2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/sched/signal.h>
20 #include <linux/pagemap.h>
21 #include <linux/writeback.h>
22 #include <linux/blkdev.h>
23 #include <linux/sort.h>
24 #include <linux/rcupdate.h>
25 #include <linux/kthread.h>
26 #include <linux/slab.h>
27 #include <linux/ratelimit.h>
28 #include <linux/percpu_counter.h>
32 #include "print-tree.h"
36 #include "free-space-cache.h"
37 #include "free-space-tree.h"
42 #undef SCRAMBLE_DELAYED_REFS
45 * control flags for do_chunk_alloc's force field
46 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
47 * if we really need one.
49 * CHUNK_ALLOC_LIMITED means to only try and allocate one
50 * if we have very few chunks already allocated. This is
51 * used as part of the clustering code to help make sure
52 * we have a good pool of storage to cluster in, without
53 * filling the FS with empty chunks
55 * CHUNK_ALLOC_FORCE means it must try to allocate one
59 CHUNK_ALLOC_NO_FORCE
= 0,
60 CHUNK_ALLOC_LIMITED
= 1,
61 CHUNK_ALLOC_FORCE
= 2,
64 static int update_block_group(struct btrfs_trans_handle
*trans
,
65 struct btrfs_fs_info
*fs_info
, u64 bytenr
,
66 u64 num_bytes
, int alloc
);
67 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
68 struct btrfs_fs_info
*fs_info
,
69 struct btrfs_delayed_ref_node
*node
, u64 parent
,
70 u64 root_objectid
, u64 owner_objectid
,
71 u64 owner_offset
, int refs_to_drop
,
72 struct btrfs_delayed_extent_op
*extra_op
);
73 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
74 struct extent_buffer
*leaf
,
75 struct btrfs_extent_item
*ei
);
76 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
77 struct btrfs_fs_info
*fs_info
,
78 u64 parent
, u64 root_objectid
,
79 u64 flags
, u64 owner
, u64 offset
,
80 struct btrfs_key
*ins
, int ref_mod
);
81 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
82 struct btrfs_fs_info
*fs_info
,
83 u64 parent
, u64 root_objectid
,
84 u64 flags
, struct btrfs_disk_key
*key
,
85 int level
, struct btrfs_key
*ins
);
86 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
87 struct btrfs_fs_info
*fs_info
, u64 flags
,
89 static int find_next_key(struct btrfs_path
*path
, int level
,
90 struct btrfs_key
*key
);
91 static void dump_space_info(struct btrfs_fs_info
*fs_info
,
92 struct btrfs_space_info
*info
, u64 bytes
,
93 int dump_block_groups
);
94 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache
*cache
,
95 u64 ram_bytes
, u64 num_bytes
, int delalloc
);
96 static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache
*cache
,
97 u64 num_bytes
, int delalloc
);
98 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
100 static int __reserve_metadata_bytes(struct btrfs_root
*root
,
101 struct btrfs_space_info
*space_info
,
103 enum btrfs_reserve_flush_enum flush
);
104 static void space_info_add_new_bytes(struct btrfs_fs_info
*fs_info
,
105 struct btrfs_space_info
*space_info
,
107 static void space_info_add_old_bytes(struct btrfs_fs_info
*fs_info
,
108 struct btrfs_space_info
*space_info
,
112 block_group_cache_done(struct btrfs_block_group_cache
*cache
)
115 return cache
->cached
== BTRFS_CACHE_FINISHED
||
116 cache
->cached
== BTRFS_CACHE_ERROR
;
119 static int block_group_bits(struct btrfs_block_group_cache
*cache
, u64 bits
)
121 return (cache
->flags
& bits
) == bits
;
124 void btrfs_get_block_group(struct btrfs_block_group_cache
*cache
)
126 atomic_inc(&cache
->count
);
129 void btrfs_put_block_group(struct btrfs_block_group_cache
*cache
)
131 if (atomic_dec_and_test(&cache
->count
)) {
132 WARN_ON(cache
->pinned
> 0);
133 WARN_ON(cache
->reserved
> 0);
134 kfree(cache
->free_space_ctl
);
140 * this adds the block group to the fs_info rb tree for the block group
143 static int btrfs_add_block_group_cache(struct btrfs_fs_info
*info
,
144 struct btrfs_block_group_cache
*block_group
)
147 struct rb_node
*parent
= NULL
;
148 struct btrfs_block_group_cache
*cache
;
150 spin_lock(&info
->block_group_cache_lock
);
151 p
= &info
->block_group_cache_tree
.rb_node
;
155 cache
= rb_entry(parent
, struct btrfs_block_group_cache
,
157 if (block_group
->key
.objectid
< cache
->key
.objectid
) {
159 } else if (block_group
->key
.objectid
> cache
->key
.objectid
) {
162 spin_unlock(&info
->block_group_cache_lock
);
167 rb_link_node(&block_group
->cache_node
, parent
, p
);
168 rb_insert_color(&block_group
->cache_node
,
169 &info
->block_group_cache_tree
);
171 if (info
->first_logical_byte
> block_group
->key
.objectid
)
172 info
->first_logical_byte
= block_group
->key
.objectid
;
174 spin_unlock(&info
->block_group_cache_lock
);
180 * This will return the block group at or after bytenr if contains is 0, else
181 * it will return the block group that contains the bytenr
183 static struct btrfs_block_group_cache
*
184 block_group_cache_tree_search(struct btrfs_fs_info
*info
, u64 bytenr
,
187 struct btrfs_block_group_cache
*cache
, *ret
= NULL
;
191 spin_lock(&info
->block_group_cache_lock
);
192 n
= info
->block_group_cache_tree
.rb_node
;
195 cache
= rb_entry(n
, struct btrfs_block_group_cache
,
197 end
= cache
->key
.objectid
+ cache
->key
.offset
- 1;
198 start
= cache
->key
.objectid
;
200 if (bytenr
< start
) {
201 if (!contains
&& (!ret
|| start
< ret
->key
.objectid
))
204 } else if (bytenr
> start
) {
205 if (contains
&& bytenr
<= end
) {
216 btrfs_get_block_group(ret
);
217 if (bytenr
== 0 && info
->first_logical_byte
> ret
->key
.objectid
)
218 info
->first_logical_byte
= ret
->key
.objectid
;
220 spin_unlock(&info
->block_group_cache_lock
);
225 static int add_excluded_extent(struct btrfs_fs_info
*fs_info
,
226 u64 start
, u64 num_bytes
)
228 u64 end
= start
+ num_bytes
- 1;
229 set_extent_bits(&fs_info
->freed_extents
[0],
230 start
, end
, EXTENT_UPTODATE
);
231 set_extent_bits(&fs_info
->freed_extents
[1],
232 start
, end
, EXTENT_UPTODATE
);
236 static void free_excluded_extents(struct btrfs_fs_info
*fs_info
,
237 struct btrfs_block_group_cache
*cache
)
241 start
= cache
->key
.objectid
;
242 end
= start
+ cache
->key
.offset
- 1;
244 clear_extent_bits(&fs_info
->freed_extents
[0],
245 start
, end
, EXTENT_UPTODATE
);
246 clear_extent_bits(&fs_info
->freed_extents
[1],
247 start
, end
, EXTENT_UPTODATE
);
250 static int exclude_super_stripes(struct btrfs_fs_info
*fs_info
,
251 struct btrfs_block_group_cache
*cache
)
258 if (cache
->key
.objectid
< BTRFS_SUPER_INFO_OFFSET
) {
259 stripe_len
= BTRFS_SUPER_INFO_OFFSET
- cache
->key
.objectid
;
260 cache
->bytes_super
+= stripe_len
;
261 ret
= add_excluded_extent(fs_info
, cache
->key
.objectid
,
267 for (i
= 0; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
268 bytenr
= btrfs_sb_offset(i
);
269 ret
= btrfs_rmap_block(fs_info
, cache
->key
.objectid
,
270 bytenr
, 0, &logical
, &nr
, &stripe_len
);
277 if (logical
[nr
] > cache
->key
.objectid
+
281 if (logical
[nr
] + stripe_len
<= cache
->key
.objectid
)
285 if (start
< cache
->key
.objectid
) {
286 start
= cache
->key
.objectid
;
287 len
= (logical
[nr
] + stripe_len
) - start
;
289 len
= min_t(u64
, stripe_len
,
290 cache
->key
.objectid
+
291 cache
->key
.offset
- start
);
294 cache
->bytes_super
+= len
;
295 ret
= add_excluded_extent(fs_info
, start
, len
);
307 static struct btrfs_caching_control
*
308 get_caching_control(struct btrfs_block_group_cache
*cache
)
310 struct btrfs_caching_control
*ctl
;
312 spin_lock(&cache
->lock
);
313 if (!cache
->caching_ctl
) {
314 spin_unlock(&cache
->lock
);
318 ctl
= cache
->caching_ctl
;
319 atomic_inc(&ctl
->count
);
320 spin_unlock(&cache
->lock
);
324 static void put_caching_control(struct btrfs_caching_control
*ctl
)
326 if (atomic_dec_and_test(&ctl
->count
))
330 #ifdef CONFIG_BTRFS_DEBUG
331 static void fragment_free_space(struct btrfs_block_group_cache
*block_group
)
333 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
334 u64 start
= block_group
->key
.objectid
;
335 u64 len
= block_group
->key
.offset
;
336 u64 chunk
= block_group
->flags
& BTRFS_BLOCK_GROUP_METADATA
?
337 fs_info
->nodesize
: fs_info
->sectorsize
;
338 u64 step
= chunk
<< 1;
340 while (len
> chunk
) {
341 btrfs_remove_free_space(block_group
, start
, chunk
);
352 * this is only called by cache_block_group, since we could have freed extents
353 * we need to check the pinned_extents for any extents that can't be used yet
354 * since their free space will be released as soon as the transaction commits.
356 u64
add_new_free_space(struct btrfs_block_group_cache
*block_group
,
357 struct btrfs_fs_info
*info
, u64 start
, u64 end
)
359 u64 extent_start
, extent_end
, size
, total_added
= 0;
362 while (start
< end
) {
363 ret
= find_first_extent_bit(info
->pinned_extents
, start
,
364 &extent_start
, &extent_end
,
365 EXTENT_DIRTY
| EXTENT_UPTODATE
,
370 if (extent_start
<= start
) {
371 start
= extent_end
+ 1;
372 } else if (extent_start
> start
&& extent_start
< end
) {
373 size
= extent_start
- start
;
375 ret
= btrfs_add_free_space(block_group
, start
,
377 BUG_ON(ret
); /* -ENOMEM or logic error */
378 start
= extent_end
+ 1;
387 ret
= btrfs_add_free_space(block_group
, start
, size
);
388 BUG_ON(ret
); /* -ENOMEM or logic error */
394 static int load_extent_tree_free(struct btrfs_caching_control
*caching_ctl
)
396 struct btrfs_block_group_cache
*block_group
= caching_ctl
->block_group
;
397 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
398 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
399 struct btrfs_path
*path
;
400 struct extent_buffer
*leaf
;
401 struct btrfs_key key
;
408 path
= btrfs_alloc_path();
412 last
= max_t(u64
, block_group
->key
.objectid
, BTRFS_SUPER_INFO_OFFSET
);
414 #ifdef CONFIG_BTRFS_DEBUG
416 * If we're fragmenting we don't want to make anybody think we can
417 * allocate from this block group until we've had a chance to fragment
420 if (btrfs_should_fragment_free_space(block_group
))
424 * We don't want to deadlock with somebody trying to allocate a new
425 * extent for the extent root while also trying to search the extent
426 * root to add free space. So we skip locking and search the commit
427 * root, since its read-only
429 path
->skip_locking
= 1;
430 path
->search_commit_root
= 1;
431 path
->reada
= READA_FORWARD
;
435 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
438 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
442 leaf
= path
->nodes
[0];
443 nritems
= btrfs_header_nritems(leaf
);
446 if (btrfs_fs_closing(fs_info
) > 1) {
451 if (path
->slots
[0] < nritems
) {
452 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
454 ret
= find_next_key(path
, 0, &key
);
458 if (need_resched() ||
459 rwsem_is_contended(&fs_info
->commit_root_sem
)) {
461 caching_ctl
->progress
= last
;
462 btrfs_release_path(path
);
463 up_read(&fs_info
->commit_root_sem
);
464 mutex_unlock(&caching_ctl
->mutex
);
466 mutex_lock(&caching_ctl
->mutex
);
467 down_read(&fs_info
->commit_root_sem
);
471 ret
= btrfs_next_leaf(extent_root
, path
);
476 leaf
= path
->nodes
[0];
477 nritems
= btrfs_header_nritems(leaf
);
481 if (key
.objectid
< last
) {
484 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
487 caching_ctl
->progress
= last
;
488 btrfs_release_path(path
);
492 if (key
.objectid
< block_group
->key
.objectid
) {
497 if (key
.objectid
>= block_group
->key
.objectid
+
498 block_group
->key
.offset
)
501 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
||
502 key
.type
== BTRFS_METADATA_ITEM_KEY
) {
503 total_found
+= add_new_free_space(block_group
,
506 if (key
.type
== BTRFS_METADATA_ITEM_KEY
)
507 last
= key
.objectid
+
510 last
= key
.objectid
+ key
.offset
;
512 if (total_found
> CACHING_CTL_WAKE_UP
) {
515 wake_up(&caching_ctl
->wait
);
522 total_found
+= add_new_free_space(block_group
, fs_info
, last
,
523 block_group
->key
.objectid
+
524 block_group
->key
.offset
);
525 caching_ctl
->progress
= (u64
)-1;
528 btrfs_free_path(path
);
532 static noinline
void caching_thread(struct btrfs_work
*work
)
534 struct btrfs_block_group_cache
*block_group
;
535 struct btrfs_fs_info
*fs_info
;
536 struct btrfs_caching_control
*caching_ctl
;
537 struct btrfs_root
*extent_root
;
540 caching_ctl
= container_of(work
, struct btrfs_caching_control
, work
);
541 block_group
= caching_ctl
->block_group
;
542 fs_info
= block_group
->fs_info
;
543 extent_root
= fs_info
->extent_root
;
545 mutex_lock(&caching_ctl
->mutex
);
546 down_read(&fs_info
->commit_root_sem
);
548 if (btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
))
549 ret
= load_free_space_tree(caching_ctl
);
551 ret
= load_extent_tree_free(caching_ctl
);
553 spin_lock(&block_group
->lock
);
554 block_group
->caching_ctl
= NULL
;
555 block_group
->cached
= ret
? BTRFS_CACHE_ERROR
: BTRFS_CACHE_FINISHED
;
556 spin_unlock(&block_group
->lock
);
558 #ifdef CONFIG_BTRFS_DEBUG
559 if (btrfs_should_fragment_free_space(block_group
)) {
562 spin_lock(&block_group
->space_info
->lock
);
563 spin_lock(&block_group
->lock
);
564 bytes_used
= block_group
->key
.offset
-
565 btrfs_block_group_used(&block_group
->item
);
566 block_group
->space_info
->bytes_used
+= bytes_used
>> 1;
567 spin_unlock(&block_group
->lock
);
568 spin_unlock(&block_group
->space_info
->lock
);
569 fragment_free_space(block_group
);
573 caching_ctl
->progress
= (u64
)-1;
575 up_read(&fs_info
->commit_root_sem
);
576 free_excluded_extents(fs_info
, block_group
);
577 mutex_unlock(&caching_ctl
->mutex
);
579 wake_up(&caching_ctl
->wait
);
581 put_caching_control(caching_ctl
);
582 btrfs_put_block_group(block_group
);
585 static int cache_block_group(struct btrfs_block_group_cache
*cache
,
589 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
590 struct btrfs_caching_control
*caching_ctl
;
593 caching_ctl
= kzalloc(sizeof(*caching_ctl
), GFP_NOFS
);
597 INIT_LIST_HEAD(&caching_ctl
->list
);
598 mutex_init(&caching_ctl
->mutex
);
599 init_waitqueue_head(&caching_ctl
->wait
);
600 caching_ctl
->block_group
= cache
;
601 caching_ctl
->progress
= cache
->key
.objectid
;
602 atomic_set(&caching_ctl
->count
, 1);
603 btrfs_init_work(&caching_ctl
->work
, btrfs_cache_helper
,
604 caching_thread
, NULL
, NULL
);
606 spin_lock(&cache
->lock
);
608 * This should be a rare occasion, but this could happen I think in the
609 * case where one thread starts to load the space cache info, and then
610 * some other thread starts a transaction commit which tries to do an
611 * allocation while the other thread is still loading the space cache
612 * info. The previous loop should have kept us from choosing this block
613 * group, but if we've moved to the state where we will wait on caching
614 * block groups we need to first check if we're doing a fast load here,
615 * so we can wait for it to finish, otherwise we could end up allocating
616 * from a block group who's cache gets evicted for one reason or
619 while (cache
->cached
== BTRFS_CACHE_FAST
) {
620 struct btrfs_caching_control
*ctl
;
622 ctl
= cache
->caching_ctl
;
623 atomic_inc(&ctl
->count
);
624 prepare_to_wait(&ctl
->wait
, &wait
, TASK_UNINTERRUPTIBLE
);
625 spin_unlock(&cache
->lock
);
629 finish_wait(&ctl
->wait
, &wait
);
630 put_caching_control(ctl
);
631 spin_lock(&cache
->lock
);
634 if (cache
->cached
!= BTRFS_CACHE_NO
) {
635 spin_unlock(&cache
->lock
);
639 WARN_ON(cache
->caching_ctl
);
640 cache
->caching_ctl
= caching_ctl
;
641 cache
->cached
= BTRFS_CACHE_FAST
;
642 spin_unlock(&cache
->lock
);
644 if (fs_info
->mount_opt
& BTRFS_MOUNT_SPACE_CACHE
) {
645 mutex_lock(&caching_ctl
->mutex
);
646 ret
= load_free_space_cache(fs_info
, cache
);
648 spin_lock(&cache
->lock
);
650 cache
->caching_ctl
= NULL
;
651 cache
->cached
= BTRFS_CACHE_FINISHED
;
652 cache
->last_byte_to_unpin
= (u64
)-1;
653 caching_ctl
->progress
= (u64
)-1;
655 if (load_cache_only
) {
656 cache
->caching_ctl
= NULL
;
657 cache
->cached
= BTRFS_CACHE_NO
;
659 cache
->cached
= BTRFS_CACHE_STARTED
;
660 cache
->has_caching_ctl
= 1;
663 spin_unlock(&cache
->lock
);
664 #ifdef CONFIG_BTRFS_DEBUG
666 btrfs_should_fragment_free_space(cache
)) {
669 spin_lock(&cache
->space_info
->lock
);
670 spin_lock(&cache
->lock
);
671 bytes_used
= cache
->key
.offset
-
672 btrfs_block_group_used(&cache
->item
);
673 cache
->space_info
->bytes_used
+= bytes_used
>> 1;
674 spin_unlock(&cache
->lock
);
675 spin_unlock(&cache
->space_info
->lock
);
676 fragment_free_space(cache
);
679 mutex_unlock(&caching_ctl
->mutex
);
681 wake_up(&caching_ctl
->wait
);
683 put_caching_control(caching_ctl
);
684 free_excluded_extents(fs_info
, cache
);
689 * We're either using the free space tree or no caching at all.
690 * Set cached to the appropriate value and wakeup any waiters.
692 spin_lock(&cache
->lock
);
693 if (load_cache_only
) {
694 cache
->caching_ctl
= NULL
;
695 cache
->cached
= BTRFS_CACHE_NO
;
697 cache
->cached
= BTRFS_CACHE_STARTED
;
698 cache
->has_caching_ctl
= 1;
700 spin_unlock(&cache
->lock
);
701 wake_up(&caching_ctl
->wait
);
704 if (load_cache_only
) {
705 put_caching_control(caching_ctl
);
709 down_write(&fs_info
->commit_root_sem
);
710 atomic_inc(&caching_ctl
->count
);
711 list_add_tail(&caching_ctl
->list
, &fs_info
->caching_block_groups
);
712 up_write(&fs_info
->commit_root_sem
);
714 btrfs_get_block_group(cache
);
716 btrfs_queue_work(fs_info
->caching_workers
, &caching_ctl
->work
);
722 * return the block group that starts at or after bytenr
724 static struct btrfs_block_group_cache
*
725 btrfs_lookup_first_block_group(struct btrfs_fs_info
*info
, u64 bytenr
)
727 return block_group_cache_tree_search(info
, bytenr
, 0);
731 * return the block group that contains the given bytenr
733 struct btrfs_block_group_cache
*btrfs_lookup_block_group(
734 struct btrfs_fs_info
*info
,
737 return block_group_cache_tree_search(info
, bytenr
, 1);
740 static struct btrfs_space_info
*__find_space_info(struct btrfs_fs_info
*info
,
743 struct list_head
*head
= &info
->space_info
;
744 struct btrfs_space_info
*found
;
746 flags
&= BTRFS_BLOCK_GROUP_TYPE_MASK
;
749 list_for_each_entry_rcu(found
, head
, list
) {
750 if (found
->flags
& flags
) {
760 * after adding space to the filesystem, we need to clear the full flags
761 * on all the space infos.
763 void btrfs_clear_space_info_full(struct btrfs_fs_info
*info
)
765 struct list_head
*head
= &info
->space_info
;
766 struct btrfs_space_info
*found
;
769 list_for_each_entry_rcu(found
, head
, list
)
774 /* simple helper to search for an existing data extent at a given offset */
775 int btrfs_lookup_data_extent(struct btrfs_fs_info
*fs_info
, u64 start
, u64 len
)
778 struct btrfs_key key
;
779 struct btrfs_path
*path
;
781 path
= btrfs_alloc_path();
785 key
.objectid
= start
;
787 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
788 ret
= btrfs_search_slot(NULL
, fs_info
->extent_root
, &key
, path
, 0, 0);
789 btrfs_free_path(path
);
794 * helper function to lookup reference count and flags of a tree block.
796 * the head node for delayed ref is used to store the sum of all the
797 * reference count modifications queued up in the rbtree. the head
798 * node may also store the extent flags to set. This way you can check
799 * to see what the reference count and extent flags would be if all of
800 * the delayed refs are not processed.
802 int btrfs_lookup_extent_info(struct btrfs_trans_handle
*trans
,
803 struct btrfs_fs_info
*fs_info
, u64 bytenr
,
804 u64 offset
, int metadata
, u64
*refs
, u64
*flags
)
806 struct btrfs_delayed_ref_head
*head
;
807 struct btrfs_delayed_ref_root
*delayed_refs
;
808 struct btrfs_path
*path
;
809 struct btrfs_extent_item
*ei
;
810 struct extent_buffer
*leaf
;
811 struct btrfs_key key
;
818 * If we don't have skinny metadata, don't bother doing anything
821 if (metadata
&& !btrfs_fs_incompat(fs_info
, SKINNY_METADATA
)) {
822 offset
= fs_info
->nodesize
;
826 path
= btrfs_alloc_path();
831 path
->skip_locking
= 1;
832 path
->search_commit_root
= 1;
836 key
.objectid
= bytenr
;
839 key
.type
= BTRFS_METADATA_ITEM_KEY
;
841 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
843 ret
= btrfs_search_slot(trans
, fs_info
->extent_root
, &key
, path
, 0, 0);
847 if (ret
> 0 && metadata
&& key
.type
== BTRFS_METADATA_ITEM_KEY
) {
848 if (path
->slots
[0]) {
850 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
852 if (key
.objectid
== bytenr
&&
853 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
854 key
.offset
== fs_info
->nodesize
)
860 leaf
= path
->nodes
[0];
861 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
862 if (item_size
>= sizeof(*ei
)) {
863 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
864 struct btrfs_extent_item
);
865 num_refs
= btrfs_extent_refs(leaf
, ei
);
866 extent_flags
= btrfs_extent_flags(leaf
, ei
);
868 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
869 struct btrfs_extent_item_v0
*ei0
;
870 BUG_ON(item_size
!= sizeof(*ei0
));
871 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
872 struct btrfs_extent_item_v0
);
873 num_refs
= btrfs_extent_refs_v0(leaf
, ei0
);
874 /* FIXME: this isn't correct for data */
875 extent_flags
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
880 BUG_ON(num_refs
== 0);
890 delayed_refs
= &trans
->transaction
->delayed_refs
;
891 spin_lock(&delayed_refs
->lock
);
892 head
= btrfs_find_delayed_ref_head(delayed_refs
, bytenr
);
894 if (!mutex_trylock(&head
->mutex
)) {
895 atomic_inc(&head
->node
.refs
);
896 spin_unlock(&delayed_refs
->lock
);
898 btrfs_release_path(path
);
901 * Mutex was contended, block until it's released and try
904 mutex_lock(&head
->mutex
);
905 mutex_unlock(&head
->mutex
);
906 btrfs_put_delayed_ref(&head
->node
);
909 spin_lock(&head
->lock
);
910 if (head
->extent_op
&& head
->extent_op
->update_flags
)
911 extent_flags
|= head
->extent_op
->flags_to_set
;
913 BUG_ON(num_refs
== 0);
915 num_refs
+= head
->node
.ref_mod
;
916 spin_unlock(&head
->lock
);
917 mutex_unlock(&head
->mutex
);
919 spin_unlock(&delayed_refs
->lock
);
921 WARN_ON(num_refs
== 0);
925 *flags
= extent_flags
;
927 btrfs_free_path(path
);
932 * Back reference rules. Back refs have three main goals:
934 * 1) differentiate between all holders of references to an extent so that
935 * when a reference is dropped we can make sure it was a valid reference
936 * before freeing the extent.
938 * 2) Provide enough information to quickly find the holders of an extent
939 * if we notice a given block is corrupted or bad.
941 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
942 * maintenance. This is actually the same as #2, but with a slightly
943 * different use case.
945 * There are two kinds of back refs. The implicit back refs is optimized
946 * for pointers in non-shared tree blocks. For a given pointer in a block,
947 * back refs of this kind provide information about the block's owner tree
948 * and the pointer's key. These information allow us to find the block by
949 * b-tree searching. The full back refs is for pointers in tree blocks not
950 * referenced by their owner trees. The location of tree block is recorded
951 * in the back refs. Actually the full back refs is generic, and can be
952 * used in all cases the implicit back refs is used. The major shortcoming
953 * of the full back refs is its overhead. Every time a tree block gets
954 * COWed, we have to update back refs entry for all pointers in it.
956 * For a newly allocated tree block, we use implicit back refs for
957 * pointers in it. This means most tree related operations only involve
958 * implicit back refs. For a tree block created in old transaction, the
959 * only way to drop a reference to it is COW it. So we can detect the
960 * event that tree block loses its owner tree's reference and do the
961 * back refs conversion.
963 * When a tree block is COWed through a tree, there are four cases:
965 * The reference count of the block is one and the tree is the block's
966 * owner tree. Nothing to do in this case.
968 * The reference count of the block is one and the tree is not the
969 * block's owner tree. In this case, full back refs is used for pointers
970 * in the block. Remove these full back refs, add implicit back refs for
971 * every pointers in the new block.
973 * The reference count of the block is greater than one and the tree is
974 * the block's owner tree. In this case, implicit back refs is used for
975 * pointers in the block. Add full back refs for every pointers in the
976 * block, increase lower level extents' reference counts. The original
977 * implicit back refs are entailed to the new block.
979 * The reference count of the block is greater than one and the tree is
980 * not the block's owner tree. Add implicit back refs for every pointer in
981 * the new block, increase lower level extents' reference count.
983 * Back Reference Key composing:
985 * The key objectid corresponds to the first byte in the extent,
986 * The key type is used to differentiate between types of back refs.
987 * There are different meanings of the key offset for different types
990 * File extents can be referenced by:
992 * - multiple snapshots, subvolumes, or different generations in one subvol
993 * - different files inside a single subvolume
994 * - different offsets inside a file (bookend extents in file.c)
996 * The extent ref structure for the implicit back refs has fields for:
998 * - Objectid of the subvolume root
999 * - objectid of the file holding the reference
1000 * - original offset in the file
1001 * - how many bookend extents
1003 * The key offset for the implicit back refs is hash of the first
1006 * The extent ref structure for the full back refs has field for:
1008 * - number of pointers in the tree leaf
1010 * The key offset for the implicit back refs is the first byte of
1013 * When a file extent is allocated, The implicit back refs is used.
1014 * the fields are filled in:
1016 * (root_key.objectid, inode objectid, offset in file, 1)
1018 * When a file extent is removed file truncation, we find the
1019 * corresponding implicit back refs and check the following fields:
1021 * (btrfs_header_owner(leaf), inode objectid, offset in file)
1023 * Btree extents can be referenced by:
1025 * - Different subvolumes
1027 * Both the implicit back refs and the full back refs for tree blocks
1028 * only consist of key. The key offset for the implicit back refs is
1029 * objectid of block's owner tree. The key offset for the full back refs
1030 * is the first byte of parent block.
1032 * When implicit back refs is used, information about the lowest key and
1033 * level of the tree block are required. These information are stored in
1034 * tree block info structure.
1037 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1038 static int convert_extent_item_v0(struct btrfs_trans_handle
*trans
,
1039 struct btrfs_fs_info
*fs_info
,
1040 struct btrfs_path
*path
,
1041 u64 owner
, u32 extra_size
)
1043 struct btrfs_root
*root
= fs_info
->extent_root
;
1044 struct btrfs_extent_item
*item
;
1045 struct btrfs_extent_item_v0
*ei0
;
1046 struct btrfs_extent_ref_v0
*ref0
;
1047 struct btrfs_tree_block_info
*bi
;
1048 struct extent_buffer
*leaf
;
1049 struct btrfs_key key
;
1050 struct btrfs_key found_key
;
1051 u32 new_size
= sizeof(*item
);
1055 leaf
= path
->nodes
[0];
1056 BUG_ON(btrfs_item_size_nr(leaf
, path
->slots
[0]) != sizeof(*ei0
));
1058 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1059 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1060 struct btrfs_extent_item_v0
);
1061 refs
= btrfs_extent_refs_v0(leaf
, ei0
);
1063 if (owner
== (u64
)-1) {
1065 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1066 ret
= btrfs_next_leaf(root
, path
);
1069 BUG_ON(ret
> 0); /* Corruption */
1070 leaf
= path
->nodes
[0];
1072 btrfs_item_key_to_cpu(leaf
, &found_key
,
1074 BUG_ON(key
.objectid
!= found_key
.objectid
);
1075 if (found_key
.type
!= BTRFS_EXTENT_REF_V0_KEY
) {
1079 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1080 struct btrfs_extent_ref_v0
);
1081 owner
= btrfs_ref_objectid_v0(leaf
, ref0
);
1085 btrfs_release_path(path
);
1087 if (owner
< BTRFS_FIRST_FREE_OBJECTID
)
1088 new_size
+= sizeof(*bi
);
1090 new_size
-= sizeof(*ei0
);
1091 ret
= btrfs_search_slot(trans
, root
, &key
, path
,
1092 new_size
+ extra_size
, 1);
1095 BUG_ON(ret
); /* Corruption */
1097 btrfs_extend_item(fs_info
, path
, new_size
);
1099 leaf
= path
->nodes
[0];
1100 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1101 btrfs_set_extent_refs(leaf
, item
, refs
);
1102 /* FIXME: get real generation */
1103 btrfs_set_extent_generation(leaf
, item
, 0);
1104 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1105 btrfs_set_extent_flags(leaf
, item
,
1106 BTRFS_EXTENT_FLAG_TREE_BLOCK
|
1107 BTRFS_BLOCK_FLAG_FULL_BACKREF
);
1108 bi
= (struct btrfs_tree_block_info
*)(item
+ 1);
1109 /* FIXME: get first key of the block */
1110 memzero_extent_buffer(leaf
, (unsigned long)bi
, sizeof(*bi
));
1111 btrfs_set_tree_block_level(leaf
, bi
, (int)owner
);
1113 btrfs_set_extent_flags(leaf
, item
, BTRFS_EXTENT_FLAG_DATA
);
1115 btrfs_mark_buffer_dirty(leaf
);
1120 static u64
hash_extent_data_ref(u64 root_objectid
, u64 owner
, u64 offset
)
1122 u32 high_crc
= ~(u32
)0;
1123 u32 low_crc
= ~(u32
)0;
1126 lenum
= cpu_to_le64(root_objectid
);
1127 high_crc
= btrfs_crc32c(high_crc
, &lenum
, sizeof(lenum
));
1128 lenum
= cpu_to_le64(owner
);
1129 low_crc
= btrfs_crc32c(low_crc
, &lenum
, sizeof(lenum
));
1130 lenum
= cpu_to_le64(offset
);
1131 low_crc
= btrfs_crc32c(low_crc
, &lenum
, sizeof(lenum
));
1133 return ((u64
)high_crc
<< 31) ^ (u64
)low_crc
;
1136 static u64
hash_extent_data_ref_item(struct extent_buffer
*leaf
,
1137 struct btrfs_extent_data_ref
*ref
)
1139 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf
, ref
),
1140 btrfs_extent_data_ref_objectid(leaf
, ref
),
1141 btrfs_extent_data_ref_offset(leaf
, ref
));
1144 static int match_extent_data_ref(struct extent_buffer
*leaf
,
1145 struct btrfs_extent_data_ref
*ref
,
1146 u64 root_objectid
, u64 owner
, u64 offset
)
1148 if (btrfs_extent_data_ref_root(leaf
, ref
) != root_objectid
||
1149 btrfs_extent_data_ref_objectid(leaf
, ref
) != owner
||
1150 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
1155 static noinline
int lookup_extent_data_ref(struct btrfs_trans_handle
*trans
,
1156 struct btrfs_fs_info
*fs_info
,
1157 struct btrfs_path
*path
,
1158 u64 bytenr
, u64 parent
,
1160 u64 owner
, u64 offset
)
1162 struct btrfs_root
*root
= fs_info
->extent_root
;
1163 struct btrfs_key key
;
1164 struct btrfs_extent_data_ref
*ref
;
1165 struct extent_buffer
*leaf
;
1171 key
.objectid
= bytenr
;
1173 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1174 key
.offset
= parent
;
1176 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1177 key
.offset
= hash_extent_data_ref(root_objectid
,
1182 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1191 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1192 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1193 btrfs_release_path(path
);
1194 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1205 leaf
= path
->nodes
[0];
1206 nritems
= btrfs_header_nritems(leaf
);
1208 if (path
->slots
[0] >= nritems
) {
1209 ret
= btrfs_next_leaf(root
, path
);
1215 leaf
= path
->nodes
[0];
1216 nritems
= btrfs_header_nritems(leaf
);
1220 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1221 if (key
.objectid
!= bytenr
||
1222 key
.type
!= BTRFS_EXTENT_DATA_REF_KEY
)
1225 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1226 struct btrfs_extent_data_ref
);
1228 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1231 btrfs_release_path(path
);
1243 static noinline
int insert_extent_data_ref(struct btrfs_trans_handle
*trans
,
1244 struct btrfs_fs_info
*fs_info
,
1245 struct btrfs_path
*path
,
1246 u64 bytenr
, u64 parent
,
1247 u64 root_objectid
, u64 owner
,
1248 u64 offset
, int refs_to_add
)
1250 struct btrfs_root
*root
= fs_info
->extent_root
;
1251 struct btrfs_key key
;
1252 struct extent_buffer
*leaf
;
1257 key
.objectid
= bytenr
;
1259 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1260 key
.offset
= parent
;
1261 size
= sizeof(struct btrfs_shared_data_ref
);
1263 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1264 key
.offset
= hash_extent_data_ref(root_objectid
,
1266 size
= sizeof(struct btrfs_extent_data_ref
);
1269 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, size
);
1270 if (ret
&& ret
!= -EEXIST
)
1273 leaf
= path
->nodes
[0];
1275 struct btrfs_shared_data_ref
*ref
;
1276 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1277 struct btrfs_shared_data_ref
);
1279 btrfs_set_shared_data_ref_count(leaf
, ref
, refs_to_add
);
1281 num_refs
= btrfs_shared_data_ref_count(leaf
, ref
);
1282 num_refs
+= refs_to_add
;
1283 btrfs_set_shared_data_ref_count(leaf
, ref
, num_refs
);
1286 struct btrfs_extent_data_ref
*ref
;
1287 while (ret
== -EEXIST
) {
1288 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1289 struct btrfs_extent_data_ref
);
1290 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1293 btrfs_release_path(path
);
1295 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1297 if (ret
&& ret
!= -EEXIST
)
1300 leaf
= path
->nodes
[0];
1302 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1303 struct btrfs_extent_data_ref
);
1305 btrfs_set_extent_data_ref_root(leaf
, ref
,
1307 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
1308 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
1309 btrfs_set_extent_data_ref_count(leaf
, ref
, refs_to_add
);
1311 num_refs
= btrfs_extent_data_ref_count(leaf
, ref
);
1312 num_refs
+= refs_to_add
;
1313 btrfs_set_extent_data_ref_count(leaf
, ref
, num_refs
);
1316 btrfs_mark_buffer_dirty(leaf
);
1319 btrfs_release_path(path
);
1323 static noinline
int remove_extent_data_ref(struct btrfs_trans_handle
*trans
,
1324 struct btrfs_fs_info
*fs_info
,
1325 struct btrfs_path
*path
,
1326 int refs_to_drop
, int *last_ref
)
1328 struct btrfs_key key
;
1329 struct btrfs_extent_data_ref
*ref1
= NULL
;
1330 struct btrfs_shared_data_ref
*ref2
= NULL
;
1331 struct extent_buffer
*leaf
;
1335 leaf
= path
->nodes
[0];
1336 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1338 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1339 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1340 struct btrfs_extent_data_ref
);
1341 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1342 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1343 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1344 struct btrfs_shared_data_ref
);
1345 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1346 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1347 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1348 struct btrfs_extent_ref_v0
*ref0
;
1349 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1350 struct btrfs_extent_ref_v0
);
1351 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1357 BUG_ON(num_refs
< refs_to_drop
);
1358 num_refs
-= refs_to_drop
;
1360 if (num_refs
== 0) {
1361 ret
= btrfs_del_item(trans
, fs_info
->extent_root
, path
);
1364 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
)
1365 btrfs_set_extent_data_ref_count(leaf
, ref1
, num_refs
);
1366 else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
)
1367 btrfs_set_shared_data_ref_count(leaf
, ref2
, num_refs
);
1368 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1370 struct btrfs_extent_ref_v0
*ref0
;
1371 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1372 struct btrfs_extent_ref_v0
);
1373 btrfs_set_ref_count_v0(leaf
, ref0
, num_refs
);
1376 btrfs_mark_buffer_dirty(leaf
);
1381 static noinline u32
extent_data_ref_count(struct btrfs_path
*path
,
1382 struct btrfs_extent_inline_ref
*iref
)
1384 struct btrfs_key key
;
1385 struct extent_buffer
*leaf
;
1386 struct btrfs_extent_data_ref
*ref1
;
1387 struct btrfs_shared_data_ref
*ref2
;
1390 leaf
= path
->nodes
[0];
1391 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1393 if (btrfs_extent_inline_ref_type(leaf
, iref
) ==
1394 BTRFS_EXTENT_DATA_REF_KEY
) {
1395 ref1
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1396 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1398 ref2
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1399 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1401 } else if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1402 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1403 struct btrfs_extent_data_ref
);
1404 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1405 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1406 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1407 struct btrfs_shared_data_ref
);
1408 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1409 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1410 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1411 struct btrfs_extent_ref_v0
*ref0
;
1412 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1413 struct btrfs_extent_ref_v0
);
1414 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1422 static noinline
int lookup_tree_block_ref(struct btrfs_trans_handle
*trans
,
1423 struct btrfs_fs_info
*fs_info
,
1424 struct btrfs_path
*path
,
1425 u64 bytenr
, u64 parent
,
1428 struct btrfs_root
*root
= fs_info
->extent_root
;
1429 struct btrfs_key key
;
1432 key
.objectid
= bytenr
;
1434 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1435 key
.offset
= parent
;
1437 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1438 key
.offset
= root_objectid
;
1441 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1444 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1445 if (ret
== -ENOENT
&& parent
) {
1446 btrfs_release_path(path
);
1447 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1448 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1456 static noinline
int insert_tree_block_ref(struct btrfs_trans_handle
*trans
,
1457 struct btrfs_fs_info
*fs_info
,
1458 struct btrfs_path
*path
,
1459 u64 bytenr
, u64 parent
,
1462 struct btrfs_key key
;
1465 key
.objectid
= bytenr
;
1467 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1468 key
.offset
= parent
;
1470 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1471 key
.offset
= root_objectid
;
1474 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
,
1476 btrfs_release_path(path
);
1480 static inline int extent_ref_type(u64 parent
, u64 owner
)
1483 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1485 type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1487 type
= BTRFS_TREE_BLOCK_REF_KEY
;
1490 type
= BTRFS_SHARED_DATA_REF_KEY
;
1492 type
= BTRFS_EXTENT_DATA_REF_KEY
;
1497 static int find_next_key(struct btrfs_path
*path
, int level
,
1498 struct btrfs_key
*key
)
1501 for (; level
< BTRFS_MAX_LEVEL
; level
++) {
1502 if (!path
->nodes
[level
])
1504 if (path
->slots
[level
] + 1 >=
1505 btrfs_header_nritems(path
->nodes
[level
]))
1508 btrfs_item_key_to_cpu(path
->nodes
[level
], key
,
1509 path
->slots
[level
] + 1);
1511 btrfs_node_key_to_cpu(path
->nodes
[level
], key
,
1512 path
->slots
[level
] + 1);
1519 * look for inline back ref. if back ref is found, *ref_ret is set
1520 * to the address of inline back ref, and 0 is returned.
1522 * if back ref isn't found, *ref_ret is set to the address where it
1523 * should be inserted, and -ENOENT is returned.
1525 * if insert is true and there are too many inline back refs, the path
1526 * points to the extent item, and -EAGAIN is returned.
1528 * NOTE: inline back refs are ordered in the same way that back ref
1529 * items in the tree are ordered.
1531 static noinline_for_stack
1532 int lookup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1533 struct btrfs_fs_info
*fs_info
,
1534 struct btrfs_path
*path
,
1535 struct btrfs_extent_inline_ref
**ref_ret
,
1536 u64 bytenr
, u64 num_bytes
,
1537 u64 parent
, u64 root_objectid
,
1538 u64 owner
, u64 offset
, int insert
)
1540 struct btrfs_root
*root
= fs_info
->extent_root
;
1541 struct btrfs_key key
;
1542 struct extent_buffer
*leaf
;
1543 struct btrfs_extent_item
*ei
;
1544 struct btrfs_extent_inline_ref
*iref
;
1554 bool skinny_metadata
= btrfs_fs_incompat(fs_info
, SKINNY_METADATA
);
1556 key
.objectid
= bytenr
;
1557 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1558 key
.offset
= num_bytes
;
1560 want
= extent_ref_type(parent
, owner
);
1562 extra_size
= btrfs_extent_inline_ref_size(want
);
1563 path
->keep_locks
= 1;
1568 * Owner is our parent level, so we can just add one to get the level
1569 * for the block we are interested in.
1571 if (skinny_metadata
&& owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1572 key
.type
= BTRFS_METADATA_ITEM_KEY
;
1577 ret
= btrfs_search_slot(trans
, root
, &key
, path
, extra_size
, 1);
1584 * We may be a newly converted file system which still has the old fat
1585 * extent entries for metadata, so try and see if we have one of those.
1587 if (ret
> 0 && skinny_metadata
) {
1588 skinny_metadata
= false;
1589 if (path
->slots
[0]) {
1591 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
1593 if (key
.objectid
== bytenr
&&
1594 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
1595 key
.offset
== num_bytes
)
1599 key
.objectid
= bytenr
;
1600 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1601 key
.offset
= num_bytes
;
1602 btrfs_release_path(path
);
1607 if (ret
&& !insert
) {
1610 } else if (WARN_ON(ret
)) {
1615 leaf
= path
->nodes
[0];
1616 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1617 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1618 if (item_size
< sizeof(*ei
)) {
1623 ret
= convert_extent_item_v0(trans
, fs_info
, path
, owner
,
1629 leaf
= path
->nodes
[0];
1630 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1633 BUG_ON(item_size
< sizeof(*ei
));
1635 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1636 flags
= btrfs_extent_flags(leaf
, ei
);
1638 ptr
= (unsigned long)(ei
+ 1);
1639 end
= (unsigned long)ei
+ item_size
;
1641 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
&& !skinny_metadata
) {
1642 ptr
+= sizeof(struct btrfs_tree_block_info
);
1652 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1653 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1657 ptr
+= btrfs_extent_inline_ref_size(type
);
1661 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1662 struct btrfs_extent_data_ref
*dref
;
1663 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1664 if (match_extent_data_ref(leaf
, dref
, root_objectid
,
1669 if (hash_extent_data_ref_item(leaf
, dref
) <
1670 hash_extent_data_ref(root_objectid
, owner
, offset
))
1674 ref_offset
= btrfs_extent_inline_ref_offset(leaf
, iref
);
1676 if (parent
== ref_offset
) {
1680 if (ref_offset
< parent
)
1683 if (root_objectid
== ref_offset
) {
1687 if (ref_offset
< root_objectid
)
1691 ptr
+= btrfs_extent_inline_ref_size(type
);
1693 if (err
== -ENOENT
&& insert
) {
1694 if (item_size
+ extra_size
>=
1695 BTRFS_MAX_EXTENT_ITEM_SIZE(root
)) {
1700 * To add new inline back ref, we have to make sure
1701 * there is no corresponding back ref item.
1702 * For simplicity, we just do not add new inline back
1703 * ref if there is any kind of item for this block
1705 if (find_next_key(path
, 0, &key
) == 0 &&
1706 key
.objectid
== bytenr
&&
1707 key
.type
< BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1712 *ref_ret
= (struct btrfs_extent_inline_ref
*)ptr
;
1715 path
->keep_locks
= 0;
1716 btrfs_unlock_up_safe(path
, 1);
1722 * helper to add new inline back ref
1724 static noinline_for_stack
1725 void setup_inline_extent_backref(struct btrfs_fs_info
*fs_info
,
1726 struct btrfs_path
*path
,
1727 struct btrfs_extent_inline_ref
*iref
,
1728 u64 parent
, u64 root_objectid
,
1729 u64 owner
, u64 offset
, int refs_to_add
,
1730 struct btrfs_delayed_extent_op
*extent_op
)
1732 struct extent_buffer
*leaf
;
1733 struct btrfs_extent_item
*ei
;
1736 unsigned long item_offset
;
1741 leaf
= path
->nodes
[0];
1742 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1743 item_offset
= (unsigned long)iref
- (unsigned long)ei
;
1745 type
= extent_ref_type(parent
, owner
);
1746 size
= btrfs_extent_inline_ref_size(type
);
1748 btrfs_extend_item(fs_info
, path
, size
);
1750 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1751 refs
= btrfs_extent_refs(leaf
, ei
);
1752 refs
+= refs_to_add
;
1753 btrfs_set_extent_refs(leaf
, ei
, refs
);
1755 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1757 ptr
= (unsigned long)ei
+ item_offset
;
1758 end
= (unsigned long)ei
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
1759 if (ptr
< end
- size
)
1760 memmove_extent_buffer(leaf
, ptr
+ size
, ptr
,
1763 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1764 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
1765 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1766 struct btrfs_extent_data_ref
*dref
;
1767 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1768 btrfs_set_extent_data_ref_root(leaf
, dref
, root_objectid
);
1769 btrfs_set_extent_data_ref_objectid(leaf
, dref
, owner
);
1770 btrfs_set_extent_data_ref_offset(leaf
, dref
, offset
);
1771 btrfs_set_extent_data_ref_count(leaf
, dref
, refs_to_add
);
1772 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1773 struct btrfs_shared_data_ref
*sref
;
1774 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1775 btrfs_set_shared_data_ref_count(leaf
, sref
, refs_to_add
);
1776 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1777 } else if (type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
1778 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1780 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
1782 btrfs_mark_buffer_dirty(leaf
);
1785 static int lookup_extent_backref(struct btrfs_trans_handle
*trans
,
1786 struct btrfs_fs_info
*fs_info
,
1787 struct btrfs_path
*path
,
1788 struct btrfs_extent_inline_ref
**ref_ret
,
1789 u64 bytenr
, u64 num_bytes
, u64 parent
,
1790 u64 root_objectid
, u64 owner
, u64 offset
)
1794 ret
= lookup_inline_extent_backref(trans
, fs_info
, path
, ref_ret
,
1795 bytenr
, num_bytes
, parent
,
1796 root_objectid
, owner
, offset
, 0);
1800 btrfs_release_path(path
);
1803 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1804 ret
= lookup_tree_block_ref(trans
, fs_info
, path
, bytenr
,
1805 parent
, root_objectid
);
1807 ret
= lookup_extent_data_ref(trans
, fs_info
, path
, bytenr
,
1808 parent
, root_objectid
, owner
,
1815 * helper to update/remove inline back ref
1817 static noinline_for_stack
1818 void update_inline_extent_backref(struct btrfs_fs_info
*fs_info
,
1819 struct btrfs_path
*path
,
1820 struct btrfs_extent_inline_ref
*iref
,
1822 struct btrfs_delayed_extent_op
*extent_op
,
1825 struct extent_buffer
*leaf
;
1826 struct btrfs_extent_item
*ei
;
1827 struct btrfs_extent_data_ref
*dref
= NULL
;
1828 struct btrfs_shared_data_ref
*sref
= NULL
;
1836 leaf
= path
->nodes
[0];
1837 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1838 refs
= btrfs_extent_refs(leaf
, ei
);
1839 WARN_ON(refs_to_mod
< 0 && refs
+ refs_to_mod
<= 0);
1840 refs
+= refs_to_mod
;
1841 btrfs_set_extent_refs(leaf
, ei
, refs
);
1843 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1845 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1847 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1848 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1849 refs
= btrfs_extent_data_ref_count(leaf
, dref
);
1850 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1851 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1852 refs
= btrfs_shared_data_ref_count(leaf
, sref
);
1855 BUG_ON(refs_to_mod
!= -1);
1858 BUG_ON(refs_to_mod
< 0 && refs
< -refs_to_mod
);
1859 refs
+= refs_to_mod
;
1862 if (type
== BTRFS_EXTENT_DATA_REF_KEY
)
1863 btrfs_set_extent_data_ref_count(leaf
, dref
, refs
);
1865 btrfs_set_shared_data_ref_count(leaf
, sref
, refs
);
1868 size
= btrfs_extent_inline_ref_size(type
);
1869 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1870 ptr
= (unsigned long)iref
;
1871 end
= (unsigned long)ei
+ item_size
;
1872 if (ptr
+ size
< end
)
1873 memmove_extent_buffer(leaf
, ptr
, ptr
+ size
,
1876 btrfs_truncate_item(fs_info
, path
, item_size
, 1);
1878 btrfs_mark_buffer_dirty(leaf
);
1881 static noinline_for_stack
1882 int insert_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1883 struct btrfs_fs_info
*fs_info
,
1884 struct btrfs_path
*path
,
1885 u64 bytenr
, u64 num_bytes
, u64 parent
,
1886 u64 root_objectid
, u64 owner
,
1887 u64 offset
, int refs_to_add
,
1888 struct btrfs_delayed_extent_op
*extent_op
)
1890 struct btrfs_extent_inline_ref
*iref
;
1893 ret
= lookup_inline_extent_backref(trans
, fs_info
, path
, &iref
,
1894 bytenr
, num_bytes
, parent
,
1895 root_objectid
, owner
, offset
, 1);
1897 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
);
1898 update_inline_extent_backref(fs_info
, path
, iref
,
1899 refs_to_add
, extent_op
, NULL
);
1900 } else if (ret
== -ENOENT
) {
1901 setup_inline_extent_backref(fs_info
, path
, iref
, parent
,
1902 root_objectid
, owner
, offset
,
1903 refs_to_add
, extent_op
);
1909 static int insert_extent_backref(struct btrfs_trans_handle
*trans
,
1910 struct btrfs_fs_info
*fs_info
,
1911 struct btrfs_path
*path
,
1912 u64 bytenr
, u64 parent
, u64 root_objectid
,
1913 u64 owner
, u64 offset
, int refs_to_add
)
1916 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1917 BUG_ON(refs_to_add
!= 1);
1918 ret
= insert_tree_block_ref(trans
, fs_info
, path
, bytenr
,
1919 parent
, root_objectid
);
1921 ret
= insert_extent_data_ref(trans
, fs_info
, path
, bytenr
,
1922 parent
, root_objectid
,
1923 owner
, offset
, refs_to_add
);
1928 static int remove_extent_backref(struct btrfs_trans_handle
*trans
,
1929 struct btrfs_fs_info
*fs_info
,
1930 struct btrfs_path
*path
,
1931 struct btrfs_extent_inline_ref
*iref
,
1932 int refs_to_drop
, int is_data
, int *last_ref
)
1936 BUG_ON(!is_data
&& refs_to_drop
!= 1);
1938 update_inline_extent_backref(fs_info
, path
, iref
,
1939 -refs_to_drop
, NULL
, last_ref
);
1940 } else if (is_data
) {
1941 ret
= remove_extent_data_ref(trans
, fs_info
, path
, refs_to_drop
,
1945 ret
= btrfs_del_item(trans
, fs_info
->extent_root
, path
);
1950 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
1951 static int btrfs_issue_discard(struct block_device
*bdev
, u64 start
, u64 len
,
1952 u64
*discarded_bytes
)
1955 u64 bytes_left
, end
;
1956 u64 aligned_start
= ALIGN(start
, 1 << 9);
1958 if (WARN_ON(start
!= aligned_start
)) {
1959 len
-= aligned_start
- start
;
1960 len
= round_down(len
, 1 << 9);
1961 start
= aligned_start
;
1964 *discarded_bytes
= 0;
1972 /* Skip any superblocks on this device. */
1973 for (j
= 0; j
< BTRFS_SUPER_MIRROR_MAX
; j
++) {
1974 u64 sb_start
= btrfs_sb_offset(j
);
1975 u64 sb_end
= sb_start
+ BTRFS_SUPER_INFO_SIZE
;
1976 u64 size
= sb_start
- start
;
1978 if (!in_range(sb_start
, start
, bytes_left
) &&
1979 !in_range(sb_end
, start
, bytes_left
) &&
1980 !in_range(start
, sb_start
, BTRFS_SUPER_INFO_SIZE
))
1984 * Superblock spans beginning of range. Adjust start and
1987 if (sb_start
<= start
) {
1988 start
+= sb_end
- start
;
1993 bytes_left
= end
- start
;
1998 ret
= blkdev_issue_discard(bdev
, start
>> 9, size
>> 9,
2001 *discarded_bytes
+= size
;
2002 else if (ret
!= -EOPNOTSUPP
)
2011 bytes_left
= end
- start
;
2015 ret
= blkdev_issue_discard(bdev
, start
>> 9, bytes_left
>> 9,
2018 *discarded_bytes
+= bytes_left
;
2023 int btrfs_discard_extent(struct btrfs_fs_info
*fs_info
, u64 bytenr
,
2024 u64 num_bytes
, u64
*actual_bytes
)
2027 u64 discarded_bytes
= 0;
2028 struct btrfs_bio
*bbio
= NULL
;
2032 * Avoid races with device replace and make sure our bbio has devices
2033 * associated to its stripes that don't go away while we are discarding.
2035 btrfs_bio_counter_inc_blocked(fs_info
);
2036 /* Tell the block device(s) that the sectors can be discarded */
2037 ret
= btrfs_map_block(fs_info
, BTRFS_MAP_DISCARD
, bytenr
, &num_bytes
,
2039 /* Error condition is -ENOMEM */
2041 struct btrfs_bio_stripe
*stripe
= bbio
->stripes
;
2045 for (i
= 0; i
< bbio
->num_stripes
; i
++, stripe
++) {
2047 if (!stripe
->dev
->can_discard
)
2050 ret
= btrfs_issue_discard(stripe
->dev
->bdev
,
2055 discarded_bytes
+= bytes
;
2056 else if (ret
!= -EOPNOTSUPP
)
2057 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2060 * Just in case we get back EOPNOTSUPP for some reason,
2061 * just ignore the return value so we don't screw up
2062 * people calling discard_extent.
2066 btrfs_put_bbio(bbio
);
2068 btrfs_bio_counter_dec(fs_info
);
2071 *actual_bytes
= discarded_bytes
;
2074 if (ret
== -EOPNOTSUPP
)
2079 /* Can return -ENOMEM */
2080 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
2081 struct btrfs_fs_info
*fs_info
,
2082 u64 bytenr
, u64 num_bytes
, u64 parent
,
2083 u64 root_objectid
, u64 owner
, u64 offset
)
2087 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
&&
2088 root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
2090 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
2091 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
2093 parent
, root_objectid
, (int)owner
,
2094 BTRFS_ADD_DELAYED_REF
, NULL
);
2096 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
2097 num_bytes
, parent
, root_objectid
,
2099 BTRFS_ADD_DELAYED_REF
);
2104 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
2105 struct btrfs_fs_info
*fs_info
,
2106 struct btrfs_delayed_ref_node
*node
,
2107 u64 parent
, u64 root_objectid
,
2108 u64 owner
, u64 offset
, int refs_to_add
,
2109 struct btrfs_delayed_extent_op
*extent_op
)
2111 struct btrfs_path
*path
;
2112 struct extent_buffer
*leaf
;
2113 struct btrfs_extent_item
*item
;
2114 struct btrfs_key key
;
2115 u64 bytenr
= node
->bytenr
;
2116 u64 num_bytes
= node
->num_bytes
;
2120 path
= btrfs_alloc_path();
2124 path
->reada
= READA_FORWARD
;
2125 path
->leave_spinning
= 1;
2126 /* this will setup the path even if it fails to insert the back ref */
2127 ret
= insert_inline_extent_backref(trans
, fs_info
, path
, bytenr
,
2128 num_bytes
, parent
, root_objectid
,
2130 refs_to_add
, extent_op
);
2131 if ((ret
< 0 && ret
!= -EAGAIN
) || !ret
)
2135 * Ok we had -EAGAIN which means we didn't have space to insert and
2136 * inline extent ref, so just update the reference count and add a
2139 leaf
= path
->nodes
[0];
2140 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2141 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2142 refs
= btrfs_extent_refs(leaf
, item
);
2143 btrfs_set_extent_refs(leaf
, item
, refs
+ refs_to_add
);
2145 __run_delayed_extent_op(extent_op
, leaf
, item
);
2147 btrfs_mark_buffer_dirty(leaf
);
2148 btrfs_release_path(path
);
2150 path
->reada
= READA_FORWARD
;
2151 path
->leave_spinning
= 1;
2152 /* now insert the actual backref */
2153 ret
= insert_extent_backref(trans
, fs_info
, path
, bytenr
, parent
,
2154 root_objectid
, owner
, offset
, refs_to_add
);
2156 btrfs_abort_transaction(trans
, ret
);
2158 btrfs_free_path(path
);
2162 static int run_delayed_data_ref(struct btrfs_trans_handle
*trans
,
2163 struct btrfs_fs_info
*fs_info
,
2164 struct btrfs_delayed_ref_node
*node
,
2165 struct btrfs_delayed_extent_op
*extent_op
,
2166 int insert_reserved
)
2169 struct btrfs_delayed_data_ref
*ref
;
2170 struct btrfs_key ins
;
2175 ins
.objectid
= node
->bytenr
;
2176 ins
.offset
= node
->num_bytes
;
2177 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2179 ref
= btrfs_delayed_node_to_data_ref(node
);
2180 trace_run_delayed_data_ref(fs_info
, node
, ref
, node
->action
);
2182 if (node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2183 parent
= ref
->parent
;
2184 ref_root
= ref
->root
;
2186 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2188 flags
|= extent_op
->flags_to_set
;
2189 ret
= alloc_reserved_file_extent(trans
, fs_info
,
2190 parent
, ref_root
, flags
,
2191 ref
->objectid
, ref
->offset
,
2192 &ins
, node
->ref_mod
);
2193 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2194 ret
= __btrfs_inc_extent_ref(trans
, fs_info
, node
, parent
,
2195 ref_root
, ref
->objectid
,
2196 ref
->offset
, node
->ref_mod
,
2198 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2199 ret
= __btrfs_free_extent(trans
, fs_info
, node
, parent
,
2200 ref_root
, ref
->objectid
,
2201 ref
->offset
, node
->ref_mod
,
2209 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
2210 struct extent_buffer
*leaf
,
2211 struct btrfs_extent_item
*ei
)
2213 u64 flags
= btrfs_extent_flags(leaf
, ei
);
2214 if (extent_op
->update_flags
) {
2215 flags
|= extent_op
->flags_to_set
;
2216 btrfs_set_extent_flags(leaf
, ei
, flags
);
2219 if (extent_op
->update_key
) {
2220 struct btrfs_tree_block_info
*bi
;
2221 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
));
2222 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
2223 btrfs_set_tree_block_key(leaf
, bi
, &extent_op
->key
);
2227 static int run_delayed_extent_op(struct btrfs_trans_handle
*trans
,
2228 struct btrfs_fs_info
*fs_info
,
2229 struct btrfs_delayed_ref_node
*node
,
2230 struct btrfs_delayed_extent_op
*extent_op
)
2232 struct btrfs_key key
;
2233 struct btrfs_path
*path
;
2234 struct btrfs_extent_item
*ei
;
2235 struct extent_buffer
*leaf
;
2239 int metadata
= !extent_op
->is_data
;
2244 if (metadata
&& !btrfs_fs_incompat(fs_info
, SKINNY_METADATA
))
2247 path
= btrfs_alloc_path();
2251 key
.objectid
= node
->bytenr
;
2254 key
.type
= BTRFS_METADATA_ITEM_KEY
;
2255 key
.offset
= extent_op
->level
;
2257 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2258 key
.offset
= node
->num_bytes
;
2262 path
->reada
= READA_FORWARD
;
2263 path
->leave_spinning
= 1;
2264 ret
= btrfs_search_slot(trans
, fs_info
->extent_root
, &key
, path
, 0, 1);
2271 if (path
->slots
[0] > 0) {
2273 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
2275 if (key
.objectid
== node
->bytenr
&&
2276 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
2277 key
.offset
== node
->num_bytes
)
2281 btrfs_release_path(path
);
2284 key
.objectid
= node
->bytenr
;
2285 key
.offset
= node
->num_bytes
;
2286 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2295 leaf
= path
->nodes
[0];
2296 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2297 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2298 if (item_size
< sizeof(*ei
)) {
2299 ret
= convert_extent_item_v0(trans
, fs_info
, path
, (u64
)-1, 0);
2304 leaf
= path
->nodes
[0];
2305 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2308 BUG_ON(item_size
< sizeof(*ei
));
2309 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2310 __run_delayed_extent_op(extent_op
, leaf
, ei
);
2312 btrfs_mark_buffer_dirty(leaf
);
2314 btrfs_free_path(path
);
2318 static int run_delayed_tree_ref(struct btrfs_trans_handle
*trans
,
2319 struct btrfs_fs_info
*fs_info
,
2320 struct btrfs_delayed_ref_node
*node
,
2321 struct btrfs_delayed_extent_op
*extent_op
,
2322 int insert_reserved
)
2325 struct btrfs_delayed_tree_ref
*ref
;
2326 struct btrfs_key ins
;
2329 bool skinny_metadata
= btrfs_fs_incompat(fs_info
, SKINNY_METADATA
);
2331 ref
= btrfs_delayed_node_to_tree_ref(node
);
2332 trace_run_delayed_tree_ref(fs_info
, node
, ref
, node
->action
);
2334 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2335 parent
= ref
->parent
;
2336 ref_root
= ref
->root
;
2338 ins
.objectid
= node
->bytenr
;
2339 if (skinny_metadata
) {
2340 ins
.offset
= ref
->level
;
2341 ins
.type
= BTRFS_METADATA_ITEM_KEY
;
2343 ins
.offset
= node
->num_bytes
;
2344 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2347 if (node
->ref_mod
!= 1) {
2349 "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu",
2350 node
->bytenr
, node
->ref_mod
, node
->action
, ref_root
,
2354 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2355 BUG_ON(!extent_op
|| !extent_op
->update_flags
);
2356 ret
= alloc_reserved_tree_block(trans
, fs_info
,
2358 extent_op
->flags_to_set
,
2361 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2362 ret
= __btrfs_inc_extent_ref(trans
, fs_info
, node
,
2366 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2367 ret
= __btrfs_free_extent(trans
, fs_info
, node
,
2369 ref
->level
, 0, 1, extent_op
);
2376 /* helper function to actually process a single delayed ref entry */
2377 static int run_one_delayed_ref(struct btrfs_trans_handle
*trans
,
2378 struct btrfs_fs_info
*fs_info
,
2379 struct btrfs_delayed_ref_node
*node
,
2380 struct btrfs_delayed_extent_op
*extent_op
,
2381 int insert_reserved
)
2385 if (trans
->aborted
) {
2386 if (insert_reserved
)
2387 btrfs_pin_extent(fs_info
, node
->bytenr
,
2388 node
->num_bytes
, 1);
2392 if (btrfs_delayed_ref_is_head(node
)) {
2393 struct btrfs_delayed_ref_head
*head
;
2395 * we've hit the end of the chain and we were supposed
2396 * to insert this extent into the tree. But, it got
2397 * deleted before we ever needed to insert it, so all
2398 * we have to do is clean up the accounting
2401 head
= btrfs_delayed_node_to_head(node
);
2402 trace_run_delayed_ref_head(fs_info
, node
, head
, node
->action
);
2404 if (insert_reserved
) {
2405 btrfs_pin_extent(fs_info
, node
->bytenr
,
2406 node
->num_bytes
, 1);
2407 if (head
->is_data
) {
2408 ret
= btrfs_del_csums(trans
, fs_info
,
2414 /* Also free its reserved qgroup space */
2415 btrfs_qgroup_free_delayed_ref(fs_info
, head
->qgroup_ref_root
,
2416 head
->qgroup_reserved
);
2420 if (node
->type
== BTRFS_TREE_BLOCK_REF_KEY
||
2421 node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2422 ret
= run_delayed_tree_ref(trans
, fs_info
, node
, extent_op
,
2424 else if (node
->type
== BTRFS_EXTENT_DATA_REF_KEY
||
2425 node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2426 ret
= run_delayed_data_ref(trans
, fs_info
, node
, extent_op
,
2433 static inline struct btrfs_delayed_ref_node
*
2434 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
2436 struct btrfs_delayed_ref_node
*ref
;
2438 if (list_empty(&head
->ref_list
))
2442 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2443 * This is to prevent a ref count from going down to zero, which deletes
2444 * the extent item from the extent tree, when there still are references
2445 * to add, which would fail because they would not find the extent item.
2447 if (!list_empty(&head
->ref_add_list
))
2448 return list_first_entry(&head
->ref_add_list
,
2449 struct btrfs_delayed_ref_node
, add_list
);
2451 ref
= list_first_entry(&head
->ref_list
, struct btrfs_delayed_ref_node
,
2453 ASSERT(list_empty(&ref
->add_list
));
2458 * Returns 0 on success or if called with an already aborted transaction.
2459 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2461 static noinline
int __btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2462 struct btrfs_fs_info
*fs_info
,
2465 struct btrfs_delayed_ref_root
*delayed_refs
;
2466 struct btrfs_delayed_ref_node
*ref
;
2467 struct btrfs_delayed_ref_head
*locked_ref
= NULL
;
2468 struct btrfs_delayed_extent_op
*extent_op
;
2469 ktime_t start
= ktime_get();
2471 unsigned long count
= 0;
2472 unsigned long actual_count
= 0;
2473 int must_insert_reserved
= 0;
2475 delayed_refs
= &trans
->transaction
->delayed_refs
;
2481 spin_lock(&delayed_refs
->lock
);
2482 locked_ref
= btrfs_select_ref_head(trans
);
2484 spin_unlock(&delayed_refs
->lock
);
2488 /* grab the lock that says we are going to process
2489 * all the refs for this head */
2490 ret
= btrfs_delayed_ref_lock(trans
, locked_ref
);
2491 spin_unlock(&delayed_refs
->lock
);
2493 * we may have dropped the spin lock to get the head
2494 * mutex lock, and that might have given someone else
2495 * time to free the head. If that's true, it has been
2496 * removed from our list and we can move on.
2498 if (ret
== -EAGAIN
) {
2506 * We need to try and merge add/drops of the same ref since we
2507 * can run into issues with relocate dropping the implicit ref
2508 * and then it being added back again before the drop can
2509 * finish. If we merged anything we need to re-loop so we can
2511 * Or we can get node references of the same type that weren't
2512 * merged when created due to bumps in the tree mod seq, and
2513 * we need to merge them to prevent adding an inline extent
2514 * backref before dropping it (triggering a BUG_ON at
2515 * insert_inline_extent_backref()).
2517 spin_lock(&locked_ref
->lock
);
2518 btrfs_merge_delayed_refs(trans
, fs_info
, delayed_refs
,
2522 * locked_ref is the head node, so we have to go one
2523 * node back for any delayed ref updates
2525 ref
= select_delayed_ref(locked_ref
);
2527 if (ref
&& ref
->seq
&&
2528 btrfs_check_delayed_seq(fs_info
, delayed_refs
, ref
->seq
)) {
2529 spin_unlock(&locked_ref
->lock
);
2530 spin_lock(&delayed_refs
->lock
);
2531 locked_ref
->processing
= 0;
2532 delayed_refs
->num_heads_ready
++;
2533 spin_unlock(&delayed_refs
->lock
);
2534 btrfs_delayed_ref_unlock(locked_ref
);
2542 * record the must insert reserved flag before we
2543 * drop the spin lock.
2545 must_insert_reserved
= locked_ref
->must_insert_reserved
;
2546 locked_ref
->must_insert_reserved
= 0;
2548 extent_op
= locked_ref
->extent_op
;
2549 locked_ref
->extent_op
= NULL
;
2554 /* All delayed refs have been processed, Go ahead
2555 * and send the head node to run_one_delayed_ref,
2556 * so that any accounting fixes can happen
2558 ref
= &locked_ref
->node
;
2560 if (extent_op
&& must_insert_reserved
) {
2561 btrfs_free_delayed_extent_op(extent_op
);
2566 spin_unlock(&locked_ref
->lock
);
2567 ret
= run_delayed_extent_op(trans
, fs_info
,
2569 btrfs_free_delayed_extent_op(extent_op
);
2573 * Need to reset must_insert_reserved if
2574 * there was an error so the abort stuff
2575 * can cleanup the reserved space
2578 if (must_insert_reserved
)
2579 locked_ref
->must_insert_reserved
= 1;
2580 spin_lock(&delayed_refs
->lock
);
2581 locked_ref
->processing
= 0;
2582 delayed_refs
->num_heads_ready
++;
2583 spin_unlock(&delayed_refs
->lock
);
2584 btrfs_debug(fs_info
,
2585 "run_delayed_extent_op returned %d",
2587 btrfs_delayed_ref_unlock(locked_ref
);
2594 * Need to drop our head ref lock and re-acquire the
2595 * delayed ref lock and then re-check to make sure
2598 spin_unlock(&locked_ref
->lock
);
2599 spin_lock(&delayed_refs
->lock
);
2600 spin_lock(&locked_ref
->lock
);
2601 if (!list_empty(&locked_ref
->ref_list
) ||
2602 locked_ref
->extent_op
) {
2603 spin_unlock(&locked_ref
->lock
);
2604 spin_unlock(&delayed_refs
->lock
);
2608 delayed_refs
->num_heads
--;
2609 rb_erase(&locked_ref
->href_node
,
2610 &delayed_refs
->href_root
);
2611 spin_unlock(&delayed_refs
->lock
);
2615 list_del(&ref
->list
);
2616 if (!list_empty(&ref
->add_list
))
2617 list_del(&ref
->add_list
);
2619 atomic_dec(&delayed_refs
->num_entries
);
2621 if (!btrfs_delayed_ref_is_head(ref
)) {
2623 * when we play the delayed ref, also correct the
2626 switch (ref
->action
) {
2627 case BTRFS_ADD_DELAYED_REF
:
2628 case BTRFS_ADD_DELAYED_EXTENT
:
2629 locked_ref
->node
.ref_mod
-= ref
->ref_mod
;
2631 case BTRFS_DROP_DELAYED_REF
:
2632 locked_ref
->node
.ref_mod
+= ref
->ref_mod
;
2638 spin_unlock(&locked_ref
->lock
);
2640 ret
= run_one_delayed_ref(trans
, fs_info
, ref
, extent_op
,
2641 must_insert_reserved
);
2643 btrfs_free_delayed_extent_op(extent_op
);
2645 spin_lock(&delayed_refs
->lock
);
2646 locked_ref
->processing
= 0;
2647 delayed_refs
->num_heads_ready
++;
2648 spin_unlock(&delayed_refs
->lock
);
2649 btrfs_delayed_ref_unlock(locked_ref
);
2650 btrfs_put_delayed_ref(ref
);
2651 btrfs_debug(fs_info
, "run_one_delayed_ref returned %d",
2657 * If this node is a head, that means all the refs in this head
2658 * have been dealt with, and we will pick the next head to deal
2659 * with, so we must unlock the head and drop it from the cluster
2660 * list before we release it.
2662 if (btrfs_delayed_ref_is_head(ref
)) {
2663 if (locked_ref
->is_data
&&
2664 locked_ref
->total_ref_mod
< 0) {
2665 spin_lock(&delayed_refs
->lock
);
2666 delayed_refs
->pending_csums
-= ref
->num_bytes
;
2667 spin_unlock(&delayed_refs
->lock
);
2669 btrfs_delayed_ref_unlock(locked_ref
);
2672 btrfs_put_delayed_ref(ref
);
2678 * We don't want to include ref heads since we can have empty ref heads
2679 * and those will drastically skew our runtime down since we just do
2680 * accounting, no actual extent tree updates.
2682 if (actual_count
> 0) {
2683 u64 runtime
= ktime_to_ns(ktime_sub(ktime_get(), start
));
2687 * We weigh the current average higher than our current runtime
2688 * to avoid large swings in the average.
2690 spin_lock(&delayed_refs
->lock
);
2691 avg
= fs_info
->avg_delayed_ref_runtime
* 3 + runtime
;
2692 fs_info
->avg_delayed_ref_runtime
= avg
>> 2; /* div by 4 */
2693 spin_unlock(&delayed_refs
->lock
);
2698 #ifdef SCRAMBLE_DELAYED_REFS
2700 * Normally delayed refs get processed in ascending bytenr order. This
2701 * correlates in most cases to the order added. To expose dependencies on this
2702 * order, we start to process the tree in the middle instead of the beginning
2704 static u64
find_middle(struct rb_root
*root
)
2706 struct rb_node
*n
= root
->rb_node
;
2707 struct btrfs_delayed_ref_node
*entry
;
2710 u64 first
= 0, last
= 0;
2714 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2715 first
= entry
->bytenr
;
2719 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2720 last
= entry
->bytenr
;
2725 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2726 WARN_ON(!entry
->in_tree
);
2728 middle
= entry
->bytenr
;
2741 static inline u64
heads_to_leaves(struct btrfs_fs_info
*fs_info
, u64 heads
)
2745 num_bytes
= heads
* (sizeof(struct btrfs_extent_item
) +
2746 sizeof(struct btrfs_extent_inline_ref
));
2747 if (!btrfs_fs_incompat(fs_info
, SKINNY_METADATA
))
2748 num_bytes
+= heads
* sizeof(struct btrfs_tree_block_info
);
2751 * We don't ever fill up leaves all the way so multiply by 2 just to be
2752 * closer to what we're really going to want to use.
2754 return div_u64(num_bytes
, BTRFS_LEAF_DATA_SIZE(fs_info
));
2758 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2759 * would require to store the csums for that many bytes.
2761 u64
btrfs_csum_bytes_to_leaves(struct btrfs_fs_info
*fs_info
, u64 csum_bytes
)
2764 u64 num_csums_per_leaf
;
2767 csum_size
= BTRFS_MAX_ITEM_SIZE(fs_info
);
2768 num_csums_per_leaf
= div64_u64(csum_size
,
2769 (u64
)btrfs_super_csum_size(fs_info
->super_copy
));
2770 num_csums
= div64_u64(csum_bytes
, fs_info
->sectorsize
);
2771 num_csums
+= num_csums_per_leaf
- 1;
2772 num_csums
= div64_u64(num_csums
, num_csums_per_leaf
);
2776 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle
*trans
,
2777 struct btrfs_fs_info
*fs_info
)
2779 struct btrfs_block_rsv
*global_rsv
;
2780 u64 num_heads
= trans
->transaction
->delayed_refs
.num_heads_ready
;
2781 u64 csum_bytes
= trans
->transaction
->delayed_refs
.pending_csums
;
2782 u64 num_dirty_bgs
= trans
->transaction
->num_dirty_bgs
;
2783 u64 num_bytes
, num_dirty_bgs_bytes
;
2786 num_bytes
= btrfs_calc_trans_metadata_size(fs_info
, 1);
2787 num_heads
= heads_to_leaves(fs_info
, num_heads
);
2789 num_bytes
+= (num_heads
- 1) * fs_info
->nodesize
;
2791 num_bytes
+= btrfs_csum_bytes_to_leaves(fs_info
, csum_bytes
) *
2793 num_dirty_bgs_bytes
= btrfs_calc_trans_metadata_size(fs_info
,
2795 global_rsv
= &fs_info
->global_block_rsv
;
2798 * If we can't allocate any more chunks lets make sure we have _lots_ of
2799 * wiggle room since running delayed refs can create more delayed refs.
2801 if (global_rsv
->space_info
->full
) {
2802 num_dirty_bgs_bytes
<<= 1;
2806 spin_lock(&global_rsv
->lock
);
2807 if (global_rsv
->reserved
<= num_bytes
+ num_dirty_bgs_bytes
)
2809 spin_unlock(&global_rsv
->lock
);
2813 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle
*trans
,
2814 struct btrfs_fs_info
*fs_info
)
2817 atomic_read(&trans
->transaction
->delayed_refs
.num_entries
);
2822 avg_runtime
= fs_info
->avg_delayed_ref_runtime
;
2823 val
= num_entries
* avg_runtime
;
2824 if (val
>= NSEC_PER_SEC
)
2826 if (val
>= NSEC_PER_SEC
/ 2)
2829 return btrfs_check_space_for_delayed_refs(trans
, fs_info
);
2832 struct async_delayed_refs
{
2833 struct btrfs_root
*root
;
2838 struct completion wait
;
2839 struct btrfs_work work
;
2842 static inline struct async_delayed_refs
*
2843 to_async_delayed_refs(struct btrfs_work
*work
)
2845 return container_of(work
, struct async_delayed_refs
, work
);
2848 static void delayed_ref_async_start(struct btrfs_work
*work
)
2850 struct async_delayed_refs
*async
= to_async_delayed_refs(work
);
2851 struct btrfs_trans_handle
*trans
;
2852 struct btrfs_fs_info
*fs_info
= async
->root
->fs_info
;
2855 /* if the commit is already started, we don't need to wait here */
2856 if (btrfs_transaction_blocked(fs_info
))
2859 trans
= btrfs_join_transaction(async
->root
);
2860 if (IS_ERR(trans
)) {
2861 async
->error
= PTR_ERR(trans
);
2866 * trans->sync means that when we call end_transaction, we won't
2867 * wait on delayed refs
2871 /* Don't bother flushing if we got into a different transaction */
2872 if (trans
->transid
> async
->transid
)
2875 ret
= btrfs_run_delayed_refs(trans
, fs_info
, async
->count
);
2879 ret
= btrfs_end_transaction(trans
);
2880 if (ret
&& !async
->error
)
2884 complete(&async
->wait
);
2889 int btrfs_async_run_delayed_refs(struct btrfs_fs_info
*fs_info
,
2890 unsigned long count
, u64 transid
, int wait
)
2892 struct async_delayed_refs
*async
;
2895 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
2899 async
->root
= fs_info
->tree_root
;
2900 async
->count
= count
;
2902 async
->transid
= transid
;
2907 init_completion(&async
->wait
);
2909 btrfs_init_work(&async
->work
, btrfs_extent_refs_helper
,
2910 delayed_ref_async_start
, NULL
, NULL
);
2912 btrfs_queue_work(fs_info
->extent_workers
, &async
->work
);
2915 wait_for_completion(&async
->wait
);
2924 * this starts processing the delayed reference count updates and
2925 * extent insertions we have queued up so far. count can be
2926 * 0, which means to process everything in the tree at the start
2927 * of the run (but not newly added entries), or it can be some target
2928 * number you'd like to process.
2930 * Returns 0 on success or if called with an aborted transaction
2931 * Returns <0 on error and aborts the transaction
2933 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2934 struct btrfs_fs_info
*fs_info
, unsigned long count
)
2936 struct rb_node
*node
;
2937 struct btrfs_delayed_ref_root
*delayed_refs
;
2938 struct btrfs_delayed_ref_head
*head
;
2940 int run_all
= count
== (unsigned long)-1;
2941 bool can_flush_pending_bgs
= trans
->can_flush_pending_bgs
;
2943 /* We'll clean this up in btrfs_cleanup_transaction */
2947 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE
, &fs_info
->flags
))
2950 delayed_refs
= &trans
->transaction
->delayed_refs
;
2952 count
= atomic_read(&delayed_refs
->num_entries
) * 2;
2955 #ifdef SCRAMBLE_DELAYED_REFS
2956 delayed_refs
->run_delayed_start
= find_middle(&delayed_refs
->root
);
2958 trans
->can_flush_pending_bgs
= false;
2959 ret
= __btrfs_run_delayed_refs(trans
, fs_info
, count
);
2961 btrfs_abort_transaction(trans
, ret
);
2966 if (!list_empty(&trans
->new_bgs
))
2967 btrfs_create_pending_block_groups(trans
, fs_info
);
2969 spin_lock(&delayed_refs
->lock
);
2970 node
= rb_first(&delayed_refs
->href_root
);
2972 spin_unlock(&delayed_refs
->lock
);
2977 head
= rb_entry(node
, struct btrfs_delayed_ref_head
,
2979 if (btrfs_delayed_ref_is_head(&head
->node
)) {
2980 struct btrfs_delayed_ref_node
*ref
;
2983 atomic_inc(&ref
->refs
);
2985 spin_unlock(&delayed_refs
->lock
);
2987 * Mutex was contended, block until it's
2988 * released and try again
2990 mutex_lock(&head
->mutex
);
2991 mutex_unlock(&head
->mutex
);
2993 btrfs_put_delayed_ref(ref
);
2999 node
= rb_next(node
);
3001 spin_unlock(&delayed_refs
->lock
);
3006 assert_qgroups_uptodate(trans
);
3007 trans
->can_flush_pending_bgs
= can_flush_pending_bgs
;
3011 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
3012 struct btrfs_fs_info
*fs_info
,
3013 u64 bytenr
, u64 num_bytes
, u64 flags
,
3014 int level
, int is_data
)
3016 struct btrfs_delayed_extent_op
*extent_op
;
3019 extent_op
= btrfs_alloc_delayed_extent_op();
3023 extent_op
->flags_to_set
= flags
;
3024 extent_op
->update_flags
= true;
3025 extent_op
->update_key
= false;
3026 extent_op
->is_data
= is_data
? true : false;
3027 extent_op
->level
= level
;
3029 ret
= btrfs_add_delayed_extent_op(fs_info
, trans
, bytenr
,
3030 num_bytes
, extent_op
);
3032 btrfs_free_delayed_extent_op(extent_op
);
3036 static noinline
int check_delayed_ref(struct btrfs_root
*root
,
3037 struct btrfs_path
*path
,
3038 u64 objectid
, u64 offset
, u64 bytenr
)
3040 struct btrfs_delayed_ref_head
*head
;
3041 struct btrfs_delayed_ref_node
*ref
;
3042 struct btrfs_delayed_data_ref
*data_ref
;
3043 struct btrfs_delayed_ref_root
*delayed_refs
;
3044 struct btrfs_transaction
*cur_trans
;
3047 cur_trans
= root
->fs_info
->running_transaction
;
3051 delayed_refs
= &cur_trans
->delayed_refs
;
3052 spin_lock(&delayed_refs
->lock
);
3053 head
= btrfs_find_delayed_ref_head(delayed_refs
, bytenr
);
3055 spin_unlock(&delayed_refs
->lock
);
3059 if (!mutex_trylock(&head
->mutex
)) {
3060 atomic_inc(&head
->node
.refs
);
3061 spin_unlock(&delayed_refs
->lock
);
3063 btrfs_release_path(path
);
3066 * Mutex was contended, block until it's released and let
3069 mutex_lock(&head
->mutex
);
3070 mutex_unlock(&head
->mutex
);
3071 btrfs_put_delayed_ref(&head
->node
);
3074 spin_unlock(&delayed_refs
->lock
);
3076 spin_lock(&head
->lock
);
3077 list_for_each_entry(ref
, &head
->ref_list
, list
) {
3078 /* If it's a shared ref we know a cross reference exists */
3079 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
) {
3084 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
3087 * If our ref doesn't match the one we're currently looking at
3088 * then we have a cross reference.
3090 if (data_ref
->root
!= root
->root_key
.objectid
||
3091 data_ref
->objectid
!= objectid
||
3092 data_ref
->offset
!= offset
) {
3097 spin_unlock(&head
->lock
);
3098 mutex_unlock(&head
->mutex
);
3102 static noinline
int check_committed_ref(struct btrfs_root
*root
,
3103 struct btrfs_path
*path
,
3104 u64 objectid
, u64 offset
, u64 bytenr
)
3106 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3107 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
3108 struct extent_buffer
*leaf
;
3109 struct btrfs_extent_data_ref
*ref
;
3110 struct btrfs_extent_inline_ref
*iref
;
3111 struct btrfs_extent_item
*ei
;
3112 struct btrfs_key key
;
3116 key
.objectid
= bytenr
;
3117 key
.offset
= (u64
)-1;
3118 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
3120 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
3123 BUG_ON(ret
== 0); /* Corruption */
3126 if (path
->slots
[0] == 0)
3130 leaf
= path
->nodes
[0];
3131 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
3133 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
3137 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
3138 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3139 if (item_size
< sizeof(*ei
)) {
3140 WARN_ON(item_size
!= sizeof(struct btrfs_extent_item_v0
));
3144 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
3146 if (item_size
!= sizeof(*ei
) +
3147 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
3150 if (btrfs_extent_generation(leaf
, ei
) <=
3151 btrfs_root_last_snapshot(&root
->root_item
))
3154 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
3155 if (btrfs_extent_inline_ref_type(leaf
, iref
) !=
3156 BTRFS_EXTENT_DATA_REF_KEY
)
3159 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
3160 if (btrfs_extent_refs(leaf
, ei
) !=
3161 btrfs_extent_data_ref_count(leaf
, ref
) ||
3162 btrfs_extent_data_ref_root(leaf
, ref
) !=
3163 root
->root_key
.objectid
||
3164 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
3165 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
3173 int btrfs_cross_ref_exist(struct btrfs_root
*root
, u64 objectid
, u64 offset
,
3176 struct btrfs_path
*path
;
3180 path
= btrfs_alloc_path();
3185 ret
= check_committed_ref(root
, path
, objectid
,
3187 if (ret
&& ret
!= -ENOENT
)
3190 ret2
= check_delayed_ref(root
, path
, objectid
,
3192 } while (ret2
== -EAGAIN
);
3194 if (ret2
&& ret2
!= -ENOENT
) {
3199 if (ret
!= -ENOENT
|| ret2
!= -ENOENT
)
3202 btrfs_free_path(path
);
3203 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
3208 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
3209 struct btrfs_root
*root
,
3210 struct extent_buffer
*buf
,
3211 int full_backref
, int inc
)
3213 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3219 struct btrfs_key key
;
3220 struct btrfs_file_extent_item
*fi
;
3224 int (*process_func
)(struct btrfs_trans_handle
*,
3225 struct btrfs_fs_info
*,
3226 u64
, u64
, u64
, u64
, u64
, u64
);
3229 if (btrfs_is_testing(fs_info
))
3232 ref_root
= btrfs_header_owner(buf
);
3233 nritems
= btrfs_header_nritems(buf
);
3234 level
= btrfs_header_level(buf
);
3236 if (!test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) && level
== 0)
3240 process_func
= btrfs_inc_extent_ref
;
3242 process_func
= btrfs_free_extent
;
3245 parent
= buf
->start
;
3249 for (i
= 0; i
< nritems
; i
++) {
3251 btrfs_item_key_to_cpu(buf
, &key
, i
);
3252 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
3254 fi
= btrfs_item_ptr(buf
, i
,
3255 struct btrfs_file_extent_item
);
3256 if (btrfs_file_extent_type(buf
, fi
) ==
3257 BTRFS_FILE_EXTENT_INLINE
)
3259 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
3263 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
3264 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
3265 ret
= process_func(trans
, fs_info
, bytenr
, num_bytes
,
3266 parent
, ref_root
, key
.objectid
,
3271 bytenr
= btrfs_node_blockptr(buf
, i
);
3272 num_bytes
= fs_info
->nodesize
;
3273 ret
= process_func(trans
, fs_info
, bytenr
, num_bytes
,
3274 parent
, ref_root
, level
- 1, 0);
3284 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3285 struct extent_buffer
*buf
, int full_backref
)
3287 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1);
3290 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3291 struct extent_buffer
*buf
, int full_backref
)
3293 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0);
3296 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
3297 struct btrfs_fs_info
*fs_info
,
3298 struct btrfs_path
*path
,
3299 struct btrfs_block_group_cache
*cache
)
3302 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
3304 struct extent_buffer
*leaf
;
3306 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
3313 leaf
= path
->nodes
[0];
3314 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
3315 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
3316 btrfs_mark_buffer_dirty(leaf
);
3318 btrfs_release_path(path
);
3323 static struct btrfs_block_group_cache
*
3324 next_block_group(struct btrfs_fs_info
*fs_info
,
3325 struct btrfs_block_group_cache
*cache
)
3327 struct rb_node
*node
;
3329 spin_lock(&fs_info
->block_group_cache_lock
);
3331 /* If our block group was removed, we need a full search. */
3332 if (RB_EMPTY_NODE(&cache
->cache_node
)) {
3333 const u64 next_bytenr
= cache
->key
.objectid
+ cache
->key
.offset
;
3335 spin_unlock(&fs_info
->block_group_cache_lock
);
3336 btrfs_put_block_group(cache
);
3337 cache
= btrfs_lookup_first_block_group(fs_info
, next_bytenr
); return cache
;
3339 node
= rb_next(&cache
->cache_node
);
3340 btrfs_put_block_group(cache
);
3342 cache
= rb_entry(node
, struct btrfs_block_group_cache
,
3344 btrfs_get_block_group(cache
);
3347 spin_unlock(&fs_info
->block_group_cache_lock
);
3351 static int cache_save_setup(struct btrfs_block_group_cache
*block_group
,
3352 struct btrfs_trans_handle
*trans
,
3353 struct btrfs_path
*path
)
3355 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
3356 struct btrfs_root
*root
= fs_info
->tree_root
;
3357 struct inode
*inode
= NULL
;
3359 int dcs
= BTRFS_DC_ERROR
;
3365 * If this block group is smaller than 100 megs don't bother caching the
3368 if (block_group
->key
.offset
< (100 * SZ_1M
)) {
3369 spin_lock(&block_group
->lock
);
3370 block_group
->disk_cache_state
= BTRFS_DC_WRITTEN
;
3371 spin_unlock(&block_group
->lock
);
3378 inode
= lookup_free_space_inode(fs_info
, block_group
, path
);
3379 if (IS_ERR(inode
) && PTR_ERR(inode
) != -ENOENT
) {
3380 ret
= PTR_ERR(inode
);
3381 btrfs_release_path(path
);
3385 if (IS_ERR(inode
)) {
3389 if (block_group
->ro
)
3392 ret
= create_free_space_inode(fs_info
, trans
, block_group
,
3399 /* We've already setup this transaction, go ahead and exit */
3400 if (block_group
->cache_generation
== trans
->transid
&&
3401 i_size_read(inode
)) {
3402 dcs
= BTRFS_DC_SETUP
;
3407 * We want to set the generation to 0, that way if anything goes wrong
3408 * from here on out we know not to trust this cache when we load up next
3411 BTRFS_I(inode
)->generation
= 0;
3412 ret
= btrfs_update_inode(trans
, root
, inode
);
3415 * So theoretically we could recover from this, simply set the
3416 * super cache generation to 0 so we know to invalidate the
3417 * cache, but then we'd have to keep track of the block groups
3418 * that fail this way so we know we _have_ to reset this cache
3419 * before the next commit or risk reading stale cache. So to
3420 * limit our exposure to horrible edge cases lets just abort the
3421 * transaction, this only happens in really bad situations
3424 btrfs_abort_transaction(trans
, ret
);
3429 if (i_size_read(inode
) > 0) {
3430 ret
= btrfs_check_trunc_cache_free_space(fs_info
,
3431 &fs_info
->global_block_rsv
);
3435 ret
= btrfs_truncate_free_space_cache(trans
, NULL
, inode
);
3440 spin_lock(&block_group
->lock
);
3441 if (block_group
->cached
!= BTRFS_CACHE_FINISHED
||
3442 !btrfs_test_opt(fs_info
, SPACE_CACHE
)) {
3444 * don't bother trying to write stuff out _if_
3445 * a) we're not cached,
3446 * b) we're with nospace_cache mount option.
3448 dcs
= BTRFS_DC_WRITTEN
;
3449 spin_unlock(&block_group
->lock
);
3452 spin_unlock(&block_group
->lock
);
3455 * We hit an ENOSPC when setting up the cache in this transaction, just
3456 * skip doing the setup, we've already cleared the cache so we're safe.
3458 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC
, &trans
->transaction
->flags
)) {
3464 * Try to preallocate enough space based on how big the block group is.
3465 * Keep in mind this has to include any pinned space which could end up
3466 * taking up quite a bit since it's not folded into the other space
3469 num_pages
= div_u64(block_group
->key
.offset
, SZ_256M
);
3474 num_pages
*= PAGE_SIZE
;
3476 ret
= btrfs_check_data_free_space(inode
, 0, num_pages
);
3480 ret
= btrfs_prealloc_file_range_trans(inode
, trans
, 0, 0, num_pages
,
3481 num_pages
, num_pages
,
3484 * Our cache requires contiguous chunks so that we don't modify a bunch
3485 * of metadata or split extents when writing the cache out, which means
3486 * we can enospc if we are heavily fragmented in addition to just normal
3487 * out of space conditions. So if we hit this just skip setting up any
3488 * other block groups for this transaction, maybe we'll unpin enough
3489 * space the next time around.
3492 dcs
= BTRFS_DC_SETUP
;
3493 else if (ret
== -ENOSPC
)
3494 set_bit(BTRFS_TRANS_CACHE_ENOSPC
, &trans
->transaction
->flags
);
3499 btrfs_release_path(path
);
3501 spin_lock(&block_group
->lock
);
3502 if (!ret
&& dcs
== BTRFS_DC_SETUP
)
3503 block_group
->cache_generation
= trans
->transid
;
3504 block_group
->disk_cache_state
= dcs
;
3505 spin_unlock(&block_group
->lock
);
3510 int btrfs_setup_space_cache(struct btrfs_trans_handle
*trans
,
3511 struct btrfs_fs_info
*fs_info
)
3513 struct btrfs_block_group_cache
*cache
, *tmp
;
3514 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3515 struct btrfs_path
*path
;
3517 if (list_empty(&cur_trans
->dirty_bgs
) ||
3518 !btrfs_test_opt(fs_info
, SPACE_CACHE
))
3521 path
= btrfs_alloc_path();
3525 /* Could add new block groups, use _safe just in case */
3526 list_for_each_entry_safe(cache
, tmp
, &cur_trans
->dirty_bgs
,
3528 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
)
3529 cache_save_setup(cache
, trans
, path
);
3532 btrfs_free_path(path
);
3537 * transaction commit does final block group cache writeback during a
3538 * critical section where nothing is allowed to change the FS. This is
3539 * required in order for the cache to actually match the block group,
3540 * but can introduce a lot of latency into the commit.
3542 * So, btrfs_start_dirty_block_groups is here to kick off block group
3543 * cache IO. There's a chance we'll have to redo some of it if the
3544 * block group changes again during the commit, but it greatly reduces
3545 * the commit latency by getting rid of the easy block groups while
3546 * we're still allowing others to join the commit.
3548 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle
*trans
,
3549 struct btrfs_fs_info
*fs_info
)
3551 struct btrfs_block_group_cache
*cache
;
3552 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3555 struct btrfs_path
*path
= NULL
;
3557 struct list_head
*io
= &cur_trans
->io_bgs
;
3558 int num_started
= 0;
3561 spin_lock(&cur_trans
->dirty_bgs_lock
);
3562 if (list_empty(&cur_trans
->dirty_bgs
)) {
3563 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3566 list_splice_init(&cur_trans
->dirty_bgs
, &dirty
);
3567 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3571 * make sure all the block groups on our dirty list actually
3574 btrfs_create_pending_block_groups(trans
, fs_info
);
3577 path
= btrfs_alloc_path();
3583 * cache_write_mutex is here only to save us from balance or automatic
3584 * removal of empty block groups deleting this block group while we are
3585 * writing out the cache
3587 mutex_lock(&trans
->transaction
->cache_write_mutex
);
3588 while (!list_empty(&dirty
)) {
3589 cache
= list_first_entry(&dirty
,
3590 struct btrfs_block_group_cache
,
3593 * this can happen if something re-dirties a block
3594 * group that is already under IO. Just wait for it to
3595 * finish and then do it all again
3597 if (!list_empty(&cache
->io_list
)) {
3598 list_del_init(&cache
->io_list
);
3599 btrfs_wait_cache_io(trans
, cache
, path
);
3600 btrfs_put_block_group(cache
);
3605 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3606 * if it should update the cache_state. Don't delete
3607 * until after we wait.
3609 * Since we're not running in the commit critical section
3610 * we need the dirty_bgs_lock to protect from update_block_group
3612 spin_lock(&cur_trans
->dirty_bgs_lock
);
3613 list_del_init(&cache
->dirty_list
);
3614 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3618 cache_save_setup(cache
, trans
, path
);
3620 if (cache
->disk_cache_state
== BTRFS_DC_SETUP
) {
3621 cache
->io_ctl
.inode
= NULL
;
3622 ret
= btrfs_write_out_cache(fs_info
, trans
,
3624 if (ret
== 0 && cache
->io_ctl
.inode
) {
3629 * the cache_write_mutex is protecting
3632 list_add_tail(&cache
->io_list
, io
);
3635 * if we failed to write the cache, the
3636 * generation will be bad and life goes on
3642 ret
= write_one_cache_group(trans
, fs_info
,
3645 * Our block group might still be attached to the list
3646 * of new block groups in the transaction handle of some
3647 * other task (struct btrfs_trans_handle->new_bgs). This
3648 * means its block group item isn't yet in the extent
3649 * tree. If this happens ignore the error, as we will
3650 * try again later in the critical section of the
3651 * transaction commit.
3653 if (ret
== -ENOENT
) {
3655 spin_lock(&cur_trans
->dirty_bgs_lock
);
3656 if (list_empty(&cache
->dirty_list
)) {
3657 list_add_tail(&cache
->dirty_list
,
3658 &cur_trans
->dirty_bgs
);
3659 btrfs_get_block_group(cache
);
3661 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3663 btrfs_abort_transaction(trans
, ret
);
3667 /* if its not on the io list, we need to put the block group */
3669 btrfs_put_block_group(cache
);
3675 * Avoid blocking other tasks for too long. It might even save
3676 * us from writing caches for block groups that are going to be
3679 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
3680 mutex_lock(&trans
->transaction
->cache_write_mutex
);
3682 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
3685 * go through delayed refs for all the stuff we've just kicked off
3686 * and then loop back (just once)
3688 ret
= btrfs_run_delayed_refs(trans
, fs_info
, 0);
3689 if (!ret
&& loops
== 0) {
3691 spin_lock(&cur_trans
->dirty_bgs_lock
);
3692 list_splice_init(&cur_trans
->dirty_bgs
, &dirty
);
3694 * dirty_bgs_lock protects us from concurrent block group
3695 * deletes too (not just cache_write_mutex).
3697 if (!list_empty(&dirty
)) {
3698 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3701 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3702 } else if (ret
< 0) {
3703 btrfs_cleanup_dirty_bgs(cur_trans
, fs_info
);
3706 btrfs_free_path(path
);
3710 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
3711 struct btrfs_fs_info
*fs_info
)
3713 struct btrfs_block_group_cache
*cache
;
3714 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3717 struct btrfs_path
*path
;
3718 struct list_head
*io
= &cur_trans
->io_bgs
;
3719 int num_started
= 0;
3721 path
= btrfs_alloc_path();
3726 * Even though we are in the critical section of the transaction commit,
3727 * we can still have concurrent tasks adding elements to this
3728 * transaction's list of dirty block groups. These tasks correspond to
3729 * endio free space workers started when writeback finishes for a
3730 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3731 * allocate new block groups as a result of COWing nodes of the root
3732 * tree when updating the free space inode. The writeback for the space
3733 * caches is triggered by an earlier call to
3734 * btrfs_start_dirty_block_groups() and iterations of the following
3736 * Also we want to do the cache_save_setup first and then run the
3737 * delayed refs to make sure we have the best chance at doing this all
3740 spin_lock(&cur_trans
->dirty_bgs_lock
);
3741 while (!list_empty(&cur_trans
->dirty_bgs
)) {
3742 cache
= list_first_entry(&cur_trans
->dirty_bgs
,
3743 struct btrfs_block_group_cache
,
3747 * this can happen if cache_save_setup re-dirties a block
3748 * group that is already under IO. Just wait for it to
3749 * finish and then do it all again
3751 if (!list_empty(&cache
->io_list
)) {
3752 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3753 list_del_init(&cache
->io_list
);
3754 btrfs_wait_cache_io(trans
, cache
, path
);
3755 btrfs_put_block_group(cache
);
3756 spin_lock(&cur_trans
->dirty_bgs_lock
);
3760 * don't remove from the dirty list until after we've waited
3763 list_del_init(&cache
->dirty_list
);
3764 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3767 cache_save_setup(cache
, trans
, path
);
3770 ret
= btrfs_run_delayed_refs(trans
, fs_info
,
3771 (unsigned long) -1);
3773 if (!ret
&& cache
->disk_cache_state
== BTRFS_DC_SETUP
) {
3774 cache
->io_ctl
.inode
= NULL
;
3775 ret
= btrfs_write_out_cache(fs_info
, trans
,
3777 if (ret
== 0 && cache
->io_ctl
.inode
) {
3780 list_add_tail(&cache
->io_list
, io
);
3783 * if we failed to write the cache, the
3784 * generation will be bad and life goes on
3790 ret
= write_one_cache_group(trans
, fs_info
,
3793 * One of the free space endio workers might have
3794 * created a new block group while updating a free space
3795 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3796 * and hasn't released its transaction handle yet, in
3797 * which case the new block group is still attached to
3798 * its transaction handle and its creation has not
3799 * finished yet (no block group item in the extent tree
3800 * yet, etc). If this is the case, wait for all free
3801 * space endio workers to finish and retry. This is a
3802 * a very rare case so no need for a more efficient and
3805 if (ret
== -ENOENT
) {
3806 wait_event(cur_trans
->writer_wait
,
3807 atomic_read(&cur_trans
->num_writers
) == 1);
3808 ret
= write_one_cache_group(trans
, fs_info
,
3812 btrfs_abort_transaction(trans
, ret
);
3815 /* if its not on the io list, we need to put the block group */
3817 btrfs_put_block_group(cache
);
3818 spin_lock(&cur_trans
->dirty_bgs_lock
);
3820 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3822 while (!list_empty(io
)) {
3823 cache
= list_first_entry(io
, struct btrfs_block_group_cache
,
3825 list_del_init(&cache
->io_list
);
3826 btrfs_wait_cache_io(trans
, cache
, path
);
3827 btrfs_put_block_group(cache
);
3830 btrfs_free_path(path
);
3834 int btrfs_extent_readonly(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
3836 struct btrfs_block_group_cache
*block_group
;
3839 block_group
= btrfs_lookup_block_group(fs_info
, bytenr
);
3840 if (!block_group
|| block_group
->ro
)
3843 btrfs_put_block_group(block_group
);
3847 bool btrfs_inc_nocow_writers(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
3849 struct btrfs_block_group_cache
*bg
;
3852 bg
= btrfs_lookup_block_group(fs_info
, bytenr
);
3856 spin_lock(&bg
->lock
);
3860 atomic_inc(&bg
->nocow_writers
);
3861 spin_unlock(&bg
->lock
);
3863 /* no put on block group, done by btrfs_dec_nocow_writers */
3865 btrfs_put_block_group(bg
);
3871 void btrfs_dec_nocow_writers(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
3873 struct btrfs_block_group_cache
*bg
;
3875 bg
= btrfs_lookup_block_group(fs_info
, bytenr
);
3877 if (atomic_dec_and_test(&bg
->nocow_writers
))
3878 wake_up_atomic_t(&bg
->nocow_writers
);
3880 * Once for our lookup and once for the lookup done by a previous call
3881 * to btrfs_inc_nocow_writers()
3883 btrfs_put_block_group(bg
);
3884 btrfs_put_block_group(bg
);
3887 static int btrfs_wait_nocow_writers_atomic_t(atomic_t
*a
)
3893 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache
*bg
)
3895 wait_on_atomic_t(&bg
->nocow_writers
,
3896 btrfs_wait_nocow_writers_atomic_t
,
3897 TASK_UNINTERRUPTIBLE
);
3900 static const char *alloc_name(u64 flags
)
3903 case BTRFS_BLOCK_GROUP_METADATA
|BTRFS_BLOCK_GROUP_DATA
:
3905 case BTRFS_BLOCK_GROUP_METADATA
:
3907 case BTRFS_BLOCK_GROUP_DATA
:
3909 case BTRFS_BLOCK_GROUP_SYSTEM
:
3913 return "invalid-combination";
3917 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
3918 u64 total_bytes
, u64 bytes_used
,
3920 struct btrfs_space_info
**space_info
)
3922 struct btrfs_space_info
*found
;
3927 if (flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3928 BTRFS_BLOCK_GROUP_RAID10
))
3933 found
= __find_space_info(info
, flags
);
3935 spin_lock(&found
->lock
);
3936 found
->total_bytes
+= total_bytes
;
3937 found
->disk_total
+= total_bytes
* factor
;
3938 found
->bytes_used
+= bytes_used
;
3939 found
->disk_used
+= bytes_used
* factor
;
3940 found
->bytes_readonly
+= bytes_readonly
;
3941 if (total_bytes
> 0)
3943 space_info_add_new_bytes(info
, found
, total_bytes
-
3944 bytes_used
- bytes_readonly
);
3945 spin_unlock(&found
->lock
);
3946 *space_info
= found
;
3949 found
= kzalloc(sizeof(*found
), GFP_NOFS
);
3953 ret
= percpu_counter_init(&found
->total_bytes_pinned
, 0, GFP_KERNEL
);
3959 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
3960 INIT_LIST_HEAD(&found
->block_groups
[i
]);
3961 init_rwsem(&found
->groups_sem
);
3962 spin_lock_init(&found
->lock
);
3963 found
->flags
= flags
& BTRFS_BLOCK_GROUP_TYPE_MASK
;
3964 found
->total_bytes
= total_bytes
;
3965 found
->disk_total
= total_bytes
* factor
;
3966 found
->bytes_used
= bytes_used
;
3967 found
->disk_used
= bytes_used
* factor
;
3968 found
->bytes_pinned
= 0;
3969 found
->bytes_reserved
= 0;
3970 found
->bytes_readonly
= bytes_readonly
;
3971 found
->bytes_may_use
= 0;
3973 found
->max_extent_size
= 0;
3974 found
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3975 found
->chunk_alloc
= 0;
3977 init_waitqueue_head(&found
->wait
);
3978 INIT_LIST_HEAD(&found
->ro_bgs
);
3979 INIT_LIST_HEAD(&found
->tickets
);
3980 INIT_LIST_HEAD(&found
->priority_tickets
);
3982 ret
= kobject_init_and_add(&found
->kobj
, &space_info_ktype
,
3983 info
->space_info_kobj
, "%s",
3984 alloc_name(found
->flags
));
3990 *space_info
= found
;
3991 list_add_rcu(&found
->list
, &info
->space_info
);
3992 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3993 info
->data_sinfo
= found
;
3998 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
4000 u64 extra_flags
= chunk_to_extended(flags
) &
4001 BTRFS_EXTENDED_PROFILE_MASK
;
4003 write_seqlock(&fs_info
->profiles_lock
);
4004 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
4005 fs_info
->avail_data_alloc_bits
|= extra_flags
;
4006 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
4007 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
4008 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
4009 fs_info
->avail_system_alloc_bits
|= extra_flags
;
4010 write_sequnlock(&fs_info
->profiles_lock
);
4014 * returns target flags in extended format or 0 if restripe for this
4015 * chunk_type is not in progress
4017 * should be called with either volume_mutex or balance_lock held
4019 static u64
get_restripe_target(struct btrfs_fs_info
*fs_info
, u64 flags
)
4021 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
4027 if (flags
& BTRFS_BLOCK_GROUP_DATA
&&
4028 bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
4029 target
= BTRFS_BLOCK_GROUP_DATA
| bctl
->data
.target
;
4030 } else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
&&
4031 bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
4032 target
= BTRFS_BLOCK_GROUP_SYSTEM
| bctl
->sys
.target
;
4033 } else if (flags
& BTRFS_BLOCK_GROUP_METADATA
&&
4034 bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
4035 target
= BTRFS_BLOCK_GROUP_METADATA
| bctl
->meta
.target
;
4042 * @flags: available profiles in extended format (see ctree.h)
4044 * Returns reduced profile in chunk format. If profile changing is in
4045 * progress (either running or paused) picks the target profile (if it's
4046 * already available), otherwise falls back to plain reducing.
4048 static u64
btrfs_reduce_alloc_profile(struct btrfs_fs_info
*fs_info
, u64 flags
)
4050 u64 num_devices
= fs_info
->fs_devices
->rw_devices
;
4056 * see if restripe for this chunk_type is in progress, if so
4057 * try to reduce to the target profile
4059 spin_lock(&fs_info
->balance_lock
);
4060 target
= get_restripe_target(fs_info
, flags
);
4062 /* pick target profile only if it's already available */
4063 if ((flags
& target
) & BTRFS_EXTENDED_PROFILE_MASK
) {
4064 spin_unlock(&fs_info
->balance_lock
);
4065 return extended_to_chunk(target
);
4068 spin_unlock(&fs_info
->balance_lock
);
4070 /* First, mask out the RAID levels which aren't possible */
4071 for (raid_type
= 0; raid_type
< BTRFS_NR_RAID_TYPES
; raid_type
++) {
4072 if (num_devices
>= btrfs_raid_array
[raid_type
].devs_min
)
4073 allowed
|= btrfs_raid_group
[raid_type
];
4077 if (allowed
& BTRFS_BLOCK_GROUP_RAID6
)
4078 allowed
= BTRFS_BLOCK_GROUP_RAID6
;
4079 else if (allowed
& BTRFS_BLOCK_GROUP_RAID5
)
4080 allowed
= BTRFS_BLOCK_GROUP_RAID5
;
4081 else if (allowed
& BTRFS_BLOCK_GROUP_RAID10
)
4082 allowed
= BTRFS_BLOCK_GROUP_RAID10
;
4083 else if (allowed
& BTRFS_BLOCK_GROUP_RAID1
)
4084 allowed
= BTRFS_BLOCK_GROUP_RAID1
;
4085 else if (allowed
& BTRFS_BLOCK_GROUP_RAID0
)
4086 allowed
= BTRFS_BLOCK_GROUP_RAID0
;
4088 flags
&= ~BTRFS_BLOCK_GROUP_PROFILE_MASK
;
4090 return extended_to_chunk(flags
| allowed
);
4093 static u64
get_alloc_profile(struct btrfs_fs_info
*fs_info
, u64 orig_flags
)
4100 seq
= read_seqbegin(&fs_info
->profiles_lock
);
4102 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
4103 flags
|= fs_info
->avail_data_alloc_bits
;
4104 else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
4105 flags
|= fs_info
->avail_system_alloc_bits
;
4106 else if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
4107 flags
|= fs_info
->avail_metadata_alloc_bits
;
4108 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
4110 return btrfs_reduce_alloc_profile(fs_info
, flags
);
4113 u64
btrfs_get_alloc_profile(struct btrfs_root
*root
, int data
)
4115 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4120 flags
= BTRFS_BLOCK_GROUP_DATA
;
4121 else if (root
== fs_info
->chunk_root
)
4122 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
4124 flags
= BTRFS_BLOCK_GROUP_METADATA
;
4126 ret
= get_alloc_profile(fs_info
, flags
);
4130 static u64
btrfs_space_info_used(struct btrfs_space_info
*s_info
,
4131 bool may_use_included
)
4134 return s_info
->bytes_used
+ s_info
->bytes_reserved
+
4135 s_info
->bytes_pinned
+ s_info
->bytes_readonly
+
4136 (may_use_included
? s_info
->bytes_may_use
: 0);
4139 int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode
*inode
, u64 bytes
)
4141 struct btrfs_space_info
*data_sinfo
;
4142 struct btrfs_root
*root
= inode
->root
;
4143 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4146 int need_commit
= 2;
4147 int have_pinned_space
;
4149 /* make sure bytes are sectorsize aligned */
4150 bytes
= ALIGN(bytes
, fs_info
->sectorsize
);
4152 if (btrfs_is_free_space_inode(inode
)) {
4154 ASSERT(current
->journal_info
);
4157 data_sinfo
= fs_info
->data_sinfo
;
4162 /* make sure we have enough space to handle the data first */
4163 spin_lock(&data_sinfo
->lock
);
4164 used
= btrfs_space_info_used(data_sinfo
, true);
4166 if (used
+ bytes
> data_sinfo
->total_bytes
) {
4167 struct btrfs_trans_handle
*trans
;
4170 * if we don't have enough free bytes in this space then we need
4171 * to alloc a new chunk.
4173 if (!data_sinfo
->full
) {
4176 data_sinfo
->force_alloc
= CHUNK_ALLOC_FORCE
;
4177 spin_unlock(&data_sinfo
->lock
);
4179 alloc_target
= btrfs_get_alloc_profile(root
, 1);
4181 * It is ugly that we don't call nolock join
4182 * transaction for the free space inode case here.
4183 * But it is safe because we only do the data space
4184 * reservation for the free space cache in the
4185 * transaction context, the common join transaction
4186 * just increase the counter of the current transaction
4187 * handler, doesn't try to acquire the trans_lock of
4190 trans
= btrfs_join_transaction(root
);
4192 return PTR_ERR(trans
);
4194 ret
= do_chunk_alloc(trans
, fs_info
, alloc_target
,
4195 CHUNK_ALLOC_NO_FORCE
);
4196 btrfs_end_transaction(trans
);
4201 have_pinned_space
= 1;
4207 data_sinfo
= fs_info
->data_sinfo
;
4213 * If we don't have enough pinned space to deal with this
4214 * allocation, and no removed chunk in current transaction,
4215 * don't bother committing the transaction.
4217 have_pinned_space
= percpu_counter_compare(
4218 &data_sinfo
->total_bytes_pinned
,
4219 used
+ bytes
- data_sinfo
->total_bytes
);
4220 spin_unlock(&data_sinfo
->lock
);
4222 /* commit the current transaction and try again */
4225 !atomic_read(&fs_info
->open_ioctl_trans
)) {
4228 if (need_commit
> 0) {
4229 btrfs_start_delalloc_roots(fs_info
, 0, -1);
4230 btrfs_wait_ordered_roots(fs_info
, -1, 0,
4234 trans
= btrfs_join_transaction(root
);
4236 return PTR_ERR(trans
);
4237 if (have_pinned_space
>= 0 ||
4238 test_bit(BTRFS_TRANS_HAVE_FREE_BGS
,
4239 &trans
->transaction
->flags
) ||
4241 ret
= btrfs_commit_transaction(trans
);
4245 * The cleaner kthread might still be doing iput
4246 * operations. Wait for it to finish so that
4247 * more space is released.
4249 mutex_lock(&fs_info
->cleaner_delayed_iput_mutex
);
4250 mutex_unlock(&fs_info
->cleaner_delayed_iput_mutex
);
4253 btrfs_end_transaction(trans
);
4257 trace_btrfs_space_reservation(fs_info
,
4258 "space_info:enospc",
4259 data_sinfo
->flags
, bytes
, 1);
4262 data_sinfo
->bytes_may_use
+= bytes
;
4263 trace_btrfs_space_reservation(fs_info
, "space_info",
4264 data_sinfo
->flags
, bytes
, 1);
4265 spin_unlock(&data_sinfo
->lock
);
4271 * New check_data_free_space() with ability for precious data reservation
4272 * Will replace old btrfs_check_data_free_space(), but for patch split,
4273 * add a new function first and then replace it.
4275 int btrfs_check_data_free_space(struct inode
*inode
, u64 start
, u64 len
)
4277 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
4280 /* align the range */
4281 len
= round_up(start
+ len
, fs_info
->sectorsize
) -
4282 round_down(start
, fs_info
->sectorsize
);
4283 start
= round_down(start
, fs_info
->sectorsize
);
4285 ret
= btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode
), len
);
4289 /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */
4290 ret
= btrfs_qgroup_reserve_data(inode
, start
, len
);
4292 btrfs_free_reserved_data_space_noquota(inode
, start
, len
);
4297 * Called if we need to clear a data reservation for this inode
4298 * Normally in a error case.
4300 * This one will *NOT* use accurate qgroup reserved space API, just for case
4301 * which we can't sleep and is sure it won't affect qgroup reserved space.
4302 * Like clear_bit_hook().
4304 void btrfs_free_reserved_data_space_noquota(struct inode
*inode
, u64 start
,
4307 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
4308 struct btrfs_space_info
*data_sinfo
;
4310 /* Make sure the range is aligned to sectorsize */
4311 len
= round_up(start
+ len
, fs_info
->sectorsize
) -
4312 round_down(start
, fs_info
->sectorsize
);
4313 start
= round_down(start
, fs_info
->sectorsize
);
4315 data_sinfo
= fs_info
->data_sinfo
;
4316 spin_lock(&data_sinfo
->lock
);
4317 if (WARN_ON(data_sinfo
->bytes_may_use
< len
))
4318 data_sinfo
->bytes_may_use
= 0;
4320 data_sinfo
->bytes_may_use
-= len
;
4321 trace_btrfs_space_reservation(fs_info
, "space_info",
4322 data_sinfo
->flags
, len
, 0);
4323 spin_unlock(&data_sinfo
->lock
);
4327 * Called if we need to clear a data reservation for this inode
4328 * Normally in a error case.
4330 * This one will handle the per-inode data rsv map for accurate reserved
4333 void btrfs_free_reserved_data_space(struct inode
*inode
, u64 start
, u64 len
)
4335 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4337 /* Make sure the range is aligned to sectorsize */
4338 len
= round_up(start
+ len
, root
->fs_info
->sectorsize
) -
4339 round_down(start
, root
->fs_info
->sectorsize
);
4340 start
= round_down(start
, root
->fs_info
->sectorsize
);
4342 btrfs_free_reserved_data_space_noquota(inode
, start
, len
);
4343 btrfs_qgroup_free_data(inode
, start
, len
);
4346 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
4348 struct list_head
*head
= &info
->space_info
;
4349 struct btrfs_space_info
*found
;
4352 list_for_each_entry_rcu(found
, head
, list
) {
4353 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
4354 found
->force_alloc
= CHUNK_ALLOC_FORCE
;
4359 static inline u64
calc_global_rsv_need_space(struct btrfs_block_rsv
*global
)
4361 return (global
->size
<< 1);
4364 static int should_alloc_chunk(struct btrfs_fs_info
*fs_info
,
4365 struct btrfs_space_info
*sinfo
, int force
)
4367 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
4368 u64 num_bytes
= sinfo
->total_bytes
- sinfo
->bytes_readonly
;
4369 u64 num_allocated
= sinfo
->bytes_used
+ sinfo
->bytes_reserved
;
4372 if (force
== CHUNK_ALLOC_FORCE
)
4376 * We need to take into account the global rsv because for all intents
4377 * and purposes it's used space. Don't worry about locking the
4378 * global_rsv, it doesn't change except when the transaction commits.
4380 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
4381 num_allocated
+= calc_global_rsv_need_space(global_rsv
);
4384 * in limited mode, we want to have some free space up to
4385 * about 1% of the FS size.
4387 if (force
== CHUNK_ALLOC_LIMITED
) {
4388 thresh
= btrfs_super_total_bytes(fs_info
->super_copy
);
4389 thresh
= max_t(u64
, SZ_64M
, div_factor_fine(thresh
, 1));
4391 if (num_bytes
- num_allocated
< thresh
)
4395 if (num_allocated
+ SZ_2M
< div_factor(num_bytes
, 8))
4400 static u64
get_profile_num_devs(struct btrfs_fs_info
*fs_info
, u64 type
)
4404 if (type
& (BTRFS_BLOCK_GROUP_RAID10
|
4405 BTRFS_BLOCK_GROUP_RAID0
|
4406 BTRFS_BLOCK_GROUP_RAID5
|
4407 BTRFS_BLOCK_GROUP_RAID6
))
4408 num_dev
= fs_info
->fs_devices
->rw_devices
;
4409 else if (type
& BTRFS_BLOCK_GROUP_RAID1
)
4412 num_dev
= 1; /* DUP or single */
4418 * If @is_allocation is true, reserve space in the system space info necessary
4419 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4422 void check_system_chunk(struct btrfs_trans_handle
*trans
,
4423 struct btrfs_fs_info
*fs_info
, u64 type
)
4425 struct btrfs_space_info
*info
;
4432 * Needed because we can end up allocating a system chunk and for an
4433 * atomic and race free space reservation in the chunk block reserve.
4435 ASSERT(mutex_is_locked(&fs_info
->chunk_mutex
));
4437 info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
4438 spin_lock(&info
->lock
);
4439 left
= info
->total_bytes
- btrfs_space_info_used(info
, true);
4440 spin_unlock(&info
->lock
);
4442 num_devs
= get_profile_num_devs(fs_info
, type
);
4444 /* num_devs device items to update and 1 chunk item to add or remove */
4445 thresh
= btrfs_calc_trunc_metadata_size(fs_info
, num_devs
) +
4446 btrfs_calc_trans_metadata_size(fs_info
, 1);
4448 if (left
< thresh
&& btrfs_test_opt(fs_info
, ENOSPC_DEBUG
)) {
4449 btrfs_info(fs_info
, "left=%llu, need=%llu, flags=%llu",
4450 left
, thresh
, type
);
4451 dump_space_info(fs_info
, info
, 0, 0);
4454 if (left
< thresh
) {
4457 flags
= btrfs_get_alloc_profile(fs_info
->chunk_root
, 0);
4459 * Ignore failure to create system chunk. We might end up not
4460 * needing it, as we might not need to COW all nodes/leafs from
4461 * the paths we visit in the chunk tree (they were already COWed
4462 * or created in the current transaction for example).
4464 ret
= btrfs_alloc_chunk(trans
, fs_info
, flags
);
4468 ret
= btrfs_block_rsv_add(fs_info
->chunk_root
,
4469 &fs_info
->chunk_block_rsv
,
4470 thresh
, BTRFS_RESERVE_NO_FLUSH
);
4472 trans
->chunk_bytes_reserved
+= thresh
;
4477 * If force is CHUNK_ALLOC_FORCE:
4478 * - return 1 if it successfully allocates a chunk,
4479 * - return errors including -ENOSPC otherwise.
4480 * If force is NOT CHUNK_ALLOC_FORCE:
4481 * - return 0 if it doesn't need to allocate a new chunk,
4482 * - return 1 if it successfully allocates a chunk,
4483 * - return errors including -ENOSPC otherwise.
4485 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
4486 struct btrfs_fs_info
*fs_info
, u64 flags
, int force
)
4488 struct btrfs_space_info
*space_info
;
4489 int wait_for_alloc
= 0;
4492 /* Don't re-enter if we're already allocating a chunk */
4493 if (trans
->allocating_chunk
)
4496 space_info
= __find_space_info(fs_info
, flags
);
4498 ret
= update_space_info(fs_info
, flags
, 0, 0, 0, &space_info
);
4499 BUG_ON(ret
); /* -ENOMEM */
4501 BUG_ON(!space_info
); /* Logic error */
4504 spin_lock(&space_info
->lock
);
4505 if (force
< space_info
->force_alloc
)
4506 force
= space_info
->force_alloc
;
4507 if (space_info
->full
) {
4508 if (should_alloc_chunk(fs_info
, space_info
, force
))
4512 spin_unlock(&space_info
->lock
);
4516 if (!should_alloc_chunk(fs_info
, space_info
, force
)) {
4517 spin_unlock(&space_info
->lock
);
4519 } else if (space_info
->chunk_alloc
) {
4522 space_info
->chunk_alloc
= 1;
4525 spin_unlock(&space_info
->lock
);
4527 mutex_lock(&fs_info
->chunk_mutex
);
4530 * The chunk_mutex is held throughout the entirety of a chunk
4531 * allocation, so once we've acquired the chunk_mutex we know that the
4532 * other guy is done and we need to recheck and see if we should
4535 if (wait_for_alloc
) {
4536 mutex_unlock(&fs_info
->chunk_mutex
);
4541 trans
->allocating_chunk
= true;
4544 * If we have mixed data/metadata chunks we want to make sure we keep
4545 * allocating mixed chunks instead of individual chunks.
4547 if (btrfs_mixed_space_info(space_info
))
4548 flags
|= (BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
);
4551 * if we're doing a data chunk, go ahead and make sure that
4552 * we keep a reasonable number of metadata chunks allocated in the
4555 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
4556 fs_info
->data_chunk_allocations
++;
4557 if (!(fs_info
->data_chunk_allocations
%
4558 fs_info
->metadata_ratio
))
4559 force_metadata_allocation(fs_info
);
4563 * Check if we have enough space in SYSTEM chunk because we may need
4564 * to update devices.
4566 check_system_chunk(trans
, fs_info
, flags
);
4568 ret
= btrfs_alloc_chunk(trans
, fs_info
, flags
);
4569 trans
->allocating_chunk
= false;
4571 spin_lock(&space_info
->lock
);
4572 if (ret
< 0 && ret
!= -ENOSPC
)
4575 space_info
->full
= 1;
4579 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
4581 space_info
->chunk_alloc
= 0;
4582 spin_unlock(&space_info
->lock
);
4583 mutex_unlock(&fs_info
->chunk_mutex
);
4585 * When we allocate a new chunk we reserve space in the chunk block
4586 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4587 * add new nodes/leafs to it if we end up needing to do it when
4588 * inserting the chunk item and updating device items as part of the
4589 * second phase of chunk allocation, performed by
4590 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4591 * large number of new block groups to create in our transaction
4592 * handle's new_bgs list to avoid exhausting the chunk block reserve
4593 * in extreme cases - like having a single transaction create many new
4594 * block groups when starting to write out the free space caches of all
4595 * the block groups that were made dirty during the lifetime of the
4598 if (trans
->can_flush_pending_bgs
&&
4599 trans
->chunk_bytes_reserved
>= (u64
)SZ_2M
) {
4600 btrfs_create_pending_block_groups(trans
, fs_info
);
4601 btrfs_trans_release_chunk_metadata(trans
);
4606 static int can_overcommit(struct btrfs_root
*root
,
4607 struct btrfs_space_info
*space_info
, u64 bytes
,
4608 enum btrfs_reserve_flush_enum flush
)
4610 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4611 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
4617 /* Don't overcommit when in mixed mode. */
4618 if (space_info
->flags
& BTRFS_BLOCK_GROUP_DATA
)
4621 profile
= btrfs_get_alloc_profile(root
, 0);
4622 used
= btrfs_space_info_used(space_info
, false);
4625 * We only want to allow over committing if we have lots of actual space
4626 * free, but if we don't have enough space to handle the global reserve
4627 * space then we could end up having a real enospc problem when trying
4628 * to allocate a chunk or some other such important allocation.
4630 spin_lock(&global_rsv
->lock
);
4631 space_size
= calc_global_rsv_need_space(global_rsv
);
4632 spin_unlock(&global_rsv
->lock
);
4633 if (used
+ space_size
>= space_info
->total_bytes
)
4636 used
+= space_info
->bytes_may_use
;
4638 spin_lock(&fs_info
->free_chunk_lock
);
4639 avail
= fs_info
->free_chunk_space
;
4640 spin_unlock(&fs_info
->free_chunk_lock
);
4643 * If we have dup, raid1 or raid10 then only half of the free
4644 * space is actually useable. For raid56, the space info used
4645 * doesn't include the parity drive, so we don't have to
4648 if (profile
& (BTRFS_BLOCK_GROUP_DUP
|
4649 BTRFS_BLOCK_GROUP_RAID1
|
4650 BTRFS_BLOCK_GROUP_RAID10
))
4654 * If we aren't flushing all things, let us overcommit up to
4655 * 1/2th of the space. If we can flush, don't let us overcommit
4656 * too much, let it overcommit up to 1/8 of the space.
4658 if (flush
== BTRFS_RESERVE_FLUSH_ALL
)
4663 if (used
+ bytes
< space_info
->total_bytes
+ avail
)
4668 static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info
*fs_info
,
4669 unsigned long nr_pages
, int nr_items
)
4671 struct super_block
*sb
= fs_info
->sb
;
4673 if (down_read_trylock(&sb
->s_umount
)) {
4674 writeback_inodes_sb_nr(sb
, nr_pages
, WB_REASON_FS_FREE_SPACE
);
4675 up_read(&sb
->s_umount
);
4678 * We needn't worry the filesystem going from r/w to r/o though
4679 * we don't acquire ->s_umount mutex, because the filesystem
4680 * should guarantee the delalloc inodes list be empty after
4681 * the filesystem is readonly(all dirty pages are written to
4684 btrfs_start_delalloc_roots(fs_info
, 0, nr_items
);
4685 if (!current
->journal_info
)
4686 btrfs_wait_ordered_roots(fs_info
, nr_items
, 0, (u64
)-1);
4690 static inline int calc_reclaim_items_nr(struct btrfs_fs_info
*fs_info
,
4696 bytes
= btrfs_calc_trans_metadata_size(fs_info
, 1);
4697 nr
= (int)div64_u64(to_reclaim
, bytes
);
4703 #define EXTENT_SIZE_PER_ITEM SZ_256K
4706 * shrink metadata reservation for delalloc
4708 static void shrink_delalloc(struct btrfs_root
*root
, u64 to_reclaim
, u64 orig
,
4711 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4712 struct btrfs_block_rsv
*block_rsv
;
4713 struct btrfs_space_info
*space_info
;
4714 struct btrfs_trans_handle
*trans
;
4718 unsigned long nr_pages
;
4721 enum btrfs_reserve_flush_enum flush
;
4723 /* Calc the number of the pages we need flush for space reservation */
4724 items
= calc_reclaim_items_nr(fs_info
, to_reclaim
);
4725 to_reclaim
= (u64
)items
* EXTENT_SIZE_PER_ITEM
;
4727 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4728 block_rsv
= &fs_info
->delalloc_block_rsv
;
4729 space_info
= block_rsv
->space_info
;
4731 delalloc_bytes
= percpu_counter_sum_positive(
4732 &fs_info
->delalloc_bytes
);
4733 if (delalloc_bytes
== 0) {
4737 btrfs_wait_ordered_roots(fs_info
, items
, 0, (u64
)-1);
4742 while (delalloc_bytes
&& loops
< 3) {
4743 max_reclaim
= min(delalloc_bytes
, to_reclaim
);
4744 nr_pages
= max_reclaim
>> PAGE_SHIFT
;
4745 btrfs_writeback_inodes_sb_nr(fs_info
, nr_pages
, items
);
4747 * We need to wait for the async pages to actually start before
4750 max_reclaim
= atomic_read(&fs_info
->async_delalloc_pages
);
4754 if (max_reclaim
<= nr_pages
)
4757 max_reclaim
-= nr_pages
;
4759 wait_event(fs_info
->async_submit_wait
,
4760 atomic_read(&fs_info
->async_delalloc_pages
) <=
4764 flush
= BTRFS_RESERVE_FLUSH_ALL
;
4766 flush
= BTRFS_RESERVE_NO_FLUSH
;
4767 spin_lock(&space_info
->lock
);
4768 if (can_overcommit(root
, space_info
, orig
, flush
)) {
4769 spin_unlock(&space_info
->lock
);
4772 if (list_empty(&space_info
->tickets
) &&
4773 list_empty(&space_info
->priority_tickets
)) {
4774 spin_unlock(&space_info
->lock
);
4777 spin_unlock(&space_info
->lock
);
4780 if (wait_ordered
&& !trans
) {
4781 btrfs_wait_ordered_roots(fs_info
, items
, 0, (u64
)-1);
4783 time_left
= schedule_timeout_killable(1);
4787 delalloc_bytes
= percpu_counter_sum_positive(
4788 &fs_info
->delalloc_bytes
);
4793 * maybe_commit_transaction - possibly commit the transaction if its ok to
4794 * @root - the root we're allocating for
4795 * @bytes - the number of bytes we want to reserve
4796 * @force - force the commit
4798 * This will check to make sure that committing the transaction will actually
4799 * get us somewhere and then commit the transaction if it does. Otherwise it
4800 * will return -ENOSPC.
4802 static int may_commit_transaction(struct btrfs_fs_info
*fs_info
,
4803 struct btrfs_space_info
*space_info
,
4804 u64 bytes
, int force
)
4806 struct btrfs_block_rsv
*delayed_rsv
= &fs_info
->delayed_block_rsv
;
4807 struct btrfs_trans_handle
*trans
;
4809 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4816 /* See if there is enough pinned space to make this reservation */
4817 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4822 * See if there is some space in the delayed insertion reservation for
4825 if (space_info
!= delayed_rsv
->space_info
)
4828 spin_lock(&delayed_rsv
->lock
);
4829 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4830 bytes
- delayed_rsv
->size
) >= 0) {
4831 spin_unlock(&delayed_rsv
->lock
);
4834 spin_unlock(&delayed_rsv
->lock
);
4837 trans
= btrfs_join_transaction(fs_info
->fs_root
);
4841 return btrfs_commit_transaction(trans
);
4844 struct reserve_ticket
{
4847 struct list_head list
;
4848 wait_queue_head_t wait
;
4851 static int flush_space(struct btrfs_fs_info
*fs_info
,
4852 struct btrfs_space_info
*space_info
, u64 num_bytes
,
4853 u64 orig_bytes
, int state
)
4855 struct btrfs_root
*root
= fs_info
->fs_root
;
4856 struct btrfs_trans_handle
*trans
;
4861 case FLUSH_DELAYED_ITEMS_NR
:
4862 case FLUSH_DELAYED_ITEMS
:
4863 if (state
== FLUSH_DELAYED_ITEMS_NR
)
4864 nr
= calc_reclaim_items_nr(fs_info
, num_bytes
) * 2;
4868 trans
= btrfs_join_transaction(root
);
4869 if (IS_ERR(trans
)) {
4870 ret
= PTR_ERR(trans
);
4873 ret
= btrfs_run_delayed_items_nr(trans
, fs_info
, nr
);
4874 btrfs_end_transaction(trans
);
4876 case FLUSH_DELALLOC
:
4877 case FLUSH_DELALLOC_WAIT
:
4878 shrink_delalloc(root
, num_bytes
* 2, orig_bytes
,
4879 state
== FLUSH_DELALLOC_WAIT
);
4882 trans
= btrfs_join_transaction(root
);
4883 if (IS_ERR(trans
)) {
4884 ret
= PTR_ERR(trans
);
4887 ret
= do_chunk_alloc(trans
, fs_info
,
4888 btrfs_get_alloc_profile(root
, 0),
4889 CHUNK_ALLOC_NO_FORCE
);
4890 btrfs_end_transaction(trans
);
4891 if (ret
> 0 || ret
== -ENOSPC
)
4895 ret
= may_commit_transaction(fs_info
, space_info
,
4903 trace_btrfs_flush_space(fs_info
, space_info
->flags
, num_bytes
,
4904 orig_bytes
, state
, ret
);
4909 btrfs_calc_reclaim_metadata_size(struct btrfs_root
*root
,
4910 struct btrfs_space_info
*space_info
)
4912 struct reserve_ticket
*ticket
;
4917 list_for_each_entry(ticket
, &space_info
->tickets
, list
)
4918 to_reclaim
+= ticket
->bytes
;
4919 list_for_each_entry(ticket
, &space_info
->priority_tickets
, list
)
4920 to_reclaim
+= ticket
->bytes
;
4924 to_reclaim
= min_t(u64
, num_online_cpus() * SZ_1M
, SZ_16M
);
4925 if (can_overcommit(root
, space_info
, to_reclaim
,
4926 BTRFS_RESERVE_FLUSH_ALL
))
4929 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4930 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4931 space_info
->bytes_may_use
;
4932 if (can_overcommit(root
, space_info
, SZ_1M
, BTRFS_RESERVE_FLUSH_ALL
))
4933 expected
= div_factor_fine(space_info
->total_bytes
, 95);
4935 expected
= div_factor_fine(space_info
->total_bytes
, 90);
4937 if (used
> expected
)
4938 to_reclaim
= used
- expected
;
4941 to_reclaim
= min(to_reclaim
, space_info
->bytes_may_use
+
4942 space_info
->bytes_reserved
);
4946 static inline int need_do_async_reclaim(struct btrfs_space_info
*space_info
,
4947 struct btrfs_root
*root
, u64 used
)
4949 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4950 u64 thresh
= div_factor_fine(space_info
->total_bytes
, 98);
4952 /* If we're just plain full then async reclaim just slows us down. */
4953 if ((space_info
->bytes_used
+ space_info
->bytes_reserved
) >= thresh
)
4956 if (!btrfs_calc_reclaim_metadata_size(root
, space_info
))
4959 return (used
>= thresh
&& !btrfs_fs_closing(fs_info
) &&
4960 !test_bit(BTRFS_FS_STATE_REMOUNTING
, &fs_info
->fs_state
));
4963 static void wake_all_tickets(struct list_head
*head
)
4965 struct reserve_ticket
*ticket
;
4967 while (!list_empty(head
)) {
4968 ticket
= list_first_entry(head
, struct reserve_ticket
, list
);
4969 list_del_init(&ticket
->list
);
4970 ticket
->error
= -ENOSPC
;
4971 wake_up(&ticket
->wait
);
4976 * This is for normal flushers, we can wait all goddamned day if we want to. We
4977 * will loop and continuously try to flush as long as we are making progress.
4978 * We count progress as clearing off tickets each time we have to loop.
4980 static void btrfs_async_reclaim_metadata_space(struct work_struct
*work
)
4982 struct btrfs_fs_info
*fs_info
;
4983 struct btrfs_space_info
*space_info
;
4986 int commit_cycles
= 0;
4987 u64 last_tickets_id
;
4989 fs_info
= container_of(work
, struct btrfs_fs_info
, async_reclaim_work
);
4990 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4992 spin_lock(&space_info
->lock
);
4993 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
->fs_root
,
4996 space_info
->flush
= 0;
4997 spin_unlock(&space_info
->lock
);
5000 last_tickets_id
= space_info
->tickets_id
;
5001 spin_unlock(&space_info
->lock
);
5003 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
5005 struct reserve_ticket
*ticket
;
5008 ret
= flush_space(fs_info
, space_info
, to_reclaim
, to_reclaim
,
5010 spin_lock(&space_info
->lock
);
5011 if (list_empty(&space_info
->tickets
)) {
5012 space_info
->flush
= 0;
5013 spin_unlock(&space_info
->lock
);
5016 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
->fs_root
,
5018 ticket
= list_first_entry(&space_info
->tickets
,
5019 struct reserve_ticket
, list
);
5020 if (last_tickets_id
== space_info
->tickets_id
) {
5023 last_tickets_id
= space_info
->tickets_id
;
5024 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
5029 if (flush_state
> COMMIT_TRANS
) {
5031 if (commit_cycles
> 2) {
5032 wake_all_tickets(&space_info
->tickets
);
5033 space_info
->flush
= 0;
5035 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
5038 spin_unlock(&space_info
->lock
);
5039 } while (flush_state
<= COMMIT_TRANS
);
5042 void btrfs_init_async_reclaim_work(struct work_struct
*work
)
5044 INIT_WORK(work
, btrfs_async_reclaim_metadata_space
);
5047 static void priority_reclaim_metadata_space(struct btrfs_fs_info
*fs_info
,
5048 struct btrfs_space_info
*space_info
,
5049 struct reserve_ticket
*ticket
)
5052 int flush_state
= FLUSH_DELAYED_ITEMS_NR
;
5054 spin_lock(&space_info
->lock
);
5055 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
->fs_root
,
5058 spin_unlock(&space_info
->lock
);
5061 spin_unlock(&space_info
->lock
);
5064 flush_space(fs_info
, space_info
, to_reclaim
, to_reclaim
,
5067 spin_lock(&space_info
->lock
);
5068 if (ticket
->bytes
== 0) {
5069 spin_unlock(&space_info
->lock
);
5072 spin_unlock(&space_info
->lock
);
5075 * Priority flushers can't wait on delalloc without
5078 if (flush_state
== FLUSH_DELALLOC
||
5079 flush_state
== FLUSH_DELALLOC_WAIT
)
5080 flush_state
= ALLOC_CHUNK
;
5081 } while (flush_state
< COMMIT_TRANS
);
5084 static int wait_reserve_ticket(struct btrfs_fs_info
*fs_info
,
5085 struct btrfs_space_info
*space_info
,
5086 struct reserve_ticket
*ticket
, u64 orig_bytes
)
5092 spin_lock(&space_info
->lock
);
5093 while (ticket
->bytes
> 0 && ticket
->error
== 0) {
5094 ret
= prepare_to_wait_event(&ticket
->wait
, &wait
, TASK_KILLABLE
);
5099 spin_unlock(&space_info
->lock
);
5103 finish_wait(&ticket
->wait
, &wait
);
5104 spin_lock(&space_info
->lock
);
5107 ret
= ticket
->error
;
5108 if (!list_empty(&ticket
->list
))
5109 list_del_init(&ticket
->list
);
5110 if (ticket
->bytes
&& ticket
->bytes
< orig_bytes
) {
5111 u64 num_bytes
= orig_bytes
- ticket
->bytes
;
5112 space_info
->bytes_may_use
-= num_bytes
;
5113 trace_btrfs_space_reservation(fs_info
, "space_info",
5114 space_info
->flags
, num_bytes
, 0);
5116 spin_unlock(&space_info
->lock
);
5122 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5123 * @root - the root we're allocating for
5124 * @space_info - the space info we want to allocate from
5125 * @orig_bytes - the number of bytes we want
5126 * @flush - whether or not we can flush to make our reservation
5128 * This will reserve orig_bytes number of bytes from the space info associated
5129 * with the block_rsv. If there is not enough space it will make an attempt to
5130 * flush out space to make room. It will do this by flushing delalloc if
5131 * possible or committing the transaction. If flush is 0 then no attempts to
5132 * regain reservations will be made and this will fail if there is not enough
5135 static int __reserve_metadata_bytes(struct btrfs_root
*root
,
5136 struct btrfs_space_info
*space_info
,
5138 enum btrfs_reserve_flush_enum flush
)
5140 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5141 struct reserve_ticket ticket
;
5146 ASSERT(!current
->journal_info
|| flush
!= BTRFS_RESERVE_FLUSH_ALL
);
5148 spin_lock(&space_info
->lock
);
5150 used
= btrfs_space_info_used(space_info
, true);
5153 * If we have enough space then hooray, make our reservation and carry
5154 * on. If not see if we can overcommit, and if we can, hooray carry on.
5155 * If not things get more complicated.
5157 if (used
+ orig_bytes
<= space_info
->total_bytes
) {
5158 space_info
->bytes_may_use
+= orig_bytes
;
5159 trace_btrfs_space_reservation(fs_info
, "space_info",
5160 space_info
->flags
, orig_bytes
, 1);
5162 } else if (can_overcommit(root
, space_info
, orig_bytes
, flush
)) {
5163 space_info
->bytes_may_use
+= orig_bytes
;
5164 trace_btrfs_space_reservation(fs_info
, "space_info",
5165 space_info
->flags
, orig_bytes
, 1);
5170 * If we couldn't make a reservation then setup our reservation ticket
5171 * and kick the async worker if it's not already running.
5173 * If we are a priority flusher then we just need to add our ticket to
5174 * the list and we will do our own flushing further down.
5176 if (ret
&& flush
!= BTRFS_RESERVE_NO_FLUSH
) {
5177 ticket
.bytes
= orig_bytes
;
5179 init_waitqueue_head(&ticket
.wait
);
5180 if (flush
== BTRFS_RESERVE_FLUSH_ALL
) {
5181 list_add_tail(&ticket
.list
, &space_info
->tickets
);
5182 if (!space_info
->flush
) {
5183 space_info
->flush
= 1;
5184 trace_btrfs_trigger_flush(fs_info
,
5188 queue_work(system_unbound_wq
,
5189 &root
->fs_info
->async_reclaim_work
);
5192 list_add_tail(&ticket
.list
,
5193 &space_info
->priority_tickets
);
5195 } else if (!ret
&& space_info
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
5198 * We will do the space reservation dance during log replay,
5199 * which means we won't have fs_info->fs_root set, so don't do
5200 * the async reclaim as we will panic.
5202 if (!test_bit(BTRFS_FS_LOG_RECOVERING
, &fs_info
->flags
) &&
5203 need_do_async_reclaim(space_info
, root
, used
) &&
5204 !work_busy(&fs_info
->async_reclaim_work
)) {
5205 trace_btrfs_trigger_flush(fs_info
, space_info
->flags
,
5206 orig_bytes
, flush
, "preempt");
5207 queue_work(system_unbound_wq
,
5208 &fs_info
->async_reclaim_work
);
5211 spin_unlock(&space_info
->lock
);
5212 if (!ret
|| flush
== BTRFS_RESERVE_NO_FLUSH
)
5215 if (flush
== BTRFS_RESERVE_FLUSH_ALL
)
5216 return wait_reserve_ticket(fs_info
, space_info
, &ticket
,
5220 priority_reclaim_metadata_space(fs_info
, space_info
, &ticket
);
5221 spin_lock(&space_info
->lock
);
5223 if (ticket
.bytes
< orig_bytes
) {
5224 u64 num_bytes
= orig_bytes
- ticket
.bytes
;
5225 space_info
->bytes_may_use
-= num_bytes
;
5226 trace_btrfs_space_reservation(fs_info
, "space_info",
5231 list_del_init(&ticket
.list
);
5234 spin_unlock(&space_info
->lock
);
5235 ASSERT(list_empty(&ticket
.list
));
5240 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5241 * @root - the root we're allocating for
5242 * @block_rsv - the block_rsv we're allocating for
5243 * @orig_bytes - the number of bytes we want
5244 * @flush - whether or not we can flush to make our reservation
5246 * This will reserve orgi_bytes number of bytes from the space info associated
5247 * with the block_rsv. If there is not enough space it will make an attempt to
5248 * flush out space to make room. It will do this by flushing delalloc if
5249 * possible or committing the transaction. If flush is 0 then no attempts to
5250 * regain reservations will be made and this will fail if there is not enough
5253 static int reserve_metadata_bytes(struct btrfs_root
*root
,
5254 struct btrfs_block_rsv
*block_rsv
,
5256 enum btrfs_reserve_flush_enum flush
)
5258 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5259 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5262 ret
= __reserve_metadata_bytes(root
, block_rsv
->space_info
, orig_bytes
,
5264 if (ret
== -ENOSPC
&&
5265 unlikely(root
->orphan_cleanup_state
== ORPHAN_CLEANUP_STARTED
)) {
5266 if (block_rsv
!= global_rsv
&&
5267 !block_rsv_use_bytes(global_rsv
, orig_bytes
))
5271 trace_btrfs_space_reservation(fs_info
, "space_info:enospc",
5272 block_rsv
->space_info
->flags
,
5277 static struct btrfs_block_rsv
*get_block_rsv(
5278 const struct btrfs_trans_handle
*trans
,
5279 const struct btrfs_root
*root
)
5281 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5282 struct btrfs_block_rsv
*block_rsv
= NULL
;
5284 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
5285 (root
== fs_info
->csum_root
&& trans
->adding_csums
) ||
5286 (root
== fs_info
->uuid_root
))
5287 block_rsv
= trans
->block_rsv
;
5290 block_rsv
= root
->block_rsv
;
5293 block_rsv
= &fs_info
->empty_block_rsv
;
5298 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
5302 spin_lock(&block_rsv
->lock
);
5303 if (block_rsv
->reserved
>= num_bytes
) {
5304 block_rsv
->reserved
-= num_bytes
;
5305 if (block_rsv
->reserved
< block_rsv
->size
)
5306 block_rsv
->full
= 0;
5309 spin_unlock(&block_rsv
->lock
);
5313 static void block_rsv_add_bytes(struct btrfs_block_rsv
*block_rsv
,
5314 u64 num_bytes
, int update_size
)
5316 spin_lock(&block_rsv
->lock
);
5317 block_rsv
->reserved
+= num_bytes
;
5319 block_rsv
->size
+= num_bytes
;
5320 else if (block_rsv
->reserved
>= block_rsv
->size
)
5321 block_rsv
->full
= 1;
5322 spin_unlock(&block_rsv
->lock
);
5325 int btrfs_cond_migrate_bytes(struct btrfs_fs_info
*fs_info
,
5326 struct btrfs_block_rsv
*dest
, u64 num_bytes
,
5329 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5332 if (global_rsv
->space_info
!= dest
->space_info
)
5335 spin_lock(&global_rsv
->lock
);
5336 min_bytes
= div_factor(global_rsv
->size
, min_factor
);
5337 if (global_rsv
->reserved
< min_bytes
+ num_bytes
) {
5338 spin_unlock(&global_rsv
->lock
);
5341 global_rsv
->reserved
-= num_bytes
;
5342 if (global_rsv
->reserved
< global_rsv
->size
)
5343 global_rsv
->full
= 0;
5344 spin_unlock(&global_rsv
->lock
);
5346 block_rsv_add_bytes(dest
, num_bytes
, 1);
5351 * This is for space we already have accounted in space_info->bytes_may_use, so
5352 * basically when we're returning space from block_rsv's.
5354 static void space_info_add_old_bytes(struct btrfs_fs_info
*fs_info
,
5355 struct btrfs_space_info
*space_info
,
5358 struct reserve_ticket
*ticket
;
5359 struct list_head
*head
;
5361 enum btrfs_reserve_flush_enum flush
= BTRFS_RESERVE_NO_FLUSH
;
5362 bool check_overcommit
= false;
5364 spin_lock(&space_info
->lock
);
5365 head
= &space_info
->priority_tickets
;
5368 * If we are over our limit then we need to check and see if we can
5369 * overcommit, and if we can't then we just need to free up our space
5370 * and not satisfy any requests.
5372 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
5373 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
5374 space_info
->bytes_may_use
;
5375 if (used
- num_bytes
>= space_info
->total_bytes
)
5376 check_overcommit
= true;
5378 while (!list_empty(head
) && num_bytes
) {
5379 ticket
= list_first_entry(head
, struct reserve_ticket
,
5382 * We use 0 bytes because this space is already reserved, so
5383 * adding the ticket space would be a double count.
5385 if (check_overcommit
&&
5386 !can_overcommit(fs_info
->extent_root
, space_info
, 0,
5389 if (num_bytes
>= ticket
->bytes
) {
5390 list_del_init(&ticket
->list
);
5391 num_bytes
-= ticket
->bytes
;
5393 space_info
->tickets_id
++;
5394 wake_up(&ticket
->wait
);
5396 ticket
->bytes
-= num_bytes
;
5401 if (num_bytes
&& head
== &space_info
->priority_tickets
) {
5402 head
= &space_info
->tickets
;
5403 flush
= BTRFS_RESERVE_FLUSH_ALL
;
5406 space_info
->bytes_may_use
-= num_bytes
;
5407 trace_btrfs_space_reservation(fs_info
, "space_info",
5408 space_info
->flags
, num_bytes
, 0);
5409 spin_unlock(&space_info
->lock
);
5413 * This is for newly allocated space that isn't accounted in
5414 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5415 * we use this helper.
5417 static void space_info_add_new_bytes(struct btrfs_fs_info
*fs_info
,
5418 struct btrfs_space_info
*space_info
,
5421 struct reserve_ticket
*ticket
;
5422 struct list_head
*head
= &space_info
->priority_tickets
;
5425 while (!list_empty(head
) && num_bytes
) {
5426 ticket
= list_first_entry(head
, struct reserve_ticket
,
5428 if (num_bytes
>= ticket
->bytes
) {
5429 trace_btrfs_space_reservation(fs_info
, "space_info",
5432 list_del_init(&ticket
->list
);
5433 num_bytes
-= ticket
->bytes
;
5434 space_info
->bytes_may_use
+= ticket
->bytes
;
5436 space_info
->tickets_id
++;
5437 wake_up(&ticket
->wait
);
5439 trace_btrfs_space_reservation(fs_info
, "space_info",
5442 space_info
->bytes_may_use
+= num_bytes
;
5443 ticket
->bytes
-= num_bytes
;
5448 if (num_bytes
&& head
== &space_info
->priority_tickets
) {
5449 head
= &space_info
->tickets
;
5454 static void block_rsv_release_bytes(struct btrfs_fs_info
*fs_info
,
5455 struct btrfs_block_rsv
*block_rsv
,
5456 struct btrfs_block_rsv
*dest
, u64 num_bytes
)
5458 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
5460 spin_lock(&block_rsv
->lock
);
5461 if (num_bytes
== (u64
)-1)
5462 num_bytes
= block_rsv
->size
;
5463 block_rsv
->size
-= num_bytes
;
5464 if (block_rsv
->reserved
>= block_rsv
->size
) {
5465 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
5466 block_rsv
->reserved
= block_rsv
->size
;
5467 block_rsv
->full
= 1;
5471 spin_unlock(&block_rsv
->lock
);
5473 if (num_bytes
> 0) {
5475 spin_lock(&dest
->lock
);
5479 bytes_to_add
= dest
->size
- dest
->reserved
;
5480 bytes_to_add
= min(num_bytes
, bytes_to_add
);
5481 dest
->reserved
+= bytes_to_add
;
5482 if (dest
->reserved
>= dest
->size
)
5484 num_bytes
-= bytes_to_add
;
5486 spin_unlock(&dest
->lock
);
5489 space_info_add_old_bytes(fs_info
, space_info
,
5494 int btrfs_block_rsv_migrate(struct btrfs_block_rsv
*src
,
5495 struct btrfs_block_rsv
*dst
, u64 num_bytes
,
5500 ret
= block_rsv_use_bytes(src
, num_bytes
);
5504 block_rsv_add_bytes(dst
, num_bytes
, update_size
);
5508 void btrfs_init_block_rsv(struct btrfs_block_rsv
*rsv
, unsigned short type
)
5510 memset(rsv
, 0, sizeof(*rsv
));
5511 spin_lock_init(&rsv
->lock
);
5515 struct btrfs_block_rsv
*btrfs_alloc_block_rsv(struct btrfs_fs_info
*fs_info
,
5516 unsigned short type
)
5518 struct btrfs_block_rsv
*block_rsv
;
5520 block_rsv
= kmalloc(sizeof(*block_rsv
), GFP_NOFS
);
5524 btrfs_init_block_rsv(block_rsv
, type
);
5525 block_rsv
->space_info
= __find_space_info(fs_info
,
5526 BTRFS_BLOCK_GROUP_METADATA
);
5530 void btrfs_free_block_rsv(struct btrfs_fs_info
*fs_info
,
5531 struct btrfs_block_rsv
*rsv
)
5535 btrfs_block_rsv_release(fs_info
, rsv
, (u64
)-1);
5539 void __btrfs_free_block_rsv(struct btrfs_block_rsv
*rsv
)
5544 int btrfs_block_rsv_add(struct btrfs_root
*root
,
5545 struct btrfs_block_rsv
*block_rsv
, u64 num_bytes
,
5546 enum btrfs_reserve_flush_enum flush
)
5553 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
5555 block_rsv_add_bytes(block_rsv
, num_bytes
, 1);
5562 int btrfs_block_rsv_check(struct btrfs_block_rsv
*block_rsv
, int min_factor
)
5570 spin_lock(&block_rsv
->lock
);
5571 num_bytes
= div_factor(block_rsv
->size
, min_factor
);
5572 if (block_rsv
->reserved
>= num_bytes
)
5574 spin_unlock(&block_rsv
->lock
);
5579 int btrfs_block_rsv_refill(struct btrfs_root
*root
,
5580 struct btrfs_block_rsv
*block_rsv
, u64 min_reserved
,
5581 enum btrfs_reserve_flush_enum flush
)
5589 spin_lock(&block_rsv
->lock
);
5590 num_bytes
= min_reserved
;
5591 if (block_rsv
->reserved
>= num_bytes
)
5594 num_bytes
-= block_rsv
->reserved
;
5595 spin_unlock(&block_rsv
->lock
);
5600 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
5602 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
5609 void btrfs_block_rsv_release(struct btrfs_fs_info
*fs_info
,
5610 struct btrfs_block_rsv
*block_rsv
,
5613 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5615 if (global_rsv
== block_rsv
||
5616 block_rsv
->space_info
!= global_rsv
->space_info
)
5618 block_rsv_release_bytes(fs_info
, block_rsv
, global_rsv
, num_bytes
);
5621 static void update_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5623 struct btrfs_block_rsv
*block_rsv
= &fs_info
->global_block_rsv
;
5624 struct btrfs_space_info
*sinfo
= block_rsv
->space_info
;
5628 * The global block rsv is based on the size of the extent tree, the
5629 * checksum tree and the root tree. If the fs is empty we want to set
5630 * it to a minimal amount for safety.
5632 num_bytes
= btrfs_root_used(&fs_info
->extent_root
->root_item
) +
5633 btrfs_root_used(&fs_info
->csum_root
->root_item
) +
5634 btrfs_root_used(&fs_info
->tree_root
->root_item
);
5635 num_bytes
= max_t(u64
, num_bytes
, SZ_16M
);
5637 spin_lock(&sinfo
->lock
);
5638 spin_lock(&block_rsv
->lock
);
5640 block_rsv
->size
= min_t(u64
, num_bytes
, SZ_512M
);
5642 if (block_rsv
->reserved
< block_rsv
->size
) {
5643 num_bytes
= btrfs_space_info_used(sinfo
, true);
5644 if (sinfo
->total_bytes
> num_bytes
) {
5645 num_bytes
= sinfo
->total_bytes
- num_bytes
;
5646 num_bytes
= min(num_bytes
,
5647 block_rsv
->size
- block_rsv
->reserved
);
5648 block_rsv
->reserved
+= num_bytes
;
5649 sinfo
->bytes_may_use
+= num_bytes
;
5650 trace_btrfs_space_reservation(fs_info
, "space_info",
5651 sinfo
->flags
, num_bytes
,
5654 } else if (block_rsv
->reserved
> block_rsv
->size
) {
5655 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
5656 sinfo
->bytes_may_use
-= num_bytes
;
5657 trace_btrfs_space_reservation(fs_info
, "space_info",
5658 sinfo
->flags
, num_bytes
, 0);
5659 block_rsv
->reserved
= block_rsv
->size
;
5662 if (block_rsv
->reserved
== block_rsv
->size
)
5663 block_rsv
->full
= 1;
5665 block_rsv
->full
= 0;
5667 spin_unlock(&block_rsv
->lock
);
5668 spin_unlock(&sinfo
->lock
);
5671 static void init_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5673 struct btrfs_space_info
*space_info
;
5675 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
5676 fs_info
->chunk_block_rsv
.space_info
= space_info
;
5678 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
5679 fs_info
->global_block_rsv
.space_info
= space_info
;
5680 fs_info
->delalloc_block_rsv
.space_info
= space_info
;
5681 fs_info
->trans_block_rsv
.space_info
= space_info
;
5682 fs_info
->empty_block_rsv
.space_info
= space_info
;
5683 fs_info
->delayed_block_rsv
.space_info
= space_info
;
5685 fs_info
->extent_root
->block_rsv
= &fs_info
->global_block_rsv
;
5686 fs_info
->csum_root
->block_rsv
= &fs_info
->global_block_rsv
;
5687 fs_info
->dev_root
->block_rsv
= &fs_info
->global_block_rsv
;
5688 fs_info
->tree_root
->block_rsv
= &fs_info
->global_block_rsv
;
5689 if (fs_info
->quota_root
)
5690 fs_info
->quota_root
->block_rsv
= &fs_info
->global_block_rsv
;
5691 fs_info
->chunk_root
->block_rsv
= &fs_info
->chunk_block_rsv
;
5693 update_global_block_rsv(fs_info
);
5696 static void release_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5698 block_rsv_release_bytes(fs_info
, &fs_info
->global_block_rsv
, NULL
,
5700 WARN_ON(fs_info
->delalloc_block_rsv
.size
> 0);
5701 WARN_ON(fs_info
->delalloc_block_rsv
.reserved
> 0);
5702 WARN_ON(fs_info
->trans_block_rsv
.size
> 0);
5703 WARN_ON(fs_info
->trans_block_rsv
.reserved
> 0);
5704 WARN_ON(fs_info
->chunk_block_rsv
.size
> 0);
5705 WARN_ON(fs_info
->chunk_block_rsv
.reserved
> 0);
5706 WARN_ON(fs_info
->delayed_block_rsv
.size
> 0);
5707 WARN_ON(fs_info
->delayed_block_rsv
.reserved
> 0);
5710 void btrfs_trans_release_metadata(struct btrfs_trans_handle
*trans
,
5711 struct btrfs_fs_info
*fs_info
)
5713 if (!trans
->block_rsv
)
5716 if (!trans
->bytes_reserved
)
5719 trace_btrfs_space_reservation(fs_info
, "transaction",
5720 trans
->transid
, trans
->bytes_reserved
, 0);
5721 btrfs_block_rsv_release(fs_info
, trans
->block_rsv
,
5722 trans
->bytes_reserved
);
5723 trans
->bytes_reserved
= 0;
5727 * To be called after all the new block groups attached to the transaction
5728 * handle have been created (btrfs_create_pending_block_groups()).
5730 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle
*trans
)
5732 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
5734 if (!trans
->chunk_bytes_reserved
)
5737 WARN_ON_ONCE(!list_empty(&trans
->new_bgs
));
5739 block_rsv_release_bytes(fs_info
, &fs_info
->chunk_block_rsv
, NULL
,
5740 trans
->chunk_bytes_reserved
);
5741 trans
->chunk_bytes_reserved
= 0;
5744 /* Can only return 0 or -ENOSPC */
5745 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle
*trans
,
5746 struct btrfs_inode
*inode
)
5748 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
5749 struct btrfs_root
*root
= inode
->root
;
5751 * We always use trans->block_rsv here as we will have reserved space
5752 * for our orphan when starting the transaction, using get_block_rsv()
5753 * here will sometimes make us choose the wrong block rsv as we could be
5754 * doing a reloc inode for a non refcounted root.
5756 struct btrfs_block_rsv
*src_rsv
= trans
->block_rsv
;
5757 struct btrfs_block_rsv
*dst_rsv
= root
->orphan_block_rsv
;
5760 * We need to hold space in order to delete our orphan item once we've
5761 * added it, so this takes the reservation so we can release it later
5762 * when we are truly done with the orphan item.
5764 u64 num_bytes
= btrfs_calc_trans_metadata_size(fs_info
, 1);
5766 trace_btrfs_space_reservation(fs_info
, "orphan", btrfs_ino(inode
),
5768 return btrfs_block_rsv_migrate(src_rsv
, dst_rsv
, num_bytes
, 1);
5771 void btrfs_orphan_release_metadata(struct btrfs_inode
*inode
)
5773 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
5774 struct btrfs_root
*root
= inode
->root
;
5775 u64 num_bytes
= btrfs_calc_trans_metadata_size(fs_info
, 1);
5777 trace_btrfs_space_reservation(fs_info
, "orphan", btrfs_ino(inode
),
5779 btrfs_block_rsv_release(fs_info
, root
->orphan_block_rsv
, num_bytes
);
5783 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5784 * root: the root of the parent directory
5785 * rsv: block reservation
5786 * items: the number of items that we need do reservation
5787 * qgroup_reserved: used to return the reserved size in qgroup
5789 * This function is used to reserve the space for snapshot/subvolume
5790 * creation and deletion. Those operations are different with the
5791 * common file/directory operations, they change two fs/file trees
5792 * and root tree, the number of items that the qgroup reserves is
5793 * different with the free space reservation. So we can not use
5794 * the space reservation mechanism in start_transaction().
5796 int btrfs_subvolume_reserve_metadata(struct btrfs_root
*root
,
5797 struct btrfs_block_rsv
*rsv
,
5799 u64
*qgroup_reserved
,
5800 bool use_global_rsv
)
5804 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5805 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5807 if (test_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
)) {
5808 /* One for parent inode, two for dir entries */
5809 num_bytes
= 3 * fs_info
->nodesize
;
5810 ret
= btrfs_qgroup_reserve_meta(root
, num_bytes
, true);
5817 *qgroup_reserved
= num_bytes
;
5819 num_bytes
= btrfs_calc_trans_metadata_size(fs_info
, items
);
5820 rsv
->space_info
= __find_space_info(fs_info
,
5821 BTRFS_BLOCK_GROUP_METADATA
);
5822 ret
= btrfs_block_rsv_add(root
, rsv
, num_bytes
,
5823 BTRFS_RESERVE_FLUSH_ALL
);
5825 if (ret
== -ENOSPC
&& use_global_rsv
)
5826 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
, num_bytes
, 1);
5828 if (ret
&& *qgroup_reserved
)
5829 btrfs_qgroup_free_meta(root
, *qgroup_reserved
);
5834 void btrfs_subvolume_release_metadata(struct btrfs_fs_info
*fs_info
,
5835 struct btrfs_block_rsv
*rsv
)
5837 btrfs_block_rsv_release(fs_info
, rsv
, (u64
)-1);
5841 * drop_outstanding_extent - drop an outstanding extent
5842 * @inode: the inode we're dropping the extent for
5843 * @num_bytes: the number of bytes we're releasing.
5845 * This is called when we are freeing up an outstanding extent, either called
5846 * after an error or after an extent is written. This will return the number of
5847 * reserved extents that need to be freed. This must be called with
5848 * BTRFS_I(inode)->lock held.
5850 static unsigned drop_outstanding_extent(struct btrfs_inode
*inode
,
5853 unsigned drop_inode_space
= 0;
5854 unsigned dropped_extents
= 0;
5855 unsigned num_extents
;
5857 num_extents
= count_max_extents(num_bytes
);
5858 ASSERT(num_extents
);
5859 ASSERT(inode
->outstanding_extents
>= num_extents
);
5860 inode
->outstanding_extents
-= num_extents
;
5862 if (inode
->outstanding_extents
== 0 &&
5863 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5864 &inode
->runtime_flags
))
5865 drop_inode_space
= 1;
5868 * If we have more or the same amount of outstanding extents than we have
5869 * reserved then we need to leave the reserved extents count alone.
5871 if (inode
->outstanding_extents
>= inode
->reserved_extents
)
5872 return drop_inode_space
;
5874 dropped_extents
= inode
->reserved_extents
- inode
->outstanding_extents
;
5875 inode
->reserved_extents
-= dropped_extents
;
5876 return dropped_extents
+ drop_inode_space
;
5880 * calc_csum_metadata_size - return the amount of metadata space that must be
5881 * reserved/freed for the given bytes.
5882 * @inode: the inode we're manipulating
5883 * @num_bytes: the number of bytes in question
5884 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5886 * This adjusts the number of csum_bytes in the inode and then returns the
5887 * correct amount of metadata that must either be reserved or freed. We
5888 * calculate how many checksums we can fit into one leaf and then divide the
5889 * number of bytes that will need to be checksumed by this value to figure out
5890 * how many checksums will be required. If we are adding bytes then the number
5891 * may go up and we will return the number of additional bytes that must be
5892 * reserved. If it is going down we will return the number of bytes that must
5895 * This must be called with BTRFS_I(inode)->lock held.
5897 static u64
calc_csum_metadata_size(struct btrfs_inode
*inode
, u64 num_bytes
,
5900 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
5901 u64 old_csums
, num_csums
;
5903 if (inode
->flags
& BTRFS_INODE_NODATASUM
&& inode
->csum_bytes
== 0)
5906 old_csums
= btrfs_csum_bytes_to_leaves(fs_info
, inode
->csum_bytes
);
5908 inode
->csum_bytes
+= num_bytes
;
5910 inode
->csum_bytes
-= num_bytes
;
5911 num_csums
= btrfs_csum_bytes_to_leaves(fs_info
, inode
->csum_bytes
);
5913 /* No change, no need to reserve more */
5914 if (old_csums
== num_csums
)
5918 return btrfs_calc_trans_metadata_size(fs_info
,
5919 num_csums
- old_csums
);
5921 return btrfs_calc_trans_metadata_size(fs_info
, old_csums
- num_csums
);
5924 int btrfs_delalloc_reserve_metadata(struct btrfs_inode
*inode
, u64 num_bytes
)
5926 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
5927 struct btrfs_root
*root
= inode
->root
;
5928 struct btrfs_block_rsv
*block_rsv
= &fs_info
->delalloc_block_rsv
;
5931 unsigned nr_extents
;
5932 enum btrfs_reserve_flush_enum flush
= BTRFS_RESERVE_FLUSH_ALL
;
5934 bool delalloc_lock
= true;
5937 bool release_extra
= false;
5939 /* If we are a free space inode we need to not flush since we will be in
5940 * the middle of a transaction commit. We also don't need the delalloc
5941 * mutex since we won't race with anybody. We need this mostly to make
5942 * lockdep shut its filthy mouth.
5944 * If we have a transaction open (can happen if we call truncate_block
5945 * from truncate), then we need FLUSH_LIMIT so we don't deadlock.
5947 if (btrfs_is_free_space_inode(inode
)) {
5948 flush
= BTRFS_RESERVE_NO_FLUSH
;
5949 delalloc_lock
= false;
5950 } else if (current
->journal_info
) {
5951 flush
= BTRFS_RESERVE_FLUSH_LIMIT
;
5954 if (flush
!= BTRFS_RESERVE_NO_FLUSH
&&
5955 btrfs_transaction_in_commit(fs_info
))
5956 schedule_timeout(1);
5959 mutex_lock(&inode
->delalloc_mutex
);
5961 num_bytes
= ALIGN(num_bytes
, fs_info
->sectorsize
);
5963 spin_lock(&inode
->lock
);
5964 nr_extents
= count_max_extents(num_bytes
);
5965 inode
->outstanding_extents
+= nr_extents
;
5968 if (inode
->outstanding_extents
> inode
->reserved_extents
)
5969 nr_extents
+= inode
->outstanding_extents
-
5970 inode
->reserved_extents
;
5972 /* We always want to reserve a slot for updating the inode. */
5973 to_reserve
= btrfs_calc_trans_metadata_size(fs_info
, nr_extents
+ 1);
5974 to_reserve
+= calc_csum_metadata_size(inode
, num_bytes
, 1);
5975 csum_bytes
= inode
->csum_bytes
;
5976 spin_unlock(&inode
->lock
);
5978 if (test_bit(BTRFS_FS_QUOTA_ENABLED
, &fs_info
->flags
)) {
5979 ret
= btrfs_qgroup_reserve_meta(root
,
5980 nr_extents
* fs_info
->nodesize
, true);
5985 ret
= btrfs_block_rsv_add(root
, block_rsv
, to_reserve
, flush
);
5986 if (unlikely(ret
)) {
5987 btrfs_qgroup_free_meta(root
,
5988 nr_extents
* fs_info
->nodesize
);
5992 spin_lock(&inode
->lock
);
5993 if (test_and_set_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5994 &inode
->runtime_flags
)) {
5995 to_reserve
-= btrfs_calc_trans_metadata_size(fs_info
, 1);
5996 release_extra
= true;
5998 inode
->reserved_extents
+= nr_extents
;
5999 spin_unlock(&inode
->lock
);
6002 mutex_unlock(&inode
->delalloc_mutex
);
6005 trace_btrfs_space_reservation(fs_info
, "delalloc",
6006 btrfs_ino(inode
), to_reserve
, 1);
6008 btrfs_block_rsv_release(fs_info
, block_rsv
,
6009 btrfs_calc_trans_metadata_size(fs_info
, 1));
6013 spin_lock(&inode
->lock
);
6014 dropped
= drop_outstanding_extent(inode
, num_bytes
);
6016 * If the inodes csum_bytes is the same as the original
6017 * csum_bytes then we know we haven't raced with any free()ers
6018 * so we can just reduce our inodes csum bytes and carry on.
6020 if (inode
->csum_bytes
== csum_bytes
) {
6021 calc_csum_metadata_size(inode
, num_bytes
, 0);
6023 u64 orig_csum_bytes
= inode
->csum_bytes
;
6027 * This is tricky, but first we need to figure out how much we
6028 * freed from any free-ers that occurred during this
6029 * reservation, so we reset ->csum_bytes to the csum_bytes
6030 * before we dropped our lock, and then call the free for the
6031 * number of bytes that were freed while we were trying our
6034 bytes
= csum_bytes
- inode
->csum_bytes
;
6035 inode
->csum_bytes
= csum_bytes
;
6036 to_free
= calc_csum_metadata_size(inode
, bytes
, 0);
6040 * Now we need to see how much we would have freed had we not
6041 * been making this reservation and our ->csum_bytes were not
6042 * artificially inflated.
6044 inode
->csum_bytes
= csum_bytes
- num_bytes
;
6045 bytes
= csum_bytes
- orig_csum_bytes
;
6046 bytes
= calc_csum_metadata_size(inode
, bytes
, 0);
6049 * Now reset ->csum_bytes to what it should be. If bytes is
6050 * more than to_free then we would have freed more space had we
6051 * not had an artificially high ->csum_bytes, so we need to free
6052 * the remainder. If bytes is the same or less then we don't
6053 * need to do anything, the other free-ers did the correct
6056 inode
->csum_bytes
= orig_csum_bytes
- num_bytes
;
6057 if (bytes
> to_free
)
6058 to_free
= bytes
- to_free
;
6062 spin_unlock(&inode
->lock
);
6064 to_free
+= btrfs_calc_trans_metadata_size(fs_info
, dropped
);
6067 btrfs_block_rsv_release(fs_info
, block_rsv
, to_free
);
6068 trace_btrfs_space_reservation(fs_info
, "delalloc",
6069 btrfs_ino(inode
), to_free
, 0);
6072 mutex_unlock(&inode
->delalloc_mutex
);
6077 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
6078 * @inode: the inode to release the reservation for
6079 * @num_bytes: the number of bytes we're releasing
6081 * This will release the metadata reservation for an inode. This can be called
6082 * once we complete IO for a given set of bytes to release their metadata
6085 void btrfs_delalloc_release_metadata(struct btrfs_inode
*inode
, u64 num_bytes
)
6087 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
6091 num_bytes
= ALIGN(num_bytes
, fs_info
->sectorsize
);
6092 spin_lock(&inode
->lock
);
6093 dropped
= drop_outstanding_extent(inode
, num_bytes
);
6096 to_free
= calc_csum_metadata_size(inode
, num_bytes
, 0);
6097 spin_unlock(&inode
->lock
);
6099 to_free
+= btrfs_calc_trans_metadata_size(fs_info
, dropped
);
6101 if (btrfs_is_testing(fs_info
))
6104 trace_btrfs_space_reservation(fs_info
, "delalloc", btrfs_ino(inode
),
6107 btrfs_block_rsv_release(fs_info
, &fs_info
->delalloc_block_rsv
, to_free
);
6111 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6113 * @inode: inode we're writing to
6114 * @start: start range we are writing to
6115 * @len: how long the range we are writing to
6117 * This will do the following things
6119 * o reserve space in data space info for num bytes
6120 * and reserve precious corresponding qgroup space
6121 * (Done in check_data_free_space)
6123 * o reserve space for metadata space, based on the number of outstanding
6124 * extents and how much csums will be needed
6125 * also reserve metadata space in a per root over-reserve method.
6126 * o add to the inodes->delalloc_bytes
6127 * o add it to the fs_info's delalloc inodes list.
6128 * (Above 3 all done in delalloc_reserve_metadata)
6130 * Return 0 for success
6131 * Return <0 for error(-ENOSPC or -EQUOT)
6133 int btrfs_delalloc_reserve_space(struct inode
*inode
, u64 start
, u64 len
)
6137 ret
= btrfs_check_data_free_space(inode
, start
, len
);
6140 ret
= btrfs_delalloc_reserve_metadata(BTRFS_I(inode
), len
);
6142 btrfs_free_reserved_data_space(inode
, start
, len
);
6147 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6148 * @inode: inode we're releasing space for
6149 * @start: start position of the space already reserved
6150 * @len: the len of the space already reserved
6152 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
6153 * called in the case that we don't need the metadata AND data reservations
6154 * anymore. So if there is an error or we insert an inline extent.
6156 * This function will release the metadata space that was not used and will
6157 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
6158 * list if there are no delalloc bytes left.
6159 * Also it will handle the qgroup reserved space.
6161 void btrfs_delalloc_release_space(struct inode
*inode
, u64 start
, u64 len
)
6163 btrfs_delalloc_release_metadata(BTRFS_I(inode
), len
);
6164 btrfs_free_reserved_data_space(inode
, start
, len
);
6167 static int update_block_group(struct btrfs_trans_handle
*trans
,
6168 struct btrfs_fs_info
*info
, u64 bytenr
,
6169 u64 num_bytes
, int alloc
)
6171 struct btrfs_block_group_cache
*cache
= NULL
;
6172 u64 total
= num_bytes
;
6177 /* block accounting for super block */
6178 spin_lock(&info
->delalloc_root_lock
);
6179 old_val
= btrfs_super_bytes_used(info
->super_copy
);
6181 old_val
+= num_bytes
;
6183 old_val
-= num_bytes
;
6184 btrfs_set_super_bytes_used(info
->super_copy
, old_val
);
6185 spin_unlock(&info
->delalloc_root_lock
);
6188 cache
= btrfs_lookup_block_group(info
, bytenr
);
6191 if (cache
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
6192 BTRFS_BLOCK_GROUP_RAID1
|
6193 BTRFS_BLOCK_GROUP_RAID10
))
6198 * If this block group has free space cache written out, we
6199 * need to make sure to load it if we are removing space. This
6200 * is because we need the unpinning stage to actually add the
6201 * space back to the block group, otherwise we will leak space.
6203 if (!alloc
&& cache
->cached
== BTRFS_CACHE_NO
)
6204 cache_block_group(cache
, 1);
6206 byte_in_group
= bytenr
- cache
->key
.objectid
;
6207 WARN_ON(byte_in_group
> cache
->key
.offset
);
6209 spin_lock(&cache
->space_info
->lock
);
6210 spin_lock(&cache
->lock
);
6212 if (btrfs_test_opt(info
, SPACE_CACHE
) &&
6213 cache
->disk_cache_state
< BTRFS_DC_CLEAR
)
6214 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
6216 old_val
= btrfs_block_group_used(&cache
->item
);
6217 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
6219 old_val
+= num_bytes
;
6220 btrfs_set_block_group_used(&cache
->item
, old_val
);
6221 cache
->reserved
-= num_bytes
;
6222 cache
->space_info
->bytes_reserved
-= num_bytes
;
6223 cache
->space_info
->bytes_used
+= num_bytes
;
6224 cache
->space_info
->disk_used
+= num_bytes
* factor
;
6225 spin_unlock(&cache
->lock
);
6226 spin_unlock(&cache
->space_info
->lock
);
6228 old_val
-= num_bytes
;
6229 btrfs_set_block_group_used(&cache
->item
, old_val
);
6230 cache
->pinned
+= num_bytes
;
6231 cache
->space_info
->bytes_pinned
+= num_bytes
;
6232 cache
->space_info
->bytes_used
-= num_bytes
;
6233 cache
->space_info
->disk_used
-= num_bytes
* factor
;
6234 spin_unlock(&cache
->lock
);
6235 spin_unlock(&cache
->space_info
->lock
);
6237 trace_btrfs_space_reservation(info
, "pinned",
6238 cache
->space_info
->flags
,
6240 set_extent_dirty(info
->pinned_extents
,
6241 bytenr
, bytenr
+ num_bytes
- 1,
6242 GFP_NOFS
| __GFP_NOFAIL
);
6245 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
6246 if (list_empty(&cache
->dirty_list
)) {
6247 list_add_tail(&cache
->dirty_list
,
6248 &trans
->transaction
->dirty_bgs
);
6249 trans
->transaction
->num_dirty_bgs
++;
6250 btrfs_get_block_group(cache
);
6252 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
6255 * No longer have used bytes in this block group, queue it for
6256 * deletion. We do this after adding the block group to the
6257 * dirty list to avoid races between cleaner kthread and space
6260 if (!alloc
&& old_val
== 0) {
6261 spin_lock(&info
->unused_bgs_lock
);
6262 if (list_empty(&cache
->bg_list
)) {
6263 btrfs_get_block_group(cache
);
6264 list_add_tail(&cache
->bg_list
,
6267 spin_unlock(&info
->unused_bgs_lock
);
6270 btrfs_put_block_group(cache
);
6272 bytenr
+= num_bytes
;
6277 static u64
first_logical_byte(struct btrfs_fs_info
*fs_info
, u64 search_start
)
6279 struct btrfs_block_group_cache
*cache
;
6282 spin_lock(&fs_info
->block_group_cache_lock
);
6283 bytenr
= fs_info
->first_logical_byte
;
6284 spin_unlock(&fs_info
->block_group_cache_lock
);
6286 if (bytenr
< (u64
)-1)
6289 cache
= btrfs_lookup_first_block_group(fs_info
, search_start
);
6293 bytenr
= cache
->key
.objectid
;
6294 btrfs_put_block_group(cache
);
6299 static int pin_down_extent(struct btrfs_fs_info
*fs_info
,
6300 struct btrfs_block_group_cache
*cache
,
6301 u64 bytenr
, u64 num_bytes
, int reserved
)
6303 spin_lock(&cache
->space_info
->lock
);
6304 spin_lock(&cache
->lock
);
6305 cache
->pinned
+= num_bytes
;
6306 cache
->space_info
->bytes_pinned
+= num_bytes
;
6308 cache
->reserved
-= num_bytes
;
6309 cache
->space_info
->bytes_reserved
-= num_bytes
;
6311 spin_unlock(&cache
->lock
);
6312 spin_unlock(&cache
->space_info
->lock
);
6314 trace_btrfs_space_reservation(fs_info
, "pinned",
6315 cache
->space_info
->flags
, num_bytes
, 1);
6316 set_extent_dirty(fs_info
->pinned_extents
, bytenr
,
6317 bytenr
+ num_bytes
- 1, GFP_NOFS
| __GFP_NOFAIL
);
6322 * this function must be called within transaction
6324 int btrfs_pin_extent(struct btrfs_fs_info
*fs_info
,
6325 u64 bytenr
, u64 num_bytes
, int reserved
)
6327 struct btrfs_block_group_cache
*cache
;
6329 cache
= btrfs_lookup_block_group(fs_info
, bytenr
);
6330 BUG_ON(!cache
); /* Logic error */
6332 pin_down_extent(fs_info
, cache
, bytenr
, num_bytes
, reserved
);
6334 btrfs_put_block_group(cache
);
6339 * this function must be called within transaction
6341 int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info
*fs_info
,
6342 u64 bytenr
, u64 num_bytes
)
6344 struct btrfs_block_group_cache
*cache
;
6347 cache
= btrfs_lookup_block_group(fs_info
, bytenr
);
6352 * pull in the free space cache (if any) so that our pin
6353 * removes the free space from the cache. We have load_only set
6354 * to one because the slow code to read in the free extents does check
6355 * the pinned extents.
6357 cache_block_group(cache
, 1);
6359 pin_down_extent(fs_info
, cache
, bytenr
, num_bytes
, 0);
6361 /* remove us from the free space cache (if we're there at all) */
6362 ret
= btrfs_remove_free_space(cache
, bytenr
, num_bytes
);
6363 btrfs_put_block_group(cache
);
6367 static int __exclude_logged_extent(struct btrfs_fs_info
*fs_info
,
6368 u64 start
, u64 num_bytes
)
6371 struct btrfs_block_group_cache
*block_group
;
6372 struct btrfs_caching_control
*caching_ctl
;
6374 block_group
= btrfs_lookup_block_group(fs_info
, start
);
6378 cache_block_group(block_group
, 0);
6379 caching_ctl
= get_caching_control(block_group
);
6383 BUG_ON(!block_group_cache_done(block_group
));
6384 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
6386 mutex_lock(&caching_ctl
->mutex
);
6388 if (start
>= caching_ctl
->progress
) {
6389 ret
= add_excluded_extent(fs_info
, start
, num_bytes
);
6390 } else if (start
+ num_bytes
<= caching_ctl
->progress
) {
6391 ret
= btrfs_remove_free_space(block_group
,
6394 num_bytes
= caching_ctl
->progress
- start
;
6395 ret
= btrfs_remove_free_space(block_group
,
6400 num_bytes
= (start
+ num_bytes
) -
6401 caching_ctl
->progress
;
6402 start
= caching_ctl
->progress
;
6403 ret
= add_excluded_extent(fs_info
, start
, num_bytes
);
6406 mutex_unlock(&caching_ctl
->mutex
);
6407 put_caching_control(caching_ctl
);
6409 btrfs_put_block_group(block_group
);
6413 int btrfs_exclude_logged_extents(struct btrfs_fs_info
*fs_info
,
6414 struct extent_buffer
*eb
)
6416 struct btrfs_file_extent_item
*item
;
6417 struct btrfs_key key
;
6421 if (!btrfs_fs_incompat(fs_info
, MIXED_GROUPS
))
6424 for (i
= 0; i
< btrfs_header_nritems(eb
); i
++) {
6425 btrfs_item_key_to_cpu(eb
, &key
, i
);
6426 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
6428 item
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
6429 found_type
= btrfs_file_extent_type(eb
, item
);
6430 if (found_type
== BTRFS_FILE_EXTENT_INLINE
)
6432 if (btrfs_file_extent_disk_bytenr(eb
, item
) == 0)
6434 key
.objectid
= btrfs_file_extent_disk_bytenr(eb
, item
);
6435 key
.offset
= btrfs_file_extent_disk_num_bytes(eb
, item
);
6436 __exclude_logged_extent(fs_info
, key
.objectid
, key
.offset
);
6443 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache
*bg
)
6445 atomic_inc(&bg
->reservations
);
6448 void btrfs_dec_block_group_reservations(struct btrfs_fs_info
*fs_info
,
6451 struct btrfs_block_group_cache
*bg
;
6453 bg
= btrfs_lookup_block_group(fs_info
, start
);
6455 if (atomic_dec_and_test(&bg
->reservations
))
6456 wake_up_atomic_t(&bg
->reservations
);
6457 btrfs_put_block_group(bg
);
6460 static int btrfs_wait_bg_reservations_atomic_t(atomic_t
*a
)
6466 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache
*bg
)
6468 struct btrfs_space_info
*space_info
= bg
->space_info
;
6472 if (!(bg
->flags
& BTRFS_BLOCK_GROUP_DATA
))
6476 * Our block group is read only but before we set it to read only,
6477 * some task might have had allocated an extent from it already, but it
6478 * has not yet created a respective ordered extent (and added it to a
6479 * root's list of ordered extents).
6480 * Therefore wait for any task currently allocating extents, since the
6481 * block group's reservations counter is incremented while a read lock
6482 * on the groups' semaphore is held and decremented after releasing
6483 * the read access on that semaphore and creating the ordered extent.
6485 down_write(&space_info
->groups_sem
);
6486 up_write(&space_info
->groups_sem
);
6488 wait_on_atomic_t(&bg
->reservations
,
6489 btrfs_wait_bg_reservations_atomic_t
,
6490 TASK_UNINTERRUPTIBLE
);
6494 * btrfs_add_reserved_bytes - update the block_group and space info counters
6495 * @cache: The cache we are manipulating
6496 * @ram_bytes: The number of bytes of file content, and will be same to
6497 * @num_bytes except for the compress path.
6498 * @num_bytes: The number of bytes in question
6499 * @delalloc: The blocks are allocated for the delalloc write
6501 * This is called by the allocator when it reserves space. If this is a
6502 * reservation and the block group has become read only we cannot make the
6503 * reservation and return -EAGAIN, otherwise this function always succeeds.
6505 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache
*cache
,
6506 u64 ram_bytes
, u64 num_bytes
, int delalloc
)
6508 struct btrfs_space_info
*space_info
= cache
->space_info
;
6511 spin_lock(&space_info
->lock
);
6512 spin_lock(&cache
->lock
);
6516 cache
->reserved
+= num_bytes
;
6517 space_info
->bytes_reserved
+= num_bytes
;
6519 trace_btrfs_space_reservation(cache
->fs_info
,
6520 "space_info", space_info
->flags
,
6522 space_info
->bytes_may_use
-= ram_bytes
;
6524 cache
->delalloc_bytes
+= num_bytes
;
6526 spin_unlock(&cache
->lock
);
6527 spin_unlock(&space_info
->lock
);
6532 * btrfs_free_reserved_bytes - update the block_group and space info counters
6533 * @cache: The cache we are manipulating
6534 * @num_bytes: The number of bytes in question
6535 * @delalloc: The blocks are allocated for the delalloc write
6537 * This is called by somebody who is freeing space that was never actually used
6538 * on disk. For example if you reserve some space for a new leaf in transaction
6539 * A and before transaction A commits you free that leaf, you call this with
6540 * reserve set to 0 in order to clear the reservation.
6543 static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache
*cache
,
6544 u64 num_bytes
, int delalloc
)
6546 struct btrfs_space_info
*space_info
= cache
->space_info
;
6549 spin_lock(&space_info
->lock
);
6550 spin_lock(&cache
->lock
);
6552 space_info
->bytes_readonly
+= num_bytes
;
6553 cache
->reserved
-= num_bytes
;
6554 space_info
->bytes_reserved
-= num_bytes
;
6557 cache
->delalloc_bytes
-= num_bytes
;
6558 spin_unlock(&cache
->lock
);
6559 spin_unlock(&space_info
->lock
);
6562 void btrfs_prepare_extent_commit(struct btrfs_fs_info
*fs_info
)
6564 struct btrfs_caching_control
*next
;
6565 struct btrfs_caching_control
*caching_ctl
;
6566 struct btrfs_block_group_cache
*cache
;
6568 down_write(&fs_info
->commit_root_sem
);
6570 list_for_each_entry_safe(caching_ctl
, next
,
6571 &fs_info
->caching_block_groups
, list
) {
6572 cache
= caching_ctl
->block_group
;
6573 if (block_group_cache_done(cache
)) {
6574 cache
->last_byte_to_unpin
= (u64
)-1;
6575 list_del_init(&caching_ctl
->list
);
6576 put_caching_control(caching_ctl
);
6578 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
6582 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
6583 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
6585 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
6587 up_write(&fs_info
->commit_root_sem
);
6589 update_global_block_rsv(fs_info
);
6593 * Returns the free cluster for the given space info and sets empty_cluster to
6594 * what it should be based on the mount options.
6596 static struct btrfs_free_cluster
*
6597 fetch_cluster_info(struct btrfs_fs_info
*fs_info
,
6598 struct btrfs_space_info
*space_info
, u64
*empty_cluster
)
6600 struct btrfs_free_cluster
*ret
= NULL
;
6601 bool ssd
= btrfs_test_opt(fs_info
, SSD
);
6604 if (btrfs_mixed_space_info(space_info
))
6608 *empty_cluster
= SZ_2M
;
6609 if (space_info
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
6610 ret
= &fs_info
->meta_alloc_cluster
;
6612 *empty_cluster
= SZ_64K
;
6613 } else if ((space_info
->flags
& BTRFS_BLOCK_GROUP_DATA
) && ssd
) {
6614 ret
= &fs_info
->data_alloc_cluster
;
6620 static int unpin_extent_range(struct btrfs_fs_info
*fs_info
,
6622 const bool return_free_space
)
6624 struct btrfs_block_group_cache
*cache
= NULL
;
6625 struct btrfs_space_info
*space_info
;
6626 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
6627 struct btrfs_free_cluster
*cluster
= NULL
;
6629 u64 total_unpinned
= 0;
6630 u64 empty_cluster
= 0;
6633 while (start
<= end
) {
6636 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
6638 btrfs_put_block_group(cache
);
6640 cache
= btrfs_lookup_block_group(fs_info
, start
);
6641 BUG_ON(!cache
); /* Logic error */
6643 cluster
= fetch_cluster_info(fs_info
,
6646 empty_cluster
<<= 1;
6649 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
6650 len
= min(len
, end
+ 1 - start
);
6652 if (start
< cache
->last_byte_to_unpin
) {
6653 len
= min(len
, cache
->last_byte_to_unpin
- start
);
6654 if (return_free_space
)
6655 btrfs_add_free_space(cache
, start
, len
);
6659 total_unpinned
+= len
;
6660 space_info
= cache
->space_info
;
6663 * If this space cluster has been marked as fragmented and we've
6664 * unpinned enough in this block group to potentially allow a
6665 * cluster to be created inside of it go ahead and clear the
6668 if (cluster
&& cluster
->fragmented
&&
6669 total_unpinned
> empty_cluster
) {
6670 spin_lock(&cluster
->lock
);
6671 cluster
->fragmented
= 0;
6672 spin_unlock(&cluster
->lock
);
6675 spin_lock(&space_info
->lock
);
6676 spin_lock(&cache
->lock
);
6677 cache
->pinned
-= len
;
6678 space_info
->bytes_pinned
-= len
;
6680 trace_btrfs_space_reservation(fs_info
, "pinned",
6681 space_info
->flags
, len
, 0);
6682 space_info
->max_extent_size
= 0;
6683 percpu_counter_add(&space_info
->total_bytes_pinned
, -len
);
6685 space_info
->bytes_readonly
+= len
;
6688 spin_unlock(&cache
->lock
);
6689 if (!readonly
&& return_free_space
&&
6690 global_rsv
->space_info
== space_info
) {
6692 WARN_ON(!return_free_space
);
6693 spin_lock(&global_rsv
->lock
);
6694 if (!global_rsv
->full
) {
6695 to_add
= min(len
, global_rsv
->size
-
6696 global_rsv
->reserved
);
6697 global_rsv
->reserved
+= to_add
;
6698 space_info
->bytes_may_use
+= to_add
;
6699 if (global_rsv
->reserved
>= global_rsv
->size
)
6700 global_rsv
->full
= 1;
6701 trace_btrfs_space_reservation(fs_info
,
6707 spin_unlock(&global_rsv
->lock
);
6708 /* Add to any tickets we may have */
6710 space_info_add_new_bytes(fs_info
, space_info
,
6713 spin_unlock(&space_info
->lock
);
6717 btrfs_put_block_group(cache
);
6721 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
6722 struct btrfs_fs_info
*fs_info
)
6724 struct btrfs_block_group_cache
*block_group
, *tmp
;
6725 struct list_head
*deleted_bgs
;
6726 struct extent_io_tree
*unpin
;
6731 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
6732 unpin
= &fs_info
->freed_extents
[1];
6734 unpin
= &fs_info
->freed_extents
[0];
6736 while (!trans
->aborted
) {
6737 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
6738 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
6739 EXTENT_DIRTY
, NULL
);
6741 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
6745 if (btrfs_test_opt(fs_info
, DISCARD
))
6746 ret
= btrfs_discard_extent(fs_info
, start
,
6747 end
+ 1 - start
, NULL
);
6749 clear_extent_dirty(unpin
, start
, end
);
6750 unpin_extent_range(fs_info
, start
, end
, true);
6751 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
6756 * Transaction is finished. We don't need the lock anymore. We
6757 * do need to clean up the block groups in case of a transaction
6760 deleted_bgs
= &trans
->transaction
->deleted_bgs
;
6761 list_for_each_entry_safe(block_group
, tmp
, deleted_bgs
, bg_list
) {
6765 if (!trans
->aborted
)
6766 ret
= btrfs_discard_extent(fs_info
,
6767 block_group
->key
.objectid
,
6768 block_group
->key
.offset
,
6771 list_del_init(&block_group
->bg_list
);
6772 btrfs_put_block_group_trimming(block_group
);
6773 btrfs_put_block_group(block_group
);
6776 const char *errstr
= btrfs_decode_error(ret
);
6778 "Discard failed while removing blockgroup: errno=%d %s\n",
6786 static void add_pinned_bytes(struct btrfs_fs_info
*fs_info
, u64 num_bytes
,
6787 u64 owner
, u64 root_objectid
)
6789 struct btrfs_space_info
*space_info
;
6792 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
6793 if (root_objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
6794 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
6796 flags
= BTRFS_BLOCK_GROUP_METADATA
;
6798 flags
= BTRFS_BLOCK_GROUP_DATA
;
6801 space_info
= __find_space_info(fs_info
, flags
);
6802 BUG_ON(!space_info
); /* Logic bug */
6803 percpu_counter_add(&space_info
->total_bytes_pinned
, num_bytes
);
6807 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
6808 struct btrfs_fs_info
*info
,
6809 struct btrfs_delayed_ref_node
*node
, u64 parent
,
6810 u64 root_objectid
, u64 owner_objectid
,
6811 u64 owner_offset
, int refs_to_drop
,
6812 struct btrfs_delayed_extent_op
*extent_op
)
6814 struct btrfs_key key
;
6815 struct btrfs_path
*path
;
6816 struct btrfs_root
*extent_root
= info
->extent_root
;
6817 struct extent_buffer
*leaf
;
6818 struct btrfs_extent_item
*ei
;
6819 struct btrfs_extent_inline_ref
*iref
;
6822 int extent_slot
= 0;
6823 int found_extent
= 0;
6827 u64 bytenr
= node
->bytenr
;
6828 u64 num_bytes
= node
->num_bytes
;
6830 bool skinny_metadata
= btrfs_fs_incompat(info
, SKINNY_METADATA
);
6832 path
= btrfs_alloc_path();
6836 path
->reada
= READA_FORWARD
;
6837 path
->leave_spinning
= 1;
6839 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
6840 BUG_ON(!is_data
&& refs_to_drop
!= 1);
6843 skinny_metadata
= 0;
6845 ret
= lookup_extent_backref(trans
, info
, path
, &iref
,
6846 bytenr
, num_bytes
, parent
,
6847 root_objectid
, owner_objectid
,
6850 extent_slot
= path
->slots
[0];
6851 while (extent_slot
>= 0) {
6852 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
6854 if (key
.objectid
!= bytenr
)
6856 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
6857 key
.offset
== num_bytes
) {
6861 if (key
.type
== BTRFS_METADATA_ITEM_KEY
&&
6862 key
.offset
== owner_objectid
) {
6866 if (path
->slots
[0] - extent_slot
> 5)
6870 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6871 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
6872 if (found_extent
&& item_size
< sizeof(*ei
))
6875 if (!found_extent
) {
6877 ret
= remove_extent_backref(trans
, info
, path
, NULL
,
6879 is_data
, &last_ref
);
6881 btrfs_abort_transaction(trans
, ret
);
6884 btrfs_release_path(path
);
6885 path
->leave_spinning
= 1;
6887 key
.objectid
= bytenr
;
6888 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6889 key
.offset
= num_bytes
;
6891 if (!is_data
&& skinny_metadata
) {
6892 key
.type
= BTRFS_METADATA_ITEM_KEY
;
6893 key
.offset
= owner_objectid
;
6896 ret
= btrfs_search_slot(trans
, extent_root
,
6898 if (ret
> 0 && skinny_metadata
&& path
->slots
[0]) {
6900 * Couldn't find our skinny metadata item,
6901 * see if we have ye olde extent item.
6904 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
6906 if (key
.objectid
== bytenr
&&
6907 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
6908 key
.offset
== num_bytes
)
6912 if (ret
> 0 && skinny_metadata
) {
6913 skinny_metadata
= false;
6914 key
.objectid
= bytenr
;
6915 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6916 key
.offset
= num_bytes
;
6917 btrfs_release_path(path
);
6918 ret
= btrfs_search_slot(trans
, extent_root
,
6924 "umm, got %d back from search, was looking for %llu",
6927 btrfs_print_leaf(info
, path
->nodes
[0]);
6930 btrfs_abort_transaction(trans
, ret
);
6933 extent_slot
= path
->slots
[0];
6935 } else if (WARN_ON(ret
== -ENOENT
)) {
6936 btrfs_print_leaf(info
, path
->nodes
[0]);
6938 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6939 bytenr
, parent
, root_objectid
, owner_objectid
,
6941 btrfs_abort_transaction(trans
, ret
);
6944 btrfs_abort_transaction(trans
, ret
);
6948 leaf
= path
->nodes
[0];
6949 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
6950 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6951 if (item_size
< sizeof(*ei
)) {
6952 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
6953 ret
= convert_extent_item_v0(trans
, info
, path
, owner_objectid
,
6956 btrfs_abort_transaction(trans
, ret
);
6960 btrfs_release_path(path
);
6961 path
->leave_spinning
= 1;
6963 key
.objectid
= bytenr
;
6964 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6965 key
.offset
= num_bytes
;
6967 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
6971 "umm, got %d back from search, was looking for %llu",
6973 btrfs_print_leaf(info
, path
->nodes
[0]);
6976 btrfs_abort_transaction(trans
, ret
);
6980 extent_slot
= path
->slots
[0];
6981 leaf
= path
->nodes
[0];
6982 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
6985 BUG_ON(item_size
< sizeof(*ei
));
6986 ei
= btrfs_item_ptr(leaf
, extent_slot
,
6987 struct btrfs_extent_item
);
6988 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
6989 key
.type
== BTRFS_EXTENT_ITEM_KEY
) {
6990 struct btrfs_tree_block_info
*bi
;
6991 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
6992 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
6993 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
6996 refs
= btrfs_extent_refs(leaf
, ei
);
6997 if (refs
< refs_to_drop
) {
6999 "trying to drop %d refs but we only have %Lu for bytenr %Lu",
7000 refs_to_drop
, refs
, bytenr
);
7002 btrfs_abort_transaction(trans
, ret
);
7005 refs
-= refs_to_drop
;
7009 __run_delayed_extent_op(extent_op
, leaf
, ei
);
7011 * In the case of inline back ref, reference count will
7012 * be updated by remove_extent_backref
7015 BUG_ON(!found_extent
);
7017 btrfs_set_extent_refs(leaf
, ei
, refs
);
7018 btrfs_mark_buffer_dirty(leaf
);
7021 ret
= remove_extent_backref(trans
, info
, path
,
7023 is_data
, &last_ref
);
7025 btrfs_abort_transaction(trans
, ret
);
7029 add_pinned_bytes(info
, -num_bytes
, owner_objectid
,
7033 BUG_ON(is_data
&& refs_to_drop
!=
7034 extent_data_ref_count(path
, iref
));
7036 BUG_ON(path
->slots
[0] != extent_slot
);
7038 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
7039 path
->slots
[0] = extent_slot
;
7045 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
7048 btrfs_abort_transaction(trans
, ret
);
7051 btrfs_release_path(path
);
7054 ret
= btrfs_del_csums(trans
, info
, bytenr
, num_bytes
);
7056 btrfs_abort_transaction(trans
, ret
);
7061 ret
= add_to_free_space_tree(trans
, info
, bytenr
, num_bytes
);
7063 btrfs_abort_transaction(trans
, ret
);
7067 ret
= update_block_group(trans
, info
, bytenr
, num_bytes
, 0);
7069 btrfs_abort_transaction(trans
, ret
);
7073 btrfs_release_path(path
);
7076 btrfs_free_path(path
);
7081 * when we free an block, it is possible (and likely) that we free the last
7082 * delayed ref for that extent as well. This searches the delayed ref tree for
7083 * a given extent, and if there are no other delayed refs to be processed, it
7084 * removes it from the tree.
7086 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
7089 struct btrfs_delayed_ref_head
*head
;
7090 struct btrfs_delayed_ref_root
*delayed_refs
;
7093 delayed_refs
= &trans
->transaction
->delayed_refs
;
7094 spin_lock(&delayed_refs
->lock
);
7095 head
= btrfs_find_delayed_ref_head(delayed_refs
, bytenr
);
7097 goto out_delayed_unlock
;
7099 spin_lock(&head
->lock
);
7100 if (!list_empty(&head
->ref_list
))
7103 if (head
->extent_op
) {
7104 if (!head
->must_insert_reserved
)
7106 btrfs_free_delayed_extent_op(head
->extent_op
);
7107 head
->extent_op
= NULL
;
7111 * waiting for the lock here would deadlock. If someone else has it
7112 * locked they are already in the process of dropping it anyway
7114 if (!mutex_trylock(&head
->mutex
))
7118 * at this point we have a head with no other entries. Go
7119 * ahead and process it.
7121 head
->node
.in_tree
= 0;
7122 rb_erase(&head
->href_node
, &delayed_refs
->href_root
);
7124 atomic_dec(&delayed_refs
->num_entries
);
7127 * we don't take a ref on the node because we're removing it from the
7128 * tree, so we just steal the ref the tree was holding.
7130 delayed_refs
->num_heads
--;
7131 if (head
->processing
== 0)
7132 delayed_refs
->num_heads_ready
--;
7133 head
->processing
= 0;
7134 spin_unlock(&head
->lock
);
7135 spin_unlock(&delayed_refs
->lock
);
7137 BUG_ON(head
->extent_op
);
7138 if (head
->must_insert_reserved
)
7141 mutex_unlock(&head
->mutex
);
7142 btrfs_put_delayed_ref(&head
->node
);
7145 spin_unlock(&head
->lock
);
7148 spin_unlock(&delayed_refs
->lock
);
7152 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
7153 struct btrfs_root
*root
,
7154 struct extent_buffer
*buf
,
7155 u64 parent
, int last_ref
)
7157 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7161 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
7162 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
,
7163 buf
->start
, buf
->len
,
7165 root
->root_key
.objectid
,
7166 btrfs_header_level(buf
),
7167 BTRFS_DROP_DELAYED_REF
, NULL
);
7168 BUG_ON(ret
); /* -ENOMEM */
7174 if (btrfs_header_generation(buf
) == trans
->transid
) {
7175 struct btrfs_block_group_cache
*cache
;
7177 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
7178 ret
= check_ref_cleanup(trans
, buf
->start
);
7183 cache
= btrfs_lookup_block_group(fs_info
, buf
->start
);
7185 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
7186 pin_down_extent(fs_info
, cache
, buf
->start
,
7188 btrfs_put_block_group(cache
);
7192 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
7194 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
7195 btrfs_free_reserved_bytes(cache
, buf
->len
, 0);
7196 btrfs_put_block_group(cache
);
7197 trace_btrfs_reserved_extent_free(fs_info
, buf
->start
, buf
->len
);
7202 add_pinned_bytes(fs_info
, buf
->len
, btrfs_header_level(buf
),
7203 root
->root_key
.objectid
);
7206 * Deleting the buffer, clear the corrupt flag since it doesn't matter
7209 clear_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
);
7212 /* Can return -ENOMEM */
7213 int btrfs_free_extent(struct btrfs_trans_handle
*trans
,
7214 struct btrfs_fs_info
*fs_info
,
7215 u64 bytenr
, u64 num_bytes
, u64 parent
, u64 root_objectid
,
7216 u64 owner
, u64 offset
)
7220 if (btrfs_is_testing(fs_info
))
7223 add_pinned_bytes(fs_info
, num_bytes
, owner
, root_objectid
);
7226 * tree log blocks never actually go into the extent allocation
7227 * tree, just update pinning info and exit early.
7229 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
7230 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
7231 /* unlocks the pinned mutex */
7232 btrfs_pin_extent(fs_info
, bytenr
, num_bytes
, 1);
7234 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
7235 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
7237 parent
, root_objectid
, (int)owner
,
7238 BTRFS_DROP_DELAYED_REF
, NULL
);
7240 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
7242 parent
, root_objectid
, owner
,
7244 BTRFS_DROP_DELAYED_REF
);
7250 * when we wait for progress in the block group caching, its because
7251 * our allocation attempt failed at least once. So, we must sleep
7252 * and let some progress happen before we try again.
7254 * This function will sleep at least once waiting for new free space to
7255 * show up, and then it will check the block group free space numbers
7256 * for our min num_bytes. Another option is to have it go ahead
7257 * and look in the rbtree for a free extent of a given size, but this
7260 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7261 * any of the information in this block group.
7263 static noinline
void
7264 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
7267 struct btrfs_caching_control
*caching_ctl
;
7269 caching_ctl
= get_caching_control(cache
);
7273 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
7274 (cache
->free_space_ctl
->free_space
>= num_bytes
));
7276 put_caching_control(caching_ctl
);
7280 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
7282 struct btrfs_caching_control
*caching_ctl
;
7285 caching_ctl
= get_caching_control(cache
);
7287 return (cache
->cached
== BTRFS_CACHE_ERROR
) ? -EIO
: 0;
7289 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
7290 if (cache
->cached
== BTRFS_CACHE_ERROR
)
7292 put_caching_control(caching_ctl
);
7296 int __get_raid_index(u64 flags
)
7298 if (flags
& BTRFS_BLOCK_GROUP_RAID10
)
7299 return BTRFS_RAID_RAID10
;
7300 else if (flags
& BTRFS_BLOCK_GROUP_RAID1
)
7301 return BTRFS_RAID_RAID1
;
7302 else if (flags
& BTRFS_BLOCK_GROUP_DUP
)
7303 return BTRFS_RAID_DUP
;
7304 else if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
7305 return BTRFS_RAID_RAID0
;
7306 else if (flags
& BTRFS_BLOCK_GROUP_RAID5
)
7307 return BTRFS_RAID_RAID5
;
7308 else if (flags
& BTRFS_BLOCK_GROUP_RAID6
)
7309 return BTRFS_RAID_RAID6
;
7311 return BTRFS_RAID_SINGLE
; /* BTRFS_BLOCK_GROUP_SINGLE */
7314 int get_block_group_index(struct btrfs_block_group_cache
*cache
)
7316 return __get_raid_index(cache
->flags
);
7319 static const char *btrfs_raid_type_names
[BTRFS_NR_RAID_TYPES
] = {
7320 [BTRFS_RAID_RAID10
] = "raid10",
7321 [BTRFS_RAID_RAID1
] = "raid1",
7322 [BTRFS_RAID_DUP
] = "dup",
7323 [BTRFS_RAID_RAID0
] = "raid0",
7324 [BTRFS_RAID_SINGLE
] = "single",
7325 [BTRFS_RAID_RAID5
] = "raid5",
7326 [BTRFS_RAID_RAID6
] = "raid6",
7329 static const char *get_raid_name(enum btrfs_raid_types type
)
7331 if (type
>= BTRFS_NR_RAID_TYPES
)
7334 return btrfs_raid_type_names
[type
];
7337 enum btrfs_loop_type
{
7338 LOOP_CACHING_NOWAIT
= 0,
7339 LOOP_CACHING_WAIT
= 1,
7340 LOOP_ALLOC_CHUNK
= 2,
7341 LOOP_NO_EMPTY_SIZE
= 3,
7345 btrfs_lock_block_group(struct btrfs_block_group_cache
*cache
,
7349 down_read(&cache
->data_rwsem
);
7353 btrfs_grab_block_group(struct btrfs_block_group_cache
*cache
,
7356 btrfs_get_block_group(cache
);
7358 down_read(&cache
->data_rwsem
);
7361 static struct btrfs_block_group_cache
*
7362 btrfs_lock_cluster(struct btrfs_block_group_cache
*block_group
,
7363 struct btrfs_free_cluster
*cluster
,
7366 struct btrfs_block_group_cache
*used_bg
= NULL
;
7368 spin_lock(&cluster
->refill_lock
);
7370 used_bg
= cluster
->block_group
;
7374 if (used_bg
== block_group
)
7377 btrfs_get_block_group(used_bg
);
7382 if (down_read_trylock(&used_bg
->data_rwsem
))
7385 spin_unlock(&cluster
->refill_lock
);
7387 /* We should only have one-level nested. */
7388 down_read_nested(&used_bg
->data_rwsem
, SINGLE_DEPTH_NESTING
);
7390 spin_lock(&cluster
->refill_lock
);
7391 if (used_bg
== cluster
->block_group
)
7394 up_read(&used_bg
->data_rwsem
);
7395 btrfs_put_block_group(used_bg
);
7400 btrfs_release_block_group(struct btrfs_block_group_cache
*cache
,
7404 up_read(&cache
->data_rwsem
);
7405 btrfs_put_block_group(cache
);
7409 * walks the btree of allocated extents and find a hole of a given size.
7410 * The key ins is changed to record the hole:
7411 * ins->objectid == start position
7412 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7413 * ins->offset == the size of the hole.
7414 * Any available blocks before search_start are skipped.
7416 * If there is no suitable free space, we will record the max size of
7417 * the free space extent currently.
7419 static noinline
int find_free_extent(struct btrfs_fs_info
*fs_info
,
7420 u64 ram_bytes
, u64 num_bytes
, u64 empty_size
,
7421 u64 hint_byte
, struct btrfs_key
*ins
,
7422 u64 flags
, int delalloc
)
7425 struct btrfs_root
*root
= fs_info
->extent_root
;
7426 struct btrfs_free_cluster
*last_ptr
= NULL
;
7427 struct btrfs_block_group_cache
*block_group
= NULL
;
7428 u64 search_start
= 0;
7429 u64 max_extent_size
= 0;
7430 u64 empty_cluster
= 0;
7431 struct btrfs_space_info
*space_info
;
7433 int index
= __get_raid_index(flags
);
7434 bool failed_cluster_refill
= false;
7435 bool failed_alloc
= false;
7436 bool use_cluster
= true;
7437 bool have_caching_bg
= false;
7438 bool orig_have_caching_bg
= false;
7439 bool full_search
= false;
7441 WARN_ON(num_bytes
< fs_info
->sectorsize
);
7442 ins
->type
= BTRFS_EXTENT_ITEM_KEY
;
7446 trace_find_free_extent(fs_info
, num_bytes
, empty_size
, flags
);
7448 space_info
= __find_space_info(fs_info
, flags
);
7450 btrfs_err(fs_info
, "No space info for %llu", flags
);
7455 * If our free space is heavily fragmented we may not be able to make
7456 * big contiguous allocations, so instead of doing the expensive search
7457 * for free space, simply return ENOSPC with our max_extent_size so we
7458 * can go ahead and search for a more manageable chunk.
7460 * If our max_extent_size is large enough for our allocation simply
7461 * disable clustering since we will likely not be able to find enough
7462 * space to create a cluster and induce latency trying.
7464 if (unlikely(space_info
->max_extent_size
)) {
7465 spin_lock(&space_info
->lock
);
7466 if (space_info
->max_extent_size
&&
7467 num_bytes
> space_info
->max_extent_size
) {
7468 ins
->offset
= space_info
->max_extent_size
;
7469 spin_unlock(&space_info
->lock
);
7471 } else if (space_info
->max_extent_size
) {
7472 use_cluster
= false;
7474 spin_unlock(&space_info
->lock
);
7477 last_ptr
= fetch_cluster_info(fs_info
, space_info
, &empty_cluster
);
7479 spin_lock(&last_ptr
->lock
);
7480 if (last_ptr
->block_group
)
7481 hint_byte
= last_ptr
->window_start
;
7482 if (last_ptr
->fragmented
) {
7484 * We still set window_start so we can keep track of the
7485 * last place we found an allocation to try and save
7488 hint_byte
= last_ptr
->window_start
;
7489 use_cluster
= false;
7491 spin_unlock(&last_ptr
->lock
);
7494 search_start
= max(search_start
, first_logical_byte(fs_info
, 0));
7495 search_start
= max(search_start
, hint_byte
);
7496 if (search_start
== hint_byte
) {
7497 block_group
= btrfs_lookup_block_group(fs_info
, search_start
);
7499 * we don't want to use the block group if it doesn't match our
7500 * allocation bits, or if its not cached.
7502 * However if we are re-searching with an ideal block group
7503 * picked out then we don't care that the block group is cached.
7505 if (block_group
&& block_group_bits(block_group
, flags
) &&
7506 block_group
->cached
!= BTRFS_CACHE_NO
) {
7507 down_read(&space_info
->groups_sem
);
7508 if (list_empty(&block_group
->list
) ||
7511 * someone is removing this block group,
7512 * we can't jump into the have_block_group
7513 * target because our list pointers are not
7516 btrfs_put_block_group(block_group
);
7517 up_read(&space_info
->groups_sem
);
7519 index
= get_block_group_index(block_group
);
7520 btrfs_lock_block_group(block_group
, delalloc
);
7521 goto have_block_group
;
7523 } else if (block_group
) {
7524 btrfs_put_block_group(block_group
);
7528 have_caching_bg
= false;
7529 if (index
== 0 || index
== __get_raid_index(flags
))
7531 down_read(&space_info
->groups_sem
);
7532 list_for_each_entry(block_group
, &space_info
->block_groups
[index
],
7537 btrfs_grab_block_group(block_group
, delalloc
);
7538 search_start
= block_group
->key
.objectid
;
7541 * this can happen if we end up cycling through all the
7542 * raid types, but we want to make sure we only allocate
7543 * for the proper type.
7545 if (!block_group_bits(block_group
, flags
)) {
7546 u64 extra
= BTRFS_BLOCK_GROUP_DUP
|
7547 BTRFS_BLOCK_GROUP_RAID1
|
7548 BTRFS_BLOCK_GROUP_RAID5
|
7549 BTRFS_BLOCK_GROUP_RAID6
|
7550 BTRFS_BLOCK_GROUP_RAID10
;
7553 * if they asked for extra copies and this block group
7554 * doesn't provide them, bail. This does allow us to
7555 * fill raid0 from raid1.
7557 if ((flags
& extra
) && !(block_group
->flags
& extra
))
7562 cached
= block_group_cache_done(block_group
);
7563 if (unlikely(!cached
)) {
7564 have_caching_bg
= true;
7565 ret
= cache_block_group(block_group
, 0);
7570 if (unlikely(block_group
->cached
== BTRFS_CACHE_ERROR
))
7572 if (unlikely(block_group
->ro
))
7576 * Ok we want to try and use the cluster allocator, so
7579 if (last_ptr
&& use_cluster
) {
7580 struct btrfs_block_group_cache
*used_block_group
;
7581 unsigned long aligned_cluster
;
7583 * the refill lock keeps out other
7584 * people trying to start a new cluster
7586 used_block_group
= btrfs_lock_cluster(block_group
,
7589 if (!used_block_group
)
7590 goto refill_cluster
;
7592 if (used_block_group
!= block_group
&&
7593 (used_block_group
->ro
||
7594 !block_group_bits(used_block_group
, flags
)))
7595 goto release_cluster
;
7597 offset
= btrfs_alloc_from_cluster(used_block_group
,
7600 used_block_group
->key
.objectid
,
7603 /* we have a block, we're done */
7604 spin_unlock(&last_ptr
->refill_lock
);
7605 trace_btrfs_reserve_extent_cluster(fs_info
,
7607 search_start
, num_bytes
);
7608 if (used_block_group
!= block_group
) {
7609 btrfs_release_block_group(block_group
,
7611 block_group
= used_block_group
;
7616 WARN_ON(last_ptr
->block_group
!= used_block_group
);
7618 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7619 * set up a new clusters, so lets just skip it
7620 * and let the allocator find whatever block
7621 * it can find. If we reach this point, we
7622 * will have tried the cluster allocator
7623 * plenty of times and not have found
7624 * anything, so we are likely way too
7625 * fragmented for the clustering stuff to find
7628 * However, if the cluster is taken from the
7629 * current block group, release the cluster
7630 * first, so that we stand a better chance of
7631 * succeeding in the unclustered
7633 if (loop
>= LOOP_NO_EMPTY_SIZE
&&
7634 used_block_group
!= block_group
) {
7635 spin_unlock(&last_ptr
->refill_lock
);
7636 btrfs_release_block_group(used_block_group
,
7638 goto unclustered_alloc
;
7642 * this cluster didn't work out, free it and
7645 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
7647 if (used_block_group
!= block_group
)
7648 btrfs_release_block_group(used_block_group
,
7651 if (loop
>= LOOP_NO_EMPTY_SIZE
) {
7652 spin_unlock(&last_ptr
->refill_lock
);
7653 goto unclustered_alloc
;
7656 aligned_cluster
= max_t(unsigned long,
7657 empty_cluster
+ empty_size
,
7658 block_group
->full_stripe_len
);
7660 /* allocate a cluster in this block group */
7661 ret
= btrfs_find_space_cluster(fs_info
, block_group
,
7662 last_ptr
, search_start
,
7667 * now pull our allocation out of this
7670 offset
= btrfs_alloc_from_cluster(block_group
,
7676 /* we found one, proceed */
7677 spin_unlock(&last_ptr
->refill_lock
);
7678 trace_btrfs_reserve_extent_cluster(fs_info
,
7679 block_group
, search_start
,
7683 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
7684 && !failed_cluster_refill
) {
7685 spin_unlock(&last_ptr
->refill_lock
);
7687 failed_cluster_refill
= true;
7688 wait_block_group_cache_progress(block_group
,
7689 num_bytes
+ empty_cluster
+ empty_size
);
7690 goto have_block_group
;
7694 * at this point we either didn't find a cluster
7695 * or we weren't able to allocate a block from our
7696 * cluster. Free the cluster we've been trying
7697 * to use, and go to the next block group
7699 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
7700 spin_unlock(&last_ptr
->refill_lock
);
7706 * We are doing an unclustered alloc, set the fragmented flag so
7707 * we don't bother trying to setup a cluster again until we get
7710 if (unlikely(last_ptr
)) {
7711 spin_lock(&last_ptr
->lock
);
7712 last_ptr
->fragmented
= 1;
7713 spin_unlock(&last_ptr
->lock
);
7716 struct btrfs_free_space_ctl
*ctl
=
7717 block_group
->free_space_ctl
;
7719 spin_lock(&ctl
->tree_lock
);
7720 if (ctl
->free_space
<
7721 num_bytes
+ empty_cluster
+ empty_size
) {
7722 if (ctl
->free_space
> max_extent_size
)
7723 max_extent_size
= ctl
->free_space
;
7724 spin_unlock(&ctl
->tree_lock
);
7727 spin_unlock(&ctl
->tree_lock
);
7730 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
7731 num_bytes
, empty_size
,
7734 * If we didn't find a chunk, and we haven't failed on this
7735 * block group before, and this block group is in the middle of
7736 * caching and we are ok with waiting, then go ahead and wait
7737 * for progress to be made, and set failed_alloc to true.
7739 * If failed_alloc is true then we've already waited on this
7740 * block group once and should move on to the next block group.
7742 if (!offset
&& !failed_alloc
&& !cached
&&
7743 loop
> LOOP_CACHING_NOWAIT
) {
7744 wait_block_group_cache_progress(block_group
,
7745 num_bytes
+ empty_size
);
7746 failed_alloc
= true;
7747 goto have_block_group
;
7748 } else if (!offset
) {
7752 search_start
= ALIGN(offset
, fs_info
->stripesize
);
7754 /* move on to the next group */
7755 if (search_start
+ num_bytes
>
7756 block_group
->key
.objectid
+ block_group
->key
.offset
) {
7757 btrfs_add_free_space(block_group
, offset
, num_bytes
);
7761 if (offset
< search_start
)
7762 btrfs_add_free_space(block_group
, offset
,
7763 search_start
- offset
);
7764 BUG_ON(offset
> search_start
);
7766 ret
= btrfs_add_reserved_bytes(block_group
, ram_bytes
,
7767 num_bytes
, delalloc
);
7768 if (ret
== -EAGAIN
) {
7769 btrfs_add_free_space(block_group
, offset
, num_bytes
);
7772 btrfs_inc_block_group_reservations(block_group
);
7774 /* we are all good, lets return */
7775 ins
->objectid
= search_start
;
7776 ins
->offset
= num_bytes
;
7778 trace_btrfs_reserve_extent(fs_info
, block_group
,
7779 search_start
, num_bytes
);
7780 btrfs_release_block_group(block_group
, delalloc
);
7783 failed_cluster_refill
= false;
7784 failed_alloc
= false;
7785 BUG_ON(index
!= get_block_group_index(block_group
));
7786 btrfs_release_block_group(block_group
, delalloc
);
7788 up_read(&space_info
->groups_sem
);
7790 if ((loop
== LOOP_CACHING_NOWAIT
) && have_caching_bg
7791 && !orig_have_caching_bg
)
7792 orig_have_caching_bg
= true;
7794 if (!ins
->objectid
&& loop
>= LOOP_CACHING_WAIT
&& have_caching_bg
)
7797 if (!ins
->objectid
&& ++index
< BTRFS_NR_RAID_TYPES
)
7801 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7802 * caching kthreads as we move along
7803 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7804 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7805 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7808 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
) {
7810 if (loop
== LOOP_CACHING_NOWAIT
) {
7812 * We want to skip the LOOP_CACHING_WAIT step if we
7813 * don't have any uncached bgs and we've already done a
7814 * full search through.
7816 if (orig_have_caching_bg
|| !full_search
)
7817 loop
= LOOP_CACHING_WAIT
;
7819 loop
= LOOP_ALLOC_CHUNK
;
7824 if (loop
== LOOP_ALLOC_CHUNK
) {
7825 struct btrfs_trans_handle
*trans
;
7828 trans
= current
->journal_info
;
7832 trans
= btrfs_join_transaction(root
);
7834 if (IS_ERR(trans
)) {
7835 ret
= PTR_ERR(trans
);
7839 ret
= do_chunk_alloc(trans
, fs_info
, flags
,
7843 * If we can't allocate a new chunk we've already looped
7844 * through at least once, move on to the NO_EMPTY_SIZE
7848 loop
= LOOP_NO_EMPTY_SIZE
;
7851 * Do not bail out on ENOSPC since we
7852 * can do more things.
7854 if (ret
< 0 && ret
!= -ENOSPC
)
7855 btrfs_abort_transaction(trans
, ret
);
7859 btrfs_end_transaction(trans
);
7864 if (loop
== LOOP_NO_EMPTY_SIZE
) {
7866 * Don't loop again if we already have no empty_size and
7869 if (empty_size
== 0 &&
7870 empty_cluster
== 0) {
7879 } else if (!ins
->objectid
) {
7881 } else if (ins
->objectid
) {
7882 if (!use_cluster
&& last_ptr
) {
7883 spin_lock(&last_ptr
->lock
);
7884 last_ptr
->window_start
= ins
->objectid
;
7885 spin_unlock(&last_ptr
->lock
);
7890 if (ret
== -ENOSPC
) {
7891 spin_lock(&space_info
->lock
);
7892 space_info
->max_extent_size
= max_extent_size
;
7893 spin_unlock(&space_info
->lock
);
7894 ins
->offset
= max_extent_size
;
7899 static void dump_space_info(struct btrfs_fs_info
*fs_info
,
7900 struct btrfs_space_info
*info
, u64 bytes
,
7901 int dump_block_groups
)
7903 struct btrfs_block_group_cache
*cache
;
7906 spin_lock(&info
->lock
);
7907 btrfs_info(fs_info
, "space_info %llu has %llu free, is %sfull",
7909 info
->total_bytes
- btrfs_space_info_used(info
, true),
7910 info
->full
? "" : "not ");
7912 "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu",
7913 info
->total_bytes
, info
->bytes_used
, info
->bytes_pinned
,
7914 info
->bytes_reserved
, info
->bytes_may_use
,
7915 info
->bytes_readonly
);
7916 spin_unlock(&info
->lock
);
7918 if (!dump_block_groups
)
7921 down_read(&info
->groups_sem
);
7923 list_for_each_entry(cache
, &info
->block_groups
[index
], list
) {
7924 spin_lock(&cache
->lock
);
7926 "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s",
7927 cache
->key
.objectid
, cache
->key
.offset
,
7928 btrfs_block_group_used(&cache
->item
), cache
->pinned
,
7929 cache
->reserved
, cache
->ro
? "[readonly]" : "");
7930 btrfs_dump_free_space(cache
, bytes
);
7931 spin_unlock(&cache
->lock
);
7933 if (++index
< BTRFS_NR_RAID_TYPES
)
7935 up_read(&info
->groups_sem
);
7938 int btrfs_reserve_extent(struct btrfs_root
*root
, u64 ram_bytes
,
7939 u64 num_bytes
, u64 min_alloc_size
,
7940 u64 empty_size
, u64 hint_byte
,
7941 struct btrfs_key
*ins
, int is_data
, int delalloc
)
7943 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7944 bool final_tried
= num_bytes
== min_alloc_size
;
7948 flags
= btrfs_get_alloc_profile(root
, is_data
);
7950 WARN_ON(num_bytes
< fs_info
->sectorsize
);
7951 ret
= find_free_extent(fs_info
, ram_bytes
, num_bytes
, empty_size
,
7952 hint_byte
, ins
, flags
, delalloc
);
7953 if (!ret
&& !is_data
) {
7954 btrfs_dec_block_group_reservations(fs_info
, ins
->objectid
);
7955 } else if (ret
== -ENOSPC
) {
7956 if (!final_tried
&& ins
->offset
) {
7957 num_bytes
= min(num_bytes
>> 1, ins
->offset
);
7958 num_bytes
= round_down(num_bytes
,
7959 fs_info
->sectorsize
);
7960 num_bytes
= max(num_bytes
, min_alloc_size
);
7961 ram_bytes
= num_bytes
;
7962 if (num_bytes
== min_alloc_size
)
7965 } else if (btrfs_test_opt(fs_info
, ENOSPC_DEBUG
)) {
7966 struct btrfs_space_info
*sinfo
;
7968 sinfo
= __find_space_info(fs_info
, flags
);
7970 "allocation failed flags %llu, wanted %llu",
7973 dump_space_info(fs_info
, sinfo
, num_bytes
, 1);
7980 static int __btrfs_free_reserved_extent(struct btrfs_fs_info
*fs_info
,
7982 int pin
, int delalloc
)
7984 struct btrfs_block_group_cache
*cache
;
7987 cache
= btrfs_lookup_block_group(fs_info
, start
);
7989 btrfs_err(fs_info
, "Unable to find block group for %llu",
7995 pin_down_extent(fs_info
, cache
, start
, len
, 1);
7997 if (btrfs_test_opt(fs_info
, DISCARD
))
7998 ret
= btrfs_discard_extent(fs_info
, start
, len
, NULL
);
7999 btrfs_add_free_space(cache
, start
, len
);
8000 btrfs_free_reserved_bytes(cache
, len
, delalloc
);
8001 trace_btrfs_reserved_extent_free(fs_info
, start
, len
);
8004 btrfs_put_block_group(cache
);
8008 int btrfs_free_reserved_extent(struct btrfs_fs_info
*fs_info
,
8009 u64 start
, u64 len
, int delalloc
)
8011 return __btrfs_free_reserved_extent(fs_info
, start
, len
, 0, delalloc
);
8014 int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info
*fs_info
,
8017 return __btrfs_free_reserved_extent(fs_info
, start
, len
, 1, 0);
8020 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
8021 struct btrfs_fs_info
*fs_info
,
8022 u64 parent
, u64 root_objectid
,
8023 u64 flags
, u64 owner
, u64 offset
,
8024 struct btrfs_key
*ins
, int ref_mod
)
8027 struct btrfs_extent_item
*extent_item
;
8028 struct btrfs_extent_inline_ref
*iref
;
8029 struct btrfs_path
*path
;
8030 struct extent_buffer
*leaf
;
8035 type
= BTRFS_SHARED_DATA_REF_KEY
;
8037 type
= BTRFS_EXTENT_DATA_REF_KEY
;
8039 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
8041 path
= btrfs_alloc_path();
8045 path
->leave_spinning
= 1;
8046 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
8049 btrfs_free_path(path
);
8053 leaf
= path
->nodes
[0];
8054 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
8055 struct btrfs_extent_item
);
8056 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
8057 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
8058 btrfs_set_extent_flags(leaf
, extent_item
,
8059 flags
| BTRFS_EXTENT_FLAG_DATA
);
8061 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
8062 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
8064 struct btrfs_shared_data_ref
*ref
;
8065 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
8066 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
8067 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
8069 struct btrfs_extent_data_ref
*ref
;
8070 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
8071 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
8072 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
8073 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
8074 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
8077 btrfs_mark_buffer_dirty(path
->nodes
[0]);
8078 btrfs_free_path(path
);
8080 ret
= remove_from_free_space_tree(trans
, fs_info
, ins
->objectid
,
8085 ret
= update_block_group(trans
, fs_info
, ins
->objectid
, ins
->offset
, 1);
8086 if (ret
) { /* -ENOENT, logic error */
8087 btrfs_err(fs_info
, "update block group failed for %llu %llu",
8088 ins
->objectid
, ins
->offset
);
8091 trace_btrfs_reserved_extent_alloc(fs_info
, ins
->objectid
, ins
->offset
);
8095 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
8096 struct btrfs_fs_info
*fs_info
,
8097 u64 parent
, u64 root_objectid
,
8098 u64 flags
, struct btrfs_disk_key
*key
,
8099 int level
, struct btrfs_key
*ins
)
8102 struct btrfs_extent_item
*extent_item
;
8103 struct btrfs_tree_block_info
*block_info
;
8104 struct btrfs_extent_inline_ref
*iref
;
8105 struct btrfs_path
*path
;
8106 struct extent_buffer
*leaf
;
8107 u32 size
= sizeof(*extent_item
) + sizeof(*iref
);
8108 u64 num_bytes
= ins
->offset
;
8109 bool skinny_metadata
= btrfs_fs_incompat(fs_info
, SKINNY_METADATA
);
8111 if (!skinny_metadata
)
8112 size
+= sizeof(*block_info
);
8114 path
= btrfs_alloc_path();
8116 btrfs_free_and_pin_reserved_extent(fs_info
, ins
->objectid
,
8121 path
->leave_spinning
= 1;
8122 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
8125 btrfs_free_path(path
);
8126 btrfs_free_and_pin_reserved_extent(fs_info
, ins
->objectid
,
8131 leaf
= path
->nodes
[0];
8132 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
8133 struct btrfs_extent_item
);
8134 btrfs_set_extent_refs(leaf
, extent_item
, 1);
8135 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
8136 btrfs_set_extent_flags(leaf
, extent_item
,
8137 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
8139 if (skinny_metadata
) {
8140 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
8141 num_bytes
= fs_info
->nodesize
;
8143 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
8144 btrfs_set_tree_block_key(leaf
, block_info
, key
);
8145 btrfs_set_tree_block_level(leaf
, block_info
, level
);
8146 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
8150 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
8151 btrfs_set_extent_inline_ref_type(leaf
, iref
,
8152 BTRFS_SHARED_BLOCK_REF_KEY
);
8153 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
8155 btrfs_set_extent_inline_ref_type(leaf
, iref
,
8156 BTRFS_TREE_BLOCK_REF_KEY
);
8157 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
8160 btrfs_mark_buffer_dirty(leaf
);
8161 btrfs_free_path(path
);
8163 ret
= remove_from_free_space_tree(trans
, fs_info
, ins
->objectid
,
8168 ret
= update_block_group(trans
, fs_info
, ins
->objectid
,
8169 fs_info
->nodesize
, 1);
8170 if (ret
) { /* -ENOENT, logic error */
8171 btrfs_err(fs_info
, "update block group failed for %llu %llu",
8172 ins
->objectid
, ins
->offset
);
8176 trace_btrfs_reserved_extent_alloc(fs_info
, ins
->objectid
,
8181 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
8182 u64 root_objectid
, u64 owner
,
8183 u64 offset
, u64 ram_bytes
,
8184 struct btrfs_key
*ins
)
8186 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
8189 BUG_ON(root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
8191 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, ins
->objectid
,
8193 root_objectid
, owner
, offset
,
8194 ram_bytes
, BTRFS_ADD_DELAYED_EXTENT
);
8199 * this is used by the tree logging recovery code. It records that
8200 * an extent has been allocated and makes sure to clear the free
8201 * space cache bits as well
8203 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
8204 struct btrfs_fs_info
*fs_info
,
8205 u64 root_objectid
, u64 owner
, u64 offset
,
8206 struct btrfs_key
*ins
)
8209 struct btrfs_block_group_cache
*block_group
;
8210 struct btrfs_space_info
*space_info
;
8213 * Mixed block groups will exclude before processing the log so we only
8214 * need to do the exclude dance if this fs isn't mixed.
8216 if (!btrfs_fs_incompat(fs_info
, MIXED_GROUPS
)) {
8217 ret
= __exclude_logged_extent(fs_info
, ins
->objectid
,
8223 block_group
= btrfs_lookup_block_group(fs_info
, ins
->objectid
);
8227 space_info
= block_group
->space_info
;
8228 spin_lock(&space_info
->lock
);
8229 spin_lock(&block_group
->lock
);
8230 space_info
->bytes_reserved
+= ins
->offset
;
8231 block_group
->reserved
+= ins
->offset
;
8232 spin_unlock(&block_group
->lock
);
8233 spin_unlock(&space_info
->lock
);
8235 ret
= alloc_reserved_file_extent(trans
, fs_info
, 0, root_objectid
,
8236 0, owner
, offset
, ins
, 1);
8237 btrfs_put_block_group(block_group
);
8241 static struct extent_buffer
*
8242 btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
8243 u64 bytenr
, int level
)
8245 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8246 struct extent_buffer
*buf
;
8248 buf
= btrfs_find_create_tree_block(fs_info
, bytenr
);
8252 btrfs_set_header_generation(buf
, trans
->transid
);
8253 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, buf
, level
);
8254 btrfs_tree_lock(buf
);
8255 clean_tree_block(fs_info
, buf
);
8256 clear_bit(EXTENT_BUFFER_STALE
, &buf
->bflags
);
8258 btrfs_set_lock_blocking(buf
);
8259 set_extent_buffer_uptodate(buf
);
8261 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
8262 buf
->log_index
= root
->log_transid
% 2;
8264 * we allow two log transactions at a time, use different
8265 * EXENT bit to differentiate dirty pages.
8267 if (buf
->log_index
== 0)
8268 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
8269 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
8271 set_extent_new(&root
->dirty_log_pages
, buf
->start
,
8272 buf
->start
+ buf
->len
- 1);
8274 buf
->log_index
= -1;
8275 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
8276 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
8278 trans
->dirty
= true;
8279 /* this returns a buffer locked for blocking */
8283 static struct btrfs_block_rsv
*
8284 use_block_rsv(struct btrfs_trans_handle
*trans
,
8285 struct btrfs_root
*root
, u32 blocksize
)
8287 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8288 struct btrfs_block_rsv
*block_rsv
;
8289 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
8291 bool global_updated
= false;
8293 block_rsv
= get_block_rsv(trans
, root
);
8295 if (unlikely(block_rsv
->size
== 0))
8298 ret
= block_rsv_use_bytes(block_rsv
, blocksize
);
8302 if (block_rsv
->failfast
)
8303 return ERR_PTR(ret
);
8305 if (block_rsv
->type
== BTRFS_BLOCK_RSV_GLOBAL
&& !global_updated
) {
8306 global_updated
= true;
8307 update_global_block_rsv(fs_info
);
8311 if (btrfs_test_opt(fs_info
, ENOSPC_DEBUG
)) {
8312 static DEFINE_RATELIMIT_STATE(_rs
,
8313 DEFAULT_RATELIMIT_INTERVAL
* 10,
8314 /*DEFAULT_RATELIMIT_BURST*/ 1);
8315 if (__ratelimit(&_rs
))
8317 "BTRFS: block rsv returned %d\n", ret
);
8320 ret
= reserve_metadata_bytes(root
, block_rsv
, blocksize
,
8321 BTRFS_RESERVE_NO_FLUSH
);
8325 * If we couldn't reserve metadata bytes try and use some from
8326 * the global reserve if its space type is the same as the global
8329 if (block_rsv
->type
!= BTRFS_BLOCK_RSV_GLOBAL
&&
8330 block_rsv
->space_info
== global_rsv
->space_info
) {
8331 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
8335 return ERR_PTR(ret
);
8338 static void unuse_block_rsv(struct btrfs_fs_info
*fs_info
,
8339 struct btrfs_block_rsv
*block_rsv
, u32 blocksize
)
8341 block_rsv_add_bytes(block_rsv
, blocksize
, 0);
8342 block_rsv_release_bytes(fs_info
, block_rsv
, NULL
, 0);
8346 * finds a free extent and does all the dirty work required for allocation
8347 * returns the tree buffer or an ERR_PTR on error.
8349 struct extent_buffer
*btrfs_alloc_tree_block(struct btrfs_trans_handle
*trans
,
8350 struct btrfs_root
*root
,
8351 u64 parent
, u64 root_objectid
,
8352 const struct btrfs_disk_key
*key
,
8353 int level
, u64 hint
,
8356 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8357 struct btrfs_key ins
;
8358 struct btrfs_block_rsv
*block_rsv
;
8359 struct extent_buffer
*buf
;
8360 struct btrfs_delayed_extent_op
*extent_op
;
8363 u32 blocksize
= fs_info
->nodesize
;
8364 bool skinny_metadata
= btrfs_fs_incompat(fs_info
, SKINNY_METADATA
);
8366 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8367 if (btrfs_is_testing(fs_info
)) {
8368 buf
= btrfs_init_new_buffer(trans
, root
, root
->alloc_bytenr
,
8371 root
->alloc_bytenr
+= blocksize
;
8376 block_rsv
= use_block_rsv(trans
, root
, blocksize
);
8377 if (IS_ERR(block_rsv
))
8378 return ERR_CAST(block_rsv
);
8380 ret
= btrfs_reserve_extent(root
, blocksize
, blocksize
, blocksize
,
8381 empty_size
, hint
, &ins
, 0, 0);
8385 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
, level
);
8388 goto out_free_reserved
;
8391 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
8393 parent
= ins
.objectid
;
8394 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
8398 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
8399 extent_op
= btrfs_alloc_delayed_extent_op();
8405 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
8407 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
8408 extent_op
->flags_to_set
= flags
;
8409 extent_op
->update_key
= skinny_metadata
? false : true;
8410 extent_op
->update_flags
= true;
8411 extent_op
->is_data
= false;
8412 extent_op
->level
= level
;
8414 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
,
8415 ins
.objectid
, ins
.offset
,
8416 parent
, root_objectid
, level
,
8417 BTRFS_ADD_DELAYED_EXTENT
,
8420 goto out_free_delayed
;
8425 btrfs_free_delayed_extent_op(extent_op
);
8427 free_extent_buffer(buf
);
8429 btrfs_free_reserved_extent(fs_info
, ins
.objectid
, ins
.offset
, 0);
8431 unuse_block_rsv(fs_info
, block_rsv
, blocksize
);
8432 return ERR_PTR(ret
);
8435 struct walk_control
{
8436 u64 refs
[BTRFS_MAX_LEVEL
];
8437 u64 flags
[BTRFS_MAX_LEVEL
];
8438 struct btrfs_key update_progress
;
8449 #define DROP_REFERENCE 1
8450 #define UPDATE_BACKREF 2
8452 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
8453 struct btrfs_root
*root
,
8454 struct walk_control
*wc
,
8455 struct btrfs_path
*path
)
8457 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8463 struct btrfs_key key
;
8464 struct extent_buffer
*eb
;
8469 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
8470 wc
->reada_count
= wc
->reada_count
* 2 / 3;
8471 wc
->reada_count
= max(wc
->reada_count
, 2);
8473 wc
->reada_count
= wc
->reada_count
* 3 / 2;
8474 wc
->reada_count
= min_t(int, wc
->reada_count
,
8475 BTRFS_NODEPTRS_PER_BLOCK(fs_info
));
8478 eb
= path
->nodes
[wc
->level
];
8479 nritems
= btrfs_header_nritems(eb
);
8481 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
8482 if (nread
>= wc
->reada_count
)
8486 bytenr
= btrfs_node_blockptr(eb
, slot
);
8487 generation
= btrfs_node_ptr_generation(eb
, slot
);
8489 if (slot
== path
->slots
[wc
->level
])
8492 if (wc
->stage
== UPDATE_BACKREF
&&
8493 generation
<= root
->root_key
.offset
)
8496 /* We don't lock the tree block, it's OK to be racy here */
8497 ret
= btrfs_lookup_extent_info(trans
, fs_info
, bytenr
,
8498 wc
->level
- 1, 1, &refs
,
8500 /* We don't care about errors in readahead. */
8505 if (wc
->stage
== DROP_REFERENCE
) {
8509 if (wc
->level
== 1 &&
8510 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8512 if (!wc
->update_ref
||
8513 generation
<= root
->root_key
.offset
)
8515 btrfs_node_key_to_cpu(eb
, &key
, slot
);
8516 ret
= btrfs_comp_cpu_keys(&key
,
8517 &wc
->update_progress
);
8521 if (wc
->level
== 1 &&
8522 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8526 readahead_tree_block(fs_info
, bytenr
);
8529 wc
->reada_slot
= slot
;
8533 * helper to process tree block while walking down the tree.
8535 * when wc->stage == UPDATE_BACKREF, this function updates
8536 * back refs for pointers in the block.
8538 * NOTE: return value 1 means we should stop walking down.
8540 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
8541 struct btrfs_root
*root
,
8542 struct btrfs_path
*path
,
8543 struct walk_control
*wc
, int lookup_info
)
8545 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8546 int level
= wc
->level
;
8547 struct extent_buffer
*eb
= path
->nodes
[level
];
8548 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
8551 if (wc
->stage
== UPDATE_BACKREF
&&
8552 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
8556 * when reference count of tree block is 1, it won't increase
8557 * again. once full backref flag is set, we never clear it.
8560 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
8561 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
8562 BUG_ON(!path
->locks
[level
]);
8563 ret
= btrfs_lookup_extent_info(trans
, fs_info
,
8564 eb
->start
, level
, 1,
8567 BUG_ON(ret
== -ENOMEM
);
8570 BUG_ON(wc
->refs
[level
] == 0);
8573 if (wc
->stage
== DROP_REFERENCE
) {
8574 if (wc
->refs
[level
] > 1)
8577 if (path
->locks
[level
] && !wc
->keep_locks
) {
8578 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8579 path
->locks
[level
] = 0;
8584 /* wc->stage == UPDATE_BACKREF */
8585 if (!(wc
->flags
[level
] & flag
)) {
8586 BUG_ON(!path
->locks
[level
]);
8587 ret
= btrfs_inc_ref(trans
, root
, eb
, 1);
8588 BUG_ON(ret
); /* -ENOMEM */
8589 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
8590 BUG_ON(ret
); /* -ENOMEM */
8591 ret
= btrfs_set_disk_extent_flags(trans
, fs_info
, eb
->start
,
8593 btrfs_header_level(eb
), 0);
8594 BUG_ON(ret
); /* -ENOMEM */
8595 wc
->flags
[level
] |= flag
;
8599 * the block is shared by multiple trees, so it's not good to
8600 * keep the tree lock
8602 if (path
->locks
[level
] && level
> 0) {
8603 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8604 path
->locks
[level
] = 0;
8610 * helper to process tree block pointer.
8612 * when wc->stage == DROP_REFERENCE, this function checks
8613 * reference count of the block pointed to. if the block
8614 * is shared and we need update back refs for the subtree
8615 * rooted at the block, this function changes wc->stage to
8616 * UPDATE_BACKREF. if the block is shared and there is no
8617 * need to update back, this function drops the reference
8620 * NOTE: return value 1 means we should stop walking down.
8622 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
8623 struct btrfs_root
*root
,
8624 struct btrfs_path
*path
,
8625 struct walk_control
*wc
, int *lookup_info
)
8627 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8632 struct btrfs_key key
;
8633 struct extent_buffer
*next
;
8634 int level
= wc
->level
;
8637 bool need_account
= false;
8639 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
8640 path
->slots
[level
]);
8642 * if the lower level block was created before the snapshot
8643 * was created, we know there is no need to update back refs
8646 if (wc
->stage
== UPDATE_BACKREF
&&
8647 generation
<= root
->root_key
.offset
) {
8652 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
8653 blocksize
= fs_info
->nodesize
;
8655 next
= find_extent_buffer(fs_info
, bytenr
);
8657 next
= btrfs_find_create_tree_block(fs_info
, bytenr
);
8659 return PTR_ERR(next
);
8661 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, next
,
8665 btrfs_tree_lock(next
);
8666 btrfs_set_lock_blocking(next
);
8668 ret
= btrfs_lookup_extent_info(trans
, fs_info
, bytenr
, level
- 1, 1,
8669 &wc
->refs
[level
- 1],
8670 &wc
->flags
[level
- 1]);
8674 if (unlikely(wc
->refs
[level
- 1] == 0)) {
8675 btrfs_err(fs_info
, "Missing references.");
8681 if (wc
->stage
== DROP_REFERENCE
) {
8682 if (wc
->refs
[level
- 1] > 1) {
8683 need_account
= true;
8685 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8688 if (!wc
->update_ref
||
8689 generation
<= root
->root_key
.offset
)
8692 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
8693 path
->slots
[level
]);
8694 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
8698 wc
->stage
= UPDATE_BACKREF
;
8699 wc
->shared_level
= level
- 1;
8703 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8707 if (!btrfs_buffer_uptodate(next
, generation
, 0)) {
8708 btrfs_tree_unlock(next
);
8709 free_extent_buffer(next
);
8715 if (reada
&& level
== 1)
8716 reada_walk_down(trans
, root
, wc
, path
);
8717 next
= read_tree_block(fs_info
, bytenr
, generation
);
8719 return PTR_ERR(next
);
8720 } else if (!extent_buffer_uptodate(next
)) {
8721 free_extent_buffer(next
);
8724 btrfs_tree_lock(next
);
8725 btrfs_set_lock_blocking(next
);
8729 ASSERT(level
== btrfs_header_level(next
));
8730 if (level
!= btrfs_header_level(next
)) {
8731 btrfs_err(root
->fs_info
, "mismatched level");
8735 path
->nodes
[level
] = next
;
8736 path
->slots
[level
] = 0;
8737 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8743 wc
->refs
[level
- 1] = 0;
8744 wc
->flags
[level
- 1] = 0;
8745 if (wc
->stage
== DROP_REFERENCE
) {
8746 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
8747 parent
= path
->nodes
[level
]->start
;
8749 ASSERT(root
->root_key
.objectid
==
8750 btrfs_header_owner(path
->nodes
[level
]));
8751 if (root
->root_key
.objectid
!=
8752 btrfs_header_owner(path
->nodes
[level
])) {
8753 btrfs_err(root
->fs_info
,
8754 "mismatched block owner");
8762 ret
= btrfs_qgroup_trace_subtree(trans
, root
, next
,
8763 generation
, level
- 1);
8765 btrfs_err_rl(fs_info
,
8766 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
8770 ret
= btrfs_free_extent(trans
, fs_info
, bytenr
, blocksize
,
8771 parent
, root
->root_key
.objectid
,
8781 btrfs_tree_unlock(next
);
8782 free_extent_buffer(next
);
8788 * helper to process tree block while walking up the tree.
8790 * when wc->stage == DROP_REFERENCE, this function drops
8791 * reference count on the block.
8793 * when wc->stage == UPDATE_BACKREF, this function changes
8794 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8795 * to UPDATE_BACKREF previously while processing the block.
8797 * NOTE: return value 1 means we should stop walking up.
8799 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
8800 struct btrfs_root
*root
,
8801 struct btrfs_path
*path
,
8802 struct walk_control
*wc
)
8804 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8806 int level
= wc
->level
;
8807 struct extent_buffer
*eb
= path
->nodes
[level
];
8810 if (wc
->stage
== UPDATE_BACKREF
) {
8811 BUG_ON(wc
->shared_level
< level
);
8812 if (level
< wc
->shared_level
)
8815 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
8819 wc
->stage
= DROP_REFERENCE
;
8820 wc
->shared_level
= -1;
8821 path
->slots
[level
] = 0;
8824 * check reference count again if the block isn't locked.
8825 * we should start walking down the tree again if reference
8828 if (!path
->locks
[level
]) {
8830 btrfs_tree_lock(eb
);
8831 btrfs_set_lock_blocking(eb
);
8832 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8834 ret
= btrfs_lookup_extent_info(trans
, fs_info
,
8835 eb
->start
, level
, 1,
8839 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8840 path
->locks
[level
] = 0;
8843 BUG_ON(wc
->refs
[level
] == 0);
8844 if (wc
->refs
[level
] == 1) {
8845 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8846 path
->locks
[level
] = 0;
8852 /* wc->stage == DROP_REFERENCE */
8853 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
8855 if (wc
->refs
[level
] == 1) {
8857 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8858 ret
= btrfs_dec_ref(trans
, root
, eb
, 1);
8860 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
8861 BUG_ON(ret
); /* -ENOMEM */
8862 ret
= btrfs_qgroup_trace_leaf_items(trans
, fs_info
, eb
);
8864 btrfs_err_rl(fs_info
,
8865 "error %d accounting leaf items. Quota is out of sync, rescan required.",
8869 /* make block locked assertion in clean_tree_block happy */
8870 if (!path
->locks
[level
] &&
8871 btrfs_header_generation(eb
) == trans
->transid
) {
8872 btrfs_tree_lock(eb
);
8873 btrfs_set_lock_blocking(eb
);
8874 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8876 clean_tree_block(fs_info
, eb
);
8879 if (eb
== root
->node
) {
8880 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8883 BUG_ON(root
->root_key
.objectid
!=
8884 btrfs_header_owner(eb
));
8886 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8887 parent
= path
->nodes
[level
+ 1]->start
;
8889 BUG_ON(root
->root_key
.objectid
!=
8890 btrfs_header_owner(path
->nodes
[level
+ 1]));
8893 btrfs_free_tree_block(trans
, root
, eb
, parent
, wc
->refs
[level
] == 1);
8895 wc
->refs
[level
] = 0;
8896 wc
->flags
[level
] = 0;
8900 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
8901 struct btrfs_root
*root
,
8902 struct btrfs_path
*path
,
8903 struct walk_control
*wc
)
8905 int level
= wc
->level
;
8906 int lookup_info
= 1;
8909 while (level
>= 0) {
8910 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
8917 if (path
->slots
[level
] >=
8918 btrfs_header_nritems(path
->nodes
[level
]))
8921 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
8923 path
->slots
[level
]++;
8932 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
8933 struct btrfs_root
*root
,
8934 struct btrfs_path
*path
,
8935 struct walk_control
*wc
, int max_level
)
8937 int level
= wc
->level
;
8940 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
8941 while (level
< max_level
&& path
->nodes
[level
]) {
8943 if (path
->slots
[level
] + 1 <
8944 btrfs_header_nritems(path
->nodes
[level
])) {
8945 path
->slots
[level
]++;
8948 ret
= walk_up_proc(trans
, root
, path
, wc
);
8952 if (path
->locks
[level
]) {
8953 btrfs_tree_unlock_rw(path
->nodes
[level
],
8954 path
->locks
[level
]);
8955 path
->locks
[level
] = 0;
8957 free_extent_buffer(path
->nodes
[level
]);
8958 path
->nodes
[level
] = NULL
;
8966 * drop a subvolume tree.
8968 * this function traverses the tree freeing any blocks that only
8969 * referenced by the tree.
8971 * when a shared tree block is found. this function decreases its
8972 * reference count by one. if update_ref is true, this function
8973 * also make sure backrefs for the shared block and all lower level
8974 * blocks are properly updated.
8976 * If called with for_reloc == 0, may exit early with -EAGAIN
8978 int btrfs_drop_snapshot(struct btrfs_root
*root
,
8979 struct btrfs_block_rsv
*block_rsv
, int update_ref
,
8982 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8983 struct btrfs_path
*path
;
8984 struct btrfs_trans_handle
*trans
;
8985 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
8986 struct btrfs_root_item
*root_item
= &root
->root_item
;
8987 struct walk_control
*wc
;
8988 struct btrfs_key key
;
8992 bool root_dropped
= false;
8994 btrfs_debug(fs_info
, "Drop subvolume %llu", root
->objectid
);
8996 path
= btrfs_alloc_path();
9002 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
9004 btrfs_free_path(path
);
9009 trans
= btrfs_start_transaction(tree_root
, 0);
9010 if (IS_ERR(trans
)) {
9011 err
= PTR_ERR(trans
);
9016 trans
->block_rsv
= block_rsv
;
9018 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
9019 level
= btrfs_header_level(root
->node
);
9020 path
->nodes
[level
] = btrfs_lock_root_node(root
);
9021 btrfs_set_lock_blocking(path
->nodes
[level
]);
9022 path
->slots
[level
] = 0;
9023 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
9024 memset(&wc
->update_progress
, 0,
9025 sizeof(wc
->update_progress
));
9027 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
9028 memcpy(&wc
->update_progress
, &key
,
9029 sizeof(wc
->update_progress
));
9031 level
= root_item
->drop_level
;
9033 path
->lowest_level
= level
;
9034 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
9035 path
->lowest_level
= 0;
9043 * unlock our path, this is safe because only this
9044 * function is allowed to delete this snapshot
9046 btrfs_unlock_up_safe(path
, 0);
9048 level
= btrfs_header_level(root
->node
);
9050 btrfs_tree_lock(path
->nodes
[level
]);
9051 btrfs_set_lock_blocking(path
->nodes
[level
]);
9052 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
9054 ret
= btrfs_lookup_extent_info(trans
, fs_info
,
9055 path
->nodes
[level
]->start
,
9056 level
, 1, &wc
->refs
[level
],
9062 BUG_ON(wc
->refs
[level
] == 0);
9064 if (level
== root_item
->drop_level
)
9067 btrfs_tree_unlock(path
->nodes
[level
]);
9068 path
->locks
[level
] = 0;
9069 WARN_ON(wc
->refs
[level
] != 1);
9075 wc
->shared_level
= -1;
9076 wc
->stage
= DROP_REFERENCE
;
9077 wc
->update_ref
= update_ref
;
9079 wc
->for_reloc
= for_reloc
;
9080 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(fs_info
);
9084 ret
= walk_down_tree(trans
, root
, path
, wc
);
9090 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
9097 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
9101 if (wc
->stage
== DROP_REFERENCE
) {
9103 btrfs_node_key(path
->nodes
[level
],
9104 &root_item
->drop_progress
,
9105 path
->slots
[level
]);
9106 root_item
->drop_level
= level
;
9109 BUG_ON(wc
->level
== 0);
9110 if (btrfs_should_end_transaction(trans
) ||
9111 (!for_reloc
&& btrfs_need_cleaner_sleep(fs_info
))) {
9112 ret
= btrfs_update_root(trans
, tree_root
,
9116 btrfs_abort_transaction(trans
, ret
);
9121 btrfs_end_transaction_throttle(trans
);
9122 if (!for_reloc
&& btrfs_need_cleaner_sleep(fs_info
)) {
9123 btrfs_debug(fs_info
,
9124 "drop snapshot early exit");
9129 trans
= btrfs_start_transaction(tree_root
, 0);
9130 if (IS_ERR(trans
)) {
9131 err
= PTR_ERR(trans
);
9135 trans
->block_rsv
= block_rsv
;
9138 btrfs_release_path(path
);
9142 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
9144 btrfs_abort_transaction(trans
, ret
);
9148 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
9149 ret
= btrfs_find_root(tree_root
, &root
->root_key
, path
,
9152 btrfs_abort_transaction(trans
, ret
);
9155 } else if (ret
> 0) {
9156 /* if we fail to delete the orphan item this time
9157 * around, it'll get picked up the next time.
9159 * The most common failure here is just -ENOENT.
9161 btrfs_del_orphan_item(trans
, tree_root
,
9162 root
->root_key
.objectid
);
9166 if (test_bit(BTRFS_ROOT_IN_RADIX
, &root
->state
)) {
9167 btrfs_add_dropped_root(trans
, root
);
9169 free_extent_buffer(root
->node
);
9170 free_extent_buffer(root
->commit_root
);
9171 btrfs_put_fs_root(root
);
9173 root_dropped
= true;
9175 btrfs_end_transaction_throttle(trans
);
9178 btrfs_free_path(path
);
9181 * So if we need to stop dropping the snapshot for whatever reason we
9182 * need to make sure to add it back to the dead root list so that we
9183 * keep trying to do the work later. This also cleans up roots if we
9184 * don't have it in the radix (like when we recover after a power fail
9185 * or unmount) so we don't leak memory.
9187 if (!for_reloc
&& root_dropped
== false)
9188 btrfs_add_dead_root(root
);
9189 if (err
&& err
!= -EAGAIN
)
9190 btrfs_handle_fs_error(fs_info
, err
, NULL
);
9195 * drop subtree rooted at tree block 'node'.
9197 * NOTE: this function will unlock and release tree block 'node'
9198 * only used by relocation code
9200 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
9201 struct btrfs_root
*root
,
9202 struct extent_buffer
*node
,
9203 struct extent_buffer
*parent
)
9205 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
9206 struct btrfs_path
*path
;
9207 struct walk_control
*wc
;
9213 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
9215 path
= btrfs_alloc_path();
9219 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
9221 btrfs_free_path(path
);
9225 btrfs_assert_tree_locked(parent
);
9226 parent_level
= btrfs_header_level(parent
);
9227 extent_buffer_get(parent
);
9228 path
->nodes
[parent_level
] = parent
;
9229 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
9231 btrfs_assert_tree_locked(node
);
9232 level
= btrfs_header_level(node
);
9233 path
->nodes
[level
] = node
;
9234 path
->slots
[level
] = 0;
9235 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
9237 wc
->refs
[parent_level
] = 1;
9238 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
9240 wc
->shared_level
= -1;
9241 wc
->stage
= DROP_REFERENCE
;
9245 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(fs_info
);
9248 wret
= walk_down_tree(trans
, root
, path
, wc
);
9254 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
9262 btrfs_free_path(path
);
9266 static u64
update_block_group_flags(struct btrfs_fs_info
*fs_info
, u64 flags
)
9272 * if restripe for this chunk_type is on pick target profile and
9273 * return, otherwise do the usual balance
9275 stripped
= get_restripe_target(fs_info
, flags
);
9277 return extended_to_chunk(stripped
);
9279 num_devices
= fs_info
->fs_devices
->rw_devices
;
9281 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
9282 BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
|
9283 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
9285 if (num_devices
== 1) {
9286 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
9287 stripped
= flags
& ~stripped
;
9289 /* turn raid0 into single device chunks */
9290 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
9293 /* turn mirroring into duplication */
9294 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
9295 BTRFS_BLOCK_GROUP_RAID10
))
9296 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
9298 /* they already had raid on here, just return */
9299 if (flags
& stripped
)
9302 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
9303 stripped
= flags
& ~stripped
;
9305 /* switch duplicated blocks with raid1 */
9306 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
9307 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
9309 /* this is drive concat, leave it alone */
9315 static int inc_block_group_ro(struct btrfs_block_group_cache
*cache
, int force
)
9317 struct btrfs_space_info
*sinfo
= cache
->space_info
;
9319 u64 min_allocable_bytes
;
9323 * We need some metadata space and system metadata space for
9324 * allocating chunks in some corner cases until we force to set
9325 * it to be readonly.
9328 (BTRFS_BLOCK_GROUP_SYSTEM
| BTRFS_BLOCK_GROUP_METADATA
)) &&
9330 min_allocable_bytes
= SZ_1M
;
9332 min_allocable_bytes
= 0;
9334 spin_lock(&sinfo
->lock
);
9335 spin_lock(&cache
->lock
);
9343 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
9344 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
9346 if (btrfs_space_info_used(sinfo
, true) + num_bytes
+
9347 min_allocable_bytes
<= sinfo
->total_bytes
) {
9348 sinfo
->bytes_readonly
+= num_bytes
;
9350 list_add_tail(&cache
->ro_list
, &sinfo
->ro_bgs
);
9354 spin_unlock(&cache
->lock
);
9355 spin_unlock(&sinfo
->lock
);
9359 int btrfs_inc_block_group_ro(struct btrfs_fs_info
*fs_info
,
9360 struct btrfs_block_group_cache
*cache
)
9363 struct btrfs_trans_handle
*trans
;
9368 trans
= btrfs_join_transaction(fs_info
->extent_root
);
9370 return PTR_ERR(trans
);
9373 * we're not allowed to set block groups readonly after the dirty
9374 * block groups cache has started writing. If it already started,
9375 * back off and let this transaction commit
9377 mutex_lock(&fs_info
->ro_block_group_mutex
);
9378 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN
, &trans
->transaction
->flags
)) {
9379 u64 transid
= trans
->transid
;
9381 mutex_unlock(&fs_info
->ro_block_group_mutex
);
9382 btrfs_end_transaction(trans
);
9384 ret
= btrfs_wait_for_commit(fs_info
, transid
);
9391 * if we are changing raid levels, try to allocate a corresponding
9392 * block group with the new raid level.
9394 alloc_flags
= update_block_group_flags(fs_info
, cache
->flags
);
9395 if (alloc_flags
!= cache
->flags
) {
9396 ret
= do_chunk_alloc(trans
, fs_info
, alloc_flags
,
9399 * ENOSPC is allowed here, we may have enough space
9400 * already allocated at the new raid level to
9409 ret
= inc_block_group_ro(cache
, 0);
9412 alloc_flags
= get_alloc_profile(fs_info
, cache
->space_info
->flags
);
9413 ret
= do_chunk_alloc(trans
, fs_info
, alloc_flags
,
9417 ret
= inc_block_group_ro(cache
, 0);
9419 if (cache
->flags
& BTRFS_BLOCK_GROUP_SYSTEM
) {
9420 alloc_flags
= update_block_group_flags(fs_info
, cache
->flags
);
9421 mutex_lock(&fs_info
->chunk_mutex
);
9422 check_system_chunk(trans
, fs_info
, alloc_flags
);
9423 mutex_unlock(&fs_info
->chunk_mutex
);
9425 mutex_unlock(&fs_info
->ro_block_group_mutex
);
9427 btrfs_end_transaction(trans
);
9431 int btrfs_force_chunk_alloc(struct btrfs_trans_handle
*trans
,
9432 struct btrfs_fs_info
*fs_info
, u64 type
)
9434 u64 alloc_flags
= get_alloc_profile(fs_info
, type
);
9436 return do_chunk_alloc(trans
, fs_info
, alloc_flags
, CHUNK_ALLOC_FORCE
);
9440 * helper to account the unused space of all the readonly block group in the
9441 * space_info. takes mirrors into account.
9443 u64
btrfs_account_ro_block_groups_free_space(struct btrfs_space_info
*sinfo
)
9445 struct btrfs_block_group_cache
*block_group
;
9449 /* It's df, we don't care if it's racy */
9450 if (list_empty(&sinfo
->ro_bgs
))
9453 spin_lock(&sinfo
->lock
);
9454 list_for_each_entry(block_group
, &sinfo
->ro_bgs
, ro_list
) {
9455 spin_lock(&block_group
->lock
);
9457 if (!block_group
->ro
) {
9458 spin_unlock(&block_group
->lock
);
9462 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_RAID1
|
9463 BTRFS_BLOCK_GROUP_RAID10
|
9464 BTRFS_BLOCK_GROUP_DUP
))
9469 free_bytes
+= (block_group
->key
.offset
-
9470 btrfs_block_group_used(&block_group
->item
)) *
9473 spin_unlock(&block_group
->lock
);
9475 spin_unlock(&sinfo
->lock
);
9480 void btrfs_dec_block_group_ro(struct btrfs_block_group_cache
*cache
)
9482 struct btrfs_space_info
*sinfo
= cache
->space_info
;
9487 spin_lock(&sinfo
->lock
);
9488 spin_lock(&cache
->lock
);
9490 num_bytes
= cache
->key
.offset
- cache
->reserved
-
9491 cache
->pinned
- cache
->bytes_super
-
9492 btrfs_block_group_used(&cache
->item
);
9493 sinfo
->bytes_readonly
-= num_bytes
;
9494 list_del_init(&cache
->ro_list
);
9496 spin_unlock(&cache
->lock
);
9497 spin_unlock(&sinfo
->lock
);
9501 * checks to see if its even possible to relocate this block group.
9503 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9504 * ok to go ahead and try.
9506 int btrfs_can_relocate(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
9508 struct btrfs_root
*root
= fs_info
->extent_root
;
9509 struct btrfs_block_group_cache
*block_group
;
9510 struct btrfs_space_info
*space_info
;
9511 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
9512 struct btrfs_device
*device
;
9513 struct btrfs_trans_handle
*trans
;
9523 debug
= btrfs_test_opt(fs_info
, ENOSPC_DEBUG
);
9525 block_group
= btrfs_lookup_block_group(fs_info
, bytenr
);
9527 /* odd, couldn't find the block group, leave it alone */
9531 "can't find block group for bytenr %llu",
9536 min_free
= btrfs_block_group_used(&block_group
->item
);
9538 /* no bytes used, we're good */
9542 space_info
= block_group
->space_info
;
9543 spin_lock(&space_info
->lock
);
9545 full
= space_info
->full
;
9548 * if this is the last block group we have in this space, we can't
9549 * relocate it unless we're able to allocate a new chunk below.
9551 * Otherwise, we need to make sure we have room in the space to handle
9552 * all of the extents from this block group. If we can, we're good
9554 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
9555 (btrfs_space_info_used(space_info
, false) + min_free
<
9556 space_info
->total_bytes
)) {
9557 spin_unlock(&space_info
->lock
);
9560 spin_unlock(&space_info
->lock
);
9563 * ok we don't have enough space, but maybe we have free space on our
9564 * devices to allocate new chunks for relocation, so loop through our
9565 * alloc devices and guess if we have enough space. if this block
9566 * group is going to be restriped, run checks against the target
9567 * profile instead of the current one.
9579 target
= get_restripe_target(fs_info
, block_group
->flags
);
9581 index
= __get_raid_index(extended_to_chunk(target
));
9584 * this is just a balance, so if we were marked as full
9585 * we know there is no space for a new chunk
9590 "no space to alloc new chunk for block group %llu",
9591 block_group
->key
.objectid
);
9595 index
= get_block_group_index(block_group
);
9598 if (index
== BTRFS_RAID_RAID10
) {
9602 } else if (index
== BTRFS_RAID_RAID1
) {
9604 } else if (index
== BTRFS_RAID_DUP
) {
9607 } else if (index
== BTRFS_RAID_RAID0
) {
9608 dev_min
= fs_devices
->rw_devices
;
9609 min_free
= div64_u64(min_free
, dev_min
);
9612 /* We need to do this so that we can look at pending chunks */
9613 trans
= btrfs_join_transaction(root
);
9614 if (IS_ERR(trans
)) {
9615 ret
= PTR_ERR(trans
);
9619 mutex_lock(&fs_info
->chunk_mutex
);
9620 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
9624 * check to make sure we can actually find a chunk with enough
9625 * space to fit our block group in.
9627 if (device
->total_bytes
> device
->bytes_used
+ min_free
&&
9628 !device
->is_tgtdev_for_dev_replace
) {
9629 ret
= find_free_dev_extent(trans
, device
, min_free
,
9634 if (dev_nr
>= dev_min
)
9640 if (debug
&& ret
== -1)
9642 "no space to allocate a new chunk for block group %llu",
9643 block_group
->key
.objectid
);
9644 mutex_unlock(&fs_info
->chunk_mutex
);
9645 btrfs_end_transaction(trans
);
9647 btrfs_put_block_group(block_group
);
9651 static int find_first_block_group(struct btrfs_fs_info
*fs_info
,
9652 struct btrfs_path
*path
,
9653 struct btrfs_key
*key
)
9655 struct btrfs_root
*root
= fs_info
->extent_root
;
9657 struct btrfs_key found_key
;
9658 struct extent_buffer
*leaf
;
9661 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
9666 slot
= path
->slots
[0];
9667 leaf
= path
->nodes
[0];
9668 if (slot
>= btrfs_header_nritems(leaf
)) {
9669 ret
= btrfs_next_leaf(root
, path
);
9676 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
9678 if (found_key
.objectid
>= key
->objectid
&&
9679 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
9680 struct extent_map_tree
*em_tree
;
9681 struct extent_map
*em
;
9683 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
9684 read_lock(&em_tree
->lock
);
9685 em
= lookup_extent_mapping(em_tree
, found_key
.objectid
,
9687 read_unlock(&em_tree
->lock
);
9690 "logical %llu len %llu found bg but no related chunk",
9691 found_key
.objectid
, found_key
.offset
);
9696 free_extent_map(em
);
9705 void btrfs_put_block_group_cache(struct btrfs_fs_info
*info
)
9707 struct btrfs_block_group_cache
*block_group
;
9711 struct inode
*inode
;
9713 block_group
= btrfs_lookup_first_block_group(info
, last
);
9714 while (block_group
) {
9715 spin_lock(&block_group
->lock
);
9716 if (block_group
->iref
)
9718 spin_unlock(&block_group
->lock
);
9719 block_group
= next_block_group(info
, block_group
);
9728 inode
= block_group
->inode
;
9729 block_group
->iref
= 0;
9730 block_group
->inode
= NULL
;
9731 spin_unlock(&block_group
->lock
);
9732 ASSERT(block_group
->io_ctl
.inode
== NULL
);
9734 last
= block_group
->key
.objectid
+ block_group
->key
.offset
;
9735 btrfs_put_block_group(block_group
);
9740 * Must be called only after stopping all workers, since we could have block
9741 * group caching kthreads running, and therefore they could race with us if we
9742 * freed the block groups before stopping them.
9744 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
9746 struct btrfs_block_group_cache
*block_group
;
9747 struct btrfs_space_info
*space_info
;
9748 struct btrfs_caching_control
*caching_ctl
;
9751 down_write(&info
->commit_root_sem
);
9752 while (!list_empty(&info
->caching_block_groups
)) {
9753 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
9754 struct btrfs_caching_control
, list
);
9755 list_del(&caching_ctl
->list
);
9756 put_caching_control(caching_ctl
);
9758 up_write(&info
->commit_root_sem
);
9760 spin_lock(&info
->unused_bgs_lock
);
9761 while (!list_empty(&info
->unused_bgs
)) {
9762 block_group
= list_first_entry(&info
->unused_bgs
,
9763 struct btrfs_block_group_cache
,
9765 list_del_init(&block_group
->bg_list
);
9766 btrfs_put_block_group(block_group
);
9768 spin_unlock(&info
->unused_bgs_lock
);
9770 spin_lock(&info
->block_group_cache_lock
);
9771 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
9772 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
9774 rb_erase(&block_group
->cache_node
,
9775 &info
->block_group_cache_tree
);
9776 RB_CLEAR_NODE(&block_group
->cache_node
);
9777 spin_unlock(&info
->block_group_cache_lock
);
9779 down_write(&block_group
->space_info
->groups_sem
);
9780 list_del(&block_group
->list
);
9781 up_write(&block_group
->space_info
->groups_sem
);
9784 * We haven't cached this block group, which means we could
9785 * possibly have excluded extents on this block group.
9787 if (block_group
->cached
== BTRFS_CACHE_NO
||
9788 block_group
->cached
== BTRFS_CACHE_ERROR
)
9789 free_excluded_extents(info
, block_group
);
9791 btrfs_remove_free_space_cache(block_group
);
9792 ASSERT(block_group
->cached
!= BTRFS_CACHE_STARTED
);
9793 ASSERT(list_empty(&block_group
->dirty_list
));
9794 ASSERT(list_empty(&block_group
->io_list
));
9795 ASSERT(list_empty(&block_group
->bg_list
));
9796 ASSERT(atomic_read(&block_group
->count
) == 1);
9797 btrfs_put_block_group(block_group
);
9799 spin_lock(&info
->block_group_cache_lock
);
9801 spin_unlock(&info
->block_group_cache_lock
);
9803 /* now that all the block groups are freed, go through and
9804 * free all the space_info structs. This is only called during
9805 * the final stages of unmount, and so we know nobody is
9806 * using them. We call synchronize_rcu() once before we start,
9807 * just to be on the safe side.
9811 release_global_block_rsv(info
);
9813 while (!list_empty(&info
->space_info
)) {
9816 space_info
= list_entry(info
->space_info
.next
,
9817 struct btrfs_space_info
,
9821 * Do not hide this behind enospc_debug, this is actually
9822 * important and indicates a real bug if this happens.
9824 if (WARN_ON(space_info
->bytes_pinned
> 0 ||
9825 space_info
->bytes_reserved
> 0 ||
9826 space_info
->bytes_may_use
> 0))
9827 dump_space_info(info
, space_info
, 0, 0);
9828 list_del(&space_info
->list
);
9829 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++) {
9830 struct kobject
*kobj
;
9831 kobj
= space_info
->block_group_kobjs
[i
];
9832 space_info
->block_group_kobjs
[i
] = NULL
;
9838 kobject_del(&space_info
->kobj
);
9839 kobject_put(&space_info
->kobj
);
9844 static void __link_block_group(struct btrfs_space_info
*space_info
,
9845 struct btrfs_block_group_cache
*cache
)
9847 int index
= get_block_group_index(cache
);
9850 down_write(&space_info
->groups_sem
);
9851 if (list_empty(&space_info
->block_groups
[index
]))
9853 list_add_tail(&cache
->list
, &space_info
->block_groups
[index
]);
9854 up_write(&space_info
->groups_sem
);
9857 struct raid_kobject
*rkobj
;
9860 rkobj
= kzalloc(sizeof(*rkobj
), GFP_NOFS
);
9863 rkobj
->raid_type
= index
;
9864 kobject_init(&rkobj
->kobj
, &btrfs_raid_ktype
);
9865 ret
= kobject_add(&rkobj
->kobj
, &space_info
->kobj
,
9866 "%s", get_raid_name(index
));
9868 kobject_put(&rkobj
->kobj
);
9871 space_info
->block_group_kobjs
[index
] = &rkobj
->kobj
;
9876 btrfs_warn(cache
->fs_info
,
9877 "failed to add kobject for block cache, ignoring");
9880 static struct btrfs_block_group_cache
*
9881 btrfs_create_block_group_cache(struct btrfs_fs_info
*fs_info
,
9882 u64 start
, u64 size
)
9884 struct btrfs_block_group_cache
*cache
;
9886 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
9890 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
9892 if (!cache
->free_space_ctl
) {
9897 cache
->key
.objectid
= start
;
9898 cache
->key
.offset
= size
;
9899 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
9901 cache
->sectorsize
= fs_info
->sectorsize
;
9902 cache
->fs_info
= fs_info
;
9903 cache
->full_stripe_len
= btrfs_full_stripe_len(fs_info
,
9904 &fs_info
->mapping_tree
,
9906 set_free_space_tree_thresholds(cache
);
9908 atomic_set(&cache
->count
, 1);
9909 spin_lock_init(&cache
->lock
);
9910 init_rwsem(&cache
->data_rwsem
);
9911 INIT_LIST_HEAD(&cache
->list
);
9912 INIT_LIST_HEAD(&cache
->cluster_list
);
9913 INIT_LIST_HEAD(&cache
->bg_list
);
9914 INIT_LIST_HEAD(&cache
->ro_list
);
9915 INIT_LIST_HEAD(&cache
->dirty_list
);
9916 INIT_LIST_HEAD(&cache
->io_list
);
9917 btrfs_init_free_space_ctl(cache
);
9918 atomic_set(&cache
->trimming
, 0);
9919 mutex_init(&cache
->free_space_lock
);
9924 int btrfs_read_block_groups(struct btrfs_fs_info
*info
)
9926 struct btrfs_path
*path
;
9928 struct btrfs_block_group_cache
*cache
;
9929 struct btrfs_space_info
*space_info
;
9930 struct btrfs_key key
;
9931 struct btrfs_key found_key
;
9932 struct extent_buffer
*leaf
;
9938 feature
= btrfs_super_incompat_flags(info
->super_copy
);
9939 mixed
= !!(feature
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
);
9943 key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
9944 path
= btrfs_alloc_path();
9947 path
->reada
= READA_FORWARD
;
9949 cache_gen
= btrfs_super_cache_generation(info
->super_copy
);
9950 if (btrfs_test_opt(info
, SPACE_CACHE
) &&
9951 btrfs_super_generation(info
->super_copy
) != cache_gen
)
9953 if (btrfs_test_opt(info
, CLEAR_CACHE
))
9957 ret
= find_first_block_group(info
, path
, &key
);
9963 leaf
= path
->nodes
[0];
9964 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
9966 cache
= btrfs_create_block_group_cache(info
, found_key
.objectid
,
9975 * When we mount with old space cache, we need to
9976 * set BTRFS_DC_CLEAR and set dirty flag.
9978 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9979 * truncate the old free space cache inode and
9981 * b) Setting 'dirty flag' makes sure that we flush
9982 * the new space cache info onto disk.
9984 if (btrfs_test_opt(info
, SPACE_CACHE
))
9985 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
9988 read_extent_buffer(leaf
, &cache
->item
,
9989 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
9990 sizeof(cache
->item
));
9991 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
9993 ((cache
->flags
& BTRFS_BLOCK_GROUP_METADATA
) &&
9994 (cache
->flags
& BTRFS_BLOCK_GROUP_DATA
))) {
9996 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
9997 cache
->key
.objectid
);
10002 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
10003 btrfs_release_path(path
);
10006 * We need to exclude the super stripes now so that the space
10007 * info has super bytes accounted for, otherwise we'll think
10008 * we have more space than we actually do.
10010 ret
= exclude_super_stripes(info
, cache
);
10013 * We may have excluded something, so call this just in
10016 free_excluded_extents(info
, cache
);
10017 btrfs_put_block_group(cache
);
10022 * check for two cases, either we are full, and therefore
10023 * don't need to bother with the caching work since we won't
10024 * find any space, or we are empty, and we can just add all
10025 * the space in and be done with it. This saves us _alot_ of
10026 * time, particularly in the full case.
10028 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
10029 cache
->last_byte_to_unpin
= (u64
)-1;
10030 cache
->cached
= BTRFS_CACHE_FINISHED
;
10031 free_excluded_extents(info
, cache
);
10032 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
10033 cache
->last_byte_to_unpin
= (u64
)-1;
10034 cache
->cached
= BTRFS_CACHE_FINISHED
;
10035 add_new_free_space(cache
, info
,
10036 found_key
.objectid
,
10037 found_key
.objectid
+
10039 free_excluded_extents(info
, cache
);
10042 ret
= btrfs_add_block_group_cache(info
, cache
);
10044 btrfs_remove_free_space_cache(cache
);
10045 btrfs_put_block_group(cache
);
10049 trace_btrfs_add_block_group(info
, cache
, 0);
10050 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
10051 btrfs_block_group_used(&cache
->item
),
10052 cache
->bytes_super
, &space_info
);
10054 btrfs_remove_free_space_cache(cache
);
10055 spin_lock(&info
->block_group_cache_lock
);
10056 rb_erase(&cache
->cache_node
,
10057 &info
->block_group_cache_tree
);
10058 RB_CLEAR_NODE(&cache
->cache_node
);
10059 spin_unlock(&info
->block_group_cache_lock
);
10060 btrfs_put_block_group(cache
);
10064 cache
->space_info
= space_info
;
10066 __link_block_group(space_info
, cache
);
10068 set_avail_alloc_bits(info
, cache
->flags
);
10069 if (btrfs_chunk_readonly(info
, cache
->key
.objectid
)) {
10070 inc_block_group_ro(cache
, 1);
10071 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
10072 spin_lock(&info
->unused_bgs_lock
);
10073 /* Should always be true but just in case. */
10074 if (list_empty(&cache
->bg_list
)) {
10075 btrfs_get_block_group(cache
);
10076 list_add_tail(&cache
->bg_list
,
10077 &info
->unused_bgs
);
10079 spin_unlock(&info
->unused_bgs_lock
);
10083 list_for_each_entry_rcu(space_info
, &info
->space_info
, list
) {
10084 if (!(get_alloc_profile(info
, space_info
->flags
) &
10085 (BTRFS_BLOCK_GROUP_RAID10
|
10086 BTRFS_BLOCK_GROUP_RAID1
|
10087 BTRFS_BLOCK_GROUP_RAID5
|
10088 BTRFS_BLOCK_GROUP_RAID6
|
10089 BTRFS_BLOCK_GROUP_DUP
)))
10092 * avoid allocating from un-mirrored block group if there are
10093 * mirrored block groups.
10095 list_for_each_entry(cache
,
10096 &space_info
->block_groups
[BTRFS_RAID_RAID0
],
10098 inc_block_group_ro(cache
, 1);
10099 list_for_each_entry(cache
,
10100 &space_info
->block_groups
[BTRFS_RAID_SINGLE
],
10102 inc_block_group_ro(cache
, 1);
10105 init_global_block_rsv(info
);
10108 btrfs_free_path(path
);
10112 void btrfs_create_pending_block_groups(struct btrfs_trans_handle
*trans
,
10113 struct btrfs_fs_info
*fs_info
)
10115 struct btrfs_block_group_cache
*block_group
, *tmp
;
10116 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
10117 struct btrfs_block_group_item item
;
10118 struct btrfs_key key
;
10120 bool can_flush_pending_bgs
= trans
->can_flush_pending_bgs
;
10122 trans
->can_flush_pending_bgs
= false;
10123 list_for_each_entry_safe(block_group
, tmp
, &trans
->new_bgs
, bg_list
) {
10127 spin_lock(&block_group
->lock
);
10128 memcpy(&item
, &block_group
->item
, sizeof(item
));
10129 memcpy(&key
, &block_group
->key
, sizeof(key
));
10130 spin_unlock(&block_group
->lock
);
10132 ret
= btrfs_insert_item(trans
, extent_root
, &key
, &item
,
10135 btrfs_abort_transaction(trans
, ret
);
10136 ret
= btrfs_finish_chunk_alloc(trans
, fs_info
, key
.objectid
,
10139 btrfs_abort_transaction(trans
, ret
);
10140 add_block_group_free_space(trans
, fs_info
, block_group
);
10141 /* already aborted the transaction if it failed. */
10143 list_del_init(&block_group
->bg_list
);
10145 trans
->can_flush_pending_bgs
= can_flush_pending_bgs
;
10148 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
10149 struct btrfs_fs_info
*fs_info
, u64 bytes_used
,
10150 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
10153 struct btrfs_block_group_cache
*cache
;
10156 btrfs_set_log_full_commit(fs_info
, trans
);
10158 cache
= btrfs_create_block_group_cache(fs_info
, chunk_offset
, size
);
10162 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
10163 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
10164 btrfs_set_block_group_flags(&cache
->item
, type
);
10166 cache
->flags
= type
;
10167 cache
->last_byte_to_unpin
= (u64
)-1;
10168 cache
->cached
= BTRFS_CACHE_FINISHED
;
10169 cache
->needs_free_space
= 1;
10170 ret
= exclude_super_stripes(fs_info
, cache
);
10173 * We may have excluded something, so call this just in
10176 free_excluded_extents(fs_info
, cache
);
10177 btrfs_put_block_group(cache
);
10181 add_new_free_space(cache
, fs_info
, chunk_offset
, chunk_offset
+ size
);
10183 free_excluded_extents(fs_info
, cache
);
10185 #ifdef CONFIG_BTRFS_DEBUG
10186 if (btrfs_should_fragment_free_space(cache
)) {
10187 u64 new_bytes_used
= size
- bytes_used
;
10189 bytes_used
+= new_bytes_used
>> 1;
10190 fragment_free_space(cache
);
10194 * Call to ensure the corresponding space_info object is created and
10195 * assigned to our block group, but don't update its counters just yet.
10196 * We want our bg to be added to the rbtree with its ->space_info set.
10198 ret
= update_space_info(fs_info
, cache
->flags
, 0, 0, 0,
10199 &cache
->space_info
);
10201 btrfs_remove_free_space_cache(cache
);
10202 btrfs_put_block_group(cache
);
10206 ret
= btrfs_add_block_group_cache(fs_info
, cache
);
10208 btrfs_remove_free_space_cache(cache
);
10209 btrfs_put_block_group(cache
);
10214 * Now that our block group has its ->space_info set and is inserted in
10215 * the rbtree, update the space info's counters.
10217 trace_btrfs_add_block_group(fs_info
, cache
, 1);
10218 ret
= update_space_info(fs_info
, cache
->flags
, size
, bytes_used
,
10219 cache
->bytes_super
, &cache
->space_info
);
10221 btrfs_remove_free_space_cache(cache
);
10222 spin_lock(&fs_info
->block_group_cache_lock
);
10223 rb_erase(&cache
->cache_node
,
10224 &fs_info
->block_group_cache_tree
);
10225 RB_CLEAR_NODE(&cache
->cache_node
);
10226 spin_unlock(&fs_info
->block_group_cache_lock
);
10227 btrfs_put_block_group(cache
);
10230 update_global_block_rsv(fs_info
);
10232 __link_block_group(cache
->space_info
, cache
);
10234 list_add_tail(&cache
->bg_list
, &trans
->new_bgs
);
10236 set_avail_alloc_bits(fs_info
, type
);
10240 static void clear_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
10242 u64 extra_flags
= chunk_to_extended(flags
) &
10243 BTRFS_EXTENDED_PROFILE_MASK
;
10245 write_seqlock(&fs_info
->profiles_lock
);
10246 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
10247 fs_info
->avail_data_alloc_bits
&= ~extra_flags
;
10248 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
10249 fs_info
->avail_metadata_alloc_bits
&= ~extra_flags
;
10250 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
10251 fs_info
->avail_system_alloc_bits
&= ~extra_flags
;
10252 write_sequnlock(&fs_info
->profiles_lock
);
10255 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
10256 struct btrfs_fs_info
*fs_info
, u64 group_start
,
10257 struct extent_map
*em
)
10259 struct btrfs_root
*root
= fs_info
->extent_root
;
10260 struct btrfs_path
*path
;
10261 struct btrfs_block_group_cache
*block_group
;
10262 struct btrfs_free_cluster
*cluster
;
10263 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
10264 struct btrfs_key key
;
10265 struct inode
*inode
;
10266 struct kobject
*kobj
= NULL
;
10270 struct btrfs_caching_control
*caching_ctl
= NULL
;
10273 block_group
= btrfs_lookup_block_group(fs_info
, group_start
);
10274 BUG_ON(!block_group
);
10275 BUG_ON(!block_group
->ro
);
10278 * Free the reserved super bytes from this block group before
10281 free_excluded_extents(fs_info
, block_group
);
10283 memcpy(&key
, &block_group
->key
, sizeof(key
));
10284 index
= get_block_group_index(block_group
);
10285 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
10286 BTRFS_BLOCK_GROUP_RAID1
|
10287 BTRFS_BLOCK_GROUP_RAID10
))
10292 /* make sure this block group isn't part of an allocation cluster */
10293 cluster
= &fs_info
->data_alloc_cluster
;
10294 spin_lock(&cluster
->refill_lock
);
10295 btrfs_return_cluster_to_free_space(block_group
, cluster
);
10296 spin_unlock(&cluster
->refill_lock
);
10299 * make sure this block group isn't part of a metadata
10300 * allocation cluster
10302 cluster
= &fs_info
->meta_alloc_cluster
;
10303 spin_lock(&cluster
->refill_lock
);
10304 btrfs_return_cluster_to_free_space(block_group
, cluster
);
10305 spin_unlock(&cluster
->refill_lock
);
10307 path
= btrfs_alloc_path();
10314 * get the inode first so any iput calls done for the io_list
10315 * aren't the final iput (no unlinks allowed now)
10317 inode
= lookup_free_space_inode(fs_info
, block_group
, path
);
10319 mutex_lock(&trans
->transaction
->cache_write_mutex
);
10321 * make sure our free spache cache IO is done before remove the
10324 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
10325 if (!list_empty(&block_group
->io_list
)) {
10326 list_del_init(&block_group
->io_list
);
10328 WARN_ON(!IS_ERR(inode
) && inode
!= block_group
->io_ctl
.inode
);
10330 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
10331 btrfs_wait_cache_io(trans
, block_group
, path
);
10332 btrfs_put_block_group(block_group
);
10333 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
10336 if (!list_empty(&block_group
->dirty_list
)) {
10337 list_del_init(&block_group
->dirty_list
);
10338 btrfs_put_block_group(block_group
);
10340 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
10341 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
10343 if (!IS_ERR(inode
)) {
10344 ret
= btrfs_orphan_add(trans
, BTRFS_I(inode
));
10346 btrfs_add_delayed_iput(inode
);
10349 clear_nlink(inode
);
10350 /* One for the block groups ref */
10351 spin_lock(&block_group
->lock
);
10352 if (block_group
->iref
) {
10353 block_group
->iref
= 0;
10354 block_group
->inode
= NULL
;
10355 spin_unlock(&block_group
->lock
);
10358 spin_unlock(&block_group
->lock
);
10360 /* One for our lookup ref */
10361 btrfs_add_delayed_iput(inode
);
10364 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
10365 key
.offset
= block_group
->key
.objectid
;
10368 ret
= btrfs_search_slot(trans
, tree_root
, &key
, path
, -1, 1);
10372 btrfs_release_path(path
);
10374 ret
= btrfs_del_item(trans
, tree_root
, path
);
10377 btrfs_release_path(path
);
10380 spin_lock(&fs_info
->block_group_cache_lock
);
10381 rb_erase(&block_group
->cache_node
,
10382 &fs_info
->block_group_cache_tree
);
10383 RB_CLEAR_NODE(&block_group
->cache_node
);
10385 if (fs_info
->first_logical_byte
== block_group
->key
.objectid
)
10386 fs_info
->first_logical_byte
= (u64
)-1;
10387 spin_unlock(&fs_info
->block_group_cache_lock
);
10389 down_write(&block_group
->space_info
->groups_sem
);
10391 * we must use list_del_init so people can check to see if they
10392 * are still on the list after taking the semaphore
10394 list_del_init(&block_group
->list
);
10395 if (list_empty(&block_group
->space_info
->block_groups
[index
])) {
10396 kobj
= block_group
->space_info
->block_group_kobjs
[index
];
10397 block_group
->space_info
->block_group_kobjs
[index
] = NULL
;
10398 clear_avail_alloc_bits(fs_info
, block_group
->flags
);
10400 up_write(&block_group
->space_info
->groups_sem
);
10406 if (block_group
->has_caching_ctl
)
10407 caching_ctl
= get_caching_control(block_group
);
10408 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
10409 wait_block_group_cache_done(block_group
);
10410 if (block_group
->has_caching_ctl
) {
10411 down_write(&fs_info
->commit_root_sem
);
10412 if (!caching_ctl
) {
10413 struct btrfs_caching_control
*ctl
;
10415 list_for_each_entry(ctl
,
10416 &fs_info
->caching_block_groups
, list
)
10417 if (ctl
->block_group
== block_group
) {
10419 atomic_inc(&caching_ctl
->count
);
10424 list_del_init(&caching_ctl
->list
);
10425 up_write(&fs_info
->commit_root_sem
);
10427 /* Once for the caching bgs list and once for us. */
10428 put_caching_control(caching_ctl
);
10429 put_caching_control(caching_ctl
);
10433 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
10434 if (!list_empty(&block_group
->dirty_list
)) {
10437 if (!list_empty(&block_group
->io_list
)) {
10440 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
10441 btrfs_remove_free_space_cache(block_group
);
10443 spin_lock(&block_group
->space_info
->lock
);
10444 list_del_init(&block_group
->ro_list
);
10446 if (btrfs_test_opt(fs_info
, ENOSPC_DEBUG
)) {
10447 WARN_ON(block_group
->space_info
->total_bytes
10448 < block_group
->key
.offset
);
10449 WARN_ON(block_group
->space_info
->bytes_readonly
10450 < block_group
->key
.offset
);
10451 WARN_ON(block_group
->space_info
->disk_total
10452 < block_group
->key
.offset
* factor
);
10454 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
10455 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
10456 block_group
->space_info
->disk_total
-= block_group
->key
.offset
* factor
;
10458 spin_unlock(&block_group
->space_info
->lock
);
10460 memcpy(&key
, &block_group
->key
, sizeof(key
));
10462 mutex_lock(&fs_info
->chunk_mutex
);
10463 if (!list_empty(&em
->list
)) {
10464 /* We're in the transaction->pending_chunks list. */
10465 free_extent_map(em
);
10467 spin_lock(&block_group
->lock
);
10468 block_group
->removed
= 1;
10470 * At this point trimming can't start on this block group, because we
10471 * removed the block group from the tree fs_info->block_group_cache_tree
10472 * so no one can't find it anymore and even if someone already got this
10473 * block group before we removed it from the rbtree, they have already
10474 * incremented block_group->trimming - if they didn't, they won't find
10475 * any free space entries because we already removed them all when we
10476 * called btrfs_remove_free_space_cache().
10478 * And we must not remove the extent map from the fs_info->mapping_tree
10479 * to prevent the same logical address range and physical device space
10480 * ranges from being reused for a new block group. This is because our
10481 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10482 * completely transactionless, so while it is trimming a range the
10483 * currently running transaction might finish and a new one start,
10484 * allowing for new block groups to be created that can reuse the same
10485 * physical device locations unless we take this special care.
10487 * There may also be an implicit trim operation if the file system
10488 * is mounted with -odiscard. The same protections must remain
10489 * in place until the extents have been discarded completely when
10490 * the transaction commit has completed.
10492 remove_em
= (atomic_read(&block_group
->trimming
) == 0);
10494 * Make sure a trimmer task always sees the em in the pinned_chunks list
10495 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10496 * before checking block_group->removed).
10500 * Our em might be in trans->transaction->pending_chunks which
10501 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10502 * and so is the fs_info->pinned_chunks list.
10504 * So at this point we must be holding the chunk_mutex to avoid
10505 * any races with chunk allocation (more specifically at
10506 * volumes.c:contains_pending_extent()), to ensure it always
10507 * sees the em, either in the pending_chunks list or in the
10508 * pinned_chunks list.
10510 list_move_tail(&em
->list
, &fs_info
->pinned_chunks
);
10512 spin_unlock(&block_group
->lock
);
10515 struct extent_map_tree
*em_tree
;
10517 em_tree
= &fs_info
->mapping_tree
.map_tree
;
10518 write_lock(&em_tree
->lock
);
10520 * The em might be in the pending_chunks list, so make sure the
10521 * chunk mutex is locked, since remove_extent_mapping() will
10522 * delete us from that list.
10524 remove_extent_mapping(em_tree
, em
);
10525 write_unlock(&em_tree
->lock
);
10526 /* once for the tree */
10527 free_extent_map(em
);
10530 mutex_unlock(&fs_info
->chunk_mutex
);
10532 ret
= remove_block_group_free_space(trans
, fs_info
, block_group
);
10536 btrfs_put_block_group(block_group
);
10537 btrfs_put_block_group(block_group
);
10539 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
10545 ret
= btrfs_del_item(trans
, root
, path
);
10547 btrfs_free_path(path
);
10551 struct btrfs_trans_handle
*
10552 btrfs_start_trans_remove_block_group(struct btrfs_fs_info
*fs_info
,
10553 const u64 chunk_offset
)
10555 struct extent_map_tree
*em_tree
= &fs_info
->mapping_tree
.map_tree
;
10556 struct extent_map
*em
;
10557 struct map_lookup
*map
;
10558 unsigned int num_items
;
10560 read_lock(&em_tree
->lock
);
10561 em
= lookup_extent_mapping(em_tree
, chunk_offset
, 1);
10562 read_unlock(&em_tree
->lock
);
10563 ASSERT(em
&& em
->start
== chunk_offset
);
10566 * We need to reserve 3 + N units from the metadata space info in order
10567 * to remove a block group (done at btrfs_remove_chunk() and at
10568 * btrfs_remove_block_group()), which are used for:
10570 * 1 unit for adding the free space inode's orphan (located in the tree
10572 * 1 unit for deleting the block group item (located in the extent
10574 * 1 unit for deleting the free space item (located in tree of tree
10576 * N units for deleting N device extent items corresponding to each
10577 * stripe (located in the device tree).
10579 * In order to remove a block group we also need to reserve units in the
10580 * system space info in order to update the chunk tree (update one or
10581 * more device items and remove one chunk item), but this is done at
10582 * btrfs_remove_chunk() through a call to check_system_chunk().
10584 map
= em
->map_lookup
;
10585 num_items
= 3 + map
->num_stripes
;
10586 free_extent_map(em
);
10588 return btrfs_start_transaction_fallback_global_rsv(fs_info
->extent_root
,
10593 * Process the unused_bgs list and remove any that don't have any allocated
10594 * space inside of them.
10596 void btrfs_delete_unused_bgs(struct btrfs_fs_info
*fs_info
)
10598 struct btrfs_block_group_cache
*block_group
;
10599 struct btrfs_space_info
*space_info
;
10600 struct btrfs_trans_handle
*trans
;
10603 if (!test_bit(BTRFS_FS_OPEN
, &fs_info
->flags
))
10606 spin_lock(&fs_info
->unused_bgs_lock
);
10607 while (!list_empty(&fs_info
->unused_bgs
)) {
10611 block_group
= list_first_entry(&fs_info
->unused_bgs
,
10612 struct btrfs_block_group_cache
,
10614 list_del_init(&block_group
->bg_list
);
10616 space_info
= block_group
->space_info
;
10618 if (ret
|| btrfs_mixed_space_info(space_info
)) {
10619 btrfs_put_block_group(block_group
);
10622 spin_unlock(&fs_info
->unused_bgs_lock
);
10624 mutex_lock(&fs_info
->delete_unused_bgs_mutex
);
10626 /* Don't want to race with allocators so take the groups_sem */
10627 down_write(&space_info
->groups_sem
);
10628 spin_lock(&block_group
->lock
);
10629 if (block_group
->reserved
||
10630 btrfs_block_group_used(&block_group
->item
) ||
10632 list_is_singular(&block_group
->list
)) {
10634 * We want to bail if we made new allocations or have
10635 * outstanding allocations in this block group. We do
10636 * the ro check in case balance is currently acting on
10637 * this block group.
10639 spin_unlock(&block_group
->lock
);
10640 up_write(&space_info
->groups_sem
);
10643 spin_unlock(&block_group
->lock
);
10645 /* We don't want to force the issue, only flip if it's ok. */
10646 ret
= inc_block_group_ro(block_group
, 0);
10647 up_write(&space_info
->groups_sem
);
10654 * Want to do this before we do anything else so we can recover
10655 * properly if we fail to join the transaction.
10657 trans
= btrfs_start_trans_remove_block_group(fs_info
,
10658 block_group
->key
.objectid
);
10659 if (IS_ERR(trans
)) {
10660 btrfs_dec_block_group_ro(block_group
);
10661 ret
= PTR_ERR(trans
);
10666 * We could have pending pinned extents for this block group,
10667 * just delete them, we don't care about them anymore.
10669 start
= block_group
->key
.objectid
;
10670 end
= start
+ block_group
->key
.offset
- 1;
10672 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10673 * btrfs_finish_extent_commit(). If we are at transaction N,
10674 * another task might be running finish_extent_commit() for the
10675 * previous transaction N - 1, and have seen a range belonging
10676 * to the block group in freed_extents[] before we were able to
10677 * clear the whole block group range from freed_extents[]. This
10678 * means that task can lookup for the block group after we
10679 * unpinned it from freed_extents[] and removed it, leading to
10680 * a BUG_ON() at btrfs_unpin_extent_range().
10682 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
10683 ret
= clear_extent_bits(&fs_info
->freed_extents
[0], start
, end
,
10686 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10687 btrfs_dec_block_group_ro(block_group
);
10690 ret
= clear_extent_bits(&fs_info
->freed_extents
[1], start
, end
,
10693 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10694 btrfs_dec_block_group_ro(block_group
);
10697 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10699 /* Reset pinned so btrfs_put_block_group doesn't complain */
10700 spin_lock(&space_info
->lock
);
10701 spin_lock(&block_group
->lock
);
10703 space_info
->bytes_pinned
-= block_group
->pinned
;
10704 space_info
->bytes_readonly
+= block_group
->pinned
;
10705 percpu_counter_add(&space_info
->total_bytes_pinned
,
10706 -block_group
->pinned
);
10707 block_group
->pinned
= 0;
10709 spin_unlock(&block_group
->lock
);
10710 spin_unlock(&space_info
->lock
);
10712 /* DISCARD can flip during remount */
10713 trimming
= btrfs_test_opt(fs_info
, DISCARD
);
10715 /* Implicit trim during transaction commit. */
10717 btrfs_get_block_group_trimming(block_group
);
10720 * Btrfs_remove_chunk will abort the transaction if things go
10723 ret
= btrfs_remove_chunk(trans
, fs_info
,
10724 block_group
->key
.objectid
);
10728 btrfs_put_block_group_trimming(block_group
);
10733 * If we're not mounted with -odiscard, we can just forget
10734 * about this block group. Otherwise we'll need to wait
10735 * until transaction commit to do the actual discard.
10738 spin_lock(&fs_info
->unused_bgs_lock
);
10740 * A concurrent scrub might have added us to the list
10741 * fs_info->unused_bgs, so use a list_move operation
10742 * to add the block group to the deleted_bgs list.
10744 list_move(&block_group
->bg_list
,
10745 &trans
->transaction
->deleted_bgs
);
10746 spin_unlock(&fs_info
->unused_bgs_lock
);
10747 btrfs_get_block_group(block_group
);
10750 btrfs_end_transaction(trans
);
10752 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
10753 btrfs_put_block_group(block_group
);
10754 spin_lock(&fs_info
->unused_bgs_lock
);
10756 spin_unlock(&fs_info
->unused_bgs_lock
);
10759 int btrfs_init_space_info(struct btrfs_fs_info
*fs_info
)
10761 struct btrfs_space_info
*space_info
;
10762 struct btrfs_super_block
*disk_super
;
10768 disk_super
= fs_info
->super_copy
;
10769 if (!btrfs_super_root(disk_super
))
10772 features
= btrfs_super_incompat_flags(disk_super
);
10773 if (features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
10776 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
10777 ret
= update_space_info(fs_info
, flags
, 0, 0, 0, &space_info
);
10782 flags
= BTRFS_BLOCK_GROUP_METADATA
| BTRFS_BLOCK_GROUP_DATA
;
10783 ret
= update_space_info(fs_info
, flags
, 0, 0, 0, &space_info
);
10785 flags
= BTRFS_BLOCK_GROUP_METADATA
;
10786 ret
= update_space_info(fs_info
, flags
, 0, 0, 0, &space_info
);
10790 flags
= BTRFS_BLOCK_GROUP_DATA
;
10791 ret
= update_space_info(fs_info
, flags
, 0, 0, 0, &space_info
);
10797 int btrfs_error_unpin_extent_range(struct btrfs_fs_info
*fs_info
,
10798 u64 start
, u64 end
)
10800 return unpin_extent_range(fs_info
, start
, end
, false);
10804 * It used to be that old block groups would be left around forever.
10805 * Iterating over them would be enough to trim unused space. Since we
10806 * now automatically remove them, we also need to iterate over unallocated
10809 * We don't want a transaction for this since the discard may take a
10810 * substantial amount of time. We don't require that a transaction be
10811 * running, but we do need to take a running transaction into account
10812 * to ensure that we're not discarding chunks that were released in
10813 * the current transaction.
10815 * Holding the chunks lock will prevent other threads from allocating
10816 * or releasing chunks, but it won't prevent a running transaction
10817 * from committing and releasing the memory that the pending chunks
10818 * list head uses. For that, we need to take a reference to the
10821 static int btrfs_trim_free_extents(struct btrfs_device
*device
,
10822 u64 minlen
, u64
*trimmed
)
10824 u64 start
= 0, len
= 0;
10829 /* Not writeable = nothing to do. */
10830 if (!device
->writeable
)
10833 /* No free space = nothing to do. */
10834 if (device
->total_bytes
<= device
->bytes_used
)
10840 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
10841 struct btrfs_transaction
*trans
;
10844 ret
= mutex_lock_interruptible(&fs_info
->chunk_mutex
);
10848 down_read(&fs_info
->commit_root_sem
);
10850 spin_lock(&fs_info
->trans_lock
);
10851 trans
= fs_info
->running_transaction
;
10853 atomic_inc(&trans
->use_count
);
10854 spin_unlock(&fs_info
->trans_lock
);
10856 ret
= find_free_dev_extent_start(trans
, device
, minlen
, start
,
10859 btrfs_put_transaction(trans
);
10862 up_read(&fs_info
->commit_root_sem
);
10863 mutex_unlock(&fs_info
->chunk_mutex
);
10864 if (ret
== -ENOSPC
)
10869 ret
= btrfs_issue_discard(device
->bdev
, start
, len
, &bytes
);
10870 up_read(&fs_info
->commit_root_sem
);
10871 mutex_unlock(&fs_info
->chunk_mutex
);
10879 if (fatal_signal_pending(current
)) {
10880 ret
= -ERESTARTSYS
;
10890 int btrfs_trim_fs(struct btrfs_fs_info
*fs_info
, struct fstrim_range
*range
)
10892 struct btrfs_block_group_cache
*cache
= NULL
;
10893 struct btrfs_device
*device
;
10894 struct list_head
*devices
;
10899 u64 total_bytes
= btrfs_super_total_bytes(fs_info
->super_copy
);
10903 * try to trim all FS space, our block group may start from non-zero.
10905 if (range
->len
== total_bytes
)
10906 cache
= btrfs_lookup_first_block_group(fs_info
, range
->start
);
10908 cache
= btrfs_lookup_block_group(fs_info
, range
->start
);
10911 if (cache
->key
.objectid
>= (range
->start
+ range
->len
)) {
10912 btrfs_put_block_group(cache
);
10916 start
= max(range
->start
, cache
->key
.objectid
);
10917 end
= min(range
->start
+ range
->len
,
10918 cache
->key
.objectid
+ cache
->key
.offset
);
10920 if (end
- start
>= range
->minlen
) {
10921 if (!block_group_cache_done(cache
)) {
10922 ret
= cache_block_group(cache
, 0);
10924 btrfs_put_block_group(cache
);
10927 ret
= wait_block_group_cache_done(cache
);
10929 btrfs_put_block_group(cache
);
10933 ret
= btrfs_trim_block_group(cache
,
10939 trimmed
+= group_trimmed
;
10941 btrfs_put_block_group(cache
);
10946 cache
= next_block_group(fs_info
, cache
);
10949 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
10950 devices
= &fs_info
->fs_devices
->alloc_list
;
10951 list_for_each_entry(device
, devices
, dev_alloc_list
) {
10952 ret
= btrfs_trim_free_extents(device
, range
->minlen
,
10957 trimmed
+= group_trimmed
;
10959 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
10961 range
->len
= trimmed
;
10966 * btrfs_{start,end}_write_no_snapshoting() are similar to
10967 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10968 * data into the page cache through nocow before the subvolume is snapshoted,
10969 * but flush the data into disk after the snapshot creation, or to prevent
10970 * operations while snapshoting is ongoing and that cause the snapshot to be
10971 * inconsistent (writes followed by expanding truncates for example).
10973 void btrfs_end_write_no_snapshoting(struct btrfs_root
*root
)
10975 percpu_counter_dec(&root
->subv_writers
->counter
);
10977 * Make sure counter is updated before we wake up waiters.
10980 if (waitqueue_active(&root
->subv_writers
->wait
))
10981 wake_up(&root
->subv_writers
->wait
);
10984 int btrfs_start_write_no_snapshoting(struct btrfs_root
*root
)
10986 if (atomic_read(&root
->will_be_snapshoted
))
10989 percpu_counter_inc(&root
->subv_writers
->counter
);
10991 * Make sure counter is updated before we check for snapshot creation.
10994 if (atomic_read(&root
->will_be_snapshoted
)) {
10995 btrfs_end_write_no_snapshoting(root
);
11001 static int wait_snapshoting_atomic_t(atomic_t
*a
)
11007 void btrfs_wait_for_snapshot_creation(struct btrfs_root
*root
)
11012 ret
= btrfs_start_write_no_snapshoting(root
);
11015 wait_on_atomic_t(&root
->will_be_snapshoted
,
11016 wait_snapshoting_atomic_t
,
11017 TASK_UNINTERRUPTIBLE
);