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/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include <linux/slab.h>
26 #include <linux/ratelimit.h>
27 #include <linux/percpu_counter.h>
31 #include "print-tree.h"
35 #include "free-space-cache.h"
36 #include "free-space-tree.h"
41 #undef SCRAMBLE_DELAYED_REFS
44 * control flags for do_chunk_alloc's force field
45 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
46 * if we really need one.
48 * CHUNK_ALLOC_LIMITED means to only try and allocate one
49 * if we have very few chunks already allocated. This is
50 * used as part of the clustering code to help make sure
51 * we have a good pool of storage to cluster in, without
52 * filling the FS with empty chunks
54 * CHUNK_ALLOC_FORCE means it must try to allocate one
58 CHUNK_ALLOC_NO_FORCE
= 0,
59 CHUNK_ALLOC_LIMITED
= 1,
60 CHUNK_ALLOC_FORCE
= 2,
64 * Control how reservations are dealt with.
66 * RESERVE_FREE - freeing a reservation.
67 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
69 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
70 * bytes_may_use as the ENOSPC accounting is done elsewhere
75 RESERVE_ALLOC_NO_ACCOUNT
= 2,
78 static int update_block_group(struct btrfs_trans_handle
*trans
,
79 struct btrfs_root
*root
, u64 bytenr
,
80 u64 num_bytes
, int alloc
);
81 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
82 struct btrfs_root
*root
,
83 struct btrfs_delayed_ref_node
*node
, u64 parent
,
84 u64 root_objectid
, u64 owner_objectid
,
85 u64 owner_offset
, int refs_to_drop
,
86 struct btrfs_delayed_extent_op
*extra_op
);
87 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
88 struct extent_buffer
*leaf
,
89 struct btrfs_extent_item
*ei
);
90 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
91 struct btrfs_root
*root
,
92 u64 parent
, u64 root_objectid
,
93 u64 flags
, u64 owner
, u64 offset
,
94 struct btrfs_key
*ins
, int ref_mod
);
95 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
96 struct btrfs_root
*root
,
97 u64 parent
, u64 root_objectid
,
98 u64 flags
, struct btrfs_disk_key
*key
,
99 int level
, struct btrfs_key
*ins
);
100 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
101 struct btrfs_root
*extent_root
, u64 flags
,
103 static int find_next_key(struct btrfs_path
*path
, int level
,
104 struct btrfs_key
*key
);
105 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
106 int dump_block_groups
);
107 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
108 u64 num_bytes
, int reserve
,
110 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
112 int btrfs_pin_extent(struct btrfs_root
*root
,
113 u64 bytenr
, u64 num_bytes
, int reserved
);
114 static int __reserve_metadata_bytes(struct btrfs_root
*root
,
115 struct btrfs_space_info
*space_info
,
117 enum btrfs_reserve_flush_enum flush
);
118 static void space_info_add_new_bytes(struct btrfs_fs_info
*fs_info
,
119 struct btrfs_space_info
*space_info
,
121 static void space_info_add_old_bytes(struct btrfs_fs_info
*fs_info
,
122 struct btrfs_space_info
*space_info
,
126 block_group_cache_done(struct btrfs_block_group_cache
*cache
)
129 return cache
->cached
== BTRFS_CACHE_FINISHED
||
130 cache
->cached
== BTRFS_CACHE_ERROR
;
133 static int block_group_bits(struct btrfs_block_group_cache
*cache
, u64 bits
)
135 return (cache
->flags
& bits
) == bits
;
138 void btrfs_get_block_group(struct btrfs_block_group_cache
*cache
)
140 atomic_inc(&cache
->count
);
143 void btrfs_put_block_group(struct btrfs_block_group_cache
*cache
)
145 if (atomic_dec_and_test(&cache
->count
)) {
146 WARN_ON(cache
->pinned
> 0);
147 WARN_ON(cache
->reserved
> 0);
148 kfree(cache
->free_space_ctl
);
154 * this adds the block group to the fs_info rb tree for the block group
157 static int btrfs_add_block_group_cache(struct btrfs_fs_info
*info
,
158 struct btrfs_block_group_cache
*block_group
)
161 struct rb_node
*parent
= NULL
;
162 struct btrfs_block_group_cache
*cache
;
164 spin_lock(&info
->block_group_cache_lock
);
165 p
= &info
->block_group_cache_tree
.rb_node
;
169 cache
= rb_entry(parent
, struct btrfs_block_group_cache
,
171 if (block_group
->key
.objectid
< cache
->key
.objectid
) {
173 } else if (block_group
->key
.objectid
> cache
->key
.objectid
) {
176 spin_unlock(&info
->block_group_cache_lock
);
181 rb_link_node(&block_group
->cache_node
, parent
, p
);
182 rb_insert_color(&block_group
->cache_node
,
183 &info
->block_group_cache_tree
);
185 if (info
->first_logical_byte
> block_group
->key
.objectid
)
186 info
->first_logical_byte
= block_group
->key
.objectid
;
188 spin_unlock(&info
->block_group_cache_lock
);
194 * This will return the block group at or after bytenr if contains is 0, else
195 * it will return the block group that contains the bytenr
197 static struct btrfs_block_group_cache
*
198 block_group_cache_tree_search(struct btrfs_fs_info
*info
, u64 bytenr
,
201 struct btrfs_block_group_cache
*cache
, *ret
= NULL
;
205 spin_lock(&info
->block_group_cache_lock
);
206 n
= info
->block_group_cache_tree
.rb_node
;
209 cache
= rb_entry(n
, struct btrfs_block_group_cache
,
211 end
= cache
->key
.objectid
+ cache
->key
.offset
- 1;
212 start
= cache
->key
.objectid
;
214 if (bytenr
< start
) {
215 if (!contains
&& (!ret
|| start
< ret
->key
.objectid
))
218 } else if (bytenr
> start
) {
219 if (contains
&& bytenr
<= end
) {
230 btrfs_get_block_group(ret
);
231 if (bytenr
== 0 && info
->first_logical_byte
> ret
->key
.objectid
)
232 info
->first_logical_byte
= ret
->key
.objectid
;
234 spin_unlock(&info
->block_group_cache_lock
);
239 static int add_excluded_extent(struct btrfs_root
*root
,
240 u64 start
, u64 num_bytes
)
242 u64 end
= start
+ num_bytes
- 1;
243 set_extent_bits(&root
->fs_info
->freed_extents
[0],
244 start
, end
, EXTENT_UPTODATE
);
245 set_extent_bits(&root
->fs_info
->freed_extents
[1],
246 start
, end
, EXTENT_UPTODATE
);
250 static void free_excluded_extents(struct btrfs_root
*root
,
251 struct btrfs_block_group_cache
*cache
)
255 start
= cache
->key
.objectid
;
256 end
= start
+ cache
->key
.offset
- 1;
258 clear_extent_bits(&root
->fs_info
->freed_extents
[0],
259 start
, end
, EXTENT_UPTODATE
);
260 clear_extent_bits(&root
->fs_info
->freed_extents
[1],
261 start
, end
, EXTENT_UPTODATE
);
264 static int exclude_super_stripes(struct btrfs_root
*root
,
265 struct btrfs_block_group_cache
*cache
)
272 if (cache
->key
.objectid
< BTRFS_SUPER_INFO_OFFSET
) {
273 stripe_len
= BTRFS_SUPER_INFO_OFFSET
- cache
->key
.objectid
;
274 cache
->bytes_super
+= stripe_len
;
275 ret
= add_excluded_extent(root
, cache
->key
.objectid
,
281 for (i
= 0; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
282 bytenr
= btrfs_sb_offset(i
);
283 ret
= btrfs_rmap_block(&root
->fs_info
->mapping_tree
,
284 cache
->key
.objectid
, bytenr
,
285 0, &logical
, &nr
, &stripe_len
);
292 if (logical
[nr
] > cache
->key
.objectid
+
296 if (logical
[nr
] + stripe_len
<= cache
->key
.objectid
)
300 if (start
< cache
->key
.objectid
) {
301 start
= cache
->key
.objectid
;
302 len
= (logical
[nr
] + stripe_len
) - start
;
304 len
= min_t(u64
, stripe_len
,
305 cache
->key
.objectid
+
306 cache
->key
.offset
- start
);
309 cache
->bytes_super
+= len
;
310 ret
= add_excluded_extent(root
, start
, len
);
322 static struct btrfs_caching_control
*
323 get_caching_control(struct btrfs_block_group_cache
*cache
)
325 struct btrfs_caching_control
*ctl
;
327 spin_lock(&cache
->lock
);
328 if (!cache
->caching_ctl
) {
329 spin_unlock(&cache
->lock
);
333 ctl
= cache
->caching_ctl
;
334 atomic_inc(&ctl
->count
);
335 spin_unlock(&cache
->lock
);
339 static void put_caching_control(struct btrfs_caching_control
*ctl
)
341 if (atomic_dec_and_test(&ctl
->count
))
345 #ifdef CONFIG_BTRFS_DEBUG
346 static void fragment_free_space(struct btrfs_root
*root
,
347 struct btrfs_block_group_cache
*block_group
)
349 u64 start
= block_group
->key
.objectid
;
350 u64 len
= block_group
->key
.offset
;
351 u64 chunk
= block_group
->flags
& BTRFS_BLOCK_GROUP_METADATA
?
352 root
->nodesize
: root
->sectorsize
;
353 u64 step
= chunk
<< 1;
355 while (len
> chunk
) {
356 btrfs_remove_free_space(block_group
, start
, chunk
);
367 * this is only called by cache_block_group, since we could have freed extents
368 * we need to check the pinned_extents for any extents that can't be used yet
369 * since their free space will be released as soon as the transaction commits.
371 u64
add_new_free_space(struct btrfs_block_group_cache
*block_group
,
372 struct btrfs_fs_info
*info
, u64 start
, u64 end
)
374 u64 extent_start
, extent_end
, size
, total_added
= 0;
377 while (start
< end
) {
378 ret
= find_first_extent_bit(info
->pinned_extents
, start
,
379 &extent_start
, &extent_end
,
380 EXTENT_DIRTY
| EXTENT_UPTODATE
,
385 if (extent_start
<= start
) {
386 start
= extent_end
+ 1;
387 } else if (extent_start
> start
&& extent_start
< end
) {
388 size
= extent_start
- start
;
390 ret
= btrfs_add_free_space(block_group
, start
,
392 BUG_ON(ret
); /* -ENOMEM or logic error */
393 start
= extent_end
+ 1;
402 ret
= btrfs_add_free_space(block_group
, start
, size
);
403 BUG_ON(ret
); /* -ENOMEM or logic error */
409 static int load_extent_tree_free(struct btrfs_caching_control
*caching_ctl
)
411 struct btrfs_block_group_cache
*block_group
;
412 struct btrfs_fs_info
*fs_info
;
413 struct btrfs_root
*extent_root
;
414 struct btrfs_path
*path
;
415 struct extent_buffer
*leaf
;
416 struct btrfs_key key
;
423 block_group
= caching_ctl
->block_group
;
424 fs_info
= block_group
->fs_info
;
425 extent_root
= fs_info
->extent_root
;
427 path
= btrfs_alloc_path();
431 last
= max_t(u64
, block_group
->key
.objectid
, BTRFS_SUPER_INFO_OFFSET
);
433 #ifdef CONFIG_BTRFS_DEBUG
435 * If we're fragmenting we don't want to make anybody think we can
436 * allocate from this block group until we've had a chance to fragment
439 if (btrfs_should_fragment_free_space(extent_root
, block_group
))
443 * We don't want to deadlock with somebody trying to allocate a new
444 * extent for the extent root while also trying to search the extent
445 * root to add free space. So we skip locking and search the commit
446 * root, since its read-only
448 path
->skip_locking
= 1;
449 path
->search_commit_root
= 1;
450 path
->reada
= READA_FORWARD
;
454 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
457 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
461 leaf
= path
->nodes
[0];
462 nritems
= btrfs_header_nritems(leaf
);
465 if (btrfs_fs_closing(fs_info
) > 1) {
470 if (path
->slots
[0] < nritems
) {
471 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
473 ret
= find_next_key(path
, 0, &key
);
477 if (need_resched() ||
478 rwsem_is_contended(&fs_info
->commit_root_sem
)) {
480 caching_ctl
->progress
= last
;
481 btrfs_release_path(path
);
482 up_read(&fs_info
->commit_root_sem
);
483 mutex_unlock(&caching_ctl
->mutex
);
485 mutex_lock(&caching_ctl
->mutex
);
486 down_read(&fs_info
->commit_root_sem
);
490 ret
= btrfs_next_leaf(extent_root
, path
);
495 leaf
= path
->nodes
[0];
496 nritems
= btrfs_header_nritems(leaf
);
500 if (key
.objectid
< last
) {
503 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
506 caching_ctl
->progress
= last
;
507 btrfs_release_path(path
);
511 if (key
.objectid
< block_group
->key
.objectid
) {
516 if (key
.objectid
>= block_group
->key
.objectid
+
517 block_group
->key
.offset
)
520 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
||
521 key
.type
== BTRFS_METADATA_ITEM_KEY
) {
522 total_found
+= add_new_free_space(block_group
,
525 if (key
.type
== BTRFS_METADATA_ITEM_KEY
)
526 last
= key
.objectid
+
527 fs_info
->tree_root
->nodesize
;
529 last
= key
.objectid
+ key
.offset
;
531 if (total_found
> CACHING_CTL_WAKE_UP
) {
534 wake_up(&caching_ctl
->wait
);
541 total_found
+= add_new_free_space(block_group
, fs_info
, last
,
542 block_group
->key
.objectid
+
543 block_group
->key
.offset
);
544 caching_ctl
->progress
= (u64
)-1;
547 btrfs_free_path(path
);
551 static noinline
void caching_thread(struct btrfs_work
*work
)
553 struct btrfs_block_group_cache
*block_group
;
554 struct btrfs_fs_info
*fs_info
;
555 struct btrfs_caching_control
*caching_ctl
;
556 struct btrfs_root
*extent_root
;
559 caching_ctl
= container_of(work
, struct btrfs_caching_control
, work
);
560 block_group
= caching_ctl
->block_group
;
561 fs_info
= block_group
->fs_info
;
562 extent_root
= fs_info
->extent_root
;
564 mutex_lock(&caching_ctl
->mutex
);
565 down_read(&fs_info
->commit_root_sem
);
567 if (btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
))
568 ret
= load_free_space_tree(caching_ctl
);
570 ret
= load_extent_tree_free(caching_ctl
);
572 spin_lock(&block_group
->lock
);
573 block_group
->caching_ctl
= NULL
;
574 block_group
->cached
= ret
? BTRFS_CACHE_ERROR
: BTRFS_CACHE_FINISHED
;
575 spin_unlock(&block_group
->lock
);
577 #ifdef CONFIG_BTRFS_DEBUG
578 if (btrfs_should_fragment_free_space(extent_root
, block_group
)) {
581 spin_lock(&block_group
->space_info
->lock
);
582 spin_lock(&block_group
->lock
);
583 bytes_used
= block_group
->key
.offset
-
584 btrfs_block_group_used(&block_group
->item
);
585 block_group
->space_info
->bytes_used
+= bytes_used
>> 1;
586 spin_unlock(&block_group
->lock
);
587 spin_unlock(&block_group
->space_info
->lock
);
588 fragment_free_space(extent_root
, block_group
);
592 caching_ctl
->progress
= (u64
)-1;
594 up_read(&fs_info
->commit_root_sem
);
595 free_excluded_extents(fs_info
->extent_root
, block_group
);
596 mutex_unlock(&caching_ctl
->mutex
);
598 wake_up(&caching_ctl
->wait
);
600 put_caching_control(caching_ctl
);
601 btrfs_put_block_group(block_group
);
604 static int cache_block_group(struct btrfs_block_group_cache
*cache
,
608 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
609 struct btrfs_caching_control
*caching_ctl
;
612 caching_ctl
= kzalloc(sizeof(*caching_ctl
), GFP_NOFS
);
616 INIT_LIST_HEAD(&caching_ctl
->list
);
617 mutex_init(&caching_ctl
->mutex
);
618 init_waitqueue_head(&caching_ctl
->wait
);
619 caching_ctl
->block_group
= cache
;
620 caching_ctl
->progress
= cache
->key
.objectid
;
621 atomic_set(&caching_ctl
->count
, 1);
622 btrfs_init_work(&caching_ctl
->work
, btrfs_cache_helper
,
623 caching_thread
, NULL
, NULL
);
625 spin_lock(&cache
->lock
);
627 * This should be a rare occasion, but this could happen I think in the
628 * case where one thread starts to load the space cache info, and then
629 * some other thread starts a transaction commit which tries to do an
630 * allocation while the other thread is still loading the space cache
631 * info. The previous loop should have kept us from choosing this block
632 * group, but if we've moved to the state where we will wait on caching
633 * block groups we need to first check if we're doing a fast load here,
634 * so we can wait for it to finish, otherwise we could end up allocating
635 * from a block group who's cache gets evicted for one reason or
638 while (cache
->cached
== BTRFS_CACHE_FAST
) {
639 struct btrfs_caching_control
*ctl
;
641 ctl
= cache
->caching_ctl
;
642 atomic_inc(&ctl
->count
);
643 prepare_to_wait(&ctl
->wait
, &wait
, TASK_UNINTERRUPTIBLE
);
644 spin_unlock(&cache
->lock
);
648 finish_wait(&ctl
->wait
, &wait
);
649 put_caching_control(ctl
);
650 spin_lock(&cache
->lock
);
653 if (cache
->cached
!= BTRFS_CACHE_NO
) {
654 spin_unlock(&cache
->lock
);
658 WARN_ON(cache
->caching_ctl
);
659 cache
->caching_ctl
= caching_ctl
;
660 cache
->cached
= BTRFS_CACHE_FAST
;
661 spin_unlock(&cache
->lock
);
663 if (fs_info
->mount_opt
& BTRFS_MOUNT_SPACE_CACHE
) {
664 mutex_lock(&caching_ctl
->mutex
);
665 ret
= load_free_space_cache(fs_info
, cache
);
667 spin_lock(&cache
->lock
);
669 cache
->caching_ctl
= NULL
;
670 cache
->cached
= BTRFS_CACHE_FINISHED
;
671 cache
->last_byte_to_unpin
= (u64
)-1;
672 caching_ctl
->progress
= (u64
)-1;
674 if (load_cache_only
) {
675 cache
->caching_ctl
= NULL
;
676 cache
->cached
= BTRFS_CACHE_NO
;
678 cache
->cached
= BTRFS_CACHE_STARTED
;
679 cache
->has_caching_ctl
= 1;
682 spin_unlock(&cache
->lock
);
683 #ifdef CONFIG_BTRFS_DEBUG
685 btrfs_should_fragment_free_space(fs_info
->extent_root
,
689 spin_lock(&cache
->space_info
->lock
);
690 spin_lock(&cache
->lock
);
691 bytes_used
= cache
->key
.offset
-
692 btrfs_block_group_used(&cache
->item
);
693 cache
->space_info
->bytes_used
+= bytes_used
>> 1;
694 spin_unlock(&cache
->lock
);
695 spin_unlock(&cache
->space_info
->lock
);
696 fragment_free_space(fs_info
->extent_root
, cache
);
699 mutex_unlock(&caching_ctl
->mutex
);
701 wake_up(&caching_ctl
->wait
);
703 put_caching_control(caching_ctl
);
704 free_excluded_extents(fs_info
->extent_root
, cache
);
709 * We're either using the free space tree or no caching at all.
710 * Set cached to the appropriate value and wakeup any waiters.
712 spin_lock(&cache
->lock
);
713 if (load_cache_only
) {
714 cache
->caching_ctl
= NULL
;
715 cache
->cached
= BTRFS_CACHE_NO
;
717 cache
->cached
= BTRFS_CACHE_STARTED
;
718 cache
->has_caching_ctl
= 1;
720 spin_unlock(&cache
->lock
);
721 wake_up(&caching_ctl
->wait
);
724 if (load_cache_only
) {
725 put_caching_control(caching_ctl
);
729 down_write(&fs_info
->commit_root_sem
);
730 atomic_inc(&caching_ctl
->count
);
731 list_add_tail(&caching_ctl
->list
, &fs_info
->caching_block_groups
);
732 up_write(&fs_info
->commit_root_sem
);
734 btrfs_get_block_group(cache
);
736 btrfs_queue_work(fs_info
->caching_workers
, &caching_ctl
->work
);
742 * return the block group that starts at or after bytenr
744 static struct btrfs_block_group_cache
*
745 btrfs_lookup_first_block_group(struct btrfs_fs_info
*info
, u64 bytenr
)
747 struct btrfs_block_group_cache
*cache
;
749 cache
= block_group_cache_tree_search(info
, bytenr
, 0);
755 * return the block group that contains the given bytenr
757 struct btrfs_block_group_cache
*btrfs_lookup_block_group(
758 struct btrfs_fs_info
*info
,
761 struct btrfs_block_group_cache
*cache
;
763 cache
= block_group_cache_tree_search(info
, bytenr
, 1);
768 static struct btrfs_space_info
*__find_space_info(struct btrfs_fs_info
*info
,
771 struct list_head
*head
= &info
->space_info
;
772 struct btrfs_space_info
*found
;
774 flags
&= BTRFS_BLOCK_GROUP_TYPE_MASK
;
777 list_for_each_entry_rcu(found
, head
, list
) {
778 if (found
->flags
& flags
) {
788 * after adding space to the filesystem, we need to clear the full flags
789 * on all the space infos.
791 void btrfs_clear_space_info_full(struct btrfs_fs_info
*info
)
793 struct list_head
*head
= &info
->space_info
;
794 struct btrfs_space_info
*found
;
797 list_for_each_entry_rcu(found
, head
, list
)
802 /* simple helper to search for an existing data extent at a given offset */
803 int btrfs_lookup_data_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
806 struct btrfs_key key
;
807 struct btrfs_path
*path
;
809 path
= btrfs_alloc_path();
813 key
.objectid
= start
;
815 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
816 ret
= btrfs_search_slot(NULL
, root
->fs_info
->extent_root
, &key
, path
,
818 btrfs_free_path(path
);
823 * helper function to lookup reference count and flags of a tree block.
825 * the head node for delayed ref is used to store the sum of all the
826 * reference count modifications queued up in the rbtree. the head
827 * node may also store the extent flags to set. This way you can check
828 * to see what the reference count and extent flags would be if all of
829 * the delayed refs are not processed.
831 int btrfs_lookup_extent_info(struct btrfs_trans_handle
*trans
,
832 struct btrfs_root
*root
, u64 bytenr
,
833 u64 offset
, int metadata
, u64
*refs
, u64
*flags
)
835 struct btrfs_delayed_ref_head
*head
;
836 struct btrfs_delayed_ref_root
*delayed_refs
;
837 struct btrfs_path
*path
;
838 struct btrfs_extent_item
*ei
;
839 struct extent_buffer
*leaf
;
840 struct btrfs_key key
;
847 * If we don't have skinny metadata, don't bother doing anything
850 if (metadata
&& !btrfs_fs_incompat(root
->fs_info
, SKINNY_METADATA
)) {
851 offset
= root
->nodesize
;
855 path
= btrfs_alloc_path();
860 path
->skip_locking
= 1;
861 path
->search_commit_root
= 1;
865 key
.objectid
= bytenr
;
868 key
.type
= BTRFS_METADATA_ITEM_KEY
;
870 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
872 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
,
877 if (ret
> 0 && metadata
&& key
.type
== BTRFS_METADATA_ITEM_KEY
) {
878 if (path
->slots
[0]) {
880 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
882 if (key
.objectid
== bytenr
&&
883 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
884 key
.offset
== root
->nodesize
)
890 leaf
= path
->nodes
[0];
891 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
892 if (item_size
>= sizeof(*ei
)) {
893 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
894 struct btrfs_extent_item
);
895 num_refs
= btrfs_extent_refs(leaf
, ei
);
896 extent_flags
= btrfs_extent_flags(leaf
, ei
);
898 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
899 struct btrfs_extent_item_v0
*ei0
;
900 BUG_ON(item_size
!= sizeof(*ei0
));
901 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
902 struct btrfs_extent_item_v0
);
903 num_refs
= btrfs_extent_refs_v0(leaf
, ei0
);
904 /* FIXME: this isn't correct for data */
905 extent_flags
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
910 BUG_ON(num_refs
== 0);
920 delayed_refs
= &trans
->transaction
->delayed_refs
;
921 spin_lock(&delayed_refs
->lock
);
922 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
924 if (!mutex_trylock(&head
->mutex
)) {
925 atomic_inc(&head
->node
.refs
);
926 spin_unlock(&delayed_refs
->lock
);
928 btrfs_release_path(path
);
931 * Mutex was contended, block until it's released and try
934 mutex_lock(&head
->mutex
);
935 mutex_unlock(&head
->mutex
);
936 btrfs_put_delayed_ref(&head
->node
);
939 spin_lock(&head
->lock
);
940 if (head
->extent_op
&& head
->extent_op
->update_flags
)
941 extent_flags
|= head
->extent_op
->flags_to_set
;
943 BUG_ON(num_refs
== 0);
945 num_refs
+= head
->node
.ref_mod
;
946 spin_unlock(&head
->lock
);
947 mutex_unlock(&head
->mutex
);
949 spin_unlock(&delayed_refs
->lock
);
951 WARN_ON(num_refs
== 0);
955 *flags
= extent_flags
;
957 btrfs_free_path(path
);
962 * Back reference rules. Back refs have three main goals:
964 * 1) differentiate between all holders of references to an extent so that
965 * when a reference is dropped we can make sure it was a valid reference
966 * before freeing the extent.
968 * 2) Provide enough information to quickly find the holders of an extent
969 * if we notice a given block is corrupted or bad.
971 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
972 * maintenance. This is actually the same as #2, but with a slightly
973 * different use case.
975 * There are two kinds of back refs. The implicit back refs is optimized
976 * for pointers in non-shared tree blocks. For a given pointer in a block,
977 * back refs of this kind provide information about the block's owner tree
978 * and the pointer's key. These information allow us to find the block by
979 * b-tree searching. The full back refs is for pointers in tree blocks not
980 * referenced by their owner trees. The location of tree block is recorded
981 * in the back refs. Actually the full back refs is generic, and can be
982 * used in all cases the implicit back refs is used. The major shortcoming
983 * of the full back refs is its overhead. Every time a tree block gets
984 * COWed, we have to update back refs entry for all pointers in it.
986 * For a newly allocated tree block, we use implicit back refs for
987 * pointers in it. This means most tree related operations only involve
988 * implicit back refs. For a tree block created in old transaction, the
989 * only way to drop a reference to it is COW it. So we can detect the
990 * event that tree block loses its owner tree's reference and do the
991 * back refs conversion.
993 * When a tree block is COWed through a tree, there are four cases:
995 * The reference count of the block is one and the tree is the block's
996 * owner tree. Nothing to do in this case.
998 * The reference count of the block is one and the tree is not the
999 * block's owner tree. In this case, full back refs is used for pointers
1000 * in the block. Remove these full back refs, add implicit back refs for
1001 * every pointers in the new block.
1003 * The reference count of the block is greater than one and the tree is
1004 * the block's owner tree. In this case, implicit back refs is used for
1005 * pointers in the block. Add full back refs for every pointers in the
1006 * block, increase lower level extents' reference counts. The original
1007 * implicit back refs are entailed to the new block.
1009 * The reference count of the block is greater than one and the tree is
1010 * not the block's owner tree. Add implicit back refs for every pointer in
1011 * the new block, increase lower level extents' reference count.
1013 * Back Reference Key composing:
1015 * The key objectid corresponds to the first byte in the extent,
1016 * The key type is used to differentiate between types of back refs.
1017 * There are different meanings of the key offset for different types
1020 * File extents can be referenced by:
1022 * - multiple snapshots, subvolumes, or different generations in one subvol
1023 * - different files inside a single subvolume
1024 * - different offsets inside a file (bookend extents in file.c)
1026 * The extent ref structure for the implicit back refs has fields for:
1028 * - Objectid of the subvolume root
1029 * - objectid of the file holding the reference
1030 * - original offset in the file
1031 * - how many bookend extents
1033 * The key offset for the implicit back refs is hash of the first
1036 * The extent ref structure for the full back refs has field for:
1038 * - number of pointers in the tree leaf
1040 * The key offset for the implicit back refs is the first byte of
1043 * When a file extent is allocated, The implicit back refs is used.
1044 * the fields are filled in:
1046 * (root_key.objectid, inode objectid, offset in file, 1)
1048 * When a file extent is removed file truncation, we find the
1049 * corresponding implicit back refs and check the following fields:
1051 * (btrfs_header_owner(leaf), inode objectid, offset in file)
1053 * Btree extents can be referenced by:
1055 * - Different subvolumes
1057 * Both the implicit back refs and the full back refs for tree blocks
1058 * only consist of key. The key offset for the implicit back refs is
1059 * objectid of block's owner tree. The key offset for the full back refs
1060 * is the first byte of parent block.
1062 * When implicit back refs is used, information about the lowest key and
1063 * level of the tree block are required. These information are stored in
1064 * tree block info structure.
1067 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1068 static int convert_extent_item_v0(struct btrfs_trans_handle
*trans
,
1069 struct btrfs_root
*root
,
1070 struct btrfs_path
*path
,
1071 u64 owner
, u32 extra_size
)
1073 struct btrfs_extent_item
*item
;
1074 struct btrfs_extent_item_v0
*ei0
;
1075 struct btrfs_extent_ref_v0
*ref0
;
1076 struct btrfs_tree_block_info
*bi
;
1077 struct extent_buffer
*leaf
;
1078 struct btrfs_key key
;
1079 struct btrfs_key found_key
;
1080 u32 new_size
= sizeof(*item
);
1084 leaf
= path
->nodes
[0];
1085 BUG_ON(btrfs_item_size_nr(leaf
, path
->slots
[0]) != sizeof(*ei0
));
1087 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1088 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1089 struct btrfs_extent_item_v0
);
1090 refs
= btrfs_extent_refs_v0(leaf
, ei0
);
1092 if (owner
== (u64
)-1) {
1094 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1095 ret
= btrfs_next_leaf(root
, path
);
1098 BUG_ON(ret
> 0); /* Corruption */
1099 leaf
= path
->nodes
[0];
1101 btrfs_item_key_to_cpu(leaf
, &found_key
,
1103 BUG_ON(key
.objectid
!= found_key
.objectid
);
1104 if (found_key
.type
!= BTRFS_EXTENT_REF_V0_KEY
) {
1108 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1109 struct btrfs_extent_ref_v0
);
1110 owner
= btrfs_ref_objectid_v0(leaf
, ref0
);
1114 btrfs_release_path(path
);
1116 if (owner
< BTRFS_FIRST_FREE_OBJECTID
)
1117 new_size
+= sizeof(*bi
);
1119 new_size
-= sizeof(*ei0
);
1120 ret
= btrfs_search_slot(trans
, root
, &key
, path
,
1121 new_size
+ extra_size
, 1);
1124 BUG_ON(ret
); /* Corruption */
1126 btrfs_extend_item(root
, path
, new_size
);
1128 leaf
= path
->nodes
[0];
1129 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1130 btrfs_set_extent_refs(leaf
, item
, refs
);
1131 /* FIXME: get real generation */
1132 btrfs_set_extent_generation(leaf
, item
, 0);
1133 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1134 btrfs_set_extent_flags(leaf
, item
,
1135 BTRFS_EXTENT_FLAG_TREE_BLOCK
|
1136 BTRFS_BLOCK_FLAG_FULL_BACKREF
);
1137 bi
= (struct btrfs_tree_block_info
*)(item
+ 1);
1138 /* FIXME: get first key of the block */
1139 memset_extent_buffer(leaf
, 0, (unsigned long)bi
, sizeof(*bi
));
1140 btrfs_set_tree_block_level(leaf
, bi
, (int)owner
);
1142 btrfs_set_extent_flags(leaf
, item
, BTRFS_EXTENT_FLAG_DATA
);
1144 btrfs_mark_buffer_dirty(leaf
);
1149 static u64
hash_extent_data_ref(u64 root_objectid
, u64 owner
, u64 offset
)
1151 u32 high_crc
= ~(u32
)0;
1152 u32 low_crc
= ~(u32
)0;
1155 lenum
= cpu_to_le64(root_objectid
);
1156 high_crc
= btrfs_crc32c(high_crc
, &lenum
, sizeof(lenum
));
1157 lenum
= cpu_to_le64(owner
);
1158 low_crc
= btrfs_crc32c(low_crc
, &lenum
, sizeof(lenum
));
1159 lenum
= cpu_to_le64(offset
);
1160 low_crc
= btrfs_crc32c(low_crc
, &lenum
, sizeof(lenum
));
1162 return ((u64
)high_crc
<< 31) ^ (u64
)low_crc
;
1165 static u64
hash_extent_data_ref_item(struct extent_buffer
*leaf
,
1166 struct btrfs_extent_data_ref
*ref
)
1168 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf
, ref
),
1169 btrfs_extent_data_ref_objectid(leaf
, ref
),
1170 btrfs_extent_data_ref_offset(leaf
, ref
));
1173 static int match_extent_data_ref(struct extent_buffer
*leaf
,
1174 struct btrfs_extent_data_ref
*ref
,
1175 u64 root_objectid
, u64 owner
, u64 offset
)
1177 if (btrfs_extent_data_ref_root(leaf
, ref
) != root_objectid
||
1178 btrfs_extent_data_ref_objectid(leaf
, ref
) != owner
||
1179 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
1184 static noinline
int lookup_extent_data_ref(struct btrfs_trans_handle
*trans
,
1185 struct btrfs_root
*root
,
1186 struct btrfs_path
*path
,
1187 u64 bytenr
, u64 parent
,
1189 u64 owner
, u64 offset
)
1191 struct btrfs_key key
;
1192 struct btrfs_extent_data_ref
*ref
;
1193 struct extent_buffer
*leaf
;
1199 key
.objectid
= bytenr
;
1201 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1202 key
.offset
= parent
;
1204 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1205 key
.offset
= hash_extent_data_ref(root_objectid
,
1210 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1219 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1220 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1221 btrfs_release_path(path
);
1222 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1233 leaf
= path
->nodes
[0];
1234 nritems
= btrfs_header_nritems(leaf
);
1236 if (path
->slots
[0] >= nritems
) {
1237 ret
= btrfs_next_leaf(root
, path
);
1243 leaf
= path
->nodes
[0];
1244 nritems
= btrfs_header_nritems(leaf
);
1248 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1249 if (key
.objectid
!= bytenr
||
1250 key
.type
!= BTRFS_EXTENT_DATA_REF_KEY
)
1253 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1254 struct btrfs_extent_data_ref
);
1256 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1259 btrfs_release_path(path
);
1271 static noinline
int insert_extent_data_ref(struct btrfs_trans_handle
*trans
,
1272 struct btrfs_root
*root
,
1273 struct btrfs_path
*path
,
1274 u64 bytenr
, u64 parent
,
1275 u64 root_objectid
, u64 owner
,
1276 u64 offset
, int refs_to_add
)
1278 struct btrfs_key key
;
1279 struct extent_buffer
*leaf
;
1284 key
.objectid
= bytenr
;
1286 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1287 key
.offset
= parent
;
1288 size
= sizeof(struct btrfs_shared_data_ref
);
1290 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1291 key
.offset
= hash_extent_data_ref(root_objectid
,
1293 size
= sizeof(struct btrfs_extent_data_ref
);
1296 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, size
);
1297 if (ret
&& ret
!= -EEXIST
)
1300 leaf
= path
->nodes
[0];
1302 struct btrfs_shared_data_ref
*ref
;
1303 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1304 struct btrfs_shared_data_ref
);
1306 btrfs_set_shared_data_ref_count(leaf
, ref
, refs_to_add
);
1308 num_refs
= btrfs_shared_data_ref_count(leaf
, ref
);
1309 num_refs
+= refs_to_add
;
1310 btrfs_set_shared_data_ref_count(leaf
, ref
, num_refs
);
1313 struct btrfs_extent_data_ref
*ref
;
1314 while (ret
== -EEXIST
) {
1315 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1316 struct btrfs_extent_data_ref
);
1317 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1320 btrfs_release_path(path
);
1322 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1324 if (ret
&& ret
!= -EEXIST
)
1327 leaf
= path
->nodes
[0];
1329 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1330 struct btrfs_extent_data_ref
);
1332 btrfs_set_extent_data_ref_root(leaf
, ref
,
1334 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
1335 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
1336 btrfs_set_extent_data_ref_count(leaf
, ref
, refs_to_add
);
1338 num_refs
= btrfs_extent_data_ref_count(leaf
, ref
);
1339 num_refs
+= refs_to_add
;
1340 btrfs_set_extent_data_ref_count(leaf
, ref
, num_refs
);
1343 btrfs_mark_buffer_dirty(leaf
);
1346 btrfs_release_path(path
);
1350 static noinline
int remove_extent_data_ref(struct btrfs_trans_handle
*trans
,
1351 struct btrfs_root
*root
,
1352 struct btrfs_path
*path
,
1353 int refs_to_drop
, int *last_ref
)
1355 struct btrfs_key key
;
1356 struct btrfs_extent_data_ref
*ref1
= NULL
;
1357 struct btrfs_shared_data_ref
*ref2
= NULL
;
1358 struct extent_buffer
*leaf
;
1362 leaf
= path
->nodes
[0];
1363 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1365 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1366 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1367 struct btrfs_extent_data_ref
);
1368 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1369 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1370 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1371 struct btrfs_shared_data_ref
);
1372 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1373 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1374 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1375 struct btrfs_extent_ref_v0
*ref0
;
1376 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1377 struct btrfs_extent_ref_v0
);
1378 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1384 BUG_ON(num_refs
< refs_to_drop
);
1385 num_refs
-= refs_to_drop
;
1387 if (num_refs
== 0) {
1388 ret
= btrfs_del_item(trans
, root
, path
);
1391 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
)
1392 btrfs_set_extent_data_ref_count(leaf
, ref1
, num_refs
);
1393 else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
)
1394 btrfs_set_shared_data_ref_count(leaf
, ref2
, num_refs
);
1395 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1397 struct btrfs_extent_ref_v0
*ref0
;
1398 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1399 struct btrfs_extent_ref_v0
);
1400 btrfs_set_ref_count_v0(leaf
, ref0
, num_refs
);
1403 btrfs_mark_buffer_dirty(leaf
);
1408 static noinline u32
extent_data_ref_count(struct btrfs_path
*path
,
1409 struct btrfs_extent_inline_ref
*iref
)
1411 struct btrfs_key key
;
1412 struct extent_buffer
*leaf
;
1413 struct btrfs_extent_data_ref
*ref1
;
1414 struct btrfs_shared_data_ref
*ref2
;
1417 leaf
= path
->nodes
[0];
1418 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1420 if (btrfs_extent_inline_ref_type(leaf
, iref
) ==
1421 BTRFS_EXTENT_DATA_REF_KEY
) {
1422 ref1
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1423 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1425 ref2
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1426 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1428 } else if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1429 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1430 struct btrfs_extent_data_ref
);
1431 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1432 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1433 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1434 struct btrfs_shared_data_ref
);
1435 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1436 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1437 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1438 struct btrfs_extent_ref_v0
*ref0
;
1439 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1440 struct btrfs_extent_ref_v0
);
1441 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1449 static noinline
int lookup_tree_block_ref(struct btrfs_trans_handle
*trans
,
1450 struct btrfs_root
*root
,
1451 struct btrfs_path
*path
,
1452 u64 bytenr
, u64 parent
,
1455 struct btrfs_key key
;
1458 key
.objectid
= bytenr
;
1460 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1461 key
.offset
= parent
;
1463 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1464 key
.offset
= root_objectid
;
1467 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1470 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1471 if (ret
== -ENOENT
&& parent
) {
1472 btrfs_release_path(path
);
1473 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1474 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1482 static noinline
int insert_tree_block_ref(struct btrfs_trans_handle
*trans
,
1483 struct btrfs_root
*root
,
1484 struct btrfs_path
*path
,
1485 u64 bytenr
, u64 parent
,
1488 struct btrfs_key key
;
1491 key
.objectid
= bytenr
;
1493 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1494 key
.offset
= parent
;
1496 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1497 key
.offset
= root_objectid
;
1500 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
1501 btrfs_release_path(path
);
1505 static inline int extent_ref_type(u64 parent
, u64 owner
)
1508 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1510 type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1512 type
= BTRFS_TREE_BLOCK_REF_KEY
;
1515 type
= BTRFS_SHARED_DATA_REF_KEY
;
1517 type
= BTRFS_EXTENT_DATA_REF_KEY
;
1522 static int find_next_key(struct btrfs_path
*path
, int level
,
1523 struct btrfs_key
*key
)
1526 for (; level
< BTRFS_MAX_LEVEL
; level
++) {
1527 if (!path
->nodes
[level
])
1529 if (path
->slots
[level
] + 1 >=
1530 btrfs_header_nritems(path
->nodes
[level
]))
1533 btrfs_item_key_to_cpu(path
->nodes
[level
], key
,
1534 path
->slots
[level
] + 1);
1536 btrfs_node_key_to_cpu(path
->nodes
[level
], key
,
1537 path
->slots
[level
] + 1);
1544 * look for inline back ref. if back ref is found, *ref_ret is set
1545 * to the address of inline back ref, and 0 is returned.
1547 * if back ref isn't found, *ref_ret is set to the address where it
1548 * should be inserted, and -ENOENT is returned.
1550 * if insert is true and there are too many inline back refs, the path
1551 * points to the extent item, and -EAGAIN is returned.
1553 * NOTE: inline back refs are ordered in the same way that back ref
1554 * items in the tree are ordered.
1556 static noinline_for_stack
1557 int lookup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1558 struct btrfs_root
*root
,
1559 struct btrfs_path
*path
,
1560 struct btrfs_extent_inline_ref
**ref_ret
,
1561 u64 bytenr
, u64 num_bytes
,
1562 u64 parent
, u64 root_objectid
,
1563 u64 owner
, u64 offset
, int insert
)
1565 struct btrfs_key key
;
1566 struct extent_buffer
*leaf
;
1567 struct btrfs_extent_item
*ei
;
1568 struct btrfs_extent_inline_ref
*iref
;
1578 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
1581 key
.objectid
= bytenr
;
1582 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1583 key
.offset
= num_bytes
;
1585 want
= extent_ref_type(parent
, owner
);
1587 extra_size
= btrfs_extent_inline_ref_size(want
);
1588 path
->keep_locks
= 1;
1593 * Owner is our parent level, so we can just add one to get the level
1594 * for the block we are interested in.
1596 if (skinny_metadata
&& owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1597 key
.type
= BTRFS_METADATA_ITEM_KEY
;
1602 ret
= btrfs_search_slot(trans
, root
, &key
, path
, extra_size
, 1);
1609 * We may be a newly converted file system which still has the old fat
1610 * extent entries for metadata, so try and see if we have one of those.
1612 if (ret
> 0 && skinny_metadata
) {
1613 skinny_metadata
= false;
1614 if (path
->slots
[0]) {
1616 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
1618 if (key
.objectid
== bytenr
&&
1619 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
1620 key
.offset
== num_bytes
)
1624 key
.objectid
= bytenr
;
1625 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1626 key
.offset
= num_bytes
;
1627 btrfs_release_path(path
);
1632 if (ret
&& !insert
) {
1635 } else if (WARN_ON(ret
)) {
1640 leaf
= path
->nodes
[0];
1641 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1642 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1643 if (item_size
< sizeof(*ei
)) {
1648 ret
= convert_extent_item_v0(trans
, root
, path
, owner
,
1654 leaf
= path
->nodes
[0];
1655 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1658 BUG_ON(item_size
< sizeof(*ei
));
1660 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1661 flags
= btrfs_extent_flags(leaf
, ei
);
1663 ptr
= (unsigned long)(ei
+ 1);
1664 end
= (unsigned long)ei
+ item_size
;
1666 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
&& !skinny_metadata
) {
1667 ptr
+= sizeof(struct btrfs_tree_block_info
);
1677 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1678 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1682 ptr
+= btrfs_extent_inline_ref_size(type
);
1686 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1687 struct btrfs_extent_data_ref
*dref
;
1688 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1689 if (match_extent_data_ref(leaf
, dref
, root_objectid
,
1694 if (hash_extent_data_ref_item(leaf
, dref
) <
1695 hash_extent_data_ref(root_objectid
, owner
, offset
))
1699 ref_offset
= btrfs_extent_inline_ref_offset(leaf
, iref
);
1701 if (parent
== ref_offset
) {
1705 if (ref_offset
< parent
)
1708 if (root_objectid
== ref_offset
) {
1712 if (ref_offset
< root_objectid
)
1716 ptr
+= btrfs_extent_inline_ref_size(type
);
1718 if (err
== -ENOENT
&& insert
) {
1719 if (item_size
+ extra_size
>=
1720 BTRFS_MAX_EXTENT_ITEM_SIZE(root
)) {
1725 * To add new inline back ref, we have to make sure
1726 * there is no corresponding back ref item.
1727 * For simplicity, we just do not add new inline back
1728 * ref if there is any kind of item for this block
1730 if (find_next_key(path
, 0, &key
) == 0 &&
1731 key
.objectid
== bytenr
&&
1732 key
.type
< BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1737 *ref_ret
= (struct btrfs_extent_inline_ref
*)ptr
;
1740 path
->keep_locks
= 0;
1741 btrfs_unlock_up_safe(path
, 1);
1747 * helper to add new inline back ref
1749 static noinline_for_stack
1750 void setup_inline_extent_backref(struct btrfs_root
*root
,
1751 struct btrfs_path
*path
,
1752 struct btrfs_extent_inline_ref
*iref
,
1753 u64 parent
, u64 root_objectid
,
1754 u64 owner
, u64 offset
, int refs_to_add
,
1755 struct btrfs_delayed_extent_op
*extent_op
)
1757 struct extent_buffer
*leaf
;
1758 struct btrfs_extent_item
*ei
;
1761 unsigned long item_offset
;
1766 leaf
= path
->nodes
[0];
1767 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1768 item_offset
= (unsigned long)iref
- (unsigned long)ei
;
1770 type
= extent_ref_type(parent
, owner
);
1771 size
= btrfs_extent_inline_ref_size(type
);
1773 btrfs_extend_item(root
, path
, size
);
1775 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1776 refs
= btrfs_extent_refs(leaf
, ei
);
1777 refs
+= refs_to_add
;
1778 btrfs_set_extent_refs(leaf
, ei
, refs
);
1780 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1782 ptr
= (unsigned long)ei
+ item_offset
;
1783 end
= (unsigned long)ei
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
1784 if (ptr
< end
- size
)
1785 memmove_extent_buffer(leaf
, ptr
+ size
, ptr
,
1788 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1789 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
1790 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1791 struct btrfs_extent_data_ref
*dref
;
1792 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1793 btrfs_set_extent_data_ref_root(leaf
, dref
, root_objectid
);
1794 btrfs_set_extent_data_ref_objectid(leaf
, dref
, owner
);
1795 btrfs_set_extent_data_ref_offset(leaf
, dref
, offset
);
1796 btrfs_set_extent_data_ref_count(leaf
, dref
, refs_to_add
);
1797 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1798 struct btrfs_shared_data_ref
*sref
;
1799 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1800 btrfs_set_shared_data_ref_count(leaf
, sref
, refs_to_add
);
1801 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1802 } else if (type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
1803 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1805 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
1807 btrfs_mark_buffer_dirty(leaf
);
1810 static int lookup_extent_backref(struct btrfs_trans_handle
*trans
,
1811 struct btrfs_root
*root
,
1812 struct btrfs_path
*path
,
1813 struct btrfs_extent_inline_ref
**ref_ret
,
1814 u64 bytenr
, u64 num_bytes
, u64 parent
,
1815 u64 root_objectid
, u64 owner
, u64 offset
)
1819 ret
= lookup_inline_extent_backref(trans
, root
, path
, ref_ret
,
1820 bytenr
, num_bytes
, parent
,
1821 root_objectid
, owner
, offset
, 0);
1825 btrfs_release_path(path
);
1828 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1829 ret
= lookup_tree_block_ref(trans
, root
, path
, bytenr
, parent
,
1832 ret
= lookup_extent_data_ref(trans
, root
, path
, bytenr
, parent
,
1833 root_objectid
, owner
, offset
);
1839 * helper to update/remove inline back ref
1841 static noinline_for_stack
1842 void update_inline_extent_backref(struct btrfs_root
*root
,
1843 struct btrfs_path
*path
,
1844 struct btrfs_extent_inline_ref
*iref
,
1846 struct btrfs_delayed_extent_op
*extent_op
,
1849 struct extent_buffer
*leaf
;
1850 struct btrfs_extent_item
*ei
;
1851 struct btrfs_extent_data_ref
*dref
= NULL
;
1852 struct btrfs_shared_data_ref
*sref
= NULL
;
1860 leaf
= path
->nodes
[0];
1861 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1862 refs
= btrfs_extent_refs(leaf
, ei
);
1863 WARN_ON(refs_to_mod
< 0 && refs
+ refs_to_mod
<= 0);
1864 refs
+= refs_to_mod
;
1865 btrfs_set_extent_refs(leaf
, ei
, refs
);
1867 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1869 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1871 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1872 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1873 refs
= btrfs_extent_data_ref_count(leaf
, dref
);
1874 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1875 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1876 refs
= btrfs_shared_data_ref_count(leaf
, sref
);
1879 BUG_ON(refs_to_mod
!= -1);
1882 BUG_ON(refs_to_mod
< 0 && refs
< -refs_to_mod
);
1883 refs
+= refs_to_mod
;
1886 if (type
== BTRFS_EXTENT_DATA_REF_KEY
)
1887 btrfs_set_extent_data_ref_count(leaf
, dref
, refs
);
1889 btrfs_set_shared_data_ref_count(leaf
, sref
, refs
);
1892 size
= btrfs_extent_inline_ref_size(type
);
1893 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1894 ptr
= (unsigned long)iref
;
1895 end
= (unsigned long)ei
+ item_size
;
1896 if (ptr
+ size
< end
)
1897 memmove_extent_buffer(leaf
, ptr
, ptr
+ size
,
1900 btrfs_truncate_item(root
, path
, item_size
, 1);
1902 btrfs_mark_buffer_dirty(leaf
);
1905 static noinline_for_stack
1906 int insert_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1907 struct btrfs_root
*root
,
1908 struct btrfs_path
*path
,
1909 u64 bytenr
, u64 num_bytes
, u64 parent
,
1910 u64 root_objectid
, u64 owner
,
1911 u64 offset
, int refs_to_add
,
1912 struct btrfs_delayed_extent_op
*extent_op
)
1914 struct btrfs_extent_inline_ref
*iref
;
1917 ret
= lookup_inline_extent_backref(trans
, root
, path
, &iref
,
1918 bytenr
, num_bytes
, parent
,
1919 root_objectid
, owner
, offset
, 1);
1921 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
);
1922 update_inline_extent_backref(root
, path
, iref
,
1923 refs_to_add
, extent_op
, NULL
);
1924 } else if (ret
== -ENOENT
) {
1925 setup_inline_extent_backref(root
, path
, iref
, parent
,
1926 root_objectid
, owner
, offset
,
1927 refs_to_add
, extent_op
);
1933 static int insert_extent_backref(struct btrfs_trans_handle
*trans
,
1934 struct btrfs_root
*root
,
1935 struct btrfs_path
*path
,
1936 u64 bytenr
, u64 parent
, u64 root_objectid
,
1937 u64 owner
, u64 offset
, int refs_to_add
)
1940 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1941 BUG_ON(refs_to_add
!= 1);
1942 ret
= insert_tree_block_ref(trans
, root
, path
, bytenr
,
1943 parent
, root_objectid
);
1945 ret
= insert_extent_data_ref(trans
, root
, path
, bytenr
,
1946 parent
, root_objectid
,
1947 owner
, offset
, refs_to_add
);
1952 static int remove_extent_backref(struct btrfs_trans_handle
*trans
,
1953 struct btrfs_root
*root
,
1954 struct btrfs_path
*path
,
1955 struct btrfs_extent_inline_ref
*iref
,
1956 int refs_to_drop
, int is_data
, int *last_ref
)
1960 BUG_ON(!is_data
&& refs_to_drop
!= 1);
1962 update_inline_extent_backref(root
, path
, iref
,
1963 -refs_to_drop
, NULL
, last_ref
);
1964 } else if (is_data
) {
1965 ret
= remove_extent_data_ref(trans
, root
, path
, refs_to_drop
,
1969 ret
= btrfs_del_item(trans
, root
, path
);
1974 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
1975 static int btrfs_issue_discard(struct block_device
*bdev
, u64 start
, u64 len
,
1976 u64
*discarded_bytes
)
1979 u64 bytes_left
, end
;
1980 u64 aligned_start
= ALIGN(start
, 1 << 9);
1982 if (WARN_ON(start
!= aligned_start
)) {
1983 len
-= aligned_start
- start
;
1984 len
= round_down(len
, 1 << 9);
1985 start
= aligned_start
;
1988 *discarded_bytes
= 0;
1996 /* Skip any superblocks on this device. */
1997 for (j
= 0; j
< BTRFS_SUPER_MIRROR_MAX
; j
++) {
1998 u64 sb_start
= btrfs_sb_offset(j
);
1999 u64 sb_end
= sb_start
+ BTRFS_SUPER_INFO_SIZE
;
2000 u64 size
= sb_start
- start
;
2002 if (!in_range(sb_start
, start
, bytes_left
) &&
2003 !in_range(sb_end
, start
, bytes_left
) &&
2004 !in_range(start
, sb_start
, BTRFS_SUPER_INFO_SIZE
))
2008 * Superblock spans beginning of range. Adjust start and
2011 if (sb_start
<= start
) {
2012 start
+= sb_end
- start
;
2017 bytes_left
= end
- start
;
2022 ret
= blkdev_issue_discard(bdev
, start
>> 9, size
>> 9,
2025 *discarded_bytes
+= size
;
2026 else if (ret
!= -EOPNOTSUPP
)
2035 bytes_left
= end
- start
;
2039 ret
= blkdev_issue_discard(bdev
, start
>> 9, bytes_left
>> 9,
2042 *discarded_bytes
+= bytes_left
;
2047 int btrfs_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
2048 u64 num_bytes
, u64
*actual_bytes
)
2051 u64 discarded_bytes
= 0;
2052 struct btrfs_bio
*bbio
= NULL
;
2056 * Avoid races with device replace and make sure our bbio has devices
2057 * associated to its stripes that don't go away while we are discarding.
2059 btrfs_bio_counter_inc_blocked(root
->fs_info
);
2060 /* Tell the block device(s) that the sectors can be discarded */
2061 ret
= btrfs_map_block(root
->fs_info
, REQ_DISCARD
,
2062 bytenr
, &num_bytes
, &bbio
, 0);
2063 /* Error condition is -ENOMEM */
2065 struct btrfs_bio_stripe
*stripe
= bbio
->stripes
;
2069 for (i
= 0; i
< bbio
->num_stripes
; i
++, stripe
++) {
2071 if (!stripe
->dev
->can_discard
)
2074 ret
= btrfs_issue_discard(stripe
->dev
->bdev
,
2079 discarded_bytes
+= bytes
;
2080 else if (ret
!= -EOPNOTSUPP
)
2081 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2084 * Just in case we get back EOPNOTSUPP for some reason,
2085 * just ignore the return value so we don't screw up
2086 * people calling discard_extent.
2090 btrfs_put_bbio(bbio
);
2092 btrfs_bio_counter_dec(root
->fs_info
);
2095 *actual_bytes
= discarded_bytes
;
2098 if (ret
== -EOPNOTSUPP
)
2103 /* Can return -ENOMEM */
2104 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
2105 struct btrfs_root
*root
,
2106 u64 bytenr
, u64 num_bytes
, u64 parent
,
2107 u64 root_objectid
, u64 owner
, u64 offset
)
2110 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2112 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
&&
2113 root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
2115 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
2116 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
2118 parent
, root_objectid
, (int)owner
,
2119 BTRFS_ADD_DELAYED_REF
, NULL
);
2121 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
2122 num_bytes
, parent
, root_objectid
,
2124 BTRFS_ADD_DELAYED_REF
, NULL
);
2129 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
2130 struct btrfs_root
*root
,
2131 struct btrfs_delayed_ref_node
*node
,
2132 u64 parent
, u64 root_objectid
,
2133 u64 owner
, u64 offset
, int refs_to_add
,
2134 struct btrfs_delayed_extent_op
*extent_op
)
2136 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2137 struct btrfs_path
*path
;
2138 struct extent_buffer
*leaf
;
2139 struct btrfs_extent_item
*item
;
2140 struct btrfs_key key
;
2141 u64 bytenr
= node
->bytenr
;
2142 u64 num_bytes
= node
->num_bytes
;
2146 path
= btrfs_alloc_path();
2150 path
->reada
= READA_FORWARD
;
2151 path
->leave_spinning
= 1;
2152 /* this will setup the path even if it fails to insert the back ref */
2153 ret
= insert_inline_extent_backref(trans
, fs_info
->extent_root
, path
,
2154 bytenr
, num_bytes
, parent
,
2155 root_objectid
, owner
, offset
,
2156 refs_to_add
, extent_op
);
2157 if ((ret
< 0 && ret
!= -EAGAIN
) || !ret
)
2161 * Ok we had -EAGAIN which means we didn't have space to insert and
2162 * inline extent ref, so just update the reference count and add a
2165 leaf
= path
->nodes
[0];
2166 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2167 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2168 refs
= btrfs_extent_refs(leaf
, item
);
2169 btrfs_set_extent_refs(leaf
, item
, refs
+ refs_to_add
);
2171 __run_delayed_extent_op(extent_op
, leaf
, item
);
2173 btrfs_mark_buffer_dirty(leaf
);
2174 btrfs_release_path(path
);
2176 path
->reada
= READA_FORWARD
;
2177 path
->leave_spinning
= 1;
2178 /* now insert the actual backref */
2179 ret
= insert_extent_backref(trans
, root
->fs_info
->extent_root
,
2180 path
, bytenr
, parent
, root_objectid
,
2181 owner
, offset
, refs_to_add
);
2183 btrfs_abort_transaction(trans
, root
, ret
);
2185 btrfs_free_path(path
);
2189 static int run_delayed_data_ref(struct btrfs_trans_handle
*trans
,
2190 struct btrfs_root
*root
,
2191 struct btrfs_delayed_ref_node
*node
,
2192 struct btrfs_delayed_extent_op
*extent_op
,
2193 int insert_reserved
)
2196 struct btrfs_delayed_data_ref
*ref
;
2197 struct btrfs_key ins
;
2202 ins
.objectid
= node
->bytenr
;
2203 ins
.offset
= node
->num_bytes
;
2204 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2206 ref
= btrfs_delayed_node_to_data_ref(node
);
2207 trace_run_delayed_data_ref(node
, ref
, node
->action
);
2209 if (node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2210 parent
= ref
->parent
;
2211 ref_root
= ref
->root
;
2213 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2215 flags
|= extent_op
->flags_to_set
;
2216 ret
= alloc_reserved_file_extent(trans
, root
,
2217 parent
, ref_root
, flags
,
2218 ref
->objectid
, ref
->offset
,
2219 &ins
, node
->ref_mod
);
2220 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2221 ret
= __btrfs_inc_extent_ref(trans
, root
, node
, parent
,
2222 ref_root
, ref
->objectid
,
2223 ref
->offset
, node
->ref_mod
,
2225 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2226 ret
= __btrfs_free_extent(trans
, root
, node
, parent
,
2227 ref_root
, ref
->objectid
,
2228 ref
->offset
, node
->ref_mod
,
2236 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
2237 struct extent_buffer
*leaf
,
2238 struct btrfs_extent_item
*ei
)
2240 u64 flags
= btrfs_extent_flags(leaf
, ei
);
2241 if (extent_op
->update_flags
) {
2242 flags
|= extent_op
->flags_to_set
;
2243 btrfs_set_extent_flags(leaf
, ei
, flags
);
2246 if (extent_op
->update_key
) {
2247 struct btrfs_tree_block_info
*bi
;
2248 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
));
2249 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
2250 btrfs_set_tree_block_key(leaf
, bi
, &extent_op
->key
);
2254 static int run_delayed_extent_op(struct btrfs_trans_handle
*trans
,
2255 struct btrfs_root
*root
,
2256 struct btrfs_delayed_ref_node
*node
,
2257 struct btrfs_delayed_extent_op
*extent_op
)
2259 struct btrfs_key key
;
2260 struct btrfs_path
*path
;
2261 struct btrfs_extent_item
*ei
;
2262 struct extent_buffer
*leaf
;
2266 int metadata
= !extent_op
->is_data
;
2271 if (metadata
&& !btrfs_fs_incompat(root
->fs_info
, SKINNY_METADATA
))
2274 path
= btrfs_alloc_path();
2278 key
.objectid
= node
->bytenr
;
2281 key
.type
= BTRFS_METADATA_ITEM_KEY
;
2282 key
.offset
= extent_op
->level
;
2284 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2285 key
.offset
= node
->num_bytes
;
2289 path
->reada
= READA_FORWARD
;
2290 path
->leave_spinning
= 1;
2291 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
,
2299 if (path
->slots
[0] > 0) {
2301 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
2303 if (key
.objectid
== node
->bytenr
&&
2304 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
2305 key
.offset
== node
->num_bytes
)
2309 btrfs_release_path(path
);
2312 key
.objectid
= node
->bytenr
;
2313 key
.offset
= node
->num_bytes
;
2314 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2323 leaf
= path
->nodes
[0];
2324 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2325 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2326 if (item_size
< sizeof(*ei
)) {
2327 ret
= convert_extent_item_v0(trans
, root
->fs_info
->extent_root
,
2333 leaf
= path
->nodes
[0];
2334 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2337 BUG_ON(item_size
< sizeof(*ei
));
2338 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2339 __run_delayed_extent_op(extent_op
, leaf
, ei
);
2341 btrfs_mark_buffer_dirty(leaf
);
2343 btrfs_free_path(path
);
2347 static int run_delayed_tree_ref(struct btrfs_trans_handle
*trans
,
2348 struct btrfs_root
*root
,
2349 struct btrfs_delayed_ref_node
*node
,
2350 struct btrfs_delayed_extent_op
*extent_op
,
2351 int insert_reserved
)
2354 struct btrfs_delayed_tree_ref
*ref
;
2355 struct btrfs_key ins
;
2358 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
2361 ref
= btrfs_delayed_node_to_tree_ref(node
);
2362 trace_run_delayed_tree_ref(node
, ref
, node
->action
);
2364 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2365 parent
= ref
->parent
;
2366 ref_root
= ref
->root
;
2368 ins
.objectid
= node
->bytenr
;
2369 if (skinny_metadata
) {
2370 ins
.offset
= ref
->level
;
2371 ins
.type
= BTRFS_METADATA_ITEM_KEY
;
2373 ins
.offset
= node
->num_bytes
;
2374 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2377 BUG_ON(node
->ref_mod
!= 1);
2378 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2379 BUG_ON(!extent_op
|| !extent_op
->update_flags
);
2380 ret
= alloc_reserved_tree_block(trans
, root
,
2382 extent_op
->flags_to_set
,
2385 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2386 ret
= __btrfs_inc_extent_ref(trans
, root
, node
,
2390 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2391 ret
= __btrfs_free_extent(trans
, root
, node
,
2393 ref
->level
, 0, 1, extent_op
);
2400 /* helper function to actually process a single delayed ref entry */
2401 static int run_one_delayed_ref(struct btrfs_trans_handle
*trans
,
2402 struct btrfs_root
*root
,
2403 struct btrfs_delayed_ref_node
*node
,
2404 struct btrfs_delayed_extent_op
*extent_op
,
2405 int insert_reserved
)
2409 if (trans
->aborted
) {
2410 if (insert_reserved
)
2411 btrfs_pin_extent(root
, node
->bytenr
,
2412 node
->num_bytes
, 1);
2416 if (btrfs_delayed_ref_is_head(node
)) {
2417 struct btrfs_delayed_ref_head
*head
;
2419 * we've hit the end of the chain and we were supposed
2420 * to insert this extent into the tree. But, it got
2421 * deleted before we ever needed to insert it, so all
2422 * we have to do is clean up the accounting
2425 head
= btrfs_delayed_node_to_head(node
);
2426 trace_run_delayed_ref_head(node
, head
, node
->action
);
2428 if (insert_reserved
) {
2429 btrfs_pin_extent(root
, node
->bytenr
,
2430 node
->num_bytes
, 1);
2431 if (head
->is_data
) {
2432 ret
= btrfs_del_csums(trans
, root
,
2438 /* Also free its reserved qgroup space */
2439 btrfs_qgroup_free_delayed_ref(root
->fs_info
,
2440 head
->qgroup_ref_root
,
2441 head
->qgroup_reserved
);
2445 if (node
->type
== BTRFS_TREE_BLOCK_REF_KEY
||
2446 node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2447 ret
= run_delayed_tree_ref(trans
, root
, node
, extent_op
,
2449 else if (node
->type
== BTRFS_EXTENT_DATA_REF_KEY
||
2450 node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2451 ret
= run_delayed_data_ref(trans
, root
, node
, extent_op
,
2458 static inline struct btrfs_delayed_ref_node
*
2459 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
2461 struct btrfs_delayed_ref_node
*ref
;
2463 if (list_empty(&head
->ref_list
))
2467 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2468 * This is to prevent a ref count from going down to zero, which deletes
2469 * the extent item from the extent tree, when there still are references
2470 * to add, which would fail because they would not find the extent item.
2472 list_for_each_entry(ref
, &head
->ref_list
, list
) {
2473 if (ref
->action
== BTRFS_ADD_DELAYED_REF
)
2477 return list_entry(head
->ref_list
.next
, struct btrfs_delayed_ref_node
,
2482 * Returns 0 on success or if called with an already aborted transaction.
2483 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2485 static noinline
int __btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2486 struct btrfs_root
*root
,
2489 struct btrfs_delayed_ref_root
*delayed_refs
;
2490 struct btrfs_delayed_ref_node
*ref
;
2491 struct btrfs_delayed_ref_head
*locked_ref
= NULL
;
2492 struct btrfs_delayed_extent_op
*extent_op
;
2493 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2494 ktime_t start
= ktime_get();
2496 unsigned long count
= 0;
2497 unsigned long actual_count
= 0;
2498 int must_insert_reserved
= 0;
2500 delayed_refs
= &trans
->transaction
->delayed_refs
;
2506 spin_lock(&delayed_refs
->lock
);
2507 locked_ref
= btrfs_select_ref_head(trans
);
2509 spin_unlock(&delayed_refs
->lock
);
2513 /* grab the lock that says we are going to process
2514 * all the refs for this head */
2515 ret
= btrfs_delayed_ref_lock(trans
, locked_ref
);
2516 spin_unlock(&delayed_refs
->lock
);
2518 * we may have dropped the spin lock to get the head
2519 * mutex lock, and that might have given someone else
2520 * time to free the head. If that's true, it has been
2521 * removed from our list and we can move on.
2523 if (ret
== -EAGAIN
) {
2531 * We need to try and merge add/drops of the same ref since we
2532 * can run into issues with relocate dropping the implicit ref
2533 * and then it being added back again before the drop can
2534 * finish. If we merged anything we need to re-loop so we can
2536 * Or we can get node references of the same type that weren't
2537 * merged when created due to bumps in the tree mod seq, and
2538 * we need to merge them to prevent adding an inline extent
2539 * backref before dropping it (triggering a BUG_ON at
2540 * insert_inline_extent_backref()).
2542 spin_lock(&locked_ref
->lock
);
2543 btrfs_merge_delayed_refs(trans
, fs_info
, delayed_refs
,
2547 * locked_ref is the head node, so we have to go one
2548 * node back for any delayed ref updates
2550 ref
= select_delayed_ref(locked_ref
);
2552 if (ref
&& ref
->seq
&&
2553 btrfs_check_delayed_seq(fs_info
, delayed_refs
, ref
->seq
)) {
2554 spin_unlock(&locked_ref
->lock
);
2555 btrfs_delayed_ref_unlock(locked_ref
);
2556 spin_lock(&delayed_refs
->lock
);
2557 locked_ref
->processing
= 0;
2558 delayed_refs
->num_heads_ready
++;
2559 spin_unlock(&delayed_refs
->lock
);
2567 * record the must insert reserved flag before we
2568 * drop the spin lock.
2570 must_insert_reserved
= locked_ref
->must_insert_reserved
;
2571 locked_ref
->must_insert_reserved
= 0;
2573 extent_op
= locked_ref
->extent_op
;
2574 locked_ref
->extent_op
= NULL
;
2579 /* All delayed refs have been processed, Go ahead
2580 * and send the head node to run_one_delayed_ref,
2581 * so that any accounting fixes can happen
2583 ref
= &locked_ref
->node
;
2585 if (extent_op
&& must_insert_reserved
) {
2586 btrfs_free_delayed_extent_op(extent_op
);
2591 spin_unlock(&locked_ref
->lock
);
2592 ret
= run_delayed_extent_op(trans
, root
,
2594 btrfs_free_delayed_extent_op(extent_op
);
2598 * Need to reset must_insert_reserved if
2599 * there was an error so the abort stuff
2600 * can cleanup the reserved space
2603 if (must_insert_reserved
)
2604 locked_ref
->must_insert_reserved
= 1;
2605 locked_ref
->processing
= 0;
2606 btrfs_debug(fs_info
, "run_delayed_extent_op returned %d", ret
);
2607 btrfs_delayed_ref_unlock(locked_ref
);
2614 * Need to drop our head ref lock and re-acquire the
2615 * delayed ref lock and then re-check to make sure
2618 spin_unlock(&locked_ref
->lock
);
2619 spin_lock(&delayed_refs
->lock
);
2620 spin_lock(&locked_ref
->lock
);
2621 if (!list_empty(&locked_ref
->ref_list
) ||
2622 locked_ref
->extent_op
) {
2623 spin_unlock(&locked_ref
->lock
);
2624 spin_unlock(&delayed_refs
->lock
);
2628 delayed_refs
->num_heads
--;
2629 rb_erase(&locked_ref
->href_node
,
2630 &delayed_refs
->href_root
);
2631 spin_unlock(&delayed_refs
->lock
);
2635 list_del(&ref
->list
);
2637 atomic_dec(&delayed_refs
->num_entries
);
2639 if (!btrfs_delayed_ref_is_head(ref
)) {
2641 * when we play the delayed ref, also correct the
2644 switch (ref
->action
) {
2645 case BTRFS_ADD_DELAYED_REF
:
2646 case BTRFS_ADD_DELAYED_EXTENT
:
2647 locked_ref
->node
.ref_mod
-= ref
->ref_mod
;
2649 case BTRFS_DROP_DELAYED_REF
:
2650 locked_ref
->node
.ref_mod
+= ref
->ref_mod
;
2656 spin_unlock(&locked_ref
->lock
);
2658 ret
= run_one_delayed_ref(trans
, root
, ref
, extent_op
,
2659 must_insert_reserved
);
2661 btrfs_free_delayed_extent_op(extent_op
);
2663 locked_ref
->processing
= 0;
2664 btrfs_delayed_ref_unlock(locked_ref
);
2665 btrfs_put_delayed_ref(ref
);
2666 btrfs_debug(fs_info
, "run_one_delayed_ref returned %d", ret
);
2671 * If this node is a head, that means all the refs in this head
2672 * have been dealt with, and we will pick the next head to deal
2673 * with, so we must unlock the head and drop it from the cluster
2674 * list before we release it.
2676 if (btrfs_delayed_ref_is_head(ref
)) {
2677 if (locked_ref
->is_data
&&
2678 locked_ref
->total_ref_mod
< 0) {
2679 spin_lock(&delayed_refs
->lock
);
2680 delayed_refs
->pending_csums
-= ref
->num_bytes
;
2681 spin_unlock(&delayed_refs
->lock
);
2683 btrfs_delayed_ref_unlock(locked_ref
);
2686 btrfs_put_delayed_ref(ref
);
2692 * We don't want to include ref heads since we can have empty ref heads
2693 * and those will drastically skew our runtime down since we just do
2694 * accounting, no actual extent tree updates.
2696 if (actual_count
> 0) {
2697 u64 runtime
= ktime_to_ns(ktime_sub(ktime_get(), start
));
2701 * We weigh the current average higher than our current runtime
2702 * to avoid large swings in the average.
2704 spin_lock(&delayed_refs
->lock
);
2705 avg
= fs_info
->avg_delayed_ref_runtime
* 3 + runtime
;
2706 fs_info
->avg_delayed_ref_runtime
= avg
>> 2; /* div by 4 */
2707 spin_unlock(&delayed_refs
->lock
);
2712 #ifdef SCRAMBLE_DELAYED_REFS
2714 * Normally delayed refs get processed in ascending bytenr order. This
2715 * correlates in most cases to the order added. To expose dependencies on this
2716 * order, we start to process the tree in the middle instead of the beginning
2718 static u64
find_middle(struct rb_root
*root
)
2720 struct rb_node
*n
= root
->rb_node
;
2721 struct btrfs_delayed_ref_node
*entry
;
2724 u64 first
= 0, last
= 0;
2728 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2729 first
= entry
->bytenr
;
2733 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2734 last
= entry
->bytenr
;
2739 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2740 WARN_ON(!entry
->in_tree
);
2742 middle
= entry
->bytenr
;
2755 static inline u64
heads_to_leaves(struct btrfs_root
*root
, u64 heads
)
2759 num_bytes
= heads
* (sizeof(struct btrfs_extent_item
) +
2760 sizeof(struct btrfs_extent_inline_ref
));
2761 if (!btrfs_fs_incompat(root
->fs_info
, SKINNY_METADATA
))
2762 num_bytes
+= heads
* sizeof(struct btrfs_tree_block_info
);
2765 * We don't ever fill up leaves all the way so multiply by 2 just to be
2766 * closer to what we're really going to want to use.
2768 return div_u64(num_bytes
, BTRFS_LEAF_DATA_SIZE(root
));
2772 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2773 * would require to store the csums for that many bytes.
2775 u64
btrfs_csum_bytes_to_leaves(struct btrfs_root
*root
, u64 csum_bytes
)
2778 u64 num_csums_per_leaf
;
2781 csum_size
= BTRFS_LEAF_DATA_SIZE(root
) - sizeof(struct btrfs_item
);
2782 num_csums_per_leaf
= div64_u64(csum_size
,
2783 (u64
)btrfs_super_csum_size(root
->fs_info
->super_copy
));
2784 num_csums
= div64_u64(csum_bytes
, root
->sectorsize
);
2785 num_csums
+= num_csums_per_leaf
- 1;
2786 num_csums
= div64_u64(num_csums
, num_csums_per_leaf
);
2790 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle
*trans
,
2791 struct btrfs_root
*root
)
2793 struct btrfs_block_rsv
*global_rsv
;
2794 u64 num_heads
= trans
->transaction
->delayed_refs
.num_heads_ready
;
2795 u64 csum_bytes
= trans
->transaction
->delayed_refs
.pending_csums
;
2796 u64 num_dirty_bgs
= trans
->transaction
->num_dirty_bgs
;
2797 u64 num_bytes
, num_dirty_bgs_bytes
;
2800 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
2801 num_heads
= heads_to_leaves(root
, num_heads
);
2803 num_bytes
+= (num_heads
- 1) * root
->nodesize
;
2805 num_bytes
+= btrfs_csum_bytes_to_leaves(root
, csum_bytes
) * root
->nodesize
;
2806 num_dirty_bgs_bytes
= btrfs_calc_trans_metadata_size(root
,
2808 global_rsv
= &root
->fs_info
->global_block_rsv
;
2811 * If we can't allocate any more chunks lets make sure we have _lots_ of
2812 * wiggle room since running delayed refs can create more delayed refs.
2814 if (global_rsv
->space_info
->full
) {
2815 num_dirty_bgs_bytes
<<= 1;
2819 spin_lock(&global_rsv
->lock
);
2820 if (global_rsv
->reserved
<= num_bytes
+ num_dirty_bgs_bytes
)
2822 spin_unlock(&global_rsv
->lock
);
2826 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle
*trans
,
2827 struct btrfs_root
*root
)
2829 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2831 atomic_read(&trans
->transaction
->delayed_refs
.num_entries
);
2836 avg_runtime
= fs_info
->avg_delayed_ref_runtime
;
2837 val
= num_entries
* avg_runtime
;
2838 if (num_entries
* avg_runtime
>= NSEC_PER_SEC
)
2840 if (val
>= NSEC_PER_SEC
/ 2)
2843 return btrfs_check_space_for_delayed_refs(trans
, root
);
2846 struct async_delayed_refs
{
2847 struct btrfs_root
*root
;
2852 struct completion wait
;
2853 struct btrfs_work work
;
2856 static void delayed_ref_async_start(struct btrfs_work
*work
)
2858 struct async_delayed_refs
*async
;
2859 struct btrfs_trans_handle
*trans
;
2862 async
= container_of(work
, struct async_delayed_refs
, work
);
2864 /* if the commit is already started, we don't need to wait here */
2865 if (btrfs_transaction_blocked(async
->root
->fs_info
))
2868 trans
= btrfs_join_transaction(async
->root
);
2869 if (IS_ERR(trans
)) {
2870 async
->error
= PTR_ERR(trans
);
2875 * trans->sync means that when we call end_transaction, we won't
2876 * wait on delayed refs
2880 /* Don't bother flushing if we got into a different transaction */
2881 if (trans
->transid
> async
->transid
)
2884 ret
= btrfs_run_delayed_refs(trans
, async
->root
, async
->count
);
2888 ret
= btrfs_end_transaction(trans
, async
->root
);
2889 if (ret
&& !async
->error
)
2893 complete(&async
->wait
);
2898 int btrfs_async_run_delayed_refs(struct btrfs_root
*root
,
2899 unsigned long count
, u64 transid
, int wait
)
2901 struct async_delayed_refs
*async
;
2904 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
2908 async
->root
= root
->fs_info
->tree_root
;
2909 async
->count
= count
;
2911 async
->transid
= transid
;
2916 init_completion(&async
->wait
);
2918 btrfs_init_work(&async
->work
, btrfs_extent_refs_helper
,
2919 delayed_ref_async_start
, NULL
, NULL
);
2921 btrfs_queue_work(root
->fs_info
->extent_workers
, &async
->work
);
2924 wait_for_completion(&async
->wait
);
2933 * this starts processing the delayed reference count updates and
2934 * extent insertions we have queued up so far. count can be
2935 * 0, which means to process everything in the tree at the start
2936 * of the run (but not newly added entries), or it can be some target
2937 * number you'd like to process.
2939 * Returns 0 on success or if called with an aborted transaction
2940 * Returns <0 on error and aborts the transaction
2942 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2943 struct btrfs_root
*root
, unsigned long count
)
2945 struct rb_node
*node
;
2946 struct btrfs_delayed_ref_root
*delayed_refs
;
2947 struct btrfs_delayed_ref_head
*head
;
2949 int run_all
= count
== (unsigned long)-1;
2950 bool can_flush_pending_bgs
= trans
->can_flush_pending_bgs
;
2952 /* We'll clean this up in btrfs_cleanup_transaction */
2956 if (root
->fs_info
->creating_free_space_tree
)
2959 if (root
== root
->fs_info
->extent_root
)
2960 root
= root
->fs_info
->tree_root
;
2962 delayed_refs
= &trans
->transaction
->delayed_refs
;
2964 count
= atomic_read(&delayed_refs
->num_entries
) * 2;
2967 #ifdef SCRAMBLE_DELAYED_REFS
2968 delayed_refs
->run_delayed_start
= find_middle(&delayed_refs
->root
);
2970 trans
->can_flush_pending_bgs
= false;
2971 ret
= __btrfs_run_delayed_refs(trans
, root
, count
);
2973 btrfs_abort_transaction(trans
, root
, ret
);
2978 if (!list_empty(&trans
->new_bgs
))
2979 btrfs_create_pending_block_groups(trans
, root
);
2981 spin_lock(&delayed_refs
->lock
);
2982 node
= rb_first(&delayed_refs
->href_root
);
2984 spin_unlock(&delayed_refs
->lock
);
2987 count
= (unsigned long)-1;
2990 head
= rb_entry(node
, struct btrfs_delayed_ref_head
,
2992 if (btrfs_delayed_ref_is_head(&head
->node
)) {
2993 struct btrfs_delayed_ref_node
*ref
;
2996 atomic_inc(&ref
->refs
);
2998 spin_unlock(&delayed_refs
->lock
);
3000 * Mutex was contended, block until it's
3001 * released and try again
3003 mutex_lock(&head
->mutex
);
3004 mutex_unlock(&head
->mutex
);
3006 btrfs_put_delayed_ref(ref
);
3012 node
= rb_next(node
);
3014 spin_unlock(&delayed_refs
->lock
);
3019 assert_qgroups_uptodate(trans
);
3020 trans
->can_flush_pending_bgs
= can_flush_pending_bgs
;
3024 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
3025 struct btrfs_root
*root
,
3026 u64 bytenr
, u64 num_bytes
, u64 flags
,
3027 int level
, int is_data
)
3029 struct btrfs_delayed_extent_op
*extent_op
;
3032 extent_op
= btrfs_alloc_delayed_extent_op();
3036 extent_op
->flags_to_set
= flags
;
3037 extent_op
->update_flags
= true;
3038 extent_op
->update_key
= false;
3039 extent_op
->is_data
= is_data
? true : false;
3040 extent_op
->level
= level
;
3042 ret
= btrfs_add_delayed_extent_op(root
->fs_info
, trans
, bytenr
,
3043 num_bytes
, extent_op
);
3045 btrfs_free_delayed_extent_op(extent_op
);
3049 static noinline
int check_delayed_ref(struct btrfs_trans_handle
*trans
,
3050 struct btrfs_root
*root
,
3051 struct btrfs_path
*path
,
3052 u64 objectid
, u64 offset
, u64 bytenr
)
3054 struct btrfs_delayed_ref_head
*head
;
3055 struct btrfs_delayed_ref_node
*ref
;
3056 struct btrfs_delayed_data_ref
*data_ref
;
3057 struct btrfs_delayed_ref_root
*delayed_refs
;
3060 delayed_refs
= &trans
->transaction
->delayed_refs
;
3061 spin_lock(&delayed_refs
->lock
);
3062 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
3064 spin_unlock(&delayed_refs
->lock
);
3068 if (!mutex_trylock(&head
->mutex
)) {
3069 atomic_inc(&head
->node
.refs
);
3070 spin_unlock(&delayed_refs
->lock
);
3072 btrfs_release_path(path
);
3075 * Mutex was contended, block until it's released and let
3078 mutex_lock(&head
->mutex
);
3079 mutex_unlock(&head
->mutex
);
3080 btrfs_put_delayed_ref(&head
->node
);
3083 spin_unlock(&delayed_refs
->lock
);
3085 spin_lock(&head
->lock
);
3086 list_for_each_entry(ref
, &head
->ref_list
, list
) {
3087 /* If it's a shared ref we know a cross reference exists */
3088 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
) {
3093 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
3096 * If our ref doesn't match the one we're currently looking at
3097 * then we have a cross reference.
3099 if (data_ref
->root
!= root
->root_key
.objectid
||
3100 data_ref
->objectid
!= objectid
||
3101 data_ref
->offset
!= offset
) {
3106 spin_unlock(&head
->lock
);
3107 mutex_unlock(&head
->mutex
);
3111 static noinline
int check_committed_ref(struct btrfs_trans_handle
*trans
,
3112 struct btrfs_root
*root
,
3113 struct btrfs_path
*path
,
3114 u64 objectid
, u64 offset
, u64 bytenr
)
3116 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
3117 struct extent_buffer
*leaf
;
3118 struct btrfs_extent_data_ref
*ref
;
3119 struct btrfs_extent_inline_ref
*iref
;
3120 struct btrfs_extent_item
*ei
;
3121 struct btrfs_key key
;
3125 key
.objectid
= bytenr
;
3126 key
.offset
= (u64
)-1;
3127 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
3129 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
3132 BUG_ON(ret
== 0); /* Corruption */
3135 if (path
->slots
[0] == 0)
3139 leaf
= path
->nodes
[0];
3140 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
3142 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
3146 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
3147 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3148 if (item_size
< sizeof(*ei
)) {
3149 WARN_ON(item_size
!= sizeof(struct btrfs_extent_item_v0
));
3153 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
3155 if (item_size
!= sizeof(*ei
) +
3156 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
3159 if (btrfs_extent_generation(leaf
, ei
) <=
3160 btrfs_root_last_snapshot(&root
->root_item
))
3163 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
3164 if (btrfs_extent_inline_ref_type(leaf
, iref
) !=
3165 BTRFS_EXTENT_DATA_REF_KEY
)
3168 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
3169 if (btrfs_extent_refs(leaf
, ei
) !=
3170 btrfs_extent_data_ref_count(leaf
, ref
) ||
3171 btrfs_extent_data_ref_root(leaf
, ref
) !=
3172 root
->root_key
.objectid
||
3173 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
3174 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
3182 int btrfs_cross_ref_exist(struct btrfs_trans_handle
*trans
,
3183 struct btrfs_root
*root
,
3184 u64 objectid
, u64 offset
, u64 bytenr
)
3186 struct btrfs_path
*path
;
3190 path
= btrfs_alloc_path();
3195 ret
= check_committed_ref(trans
, root
, path
, objectid
,
3197 if (ret
&& ret
!= -ENOENT
)
3200 ret2
= check_delayed_ref(trans
, root
, path
, objectid
,
3202 } while (ret2
== -EAGAIN
);
3204 if (ret2
&& ret2
!= -ENOENT
) {
3209 if (ret
!= -ENOENT
|| ret2
!= -ENOENT
)
3212 btrfs_free_path(path
);
3213 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
3218 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
3219 struct btrfs_root
*root
,
3220 struct extent_buffer
*buf
,
3221 int full_backref
, int inc
)
3228 struct btrfs_key key
;
3229 struct btrfs_file_extent_item
*fi
;
3233 int (*process_func
)(struct btrfs_trans_handle
*, struct btrfs_root
*,
3234 u64
, u64
, u64
, u64
, u64
, u64
);
3237 if (btrfs_test_is_dummy_root(root
))
3240 ref_root
= btrfs_header_owner(buf
);
3241 nritems
= btrfs_header_nritems(buf
);
3242 level
= btrfs_header_level(buf
);
3244 if (!test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) && level
== 0)
3248 process_func
= btrfs_inc_extent_ref
;
3250 process_func
= btrfs_free_extent
;
3253 parent
= buf
->start
;
3257 for (i
= 0; i
< nritems
; i
++) {
3259 btrfs_item_key_to_cpu(buf
, &key
, i
);
3260 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
3262 fi
= btrfs_item_ptr(buf
, i
,
3263 struct btrfs_file_extent_item
);
3264 if (btrfs_file_extent_type(buf
, fi
) ==
3265 BTRFS_FILE_EXTENT_INLINE
)
3267 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
3271 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
3272 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
3273 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
3274 parent
, ref_root
, key
.objectid
,
3279 bytenr
= btrfs_node_blockptr(buf
, i
);
3280 num_bytes
= root
->nodesize
;
3281 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
3282 parent
, ref_root
, level
- 1, 0);
3292 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3293 struct extent_buffer
*buf
, int full_backref
)
3295 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1);
3298 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3299 struct extent_buffer
*buf
, int full_backref
)
3301 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0);
3304 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
3305 struct btrfs_root
*root
,
3306 struct btrfs_path
*path
,
3307 struct btrfs_block_group_cache
*cache
)
3310 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
3312 struct extent_buffer
*leaf
;
3314 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
3321 leaf
= path
->nodes
[0];
3322 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
3323 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
3324 btrfs_mark_buffer_dirty(leaf
);
3326 btrfs_release_path(path
);
3331 static struct btrfs_block_group_cache
*
3332 next_block_group(struct btrfs_root
*root
,
3333 struct btrfs_block_group_cache
*cache
)
3335 struct rb_node
*node
;
3337 spin_lock(&root
->fs_info
->block_group_cache_lock
);
3339 /* If our block group was removed, we need a full search. */
3340 if (RB_EMPTY_NODE(&cache
->cache_node
)) {
3341 const u64 next_bytenr
= cache
->key
.objectid
+ cache
->key
.offset
;
3343 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
3344 btrfs_put_block_group(cache
);
3345 cache
= btrfs_lookup_first_block_group(root
->fs_info
,
3349 node
= rb_next(&cache
->cache_node
);
3350 btrfs_put_block_group(cache
);
3352 cache
= rb_entry(node
, struct btrfs_block_group_cache
,
3354 btrfs_get_block_group(cache
);
3357 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
3361 static int cache_save_setup(struct btrfs_block_group_cache
*block_group
,
3362 struct btrfs_trans_handle
*trans
,
3363 struct btrfs_path
*path
)
3365 struct btrfs_root
*root
= block_group
->fs_info
->tree_root
;
3366 struct inode
*inode
= NULL
;
3368 int dcs
= BTRFS_DC_ERROR
;
3374 * If this block group is smaller than 100 megs don't bother caching the
3377 if (block_group
->key
.offset
< (100 * SZ_1M
)) {
3378 spin_lock(&block_group
->lock
);
3379 block_group
->disk_cache_state
= BTRFS_DC_WRITTEN
;
3380 spin_unlock(&block_group
->lock
);
3387 inode
= lookup_free_space_inode(root
, block_group
, path
);
3388 if (IS_ERR(inode
) && PTR_ERR(inode
) != -ENOENT
) {
3389 ret
= PTR_ERR(inode
);
3390 btrfs_release_path(path
);
3394 if (IS_ERR(inode
)) {
3398 if (block_group
->ro
)
3401 ret
= create_free_space_inode(root
, trans
, block_group
, path
);
3407 /* We've already setup this transaction, go ahead and exit */
3408 if (block_group
->cache_generation
== trans
->transid
&&
3409 i_size_read(inode
)) {
3410 dcs
= BTRFS_DC_SETUP
;
3415 * We want to set the generation to 0, that way if anything goes wrong
3416 * from here on out we know not to trust this cache when we load up next
3419 BTRFS_I(inode
)->generation
= 0;
3420 ret
= btrfs_update_inode(trans
, root
, inode
);
3423 * So theoretically we could recover from this, simply set the
3424 * super cache generation to 0 so we know to invalidate the
3425 * cache, but then we'd have to keep track of the block groups
3426 * that fail this way so we know we _have_ to reset this cache
3427 * before the next commit or risk reading stale cache. So to
3428 * limit our exposure to horrible edge cases lets just abort the
3429 * transaction, this only happens in really bad situations
3432 btrfs_abort_transaction(trans
, root
, ret
);
3437 if (i_size_read(inode
) > 0) {
3438 ret
= btrfs_check_trunc_cache_free_space(root
,
3439 &root
->fs_info
->global_block_rsv
);
3443 ret
= btrfs_truncate_free_space_cache(root
, trans
, NULL
, inode
);
3448 spin_lock(&block_group
->lock
);
3449 if (block_group
->cached
!= BTRFS_CACHE_FINISHED
||
3450 !btrfs_test_opt(root
, SPACE_CACHE
)) {
3452 * don't bother trying to write stuff out _if_
3453 * a) we're not cached,
3454 * b) we're with nospace_cache mount option.
3456 dcs
= BTRFS_DC_WRITTEN
;
3457 spin_unlock(&block_group
->lock
);
3460 spin_unlock(&block_group
->lock
);
3463 * We hit an ENOSPC when setting up the cache in this transaction, just
3464 * skip doing the setup, we've already cleared the cache so we're safe.
3466 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC
, &trans
->transaction
->flags
)) {
3472 * Try to preallocate enough space based on how big the block group is.
3473 * Keep in mind this has to include any pinned space which could end up
3474 * taking up quite a bit since it's not folded into the other space
3477 num_pages
= div_u64(block_group
->key
.offset
, SZ_256M
);
3482 num_pages
*= PAGE_SIZE
;
3484 ret
= btrfs_check_data_free_space(inode
, 0, num_pages
);
3488 ret
= btrfs_prealloc_file_range_trans(inode
, trans
, 0, 0, num_pages
,
3489 num_pages
, num_pages
,
3492 * Our cache requires contiguous chunks so that we don't modify a bunch
3493 * of metadata or split extents when writing the cache out, which means
3494 * we can enospc if we are heavily fragmented in addition to just normal
3495 * out of space conditions. So if we hit this just skip setting up any
3496 * other block groups for this transaction, maybe we'll unpin enough
3497 * space the next time around.
3500 dcs
= BTRFS_DC_SETUP
;
3501 else if (ret
== -ENOSPC
)
3502 set_bit(BTRFS_TRANS_CACHE_ENOSPC
, &trans
->transaction
->flags
);
3503 btrfs_free_reserved_data_space(inode
, 0, num_pages
);
3508 btrfs_release_path(path
);
3510 spin_lock(&block_group
->lock
);
3511 if (!ret
&& dcs
== BTRFS_DC_SETUP
)
3512 block_group
->cache_generation
= trans
->transid
;
3513 block_group
->disk_cache_state
= dcs
;
3514 spin_unlock(&block_group
->lock
);
3519 int btrfs_setup_space_cache(struct btrfs_trans_handle
*trans
,
3520 struct btrfs_root
*root
)
3522 struct btrfs_block_group_cache
*cache
, *tmp
;
3523 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3524 struct btrfs_path
*path
;
3526 if (list_empty(&cur_trans
->dirty_bgs
) ||
3527 !btrfs_test_opt(root
, SPACE_CACHE
))
3530 path
= btrfs_alloc_path();
3534 /* Could add new block groups, use _safe just in case */
3535 list_for_each_entry_safe(cache
, tmp
, &cur_trans
->dirty_bgs
,
3537 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
)
3538 cache_save_setup(cache
, trans
, path
);
3541 btrfs_free_path(path
);
3546 * transaction commit does final block group cache writeback during a
3547 * critical section where nothing is allowed to change the FS. This is
3548 * required in order for the cache to actually match the block group,
3549 * but can introduce a lot of latency into the commit.
3551 * So, btrfs_start_dirty_block_groups is here to kick off block group
3552 * cache IO. There's a chance we'll have to redo some of it if the
3553 * block group changes again during the commit, but it greatly reduces
3554 * the commit latency by getting rid of the easy block groups while
3555 * we're still allowing others to join the commit.
3557 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle
*trans
,
3558 struct btrfs_root
*root
)
3560 struct btrfs_block_group_cache
*cache
;
3561 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3564 struct btrfs_path
*path
= NULL
;
3566 struct list_head
*io
= &cur_trans
->io_bgs
;
3567 int num_started
= 0;
3570 spin_lock(&cur_trans
->dirty_bgs_lock
);
3571 if (list_empty(&cur_trans
->dirty_bgs
)) {
3572 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3575 list_splice_init(&cur_trans
->dirty_bgs
, &dirty
);
3576 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3580 * make sure all the block groups on our dirty list actually
3583 btrfs_create_pending_block_groups(trans
, root
);
3586 path
= btrfs_alloc_path();
3592 * cache_write_mutex is here only to save us from balance or automatic
3593 * removal of empty block groups deleting this block group while we are
3594 * writing out the cache
3596 mutex_lock(&trans
->transaction
->cache_write_mutex
);
3597 while (!list_empty(&dirty
)) {
3598 cache
= list_first_entry(&dirty
,
3599 struct btrfs_block_group_cache
,
3602 * this can happen if something re-dirties a block
3603 * group that is already under IO. Just wait for it to
3604 * finish and then do it all again
3606 if (!list_empty(&cache
->io_list
)) {
3607 list_del_init(&cache
->io_list
);
3608 btrfs_wait_cache_io(root
, trans
, cache
,
3609 &cache
->io_ctl
, path
,
3610 cache
->key
.objectid
);
3611 btrfs_put_block_group(cache
);
3616 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3617 * if it should update the cache_state. Don't delete
3618 * until after we wait.
3620 * Since we're not running in the commit critical section
3621 * we need the dirty_bgs_lock to protect from update_block_group
3623 spin_lock(&cur_trans
->dirty_bgs_lock
);
3624 list_del_init(&cache
->dirty_list
);
3625 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3629 cache_save_setup(cache
, trans
, path
);
3631 if (cache
->disk_cache_state
== BTRFS_DC_SETUP
) {
3632 cache
->io_ctl
.inode
= NULL
;
3633 ret
= btrfs_write_out_cache(root
, trans
, cache
, path
);
3634 if (ret
== 0 && cache
->io_ctl
.inode
) {
3639 * the cache_write_mutex is protecting
3642 list_add_tail(&cache
->io_list
, io
);
3645 * if we failed to write the cache, the
3646 * generation will be bad and life goes on
3652 ret
= write_one_cache_group(trans
, root
, path
, cache
);
3654 * Our block group might still be attached to the list
3655 * of new block groups in the transaction handle of some
3656 * other task (struct btrfs_trans_handle->new_bgs). This
3657 * means its block group item isn't yet in the extent
3658 * tree. If this happens ignore the error, as we will
3659 * try again later in the critical section of the
3660 * transaction commit.
3662 if (ret
== -ENOENT
) {
3664 spin_lock(&cur_trans
->dirty_bgs_lock
);
3665 if (list_empty(&cache
->dirty_list
)) {
3666 list_add_tail(&cache
->dirty_list
,
3667 &cur_trans
->dirty_bgs
);
3668 btrfs_get_block_group(cache
);
3670 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3672 btrfs_abort_transaction(trans
, root
, ret
);
3676 /* if its not on the io list, we need to put the block group */
3678 btrfs_put_block_group(cache
);
3684 * Avoid blocking other tasks for too long. It might even save
3685 * us from writing caches for block groups that are going to be
3688 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
3689 mutex_lock(&trans
->transaction
->cache_write_mutex
);
3691 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
3694 * go through delayed refs for all the stuff we've just kicked off
3695 * and then loop back (just once)
3697 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
3698 if (!ret
&& loops
== 0) {
3700 spin_lock(&cur_trans
->dirty_bgs_lock
);
3701 list_splice_init(&cur_trans
->dirty_bgs
, &dirty
);
3703 * dirty_bgs_lock protects us from concurrent block group
3704 * deletes too (not just cache_write_mutex).
3706 if (!list_empty(&dirty
)) {
3707 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3710 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3713 btrfs_free_path(path
);
3717 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
3718 struct btrfs_root
*root
)
3720 struct btrfs_block_group_cache
*cache
;
3721 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3724 struct btrfs_path
*path
;
3725 struct list_head
*io
= &cur_trans
->io_bgs
;
3726 int num_started
= 0;
3728 path
= btrfs_alloc_path();
3733 * Even though we are in the critical section of the transaction commit,
3734 * we can still have concurrent tasks adding elements to this
3735 * transaction's list of dirty block groups. These tasks correspond to
3736 * endio free space workers started when writeback finishes for a
3737 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3738 * allocate new block groups as a result of COWing nodes of the root
3739 * tree when updating the free space inode. The writeback for the space
3740 * caches is triggered by an earlier call to
3741 * btrfs_start_dirty_block_groups() and iterations of the following
3743 * Also we want to do the cache_save_setup first and then run the
3744 * delayed refs to make sure we have the best chance at doing this all
3747 spin_lock(&cur_trans
->dirty_bgs_lock
);
3748 while (!list_empty(&cur_trans
->dirty_bgs
)) {
3749 cache
= list_first_entry(&cur_trans
->dirty_bgs
,
3750 struct btrfs_block_group_cache
,
3754 * this can happen if cache_save_setup re-dirties a block
3755 * group that is already under IO. Just wait for it to
3756 * finish and then do it all again
3758 if (!list_empty(&cache
->io_list
)) {
3759 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3760 list_del_init(&cache
->io_list
);
3761 btrfs_wait_cache_io(root
, trans
, cache
,
3762 &cache
->io_ctl
, path
,
3763 cache
->key
.objectid
);
3764 btrfs_put_block_group(cache
);
3765 spin_lock(&cur_trans
->dirty_bgs_lock
);
3769 * don't remove from the dirty list until after we've waited
3772 list_del_init(&cache
->dirty_list
);
3773 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3776 cache_save_setup(cache
, trans
, path
);
3779 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long) -1);
3781 if (!ret
&& cache
->disk_cache_state
== BTRFS_DC_SETUP
) {
3782 cache
->io_ctl
.inode
= NULL
;
3783 ret
= btrfs_write_out_cache(root
, trans
, cache
, path
);
3784 if (ret
== 0 && cache
->io_ctl
.inode
) {
3787 list_add_tail(&cache
->io_list
, io
);
3790 * if we failed to write the cache, the
3791 * generation will be bad and life goes on
3797 ret
= write_one_cache_group(trans
, root
, path
, cache
);
3799 * One of the free space endio workers might have
3800 * created a new block group while updating a free space
3801 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3802 * and hasn't released its transaction handle yet, in
3803 * which case the new block group is still attached to
3804 * its transaction handle and its creation has not
3805 * finished yet (no block group item in the extent tree
3806 * yet, etc). If this is the case, wait for all free
3807 * space endio workers to finish and retry. This is a
3808 * a very rare case so no need for a more efficient and
3811 if (ret
== -ENOENT
) {
3812 wait_event(cur_trans
->writer_wait
,
3813 atomic_read(&cur_trans
->num_writers
) == 1);
3814 ret
= write_one_cache_group(trans
, root
, path
,
3818 btrfs_abort_transaction(trans
, root
, ret
);
3821 /* if its not on the io list, we need to put the block group */
3823 btrfs_put_block_group(cache
);
3824 spin_lock(&cur_trans
->dirty_bgs_lock
);
3826 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3828 while (!list_empty(io
)) {
3829 cache
= list_first_entry(io
, struct btrfs_block_group_cache
,
3831 list_del_init(&cache
->io_list
);
3832 btrfs_wait_cache_io(root
, trans
, cache
,
3833 &cache
->io_ctl
, path
, cache
->key
.objectid
);
3834 btrfs_put_block_group(cache
);
3837 btrfs_free_path(path
);
3841 int btrfs_extent_readonly(struct btrfs_root
*root
, u64 bytenr
)
3843 struct btrfs_block_group_cache
*block_group
;
3846 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
3847 if (!block_group
|| block_group
->ro
)
3850 btrfs_put_block_group(block_group
);
3854 bool btrfs_inc_nocow_writers(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
3856 struct btrfs_block_group_cache
*bg
;
3859 bg
= btrfs_lookup_block_group(fs_info
, bytenr
);
3863 spin_lock(&bg
->lock
);
3867 atomic_inc(&bg
->nocow_writers
);
3868 spin_unlock(&bg
->lock
);
3870 /* no put on block group, done by btrfs_dec_nocow_writers */
3872 btrfs_put_block_group(bg
);
3878 void btrfs_dec_nocow_writers(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
3880 struct btrfs_block_group_cache
*bg
;
3882 bg
= btrfs_lookup_block_group(fs_info
, bytenr
);
3884 if (atomic_dec_and_test(&bg
->nocow_writers
))
3885 wake_up_atomic_t(&bg
->nocow_writers
);
3887 * Once for our lookup and once for the lookup done by a previous call
3888 * to btrfs_inc_nocow_writers()
3890 btrfs_put_block_group(bg
);
3891 btrfs_put_block_group(bg
);
3894 static int btrfs_wait_nocow_writers_atomic_t(atomic_t
*a
)
3900 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache
*bg
)
3902 wait_on_atomic_t(&bg
->nocow_writers
,
3903 btrfs_wait_nocow_writers_atomic_t
,
3904 TASK_UNINTERRUPTIBLE
);
3907 static const char *alloc_name(u64 flags
)
3910 case BTRFS_BLOCK_GROUP_METADATA
|BTRFS_BLOCK_GROUP_DATA
:
3912 case BTRFS_BLOCK_GROUP_METADATA
:
3914 case BTRFS_BLOCK_GROUP_DATA
:
3916 case BTRFS_BLOCK_GROUP_SYSTEM
:
3920 return "invalid-combination";
3924 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
3925 u64 total_bytes
, u64 bytes_used
,
3927 struct btrfs_space_info
**space_info
)
3929 struct btrfs_space_info
*found
;
3934 if (flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3935 BTRFS_BLOCK_GROUP_RAID10
))
3940 found
= __find_space_info(info
, flags
);
3942 spin_lock(&found
->lock
);
3943 found
->total_bytes
+= total_bytes
;
3944 found
->disk_total
+= total_bytes
* factor
;
3945 found
->bytes_used
+= bytes_used
;
3946 found
->disk_used
+= bytes_used
* factor
;
3947 found
->bytes_readonly
+= bytes_readonly
;
3948 if (total_bytes
> 0)
3950 space_info_add_new_bytes(info
, found
, total_bytes
-
3951 bytes_used
- bytes_readonly
);
3952 spin_unlock(&found
->lock
);
3953 *space_info
= found
;
3956 found
= kzalloc(sizeof(*found
), GFP_NOFS
);
3960 ret
= percpu_counter_init(&found
->total_bytes_pinned
, 0, GFP_KERNEL
);
3966 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
3967 INIT_LIST_HEAD(&found
->block_groups
[i
]);
3968 init_rwsem(&found
->groups_sem
);
3969 spin_lock_init(&found
->lock
);
3970 found
->flags
= flags
& BTRFS_BLOCK_GROUP_TYPE_MASK
;
3971 found
->total_bytes
= total_bytes
;
3972 found
->disk_total
= total_bytes
* factor
;
3973 found
->bytes_used
= bytes_used
;
3974 found
->disk_used
= bytes_used
* factor
;
3975 found
->bytes_pinned
= 0;
3976 found
->bytes_reserved
= 0;
3977 found
->bytes_readonly
= bytes_readonly
;
3978 found
->bytes_may_use
= 0;
3980 found
->max_extent_size
= 0;
3981 found
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3982 found
->chunk_alloc
= 0;
3984 init_waitqueue_head(&found
->wait
);
3985 INIT_LIST_HEAD(&found
->ro_bgs
);
3986 INIT_LIST_HEAD(&found
->tickets
);
3987 INIT_LIST_HEAD(&found
->priority_tickets
);
3989 ret
= kobject_init_and_add(&found
->kobj
, &space_info_ktype
,
3990 info
->space_info_kobj
, "%s",
3991 alloc_name(found
->flags
));
3997 *space_info
= found
;
3998 list_add_rcu(&found
->list
, &info
->space_info
);
3999 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
4000 info
->data_sinfo
= found
;
4005 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
4007 u64 extra_flags
= chunk_to_extended(flags
) &
4008 BTRFS_EXTENDED_PROFILE_MASK
;
4010 write_seqlock(&fs_info
->profiles_lock
);
4011 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
4012 fs_info
->avail_data_alloc_bits
|= extra_flags
;
4013 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
4014 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
4015 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
4016 fs_info
->avail_system_alloc_bits
|= extra_flags
;
4017 write_sequnlock(&fs_info
->profiles_lock
);
4021 * returns target flags in extended format or 0 if restripe for this
4022 * chunk_type is not in progress
4024 * should be called with either volume_mutex or balance_lock held
4026 static u64
get_restripe_target(struct btrfs_fs_info
*fs_info
, u64 flags
)
4028 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
4034 if (flags
& BTRFS_BLOCK_GROUP_DATA
&&
4035 bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
4036 target
= BTRFS_BLOCK_GROUP_DATA
| bctl
->data
.target
;
4037 } else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
&&
4038 bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
4039 target
= BTRFS_BLOCK_GROUP_SYSTEM
| bctl
->sys
.target
;
4040 } else if (flags
& BTRFS_BLOCK_GROUP_METADATA
&&
4041 bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
4042 target
= BTRFS_BLOCK_GROUP_METADATA
| bctl
->meta
.target
;
4049 * @flags: available profiles in extended format (see ctree.h)
4051 * Returns reduced profile in chunk format. If profile changing is in
4052 * progress (either running or paused) picks the target profile (if it's
4053 * already available), otherwise falls back to plain reducing.
4055 static u64
btrfs_reduce_alloc_profile(struct btrfs_root
*root
, u64 flags
)
4057 u64 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
4063 * see if restripe for this chunk_type is in progress, if so
4064 * try to reduce to the target profile
4066 spin_lock(&root
->fs_info
->balance_lock
);
4067 target
= get_restripe_target(root
->fs_info
, flags
);
4069 /* pick target profile only if it's already available */
4070 if ((flags
& target
) & BTRFS_EXTENDED_PROFILE_MASK
) {
4071 spin_unlock(&root
->fs_info
->balance_lock
);
4072 return extended_to_chunk(target
);
4075 spin_unlock(&root
->fs_info
->balance_lock
);
4077 /* First, mask out the RAID levels which aren't possible */
4078 for (raid_type
= 0; raid_type
< BTRFS_NR_RAID_TYPES
; raid_type
++) {
4079 if (num_devices
>= btrfs_raid_array
[raid_type
].devs_min
)
4080 allowed
|= btrfs_raid_group
[raid_type
];
4084 if (allowed
& BTRFS_BLOCK_GROUP_RAID6
)
4085 allowed
= BTRFS_BLOCK_GROUP_RAID6
;
4086 else if (allowed
& BTRFS_BLOCK_GROUP_RAID5
)
4087 allowed
= BTRFS_BLOCK_GROUP_RAID5
;
4088 else if (allowed
& BTRFS_BLOCK_GROUP_RAID10
)
4089 allowed
= BTRFS_BLOCK_GROUP_RAID10
;
4090 else if (allowed
& BTRFS_BLOCK_GROUP_RAID1
)
4091 allowed
= BTRFS_BLOCK_GROUP_RAID1
;
4092 else if (allowed
& BTRFS_BLOCK_GROUP_RAID0
)
4093 allowed
= BTRFS_BLOCK_GROUP_RAID0
;
4095 flags
&= ~BTRFS_BLOCK_GROUP_PROFILE_MASK
;
4097 return extended_to_chunk(flags
| allowed
);
4100 static u64
get_alloc_profile(struct btrfs_root
*root
, u64 orig_flags
)
4107 seq
= read_seqbegin(&root
->fs_info
->profiles_lock
);
4109 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
4110 flags
|= root
->fs_info
->avail_data_alloc_bits
;
4111 else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
4112 flags
|= root
->fs_info
->avail_system_alloc_bits
;
4113 else if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
4114 flags
|= root
->fs_info
->avail_metadata_alloc_bits
;
4115 } while (read_seqretry(&root
->fs_info
->profiles_lock
, seq
));
4117 return btrfs_reduce_alloc_profile(root
, flags
);
4120 u64
btrfs_get_alloc_profile(struct btrfs_root
*root
, int data
)
4126 flags
= BTRFS_BLOCK_GROUP_DATA
;
4127 else if (root
== root
->fs_info
->chunk_root
)
4128 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
4130 flags
= BTRFS_BLOCK_GROUP_METADATA
;
4132 ret
= get_alloc_profile(root
, flags
);
4136 int btrfs_alloc_data_chunk_ondemand(struct inode
*inode
, u64 bytes
)
4138 struct btrfs_space_info
*data_sinfo
;
4139 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4140 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4143 int need_commit
= 2;
4144 int have_pinned_space
;
4146 /* make sure bytes are sectorsize aligned */
4147 bytes
= ALIGN(bytes
, root
->sectorsize
);
4149 if (btrfs_is_free_space_inode(inode
)) {
4151 ASSERT(current
->journal_info
);
4154 data_sinfo
= fs_info
->data_sinfo
;
4159 /* make sure we have enough space to handle the data first */
4160 spin_lock(&data_sinfo
->lock
);
4161 used
= data_sinfo
->bytes_used
+ data_sinfo
->bytes_reserved
+
4162 data_sinfo
->bytes_pinned
+ data_sinfo
->bytes_readonly
+
4163 data_sinfo
->bytes_may_use
;
4165 if (used
+ bytes
> data_sinfo
->total_bytes
) {
4166 struct btrfs_trans_handle
*trans
;
4169 * if we don't have enough free bytes in this space then we need
4170 * to alloc a new chunk.
4172 if (!data_sinfo
->full
) {
4175 data_sinfo
->force_alloc
= CHUNK_ALLOC_FORCE
;
4176 spin_unlock(&data_sinfo
->lock
);
4178 alloc_target
= btrfs_get_alloc_profile(root
, 1);
4180 * It is ugly that we don't call nolock join
4181 * transaction for the free space inode case here.
4182 * But it is safe because we only do the data space
4183 * reservation for the free space cache in the
4184 * transaction context, the common join transaction
4185 * just increase the counter of the current transaction
4186 * handler, doesn't try to acquire the trans_lock of
4189 trans
= btrfs_join_transaction(root
);
4191 return PTR_ERR(trans
);
4193 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
4195 CHUNK_ALLOC_NO_FORCE
);
4196 btrfs_end_transaction(trans
, root
);
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(&root
->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, (u64
)-1);
4233 trans
= btrfs_join_transaction(root
);
4235 return PTR_ERR(trans
);
4236 if (have_pinned_space
>= 0 ||
4237 test_bit(BTRFS_TRANS_HAVE_FREE_BGS
,
4238 &trans
->transaction
->flags
) ||
4240 ret
= btrfs_commit_transaction(trans
, root
);
4244 * The cleaner kthread might still be doing iput
4245 * operations. Wait for it to finish so that
4246 * more space is released.
4248 mutex_lock(&root
->fs_info
->cleaner_delayed_iput_mutex
);
4249 mutex_unlock(&root
->fs_info
->cleaner_delayed_iput_mutex
);
4252 btrfs_end_transaction(trans
, root
);
4256 trace_btrfs_space_reservation(root
->fs_info
,
4257 "space_info:enospc",
4258 data_sinfo
->flags
, bytes
, 1);
4261 data_sinfo
->bytes_may_use
+= bytes
;
4262 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
4263 data_sinfo
->flags
, bytes
, 1);
4264 spin_unlock(&data_sinfo
->lock
);
4270 * New check_data_free_space() with ability for precious data reservation
4271 * Will replace old btrfs_check_data_free_space(), but for patch split,
4272 * add a new function first and then replace it.
4274 int btrfs_check_data_free_space(struct inode
*inode
, u64 start
, u64 len
)
4276 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4279 /* align the range */
4280 len
= round_up(start
+ len
, root
->sectorsize
) -
4281 round_down(start
, root
->sectorsize
);
4282 start
= round_down(start
, root
->sectorsize
);
4284 ret
= btrfs_alloc_data_chunk_ondemand(inode
, len
);
4289 * Use new btrfs_qgroup_reserve_data to reserve precious data space
4291 * TODO: Find a good method to avoid reserve data space for NOCOW
4292 * range, but don't impact performance on quota disable case.
4294 ret
= btrfs_qgroup_reserve_data(inode
, start
, len
);
4299 * Called if we need to clear a data reservation for this inode
4300 * Normally in a error case.
4302 * This one will *NOT* use accurate qgroup reserved space API, just for case
4303 * which we can't sleep and is sure it won't affect qgroup reserved space.
4304 * Like clear_bit_hook().
4306 void btrfs_free_reserved_data_space_noquota(struct inode
*inode
, u64 start
,
4309 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4310 struct btrfs_space_info
*data_sinfo
;
4312 /* Make sure the range is aligned to sectorsize */
4313 len
= round_up(start
+ len
, root
->sectorsize
) -
4314 round_down(start
, root
->sectorsize
);
4315 start
= round_down(start
, root
->sectorsize
);
4317 data_sinfo
= root
->fs_info
->data_sinfo
;
4318 spin_lock(&data_sinfo
->lock
);
4319 if (WARN_ON(data_sinfo
->bytes_may_use
< len
))
4320 data_sinfo
->bytes_may_use
= 0;
4322 data_sinfo
->bytes_may_use
-= len
;
4323 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
4324 data_sinfo
->flags
, len
, 0);
4325 spin_unlock(&data_sinfo
->lock
);
4329 * Called if we need to clear a data reservation for this inode
4330 * Normally in a error case.
4332 * This one will handle the per-inode data rsv map for accurate reserved
4335 void btrfs_free_reserved_data_space(struct inode
*inode
, u64 start
, u64 len
)
4337 btrfs_free_reserved_data_space_noquota(inode
, start
, len
);
4338 btrfs_qgroup_free_data(inode
, start
, len
);
4341 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
4343 struct list_head
*head
= &info
->space_info
;
4344 struct btrfs_space_info
*found
;
4347 list_for_each_entry_rcu(found
, head
, list
) {
4348 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
4349 found
->force_alloc
= CHUNK_ALLOC_FORCE
;
4354 static inline u64
calc_global_rsv_need_space(struct btrfs_block_rsv
*global
)
4356 return (global
->size
<< 1);
4359 static int should_alloc_chunk(struct btrfs_root
*root
,
4360 struct btrfs_space_info
*sinfo
, int force
)
4362 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
4363 u64 num_bytes
= sinfo
->total_bytes
- sinfo
->bytes_readonly
;
4364 u64 num_allocated
= sinfo
->bytes_used
+ sinfo
->bytes_reserved
;
4367 if (force
== CHUNK_ALLOC_FORCE
)
4371 * We need to take into account the global rsv because for all intents
4372 * and purposes it's used space. Don't worry about locking the
4373 * global_rsv, it doesn't change except when the transaction commits.
4375 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
4376 num_allocated
+= calc_global_rsv_need_space(global_rsv
);
4379 * in limited mode, we want to have some free space up to
4380 * about 1% of the FS size.
4382 if (force
== CHUNK_ALLOC_LIMITED
) {
4383 thresh
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
4384 thresh
= max_t(u64
, SZ_64M
, div_factor_fine(thresh
, 1));
4386 if (num_bytes
- num_allocated
< thresh
)
4390 if (num_allocated
+ SZ_2M
< div_factor(num_bytes
, 8))
4395 static u64
get_profile_num_devs(struct btrfs_root
*root
, u64 type
)
4399 if (type
& (BTRFS_BLOCK_GROUP_RAID10
|
4400 BTRFS_BLOCK_GROUP_RAID0
|
4401 BTRFS_BLOCK_GROUP_RAID5
|
4402 BTRFS_BLOCK_GROUP_RAID6
))
4403 num_dev
= root
->fs_info
->fs_devices
->rw_devices
;
4404 else if (type
& BTRFS_BLOCK_GROUP_RAID1
)
4407 num_dev
= 1; /* DUP or single */
4413 * If @is_allocation is true, reserve space in the system space info necessary
4414 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4417 void check_system_chunk(struct btrfs_trans_handle
*trans
,
4418 struct btrfs_root
*root
,
4421 struct btrfs_space_info
*info
;
4428 * Needed because we can end up allocating a system chunk and for an
4429 * atomic and race free space reservation in the chunk block reserve.
4431 ASSERT(mutex_is_locked(&root
->fs_info
->chunk_mutex
));
4433 info
= __find_space_info(root
->fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
4434 spin_lock(&info
->lock
);
4435 left
= info
->total_bytes
- info
->bytes_used
- info
->bytes_pinned
-
4436 info
->bytes_reserved
- info
->bytes_readonly
-
4437 info
->bytes_may_use
;
4438 spin_unlock(&info
->lock
);
4440 num_devs
= get_profile_num_devs(root
, type
);
4442 /* num_devs device items to update and 1 chunk item to add or remove */
4443 thresh
= btrfs_calc_trunc_metadata_size(root
, num_devs
) +
4444 btrfs_calc_trans_metadata_size(root
, 1);
4446 if (left
< thresh
&& btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
4447 btrfs_info(root
->fs_info
, "left=%llu, need=%llu, flags=%llu",
4448 left
, thresh
, type
);
4449 dump_space_info(info
, 0, 0);
4452 if (left
< thresh
) {
4455 flags
= btrfs_get_alloc_profile(root
->fs_info
->chunk_root
, 0);
4457 * Ignore failure to create system chunk. We might end up not
4458 * needing it, as we might not need to COW all nodes/leafs from
4459 * the paths we visit in the chunk tree (they were already COWed
4460 * or created in the current transaction for example).
4462 ret
= btrfs_alloc_chunk(trans
, root
, flags
);
4466 ret
= btrfs_block_rsv_add(root
->fs_info
->chunk_root
,
4467 &root
->fs_info
->chunk_block_rsv
,
4468 thresh
, BTRFS_RESERVE_NO_FLUSH
);
4470 trans
->chunk_bytes_reserved
+= thresh
;
4474 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
4475 struct btrfs_root
*extent_root
, u64 flags
, int force
)
4477 struct btrfs_space_info
*space_info
;
4478 struct btrfs_fs_info
*fs_info
= extent_root
->fs_info
;
4479 int wait_for_alloc
= 0;
4482 /* Don't re-enter if we're already allocating a chunk */
4483 if (trans
->allocating_chunk
)
4486 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
4488 ret
= update_space_info(extent_root
->fs_info
, flags
,
4489 0, 0, 0, &space_info
);
4490 BUG_ON(ret
); /* -ENOMEM */
4492 BUG_ON(!space_info
); /* Logic error */
4495 spin_lock(&space_info
->lock
);
4496 if (force
< space_info
->force_alloc
)
4497 force
= space_info
->force_alloc
;
4498 if (space_info
->full
) {
4499 if (should_alloc_chunk(extent_root
, space_info
, force
))
4503 spin_unlock(&space_info
->lock
);
4507 if (!should_alloc_chunk(extent_root
, space_info
, force
)) {
4508 spin_unlock(&space_info
->lock
);
4510 } else if (space_info
->chunk_alloc
) {
4513 space_info
->chunk_alloc
= 1;
4516 spin_unlock(&space_info
->lock
);
4518 mutex_lock(&fs_info
->chunk_mutex
);
4521 * The chunk_mutex is held throughout the entirety of a chunk
4522 * allocation, so once we've acquired the chunk_mutex we know that the
4523 * other guy is done and we need to recheck and see if we should
4526 if (wait_for_alloc
) {
4527 mutex_unlock(&fs_info
->chunk_mutex
);
4532 trans
->allocating_chunk
= true;
4535 * If we have mixed data/metadata chunks we want to make sure we keep
4536 * allocating mixed chunks instead of individual chunks.
4538 if (btrfs_mixed_space_info(space_info
))
4539 flags
|= (BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
);
4542 * if we're doing a data chunk, go ahead and make sure that
4543 * we keep a reasonable number of metadata chunks allocated in the
4546 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
4547 fs_info
->data_chunk_allocations
++;
4548 if (!(fs_info
->data_chunk_allocations
%
4549 fs_info
->metadata_ratio
))
4550 force_metadata_allocation(fs_info
);
4554 * Check if we have enough space in SYSTEM chunk because we may need
4555 * to update devices.
4557 check_system_chunk(trans
, extent_root
, flags
);
4559 ret
= btrfs_alloc_chunk(trans
, extent_root
, flags
);
4560 trans
->allocating_chunk
= false;
4562 spin_lock(&space_info
->lock
);
4563 if (ret
< 0 && ret
!= -ENOSPC
)
4566 space_info
->full
= 1;
4570 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
4572 space_info
->chunk_alloc
= 0;
4573 spin_unlock(&space_info
->lock
);
4574 mutex_unlock(&fs_info
->chunk_mutex
);
4576 * When we allocate a new chunk we reserve space in the chunk block
4577 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4578 * add new nodes/leafs to it if we end up needing to do it when
4579 * inserting the chunk item and updating device items as part of the
4580 * second phase of chunk allocation, performed by
4581 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4582 * large number of new block groups to create in our transaction
4583 * handle's new_bgs list to avoid exhausting the chunk block reserve
4584 * in extreme cases - like having a single transaction create many new
4585 * block groups when starting to write out the free space caches of all
4586 * the block groups that were made dirty during the lifetime of the
4589 if (trans
->can_flush_pending_bgs
&&
4590 trans
->chunk_bytes_reserved
>= (u64
)SZ_2M
) {
4591 btrfs_create_pending_block_groups(trans
, trans
->root
);
4592 btrfs_trans_release_chunk_metadata(trans
);
4597 static int can_overcommit(struct btrfs_root
*root
,
4598 struct btrfs_space_info
*space_info
, u64 bytes
,
4599 enum btrfs_reserve_flush_enum flush
)
4601 struct btrfs_block_rsv
*global_rsv
;
4607 /* Don't overcommit when in mixed mode. */
4608 if (space_info
->flags
& BTRFS_BLOCK_GROUP_DATA
)
4611 BUG_ON(root
->fs_info
== NULL
);
4612 global_rsv
= &root
->fs_info
->global_block_rsv
;
4613 profile
= btrfs_get_alloc_profile(root
, 0);
4614 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4615 space_info
->bytes_pinned
+ space_info
->bytes_readonly
;
4618 * We only want to allow over committing if we have lots of actual space
4619 * free, but if we don't have enough space to handle the global reserve
4620 * space then we could end up having a real enospc problem when trying
4621 * to allocate a chunk or some other such important allocation.
4623 spin_lock(&global_rsv
->lock
);
4624 space_size
= calc_global_rsv_need_space(global_rsv
);
4625 spin_unlock(&global_rsv
->lock
);
4626 if (used
+ space_size
>= space_info
->total_bytes
)
4629 used
+= space_info
->bytes_may_use
;
4631 spin_lock(&root
->fs_info
->free_chunk_lock
);
4632 avail
= root
->fs_info
->free_chunk_space
;
4633 spin_unlock(&root
->fs_info
->free_chunk_lock
);
4636 * If we have dup, raid1 or raid10 then only half of the free
4637 * space is actually useable. For raid56, the space info used
4638 * doesn't include the parity drive, so we don't have to
4641 if (profile
& (BTRFS_BLOCK_GROUP_DUP
|
4642 BTRFS_BLOCK_GROUP_RAID1
|
4643 BTRFS_BLOCK_GROUP_RAID10
))
4647 * If we aren't flushing all things, let us overcommit up to
4648 * 1/2th of the space. If we can flush, don't let us overcommit
4649 * too much, let it overcommit up to 1/8 of the space.
4651 if (flush
== BTRFS_RESERVE_FLUSH_ALL
)
4656 if (used
+ bytes
< space_info
->total_bytes
+ avail
)
4661 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root
*root
,
4662 unsigned long nr_pages
, int nr_items
)
4664 struct super_block
*sb
= root
->fs_info
->sb
;
4666 if (down_read_trylock(&sb
->s_umount
)) {
4667 writeback_inodes_sb_nr(sb
, nr_pages
, WB_REASON_FS_FREE_SPACE
);
4668 up_read(&sb
->s_umount
);
4671 * We needn't worry the filesystem going from r/w to r/o though
4672 * we don't acquire ->s_umount mutex, because the filesystem
4673 * should guarantee the delalloc inodes list be empty after
4674 * the filesystem is readonly(all dirty pages are written to
4677 btrfs_start_delalloc_roots(root
->fs_info
, 0, nr_items
);
4678 if (!current
->journal_info
)
4679 btrfs_wait_ordered_roots(root
->fs_info
, nr_items
,
4684 static inline int calc_reclaim_items_nr(struct btrfs_root
*root
, u64 to_reclaim
)
4689 bytes
= btrfs_calc_trans_metadata_size(root
, 1);
4690 nr
= (int)div64_u64(to_reclaim
, bytes
);
4696 #define EXTENT_SIZE_PER_ITEM SZ_256K
4699 * shrink metadata reservation for delalloc
4701 static void shrink_delalloc(struct btrfs_root
*root
, u64 to_reclaim
, u64 orig
,
4704 struct btrfs_block_rsv
*block_rsv
;
4705 struct btrfs_space_info
*space_info
;
4706 struct btrfs_trans_handle
*trans
;
4710 unsigned long nr_pages
;
4713 enum btrfs_reserve_flush_enum flush
;
4715 /* Calc the number of the pages we need flush for space reservation */
4716 items
= calc_reclaim_items_nr(root
, to_reclaim
);
4717 to_reclaim
= (u64
)items
* EXTENT_SIZE_PER_ITEM
;
4719 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4720 block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
4721 space_info
= block_rsv
->space_info
;
4723 delalloc_bytes
= percpu_counter_sum_positive(
4724 &root
->fs_info
->delalloc_bytes
);
4725 if (delalloc_bytes
== 0) {
4729 btrfs_wait_ordered_roots(root
->fs_info
, items
,
4735 while (delalloc_bytes
&& loops
< 3) {
4736 max_reclaim
= min(delalloc_bytes
, to_reclaim
);
4737 nr_pages
= max_reclaim
>> PAGE_SHIFT
;
4738 btrfs_writeback_inodes_sb_nr(root
, nr_pages
, items
);
4740 * We need to wait for the async pages to actually start before
4743 max_reclaim
= atomic_read(&root
->fs_info
->async_delalloc_pages
);
4747 if (max_reclaim
<= nr_pages
)
4750 max_reclaim
-= nr_pages
;
4752 wait_event(root
->fs_info
->async_submit_wait
,
4753 atomic_read(&root
->fs_info
->async_delalloc_pages
) <=
4757 flush
= BTRFS_RESERVE_FLUSH_ALL
;
4759 flush
= BTRFS_RESERVE_NO_FLUSH
;
4760 spin_lock(&space_info
->lock
);
4761 if (can_overcommit(root
, space_info
, orig
, flush
)) {
4762 spin_unlock(&space_info
->lock
);
4765 if (list_empty(&space_info
->tickets
) &&
4766 list_empty(&space_info
->priority_tickets
)) {
4767 spin_unlock(&space_info
->lock
);
4770 spin_unlock(&space_info
->lock
);
4773 if (wait_ordered
&& !trans
) {
4774 btrfs_wait_ordered_roots(root
->fs_info
, items
,
4777 time_left
= schedule_timeout_killable(1);
4781 delalloc_bytes
= percpu_counter_sum_positive(
4782 &root
->fs_info
->delalloc_bytes
);
4787 * maybe_commit_transaction - possibly commit the transaction if its ok to
4788 * @root - the root we're allocating for
4789 * @bytes - the number of bytes we want to reserve
4790 * @force - force the commit
4792 * This will check to make sure that committing the transaction will actually
4793 * get us somewhere and then commit the transaction if it does. Otherwise it
4794 * will return -ENOSPC.
4796 static int may_commit_transaction(struct btrfs_root
*root
,
4797 struct btrfs_space_info
*space_info
,
4798 u64 bytes
, int force
)
4800 struct btrfs_block_rsv
*delayed_rsv
= &root
->fs_info
->delayed_block_rsv
;
4801 struct btrfs_trans_handle
*trans
;
4803 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4810 /* See if there is enough pinned space to make this reservation */
4811 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4816 * See if there is some space in the delayed insertion reservation for
4819 if (space_info
!= delayed_rsv
->space_info
)
4822 spin_lock(&delayed_rsv
->lock
);
4823 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4824 bytes
- delayed_rsv
->size
) >= 0) {
4825 spin_unlock(&delayed_rsv
->lock
);
4828 spin_unlock(&delayed_rsv
->lock
);
4831 trans
= btrfs_join_transaction(root
);
4835 return btrfs_commit_transaction(trans
, root
);
4838 struct reserve_ticket
{
4841 struct list_head list
;
4842 wait_queue_head_t wait
;
4845 static int flush_space(struct btrfs_root
*root
,
4846 struct btrfs_space_info
*space_info
, u64 num_bytes
,
4847 u64 orig_bytes
, int state
)
4849 struct btrfs_trans_handle
*trans
;
4854 case FLUSH_DELAYED_ITEMS_NR
:
4855 case FLUSH_DELAYED_ITEMS
:
4856 if (state
== FLUSH_DELAYED_ITEMS_NR
)
4857 nr
= calc_reclaim_items_nr(root
, num_bytes
) * 2;
4861 trans
= btrfs_join_transaction(root
);
4862 if (IS_ERR(trans
)) {
4863 ret
= PTR_ERR(trans
);
4866 ret
= btrfs_run_delayed_items_nr(trans
, root
, nr
);
4867 btrfs_end_transaction(trans
, root
);
4869 case FLUSH_DELALLOC
:
4870 case FLUSH_DELALLOC_WAIT
:
4871 shrink_delalloc(root
, num_bytes
* 2, orig_bytes
,
4872 state
== FLUSH_DELALLOC_WAIT
);
4875 trans
= btrfs_join_transaction(root
);
4876 if (IS_ERR(trans
)) {
4877 ret
= PTR_ERR(trans
);
4880 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
4881 btrfs_get_alloc_profile(root
, 0),
4882 CHUNK_ALLOC_NO_FORCE
);
4883 btrfs_end_transaction(trans
, root
);
4888 ret
= may_commit_transaction(root
, space_info
, orig_bytes
, 0);
4895 trace_btrfs_flush_space(root
->fs_info
, space_info
->flags
, num_bytes
,
4896 orig_bytes
, state
, ret
);
4901 btrfs_calc_reclaim_metadata_size(struct btrfs_root
*root
,
4902 struct btrfs_space_info
*space_info
)
4904 struct reserve_ticket
*ticket
;
4909 to_reclaim
= min_t(u64
, num_online_cpus() * SZ_1M
, SZ_16M
);
4910 if (can_overcommit(root
, space_info
, to_reclaim
,
4911 BTRFS_RESERVE_FLUSH_ALL
))
4914 list_for_each_entry(ticket
, &space_info
->tickets
, list
)
4915 to_reclaim
+= ticket
->bytes
;
4916 list_for_each_entry(ticket
, &space_info
->priority_tickets
, list
)
4917 to_reclaim
+= ticket
->bytes
;
4921 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4922 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4923 space_info
->bytes_may_use
;
4924 if (can_overcommit(root
, space_info
, SZ_1M
, BTRFS_RESERVE_FLUSH_ALL
))
4925 expected
= div_factor_fine(space_info
->total_bytes
, 95);
4927 expected
= div_factor_fine(space_info
->total_bytes
, 90);
4929 if (used
> expected
)
4930 to_reclaim
= used
- expected
;
4933 to_reclaim
= min(to_reclaim
, space_info
->bytes_may_use
+
4934 space_info
->bytes_reserved
);
4938 static inline int need_do_async_reclaim(struct btrfs_space_info
*space_info
,
4939 struct btrfs_root
*root
, u64 used
)
4941 u64 thresh
= div_factor_fine(space_info
->total_bytes
, 98);
4943 /* If we're just plain full then async reclaim just slows us down. */
4944 if ((space_info
->bytes_used
+ space_info
->bytes_reserved
) >= thresh
)
4947 if (!btrfs_calc_reclaim_metadata_size(root
, space_info
))
4950 return (used
>= thresh
&& !btrfs_fs_closing(root
->fs_info
) &&
4951 !test_bit(BTRFS_FS_STATE_REMOUNTING
,
4952 &root
->fs_info
->fs_state
));
4955 static void wake_all_tickets(struct list_head
*head
)
4957 struct reserve_ticket
*ticket
;
4959 while (!list_empty(head
)) {
4960 ticket
= list_first_entry(head
, struct reserve_ticket
, list
);
4961 list_del_init(&ticket
->list
);
4962 ticket
->error
= -ENOSPC
;
4963 wake_up(&ticket
->wait
);
4968 * This is for normal flushers, we can wait all goddamned day if we want to. We
4969 * will loop and continuously try to flush as long as we are making progress.
4970 * We count progress as clearing off tickets each time we have to loop.
4972 static void btrfs_async_reclaim_metadata_space(struct work_struct
*work
)
4974 struct reserve_ticket
*last_ticket
= NULL
;
4975 struct btrfs_fs_info
*fs_info
;
4976 struct btrfs_space_info
*space_info
;
4979 int commit_cycles
= 0;
4981 fs_info
= container_of(work
, struct btrfs_fs_info
, async_reclaim_work
);
4982 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4984 spin_lock(&space_info
->lock
);
4985 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
->fs_root
,
4988 space_info
->flush
= 0;
4989 spin_unlock(&space_info
->lock
);
4992 last_ticket
= list_first_entry(&space_info
->tickets
,
4993 struct reserve_ticket
, list
);
4994 spin_unlock(&space_info
->lock
);
4996 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
4998 struct reserve_ticket
*ticket
;
5001 ret
= flush_space(fs_info
->fs_root
, space_info
, to_reclaim
,
5002 to_reclaim
, flush_state
);
5003 spin_lock(&space_info
->lock
);
5004 if (list_empty(&space_info
->tickets
)) {
5005 space_info
->flush
= 0;
5006 spin_unlock(&space_info
->lock
);
5009 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
->fs_root
,
5011 ticket
= list_first_entry(&space_info
->tickets
,
5012 struct reserve_ticket
, list
);
5013 if (last_ticket
== ticket
) {
5016 last_ticket
= ticket
;
5017 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
5022 if (flush_state
> COMMIT_TRANS
) {
5024 if (commit_cycles
> 2) {
5025 wake_all_tickets(&space_info
->tickets
);
5026 space_info
->flush
= 0;
5028 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
5031 spin_unlock(&space_info
->lock
);
5032 } while (flush_state
<= COMMIT_TRANS
);
5035 void btrfs_init_async_reclaim_work(struct work_struct
*work
)
5037 INIT_WORK(work
, btrfs_async_reclaim_metadata_space
);
5040 static void priority_reclaim_metadata_space(struct btrfs_fs_info
*fs_info
,
5041 struct btrfs_space_info
*space_info
,
5042 struct reserve_ticket
*ticket
)
5045 int flush_state
= FLUSH_DELAYED_ITEMS_NR
;
5047 spin_lock(&space_info
->lock
);
5048 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
->fs_root
,
5051 spin_unlock(&space_info
->lock
);
5054 spin_unlock(&space_info
->lock
);
5057 flush_space(fs_info
->fs_root
, space_info
, to_reclaim
,
5058 to_reclaim
, flush_state
);
5060 spin_lock(&space_info
->lock
);
5061 if (ticket
->bytes
== 0) {
5062 spin_unlock(&space_info
->lock
);
5065 spin_unlock(&space_info
->lock
);
5068 * Priority flushers can't wait on delalloc without
5071 if (flush_state
== FLUSH_DELALLOC
||
5072 flush_state
== FLUSH_DELALLOC_WAIT
)
5073 flush_state
= ALLOC_CHUNK
;
5074 } while (flush_state
< COMMIT_TRANS
);
5077 static int wait_reserve_ticket(struct btrfs_fs_info
*fs_info
,
5078 struct btrfs_space_info
*space_info
,
5079 struct reserve_ticket
*ticket
, u64 orig_bytes
)
5085 spin_lock(&space_info
->lock
);
5086 while (ticket
->bytes
> 0 && ticket
->error
== 0) {
5087 ret
= prepare_to_wait_event(&ticket
->wait
, &wait
, TASK_KILLABLE
);
5092 spin_unlock(&space_info
->lock
);
5096 finish_wait(&ticket
->wait
, &wait
);
5097 spin_lock(&space_info
->lock
);
5100 ret
= ticket
->error
;
5101 if (!list_empty(&ticket
->list
))
5102 list_del_init(&ticket
->list
);
5103 if (ticket
->bytes
&& ticket
->bytes
< orig_bytes
) {
5104 u64 num_bytes
= orig_bytes
- ticket
->bytes
;
5105 space_info
->bytes_may_use
-= num_bytes
;
5106 trace_btrfs_space_reservation(fs_info
, "space_info",
5107 space_info
->flags
, num_bytes
, 0);
5109 spin_unlock(&space_info
->lock
);
5115 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5116 * @root - the root we're allocating for
5117 * @space_info - the space info we want to allocate from
5118 * @orig_bytes - the number of bytes we want
5119 * @flush - whether or not we can flush to make our reservation
5121 * This will reserve orig_bytes number of bytes from the space info associated
5122 * with the block_rsv. If there is not enough space it will make an attempt to
5123 * flush out space to make room. It will do this by flushing delalloc if
5124 * possible or committing the transaction. If flush is 0 then no attempts to
5125 * regain reservations will be made and this will fail if there is not enough
5128 static int __reserve_metadata_bytes(struct btrfs_root
*root
,
5129 struct btrfs_space_info
*space_info
,
5131 enum btrfs_reserve_flush_enum flush
)
5133 struct reserve_ticket ticket
;
5138 ASSERT(!current
->journal_info
|| flush
!= BTRFS_RESERVE_FLUSH_ALL
);
5140 spin_lock(&space_info
->lock
);
5142 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
5143 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
5144 space_info
->bytes_may_use
;
5147 * If we have enough space then hooray, make our reservation and carry
5148 * on. If not see if we can overcommit, and if we can, hooray carry on.
5149 * If not things get more complicated.
5151 if (used
+ orig_bytes
<= space_info
->total_bytes
) {
5152 space_info
->bytes_may_use
+= orig_bytes
;
5153 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
5154 space_info
->flags
, orig_bytes
,
5157 } else if (can_overcommit(root
, space_info
, orig_bytes
, flush
)) {
5158 space_info
->bytes_may_use
+= orig_bytes
;
5159 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
5160 space_info
->flags
, orig_bytes
,
5166 * If we couldn't make a reservation then setup our reservation ticket
5167 * and kick the async worker if it's not already running.
5169 * If we are a priority flusher then we just need to add our ticket to
5170 * the list and we will do our own flushing further down.
5172 if (ret
&& flush
!= BTRFS_RESERVE_NO_FLUSH
) {
5173 ticket
.bytes
= orig_bytes
;
5175 init_waitqueue_head(&ticket
.wait
);
5176 if (flush
== BTRFS_RESERVE_FLUSH_ALL
) {
5177 list_add_tail(&ticket
.list
, &space_info
->tickets
);
5178 if (!space_info
->flush
) {
5179 space_info
->flush
= 1;
5180 trace_btrfs_trigger_flush(root
->fs_info
,
5184 queue_work(system_unbound_wq
,
5185 &root
->fs_info
->async_reclaim_work
);
5188 list_add_tail(&ticket
.list
,
5189 &space_info
->priority_tickets
);
5191 } else if (!ret
&& space_info
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
5194 * We will do the space reservation dance during log replay,
5195 * which means we won't have fs_info->fs_root set, so don't do
5196 * the async reclaim as we will panic.
5198 if (!root
->fs_info
->log_root_recovering
&&
5199 need_do_async_reclaim(space_info
, root
, used
) &&
5200 !work_busy(&root
->fs_info
->async_reclaim_work
)) {
5201 trace_btrfs_trigger_flush(root
->fs_info
,
5205 queue_work(system_unbound_wq
,
5206 &root
->fs_info
->async_reclaim_work
);
5209 spin_unlock(&space_info
->lock
);
5210 if (!ret
|| flush
== BTRFS_RESERVE_NO_FLUSH
)
5213 if (flush
== BTRFS_RESERVE_FLUSH_ALL
)
5214 return wait_reserve_ticket(root
->fs_info
, space_info
, &ticket
,
5218 priority_reclaim_metadata_space(root
->fs_info
, space_info
, &ticket
);
5219 spin_lock(&space_info
->lock
);
5221 if (ticket
.bytes
< orig_bytes
) {
5222 u64 num_bytes
= orig_bytes
- ticket
.bytes
;
5223 space_info
->bytes_may_use
-= num_bytes
;
5224 trace_btrfs_space_reservation(root
->fs_info
,
5225 "space_info", space_info
->flags
,
5229 list_del_init(&ticket
.list
);
5232 spin_unlock(&space_info
->lock
);
5233 ASSERT(list_empty(&ticket
.list
));
5238 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5239 * @root - the root we're allocating for
5240 * @block_rsv - the block_rsv we're allocating for
5241 * @orig_bytes - the number of bytes we want
5242 * @flush - whether or not we can flush to make our reservation
5244 * This will reserve orgi_bytes number of bytes from the space info associated
5245 * with the block_rsv. If there is not enough space it will make an attempt to
5246 * flush out space to make room. It will do this by flushing delalloc if
5247 * possible or committing the transaction. If flush is 0 then no attempts to
5248 * regain reservations will be made and this will fail if there is not enough
5251 static int reserve_metadata_bytes(struct btrfs_root
*root
,
5252 struct btrfs_block_rsv
*block_rsv
,
5254 enum btrfs_reserve_flush_enum flush
)
5258 ret
= __reserve_metadata_bytes(root
, block_rsv
->space_info
, orig_bytes
,
5260 if (ret
== -ENOSPC
&&
5261 unlikely(root
->orphan_cleanup_state
== ORPHAN_CLEANUP_STARTED
)) {
5262 struct btrfs_block_rsv
*global_rsv
=
5263 &root
->fs_info
->global_block_rsv
;
5265 if (block_rsv
!= global_rsv
&&
5266 !block_rsv_use_bytes(global_rsv
, orig_bytes
))
5270 trace_btrfs_space_reservation(root
->fs_info
,
5271 "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_block_rsv
*block_rsv
= NULL
;
5283 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
5284 (root
== root
->fs_info
->csum_root
&& trans
->adding_csums
) ||
5285 (root
== root
->fs_info
->uuid_root
))
5286 block_rsv
= trans
->block_rsv
;
5289 block_rsv
= root
->block_rsv
;
5292 block_rsv
= &root
->fs_info
->empty_block_rsv
;
5297 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
5301 spin_lock(&block_rsv
->lock
);
5302 if (block_rsv
->reserved
>= num_bytes
) {
5303 block_rsv
->reserved
-= num_bytes
;
5304 if (block_rsv
->reserved
< block_rsv
->size
)
5305 block_rsv
->full
= 0;
5308 spin_unlock(&block_rsv
->lock
);
5312 static void block_rsv_add_bytes(struct btrfs_block_rsv
*block_rsv
,
5313 u64 num_bytes
, int update_size
)
5315 spin_lock(&block_rsv
->lock
);
5316 block_rsv
->reserved
+= num_bytes
;
5318 block_rsv
->size
+= num_bytes
;
5319 else if (block_rsv
->reserved
>= block_rsv
->size
)
5320 block_rsv
->full
= 1;
5321 spin_unlock(&block_rsv
->lock
);
5324 int btrfs_cond_migrate_bytes(struct btrfs_fs_info
*fs_info
,
5325 struct btrfs_block_rsv
*dest
, u64 num_bytes
,
5328 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5331 if (global_rsv
->space_info
!= dest
->space_info
)
5334 spin_lock(&global_rsv
->lock
);
5335 min_bytes
= div_factor(global_rsv
->size
, min_factor
);
5336 if (global_rsv
->reserved
< min_bytes
+ num_bytes
) {
5337 spin_unlock(&global_rsv
->lock
);
5340 global_rsv
->reserved
-= num_bytes
;
5341 if (global_rsv
->reserved
< global_rsv
->size
)
5342 global_rsv
->full
= 0;
5343 spin_unlock(&global_rsv
->lock
);
5345 block_rsv_add_bytes(dest
, num_bytes
, 1);
5350 * This is for space we already have accounted in space_info->bytes_may_use, so
5351 * basically when we're returning space from block_rsv's.
5353 static void space_info_add_old_bytes(struct btrfs_fs_info
*fs_info
,
5354 struct btrfs_space_info
*space_info
,
5357 struct reserve_ticket
*ticket
;
5358 struct list_head
*head
;
5360 enum btrfs_reserve_flush_enum flush
= BTRFS_RESERVE_NO_FLUSH
;
5361 bool check_overcommit
= false;
5363 spin_lock(&space_info
->lock
);
5364 head
= &space_info
->priority_tickets
;
5367 * If we are over our limit then we need to check and see if we can
5368 * overcommit, and if we can't then we just need to free up our space
5369 * and not satisfy any requests.
5371 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
5372 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
5373 space_info
->bytes_may_use
;
5374 if (used
- num_bytes
>= space_info
->total_bytes
)
5375 check_overcommit
= true;
5377 while (!list_empty(head
) && num_bytes
) {
5378 ticket
= list_first_entry(head
, struct reserve_ticket
,
5381 * We use 0 bytes because this space is already reserved, so
5382 * adding the ticket space would be a double count.
5384 if (check_overcommit
&&
5385 !can_overcommit(fs_info
->extent_root
, space_info
, 0,
5388 if (num_bytes
>= ticket
->bytes
) {
5389 list_del_init(&ticket
->list
);
5390 num_bytes
-= ticket
->bytes
;
5392 wake_up(&ticket
->wait
);
5394 ticket
->bytes
-= num_bytes
;
5399 if (num_bytes
&& head
== &space_info
->priority_tickets
) {
5400 head
= &space_info
->tickets
;
5401 flush
= BTRFS_RESERVE_FLUSH_ALL
;
5404 space_info
->bytes_may_use
-= num_bytes
;
5405 trace_btrfs_space_reservation(fs_info
, "space_info",
5406 space_info
->flags
, num_bytes
, 0);
5407 spin_unlock(&space_info
->lock
);
5411 * This is for newly allocated space that isn't accounted in
5412 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5413 * we use this helper.
5415 static void space_info_add_new_bytes(struct btrfs_fs_info
*fs_info
,
5416 struct btrfs_space_info
*space_info
,
5419 struct reserve_ticket
*ticket
;
5420 struct list_head
*head
= &space_info
->priority_tickets
;
5423 while (!list_empty(head
) && num_bytes
) {
5424 ticket
= list_first_entry(head
, struct reserve_ticket
,
5426 if (num_bytes
>= ticket
->bytes
) {
5427 trace_btrfs_space_reservation(fs_info
, "space_info",
5430 list_del_init(&ticket
->list
);
5431 num_bytes
-= ticket
->bytes
;
5432 space_info
->bytes_may_use
+= ticket
->bytes
;
5434 wake_up(&ticket
->wait
);
5436 trace_btrfs_space_reservation(fs_info
, "space_info",
5439 space_info
->bytes_may_use
+= num_bytes
;
5440 ticket
->bytes
-= num_bytes
;
5445 if (num_bytes
&& head
== &space_info
->priority_tickets
) {
5446 head
= &space_info
->tickets
;
5451 static void block_rsv_release_bytes(struct btrfs_fs_info
*fs_info
,
5452 struct btrfs_block_rsv
*block_rsv
,
5453 struct btrfs_block_rsv
*dest
, u64 num_bytes
)
5455 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
5457 spin_lock(&block_rsv
->lock
);
5458 if (num_bytes
== (u64
)-1)
5459 num_bytes
= block_rsv
->size
;
5460 block_rsv
->size
-= num_bytes
;
5461 if (block_rsv
->reserved
>= block_rsv
->size
) {
5462 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
5463 block_rsv
->reserved
= block_rsv
->size
;
5464 block_rsv
->full
= 1;
5468 spin_unlock(&block_rsv
->lock
);
5470 if (num_bytes
> 0) {
5472 spin_lock(&dest
->lock
);
5476 bytes_to_add
= dest
->size
- dest
->reserved
;
5477 bytes_to_add
= min(num_bytes
, bytes_to_add
);
5478 dest
->reserved
+= bytes_to_add
;
5479 if (dest
->reserved
>= dest
->size
)
5481 num_bytes
-= bytes_to_add
;
5483 spin_unlock(&dest
->lock
);
5486 space_info_add_old_bytes(fs_info
, space_info
,
5491 int btrfs_block_rsv_migrate(struct btrfs_block_rsv
*src
,
5492 struct btrfs_block_rsv
*dst
, u64 num_bytes
,
5497 ret
= block_rsv_use_bytes(src
, num_bytes
);
5501 block_rsv_add_bytes(dst
, num_bytes
, update_size
);
5505 void btrfs_init_block_rsv(struct btrfs_block_rsv
*rsv
, unsigned short type
)
5507 memset(rsv
, 0, sizeof(*rsv
));
5508 spin_lock_init(&rsv
->lock
);
5512 struct btrfs_block_rsv
*btrfs_alloc_block_rsv(struct btrfs_root
*root
,
5513 unsigned short type
)
5515 struct btrfs_block_rsv
*block_rsv
;
5516 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5518 block_rsv
= kmalloc(sizeof(*block_rsv
), GFP_NOFS
);
5522 btrfs_init_block_rsv(block_rsv
, type
);
5523 block_rsv
->space_info
= __find_space_info(fs_info
,
5524 BTRFS_BLOCK_GROUP_METADATA
);
5528 void btrfs_free_block_rsv(struct btrfs_root
*root
,
5529 struct btrfs_block_rsv
*rsv
)
5533 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
5537 void __btrfs_free_block_rsv(struct btrfs_block_rsv
*rsv
)
5542 int btrfs_block_rsv_add(struct btrfs_root
*root
,
5543 struct btrfs_block_rsv
*block_rsv
, u64 num_bytes
,
5544 enum btrfs_reserve_flush_enum flush
)
5551 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
5553 block_rsv_add_bytes(block_rsv
, num_bytes
, 1);
5560 int btrfs_block_rsv_check(struct btrfs_root
*root
,
5561 struct btrfs_block_rsv
*block_rsv
, int min_factor
)
5569 spin_lock(&block_rsv
->lock
);
5570 num_bytes
= div_factor(block_rsv
->size
, min_factor
);
5571 if (block_rsv
->reserved
>= num_bytes
)
5573 spin_unlock(&block_rsv
->lock
);
5578 int btrfs_block_rsv_refill(struct btrfs_root
*root
,
5579 struct btrfs_block_rsv
*block_rsv
, u64 min_reserved
,
5580 enum btrfs_reserve_flush_enum flush
)
5588 spin_lock(&block_rsv
->lock
);
5589 num_bytes
= min_reserved
;
5590 if (block_rsv
->reserved
>= num_bytes
)
5593 num_bytes
-= block_rsv
->reserved
;
5594 spin_unlock(&block_rsv
->lock
);
5599 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
5601 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
5608 void btrfs_block_rsv_release(struct btrfs_root
*root
,
5609 struct btrfs_block_rsv
*block_rsv
,
5612 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
5613 if (global_rsv
== block_rsv
||
5614 block_rsv
->space_info
!= global_rsv
->space_info
)
5616 block_rsv_release_bytes(root
->fs_info
, block_rsv
, global_rsv
,
5620 static void update_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5622 struct btrfs_block_rsv
*block_rsv
= &fs_info
->global_block_rsv
;
5623 struct btrfs_space_info
*sinfo
= block_rsv
->space_info
;
5627 * The global block rsv is based on the size of the extent tree, the
5628 * checksum tree and the root tree. If the fs is empty we want to set
5629 * it to a minimal amount for safety.
5631 num_bytes
= btrfs_root_used(&fs_info
->extent_root
->root_item
) +
5632 btrfs_root_used(&fs_info
->csum_root
->root_item
) +
5633 btrfs_root_used(&fs_info
->tree_root
->root_item
);
5634 num_bytes
= max_t(u64
, num_bytes
, SZ_16M
);
5636 spin_lock(&sinfo
->lock
);
5637 spin_lock(&block_rsv
->lock
);
5639 block_rsv
->size
= min_t(u64
, num_bytes
, SZ_512M
);
5641 if (block_rsv
->reserved
< block_rsv
->size
) {
5642 num_bytes
= sinfo
->bytes_used
+ sinfo
->bytes_pinned
+
5643 sinfo
->bytes_reserved
+ sinfo
->bytes_readonly
+
5644 sinfo
->bytes_may_use
;
5645 if (sinfo
->total_bytes
> num_bytes
) {
5646 num_bytes
= sinfo
->total_bytes
- num_bytes
;
5647 num_bytes
= min(num_bytes
,
5648 block_rsv
->size
- block_rsv
->reserved
);
5649 block_rsv
->reserved
+= num_bytes
;
5650 sinfo
->bytes_may_use
+= num_bytes
;
5651 trace_btrfs_space_reservation(fs_info
, "space_info",
5652 sinfo
->flags
, num_bytes
,
5655 } else if (block_rsv
->reserved
> block_rsv
->size
) {
5656 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
5657 sinfo
->bytes_may_use
-= num_bytes
;
5658 trace_btrfs_space_reservation(fs_info
, "space_info",
5659 sinfo
->flags
, num_bytes
, 0);
5660 block_rsv
->reserved
= block_rsv
->size
;
5663 if (block_rsv
->reserved
== block_rsv
->size
)
5664 block_rsv
->full
= 1;
5666 block_rsv
->full
= 0;
5668 spin_unlock(&block_rsv
->lock
);
5669 spin_unlock(&sinfo
->lock
);
5672 static void init_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5674 struct btrfs_space_info
*space_info
;
5676 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
5677 fs_info
->chunk_block_rsv
.space_info
= space_info
;
5679 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
5680 fs_info
->global_block_rsv
.space_info
= space_info
;
5681 fs_info
->delalloc_block_rsv
.space_info
= space_info
;
5682 fs_info
->trans_block_rsv
.space_info
= space_info
;
5683 fs_info
->empty_block_rsv
.space_info
= space_info
;
5684 fs_info
->delayed_block_rsv
.space_info
= space_info
;
5686 fs_info
->extent_root
->block_rsv
= &fs_info
->global_block_rsv
;
5687 fs_info
->csum_root
->block_rsv
= &fs_info
->global_block_rsv
;
5688 fs_info
->dev_root
->block_rsv
= &fs_info
->global_block_rsv
;
5689 fs_info
->tree_root
->block_rsv
= &fs_info
->global_block_rsv
;
5690 if (fs_info
->quota_root
)
5691 fs_info
->quota_root
->block_rsv
= &fs_info
->global_block_rsv
;
5692 fs_info
->chunk_root
->block_rsv
= &fs_info
->chunk_block_rsv
;
5694 update_global_block_rsv(fs_info
);
5697 static void release_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5699 block_rsv_release_bytes(fs_info
, &fs_info
->global_block_rsv
, NULL
,
5701 WARN_ON(fs_info
->delalloc_block_rsv
.size
> 0);
5702 WARN_ON(fs_info
->delalloc_block_rsv
.reserved
> 0);
5703 WARN_ON(fs_info
->trans_block_rsv
.size
> 0);
5704 WARN_ON(fs_info
->trans_block_rsv
.reserved
> 0);
5705 WARN_ON(fs_info
->chunk_block_rsv
.size
> 0);
5706 WARN_ON(fs_info
->chunk_block_rsv
.reserved
> 0);
5707 WARN_ON(fs_info
->delayed_block_rsv
.size
> 0);
5708 WARN_ON(fs_info
->delayed_block_rsv
.reserved
> 0);
5711 void btrfs_trans_release_metadata(struct btrfs_trans_handle
*trans
,
5712 struct btrfs_root
*root
)
5714 if (!trans
->block_rsv
)
5717 if (!trans
->bytes_reserved
)
5720 trace_btrfs_space_reservation(root
->fs_info
, "transaction",
5721 trans
->transid
, trans
->bytes_reserved
, 0);
5722 btrfs_block_rsv_release(root
, trans
->block_rsv
, 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
->root
->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 inode
*inode
)
5748 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5750 * We always use trans->block_rsv here as we will have reserved space
5751 * for our orphan when starting the transaction, using get_block_rsv()
5752 * here will sometimes make us choose the wrong block rsv as we could be
5753 * doing a reloc inode for a non refcounted root.
5755 struct btrfs_block_rsv
*src_rsv
= trans
->block_rsv
;
5756 struct btrfs_block_rsv
*dst_rsv
= root
->orphan_block_rsv
;
5759 * We need to hold space in order to delete our orphan item once we've
5760 * added it, so this takes the reservation so we can release it later
5761 * when we are truly done with the orphan item.
5763 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
5764 trace_btrfs_space_reservation(root
->fs_info
, "orphan",
5765 btrfs_ino(inode
), num_bytes
, 1);
5766 return btrfs_block_rsv_migrate(src_rsv
, dst_rsv
, num_bytes
, 1);
5769 void btrfs_orphan_release_metadata(struct inode
*inode
)
5771 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5772 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
5773 trace_btrfs_space_reservation(root
->fs_info
, "orphan",
5774 btrfs_ino(inode
), num_bytes
, 0);
5775 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
, num_bytes
);
5779 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5780 * root: the root of the parent directory
5781 * rsv: block reservation
5782 * items: the number of items that we need do reservation
5783 * qgroup_reserved: used to return the reserved size in qgroup
5785 * This function is used to reserve the space for snapshot/subvolume
5786 * creation and deletion. Those operations are different with the
5787 * common file/directory operations, they change two fs/file trees
5788 * and root tree, the number of items that the qgroup reserves is
5789 * different with the free space reservation. So we can not use
5790 * the space reservation mechanism in start_transaction().
5792 int btrfs_subvolume_reserve_metadata(struct btrfs_root
*root
,
5793 struct btrfs_block_rsv
*rsv
,
5795 u64
*qgroup_reserved
,
5796 bool use_global_rsv
)
5800 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
5802 if (root
->fs_info
->quota_enabled
) {
5803 /* One for parent inode, two for dir entries */
5804 num_bytes
= 3 * root
->nodesize
;
5805 ret
= btrfs_qgroup_reserve_meta(root
, num_bytes
);
5812 *qgroup_reserved
= num_bytes
;
5814 num_bytes
= btrfs_calc_trans_metadata_size(root
, items
);
5815 rsv
->space_info
= __find_space_info(root
->fs_info
,
5816 BTRFS_BLOCK_GROUP_METADATA
);
5817 ret
= btrfs_block_rsv_add(root
, rsv
, num_bytes
,
5818 BTRFS_RESERVE_FLUSH_ALL
);
5820 if (ret
== -ENOSPC
&& use_global_rsv
)
5821 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
, num_bytes
, 1);
5823 if (ret
&& *qgroup_reserved
)
5824 btrfs_qgroup_free_meta(root
, *qgroup_reserved
);
5829 void btrfs_subvolume_release_metadata(struct btrfs_root
*root
,
5830 struct btrfs_block_rsv
*rsv
,
5831 u64 qgroup_reserved
)
5833 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
5837 * drop_outstanding_extent - drop an outstanding extent
5838 * @inode: the inode we're dropping the extent for
5839 * @num_bytes: the number of bytes we're releasing.
5841 * This is called when we are freeing up an outstanding extent, either called
5842 * after an error or after an extent is written. This will return the number of
5843 * reserved extents that need to be freed. This must be called with
5844 * BTRFS_I(inode)->lock held.
5846 static unsigned drop_outstanding_extent(struct inode
*inode
, u64 num_bytes
)
5848 unsigned drop_inode_space
= 0;
5849 unsigned dropped_extents
= 0;
5850 unsigned num_extents
= 0;
5852 num_extents
= (unsigned)div64_u64(num_bytes
+
5853 BTRFS_MAX_EXTENT_SIZE
- 1,
5854 BTRFS_MAX_EXTENT_SIZE
);
5855 ASSERT(num_extents
);
5856 ASSERT(BTRFS_I(inode
)->outstanding_extents
>= num_extents
);
5857 BTRFS_I(inode
)->outstanding_extents
-= num_extents
;
5859 if (BTRFS_I(inode
)->outstanding_extents
== 0 &&
5860 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5861 &BTRFS_I(inode
)->runtime_flags
))
5862 drop_inode_space
= 1;
5865 * If we have more or the same amount of outstanding extents than we have
5866 * reserved then we need to leave the reserved extents count alone.
5868 if (BTRFS_I(inode
)->outstanding_extents
>=
5869 BTRFS_I(inode
)->reserved_extents
)
5870 return drop_inode_space
;
5872 dropped_extents
= BTRFS_I(inode
)->reserved_extents
-
5873 BTRFS_I(inode
)->outstanding_extents
;
5874 BTRFS_I(inode
)->reserved_extents
-= dropped_extents
;
5875 return dropped_extents
+ drop_inode_space
;
5879 * calc_csum_metadata_size - return the amount of metadata space that must be
5880 * reserved/freed for the given bytes.
5881 * @inode: the inode we're manipulating
5882 * @num_bytes: the number of bytes in question
5883 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5885 * This adjusts the number of csum_bytes in the inode and then returns the
5886 * correct amount of metadata that must either be reserved or freed. We
5887 * calculate how many checksums we can fit into one leaf and then divide the
5888 * number of bytes that will need to be checksumed by this value to figure out
5889 * how many checksums will be required. If we are adding bytes then the number
5890 * may go up and we will return the number of additional bytes that must be
5891 * reserved. If it is going down we will return the number of bytes that must
5894 * This must be called with BTRFS_I(inode)->lock held.
5896 static u64
calc_csum_metadata_size(struct inode
*inode
, u64 num_bytes
,
5899 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5900 u64 old_csums
, num_csums
;
5902 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
&&
5903 BTRFS_I(inode
)->csum_bytes
== 0)
5906 old_csums
= btrfs_csum_bytes_to_leaves(root
, BTRFS_I(inode
)->csum_bytes
);
5908 BTRFS_I(inode
)->csum_bytes
+= num_bytes
;
5910 BTRFS_I(inode
)->csum_bytes
-= num_bytes
;
5911 num_csums
= btrfs_csum_bytes_to_leaves(root
, BTRFS_I(inode
)->csum_bytes
);
5913 /* No change, no need to reserve more */
5914 if (old_csums
== num_csums
)
5918 return btrfs_calc_trans_metadata_size(root
,
5919 num_csums
- old_csums
);
5921 return btrfs_calc_trans_metadata_size(root
, old_csums
- num_csums
);
5924 int btrfs_delalloc_reserve_metadata(struct inode
*inode
, u64 num_bytes
)
5926 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5927 struct btrfs_block_rsv
*block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
5930 unsigned nr_extents
= 0;
5931 enum btrfs_reserve_flush_enum flush
= BTRFS_RESERVE_FLUSH_ALL
;
5933 bool delalloc_lock
= true;
5936 bool release_extra
= false;
5938 /* If we are a free space inode we need to not flush since we will be in
5939 * the middle of a transaction commit. We also don't need the delalloc
5940 * mutex since we won't race with anybody. We need this mostly to make
5941 * lockdep shut its filthy mouth.
5943 if (btrfs_is_free_space_inode(inode
)) {
5944 flush
= BTRFS_RESERVE_NO_FLUSH
;
5945 delalloc_lock
= false;
5948 if (flush
!= BTRFS_RESERVE_NO_FLUSH
&&
5949 btrfs_transaction_in_commit(root
->fs_info
))
5950 schedule_timeout(1);
5953 mutex_lock(&BTRFS_I(inode
)->delalloc_mutex
);
5955 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
5957 spin_lock(&BTRFS_I(inode
)->lock
);
5958 nr_extents
= (unsigned)div64_u64(num_bytes
+
5959 BTRFS_MAX_EXTENT_SIZE
- 1,
5960 BTRFS_MAX_EXTENT_SIZE
);
5961 BTRFS_I(inode
)->outstanding_extents
+= nr_extents
;
5964 if (BTRFS_I(inode
)->outstanding_extents
>
5965 BTRFS_I(inode
)->reserved_extents
)
5966 nr_extents
+= BTRFS_I(inode
)->outstanding_extents
-
5967 BTRFS_I(inode
)->reserved_extents
;
5969 /* We always want to reserve a slot for updating the inode. */
5970 to_reserve
= btrfs_calc_trans_metadata_size(root
, nr_extents
+ 1);
5971 to_reserve
+= calc_csum_metadata_size(inode
, num_bytes
, 1);
5972 csum_bytes
= BTRFS_I(inode
)->csum_bytes
;
5973 spin_unlock(&BTRFS_I(inode
)->lock
);
5975 if (root
->fs_info
->quota_enabled
) {
5976 ret
= btrfs_qgroup_reserve_meta(root
,
5977 nr_extents
* root
->nodesize
);
5982 ret
= btrfs_block_rsv_add(root
, block_rsv
, to_reserve
, flush
);
5983 if (unlikely(ret
)) {
5984 btrfs_qgroup_free_meta(root
, nr_extents
* root
->nodesize
);
5988 spin_lock(&BTRFS_I(inode
)->lock
);
5989 if (test_and_set_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5990 &BTRFS_I(inode
)->runtime_flags
)) {
5991 to_reserve
-= btrfs_calc_trans_metadata_size(root
, 1);
5992 release_extra
= true;
5994 BTRFS_I(inode
)->reserved_extents
+= nr_extents
;
5995 spin_unlock(&BTRFS_I(inode
)->lock
);
5998 mutex_unlock(&BTRFS_I(inode
)->delalloc_mutex
);
6001 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
6002 btrfs_ino(inode
), to_reserve
, 1);
6004 btrfs_block_rsv_release(root
, block_rsv
,
6005 btrfs_calc_trans_metadata_size(root
,
6010 spin_lock(&BTRFS_I(inode
)->lock
);
6011 dropped
= drop_outstanding_extent(inode
, num_bytes
);
6013 * If the inodes csum_bytes is the same as the original
6014 * csum_bytes then we know we haven't raced with any free()ers
6015 * so we can just reduce our inodes csum bytes and carry on.
6017 if (BTRFS_I(inode
)->csum_bytes
== csum_bytes
) {
6018 calc_csum_metadata_size(inode
, num_bytes
, 0);
6020 u64 orig_csum_bytes
= BTRFS_I(inode
)->csum_bytes
;
6024 * This is tricky, but first we need to figure out how much we
6025 * freed from any free-ers that occurred during this
6026 * reservation, so we reset ->csum_bytes to the csum_bytes
6027 * before we dropped our lock, and then call the free for the
6028 * number of bytes that were freed while we were trying our
6031 bytes
= csum_bytes
- BTRFS_I(inode
)->csum_bytes
;
6032 BTRFS_I(inode
)->csum_bytes
= csum_bytes
;
6033 to_free
= calc_csum_metadata_size(inode
, bytes
, 0);
6037 * Now we need to see how much we would have freed had we not
6038 * been making this reservation and our ->csum_bytes were not
6039 * artificially inflated.
6041 BTRFS_I(inode
)->csum_bytes
= csum_bytes
- num_bytes
;
6042 bytes
= csum_bytes
- orig_csum_bytes
;
6043 bytes
= calc_csum_metadata_size(inode
, bytes
, 0);
6046 * Now reset ->csum_bytes to what it should be. If bytes is
6047 * more than to_free then we would have freed more space had we
6048 * not had an artificially high ->csum_bytes, so we need to free
6049 * the remainder. If bytes is the same or less then we don't
6050 * need to do anything, the other free-ers did the correct
6053 BTRFS_I(inode
)->csum_bytes
= orig_csum_bytes
- num_bytes
;
6054 if (bytes
> to_free
)
6055 to_free
= bytes
- to_free
;
6059 spin_unlock(&BTRFS_I(inode
)->lock
);
6061 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
6064 btrfs_block_rsv_release(root
, block_rsv
, to_free
);
6065 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
6066 btrfs_ino(inode
), to_free
, 0);
6069 mutex_unlock(&BTRFS_I(inode
)->delalloc_mutex
);
6074 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
6075 * @inode: the inode to release the reservation for
6076 * @num_bytes: the number of bytes we're releasing
6078 * This will release the metadata reservation for an inode. This can be called
6079 * once we complete IO for a given set of bytes to release their metadata
6082 void btrfs_delalloc_release_metadata(struct inode
*inode
, u64 num_bytes
)
6084 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6088 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
6089 spin_lock(&BTRFS_I(inode
)->lock
);
6090 dropped
= drop_outstanding_extent(inode
, num_bytes
);
6093 to_free
= calc_csum_metadata_size(inode
, num_bytes
, 0);
6094 spin_unlock(&BTRFS_I(inode
)->lock
);
6096 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
6098 if (btrfs_test_is_dummy_root(root
))
6101 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
6102 btrfs_ino(inode
), to_free
, 0);
6104 btrfs_block_rsv_release(root
, &root
->fs_info
->delalloc_block_rsv
,
6109 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6111 * @inode: inode we're writing to
6112 * @start: start range we are writing to
6113 * @len: how long the range we are writing to
6115 * TODO: This function will finally replace old btrfs_delalloc_reserve_space()
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(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(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_root
*root
, u64 bytenr
,
6169 u64 num_bytes
, int alloc
)
6171 struct btrfs_block_group_cache
*cache
= NULL
;
6172 struct btrfs_fs_info
*info
= root
->fs_info
;
6173 u64 total
= num_bytes
;
6178 /* block accounting for super block */
6179 spin_lock(&info
->delalloc_root_lock
);
6180 old_val
= btrfs_super_bytes_used(info
->super_copy
);
6182 old_val
+= num_bytes
;
6184 old_val
-= num_bytes
;
6185 btrfs_set_super_bytes_used(info
->super_copy
, old_val
);
6186 spin_unlock(&info
->delalloc_root_lock
);
6189 cache
= btrfs_lookup_block_group(info
, bytenr
);
6192 if (cache
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
6193 BTRFS_BLOCK_GROUP_RAID1
|
6194 BTRFS_BLOCK_GROUP_RAID10
))
6199 * If this block group has free space cache written out, we
6200 * need to make sure to load it if we are removing space. This
6201 * is because we need the unpinning stage to actually add the
6202 * space back to the block group, otherwise we will leak space.
6204 if (!alloc
&& cache
->cached
== BTRFS_CACHE_NO
)
6205 cache_block_group(cache
, 1);
6207 byte_in_group
= bytenr
- cache
->key
.objectid
;
6208 WARN_ON(byte_in_group
> cache
->key
.offset
);
6210 spin_lock(&cache
->space_info
->lock
);
6211 spin_lock(&cache
->lock
);
6213 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
6214 cache
->disk_cache_state
< BTRFS_DC_CLEAR
)
6215 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
6217 old_val
= btrfs_block_group_used(&cache
->item
);
6218 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
6220 old_val
+= num_bytes
;
6221 btrfs_set_block_group_used(&cache
->item
, old_val
);
6222 cache
->reserved
-= num_bytes
;
6223 cache
->space_info
->bytes_reserved
-= num_bytes
;
6224 cache
->space_info
->bytes_used
+= num_bytes
;
6225 cache
->space_info
->disk_used
+= num_bytes
* factor
;
6226 spin_unlock(&cache
->lock
);
6227 spin_unlock(&cache
->space_info
->lock
);
6229 old_val
-= num_bytes
;
6230 btrfs_set_block_group_used(&cache
->item
, old_val
);
6231 cache
->pinned
+= num_bytes
;
6232 cache
->space_info
->bytes_pinned
+= num_bytes
;
6233 cache
->space_info
->bytes_used
-= num_bytes
;
6234 cache
->space_info
->disk_used
-= num_bytes
* factor
;
6235 spin_unlock(&cache
->lock
);
6236 spin_unlock(&cache
->space_info
->lock
);
6238 trace_btrfs_space_reservation(root
->fs_info
, "pinned",
6239 cache
->space_info
->flags
,
6241 set_extent_dirty(info
->pinned_extents
,
6242 bytenr
, bytenr
+ num_bytes
- 1,
6243 GFP_NOFS
| __GFP_NOFAIL
);
6246 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
6247 if (list_empty(&cache
->dirty_list
)) {
6248 list_add_tail(&cache
->dirty_list
,
6249 &trans
->transaction
->dirty_bgs
);
6250 trans
->transaction
->num_dirty_bgs
++;
6251 btrfs_get_block_group(cache
);
6253 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
6256 * No longer have used bytes in this block group, queue it for
6257 * deletion. We do this after adding the block group to the
6258 * dirty list to avoid races between cleaner kthread and space
6261 if (!alloc
&& old_val
== 0) {
6262 spin_lock(&info
->unused_bgs_lock
);
6263 if (list_empty(&cache
->bg_list
)) {
6264 btrfs_get_block_group(cache
);
6265 list_add_tail(&cache
->bg_list
,
6268 spin_unlock(&info
->unused_bgs_lock
);
6271 btrfs_put_block_group(cache
);
6273 bytenr
+= num_bytes
;
6278 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
6280 struct btrfs_block_group_cache
*cache
;
6283 spin_lock(&root
->fs_info
->block_group_cache_lock
);
6284 bytenr
= root
->fs_info
->first_logical_byte
;
6285 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
6287 if (bytenr
< (u64
)-1)
6290 cache
= btrfs_lookup_first_block_group(root
->fs_info
, search_start
);
6294 bytenr
= cache
->key
.objectid
;
6295 btrfs_put_block_group(cache
);
6300 static int pin_down_extent(struct btrfs_root
*root
,
6301 struct btrfs_block_group_cache
*cache
,
6302 u64 bytenr
, u64 num_bytes
, int reserved
)
6304 spin_lock(&cache
->space_info
->lock
);
6305 spin_lock(&cache
->lock
);
6306 cache
->pinned
+= num_bytes
;
6307 cache
->space_info
->bytes_pinned
+= num_bytes
;
6309 cache
->reserved
-= num_bytes
;
6310 cache
->space_info
->bytes_reserved
-= num_bytes
;
6312 spin_unlock(&cache
->lock
);
6313 spin_unlock(&cache
->space_info
->lock
);
6315 trace_btrfs_space_reservation(root
->fs_info
, "pinned",
6316 cache
->space_info
->flags
, num_bytes
, 1);
6317 set_extent_dirty(root
->fs_info
->pinned_extents
, bytenr
,
6318 bytenr
+ num_bytes
- 1, GFP_NOFS
| __GFP_NOFAIL
);
6323 * this function must be called within transaction
6325 int btrfs_pin_extent(struct btrfs_root
*root
,
6326 u64 bytenr
, u64 num_bytes
, int reserved
)
6328 struct btrfs_block_group_cache
*cache
;
6330 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
6331 BUG_ON(!cache
); /* Logic error */
6333 pin_down_extent(root
, cache
, bytenr
, num_bytes
, reserved
);
6335 btrfs_put_block_group(cache
);
6340 * this function must be called within transaction
6342 int btrfs_pin_extent_for_log_replay(struct btrfs_root
*root
,
6343 u64 bytenr
, u64 num_bytes
)
6345 struct btrfs_block_group_cache
*cache
;
6348 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
6353 * pull in the free space cache (if any) so that our pin
6354 * removes the free space from the cache. We have load_only set
6355 * to one because the slow code to read in the free extents does check
6356 * the pinned extents.
6358 cache_block_group(cache
, 1);
6360 pin_down_extent(root
, cache
, bytenr
, num_bytes
, 0);
6362 /* remove us from the free space cache (if we're there at all) */
6363 ret
= btrfs_remove_free_space(cache
, bytenr
, num_bytes
);
6364 btrfs_put_block_group(cache
);
6368 static int __exclude_logged_extent(struct btrfs_root
*root
, 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(root
->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(root
, 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(root
, 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_root
*log
,
6414 struct extent_buffer
*eb
)
6416 struct btrfs_file_extent_item
*item
;
6417 struct btrfs_key key
;
6421 if (!btrfs_fs_incompat(log
->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(log
, 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_update_reserved_bytes - update the block_group and space info counters
6495 * @cache: The cache we are manipulating
6496 * @num_bytes: The number of bytes in question
6497 * @reserve: One of the reservation enums
6498 * @delalloc: The blocks are allocated for the delalloc write
6500 * This is called by the allocator when it reserves space, or by somebody who is
6501 * freeing space that was never actually used on disk. For example if you
6502 * reserve some space for a new leaf in transaction A and before transaction A
6503 * commits you free that leaf, you call this with reserve set to 0 in order to
6504 * clear the reservation.
6506 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
6507 * ENOSPC accounting. For data we handle the reservation through clearing the
6508 * delalloc bits in the io_tree. We have to do this since we could end up
6509 * allocating less disk space for the amount of data we have reserved in the
6510 * case of compression.
6512 * If this is a reservation and the block group has become read only we cannot
6513 * make the reservation and return -EAGAIN, otherwise this function always
6516 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
6517 u64 num_bytes
, int reserve
, int delalloc
)
6519 struct btrfs_space_info
*space_info
= cache
->space_info
;
6522 spin_lock(&space_info
->lock
);
6523 spin_lock(&cache
->lock
);
6524 if (reserve
!= RESERVE_FREE
) {
6528 cache
->reserved
+= num_bytes
;
6529 space_info
->bytes_reserved
+= num_bytes
;
6530 if (reserve
== RESERVE_ALLOC
) {
6531 trace_btrfs_space_reservation(cache
->fs_info
,
6532 "space_info", space_info
->flags
,
6534 space_info
->bytes_may_use
-= num_bytes
;
6538 cache
->delalloc_bytes
+= num_bytes
;
6542 space_info
->bytes_readonly
+= num_bytes
;
6543 cache
->reserved
-= num_bytes
;
6544 space_info
->bytes_reserved
-= num_bytes
;
6547 cache
->delalloc_bytes
-= num_bytes
;
6549 spin_unlock(&cache
->lock
);
6550 spin_unlock(&space_info
->lock
);
6554 void btrfs_prepare_extent_commit(struct btrfs_trans_handle
*trans
,
6555 struct btrfs_root
*root
)
6557 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6558 struct btrfs_caching_control
*next
;
6559 struct btrfs_caching_control
*caching_ctl
;
6560 struct btrfs_block_group_cache
*cache
;
6562 down_write(&fs_info
->commit_root_sem
);
6564 list_for_each_entry_safe(caching_ctl
, next
,
6565 &fs_info
->caching_block_groups
, list
) {
6566 cache
= caching_ctl
->block_group
;
6567 if (block_group_cache_done(cache
)) {
6568 cache
->last_byte_to_unpin
= (u64
)-1;
6569 list_del_init(&caching_ctl
->list
);
6570 put_caching_control(caching_ctl
);
6572 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
6576 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
6577 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
6579 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
6581 up_write(&fs_info
->commit_root_sem
);
6583 update_global_block_rsv(fs_info
);
6587 * Returns the free cluster for the given space info and sets empty_cluster to
6588 * what it should be based on the mount options.
6590 static struct btrfs_free_cluster
*
6591 fetch_cluster_info(struct btrfs_root
*root
, struct btrfs_space_info
*space_info
,
6594 struct btrfs_free_cluster
*ret
= NULL
;
6595 bool ssd
= btrfs_test_opt(root
, SSD
);
6598 if (btrfs_mixed_space_info(space_info
))
6602 *empty_cluster
= SZ_2M
;
6603 if (space_info
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
6604 ret
= &root
->fs_info
->meta_alloc_cluster
;
6606 *empty_cluster
= SZ_64K
;
6607 } else if ((space_info
->flags
& BTRFS_BLOCK_GROUP_DATA
) && ssd
) {
6608 ret
= &root
->fs_info
->data_alloc_cluster
;
6614 static int unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
,
6615 const bool return_free_space
)
6617 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6618 struct btrfs_block_group_cache
*cache
= NULL
;
6619 struct btrfs_space_info
*space_info
;
6620 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
6621 struct btrfs_free_cluster
*cluster
= NULL
;
6623 u64 total_unpinned
= 0;
6624 u64 empty_cluster
= 0;
6627 while (start
<= end
) {
6630 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
6632 btrfs_put_block_group(cache
);
6634 cache
= btrfs_lookup_block_group(fs_info
, start
);
6635 BUG_ON(!cache
); /* Logic error */
6637 cluster
= fetch_cluster_info(root
,
6640 empty_cluster
<<= 1;
6643 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
6644 len
= min(len
, end
+ 1 - start
);
6646 if (start
< cache
->last_byte_to_unpin
) {
6647 len
= min(len
, cache
->last_byte_to_unpin
- start
);
6648 if (return_free_space
)
6649 btrfs_add_free_space(cache
, start
, len
);
6653 total_unpinned
+= len
;
6654 space_info
= cache
->space_info
;
6657 * If this space cluster has been marked as fragmented and we've
6658 * unpinned enough in this block group to potentially allow a
6659 * cluster to be created inside of it go ahead and clear the
6662 if (cluster
&& cluster
->fragmented
&&
6663 total_unpinned
> empty_cluster
) {
6664 spin_lock(&cluster
->lock
);
6665 cluster
->fragmented
= 0;
6666 spin_unlock(&cluster
->lock
);
6669 spin_lock(&space_info
->lock
);
6670 spin_lock(&cache
->lock
);
6671 cache
->pinned
-= len
;
6672 space_info
->bytes_pinned
-= len
;
6674 trace_btrfs_space_reservation(fs_info
, "pinned",
6675 space_info
->flags
, len
, 0);
6676 space_info
->max_extent_size
= 0;
6677 percpu_counter_add(&space_info
->total_bytes_pinned
, -len
);
6679 space_info
->bytes_readonly
+= len
;
6682 spin_unlock(&cache
->lock
);
6683 if (!readonly
&& return_free_space
&&
6684 global_rsv
->space_info
== space_info
) {
6686 WARN_ON(!return_free_space
);
6687 spin_lock(&global_rsv
->lock
);
6688 if (!global_rsv
->full
) {
6689 to_add
= min(len
, global_rsv
->size
-
6690 global_rsv
->reserved
);
6691 global_rsv
->reserved
+= to_add
;
6692 space_info
->bytes_may_use
+= to_add
;
6693 if (global_rsv
->reserved
>= global_rsv
->size
)
6694 global_rsv
->full
= 1;
6695 trace_btrfs_space_reservation(fs_info
,
6701 spin_unlock(&global_rsv
->lock
);
6702 /* Add to any tickets we may have */
6704 space_info_add_new_bytes(fs_info
, space_info
,
6707 spin_unlock(&space_info
->lock
);
6711 btrfs_put_block_group(cache
);
6715 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
6716 struct btrfs_root
*root
)
6718 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6719 struct btrfs_block_group_cache
*block_group
, *tmp
;
6720 struct list_head
*deleted_bgs
;
6721 struct extent_io_tree
*unpin
;
6726 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
6727 unpin
= &fs_info
->freed_extents
[1];
6729 unpin
= &fs_info
->freed_extents
[0];
6731 while (!trans
->aborted
) {
6732 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
6733 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
6734 EXTENT_DIRTY
, NULL
);
6736 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
6740 if (btrfs_test_opt(root
, DISCARD
))
6741 ret
= btrfs_discard_extent(root
, start
,
6742 end
+ 1 - start
, NULL
);
6744 clear_extent_dirty(unpin
, start
, end
);
6745 unpin_extent_range(root
, start
, end
, true);
6746 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
6751 * Transaction is finished. We don't need the lock anymore. We
6752 * do need to clean up the block groups in case of a transaction
6755 deleted_bgs
= &trans
->transaction
->deleted_bgs
;
6756 list_for_each_entry_safe(block_group
, tmp
, deleted_bgs
, bg_list
) {
6760 if (!trans
->aborted
)
6761 ret
= btrfs_discard_extent(root
,
6762 block_group
->key
.objectid
,
6763 block_group
->key
.offset
,
6766 list_del_init(&block_group
->bg_list
);
6767 btrfs_put_block_group_trimming(block_group
);
6768 btrfs_put_block_group(block_group
);
6771 const char *errstr
= btrfs_decode_error(ret
);
6773 "Discard failed while removing blockgroup: errno=%d %s\n",
6781 static void add_pinned_bytes(struct btrfs_fs_info
*fs_info
, u64 num_bytes
,
6782 u64 owner
, u64 root_objectid
)
6784 struct btrfs_space_info
*space_info
;
6787 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
6788 if (root_objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
6789 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
6791 flags
= BTRFS_BLOCK_GROUP_METADATA
;
6793 flags
= BTRFS_BLOCK_GROUP_DATA
;
6796 space_info
= __find_space_info(fs_info
, flags
);
6797 BUG_ON(!space_info
); /* Logic bug */
6798 percpu_counter_add(&space_info
->total_bytes_pinned
, num_bytes
);
6802 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
6803 struct btrfs_root
*root
,
6804 struct btrfs_delayed_ref_node
*node
, u64 parent
,
6805 u64 root_objectid
, u64 owner_objectid
,
6806 u64 owner_offset
, int refs_to_drop
,
6807 struct btrfs_delayed_extent_op
*extent_op
)
6809 struct btrfs_key key
;
6810 struct btrfs_path
*path
;
6811 struct btrfs_fs_info
*info
= root
->fs_info
;
6812 struct btrfs_root
*extent_root
= info
->extent_root
;
6813 struct extent_buffer
*leaf
;
6814 struct btrfs_extent_item
*ei
;
6815 struct btrfs_extent_inline_ref
*iref
;
6818 int extent_slot
= 0;
6819 int found_extent
= 0;
6823 u64 bytenr
= node
->bytenr
;
6824 u64 num_bytes
= node
->num_bytes
;
6826 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
6829 path
= btrfs_alloc_path();
6833 path
->reada
= READA_FORWARD
;
6834 path
->leave_spinning
= 1;
6836 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
6837 BUG_ON(!is_data
&& refs_to_drop
!= 1);
6840 skinny_metadata
= 0;
6842 ret
= lookup_extent_backref(trans
, extent_root
, path
, &iref
,
6843 bytenr
, num_bytes
, parent
,
6844 root_objectid
, owner_objectid
,
6847 extent_slot
= path
->slots
[0];
6848 while (extent_slot
>= 0) {
6849 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
6851 if (key
.objectid
!= bytenr
)
6853 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
6854 key
.offset
== num_bytes
) {
6858 if (key
.type
== BTRFS_METADATA_ITEM_KEY
&&
6859 key
.offset
== owner_objectid
) {
6863 if (path
->slots
[0] - extent_slot
> 5)
6867 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6868 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
6869 if (found_extent
&& item_size
< sizeof(*ei
))
6872 if (!found_extent
) {
6874 ret
= remove_extent_backref(trans
, extent_root
, path
,
6876 is_data
, &last_ref
);
6878 btrfs_abort_transaction(trans
, extent_root
, ret
);
6881 btrfs_release_path(path
);
6882 path
->leave_spinning
= 1;
6884 key
.objectid
= bytenr
;
6885 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6886 key
.offset
= num_bytes
;
6888 if (!is_data
&& skinny_metadata
) {
6889 key
.type
= BTRFS_METADATA_ITEM_KEY
;
6890 key
.offset
= owner_objectid
;
6893 ret
= btrfs_search_slot(trans
, extent_root
,
6895 if (ret
> 0 && skinny_metadata
&& path
->slots
[0]) {
6897 * Couldn't find our skinny metadata item,
6898 * see if we have ye olde extent item.
6901 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
6903 if (key
.objectid
== bytenr
&&
6904 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
6905 key
.offset
== num_bytes
)
6909 if (ret
> 0 && skinny_metadata
) {
6910 skinny_metadata
= false;
6911 key
.objectid
= bytenr
;
6912 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6913 key
.offset
= num_bytes
;
6914 btrfs_release_path(path
);
6915 ret
= btrfs_search_slot(trans
, extent_root
,
6920 btrfs_err(info
, "umm, got %d back from search, was looking for %llu",
6923 btrfs_print_leaf(extent_root
,
6927 btrfs_abort_transaction(trans
, extent_root
, ret
);
6930 extent_slot
= path
->slots
[0];
6932 } else if (WARN_ON(ret
== -ENOENT
)) {
6933 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
6935 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6936 bytenr
, parent
, root_objectid
, owner_objectid
,
6938 btrfs_abort_transaction(trans
, extent_root
, ret
);
6941 btrfs_abort_transaction(trans
, extent_root
, ret
);
6945 leaf
= path
->nodes
[0];
6946 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
6947 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6948 if (item_size
< sizeof(*ei
)) {
6949 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
6950 ret
= convert_extent_item_v0(trans
, extent_root
, path
,
6953 btrfs_abort_transaction(trans
, extent_root
, ret
);
6957 btrfs_release_path(path
);
6958 path
->leave_spinning
= 1;
6960 key
.objectid
= bytenr
;
6961 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6962 key
.offset
= num_bytes
;
6964 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
6967 btrfs_err(info
, "umm, got %d back from search, was looking for %llu",
6969 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
6972 btrfs_abort_transaction(trans
, extent_root
, ret
);
6976 extent_slot
= path
->slots
[0];
6977 leaf
= path
->nodes
[0];
6978 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
6981 BUG_ON(item_size
< sizeof(*ei
));
6982 ei
= btrfs_item_ptr(leaf
, extent_slot
,
6983 struct btrfs_extent_item
);
6984 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
6985 key
.type
== BTRFS_EXTENT_ITEM_KEY
) {
6986 struct btrfs_tree_block_info
*bi
;
6987 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
6988 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
6989 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
6992 refs
= btrfs_extent_refs(leaf
, ei
);
6993 if (refs
< refs_to_drop
) {
6994 btrfs_err(info
, "trying to drop %d refs but we only have %Lu "
6995 "for bytenr %Lu", refs_to_drop
, refs
, bytenr
);
6997 btrfs_abort_transaction(trans
, extent_root
, ret
);
7000 refs
-= refs_to_drop
;
7004 __run_delayed_extent_op(extent_op
, leaf
, ei
);
7006 * In the case of inline back ref, reference count will
7007 * be updated by remove_extent_backref
7010 BUG_ON(!found_extent
);
7012 btrfs_set_extent_refs(leaf
, ei
, refs
);
7013 btrfs_mark_buffer_dirty(leaf
);
7016 ret
= remove_extent_backref(trans
, extent_root
, path
,
7018 is_data
, &last_ref
);
7020 btrfs_abort_transaction(trans
, extent_root
, ret
);
7024 add_pinned_bytes(root
->fs_info
, -num_bytes
, owner_objectid
,
7028 BUG_ON(is_data
&& refs_to_drop
!=
7029 extent_data_ref_count(path
, iref
));
7031 BUG_ON(path
->slots
[0] != extent_slot
);
7033 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
7034 path
->slots
[0] = extent_slot
;
7040 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
7043 btrfs_abort_transaction(trans
, extent_root
, ret
);
7046 btrfs_release_path(path
);
7049 ret
= btrfs_del_csums(trans
, root
, bytenr
, num_bytes
);
7051 btrfs_abort_transaction(trans
, extent_root
, ret
);
7056 ret
= add_to_free_space_tree(trans
, root
->fs_info
, bytenr
,
7059 btrfs_abort_transaction(trans
, extent_root
, ret
);
7063 ret
= update_block_group(trans
, root
, bytenr
, num_bytes
, 0);
7065 btrfs_abort_transaction(trans
, extent_root
, ret
);
7069 btrfs_release_path(path
);
7072 btrfs_free_path(path
);
7077 * when we free an block, it is possible (and likely) that we free the last
7078 * delayed ref for that extent as well. This searches the delayed ref tree for
7079 * a given extent, and if there are no other delayed refs to be processed, it
7080 * removes it from the tree.
7082 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
7083 struct btrfs_root
*root
, u64 bytenr
)
7085 struct btrfs_delayed_ref_head
*head
;
7086 struct btrfs_delayed_ref_root
*delayed_refs
;
7089 delayed_refs
= &trans
->transaction
->delayed_refs
;
7090 spin_lock(&delayed_refs
->lock
);
7091 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
7093 goto out_delayed_unlock
;
7095 spin_lock(&head
->lock
);
7096 if (!list_empty(&head
->ref_list
))
7099 if (head
->extent_op
) {
7100 if (!head
->must_insert_reserved
)
7102 btrfs_free_delayed_extent_op(head
->extent_op
);
7103 head
->extent_op
= NULL
;
7107 * waiting for the lock here would deadlock. If someone else has it
7108 * locked they are already in the process of dropping it anyway
7110 if (!mutex_trylock(&head
->mutex
))
7114 * at this point we have a head with no other entries. Go
7115 * ahead and process it.
7117 head
->node
.in_tree
= 0;
7118 rb_erase(&head
->href_node
, &delayed_refs
->href_root
);
7120 atomic_dec(&delayed_refs
->num_entries
);
7123 * we don't take a ref on the node because we're removing it from the
7124 * tree, so we just steal the ref the tree was holding.
7126 delayed_refs
->num_heads
--;
7127 if (head
->processing
== 0)
7128 delayed_refs
->num_heads_ready
--;
7129 head
->processing
= 0;
7130 spin_unlock(&head
->lock
);
7131 spin_unlock(&delayed_refs
->lock
);
7133 BUG_ON(head
->extent_op
);
7134 if (head
->must_insert_reserved
)
7137 mutex_unlock(&head
->mutex
);
7138 btrfs_put_delayed_ref(&head
->node
);
7141 spin_unlock(&head
->lock
);
7144 spin_unlock(&delayed_refs
->lock
);
7148 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
7149 struct btrfs_root
*root
,
7150 struct extent_buffer
*buf
,
7151 u64 parent
, int last_ref
)
7156 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
7157 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
7158 buf
->start
, buf
->len
,
7159 parent
, root
->root_key
.objectid
,
7160 btrfs_header_level(buf
),
7161 BTRFS_DROP_DELAYED_REF
, NULL
);
7162 BUG_ON(ret
); /* -ENOMEM */
7168 if (btrfs_header_generation(buf
) == trans
->transid
) {
7169 struct btrfs_block_group_cache
*cache
;
7171 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
7172 ret
= check_ref_cleanup(trans
, root
, buf
->start
);
7177 cache
= btrfs_lookup_block_group(root
->fs_info
, buf
->start
);
7179 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
7180 pin_down_extent(root
, cache
, buf
->start
, buf
->len
, 1);
7181 btrfs_put_block_group(cache
);
7185 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
7187 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
7188 btrfs_update_reserved_bytes(cache
, buf
->len
, RESERVE_FREE
, 0);
7189 btrfs_put_block_group(cache
);
7190 trace_btrfs_reserved_extent_free(root
, buf
->start
, buf
->len
);
7195 add_pinned_bytes(root
->fs_info
, buf
->len
,
7196 btrfs_header_level(buf
),
7197 root
->root_key
.objectid
);
7200 * Deleting the buffer, clear the corrupt flag since it doesn't matter
7203 clear_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
);
7206 /* Can return -ENOMEM */
7207 int btrfs_free_extent(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
7208 u64 bytenr
, u64 num_bytes
, u64 parent
, u64 root_objectid
,
7209 u64 owner
, u64 offset
)
7212 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7214 if (btrfs_test_is_dummy_root(root
))
7217 add_pinned_bytes(root
->fs_info
, num_bytes
, owner
, root_objectid
);
7220 * tree log blocks never actually go into the extent allocation
7221 * tree, just update pinning info and exit early.
7223 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
7224 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
7225 /* unlocks the pinned mutex */
7226 btrfs_pin_extent(root
, bytenr
, num_bytes
, 1);
7228 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
7229 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
7231 parent
, root_objectid
, (int)owner
,
7232 BTRFS_DROP_DELAYED_REF
, NULL
);
7234 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
7236 parent
, root_objectid
, owner
,
7238 BTRFS_DROP_DELAYED_REF
, NULL
);
7244 * when we wait for progress in the block group caching, its because
7245 * our allocation attempt failed at least once. So, we must sleep
7246 * and let some progress happen before we try again.
7248 * This function will sleep at least once waiting for new free space to
7249 * show up, and then it will check the block group free space numbers
7250 * for our min num_bytes. Another option is to have it go ahead
7251 * and look in the rbtree for a free extent of a given size, but this
7254 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7255 * any of the information in this block group.
7257 static noinline
void
7258 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
7261 struct btrfs_caching_control
*caching_ctl
;
7263 caching_ctl
= get_caching_control(cache
);
7267 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
7268 (cache
->free_space_ctl
->free_space
>= num_bytes
));
7270 put_caching_control(caching_ctl
);
7274 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
7276 struct btrfs_caching_control
*caching_ctl
;
7279 caching_ctl
= get_caching_control(cache
);
7281 return (cache
->cached
== BTRFS_CACHE_ERROR
) ? -EIO
: 0;
7283 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
7284 if (cache
->cached
== BTRFS_CACHE_ERROR
)
7286 put_caching_control(caching_ctl
);
7290 int __get_raid_index(u64 flags
)
7292 if (flags
& BTRFS_BLOCK_GROUP_RAID10
)
7293 return BTRFS_RAID_RAID10
;
7294 else if (flags
& BTRFS_BLOCK_GROUP_RAID1
)
7295 return BTRFS_RAID_RAID1
;
7296 else if (flags
& BTRFS_BLOCK_GROUP_DUP
)
7297 return BTRFS_RAID_DUP
;
7298 else if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
7299 return BTRFS_RAID_RAID0
;
7300 else if (flags
& BTRFS_BLOCK_GROUP_RAID5
)
7301 return BTRFS_RAID_RAID5
;
7302 else if (flags
& BTRFS_BLOCK_GROUP_RAID6
)
7303 return BTRFS_RAID_RAID6
;
7305 return BTRFS_RAID_SINGLE
; /* BTRFS_BLOCK_GROUP_SINGLE */
7308 int get_block_group_index(struct btrfs_block_group_cache
*cache
)
7310 return __get_raid_index(cache
->flags
);
7313 static const char *btrfs_raid_type_names
[BTRFS_NR_RAID_TYPES
] = {
7314 [BTRFS_RAID_RAID10
] = "raid10",
7315 [BTRFS_RAID_RAID1
] = "raid1",
7316 [BTRFS_RAID_DUP
] = "dup",
7317 [BTRFS_RAID_RAID0
] = "raid0",
7318 [BTRFS_RAID_SINGLE
] = "single",
7319 [BTRFS_RAID_RAID5
] = "raid5",
7320 [BTRFS_RAID_RAID6
] = "raid6",
7323 static const char *get_raid_name(enum btrfs_raid_types type
)
7325 if (type
>= BTRFS_NR_RAID_TYPES
)
7328 return btrfs_raid_type_names
[type
];
7331 enum btrfs_loop_type
{
7332 LOOP_CACHING_NOWAIT
= 0,
7333 LOOP_CACHING_WAIT
= 1,
7334 LOOP_ALLOC_CHUNK
= 2,
7335 LOOP_NO_EMPTY_SIZE
= 3,
7339 btrfs_lock_block_group(struct btrfs_block_group_cache
*cache
,
7343 down_read(&cache
->data_rwsem
);
7347 btrfs_grab_block_group(struct btrfs_block_group_cache
*cache
,
7350 btrfs_get_block_group(cache
);
7352 down_read(&cache
->data_rwsem
);
7355 static struct btrfs_block_group_cache
*
7356 btrfs_lock_cluster(struct btrfs_block_group_cache
*block_group
,
7357 struct btrfs_free_cluster
*cluster
,
7360 struct btrfs_block_group_cache
*used_bg
= NULL
;
7362 spin_lock(&cluster
->refill_lock
);
7364 used_bg
= cluster
->block_group
;
7368 if (used_bg
== block_group
)
7371 btrfs_get_block_group(used_bg
);
7376 if (down_read_trylock(&used_bg
->data_rwsem
))
7379 spin_unlock(&cluster
->refill_lock
);
7381 down_read(&used_bg
->data_rwsem
);
7383 spin_lock(&cluster
->refill_lock
);
7384 if (used_bg
== cluster
->block_group
)
7387 up_read(&used_bg
->data_rwsem
);
7388 btrfs_put_block_group(used_bg
);
7393 btrfs_release_block_group(struct btrfs_block_group_cache
*cache
,
7397 up_read(&cache
->data_rwsem
);
7398 btrfs_put_block_group(cache
);
7402 * walks the btree of allocated extents and find a hole of a given size.
7403 * The key ins is changed to record the hole:
7404 * ins->objectid == start position
7405 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7406 * ins->offset == the size of the hole.
7407 * Any available blocks before search_start are skipped.
7409 * If there is no suitable free space, we will record the max size of
7410 * the free space extent currently.
7412 static noinline
int find_free_extent(struct btrfs_root
*orig_root
,
7413 u64 num_bytes
, u64 empty_size
,
7414 u64 hint_byte
, struct btrfs_key
*ins
,
7415 u64 flags
, int delalloc
)
7418 struct btrfs_root
*root
= orig_root
->fs_info
->extent_root
;
7419 struct btrfs_free_cluster
*last_ptr
= NULL
;
7420 struct btrfs_block_group_cache
*block_group
= NULL
;
7421 u64 search_start
= 0;
7422 u64 max_extent_size
= 0;
7423 u64 empty_cluster
= 0;
7424 struct btrfs_space_info
*space_info
;
7426 int index
= __get_raid_index(flags
);
7427 int alloc_type
= (flags
& BTRFS_BLOCK_GROUP_DATA
) ?
7428 RESERVE_ALLOC_NO_ACCOUNT
: RESERVE_ALLOC
;
7429 bool failed_cluster_refill
= false;
7430 bool failed_alloc
= false;
7431 bool use_cluster
= true;
7432 bool have_caching_bg
= false;
7433 bool orig_have_caching_bg
= false;
7434 bool full_search
= false;
7436 WARN_ON(num_bytes
< root
->sectorsize
);
7437 ins
->type
= BTRFS_EXTENT_ITEM_KEY
;
7441 trace_find_free_extent(orig_root
, num_bytes
, empty_size
, flags
);
7443 space_info
= __find_space_info(root
->fs_info
, flags
);
7445 btrfs_err(root
->fs_info
, "No space info for %llu", flags
);
7450 * If our free space is heavily fragmented we may not be able to make
7451 * big contiguous allocations, so instead of doing the expensive search
7452 * for free space, simply return ENOSPC with our max_extent_size so we
7453 * can go ahead and search for a more manageable chunk.
7455 * If our max_extent_size is large enough for our allocation simply
7456 * disable clustering since we will likely not be able to find enough
7457 * space to create a cluster and induce latency trying.
7459 if (unlikely(space_info
->max_extent_size
)) {
7460 spin_lock(&space_info
->lock
);
7461 if (space_info
->max_extent_size
&&
7462 num_bytes
> space_info
->max_extent_size
) {
7463 ins
->offset
= space_info
->max_extent_size
;
7464 spin_unlock(&space_info
->lock
);
7466 } else if (space_info
->max_extent_size
) {
7467 use_cluster
= false;
7469 spin_unlock(&space_info
->lock
);
7472 last_ptr
= fetch_cluster_info(orig_root
, space_info
, &empty_cluster
);
7474 spin_lock(&last_ptr
->lock
);
7475 if (last_ptr
->block_group
)
7476 hint_byte
= last_ptr
->window_start
;
7477 if (last_ptr
->fragmented
) {
7479 * We still set window_start so we can keep track of the
7480 * last place we found an allocation to try and save
7483 hint_byte
= last_ptr
->window_start
;
7484 use_cluster
= false;
7486 spin_unlock(&last_ptr
->lock
);
7489 search_start
= max(search_start
, first_logical_byte(root
, 0));
7490 search_start
= max(search_start
, hint_byte
);
7491 if (search_start
== hint_byte
) {
7492 block_group
= btrfs_lookup_block_group(root
->fs_info
,
7495 * we don't want to use the block group if it doesn't match our
7496 * allocation bits, or if its not cached.
7498 * However if we are re-searching with an ideal block group
7499 * picked out then we don't care that the block group is cached.
7501 if (block_group
&& block_group_bits(block_group
, flags
) &&
7502 block_group
->cached
!= BTRFS_CACHE_NO
) {
7503 down_read(&space_info
->groups_sem
);
7504 if (list_empty(&block_group
->list
) ||
7507 * someone is removing this block group,
7508 * we can't jump into the have_block_group
7509 * target because our list pointers are not
7512 btrfs_put_block_group(block_group
);
7513 up_read(&space_info
->groups_sem
);
7515 index
= get_block_group_index(block_group
);
7516 btrfs_lock_block_group(block_group
, delalloc
);
7517 goto have_block_group
;
7519 } else if (block_group
) {
7520 btrfs_put_block_group(block_group
);
7524 have_caching_bg
= false;
7525 if (index
== 0 || index
== __get_raid_index(flags
))
7527 down_read(&space_info
->groups_sem
);
7528 list_for_each_entry(block_group
, &space_info
->block_groups
[index
],
7533 btrfs_grab_block_group(block_group
, delalloc
);
7534 search_start
= block_group
->key
.objectid
;
7537 * this can happen if we end up cycling through all the
7538 * raid types, but we want to make sure we only allocate
7539 * for the proper type.
7541 if (!block_group_bits(block_group
, flags
)) {
7542 u64 extra
= BTRFS_BLOCK_GROUP_DUP
|
7543 BTRFS_BLOCK_GROUP_RAID1
|
7544 BTRFS_BLOCK_GROUP_RAID5
|
7545 BTRFS_BLOCK_GROUP_RAID6
|
7546 BTRFS_BLOCK_GROUP_RAID10
;
7549 * if they asked for extra copies and this block group
7550 * doesn't provide them, bail. This does allow us to
7551 * fill raid0 from raid1.
7553 if ((flags
& extra
) && !(block_group
->flags
& extra
))
7558 cached
= block_group_cache_done(block_group
);
7559 if (unlikely(!cached
)) {
7560 have_caching_bg
= true;
7561 ret
= cache_block_group(block_group
, 0);
7566 if (unlikely(block_group
->cached
== BTRFS_CACHE_ERROR
))
7568 if (unlikely(block_group
->ro
))
7572 * Ok we want to try and use the cluster allocator, so
7575 if (last_ptr
&& use_cluster
) {
7576 struct btrfs_block_group_cache
*used_block_group
;
7577 unsigned long aligned_cluster
;
7579 * the refill lock keeps out other
7580 * people trying to start a new cluster
7582 used_block_group
= btrfs_lock_cluster(block_group
,
7585 if (!used_block_group
)
7586 goto refill_cluster
;
7588 if (used_block_group
!= block_group
&&
7589 (used_block_group
->ro
||
7590 !block_group_bits(used_block_group
, flags
)))
7591 goto release_cluster
;
7593 offset
= btrfs_alloc_from_cluster(used_block_group
,
7596 used_block_group
->key
.objectid
,
7599 /* we have a block, we're done */
7600 spin_unlock(&last_ptr
->refill_lock
);
7601 trace_btrfs_reserve_extent_cluster(root
,
7603 search_start
, num_bytes
);
7604 if (used_block_group
!= block_group
) {
7605 btrfs_release_block_group(block_group
,
7607 block_group
= used_block_group
;
7612 WARN_ON(last_ptr
->block_group
!= used_block_group
);
7614 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7615 * set up a new clusters, so lets just skip it
7616 * and let the allocator find whatever block
7617 * it can find. If we reach this point, we
7618 * will have tried the cluster allocator
7619 * plenty of times and not have found
7620 * anything, so we are likely way too
7621 * fragmented for the clustering stuff to find
7624 * However, if the cluster is taken from the
7625 * current block group, release the cluster
7626 * first, so that we stand a better chance of
7627 * succeeding in the unclustered
7629 if (loop
>= LOOP_NO_EMPTY_SIZE
&&
7630 used_block_group
!= block_group
) {
7631 spin_unlock(&last_ptr
->refill_lock
);
7632 btrfs_release_block_group(used_block_group
,
7634 goto unclustered_alloc
;
7638 * this cluster didn't work out, free it and
7641 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
7643 if (used_block_group
!= block_group
)
7644 btrfs_release_block_group(used_block_group
,
7647 if (loop
>= LOOP_NO_EMPTY_SIZE
) {
7648 spin_unlock(&last_ptr
->refill_lock
);
7649 goto unclustered_alloc
;
7652 aligned_cluster
= max_t(unsigned long,
7653 empty_cluster
+ empty_size
,
7654 block_group
->full_stripe_len
);
7656 /* allocate a cluster in this block group */
7657 ret
= btrfs_find_space_cluster(root
, block_group
,
7658 last_ptr
, search_start
,
7663 * now pull our allocation out of this
7666 offset
= btrfs_alloc_from_cluster(block_group
,
7672 /* we found one, proceed */
7673 spin_unlock(&last_ptr
->refill_lock
);
7674 trace_btrfs_reserve_extent_cluster(root
,
7675 block_group
, search_start
,
7679 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
7680 && !failed_cluster_refill
) {
7681 spin_unlock(&last_ptr
->refill_lock
);
7683 failed_cluster_refill
= true;
7684 wait_block_group_cache_progress(block_group
,
7685 num_bytes
+ empty_cluster
+ empty_size
);
7686 goto have_block_group
;
7690 * at this point we either didn't find a cluster
7691 * or we weren't able to allocate a block from our
7692 * cluster. Free the cluster we've been trying
7693 * to use, and go to the next block group
7695 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
7696 spin_unlock(&last_ptr
->refill_lock
);
7702 * We are doing an unclustered alloc, set the fragmented flag so
7703 * we don't bother trying to setup a cluster again until we get
7706 if (unlikely(last_ptr
)) {
7707 spin_lock(&last_ptr
->lock
);
7708 last_ptr
->fragmented
= 1;
7709 spin_unlock(&last_ptr
->lock
);
7711 spin_lock(&block_group
->free_space_ctl
->tree_lock
);
7713 block_group
->free_space_ctl
->free_space
<
7714 num_bytes
+ empty_cluster
+ empty_size
) {
7715 if (block_group
->free_space_ctl
->free_space
>
7718 block_group
->free_space_ctl
->free_space
;
7719 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
7722 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
7724 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
7725 num_bytes
, empty_size
,
7728 * If we didn't find a chunk, and we haven't failed on this
7729 * block group before, and this block group is in the middle of
7730 * caching and we are ok with waiting, then go ahead and wait
7731 * for progress to be made, and set failed_alloc to true.
7733 * If failed_alloc is true then we've already waited on this
7734 * block group once and should move on to the next block group.
7736 if (!offset
&& !failed_alloc
&& !cached
&&
7737 loop
> LOOP_CACHING_NOWAIT
) {
7738 wait_block_group_cache_progress(block_group
,
7739 num_bytes
+ empty_size
);
7740 failed_alloc
= true;
7741 goto have_block_group
;
7742 } else if (!offset
) {
7746 search_start
= ALIGN(offset
, root
->stripesize
);
7748 /* move on to the next group */
7749 if (search_start
+ num_bytes
>
7750 block_group
->key
.objectid
+ block_group
->key
.offset
) {
7751 btrfs_add_free_space(block_group
, offset
, num_bytes
);
7755 if (offset
< search_start
)
7756 btrfs_add_free_space(block_group
, offset
,
7757 search_start
- offset
);
7758 BUG_ON(offset
> search_start
);
7760 ret
= btrfs_update_reserved_bytes(block_group
, num_bytes
,
7761 alloc_type
, delalloc
);
7762 if (ret
== -EAGAIN
) {
7763 btrfs_add_free_space(block_group
, offset
, num_bytes
);
7766 btrfs_inc_block_group_reservations(block_group
);
7768 /* we are all good, lets return */
7769 ins
->objectid
= search_start
;
7770 ins
->offset
= num_bytes
;
7772 trace_btrfs_reserve_extent(orig_root
, block_group
,
7773 search_start
, num_bytes
);
7774 btrfs_release_block_group(block_group
, delalloc
);
7777 failed_cluster_refill
= false;
7778 failed_alloc
= false;
7779 BUG_ON(index
!= get_block_group_index(block_group
));
7780 btrfs_release_block_group(block_group
, delalloc
);
7782 up_read(&space_info
->groups_sem
);
7784 if ((loop
== LOOP_CACHING_NOWAIT
) && have_caching_bg
7785 && !orig_have_caching_bg
)
7786 orig_have_caching_bg
= true;
7788 if (!ins
->objectid
&& loop
>= LOOP_CACHING_WAIT
&& have_caching_bg
)
7791 if (!ins
->objectid
&& ++index
< BTRFS_NR_RAID_TYPES
)
7795 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7796 * caching kthreads as we move along
7797 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7798 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7799 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7802 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
) {
7804 if (loop
== LOOP_CACHING_NOWAIT
) {
7806 * We want to skip the LOOP_CACHING_WAIT step if we
7807 * don't have any uncached bgs and we've already done a
7808 * full search through.
7810 if (orig_have_caching_bg
|| !full_search
)
7811 loop
= LOOP_CACHING_WAIT
;
7813 loop
= LOOP_ALLOC_CHUNK
;
7818 if (loop
== LOOP_ALLOC_CHUNK
) {
7819 struct btrfs_trans_handle
*trans
;
7822 trans
= current
->journal_info
;
7826 trans
= btrfs_join_transaction(root
);
7828 if (IS_ERR(trans
)) {
7829 ret
= PTR_ERR(trans
);
7833 ret
= do_chunk_alloc(trans
, root
, flags
,
7837 * If we can't allocate a new chunk we've already looped
7838 * through at least once, move on to the NO_EMPTY_SIZE
7842 loop
= LOOP_NO_EMPTY_SIZE
;
7845 * Do not bail out on ENOSPC since we
7846 * can do more things.
7848 if (ret
< 0 && ret
!= -ENOSPC
)
7849 btrfs_abort_transaction(trans
,
7854 btrfs_end_transaction(trans
, root
);
7859 if (loop
== LOOP_NO_EMPTY_SIZE
) {
7861 * Don't loop again if we already have no empty_size and
7864 if (empty_size
== 0 &&
7865 empty_cluster
== 0) {
7874 } else if (!ins
->objectid
) {
7876 } else if (ins
->objectid
) {
7877 if (!use_cluster
&& last_ptr
) {
7878 spin_lock(&last_ptr
->lock
);
7879 last_ptr
->window_start
= ins
->objectid
;
7880 spin_unlock(&last_ptr
->lock
);
7885 if (ret
== -ENOSPC
) {
7886 spin_lock(&space_info
->lock
);
7887 space_info
->max_extent_size
= max_extent_size
;
7888 spin_unlock(&space_info
->lock
);
7889 ins
->offset
= max_extent_size
;
7894 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
7895 int dump_block_groups
)
7897 struct btrfs_block_group_cache
*cache
;
7900 spin_lock(&info
->lock
);
7901 printk(KERN_INFO
"BTRFS: space_info %llu has %llu free, is %sfull\n",
7903 info
->total_bytes
- info
->bytes_used
- info
->bytes_pinned
-
7904 info
->bytes_reserved
- info
->bytes_readonly
,
7905 (info
->full
) ? "" : "not ");
7906 printk(KERN_INFO
"BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7907 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7908 info
->total_bytes
, info
->bytes_used
, info
->bytes_pinned
,
7909 info
->bytes_reserved
, info
->bytes_may_use
,
7910 info
->bytes_readonly
);
7911 spin_unlock(&info
->lock
);
7913 if (!dump_block_groups
)
7916 down_read(&info
->groups_sem
);
7918 list_for_each_entry(cache
, &info
->block_groups
[index
], list
) {
7919 spin_lock(&cache
->lock
);
7920 printk(KERN_INFO
"BTRFS: "
7921 "block group %llu has %llu bytes, "
7922 "%llu used %llu pinned %llu reserved %s\n",
7923 cache
->key
.objectid
, cache
->key
.offset
,
7924 btrfs_block_group_used(&cache
->item
), cache
->pinned
,
7925 cache
->reserved
, cache
->ro
? "[readonly]" : "");
7926 btrfs_dump_free_space(cache
, bytes
);
7927 spin_unlock(&cache
->lock
);
7929 if (++index
< BTRFS_NR_RAID_TYPES
)
7931 up_read(&info
->groups_sem
);
7934 int btrfs_reserve_extent(struct btrfs_root
*root
,
7935 u64 num_bytes
, u64 min_alloc_size
,
7936 u64 empty_size
, u64 hint_byte
,
7937 struct btrfs_key
*ins
, int is_data
, int delalloc
)
7939 bool final_tried
= num_bytes
== min_alloc_size
;
7943 flags
= btrfs_get_alloc_profile(root
, is_data
);
7945 WARN_ON(num_bytes
< root
->sectorsize
);
7946 ret
= find_free_extent(root
, num_bytes
, empty_size
, hint_byte
, ins
,
7948 if (!ret
&& !is_data
) {
7949 btrfs_dec_block_group_reservations(root
->fs_info
,
7951 } else if (ret
== -ENOSPC
) {
7952 if (!final_tried
&& ins
->offset
) {
7953 num_bytes
= min(num_bytes
>> 1, ins
->offset
);
7954 num_bytes
= round_down(num_bytes
, root
->sectorsize
);
7955 num_bytes
= max(num_bytes
, min_alloc_size
);
7956 if (num_bytes
== min_alloc_size
)
7959 } else if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
7960 struct btrfs_space_info
*sinfo
;
7962 sinfo
= __find_space_info(root
->fs_info
, flags
);
7963 btrfs_err(root
->fs_info
, "allocation failed flags %llu, wanted %llu",
7966 dump_space_info(sinfo
, num_bytes
, 1);
7973 static int __btrfs_free_reserved_extent(struct btrfs_root
*root
,
7975 int pin
, int delalloc
)
7977 struct btrfs_block_group_cache
*cache
;
7980 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
7982 btrfs_err(root
->fs_info
, "Unable to find block group for %llu",
7988 pin_down_extent(root
, cache
, start
, len
, 1);
7990 if (btrfs_test_opt(root
, DISCARD
))
7991 ret
= btrfs_discard_extent(root
, start
, len
, NULL
);
7992 btrfs_add_free_space(cache
, start
, len
);
7993 btrfs_update_reserved_bytes(cache
, len
, RESERVE_FREE
, delalloc
);
7994 trace_btrfs_reserved_extent_free(root
, start
, len
);
7997 btrfs_put_block_group(cache
);
8001 int btrfs_free_reserved_extent(struct btrfs_root
*root
,
8002 u64 start
, u64 len
, int delalloc
)
8004 return __btrfs_free_reserved_extent(root
, start
, len
, 0, delalloc
);
8007 int btrfs_free_and_pin_reserved_extent(struct btrfs_root
*root
,
8010 return __btrfs_free_reserved_extent(root
, start
, len
, 1, 0);
8013 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
8014 struct btrfs_root
*root
,
8015 u64 parent
, u64 root_objectid
,
8016 u64 flags
, u64 owner
, u64 offset
,
8017 struct btrfs_key
*ins
, int ref_mod
)
8020 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8021 struct btrfs_extent_item
*extent_item
;
8022 struct btrfs_extent_inline_ref
*iref
;
8023 struct btrfs_path
*path
;
8024 struct extent_buffer
*leaf
;
8029 type
= BTRFS_SHARED_DATA_REF_KEY
;
8031 type
= BTRFS_EXTENT_DATA_REF_KEY
;
8033 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
8035 path
= btrfs_alloc_path();
8039 path
->leave_spinning
= 1;
8040 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
8043 btrfs_free_path(path
);
8047 leaf
= path
->nodes
[0];
8048 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
8049 struct btrfs_extent_item
);
8050 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
8051 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
8052 btrfs_set_extent_flags(leaf
, extent_item
,
8053 flags
| BTRFS_EXTENT_FLAG_DATA
);
8055 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
8056 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
8058 struct btrfs_shared_data_ref
*ref
;
8059 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
8060 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
8061 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
8063 struct btrfs_extent_data_ref
*ref
;
8064 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
8065 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
8066 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
8067 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
8068 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
8071 btrfs_mark_buffer_dirty(path
->nodes
[0]);
8072 btrfs_free_path(path
);
8074 ret
= remove_from_free_space_tree(trans
, fs_info
, ins
->objectid
,
8079 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1);
8080 if (ret
) { /* -ENOENT, logic error */
8081 btrfs_err(fs_info
, "update block group failed for %llu %llu",
8082 ins
->objectid
, ins
->offset
);
8085 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, ins
->offset
);
8089 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
8090 struct btrfs_root
*root
,
8091 u64 parent
, u64 root_objectid
,
8092 u64 flags
, struct btrfs_disk_key
*key
,
8093 int level
, struct btrfs_key
*ins
)
8096 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8097 struct btrfs_extent_item
*extent_item
;
8098 struct btrfs_tree_block_info
*block_info
;
8099 struct btrfs_extent_inline_ref
*iref
;
8100 struct btrfs_path
*path
;
8101 struct extent_buffer
*leaf
;
8102 u32 size
= sizeof(*extent_item
) + sizeof(*iref
);
8103 u64 num_bytes
= ins
->offset
;
8104 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
8107 if (!skinny_metadata
)
8108 size
+= sizeof(*block_info
);
8110 path
= btrfs_alloc_path();
8112 btrfs_free_and_pin_reserved_extent(root
, ins
->objectid
,
8117 path
->leave_spinning
= 1;
8118 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
8121 btrfs_free_path(path
);
8122 btrfs_free_and_pin_reserved_extent(root
, ins
->objectid
,
8127 leaf
= path
->nodes
[0];
8128 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
8129 struct btrfs_extent_item
);
8130 btrfs_set_extent_refs(leaf
, extent_item
, 1);
8131 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
8132 btrfs_set_extent_flags(leaf
, extent_item
,
8133 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
8135 if (skinny_metadata
) {
8136 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
8137 num_bytes
= root
->nodesize
;
8139 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
8140 btrfs_set_tree_block_key(leaf
, block_info
, key
);
8141 btrfs_set_tree_block_level(leaf
, block_info
, level
);
8142 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
8146 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
8147 btrfs_set_extent_inline_ref_type(leaf
, iref
,
8148 BTRFS_SHARED_BLOCK_REF_KEY
);
8149 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
8151 btrfs_set_extent_inline_ref_type(leaf
, iref
,
8152 BTRFS_TREE_BLOCK_REF_KEY
);
8153 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
8156 btrfs_mark_buffer_dirty(leaf
);
8157 btrfs_free_path(path
);
8159 ret
= remove_from_free_space_tree(trans
, fs_info
, ins
->objectid
,
8164 ret
= update_block_group(trans
, root
, ins
->objectid
, root
->nodesize
,
8166 if (ret
) { /* -ENOENT, logic error */
8167 btrfs_err(fs_info
, "update block group failed for %llu %llu",
8168 ins
->objectid
, ins
->offset
);
8172 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, root
->nodesize
);
8176 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
8177 struct btrfs_root
*root
,
8178 u64 root_objectid
, u64 owner
,
8179 u64 offset
, u64 ram_bytes
,
8180 struct btrfs_key
*ins
)
8184 BUG_ON(root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
8186 ret
= btrfs_add_delayed_data_ref(root
->fs_info
, trans
, ins
->objectid
,
8188 root_objectid
, owner
, offset
,
8189 ram_bytes
, BTRFS_ADD_DELAYED_EXTENT
,
8195 * this is used by the tree logging recovery code. It records that
8196 * an extent has been allocated and makes sure to clear the free
8197 * space cache bits as well
8199 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
8200 struct btrfs_root
*root
,
8201 u64 root_objectid
, u64 owner
, u64 offset
,
8202 struct btrfs_key
*ins
)
8205 struct btrfs_block_group_cache
*block_group
;
8208 * Mixed block groups will exclude before processing the log so we only
8209 * need to do the exclude dance if this fs isn't mixed.
8211 if (!btrfs_fs_incompat(root
->fs_info
, MIXED_GROUPS
)) {
8212 ret
= __exclude_logged_extent(root
, ins
->objectid
, ins
->offset
);
8217 block_group
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
8221 ret
= btrfs_update_reserved_bytes(block_group
, ins
->offset
,
8222 RESERVE_ALLOC_NO_ACCOUNT
, 0);
8223 BUG_ON(ret
); /* logic error */
8224 ret
= alloc_reserved_file_extent(trans
, root
, 0, root_objectid
,
8225 0, owner
, offset
, ins
, 1);
8226 btrfs_put_block_group(block_group
);
8230 static struct extent_buffer
*
8231 btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
8232 u64 bytenr
, int level
)
8234 struct extent_buffer
*buf
;
8236 buf
= btrfs_find_create_tree_block(root
, bytenr
);
8240 btrfs_set_header_generation(buf
, trans
->transid
);
8241 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, buf
, level
);
8242 btrfs_tree_lock(buf
);
8243 clean_tree_block(trans
, root
->fs_info
, buf
);
8244 clear_bit(EXTENT_BUFFER_STALE
, &buf
->bflags
);
8246 btrfs_set_lock_blocking(buf
);
8247 set_extent_buffer_uptodate(buf
);
8249 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
8250 buf
->log_index
= root
->log_transid
% 2;
8252 * we allow two log transactions at a time, use different
8253 * EXENT bit to differentiate dirty pages.
8255 if (buf
->log_index
== 0)
8256 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
8257 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
8259 set_extent_new(&root
->dirty_log_pages
, buf
->start
,
8260 buf
->start
+ buf
->len
- 1);
8262 buf
->log_index
= -1;
8263 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
8264 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
8266 trans
->dirty
= true;
8267 /* this returns a buffer locked for blocking */
8271 static struct btrfs_block_rsv
*
8272 use_block_rsv(struct btrfs_trans_handle
*trans
,
8273 struct btrfs_root
*root
, u32 blocksize
)
8275 struct btrfs_block_rsv
*block_rsv
;
8276 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
8278 bool global_updated
= false;
8280 block_rsv
= get_block_rsv(trans
, root
);
8282 if (unlikely(block_rsv
->size
== 0))
8285 ret
= block_rsv_use_bytes(block_rsv
, blocksize
);
8289 if (block_rsv
->failfast
)
8290 return ERR_PTR(ret
);
8292 if (block_rsv
->type
== BTRFS_BLOCK_RSV_GLOBAL
&& !global_updated
) {
8293 global_updated
= true;
8294 update_global_block_rsv(root
->fs_info
);
8298 if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
8299 static DEFINE_RATELIMIT_STATE(_rs
,
8300 DEFAULT_RATELIMIT_INTERVAL
* 10,
8301 /*DEFAULT_RATELIMIT_BURST*/ 1);
8302 if (__ratelimit(&_rs
))
8304 "BTRFS: block rsv returned %d\n", ret
);
8307 ret
= reserve_metadata_bytes(root
, block_rsv
, blocksize
,
8308 BTRFS_RESERVE_NO_FLUSH
);
8312 * If we couldn't reserve metadata bytes try and use some from
8313 * the global reserve if its space type is the same as the global
8316 if (block_rsv
->type
!= BTRFS_BLOCK_RSV_GLOBAL
&&
8317 block_rsv
->space_info
== global_rsv
->space_info
) {
8318 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
8322 return ERR_PTR(ret
);
8325 static void unuse_block_rsv(struct btrfs_fs_info
*fs_info
,
8326 struct btrfs_block_rsv
*block_rsv
, u32 blocksize
)
8328 block_rsv_add_bytes(block_rsv
, blocksize
, 0);
8329 block_rsv_release_bytes(fs_info
, block_rsv
, NULL
, 0);
8333 * finds a free extent and does all the dirty work required for allocation
8334 * returns the tree buffer or an ERR_PTR on error.
8336 struct extent_buffer
*btrfs_alloc_tree_block(struct btrfs_trans_handle
*trans
,
8337 struct btrfs_root
*root
,
8338 u64 parent
, u64 root_objectid
,
8339 struct btrfs_disk_key
*key
, int level
,
8340 u64 hint
, u64 empty_size
)
8342 struct btrfs_key ins
;
8343 struct btrfs_block_rsv
*block_rsv
;
8344 struct extent_buffer
*buf
;
8345 struct btrfs_delayed_extent_op
*extent_op
;
8348 u32 blocksize
= root
->nodesize
;
8349 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
8352 if (btrfs_test_is_dummy_root(root
)) {
8353 buf
= btrfs_init_new_buffer(trans
, root
, root
->alloc_bytenr
,
8356 root
->alloc_bytenr
+= blocksize
;
8360 block_rsv
= use_block_rsv(trans
, root
, blocksize
);
8361 if (IS_ERR(block_rsv
))
8362 return ERR_CAST(block_rsv
);
8364 ret
= btrfs_reserve_extent(root
, blocksize
, blocksize
,
8365 empty_size
, hint
, &ins
, 0, 0);
8369 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
, level
);
8372 goto out_free_reserved
;
8375 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
8377 parent
= ins
.objectid
;
8378 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
8382 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
8383 extent_op
= btrfs_alloc_delayed_extent_op();
8389 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
8391 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
8392 extent_op
->flags_to_set
= flags
;
8393 extent_op
->update_key
= skinny_metadata
? false : true;
8394 extent_op
->update_flags
= true;
8395 extent_op
->is_data
= false;
8396 extent_op
->level
= level
;
8398 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
8399 ins
.objectid
, ins
.offset
,
8400 parent
, root_objectid
, level
,
8401 BTRFS_ADD_DELAYED_EXTENT
,
8404 goto out_free_delayed
;
8409 btrfs_free_delayed_extent_op(extent_op
);
8411 free_extent_buffer(buf
);
8413 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
, 0);
8415 unuse_block_rsv(root
->fs_info
, block_rsv
, blocksize
);
8416 return ERR_PTR(ret
);
8419 struct walk_control
{
8420 u64 refs
[BTRFS_MAX_LEVEL
];
8421 u64 flags
[BTRFS_MAX_LEVEL
];
8422 struct btrfs_key update_progress
;
8433 #define DROP_REFERENCE 1
8434 #define UPDATE_BACKREF 2
8436 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
8437 struct btrfs_root
*root
,
8438 struct walk_control
*wc
,
8439 struct btrfs_path
*path
)
8447 struct btrfs_key key
;
8448 struct extent_buffer
*eb
;
8453 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
8454 wc
->reada_count
= wc
->reada_count
* 2 / 3;
8455 wc
->reada_count
= max(wc
->reada_count
, 2);
8457 wc
->reada_count
= wc
->reada_count
* 3 / 2;
8458 wc
->reada_count
= min_t(int, wc
->reada_count
,
8459 BTRFS_NODEPTRS_PER_BLOCK(root
));
8462 eb
= path
->nodes
[wc
->level
];
8463 nritems
= btrfs_header_nritems(eb
);
8464 blocksize
= root
->nodesize
;
8466 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
8467 if (nread
>= wc
->reada_count
)
8471 bytenr
= btrfs_node_blockptr(eb
, slot
);
8472 generation
= btrfs_node_ptr_generation(eb
, slot
);
8474 if (slot
== path
->slots
[wc
->level
])
8477 if (wc
->stage
== UPDATE_BACKREF
&&
8478 generation
<= root
->root_key
.offset
)
8481 /* We don't lock the tree block, it's OK to be racy here */
8482 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
,
8483 wc
->level
- 1, 1, &refs
,
8485 /* We don't care about errors in readahead. */
8490 if (wc
->stage
== DROP_REFERENCE
) {
8494 if (wc
->level
== 1 &&
8495 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8497 if (!wc
->update_ref
||
8498 generation
<= root
->root_key
.offset
)
8500 btrfs_node_key_to_cpu(eb
, &key
, slot
);
8501 ret
= btrfs_comp_cpu_keys(&key
,
8502 &wc
->update_progress
);
8506 if (wc
->level
== 1 &&
8507 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8511 readahead_tree_block(root
, bytenr
);
8514 wc
->reada_slot
= slot
;
8518 * These may not be seen by the usual inc/dec ref code so we have to
8521 static int record_one_subtree_extent(struct btrfs_trans_handle
*trans
,
8522 struct btrfs_root
*root
, u64 bytenr
,
8525 struct btrfs_qgroup_extent_record
*qrecord
;
8526 struct btrfs_delayed_ref_root
*delayed_refs
;
8528 qrecord
= kmalloc(sizeof(*qrecord
), GFP_NOFS
);
8532 qrecord
->bytenr
= bytenr
;
8533 qrecord
->num_bytes
= num_bytes
;
8534 qrecord
->old_roots
= NULL
;
8536 delayed_refs
= &trans
->transaction
->delayed_refs
;
8537 spin_lock(&delayed_refs
->lock
);
8538 if (btrfs_qgroup_insert_dirty_extent(delayed_refs
, qrecord
))
8540 spin_unlock(&delayed_refs
->lock
);
8545 static int account_leaf_items(struct btrfs_trans_handle
*trans
,
8546 struct btrfs_root
*root
,
8547 struct extent_buffer
*eb
)
8549 int nr
= btrfs_header_nritems(eb
);
8550 int i
, extent_type
, ret
;
8551 struct btrfs_key key
;
8552 struct btrfs_file_extent_item
*fi
;
8553 u64 bytenr
, num_bytes
;
8555 /* We can be called directly from walk_up_proc() */
8556 if (!root
->fs_info
->quota_enabled
)
8559 for (i
= 0; i
< nr
; i
++) {
8560 btrfs_item_key_to_cpu(eb
, &key
, i
);
8562 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
8565 fi
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
8566 /* filter out non qgroup-accountable extents */
8567 extent_type
= btrfs_file_extent_type(eb
, fi
);
8569 if (extent_type
== BTRFS_FILE_EXTENT_INLINE
)
8572 bytenr
= btrfs_file_extent_disk_bytenr(eb
, fi
);
8576 num_bytes
= btrfs_file_extent_disk_num_bytes(eb
, fi
);
8578 ret
= record_one_subtree_extent(trans
, root
, bytenr
, num_bytes
);
8586 * Walk up the tree from the bottom, freeing leaves and any interior
8587 * nodes which have had all slots visited. If a node (leaf or
8588 * interior) is freed, the node above it will have it's slot
8589 * incremented. The root node will never be freed.
8591 * At the end of this function, we should have a path which has all
8592 * slots incremented to the next position for a search. If we need to
8593 * read a new node it will be NULL and the node above it will have the
8594 * correct slot selected for a later read.
8596 * If we increment the root nodes slot counter past the number of
8597 * elements, 1 is returned to signal completion of the search.
8599 static int adjust_slots_upwards(struct btrfs_root
*root
,
8600 struct btrfs_path
*path
, int root_level
)
8604 struct extent_buffer
*eb
;
8606 if (root_level
== 0)
8609 while (level
<= root_level
) {
8610 eb
= path
->nodes
[level
];
8611 nr
= btrfs_header_nritems(eb
);
8612 path
->slots
[level
]++;
8613 slot
= path
->slots
[level
];
8614 if (slot
>= nr
|| level
== 0) {
8616 * Don't free the root - we will detect this
8617 * condition after our loop and return a
8618 * positive value for caller to stop walking the tree.
8620 if (level
!= root_level
) {
8621 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8622 path
->locks
[level
] = 0;
8624 free_extent_buffer(eb
);
8625 path
->nodes
[level
] = NULL
;
8626 path
->slots
[level
] = 0;
8630 * We have a valid slot to walk back down
8631 * from. Stop here so caller can process these
8640 eb
= path
->nodes
[root_level
];
8641 if (path
->slots
[root_level
] >= btrfs_header_nritems(eb
))
8648 * root_eb is the subtree root and is locked before this function is called.
8650 static int account_shared_subtree(struct btrfs_trans_handle
*trans
,
8651 struct btrfs_root
*root
,
8652 struct extent_buffer
*root_eb
,
8658 struct extent_buffer
*eb
= root_eb
;
8659 struct btrfs_path
*path
= NULL
;
8661 BUG_ON(root_level
< 0 || root_level
> BTRFS_MAX_LEVEL
);
8662 BUG_ON(root_eb
== NULL
);
8664 if (!root
->fs_info
->quota_enabled
)
8667 if (!extent_buffer_uptodate(root_eb
)) {
8668 ret
= btrfs_read_buffer(root_eb
, root_gen
);
8673 if (root_level
== 0) {
8674 ret
= account_leaf_items(trans
, root
, root_eb
);
8678 path
= btrfs_alloc_path();
8683 * Walk down the tree. Missing extent blocks are filled in as
8684 * we go. Metadata is accounted every time we read a new
8687 * When we reach a leaf, we account for file extent items in it,
8688 * walk back up the tree (adjusting slot pointers as we go)
8689 * and restart the search process.
8691 extent_buffer_get(root_eb
); /* For path */
8692 path
->nodes
[root_level
] = root_eb
;
8693 path
->slots
[root_level
] = 0;
8694 path
->locks
[root_level
] = 0; /* so release_path doesn't try to unlock */
8697 while (level
>= 0) {
8698 if (path
->nodes
[level
] == NULL
) {
8703 /* We need to get child blockptr/gen from
8704 * parent before we can read it. */
8705 eb
= path
->nodes
[level
+ 1];
8706 parent_slot
= path
->slots
[level
+ 1];
8707 child_bytenr
= btrfs_node_blockptr(eb
, parent_slot
);
8708 child_gen
= btrfs_node_ptr_generation(eb
, parent_slot
);
8710 eb
= read_tree_block(root
, child_bytenr
, child_gen
);
8714 } else if (!extent_buffer_uptodate(eb
)) {
8715 free_extent_buffer(eb
);
8720 path
->nodes
[level
] = eb
;
8721 path
->slots
[level
] = 0;
8723 btrfs_tree_read_lock(eb
);
8724 btrfs_set_lock_blocking_rw(eb
, BTRFS_READ_LOCK
);
8725 path
->locks
[level
] = BTRFS_READ_LOCK_BLOCKING
;
8727 ret
= record_one_subtree_extent(trans
, root
, child_bytenr
,
8734 ret
= account_leaf_items(trans
, root
, path
->nodes
[level
]);
8738 /* Nonzero return here means we completed our search */
8739 ret
= adjust_slots_upwards(root
, path
, root_level
);
8743 /* Restart search with new slots */
8752 btrfs_free_path(path
);
8758 * helper to process tree block while walking down the tree.
8760 * when wc->stage == UPDATE_BACKREF, this function updates
8761 * back refs for pointers in the block.
8763 * NOTE: return value 1 means we should stop walking down.
8765 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
8766 struct btrfs_root
*root
,
8767 struct btrfs_path
*path
,
8768 struct walk_control
*wc
, int lookup_info
)
8770 int level
= wc
->level
;
8771 struct extent_buffer
*eb
= path
->nodes
[level
];
8772 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
8775 if (wc
->stage
== UPDATE_BACKREF
&&
8776 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
8780 * when reference count of tree block is 1, it won't increase
8781 * again. once full backref flag is set, we never clear it.
8784 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
8785 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
8786 BUG_ON(!path
->locks
[level
]);
8787 ret
= btrfs_lookup_extent_info(trans
, root
,
8788 eb
->start
, level
, 1,
8791 BUG_ON(ret
== -ENOMEM
);
8794 BUG_ON(wc
->refs
[level
] == 0);
8797 if (wc
->stage
== DROP_REFERENCE
) {
8798 if (wc
->refs
[level
] > 1)
8801 if (path
->locks
[level
] && !wc
->keep_locks
) {
8802 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8803 path
->locks
[level
] = 0;
8808 /* wc->stage == UPDATE_BACKREF */
8809 if (!(wc
->flags
[level
] & flag
)) {
8810 BUG_ON(!path
->locks
[level
]);
8811 ret
= btrfs_inc_ref(trans
, root
, eb
, 1);
8812 BUG_ON(ret
); /* -ENOMEM */
8813 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
8814 BUG_ON(ret
); /* -ENOMEM */
8815 ret
= btrfs_set_disk_extent_flags(trans
, root
, eb
->start
,
8817 btrfs_header_level(eb
), 0);
8818 BUG_ON(ret
); /* -ENOMEM */
8819 wc
->flags
[level
] |= flag
;
8823 * the block is shared by multiple trees, so it's not good to
8824 * keep the tree lock
8826 if (path
->locks
[level
] && level
> 0) {
8827 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8828 path
->locks
[level
] = 0;
8834 * helper to process tree block pointer.
8836 * when wc->stage == DROP_REFERENCE, this function checks
8837 * reference count of the block pointed to. if the block
8838 * is shared and we need update back refs for the subtree
8839 * rooted at the block, this function changes wc->stage to
8840 * UPDATE_BACKREF. if the block is shared and there is no
8841 * need to update back, this function drops the reference
8844 * NOTE: return value 1 means we should stop walking down.
8846 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
8847 struct btrfs_root
*root
,
8848 struct btrfs_path
*path
,
8849 struct walk_control
*wc
, int *lookup_info
)
8855 struct btrfs_key key
;
8856 struct extent_buffer
*next
;
8857 int level
= wc
->level
;
8860 bool need_account
= false;
8862 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
8863 path
->slots
[level
]);
8865 * if the lower level block was created before the snapshot
8866 * was created, we know there is no need to update back refs
8869 if (wc
->stage
== UPDATE_BACKREF
&&
8870 generation
<= root
->root_key
.offset
) {
8875 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
8876 blocksize
= root
->nodesize
;
8878 next
= btrfs_find_tree_block(root
->fs_info
, bytenr
);
8880 next
= btrfs_find_create_tree_block(root
, bytenr
);
8882 return PTR_ERR(next
);
8884 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, next
,
8888 btrfs_tree_lock(next
);
8889 btrfs_set_lock_blocking(next
);
8891 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, level
- 1, 1,
8892 &wc
->refs
[level
- 1],
8893 &wc
->flags
[level
- 1]);
8895 btrfs_tree_unlock(next
);
8899 if (unlikely(wc
->refs
[level
- 1] == 0)) {
8900 btrfs_err(root
->fs_info
, "Missing references.");
8905 if (wc
->stage
== DROP_REFERENCE
) {
8906 if (wc
->refs
[level
- 1] > 1) {
8907 need_account
= true;
8909 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8912 if (!wc
->update_ref
||
8913 generation
<= root
->root_key
.offset
)
8916 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
8917 path
->slots
[level
]);
8918 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
8922 wc
->stage
= UPDATE_BACKREF
;
8923 wc
->shared_level
= level
- 1;
8927 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8931 if (!btrfs_buffer_uptodate(next
, generation
, 0)) {
8932 btrfs_tree_unlock(next
);
8933 free_extent_buffer(next
);
8939 if (reada
&& level
== 1)
8940 reada_walk_down(trans
, root
, wc
, path
);
8941 next
= read_tree_block(root
, bytenr
, generation
);
8943 return PTR_ERR(next
);
8944 } else if (!extent_buffer_uptodate(next
)) {
8945 free_extent_buffer(next
);
8948 btrfs_tree_lock(next
);
8949 btrfs_set_lock_blocking(next
);
8953 BUG_ON(level
!= btrfs_header_level(next
));
8954 path
->nodes
[level
] = next
;
8955 path
->slots
[level
] = 0;
8956 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8962 wc
->refs
[level
- 1] = 0;
8963 wc
->flags
[level
- 1] = 0;
8964 if (wc
->stage
== DROP_REFERENCE
) {
8965 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
8966 parent
= path
->nodes
[level
]->start
;
8968 BUG_ON(root
->root_key
.objectid
!=
8969 btrfs_header_owner(path
->nodes
[level
]));
8974 ret
= account_shared_subtree(trans
, root
, next
,
8975 generation
, level
- 1);
8977 btrfs_err_rl(root
->fs_info
,
8979 "%d accounting shared subtree. Quota "
8980 "is out of sync, rescan required.",
8984 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
, parent
,
8985 root
->root_key
.objectid
, level
- 1, 0);
8986 BUG_ON(ret
); /* -ENOMEM */
8988 btrfs_tree_unlock(next
);
8989 free_extent_buffer(next
);
8995 * helper to process tree block while walking up the tree.
8997 * when wc->stage == DROP_REFERENCE, this function drops
8998 * reference count on the block.
9000 * when wc->stage == UPDATE_BACKREF, this function changes
9001 * wc->stage back to DROP_REFERENCE if we changed wc->stage
9002 * to UPDATE_BACKREF previously while processing the block.
9004 * NOTE: return value 1 means we should stop walking up.
9006 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
9007 struct btrfs_root
*root
,
9008 struct btrfs_path
*path
,
9009 struct walk_control
*wc
)
9012 int level
= wc
->level
;
9013 struct extent_buffer
*eb
= path
->nodes
[level
];
9016 if (wc
->stage
== UPDATE_BACKREF
) {
9017 BUG_ON(wc
->shared_level
< level
);
9018 if (level
< wc
->shared_level
)
9021 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
9025 wc
->stage
= DROP_REFERENCE
;
9026 wc
->shared_level
= -1;
9027 path
->slots
[level
] = 0;
9030 * check reference count again if the block isn't locked.
9031 * we should start walking down the tree again if reference
9034 if (!path
->locks
[level
]) {
9036 btrfs_tree_lock(eb
);
9037 btrfs_set_lock_blocking(eb
);
9038 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
9040 ret
= btrfs_lookup_extent_info(trans
, root
,
9041 eb
->start
, level
, 1,
9045 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
9046 path
->locks
[level
] = 0;
9049 BUG_ON(wc
->refs
[level
] == 0);
9050 if (wc
->refs
[level
] == 1) {
9051 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
9052 path
->locks
[level
] = 0;
9058 /* wc->stage == DROP_REFERENCE */
9059 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
9061 if (wc
->refs
[level
] == 1) {
9063 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
9064 ret
= btrfs_dec_ref(trans
, root
, eb
, 1);
9066 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
9067 BUG_ON(ret
); /* -ENOMEM */
9068 ret
= account_leaf_items(trans
, root
, eb
);
9070 btrfs_err_rl(root
->fs_info
,
9072 "%d accounting leaf items. Quota "
9073 "is out of sync, rescan required.",
9077 /* make block locked assertion in clean_tree_block happy */
9078 if (!path
->locks
[level
] &&
9079 btrfs_header_generation(eb
) == trans
->transid
) {
9080 btrfs_tree_lock(eb
);
9081 btrfs_set_lock_blocking(eb
);
9082 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
9084 clean_tree_block(trans
, root
->fs_info
, eb
);
9087 if (eb
== root
->node
) {
9088 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
9091 BUG_ON(root
->root_key
.objectid
!=
9092 btrfs_header_owner(eb
));
9094 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
9095 parent
= path
->nodes
[level
+ 1]->start
;
9097 BUG_ON(root
->root_key
.objectid
!=
9098 btrfs_header_owner(path
->nodes
[level
+ 1]));
9101 btrfs_free_tree_block(trans
, root
, eb
, parent
, wc
->refs
[level
] == 1);
9103 wc
->refs
[level
] = 0;
9104 wc
->flags
[level
] = 0;
9108 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
9109 struct btrfs_root
*root
,
9110 struct btrfs_path
*path
,
9111 struct walk_control
*wc
)
9113 int level
= wc
->level
;
9114 int lookup_info
= 1;
9117 while (level
>= 0) {
9118 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
9125 if (path
->slots
[level
] >=
9126 btrfs_header_nritems(path
->nodes
[level
]))
9129 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
9131 path
->slots
[level
]++;
9140 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
9141 struct btrfs_root
*root
,
9142 struct btrfs_path
*path
,
9143 struct walk_control
*wc
, int max_level
)
9145 int level
= wc
->level
;
9148 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
9149 while (level
< max_level
&& path
->nodes
[level
]) {
9151 if (path
->slots
[level
] + 1 <
9152 btrfs_header_nritems(path
->nodes
[level
])) {
9153 path
->slots
[level
]++;
9156 ret
= walk_up_proc(trans
, root
, path
, wc
);
9160 if (path
->locks
[level
]) {
9161 btrfs_tree_unlock_rw(path
->nodes
[level
],
9162 path
->locks
[level
]);
9163 path
->locks
[level
] = 0;
9165 free_extent_buffer(path
->nodes
[level
]);
9166 path
->nodes
[level
] = NULL
;
9174 * drop a subvolume tree.
9176 * this function traverses the tree freeing any blocks that only
9177 * referenced by the tree.
9179 * when a shared tree block is found. this function decreases its
9180 * reference count by one. if update_ref is true, this function
9181 * also make sure backrefs for the shared block and all lower level
9182 * blocks are properly updated.
9184 * If called with for_reloc == 0, may exit early with -EAGAIN
9186 int btrfs_drop_snapshot(struct btrfs_root
*root
,
9187 struct btrfs_block_rsv
*block_rsv
, int update_ref
,
9190 struct btrfs_path
*path
;
9191 struct btrfs_trans_handle
*trans
;
9192 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
9193 struct btrfs_root_item
*root_item
= &root
->root_item
;
9194 struct walk_control
*wc
;
9195 struct btrfs_key key
;
9199 bool root_dropped
= false;
9201 btrfs_debug(root
->fs_info
, "Drop subvolume %llu", root
->objectid
);
9203 path
= btrfs_alloc_path();
9209 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
9211 btrfs_free_path(path
);
9216 trans
= btrfs_start_transaction(tree_root
, 0);
9217 if (IS_ERR(trans
)) {
9218 err
= PTR_ERR(trans
);
9223 trans
->block_rsv
= block_rsv
;
9225 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
9226 level
= btrfs_header_level(root
->node
);
9227 path
->nodes
[level
] = btrfs_lock_root_node(root
);
9228 btrfs_set_lock_blocking(path
->nodes
[level
]);
9229 path
->slots
[level
] = 0;
9230 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
9231 memset(&wc
->update_progress
, 0,
9232 sizeof(wc
->update_progress
));
9234 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
9235 memcpy(&wc
->update_progress
, &key
,
9236 sizeof(wc
->update_progress
));
9238 level
= root_item
->drop_level
;
9240 path
->lowest_level
= level
;
9241 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
9242 path
->lowest_level
= 0;
9250 * unlock our path, this is safe because only this
9251 * function is allowed to delete this snapshot
9253 btrfs_unlock_up_safe(path
, 0);
9255 level
= btrfs_header_level(root
->node
);
9257 btrfs_tree_lock(path
->nodes
[level
]);
9258 btrfs_set_lock_blocking(path
->nodes
[level
]);
9259 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
9261 ret
= btrfs_lookup_extent_info(trans
, root
,
9262 path
->nodes
[level
]->start
,
9263 level
, 1, &wc
->refs
[level
],
9269 BUG_ON(wc
->refs
[level
] == 0);
9271 if (level
== root_item
->drop_level
)
9274 btrfs_tree_unlock(path
->nodes
[level
]);
9275 path
->locks
[level
] = 0;
9276 WARN_ON(wc
->refs
[level
] != 1);
9282 wc
->shared_level
= -1;
9283 wc
->stage
= DROP_REFERENCE
;
9284 wc
->update_ref
= update_ref
;
9286 wc
->for_reloc
= for_reloc
;
9287 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
9291 ret
= walk_down_tree(trans
, root
, path
, wc
);
9297 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
9304 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
9308 if (wc
->stage
== DROP_REFERENCE
) {
9310 btrfs_node_key(path
->nodes
[level
],
9311 &root_item
->drop_progress
,
9312 path
->slots
[level
]);
9313 root_item
->drop_level
= level
;
9316 BUG_ON(wc
->level
== 0);
9317 if (btrfs_should_end_transaction(trans
, tree_root
) ||
9318 (!for_reloc
&& btrfs_need_cleaner_sleep(root
))) {
9319 ret
= btrfs_update_root(trans
, tree_root
,
9323 btrfs_abort_transaction(trans
, tree_root
, ret
);
9328 btrfs_end_transaction_throttle(trans
, tree_root
);
9329 if (!for_reloc
&& btrfs_need_cleaner_sleep(root
)) {
9330 pr_debug("BTRFS: drop snapshot early exit\n");
9335 trans
= btrfs_start_transaction(tree_root
, 0);
9336 if (IS_ERR(trans
)) {
9337 err
= PTR_ERR(trans
);
9341 trans
->block_rsv
= block_rsv
;
9344 btrfs_release_path(path
);
9348 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
9350 btrfs_abort_transaction(trans
, tree_root
, ret
);
9354 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
9355 ret
= btrfs_find_root(tree_root
, &root
->root_key
, path
,
9358 btrfs_abort_transaction(trans
, tree_root
, ret
);
9361 } else if (ret
> 0) {
9362 /* if we fail to delete the orphan item this time
9363 * around, it'll get picked up the next time.
9365 * The most common failure here is just -ENOENT.
9367 btrfs_del_orphan_item(trans
, tree_root
,
9368 root
->root_key
.objectid
);
9372 if (test_bit(BTRFS_ROOT_IN_RADIX
, &root
->state
)) {
9373 btrfs_add_dropped_root(trans
, root
);
9375 free_extent_buffer(root
->node
);
9376 free_extent_buffer(root
->commit_root
);
9377 btrfs_put_fs_root(root
);
9379 root_dropped
= true;
9381 btrfs_end_transaction_throttle(trans
, tree_root
);
9384 btrfs_free_path(path
);
9387 * So if we need to stop dropping the snapshot for whatever reason we
9388 * need to make sure to add it back to the dead root list so that we
9389 * keep trying to do the work later. This also cleans up roots if we
9390 * don't have it in the radix (like when we recover after a power fail
9391 * or unmount) so we don't leak memory.
9393 if (!for_reloc
&& root_dropped
== false)
9394 btrfs_add_dead_root(root
);
9395 if (err
&& err
!= -EAGAIN
)
9396 btrfs_handle_fs_error(root
->fs_info
, err
, NULL
);
9401 * drop subtree rooted at tree block 'node'.
9403 * NOTE: this function will unlock and release tree block 'node'
9404 * only used by relocation code
9406 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
9407 struct btrfs_root
*root
,
9408 struct extent_buffer
*node
,
9409 struct extent_buffer
*parent
)
9411 struct btrfs_path
*path
;
9412 struct walk_control
*wc
;
9418 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
9420 path
= btrfs_alloc_path();
9424 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
9426 btrfs_free_path(path
);
9430 btrfs_assert_tree_locked(parent
);
9431 parent_level
= btrfs_header_level(parent
);
9432 extent_buffer_get(parent
);
9433 path
->nodes
[parent_level
] = parent
;
9434 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
9436 btrfs_assert_tree_locked(node
);
9437 level
= btrfs_header_level(node
);
9438 path
->nodes
[level
] = node
;
9439 path
->slots
[level
] = 0;
9440 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
9442 wc
->refs
[parent_level
] = 1;
9443 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
9445 wc
->shared_level
= -1;
9446 wc
->stage
= DROP_REFERENCE
;
9450 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
9453 wret
= walk_down_tree(trans
, root
, path
, wc
);
9459 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
9467 btrfs_free_path(path
);
9471 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
9477 * if restripe for this chunk_type is on pick target profile and
9478 * return, otherwise do the usual balance
9480 stripped
= get_restripe_target(root
->fs_info
, flags
);
9482 return extended_to_chunk(stripped
);
9484 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
9486 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
9487 BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
|
9488 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
9490 if (num_devices
== 1) {
9491 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
9492 stripped
= flags
& ~stripped
;
9494 /* turn raid0 into single device chunks */
9495 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
9498 /* turn mirroring into duplication */
9499 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
9500 BTRFS_BLOCK_GROUP_RAID10
))
9501 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
9503 /* they already had raid on here, just return */
9504 if (flags
& stripped
)
9507 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
9508 stripped
= flags
& ~stripped
;
9510 /* switch duplicated blocks with raid1 */
9511 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
9512 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
9514 /* this is drive concat, leave it alone */
9520 static int inc_block_group_ro(struct btrfs_block_group_cache
*cache
, int force
)
9522 struct btrfs_space_info
*sinfo
= cache
->space_info
;
9524 u64 min_allocable_bytes
;
9528 * We need some metadata space and system metadata space for
9529 * allocating chunks in some corner cases until we force to set
9530 * it to be readonly.
9533 (BTRFS_BLOCK_GROUP_SYSTEM
| BTRFS_BLOCK_GROUP_METADATA
)) &&
9535 min_allocable_bytes
= SZ_1M
;
9537 min_allocable_bytes
= 0;
9539 spin_lock(&sinfo
->lock
);
9540 spin_lock(&cache
->lock
);
9548 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
9549 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
9551 if (sinfo
->bytes_used
+ sinfo
->bytes_reserved
+ sinfo
->bytes_pinned
+
9552 sinfo
->bytes_may_use
+ sinfo
->bytes_readonly
+ num_bytes
+
9553 min_allocable_bytes
<= sinfo
->total_bytes
) {
9554 sinfo
->bytes_readonly
+= num_bytes
;
9556 list_add_tail(&cache
->ro_list
, &sinfo
->ro_bgs
);
9560 spin_unlock(&cache
->lock
);
9561 spin_unlock(&sinfo
->lock
);
9565 int btrfs_inc_block_group_ro(struct btrfs_root
*root
,
9566 struct btrfs_block_group_cache
*cache
)
9569 struct btrfs_trans_handle
*trans
;
9574 trans
= btrfs_join_transaction(root
);
9576 return PTR_ERR(trans
);
9579 * we're not allowed to set block groups readonly after the dirty
9580 * block groups cache has started writing. If it already started,
9581 * back off and let this transaction commit
9583 mutex_lock(&root
->fs_info
->ro_block_group_mutex
);
9584 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN
, &trans
->transaction
->flags
)) {
9585 u64 transid
= trans
->transid
;
9587 mutex_unlock(&root
->fs_info
->ro_block_group_mutex
);
9588 btrfs_end_transaction(trans
, root
);
9590 ret
= btrfs_wait_for_commit(root
, transid
);
9597 * if we are changing raid levels, try to allocate a corresponding
9598 * block group with the new raid level.
9600 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
9601 if (alloc_flags
!= cache
->flags
) {
9602 ret
= do_chunk_alloc(trans
, root
, alloc_flags
,
9605 * ENOSPC is allowed here, we may have enough space
9606 * already allocated at the new raid level to
9615 ret
= inc_block_group_ro(cache
, 0);
9618 alloc_flags
= get_alloc_profile(root
, cache
->space_info
->flags
);
9619 ret
= do_chunk_alloc(trans
, root
, alloc_flags
,
9623 ret
= inc_block_group_ro(cache
, 0);
9625 if (cache
->flags
& BTRFS_BLOCK_GROUP_SYSTEM
) {
9626 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
9627 lock_chunks(root
->fs_info
->chunk_root
);
9628 check_system_chunk(trans
, root
, alloc_flags
);
9629 unlock_chunks(root
->fs_info
->chunk_root
);
9631 mutex_unlock(&root
->fs_info
->ro_block_group_mutex
);
9633 btrfs_end_transaction(trans
, root
);
9637 int btrfs_force_chunk_alloc(struct btrfs_trans_handle
*trans
,
9638 struct btrfs_root
*root
, u64 type
)
9640 u64 alloc_flags
= get_alloc_profile(root
, type
);
9641 return do_chunk_alloc(trans
, root
, alloc_flags
,
9646 * helper to account the unused space of all the readonly block group in the
9647 * space_info. takes mirrors into account.
9649 u64
btrfs_account_ro_block_groups_free_space(struct btrfs_space_info
*sinfo
)
9651 struct btrfs_block_group_cache
*block_group
;
9655 /* It's df, we don't care if it's racy */
9656 if (list_empty(&sinfo
->ro_bgs
))
9659 spin_lock(&sinfo
->lock
);
9660 list_for_each_entry(block_group
, &sinfo
->ro_bgs
, ro_list
) {
9661 spin_lock(&block_group
->lock
);
9663 if (!block_group
->ro
) {
9664 spin_unlock(&block_group
->lock
);
9668 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_RAID1
|
9669 BTRFS_BLOCK_GROUP_RAID10
|
9670 BTRFS_BLOCK_GROUP_DUP
))
9675 free_bytes
+= (block_group
->key
.offset
-
9676 btrfs_block_group_used(&block_group
->item
)) *
9679 spin_unlock(&block_group
->lock
);
9681 spin_unlock(&sinfo
->lock
);
9686 void btrfs_dec_block_group_ro(struct btrfs_root
*root
,
9687 struct btrfs_block_group_cache
*cache
)
9689 struct btrfs_space_info
*sinfo
= cache
->space_info
;
9694 spin_lock(&sinfo
->lock
);
9695 spin_lock(&cache
->lock
);
9697 num_bytes
= cache
->key
.offset
- cache
->reserved
-
9698 cache
->pinned
- cache
->bytes_super
-
9699 btrfs_block_group_used(&cache
->item
);
9700 sinfo
->bytes_readonly
-= num_bytes
;
9701 list_del_init(&cache
->ro_list
);
9703 spin_unlock(&cache
->lock
);
9704 spin_unlock(&sinfo
->lock
);
9708 * checks to see if its even possible to relocate this block group.
9710 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9711 * ok to go ahead and try.
9713 int btrfs_can_relocate(struct btrfs_root
*root
, u64 bytenr
)
9715 struct btrfs_block_group_cache
*block_group
;
9716 struct btrfs_space_info
*space_info
;
9717 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
9718 struct btrfs_device
*device
;
9719 struct btrfs_trans_handle
*trans
;
9729 debug
= btrfs_test_opt(root
, ENOSPC_DEBUG
);
9731 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
9733 /* odd, couldn't find the block group, leave it alone */
9736 btrfs_warn(root
->fs_info
,
9737 "can't find block group for bytenr %llu",
9742 min_free
= btrfs_block_group_used(&block_group
->item
);
9744 /* no bytes used, we're good */
9748 space_info
= block_group
->space_info
;
9749 spin_lock(&space_info
->lock
);
9751 full
= space_info
->full
;
9754 * if this is the last block group we have in this space, we can't
9755 * relocate it unless we're able to allocate a new chunk below.
9757 * Otherwise, we need to make sure we have room in the space to handle
9758 * all of the extents from this block group. If we can, we're good
9760 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
9761 (space_info
->bytes_used
+ space_info
->bytes_reserved
+
9762 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
9763 min_free
< space_info
->total_bytes
)) {
9764 spin_unlock(&space_info
->lock
);
9767 spin_unlock(&space_info
->lock
);
9770 * ok we don't have enough space, but maybe we have free space on our
9771 * devices to allocate new chunks for relocation, so loop through our
9772 * alloc devices and guess if we have enough space. if this block
9773 * group is going to be restriped, run checks against the target
9774 * profile instead of the current one.
9786 target
= get_restripe_target(root
->fs_info
, block_group
->flags
);
9788 index
= __get_raid_index(extended_to_chunk(target
));
9791 * this is just a balance, so if we were marked as full
9792 * we know there is no space for a new chunk
9796 btrfs_warn(root
->fs_info
,
9797 "no space to alloc new chunk for block group %llu",
9798 block_group
->key
.objectid
);
9802 index
= get_block_group_index(block_group
);
9805 if (index
== BTRFS_RAID_RAID10
) {
9809 } else if (index
== BTRFS_RAID_RAID1
) {
9811 } else if (index
== BTRFS_RAID_DUP
) {
9814 } else if (index
== BTRFS_RAID_RAID0
) {
9815 dev_min
= fs_devices
->rw_devices
;
9816 min_free
= div64_u64(min_free
, dev_min
);
9819 /* We need to do this so that we can look at pending chunks */
9820 trans
= btrfs_join_transaction(root
);
9821 if (IS_ERR(trans
)) {
9822 ret
= PTR_ERR(trans
);
9826 mutex_lock(&root
->fs_info
->chunk_mutex
);
9827 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
9831 * check to make sure we can actually find a chunk with enough
9832 * space to fit our block group in.
9834 if (device
->total_bytes
> device
->bytes_used
+ min_free
&&
9835 !device
->is_tgtdev_for_dev_replace
) {
9836 ret
= find_free_dev_extent(trans
, device
, min_free
,
9841 if (dev_nr
>= dev_min
)
9847 if (debug
&& ret
== -1)
9848 btrfs_warn(root
->fs_info
,
9849 "no space to allocate a new chunk for block group %llu",
9850 block_group
->key
.objectid
);
9851 mutex_unlock(&root
->fs_info
->chunk_mutex
);
9852 btrfs_end_transaction(trans
, root
);
9854 btrfs_put_block_group(block_group
);
9858 static int find_first_block_group(struct btrfs_root
*root
,
9859 struct btrfs_path
*path
, struct btrfs_key
*key
)
9862 struct btrfs_key found_key
;
9863 struct extent_buffer
*leaf
;
9866 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
9871 slot
= path
->slots
[0];
9872 leaf
= path
->nodes
[0];
9873 if (slot
>= btrfs_header_nritems(leaf
)) {
9874 ret
= btrfs_next_leaf(root
, path
);
9881 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
9883 if (found_key
.objectid
>= key
->objectid
&&
9884 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
9894 void btrfs_put_block_group_cache(struct btrfs_fs_info
*info
)
9896 struct btrfs_block_group_cache
*block_group
;
9900 struct inode
*inode
;
9902 block_group
= btrfs_lookup_first_block_group(info
, last
);
9903 while (block_group
) {
9904 spin_lock(&block_group
->lock
);
9905 if (block_group
->iref
)
9907 spin_unlock(&block_group
->lock
);
9908 block_group
= next_block_group(info
->tree_root
,
9918 inode
= block_group
->inode
;
9919 block_group
->iref
= 0;
9920 block_group
->inode
= NULL
;
9921 spin_unlock(&block_group
->lock
);
9923 last
= block_group
->key
.objectid
+ block_group
->key
.offset
;
9924 btrfs_put_block_group(block_group
);
9928 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
9930 struct btrfs_block_group_cache
*block_group
;
9931 struct btrfs_space_info
*space_info
;
9932 struct btrfs_caching_control
*caching_ctl
;
9935 down_write(&info
->commit_root_sem
);
9936 while (!list_empty(&info
->caching_block_groups
)) {
9937 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
9938 struct btrfs_caching_control
, list
);
9939 list_del(&caching_ctl
->list
);
9940 put_caching_control(caching_ctl
);
9942 up_write(&info
->commit_root_sem
);
9944 spin_lock(&info
->unused_bgs_lock
);
9945 while (!list_empty(&info
->unused_bgs
)) {
9946 block_group
= list_first_entry(&info
->unused_bgs
,
9947 struct btrfs_block_group_cache
,
9949 list_del_init(&block_group
->bg_list
);
9950 btrfs_put_block_group(block_group
);
9952 spin_unlock(&info
->unused_bgs_lock
);
9954 spin_lock(&info
->block_group_cache_lock
);
9955 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
9956 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
9958 rb_erase(&block_group
->cache_node
,
9959 &info
->block_group_cache_tree
);
9960 RB_CLEAR_NODE(&block_group
->cache_node
);
9961 spin_unlock(&info
->block_group_cache_lock
);
9963 down_write(&block_group
->space_info
->groups_sem
);
9964 list_del(&block_group
->list
);
9965 up_write(&block_group
->space_info
->groups_sem
);
9967 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
9968 wait_block_group_cache_done(block_group
);
9971 * We haven't cached this block group, which means we could
9972 * possibly have excluded extents on this block group.
9974 if (block_group
->cached
== BTRFS_CACHE_NO
||
9975 block_group
->cached
== BTRFS_CACHE_ERROR
)
9976 free_excluded_extents(info
->extent_root
, block_group
);
9978 btrfs_remove_free_space_cache(block_group
);
9979 btrfs_put_block_group(block_group
);
9981 spin_lock(&info
->block_group_cache_lock
);
9983 spin_unlock(&info
->block_group_cache_lock
);
9985 /* now that all the block groups are freed, go through and
9986 * free all the space_info structs. This is only called during
9987 * the final stages of unmount, and so we know nobody is
9988 * using them. We call synchronize_rcu() once before we start,
9989 * just to be on the safe side.
9993 release_global_block_rsv(info
);
9995 while (!list_empty(&info
->space_info
)) {
9998 space_info
= list_entry(info
->space_info
.next
,
9999 struct btrfs_space_info
,
10003 * Do not hide this behind enospc_debug, this is actually
10004 * important and indicates a real bug if this happens.
10006 if (WARN_ON(space_info
->bytes_pinned
> 0 ||
10007 space_info
->bytes_reserved
> 0 ||
10008 space_info
->bytes_may_use
> 0))
10009 dump_space_info(space_info
, 0, 0);
10010 list_del(&space_info
->list
);
10011 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++) {
10012 struct kobject
*kobj
;
10013 kobj
= space_info
->block_group_kobjs
[i
];
10014 space_info
->block_group_kobjs
[i
] = NULL
;
10020 kobject_del(&space_info
->kobj
);
10021 kobject_put(&space_info
->kobj
);
10026 static void __link_block_group(struct btrfs_space_info
*space_info
,
10027 struct btrfs_block_group_cache
*cache
)
10029 int index
= get_block_group_index(cache
);
10030 bool first
= false;
10032 down_write(&space_info
->groups_sem
);
10033 if (list_empty(&space_info
->block_groups
[index
]))
10035 list_add_tail(&cache
->list
, &space_info
->block_groups
[index
]);
10036 up_write(&space_info
->groups_sem
);
10039 struct raid_kobject
*rkobj
;
10042 rkobj
= kzalloc(sizeof(*rkobj
), GFP_NOFS
);
10045 rkobj
->raid_type
= index
;
10046 kobject_init(&rkobj
->kobj
, &btrfs_raid_ktype
);
10047 ret
= kobject_add(&rkobj
->kobj
, &space_info
->kobj
,
10048 "%s", get_raid_name(index
));
10050 kobject_put(&rkobj
->kobj
);
10053 space_info
->block_group_kobjs
[index
] = &rkobj
->kobj
;
10058 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
10061 static struct btrfs_block_group_cache
*
10062 btrfs_create_block_group_cache(struct btrfs_root
*root
, u64 start
, u64 size
)
10064 struct btrfs_block_group_cache
*cache
;
10066 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
10070 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
10072 if (!cache
->free_space_ctl
) {
10077 cache
->key
.objectid
= start
;
10078 cache
->key
.offset
= size
;
10079 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
10081 cache
->sectorsize
= root
->sectorsize
;
10082 cache
->fs_info
= root
->fs_info
;
10083 cache
->full_stripe_len
= btrfs_full_stripe_len(root
,
10084 &root
->fs_info
->mapping_tree
,
10086 set_free_space_tree_thresholds(cache
);
10088 atomic_set(&cache
->count
, 1);
10089 spin_lock_init(&cache
->lock
);
10090 init_rwsem(&cache
->data_rwsem
);
10091 INIT_LIST_HEAD(&cache
->list
);
10092 INIT_LIST_HEAD(&cache
->cluster_list
);
10093 INIT_LIST_HEAD(&cache
->bg_list
);
10094 INIT_LIST_HEAD(&cache
->ro_list
);
10095 INIT_LIST_HEAD(&cache
->dirty_list
);
10096 INIT_LIST_HEAD(&cache
->io_list
);
10097 btrfs_init_free_space_ctl(cache
);
10098 atomic_set(&cache
->trimming
, 0);
10099 mutex_init(&cache
->free_space_lock
);
10104 int btrfs_read_block_groups(struct btrfs_root
*root
)
10106 struct btrfs_path
*path
;
10108 struct btrfs_block_group_cache
*cache
;
10109 struct btrfs_fs_info
*info
= root
->fs_info
;
10110 struct btrfs_space_info
*space_info
;
10111 struct btrfs_key key
;
10112 struct btrfs_key found_key
;
10113 struct extent_buffer
*leaf
;
10114 int need_clear
= 0;
10117 root
= info
->extent_root
;
10120 key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
10121 path
= btrfs_alloc_path();
10124 path
->reada
= READA_FORWARD
;
10126 cache_gen
= btrfs_super_cache_generation(root
->fs_info
->super_copy
);
10127 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
10128 btrfs_super_generation(root
->fs_info
->super_copy
) != cache_gen
)
10130 if (btrfs_test_opt(root
, CLEAR_CACHE
))
10134 ret
= find_first_block_group(root
, path
, &key
);
10140 leaf
= path
->nodes
[0];
10141 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
10143 cache
= btrfs_create_block_group_cache(root
, found_key
.objectid
,
10152 * When we mount with old space cache, we need to
10153 * set BTRFS_DC_CLEAR and set dirty flag.
10155 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
10156 * truncate the old free space cache inode and
10158 * b) Setting 'dirty flag' makes sure that we flush
10159 * the new space cache info onto disk.
10161 if (btrfs_test_opt(root
, SPACE_CACHE
))
10162 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
10165 read_extent_buffer(leaf
, &cache
->item
,
10166 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
10167 sizeof(cache
->item
));
10168 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
10170 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
10171 btrfs_release_path(path
);
10174 * We need to exclude the super stripes now so that the space
10175 * info has super bytes accounted for, otherwise we'll think
10176 * we have more space than we actually do.
10178 ret
= exclude_super_stripes(root
, cache
);
10181 * We may have excluded something, so call this just in
10184 free_excluded_extents(root
, cache
);
10185 btrfs_put_block_group(cache
);
10190 * check for two cases, either we are full, and therefore
10191 * don't need to bother with the caching work since we won't
10192 * find any space, or we are empty, and we can just add all
10193 * the space in and be done with it. This saves us _alot_ of
10194 * time, particularly in the full case.
10196 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
10197 cache
->last_byte_to_unpin
= (u64
)-1;
10198 cache
->cached
= BTRFS_CACHE_FINISHED
;
10199 free_excluded_extents(root
, cache
);
10200 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
10201 cache
->last_byte_to_unpin
= (u64
)-1;
10202 cache
->cached
= BTRFS_CACHE_FINISHED
;
10203 add_new_free_space(cache
, root
->fs_info
,
10204 found_key
.objectid
,
10205 found_key
.objectid
+
10207 free_excluded_extents(root
, cache
);
10210 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
10212 btrfs_remove_free_space_cache(cache
);
10213 btrfs_put_block_group(cache
);
10217 trace_btrfs_add_block_group(root
->fs_info
, cache
, 0);
10218 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
10219 btrfs_block_group_used(&cache
->item
),
10220 cache
->bytes_super
, &space_info
);
10222 btrfs_remove_free_space_cache(cache
);
10223 spin_lock(&info
->block_group_cache_lock
);
10224 rb_erase(&cache
->cache_node
,
10225 &info
->block_group_cache_tree
);
10226 RB_CLEAR_NODE(&cache
->cache_node
);
10227 spin_unlock(&info
->block_group_cache_lock
);
10228 btrfs_put_block_group(cache
);
10232 cache
->space_info
= space_info
;
10234 __link_block_group(space_info
, cache
);
10236 set_avail_alloc_bits(root
->fs_info
, cache
->flags
);
10237 if (btrfs_chunk_readonly(root
, cache
->key
.objectid
)) {
10238 inc_block_group_ro(cache
, 1);
10239 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
10240 spin_lock(&info
->unused_bgs_lock
);
10241 /* Should always be true but just in case. */
10242 if (list_empty(&cache
->bg_list
)) {
10243 btrfs_get_block_group(cache
);
10244 list_add_tail(&cache
->bg_list
,
10245 &info
->unused_bgs
);
10247 spin_unlock(&info
->unused_bgs_lock
);
10251 list_for_each_entry_rcu(space_info
, &root
->fs_info
->space_info
, list
) {
10252 if (!(get_alloc_profile(root
, space_info
->flags
) &
10253 (BTRFS_BLOCK_GROUP_RAID10
|
10254 BTRFS_BLOCK_GROUP_RAID1
|
10255 BTRFS_BLOCK_GROUP_RAID5
|
10256 BTRFS_BLOCK_GROUP_RAID6
|
10257 BTRFS_BLOCK_GROUP_DUP
)))
10260 * avoid allocating from un-mirrored block group if there are
10261 * mirrored block groups.
10263 list_for_each_entry(cache
,
10264 &space_info
->block_groups
[BTRFS_RAID_RAID0
],
10266 inc_block_group_ro(cache
, 1);
10267 list_for_each_entry(cache
,
10268 &space_info
->block_groups
[BTRFS_RAID_SINGLE
],
10270 inc_block_group_ro(cache
, 1);
10273 init_global_block_rsv(info
);
10276 btrfs_free_path(path
);
10280 void btrfs_create_pending_block_groups(struct btrfs_trans_handle
*trans
,
10281 struct btrfs_root
*root
)
10283 struct btrfs_block_group_cache
*block_group
, *tmp
;
10284 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
10285 struct btrfs_block_group_item item
;
10286 struct btrfs_key key
;
10288 bool can_flush_pending_bgs
= trans
->can_flush_pending_bgs
;
10290 trans
->can_flush_pending_bgs
= false;
10291 list_for_each_entry_safe(block_group
, tmp
, &trans
->new_bgs
, bg_list
) {
10295 spin_lock(&block_group
->lock
);
10296 memcpy(&item
, &block_group
->item
, sizeof(item
));
10297 memcpy(&key
, &block_group
->key
, sizeof(key
));
10298 spin_unlock(&block_group
->lock
);
10300 ret
= btrfs_insert_item(trans
, extent_root
, &key
, &item
,
10303 btrfs_abort_transaction(trans
, extent_root
, ret
);
10304 ret
= btrfs_finish_chunk_alloc(trans
, extent_root
,
10305 key
.objectid
, key
.offset
);
10307 btrfs_abort_transaction(trans
, extent_root
, ret
);
10308 add_block_group_free_space(trans
, root
->fs_info
, block_group
);
10309 /* already aborted the transaction if it failed. */
10311 list_del_init(&block_group
->bg_list
);
10313 trans
->can_flush_pending_bgs
= can_flush_pending_bgs
;
10316 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
10317 struct btrfs_root
*root
, u64 bytes_used
,
10318 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
10322 struct btrfs_root
*extent_root
;
10323 struct btrfs_block_group_cache
*cache
;
10324 extent_root
= root
->fs_info
->extent_root
;
10326 btrfs_set_log_full_commit(root
->fs_info
, trans
);
10328 cache
= btrfs_create_block_group_cache(root
, chunk_offset
, size
);
10332 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
10333 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
10334 btrfs_set_block_group_flags(&cache
->item
, type
);
10336 cache
->flags
= type
;
10337 cache
->last_byte_to_unpin
= (u64
)-1;
10338 cache
->cached
= BTRFS_CACHE_FINISHED
;
10339 cache
->needs_free_space
= 1;
10340 ret
= exclude_super_stripes(root
, cache
);
10343 * We may have excluded something, so call this just in
10346 free_excluded_extents(root
, cache
);
10347 btrfs_put_block_group(cache
);
10351 add_new_free_space(cache
, root
->fs_info
, chunk_offset
,
10352 chunk_offset
+ size
);
10354 free_excluded_extents(root
, cache
);
10356 #ifdef CONFIG_BTRFS_DEBUG
10357 if (btrfs_should_fragment_free_space(root
, cache
)) {
10358 u64 new_bytes_used
= size
- bytes_used
;
10360 bytes_used
+= new_bytes_used
>> 1;
10361 fragment_free_space(root
, cache
);
10365 * Call to ensure the corresponding space_info object is created and
10366 * assigned to our block group, but don't update its counters just yet.
10367 * We want our bg to be added to the rbtree with its ->space_info set.
10369 ret
= update_space_info(root
->fs_info
, cache
->flags
, 0, 0, 0,
10370 &cache
->space_info
);
10372 btrfs_remove_free_space_cache(cache
);
10373 btrfs_put_block_group(cache
);
10377 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
10379 btrfs_remove_free_space_cache(cache
);
10380 btrfs_put_block_group(cache
);
10385 * Now that our block group has its ->space_info set and is inserted in
10386 * the rbtree, update the space info's counters.
10388 trace_btrfs_add_block_group(root
->fs_info
, cache
, 1);
10389 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
10390 cache
->bytes_super
, &cache
->space_info
);
10392 btrfs_remove_free_space_cache(cache
);
10393 spin_lock(&root
->fs_info
->block_group_cache_lock
);
10394 rb_erase(&cache
->cache_node
,
10395 &root
->fs_info
->block_group_cache_tree
);
10396 RB_CLEAR_NODE(&cache
->cache_node
);
10397 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
10398 btrfs_put_block_group(cache
);
10401 update_global_block_rsv(root
->fs_info
);
10403 __link_block_group(cache
->space_info
, cache
);
10405 list_add_tail(&cache
->bg_list
, &trans
->new_bgs
);
10407 set_avail_alloc_bits(extent_root
->fs_info
, type
);
10411 static void clear_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
10413 u64 extra_flags
= chunk_to_extended(flags
) &
10414 BTRFS_EXTENDED_PROFILE_MASK
;
10416 write_seqlock(&fs_info
->profiles_lock
);
10417 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
10418 fs_info
->avail_data_alloc_bits
&= ~extra_flags
;
10419 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
10420 fs_info
->avail_metadata_alloc_bits
&= ~extra_flags
;
10421 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
10422 fs_info
->avail_system_alloc_bits
&= ~extra_flags
;
10423 write_sequnlock(&fs_info
->profiles_lock
);
10426 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
10427 struct btrfs_root
*root
, u64 group_start
,
10428 struct extent_map
*em
)
10430 struct btrfs_path
*path
;
10431 struct btrfs_block_group_cache
*block_group
;
10432 struct btrfs_free_cluster
*cluster
;
10433 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
10434 struct btrfs_key key
;
10435 struct inode
*inode
;
10436 struct kobject
*kobj
= NULL
;
10440 struct btrfs_caching_control
*caching_ctl
= NULL
;
10443 root
= root
->fs_info
->extent_root
;
10445 block_group
= btrfs_lookup_block_group(root
->fs_info
, group_start
);
10446 BUG_ON(!block_group
);
10447 BUG_ON(!block_group
->ro
);
10450 * Free the reserved super bytes from this block group before
10453 free_excluded_extents(root
, block_group
);
10455 memcpy(&key
, &block_group
->key
, sizeof(key
));
10456 index
= get_block_group_index(block_group
);
10457 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
10458 BTRFS_BLOCK_GROUP_RAID1
|
10459 BTRFS_BLOCK_GROUP_RAID10
))
10464 /* make sure this block group isn't part of an allocation cluster */
10465 cluster
= &root
->fs_info
->data_alloc_cluster
;
10466 spin_lock(&cluster
->refill_lock
);
10467 btrfs_return_cluster_to_free_space(block_group
, cluster
);
10468 spin_unlock(&cluster
->refill_lock
);
10471 * make sure this block group isn't part of a metadata
10472 * allocation cluster
10474 cluster
= &root
->fs_info
->meta_alloc_cluster
;
10475 spin_lock(&cluster
->refill_lock
);
10476 btrfs_return_cluster_to_free_space(block_group
, cluster
);
10477 spin_unlock(&cluster
->refill_lock
);
10479 path
= btrfs_alloc_path();
10486 * get the inode first so any iput calls done for the io_list
10487 * aren't the final iput (no unlinks allowed now)
10489 inode
= lookup_free_space_inode(tree_root
, block_group
, path
);
10491 mutex_lock(&trans
->transaction
->cache_write_mutex
);
10493 * make sure our free spache cache IO is done before remove the
10496 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
10497 if (!list_empty(&block_group
->io_list
)) {
10498 list_del_init(&block_group
->io_list
);
10500 WARN_ON(!IS_ERR(inode
) && inode
!= block_group
->io_ctl
.inode
);
10502 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
10503 btrfs_wait_cache_io(root
, trans
, block_group
,
10504 &block_group
->io_ctl
, path
,
10505 block_group
->key
.objectid
);
10506 btrfs_put_block_group(block_group
);
10507 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
10510 if (!list_empty(&block_group
->dirty_list
)) {
10511 list_del_init(&block_group
->dirty_list
);
10512 btrfs_put_block_group(block_group
);
10514 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
10515 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
10517 if (!IS_ERR(inode
)) {
10518 ret
= btrfs_orphan_add(trans
, inode
);
10520 btrfs_add_delayed_iput(inode
);
10523 clear_nlink(inode
);
10524 /* One for the block groups ref */
10525 spin_lock(&block_group
->lock
);
10526 if (block_group
->iref
) {
10527 block_group
->iref
= 0;
10528 block_group
->inode
= NULL
;
10529 spin_unlock(&block_group
->lock
);
10532 spin_unlock(&block_group
->lock
);
10534 /* One for our lookup ref */
10535 btrfs_add_delayed_iput(inode
);
10538 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
10539 key
.offset
= block_group
->key
.objectid
;
10542 ret
= btrfs_search_slot(trans
, tree_root
, &key
, path
, -1, 1);
10546 btrfs_release_path(path
);
10548 ret
= btrfs_del_item(trans
, tree_root
, path
);
10551 btrfs_release_path(path
);
10554 spin_lock(&root
->fs_info
->block_group_cache_lock
);
10555 rb_erase(&block_group
->cache_node
,
10556 &root
->fs_info
->block_group_cache_tree
);
10557 RB_CLEAR_NODE(&block_group
->cache_node
);
10559 if (root
->fs_info
->first_logical_byte
== block_group
->key
.objectid
)
10560 root
->fs_info
->first_logical_byte
= (u64
)-1;
10561 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
10563 down_write(&block_group
->space_info
->groups_sem
);
10565 * we must use list_del_init so people can check to see if they
10566 * are still on the list after taking the semaphore
10568 list_del_init(&block_group
->list
);
10569 if (list_empty(&block_group
->space_info
->block_groups
[index
])) {
10570 kobj
= block_group
->space_info
->block_group_kobjs
[index
];
10571 block_group
->space_info
->block_group_kobjs
[index
] = NULL
;
10572 clear_avail_alloc_bits(root
->fs_info
, block_group
->flags
);
10574 up_write(&block_group
->space_info
->groups_sem
);
10580 if (block_group
->has_caching_ctl
)
10581 caching_ctl
= get_caching_control(block_group
);
10582 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
10583 wait_block_group_cache_done(block_group
);
10584 if (block_group
->has_caching_ctl
) {
10585 down_write(&root
->fs_info
->commit_root_sem
);
10586 if (!caching_ctl
) {
10587 struct btrfs_caching_control
*ctl
;
10589 list_for_each_entry(ctl
,
10590 &root
->fs_info
->caching_block_groups
, list
)
10591 if (ctl
->block_group
== block_group
) {
10593 atomic_inc(&caching_ctl
->count
);
10598 list_del_init(&caching_ctl
->list
);
10599 up_write(&root
->fs_info
->commit_root_sem
);
10601 /* Once for the caching bgs list and once for us. */
10602 put_caching_control(caching_ctl
);
10603 put_caching_control(caching_ctl
);
10607 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
10608 if (!list_empty(&block_group
->dirty_list
)) {
10611 if (!list_empty(&block_group
->io_list
)) {
10614 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
10615 btrfs_remove_free_space_cache(block_group
);
10617 spin_lock(&block_group
->space_info
->lock
);
10618 list_del_init(&block_group
->ro_list
);
10620 if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
10621 WARN_ON(block_group
->space_info
->total_bytes
10622 < block_group
->key
.offset
);
10623 WARN_ON(block_group
->space_info
->bytes_readonly
10624 < block_group
->key
.offset
);
10625 WARN_ON(block_group
->space_info
->disk_total
10626 < block_group
->key
.offset
* factor
);
10628 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
10629 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
10630 block_group
->space_info
->disk_total
-= block_group
->key
.offset
* factor
;
10632 spin_unlock(&block_group
->space_info
->lock
);
10634 memcpy(&key
, &block_group
->key
, sizeof(key
));
10637 if (!list_empty(&em
->list
)) {
10638 /* We're in the transaction->pending_chunks list. */
10639 free_extent_map(em
);
10641 spin_lock(&block_group
->lock
);
10642 block_group
->removed
= 1;
10644 * At this point trimming can't start on this block group, because we
10645 * removed the block group from the tree fs_info->block_group_cache_tree
10646 * so no one can't find it anymore and even if someone already got this
10647 * block group before we removed it from the rbtree, they have already
10648 * incremented block_group->trimming - if they didn't, they won't find
10649 * any free space entries because we already removed them all when we
10650 * called btrfs_remove_free_space_cache().
10652 * And we must not remove the extent map from the fs_info->mapping_tree
10653 * to prevent the same logical address range and physical device space
10654 * ranges from being reused for a new block group. This is because our
10655 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10656 * completely transactionless, so while it is trimming a range the
10657 * currently running transaction might finish and a new one start,
10658 * allowing for new block groups to be created that can reuse the same
10659 * physical device locations unless we take this special care.
10661 * There may also be an implicit trim operation if the file system
10662 * is mounted with -odiscard. The same protections must remain
10663 * in place until the extents have been discarded completely when
10664 * the transaction commit has completed.
10666 remove_em
= (atomic_read(&block_group
->trimming
) == 0);
10668 * Make sure a trimmer task always sees the em in the pinned_chunks list
10669 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10670 * before checking block_group->removed).
10674 * Our em might be in trans->transaction->pending_chunks which
10675 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10676 * and so is the fs_info->pinned_chunks list.
10678 * So at this point we must be holding the chunk_mutex to avoid
10679 * any races with chunk allocation (more specifically at
10680 * volumes.c:contains_pending_extent()), to ensure it always
10681 * sees the em, either in the pending_chunks list or in the
10682 * pinned_chunks list.
10684 list_move_tail(&em
->list
, &root
->fs_info
->pinned_chunks
);
10686 spin_unlock(&block_group
->lock
);
10689 struct extent_map_tree
*em_tree
;
10691 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
10692 write_lock(&em_tree
->lock
);
10694 * The em might be in the pending_chunks list, so make sure the
10695 * chunk mutex is locked, since remove_extent_mapping() will
10696 * delete us from that list.
10698 remove_extent_mapping(em_tree
, em
);
10699 write_unlock(&em_tree
->lock
);
10700 /* once for the tree */
10701 free_extent_map(em
);
10704 unlock_chunks(root
);
10706 ret
= remove_block_group_free_space(trans
, root
->fs_info
, block_group
);
10710 btrfs_put_block_group(block_group
);
10711 btrfs_put_block_group(block_group
);
10713 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
10719 ret
= btrfs_del_item(trans
, root
, path
);
10721 btrfs_free_path(path
);
10725 struct btrfs_trans_handle
*
10726 btrfs_start_trans_remove_block_group(struct btrfs_fs_info
*fs_info
,
10727 const u64 chunk_offset
)
10729 struct extent_map_tree
*em_tree
= &fs_info
->mapping_tree
.map_tree
;
10730 struct extent_map
*em
;
10731 struct map_lookup
*map
;
10732 unsigned int num_items
;
10734 read_lock(&em_tree
->lock
);
10735 em
= lookup_extent_mapping(em_tree
, chunk_offset
, 1);
10736 read_unlock(&em_tree
->lock
);
10737 ASSERT(em
&& em
->start
== chunk_offset
);
10740 * We need to reserve 3 + N units from the metadata space info in order
10741 * to remove a block group (done at btrfs_remove_chunk() and at
10742 * btrfs_remove_block_group()), which are used for:
10744 * 1 unit for adding the free space inode's orphan (located in the tree
10746 * 1 unit for deleting the block group item (located in the extent
10748 * 1 unit for deleting the free space item (located in tree of tree
10750 * N units for deleting N device extent items corresponding to each
10751 * stripe (located in the device tree).
10753 * In order to remove a block group we also need to reserve units in the
10754 * system space info in order to update the chunk tree (update one or
10755 * more device items and remove one chunk item), but this is done at
10756 * btrfs_remove_chunk() through a call to check_system_chunk().
10758 map
= em
->map_lookup
;
10759 num_items
= 3 + map
->num_stripes
;
10760 free_extent_map(em
);
10762 return btrfs_start_transaction_fallback_global_rsv(fs_info
->extent_root
,
10767 * Process the unused_bgs list and remove any that don't have any allocated
10768 * space inside of them.
10770 void btrfs_delete_unused_bgs(struct btrfs_fs_info
*fs_info
)
10772 struct btrfs_block_group_cache
*block_group
;
10773 struct btrfs_space_info
*space_info
;
10774 struct btrfs_root
*root
= fs_info
->extent_root
;
10775 struct btrfs_trans_handle
*trans
;
10778 if (!fs_info
->open
)
10781 spin_lock(&fs_info
->unused_bgs_lock
);
10782 while (!list_empty(&fs_info
->unused_bgs
)) {
10786 block_group
= list_first_entry(&fs_info
->unused_bgs
,
10787 struct btrfs_block_group_cache
,
10789 list_del_init(&block_group
->bg_list
);
10791 space_info
= block_group
->space_info
;
10793 if (ret
|| btrfs_mixed_space_info(space_info
)) {
10794 btrfs_put_block_group(block_group
);
10797 spin_unlock(&fs_info
->unused_bgs_lock
);
10799 mutex_lock(&fs_info
->delete_unused_bgs_mutex
);
10801 /* Don't want to race with allocators so take the groups_sem */
10802 down_write(&space_info
->groups_sem
);
10803 spin_lock(&block_group
->lock
);
10804 if (block_group
->reserved
||
10805 btrfs_block_group_used(&block_group
->item
) ||
10807 list_is_singular(&block_group
->list
)) {
10809 * We want to bail if we made new allocations or have
10810 * outstanding allocations in this block group. We do
10811 * the ro check in case balance is currently acting on
10812 * this block group.
10814 spin_unlock(&block_group
->lock
);
10815 up_write(&space_info
->groups_sem
);
10818 spin_unlock(&block_group
->lock
);
10820 /* We don't want to force the issue, only flip if it's ok. */
10821 ret
= inc_block_group_ro(block_group
, 0);
10822 up_write(&space_info
->groups_sem
);
10829 * Want to do this before we do anything else so we can recover
10830 * properly if we fail to join the transaction.
10832 trans
= btrfs_start_trans_remove_block_group(fs_info
,
10833 block_group
->key
.objectid
);
10834 if (IS_ERR(trans
)) {
10835 btrfs_dec_block_group_ro(root
, block_group
);
10836 ret
= PTR_ERR(trans
);
10841 * We could have pending pinned extents for this block group,
10842 * just delete them, we don't care about them anymore.
10844 start
= block_group
->key
.objectid
;
10845 end
= start
+ block_group
->key
.offset
- 1;
10847 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10848 * btrfs_finish_extent_commit(). If we are at transaction N,
10849 * another task might be running finish_extent_commit() for the
10850 * previous transaction N - 1, and have seen a range belonging
10851 * to the block group in freed_extents[] before we were able to
10852 * clear the whole block group range from freed_extents[]. This
10853 * means that task can lookup for the block group after we
10854 * unpinned it from freed_extents[] and removed it, leading to
10855 * a BUG_ON() at btrfs_unpin_extent_range().
10857 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
10858 ret
= clear_extent_bits(&fs_info
->freed_extents
[0], start
, end
,
10861 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10862 btrfs_dec_block_group_ro(root
, block_group
);
10865 ret
= clear_extent_bits(&fs_info
->freed_extents
[1], start
, end
,
10868 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10869 btrfs_dec_block_group_ro(root
, block_group
);
10872 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10874 /* Reset pinned so btrfs_put_block_group doesn't complain */
10875 spin_lock(&space_info
->lock
);
10876 spin_lock(&block_group
->lock
);
10878 space_info
->bytes_pinned
-= block_group
->pinned
;
10879 space_info
->bytes_readonly
+= block_group
->pinned
;
10880 percpu_counter_add(&space_info
->total_bytes_pinned
,
10881 -block_group
->pinned
);
10882 block_group
->pinned
= 0;
10884 spin_unlock(&block_group
->lock
);
10885 spin_unlock(&space_info
->lock
);
10887 /* DISCARD can flip during remount */
10888 trimming
= btrfs_test_opt(root
, DISCARD
);
10890 /* Implicit trim during transaction commit. */
10892 btrfs_get_block_group_trimming(block_group
);
10895 * Btrfs_remove_chunk will abort the transaction if things go
10898 ret
= btrfs_remove_chunk(trans
, root
,
10899 block_group
->key
.objectid
);
10903 btrfs_put_block_group_trimming(block_group
);
10908 * If we're not mounted with -odiscard, we can just forget
10909 * about this block group. Otherwise we'll need to wait
10910 * until transaction commit to do the actual discard.
10913 spin_lock(&fs_info
->unused_bgs_lock
);
10915 * A concurrent scrub might have added us to the list
10916 * fs_info->unused_bgs, so use a list_move operation
10917 * to add the block group to the deleted_bgs list.
10919 list_move(&block_group
->bg_list
,
10920 &trans
->transaction
->deleted_bgs
);
10921 spin_unlock(&fs_info
->unused_bgs_lock
);
10922 btrfs_get_block_group(block_group
);
10925 btrfs_end_transaction(trans
, root
);
10927 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
10928 btrfs_put_block_group(block_group
);
10929 spin_lock(&fs_info
->unused_bgs_lock
);
10931 spin_unlock(&fs_info
->unused_bgs_lock
);
10934 int btrfs_init_space_info(struct btrfs_fs_info
*fs_info
)
10936 struct btrfs_space_info
*space_info
;
10937 struct btrfs_super_block
*disk_super
;
10943 disk_super
= fs_info
->super_copy
;
10944 if (!btrfs_super_root(disk_super
))
10947 features
= btrfs_super_incompat_flags(disk_super
);
10948 if (features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
10951 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
10952 ret
= update_space_info(fs_info
, flags
, 0, 0, 0, &space_info
);
10957 flags
= BTRFS_BLOCK_GROUP_METADATA
| BTRFS_BLOCK_GROUP_DATA
;
10958 ret
= update_space_info(fs_info
, flags
, 0, 0, 0, &space_info
);
10960 flags
= BTRFS_BLOCK_GROUP_METADATA
;
10961 ret
= update_space_info(fs_info
, flags
, 0, 0, 0, &space_info
);
10965 flags
= BTRFS_BLOCK_GROUP_DATA
;
10966 ret
= update_space_info(fs_info
, flags
, 0, 0, 0, &space_info
);
10972 int btrfs_error_unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
10974 return unpin_extent_range(root
, start
, end
, false);
10978 * It used to be that old block groups would be left around forever.
10979 * Iterating over them would be enough to trim unused space. Since we
10980 * now automatically remove them, we also need to iterate over unallocated
10983 * We don't want a transaction for this since the discard may take a
10984 * substantial amount of time. We don't require that a transaction be
10985 * running, but we do need to take a running transaction into account
10986 * to ensure that we're not discarding chunks that were released in
10987 * the current transaction.
10989 * Holding the chunks lock will prevent other threads from allocating
10990 * or releasing chunks, but it won't prevent a running transaction
10991 * from committing and releasing the memory that the pending chunks
10992 * list head uses. For that, we need to take a reference to the
10995 static int btrfs_trim_free_extents(struct btrfs_device
*device
,
10996 u64 minlen
, u64
*trimmed
)
10998 u64 start
= 0, len
= 0;
11003 /* Not writeable = nothing to do. */
11004 if (!device
->writeable
)
11007 /* No free space = nothing to do. */
11008 if (device
->total_bytes
<= device
->bytes_used
)
11014 struct btrfs_fs_info
*fs_info
= device
->dev_root
->fs_info
;
11015 struct btrfs_transaction
*trans
;
11018 ret
= mutex_lock_interruptible(&fs_info
->chunk_mutex
);
11022 down_read(&fs_info
->commit_root_sem
);
11024 spin_lock(&fs_info
->trans_lock
);
11025 trans
= fs_info
->running_transaction
;
11027 atomic_inc(&trans
->use_count
);
11028 spin_unlock(&fs_info
->trans_lock
);
11030 ret
= find_free_dev_extent_start(trans
, device
, minlen
, start
,
11033 btrfs_put_transaction(trans
);
11036 up_read(&fs_info
->commit_root_sem
);
11037 mutex_unlock(&fs_info
->chunk_mutex
);
11038 if (ret
== -ENOSPC
)
11043 ret
= btrfs_issue_discard(device
->bdev
, start
, len
, &bytes
);
11044 up_read(&fs_info
->commit_root_sem
);
11045 mutex_unlock(&fs_info
->chunk_mutex
);
11053 if (fatal_signal_pending(current
)) {
11054 ret
= -ERESTARTSYS
;
11064 int btrfs_trim_fs(struct btrfs_root
*root
, struct fstrim_range
*range
)
11066 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
11067 struct btrfs_block_group_cache
*cache
= NULL
;
11068 struct btrfs_device
*device
;
11069 struct list_head
*devices
;
11074 u64 total_bytes
= btrfs_super_total_bytes(fs_info
->super_copy
);
11078 * try to trim all FS space, our block group may start from non-zero.
11080 if (range
->len
== total_bytes
)
11081 cache
= btrfs_lookup_first_block_group(fs_info
, range
->start
);
11083 cache
= btrfs_lookup_block_group(fs_info
, range
->start
);
11086 if (cache
->key
.objectid
>= (range
->start
+ range
->len
)) {
11087 btrfs_put_block_group(cache
);
11091 start
= max(range
->start
, cache
->key
.objectid
);
11092 end
= min(range
->start
+ range
->len
,
11093 cache
->key
.objectid
+ cache
->key
.offset
);
11095 if (end
- start
>= range
->minlen
) {
11096 if (!block_group_cache_done(cache
)) {
11097 ret
= cache_block_group(cache
, 0);
11099 btrfs_put_block_group(cache
);
11102 ret
= wait_block_group_cache_done(cache
);
11104 btrfs_put_block_group(cache
);
11108 ret
= btrfs_trim_block_group(cache
,
11114 trimmed
+= group_trimmed
;
11116 btrfs_put_block_group(cache
);
11121 cache
= next_block_group(fs_info
->tree_root
, cache
);
11124 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
11125 devices
= &root
->fs_info
->fs_devices
->alloc_list
;
11126 list_for_each_entry(device
, devices
, dev_alloc_list
) {
11127 ret
= btrfs_trim_free_extents(device
, range
->minlen
,
11132 trimmed
+= group_trimmed
;
11134 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
11136 range
->len
= trimmed
;
11141 * btrfs_{start,end}_write_no_snapshoting() are similar to
11142 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
11143 * data into the page cache through nocow before the subvolume is snapshoted,
11144 * but flush the data into disk after the snapshot creation, or to prevent
11145 * operations while snapshoting is ongoing and that cause the snapshot to be
11146 * inconsistent (writes followed by expanding truncates for example).
11148 void btrfs_end_write_no_snapshoting(struct btrfs_root
*root
)
11150 percpu_counter_dec(&root
->subv_writers
->counter
);
11152 * Make sure counter is updated before we wake up waiters.
11155 if (waitqueue_active(&root
->subv_writers
->wait
))
11156 wake_up(&root
->subv_writers
->wait
);
11159 int btrfs_start_write_no_snapshoting(struct btrfs_root
*root
)
11161 if (atomic_read(&root
->will_be_snapshoted
))
11164 percpu_counter_inc(&root
->subv_writers
->counter
);
11166 * Make sure counter is updated before we check for snapshot creation.
11169 if (atomic_read(&root
->will_be_snapshoted
)) {
11170 btrfs_end_write_no_snapshoting(root
);
11176 static int wait_snapshoting_atomic_t(atomic_t
*a
)
11182 void btrfs_wait_for_snapshot_creation(struct btrfs_root
*root
)
11187 ret
= btrfs_start_write_no_snapshoting(root
);
11190 wait_on_atomic_t(&root
->will_be_snapshoted
,
11191 wait_snapshoting_atomic_t
,
11192 TASK_UNINTERRUPTIBLE
);