
->unused_bgs_lock
);
9908 spin_lock(&info
->block_group_cache_lock
);
9909 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
9910 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
9912 rb_erase(&block_group
->cache_node
,
9913 &info
->block_group_cache_tree
);
9914 RB_CLEAR_NODE(&block_group
->cache_node
);
9915 spin_unlock(&info
->block_group_cache_lock
);
9917 down_write(&block_group
->space_info
->groups_sem
);
9918 list_del(&block_group
->list
);
9919 up_write(&block_group
->space_info
->groups_sem
);
9922 * We haven't cached this block group, which means we could
9923 * possibly have excluded extents on this block group.
9925 if (block_group
->cached
== BTRFS_CACHE_NO
||
9926 block_group
->cached
== BTRFS_CACHE_ERROR
)
9927 free_excluded_extents(info
, block_group
);
9929 btrfs_remove_free_space_cache(block_group
);
9930 ASSERT(block_group
->cached
!= BTRFS_CACHE_STARTED
);
9931 ASSERT(list_empty(&block_group
->dirty_list
));
9932 ASSERT(list_empty(&block_group
->io_list
));
9933 ASSERT(list_empty(&block_group
->bg_list
));
9934 ASSERT(atomic_read(&block_group
->count
) == 1);
9935 btrfs_put_block_group(block_group
);
9937 spin_lock(&info
->block_group_cache_lock
);
9939 spin_unlock(&info
->block_group_cache_lock
);
9941 /* now that all the block groups are freed, go through and
9942 * free all the space_info structs. This is only called during
9943 * the final stages of unmount, and so we know nobody is
9944 * using them. We call synchronize_rcu() once before we start,
9945 * just to be on the safe side.
9949 release_global_block_rsv(info
);
9951 while (!list_empty(&info
->space_info
)) {
9954 space_info
= list_entry(info
->space_info
.next
,
9955 struct btrfs_space_info
,
9959 * Do not hide this behind enospc_debug, this is actually
9960 * important and indicates a real bug if this happens.
9962 if (WARN_ON(space_info
->bytes_pinned
> 0 ||
9963 space_info
->bytes_reserved
> 0 ||
9964 space_info
->bytes_may_use
> 0))
9965 dump_space_info(info
, space_info
, 0, 0);
9966 list_del(&space_info
->list
);
9967 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++) {
9968 struct kobject
*kobj
;
9969 kobj
= space_info
->block_group_kobjs
[i
];
9970 space_info
->block_group_kobjs
[i
] = NULL
;
9976 kobject_del(&space_info
->kobj
);
9977 kobject_put(&space_info
->kobj
);
9982 static void link_block_group(struct btrfs_block_group_cache
*cache
)
9984 struct btrfs_space_info
*space_info
= cache
->space_info
;
9985 int index
= get_block_group_index(cache
);
9988 down_write(&space_info
->groups_sem
);
9989 if (list_empty(&space_info
->block_groups
[index
]))
9991 list_add_tail(&cache
->list
, &space_info
->block_groups
[index
]);
9992 up_write(&space_info
->groups_sem
);
9995 struct raid_kobject
*rkobj
;
9998 rkobj
= kzalloc(sizeof(*rkobj
), GFP_NOFS
);
10001 rkobj
->raid_type
= index
;
10002 kobject_init(&rkobj
->kobj
, &btrfs_raid_ktype
);
10003 ret
= kobject_add(&rkobj
->kobj
, &space_info
->kobj
,
10004 "%s", get_raid_name(index
));
10006 kobject_put(&rkobj
->kobj
);
10009 space_info
->block_group_kobjs
[index
] = &rkobj
->kobj
;
10014 btrfs_warn(cache
->fs_info
,
10015 "failed to add kobject for block cache, ignoring");
10018 static struct btrfs_block_group_cache
*
10019 btrfs_create_block_group_cache(struct btrfs_fs_info
*fs_info
,
10020 u64 start
, u64 size
)
10022 struct btrfs_block_group_cache
*cache
;
10024 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
10028 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
10030 if (!cache
->free_space_ctl
) {
10035 cache
->key
.objectid
= start
;
10036 cache
->key
.offset
= size
;
10037 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
10039 cache
->fs_info
= fs_info
;
10040 cache
->full_stripe_len
= btrfs_full_stripe_len(fs_info
, start
);
10041 set_free_space_tree_thresholds(cache
);
10043 atomic_set(&cache
->count
, 1);
10044 spin_lock_init(&cache
->lock
);
10045 init_rwsem(&cache
->data_rwsem
);
10046 INIT_LIST_HEAD(&cache
->list
);
10047 INIT_LIST_HEAD(&cache
->cluster_list
);
10048 INIT_LIST_HEAD(&cache
->bg_list
);
10049 INIT_LIST_HEAD(&cache
->ro_list
);
10050 INIT_LIST_HEAD(&cache
->dirty_list
);
10051 INIT_LIST_HEAD(&cache
->io_list
);
10052 btrfs_init_free_space_ctl(cache
);
10053 atomic_set(&cache
->trimming
, 0);
10054 mutex_init(&cache
->free_space_lock
);
10055 btrfs_init_full_stripe_locks_tree(&cache
->full_stripe_locks_root
);
10062 * Iterate all chunks and verify that each of them has the corresponding block
10065 static int check_chunk_block_group_mappings(struct btrfs_fs_info
*fs_info
)
10067 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
10068 struct extent_map
*em
;
10069 struct btrfs_block_group_cache
*bg
;
10074 read_lock(&map_tree
->map_tree
.lock
);
10076 * lookup_extent_mapping will return the first extent map
10077 * intersecting the range, so setting @len to 1 is enough to
10078 * get the first chunk.
10080 em
= lookup_extent_mapping(&map_tree
->map_tree
, start
, 1);
10081 read_unlock(&map_tree
->map_tree
.lock
);
10085 bg
= btrfs_lookup_block_group(fs_info
, em
->start
);
10088 "chunk start=%llu len=%llu doesn't have corresponding block group",
10089 em
->start
, em
->len
);
10091 free_extent_map(em
);
10094 if (bg
->key
.objectid
!= em
->start
||
10095 bg
->key
.offset
!= em
->len
||
10096 (bg
->flags
& BTRFS_BLOCK_GROUP_TYPE_MASK
) !=
10097 (em
->map_lookup
->type
& BTRFS_BLOCK_GROUP_TYPE_MASK
)) {
10099 "chunk start=%llu len=%llu flags=0x%llx doesn't match block group start=%llu len=%llu flags=0x%llx",
10100 em
->start
, em
->len
,
10101 em
->map_lookup
->type
& BTRFS_BLOCK_GROUP_TYPE_MASK
,
10102 bg
->key
.objectid
, bg
->key
.offset
,
10103 bg
->flags
& BTRFS_BLOCK_GROUP_TYPE_MASK
);
10105 free_extent_map(em
);
10106 btrfs_put_block_group(bg
);
10109 start
= em
->start
+ em
->len
;
10110 free_extent_map(em
);
10111 btrfs_put_block_group(bg
);
10116 int btrfs_read_block_groups(struct btrfs_fs_info
*info
)
10118 struct btrfs_path
*path
;
10120 struct btrfs_block_group_cache
*cache
;
10121 struct btrfs_space_info
*space_info
;
10122 struct btrfs_key key
;
10123 struct btrfs_key found_key
;
10124 struct extent_buffer
*leaf
;
10125 int need_clear
= 0;
10130 feature
= btrfs_super_incompat_flags(info
->super_copy
);
10131 mixed
= !!(feature
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
);
10135 key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
10136 path
= btrfs_alloc_path();
10139 path
->reada
= READA_FORWARD
;
10141 cache_gen
= btrfs_super_cache_generation(info
->super_copy
);
10142 if (btrfs_test_opt(info
, SPACE_CACHE
) &&
10143 btrfs_super_generation(info
->super_copy
) != cache_gen
)
10145 if (btrfs_test_opt(info
, CLEAR_CACHE
))
10149 ret
= find_first_block_group(info
, path
, &key
);
10155 leaf
= path
->nodes
[0];
10156 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
10158 cache
= btrfs_create_block_group_cache(info
, found_key
.objectid
,
10167 * When we mount with old space cache, we need to
10168 * set BTRFS_DC_CLEAR and set dirty flag.
10170 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
10171 * truncate the old free space cache inode and
10173 * b) Setting 'dirty flag' makes sure that we flush
10174 * the new space cache info onto disk.
10176 if (btrfs_test_opt(info
, SPACE_CACHE
))
10177 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
10180 read_extent_buffer(leaf
, &cache
->item
,
10181 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
10182 sizeof(cache
->item
));
10183 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
10185 ((cache
->flags
& BTRFS_BLOCK_GROUP_METADATA
) &&
10186 (cache
->flags
& BTRFS_BLOCK_GROUP_DATA
))) {
10188 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
10189 cache
->key
.objectid
);
10190 btrfs_put_block_group(cache
);
10195 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
10196 btrfs_release_path(path
);
10199 * We need to exclude the super stripes now so that the space
10200 * info has super bytes accounted for, otherwise we'll think
10201 * we have more space than we actually do.
10203 ret
= exclude_super_stripes(info
, cache
);
10206 * We may have excluded something, so call this just in
10209 free_excluded_extents(info
, cache
);
10210 btrfs_put_block_group(cache
);
10215 * check for two cases, either we are full, and therefore
10216 * don't need to bother with the caching work since we won't
10217 * find any space, or we are empty, and we can just add all
10218 * the space in and be done with it. This saves us _alot_ of
10219 * time, particularly in the full case.
10221 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
10222 cache
->last_byte_to_unpin
= (u64
)-1;
10223 cache
->cached
= BTRFS_CACHE_FINISHED
;
10224 free_excluded_extents(info
, cache
);
10225 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
10226 cache
->last_byte_to_unpin
= (u64
)-1;
10227 cache
->cached
= BTRFS_CACHE_FINISHED
;
10228 add_new_free_space(cache
, info
,
10229 found_key
.objectid
,
10230 found_key
.objectid
+
10232 free_excluded_extents(info
, cache
);
10235 ret
= btrfs_add_block_group_cache(info
, cache
);
10237 btrfs_remove_free_space_cache(cache
);
10238 btrfs_put_block_group(cache
);
10242 trace_btrfs_add_block_group(info
, cache
, 0);
10243 update_space_info(info
, cache
->flags
, found_key
.offset
,
10244 btrfs_block_group_used(&cache
->item
),
10245 cache
->bytes_super
, &space_info
);
10247 cache
->space_info
= space_info
;
10249 link_block_group(cache
);
10251 set_avail_alloc_bits(info
, cache
->flags
);
10252 if (btrfs_chunk_readonly(info
, cache
->key
.objectid
)) {
10253 inc_block_group_ro(cache
, 1);
10254 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
10255 spin_lock(&info
->unused_bgs_lock
);
10256 /* Should always be true but just in case. */
10257 if (list_empty(&cache
->bg_list
)) {
10258 btrfs_get_block_group(cache
);
10259 list_add_tail(&cache
->bg_list
,
10260 &info
->unused_bgs
);
10262 spin_unlock(&info
->unused_bgs_lock
);
10266 list_for_each_entry_rcu(space_info
, &info
->space_info
, list
) {
10267 if (!(get_alloc_profile(info
, space_info
->flags
) &
10268 (BTRFS_BLOCK_GROUP_RAID10
|
10269 BTRFS_BLOCK_GROUP_RAID1
|
10270 BTRFS_BLOCK_GROUP_RAID5
|
10271 BTRFS_BLOCK_GROUP_RAID6
|
10272 BTRFS_BLOCK_GROUP_DUP
)))
10275 * avoid allocating from un-mirrored block group if there are
10276 * mirrored block groups.
10278 list_for_each_entry(cache
,
10279 &space_info
->block_groups
[BTRFS_RAID_RAID0
],
10281 inc_block_group_ro(cache
, 1);
10282 list_for_each_entry(cache
,
10283 &space_info
->block_groups
[BTRFS_RAID_SINGLE
],
10285 inc_block_group_ro(cache
, 1);
10288 init_global_block_rsv(info
);
10289 ret
= check_chunk_block_group_mappings(info
);
10291 btrfs_free_path(path
);
10295 void btrfs_create_pending_block_groups(struct btrfs_trans_handle
*trans
,
10296 struct btrfs_fs_info
*fs_info
)
10298 struct btrfs_block_group_cache
*block_group
;
10299 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
10300 struct btrfs_block_group_item item
;
10301 struct btrfs_key key
;
10304 if (!trans
->can_flush_pending_bgs
)
10307 while (!list_empty(&trans
->new_bgs
)) {
10308 block_group
= list_first_entry(&trans
->new_bgs
,
10309 struct btrfs_block_group_cache
,
10314 spin_lock(&block_group
->lock
);
10315 memcpy(&item
, &block_group
->item
, sizeof(item
));
10316 memcpy(&key
, &block_group
->key
, sizeof(key
));
10317 spin_unlock(&block_group
->lock
);
10319 ret
= btrfs_insert_item(trans
, extent_root
, &key
, &item
,
10322 btrfs_abort_transaction(trans
, ret
);
10323 ret
= btrfs_finish_chunk_alloc(trans
, fs_info
, key
.objectid
,
10326 btrfs_abort_transaction(trans
, ret
);
10327 add_block_group_free_space(trans
, fs_info
, block_group
);
10328 /* already aborted the transaction if it failed. */
10330 list_del_init(&block_group
->bg_list
);
10332 btrfs_trans_release_chunk_metadata(trans
);
10335 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
10336 struct btrfs_fs_info
*fs_info
, u64 bytes_used
,
10337 u64 type
, u64 chunk_offset
, u64 size
)
10339 struct btrfs_block_group_cache
*cache
;
10342 btrfs_set_log_full_commit(fs_info
, trans
);
10344 cache
= btrfs_create_block_group_cache(fs_info
, chunk_offset
, size
);
10348 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
10349 btrfs_set_block_group_chunk_objectid(&cache
->item
,
10350 BTRFS_FIRST_CHUNK_TREE_OBJECTID
);
10351 btrfs_set_block_group_flags(&cache
->item
, type
);
10353 cache
->flags
= type
;
10354 cache
->last_byte_to_unpin
= (u64
)-1;
10355 cache
->cached
= BTRFS_CACHE_FINISHED
;
10356 cache
->needs_free_space
= 1;
10357 ret
= exclude_super_stripes(fs_info
, cache
);
10360 * We may have excluded something, so call this just in
10363 free_excluded_extents(fs_info
, cache
);
10364 btrfs_put_block_group(cache
);
10368 add_new_free_space(cache
, fs_info
, chunk_offset
, chunk_offset
+ size
);
10370 free_excluded_extents(fs_info
, cache
);
10372 #ifdef CONFIG_BTRFS_DEBUG
10373 if (btrfs_should_fragment_free_space(cache
)) {
10374 u64 new_bytes_used
= size
- bytes_used
;
10376 bytes_used
+= new_bytes_used
>> 1;
10377 fragment_free_space(cache
);
10381 * Ensure the corresponding space_info object is created and
10382 * assigned to our block group. We want our bg to be added to the rbtree
10383 * with its ->space_info set.
10385 cache
->space_info
= __find_space_info(fs_info
, cache
->flags
);
10386 if (!cache
->space_info
) {
10387 ret
= create_space_info(fs_info
, cache
->flags
,
10388 &cache
->space_info
);
10390 btrfs_remove_free_space_cache(cache
);
10391 btrfs_put_block_group(cache
);
10396 ret
= btrfs_add_block_group_cache(fs_info
, cache
);
10398 btrfs_remove_free_space_cache(cache
);
10399 btrfs_put_block_group(cache
);
10404 * Now that our block group has its ->space_info set and is inserted in
10405 * the rbtree, update the space info's counters.
10407 trace_btrfs_add_block_group(fs_info
, cache
, 1);
10408 update_space_info(fs_info
, cache
->flags
, size
, bytes_used
,
10409 cache
->bytes_super
, &cache
->space_info
);
10410 update_global_block_rsv(fs_info
);
10412 link_block_group(cache
);
10414 list_add_tail(&cache
->bg_list
, &trans
->new_bgs
);
10416 set_avail_alloc_bits(fs_info
, type
);
10420 static void clear_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
10422 u64 extra_flags
= chunk_to_extended(flags
) &
10423 BTRFS_EXTENDED_PROFILE_MASK
;
10425 write_seqlock(&fs_info
->profiles_lock
);
10426 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
10427 fs_info
->avail_data_alloc_bits
&= ~extra_flags
;
10428 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
10429 fs_info
->avail_metadata_alloc_bits
&= ~extra_flags
;
10430 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
10431 fs_info
->avail_system_alloc_bits
&= ~extra_flags
;
10432 write_sequnlock(&fs_info
->profiles_lock
);
10435 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
10436 struct btrfs_fs_info
*fs_info
, u64 group_start
,
10437 struct extent_map
*em
)
10439 struct btrfs_root
*root
= fs_info
->extent_root
;
10440 struct btrfs_path
*path
;
10441 struct btrfs_block_group_cache
*block_group
;
10442 struct btrfs_free_cluster
*cluster
;
10443 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
10444 struct btrfs_key key
;
10445 struct inode
*inode
;
10446 struct kobject
*kobj
= NULL
;
10450 struct btrfs_caching_control
*caching_ctl
= NULL
;
10453 block_group
= btrfs_lookup_block_group(fs_info
, group_start
);
10454 BUG_ON(!block_group
);
10455 BUG_ON(!block_group
->ro
);
10458 * Free the reserved super bytes from this block group before
10461 free_excluded_extents(fs_info
, block_group
);
10462 btrfs_free_ref_tree_range(fs_info
, block_group
->key
.objectid
,
10463 block_group
->key
.offset
);
10465 memcpy(&key
, &block_group
->key
, sizeof(key
));
10466 index
= get_block_group_index(block_group
);
10467 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
10468 BTRFS_BLOCK_GROUP_RAID1
|
10469 BTRFS_BLOCK_GROUP_RAID10
))
10474 /* make sure this block group isn't part of an allocation cluster */
10475 cluster
= &fs_info
->data_alloc_cluster
;
10476 spin_lock(&cluster
->refill_lock
);
10477 btrfs_return_cluster_to_free_space(block_group
, cluster
);
10478 spin_unlock(&cluster
->refill_lock
);
10481 * make sure this block group isn't part of a metadata
10482 * allocation cluster
10484 cluster
= &fs_info
->meta_alloc_cluster
;
10485 spin_lock(&cluster
->refill_lock
);
10486 btrfs_return_cluster_to_free_space(block_group
, cluster
);
10487 spin_unlock(&cluster
->refill_lock
);
10489 path
= btrfs_alloc_path();
10496 * get the inode first so any iput calls done for the io_list
10497 * aren't the final iput (no unlinks allowed now)
10499 inode
= lookup_free_space_inode(fs_info
, block_group
, path
);
10501 mutex_lock(&trans
->transaction
->cache_write_mutex
);
10503 * make sure our free spache cache IO is done before remove the
10506 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
10507 if (!list_empty(&block_group
->io_list
)) {
10508 list_del_init(&block_group
->io_list
);
10510 WARN_ON(!IS_ERR(inode
) && inode
!= block_group
->io_ctl
.inode
);
10512 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
10513 btrfs_wait_cache_io(trans
, block_group
, path
);
10514 btrfs_put_block_group(block_group
);
10515 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
10518 if (!list_empty(&block_group
->dirty_list
)) {
10519 list_del_init(&block_group
->dirty_list
);
10520 btrfs_put_block_group(block_group
);
10522 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
10523 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
10525 if (!IS_ERR(inode
)) {
10526 ret
= btrfs_orphan_add(trans
, BTRFS_I(inode
));
10528 btrfs_add_delayed_iput(inode
);
10531 clear_nlink(inode
);
10532 /* One for the block groups ref */
10533 spin_lock(&block_group
->lock
);
10534 if (block_group
->iref
) {
10535 block_group
->iref
= 0;
10536 block_group
->inode
= NULL
;
10537 spin_unlock(&block_group
->lock
);
10540 spin_unlock(&block_group
->lock
);
10542 /* One for our lookup ref */
10543 btrfs_add_delayed_iput(inode
);
10546 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
10547 key
.offset
= block_group
->key
.objectid
;
10550 ret
= btrfs_search_slot(trans
, tree_root
, &key
, path
, -1, 1);
10554 btrfs_release_path(path
);
10556 ret
= btrfs_del_item(trans
, tree_root
, path
);
10559 btrfs_release_path(path
);
10562 spin_lock(&fs_info
->block_group_cache_lock
);
10563 rb_erase(&block_group
->cache_node
,
10564 &fs_info
->block_group_cache_tree
);
10565 RB_CLEAR_NODE(&block_group
->cache_node
);
10567 if (fs_info
->first_logical_byte
== block_group
->key
.objectid
)
10568 fs_info
->first_logical_byte
= (u64
)-1;
10569 spin_unlock(&fs_info
->block_group_cache_lock
);
10571 down_write(&block_group
->space_info
->groups_sem
);
10573 * we must use list_del_init so people can check to see if they
10574 * are still on the list after taking the semaphore
10576 list_del_init(&block_group
->list
);
10577 if (list_empty(&block_group
->space_info
->block_groups
[index
])) {
10578 kobj
= block_group
->space_info
->block_group_kobjs
[index
];
10579 block_group
->space_info
->block_group_kobjs
[index
] = NULL
;
10580 clear_avail_alloc_bits(fs_info
, block_group
->flags
);
10582 up_write(&block_group
->space_info
->groups_sem
);
10588 if (block_group
->has_caching_ctl
)
10589 caching_ctl
= get_caching_control(block_group
);
10590 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
10591 wait_block_group_cache_done(block_group
);
10592 if (block_group
->has_caching_ctl
) {
10593 down_write(&fs_info
->commit_root_sem
);
10594 if (!caching_ctl
) {
10595 struct btrfs_caching_control
*ctl
;
10597 list_for_each_entry(ctl
,
10598 &fs_info
->caching_block_groups
, list
)
10599 if (ctl
->block_group
== block_group
) {
10601 refcount_inc(&caching_ctl
->count
);
10606 list_del_init(&caching_ctl
->list
);
10607 up_write(&fs_info
->commit_root_sem
);
10609 /* Once for the caching bgs list and once for us. */
10610 put_caching_control(caching_ctl
);
10611 put_caching_control(caching_ctl
);
10615 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
10616 if (!list_empty(&block_group
->dirty_list
)) {
10619 if (!list_empty(&block_group
->io_list
)) {
10622 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
10623 btrfs_remove_free_space_cache(block_group
);
10625 spin_lock(&block_group
->space_info
->lock
);
10626 list_del_init(&block_group
->ro_list
);
10628 if (btrfs_test_opt(fs_info
, ENOSPC_DEBUG
)) {
10629 WARN_ON(block_group
->space_info
->total_bytes
10630 < block_group
->key
.offset
);
10631 WARN_ON(block_group
->space_info
->bytes_readonly
10632 < block_group
->key
.offset
);
10633 WARN_ON(block_group
->space_info
->disk_total
10634 < block_group
->key
.offset
* factor
);
10636 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
10637 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
10638 block_group
->space_info
->disk_total
-= block_group
->key
.offset
* factor
;
10640 spin_unlock(&block_group
->space_info
->lock
);
10642 memcpy(&key
, &block_group
->key
, sizeof(key
));
10644 mutex_lock(&fs_info
->chunk_mutex
);
10645 if (!list_empty(&em
->list
)) {
10646 /* We're in the transaction->pending_chunks list. */
10647 free_extent_map(em
);
10649 spin_lock(&block_group
->lock
);
10650 block_group
->removed
= 1;
10652 * At this point trimming can't start on this block group, because we
10653 * removed the block group from the tree fs_info->block_group_cache_tree
10654 * so no one can't find it anymore and even if someone already got this
10655 * block group before we removed it from the rbtree, they have already
10656 * incremented block_group->trimming - if they didn't, they won't find
10657 * any free space entries because we already removed them all when we
10658 * called btrfs_remove_free_space_cache().
10660 * And we must not remove the extent map from the fs_info->mapping_tree
10661 * to prevent the same logical address range and physical device space
10662 * ranges from being reused for a new block group. This is because our
10663 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10664 * completely transactionless, so while it is trimming a range the
10665 * currently running transaction might finish and a new one start,
10666 * allowing for new block groups to be created that can reuse the same
10667 * physical device locations unless we take this special care.
10669 * There may also be an implicit trim operation if the file system
10670 * is mounted with -odiscard. The same protections must remain
10671 * in place until the extents have been discarded completely when
10672 * the transaction commit has completed.
10674 remove_em
= (atomic_read(&block_group
->trimming
) == 0);
10676 * Make sure a trimmer task always sees the em in the pinned_chunks list
10677 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10678 * before checking block_group->removed).
10682 * Our em might be in trans->transaction->pending_chunks which
10683 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10684 * and so is the fs_info->pinned_chunks list.
10686 * So at this point we must be holding the chunk_mutex to avoid
10687 * any races with chunk allocation (more specifically at
10688 * volumes.c:contains_pending_extent()), to ensure it always
10689 * sees the em, either in the pending_chunks list or in the
10690 * pinned_chunks list.
10692 list_move_tail(&em
->list
, &fs_info
->pinned_chunks
);
10694 spin_unlock(&block_group
->lock
);
10696 mutex_unlock(&fs_info
->chunk_mutex
);
10698 ret
= remove_block_group_free_space(trans
, fs_info
, block_group
);
10702 btrfs_put_block_group(block_group
);
10703 btrfs_put_block_group(block_group
);
10705 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
10711 ret
= btrfs_del_item(trans
, root
, path
);
10716 struct extent_map_tree
*em_tree
;
10718 em_tree
= &fs_info
->mapping_tree
.map_tree
;
10719 write_lock(&em_tree
->lock
);
10721 * The em might be in the pending_chunks list, so make sure the
10722 * chunk mutex is locked, since remove_extent_mapping() will
10723 * delete us from that list.
10725 remove_extent_mapping(em_tree
, em
);
10726 write_unlock(&em_tree
->lock
);
10727 /* once for the tree */
10728 free_extent_map(em
);
10731 btrfs_free_path(path
);
10735 struct btrfs_trans_handle
*
10736 btrfs_start_trans_remove_block_group(struct btrfs_fs_info
*fs_info
,
10737 const u64 chunk_offset
)
10739 struct extent_map_tree
*em_tree
= &fs_info
->mapping_tree
.map_tree
;
10740 struct extent_map
*em
;
10741 struct map_lookup
*map
;
10742 unsigned int num_items
;
10744 read_lock(&em_tree
->lock
);
10745 em
= lookup_extent_mapping(em_tree
, chunk_offset
, 1);
10746 read_unlock(&em_tree
->lock
);
10747 ASSERT(em
&& em
->start
== chunk_offset
);
10750 * We need to reserve 3 + N units from the metadata space info in order
10751 * to remove a block group (done at btrfs_remove_chunk() and at
10752 * btrfs_remove_block_group()), which are used for:
10754 * 1 unit for adding the free space inode's orphan (located in the tree
10756 * 1 unit for deleting the block group item (located in the extent
10758 * 1 unit for deleting the free space item (located in tree of tree
10760 * N units for deleting N device extent items corresponding to each
10761 * stripe (located in the device tree).
10763 * In order to remove a block group we also need to reserve units in the
10764 * system space info in order to update the chunk tree (update one or
10765 * more device items and remove one chunk item), but this is done at
10766 * btrfs_remove_chunk() through a call to check_system_chunk().
10768 map
= em
->map_lookup
;
10769 num_items
= 3 + map
->num_stripes
;
10770 free_extent_map(em
);
10772 return btrfs_start_transaction_fallback_global_rsv(fs_info
->extent_root
,
10777 * Process the unused_bgs list and remove any that don't have any allocated
10778 * space inside of them.
10780 void btrfs_delete_unused_bgs(struct btrfs_fs_info
*fs_info
)
10782 struct btrfs_block_group_cache
*block_group
;
10783 struct btrfs_space_info
*space_info
;
10784 struct btrfs_trans_handle
*trans
;
10787 if (!test_bit(BTRFS_FS_OPEN
, &fs_info
->flags
))
10790 spin_lock(&fs_info
->unused_bgs_lock
);
10791 while (!list_empty(&fs_info
->unused_bgs
)) {
10795 block_group
= list_first_entry(&fs_info
->unused_bgs
,
10796 struct btrfs_block_group_cache
,
10798 list_del_init(&block_group
->bg_list
);
10800 space_info
= block_group
->space_info
;
10802 if (ret
|| btrfs_mixed_space_info(space_info
)) {
10803 btrfs_put_block_group(block_group
);
10806 spin_unlock(&fs_info
->unused_bgs_lock
);
10808 mutex_lock(&fs_info
->delete_unused_bgs_mutex
);
10810 /* Don't want to race with allocators so take the groups_sem */
10811 down_write(&space_info
->groups_sem
);
10812 spin_lock(&block_group
->lock
);
10813 if (block_group
->reserved
|| block_group
->pinned
||
10814 btrfs_block_group_used(&block_group
->item
) ||
10816 list_is_singular(&block_group
->list
)) {
10818 * We want to bail if we made new allocations or have
10819 * outstanding allocations in this block group. We do
10820 * the ro check in case balance is currently acting on
10821 * this block group.
10823 spin_unlock(&block_group
->lock
);
10824 up_write(&space_info
->groups_sem
);
10827 spin_unlock(&block_group
->lock
);
10829 /* We don't want to force the issue, only flip if it's ok. */
10830 ret
= inc_block_group_ro(block_group
, 0);
10831 up_write(&space_info
->groups_sem
);
10838 * Want to do this before we do anything else so we can recover
10839 * properly if we fail to join the transaction.
10841 trans
= btrfs_start_trans_remove_block_group(fs_info
,
10842 block_group
->key
.objectid
);
10843 if (IS_ERR(trans
)) {
10844 btrfs_dec_block_group_ro(block_group
);
10845 ret
= PTR_ERR(trans
);
10850 * We could have pending pinned extents for this block group,
10851 * just delete them, we don't care about them anymore.
10853 start
= block_group
->key
.objectid
;
10854 end
= start
+ block_group
->key
.offset
- 1;
10856 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10857 * btrfs_finish_extent_commit(). If we are at transaction N,
10858 * another task might be running finish_extent_commit() for the
10859 * previous transaction N - 1, and have seen a range belonging
10860 * to the block group in freed_extents[] before we were able to
10861 * clear the whole block group range from freed_extents[]. This
10862 * means that task can lookup for the block group after we
10863 * unpinned it from freed_extents[] and removed it, leading to
10864 * a BUG_ON() at btrfs_unpin_extent_range().
10866 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
10867 ret
= clear_extent_bits(&fs_info
->freed_extents
[0], start
, end
,
10870 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10871 btrfs_dec_block_group_ro(block_group
);
10874 ret
= clear_extent_bits(&fs_info
->freed_extents
[1], start
, end
,
10877 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10878 btrfs_dec_block_group_ro(block_group
);
10881 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10883 /* Reset pinned so btrfs_put_block_group doesn't complain */
10884 spin_lock(&space_info
->lock
);
10885 spin_lock(&block_group
->lock
);
10887 space_info
->bytes_pinned
-= block_group
->pinned
;
10888 space_info
->bytes_readonly
+= block_group
->pinned
;
10889 percpu_counter_add(&space_info
->total_bytes_pinned
,
10890 -block_group
->pinned
);
10891 block_group
->pinned
= 0;
10893 spin_unlock(&block_group
->lock
);
10894 spin_unlock(&space_info
->lock
);
10896 /* DISCARD can flip during remount */
10897 trimming
= btrfs_test_opt(fs_info
, DISCARD
);
10899 /* Implicit trim during transaction commit. */
10901 btrfs_get_block_group_trimming(block_group
);
10904 * Btrfs_remove_chunk will abort the transaction if things go
10907 ret
= btrfs_remove_chunk(trans
, fs_info
,
10908 block_group
->key
.objectid
);
10912 btrfs_put_block_group_trimming(block_group
);
10917 * If we're not mounted with -odiscard, we can just forget
10918 * about this block group. Otherwise we'll need to wait
10919 * until transaction commit to do the actual discard.
10922 spin_lock(&fs_info
->unused_bgs_lock
);
10924 * A concurrent scrub might have added us to the list
10925 * fs_info->unused_bgs, so use a list_move operation
10926 * to add the block group to the deleted_bgs list.
10928 list_move(&block_group
->bg_list
,
10929 &trans
->transaction
->deleted_bgs
);
10930 spin_unlock(&fs_info
->unused_bgs_lock
);
10931 btrfs_get_block_group(block_group
);
10934 btrfs_end_transaction(trans
);
10936 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
10937 btrfs_put_block_group(block_group
);
10938 spin_lock(&fs_info
->unused_bgs_lock
);
10940 spin_unlock(&fs_info
->unused_bgs_lock
);
10943 int btrfs_init_space_info(struct btrfs_fs_info
*fs_info
)
10945 struct btrfs_space_info
*space_info
;
10946 struct btrfs_super_block
*disk_super
;
10952 disk_super
= fs_info
->super_copy
;
10953 if (!btrfs_super_root(disk_super
))
10956 features
= btrfs_super_incompat_flags(disk_super
);
10957 if (features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
10960 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
10961 ret
= create_space_info(fs_info
, flags
, &space_info
);
10966 flags
= BTRFS_BLOCK_GROUP_METADATA
| BTRFS_BLOCK_GROUP_DATA
;
10967 ret
= create_space_info(fs_info
, flags
, &space_info
);
10969 flags
= BTRFS_BLOCK_GROUP_METADATA
;
10970 ret
= create_space_info(fs_info
, flags
, &space_info
);
10974 flags
= BTRFS_BLOCK_GROUP_DATA
;
10975 ret
= create_space_info(fs_info
, flags
, &space_info
);
10981 int btrfs_error_unpin_extent_range(struct btrfs_fs_info
*fs_info
,
10982 u64 start
, u64 end
)
10984 return unpin_extent_range(fs_info
, start
, end
, false);
10988 * It used to be that old block groups would be left around forever.
10989 * Iterating over them would be enough to trim unused space. Since we
10990 * now automatically remove them, we also need to iterate over unallocated
10993 * We don't want a transaction for this since the discard may take a
10994 * substantial amount of time. We don't require that a transaction be
10995 * running, but we do need to take a running transaction into account
10996 * to ensure that we're not discarding chunks that were released or
10997 * allocated in the current transaction.
10999 * Holding the chunks lock will prevent other threads from allocating
11000 * or releasing chunks, but it won't prevent a running transaction
11001 * from committing and releasing the memory that the pending chunks
11002 * list head uses. For that, we need to take a reference to the
11003 * transaction and hold the commit root sem. We only need to hold
11004 * it while performing the free space search since we have already
11005 * held back allocations.
11007 static int btrfs_trim_free_extents(struct btrfs_device
*device
,
11008 u64 minlen
, u64
*trimmed
)
11010 u64 start
= 0, len
= 0;
11015 /* Discard not supported = nothing to do. */
11016 if (!blk_queue_discard(bdev_get_queue(device
->bdev
)))
11019 /* Not writeable = nothing to do. */
11020 if (!device
->writeable
)
11023 /* No free space = nothing to do. */
11024 if (device
->total_bytes
<= device
->bytes_used
)
11030 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
11031 struct btrfs_transaction
*trans
;
11034 ret
= mutex_lock_interruptible(&fs_info
->chunk_mutex
);
11038 ret
= down_read_killable(&fs_info
->commit_root_sem
);
11040 mutex_unlock(&fs_info
->chunk_mutex
);
11044 spin_lock(&fs_info
->trans_lock
);
11045 trans
= fs_info
->running_transaction
;
11047 refcount_inc(&trans
->use_count
);
11048 spin_unlock(&fs_info
->trans_lock
);
11051 up_read(&fs_info
->commit_root_sem
);
11053 ret
= find_free_dev_extent_start(trans
, device
, minlen
, start
,
11056 up_read(&fs_info
->commit_root_sem
);
11057 btrfs_put_transaction(trans
);
11061 mutex_unlock(&fs_info
->chunk_mutex
);
11062 if (ret
== -ENOSPC
)
11067 ret
= btrfs_issue_discard(device
->bdev
, start
, len
, &bytes
);
11068 mutex_unlock(&fs_info
->chunk_mutex
);
11076 if (fatal_signal_pending(current
)) {
11077 ret
= -ERESTARTSYS
;
11088 * Trim the whole filesystem by:
11089 * 1) trimming the free space in each block group
11090 * 2) trimming the unallocated space on each device
11092 * This will also continue trimming even if a block group or device encounters
11093 * an error. The return value will be the last error, or 0 if nothing bad
11096 int btrfs_trim_fs(struct btrfs_fs_info
*fs_info
, struct fstrim_range
*range
)
11098 struct btrfs_block_group_cache
*cache
= NULL
;
11099 struct btrfs_device
*device
;
11100 struct list_head
*devices
;
11106 u64 dev_failed
= 0;
11111 cache
= btrfs_lookup_first_block_group(fs_info
, range
->start
);
11112 for (; cache
; cache
= next_block_group(fs_info
, cache
)) {
11113 if (cache
->key
.objectid
>= (range
->start
+ range
->len
)) {
11114 btrfs_put_block_group(cache
);
11118 start
= max(range
->start
, cache
->key
.objectid
);
11119 end
= min(range
->start
+ range
->len
,
11120 cache
->key
.objectid
+ cache
->key
.offset
);
11122 if (end
- start
>= range
->minlen
) {
11123 if (!block_group_cache_done(cache
)) {
11124 ret
= cache_block_group(cache
, 0);
11130 ret
= wait_block_group_cache_done(cache
);
11137 ret
= btrfs_trim_block_group(cache
,
11143 trimmed
+= group_trimmed
;
11153 btrfs_warn(fs_info
,
11154 "failed to trim %llu block group(s), last error %d",
11155 bg_failed
, bg_ret
);
11156 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
11157 devices
= &fs_info
->fs_devices
->devices
;
11158 list_for_each_entry(device
, devices
, dev_list
) {
11159 ret
= btrfs_trim_free_extents(device
, range
->minlen
,
11167 trimmed
+= group_trimmed
;
11169 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
11172 btrfs_warn(fs_info
,
11173 "failed to trim %llu device(s), last error %d",
11174 dev_failed
, dev_ret
);
11175 range
->len
= trimmed
;
11182 * btrfs_{start,end}_write_no_snapshotting() are similar to
11183 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
11184 * data into the page cache through nocow before the subvolume is snapshoted,
11185 * but flush the data into disk after the snapshot creation, or to prevent
11186 * operations while snapshotting is ongoing and that cause the snapshot to be
11187 * inconsistent (writes followed by expanding truncates for example).
11189 void btrfs_end_write_no_snapshotting(struct btrfs_root
*root
)
11191 percpu_counter_dec(&root
->subv_writers
->counter
);
11193 * Make sure counter is updated before we wake up waiters.
11196 if (waitqueue_active(&root
->subv_writers
->wait
))
11197 wake_up(&root
->subv_writers
->wait
);
11200 int btrfs_start_write_no_snapshotting(struct btrfs_root
*root
)
11202 if (atomic_read(&root
->will_be_snapshotted
))
11205 percpu_counter_inc(&root
->subv_writers
->counter
);
11207 * Make sure counter is updated before we check for snapshot creation.
11210 if (atomic_read(&root
->will_be_snapshotted
)) {
11211 btrfs_end_write_no_snapshotting(root
);
11217 void btrfs_wait_for_snapshot_creation(struct btrfs_root
*root
)
11222 ret
= btrfs_start_write_no_snapshotting(root
);
11225 wait_on_atomic_t(&root
->will_be_snapshotted
, atomic_t_wait
,
11226 TASK_UNINTERRUPTIBLE
);