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
);
116 block_group_cache_done(struct btrfs_block_group_cache
*cache
)
119 return cache
->cached
== BTRFS_CACHE_FINISHED
||
120 cache
->cached
== BTRFS_CACHE_ERROR
;
123 static int block_group_bits(struct btrfs_block_group_cache
*cache
, u64 bits
)
125 return (cache
->flags
& bits
) == bits
;
128 void btrfs_get_block_group(struct btrfs_block_group_cache
*cache
)
130 atomic_inc(&cache
->count
);
133 void btrfs_put_block_group(struct btrfs_block_group_cache
*cache
)
135 if (atomic_dec_and_test(&cache
->count
)) {
136 WARN_ON(cache
->pinned
> 0);
137 WARN_ON(cache
->reserved
> 0);
138 kfree(cache
->free_space_ctl
);
144 * this adds the block group to the fs_info rb tree for the block group
147 static int btrfs_add_block_group_cache(struct btrfs_fs_info
*info
,
148 struct btrfs_block_group_cache
*block_group
)
151 struct rb_node
*parent
= NULL
;
152 struct btrfs_block_group_cache
*cache
;
154 spin_lock(&info
->block_group_cache_lock
);
155 p
= &info
->block_group_cache_tree
.rb_node
;
159 cache
= rb_entry(parent
, struct btrfs_block_group_cache
,
161 if (block_group
->key
.objectid
< cache
->key
.objectid
) {
163 } else if (block_group
->key
.objectid
> cache
->key
.objectid
) {
166 spin_unlock(&info
->block_group_cache_lock
);
171 rb_link_node(&block_group
->cache_node
, parent
, p
);
172 rb_insert_color(&block_group
->cache_node
,
173 &info
->block_group_cache_tree
);
175 if (info
->first_logical_byte
> block_group
->key
.objectid
)
176 info
->first_logical_byte
= block_group
->key
.objectid
;
178 spin_unlock(&info
->block_group_cache_lock
);
184 * This will return the block group at or after bytenr if contains is 0, else
185 * it will return the block group that contains the bytenr
187 static struct btrfs_block_group_cache
*
188 block_group_cache_tree_search(struct btrfs_fs_info
*info
, u64 bytenr
,
191 struct btrfs_block_group_cache
*cache
, *ret
= NULL
;
195 spin_lock(&info
->block_group_cache_lock
);
196 n
= info
->block_group_cache_tree
.rb_node
;
199 cache
= rb_entry(n
, struct btrfs_block_group_cache
,
201 end
= cache
->key
.objectid
+ cache
->key
.offset
- 1;
202 start
= cache
->key
.objectid
;
204 if (bytenr
< start
) {
205 if (!contains
&& (!ret
|| start
< ret
->key
.objectid
))
208 } else if (bytenr
> start
) {
209 if (contains
&& bytenr
<= end
) {
220 btrfs_get_block_group(ret
);
221 if (bytenr
== 0 && info
->first_logical_byte
> ret
->key
.objectid
)
222 info
->first_logical_byte
= ret
->key
.objectid
;
224 spin_unlock(&info
->block_group_cache_lock
);
229 static int add_excluded_extent(struct btrfs_root
*root
,
230 u64 start
, u64 num_bytes
)
232 u64 end
= start
+ num_bytes
- 1;
233 set_extent_bits(&root
->fs_info
->freed_extents
[0],
234 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
235 set_extent_bits(&root
->fs_info
->freed_extents
[1],
236 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
240 static void free_excluded_extents(struct btrfs_root
*root
,
241 struct btrfs_block_group_cache
*cache
)
245 start
= cache
->key
.objectid
;
246 end
= start
+ cache
->key
.offset
- 1;
248 clear_extent_bits(&root
->fs_info
->freed_extents
[0],
249 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
250 clear_extent_bits(&root
->fs_info
->freed_extents
[1],
251 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
254 static int exclude_super_stripes(struct btrfs_root
*root
,
255 struct btrfs_block_group_cache
*cache
)
262 if (cache
->key
.objectid
< BTRFS_SUPER_INFO_OFFSET
) {
263 stripe_len
= BTRFS_SUPER_INFO_OFFSET
- cache
->key
.objectid
;
264 cache
->bytes_super
+= stripe_len
;
265 ret
= add_excluded_extent(root
, cache
->key
.objectid
,
271 for (i
= 0; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
272 bytenr
= btrfs_sb_offset(i
);
273 ret
= btrfs_rmap_block(&root
->fs_info
->mapping_tree
,
274 cache
->key
.objectid
, bytenr
,
275 0, &logical
, &nr
, &stripe_len
);
282 if (logical
[nr
] > cache
->key
.objectid
+
286 if (logical
[nr
] + stripe_len
<= cache
->key
.objectid
)
290 if (start
< cache
->key
.objectid
) {
291 start
= cache
->key
.objectid
;
292 len
= (logical
[nr
] + stripe_len
) - start
;
294 len
= min_t(u64
, stripe_len
,
295 cache
->key
.objectid
+
296 cache
->key
.offset
- start
);
299 cache
->bytes_super
+= len
;
300 ret
= add_excluded_extent(root
, start
, len
);
312 static struct btrfs_caching_control
*
313 get_caching_control(struct btrfs_block_group_cache
*cache
)
315 struct btrfs_caching_control
*ctl
;
317 spin_lock(&cache
->lock
);
318 if (!cache
->caching_ctl
) {
319 spin_unlock(&cache
->lock
);
323 ctl
= cache
->caching_ctl
;
324 atomic_inc(&ctl
->count
);
325 spin_unlock(&cache
->lock
);
329 static void put_caching_control(struct btrfs_caching_control
*ctl
)
331 if (atomic_dec_and_test(&ctl
->count
))
335 #ifdef CONFIG_BTRFS_DEBUG
336 static void fragment_free_space(struct btrfs_root
*root
,
337 struct btrfs_block_group_cache
*block_group
)
339 u64 start
= block_group
->key
.objectid
;
340 u64 len
= block_group
->key
.offset
;
341 u64 chunk
= block_group
->flags
& BTRFS_BLOCK_GROUP_METADATA
?
342 root
->nodesize
: root
->sectorsize
;
343 u64 step
= chunk
<< 1;
345 while (len
> chunk
) {
346 btrfs_remove_free_space(block_group
, start
, chunk
);
357 * this is only called by cache_block_group, since we could have freed extents
358 * we need to check the pinned_extents for any extents that can't be used yet
359 * since their free space will be released as soon as the transaction commits.
361 u64
add_new_free_space(struct btrfs_block_group_cache
*block_group
,
362 struct btrfs_fs_info
*info
, u64 start
, u64 end
)
364 u64 extent_start
, extent_end
, size
, total_added
= 0;
367 while (start
< end
) {
368 ret
= find_first_extent_bit(info
->pinned_extents
, start
,
369 &extent_start
, &extent_end
,
370 EXTENT_DIRTY
| EXTENT_UPTODATE
,
375 if (extent_start
<= start
) {
376 start
= extent_end
+ 1;
377 } else if (extent_start
> start
&& extent_start
< end
) {
378 size
= extent_start
- start
;
380 ret
= btrfs_add_free_space(block_group
, start
,
382 BUG_ON(ret
); /* -ENOMEM or logic error */
383 start
= extent_end
+ 1;
392 ret
= btrfs_add_free_space(block_group
, start
, size
);
393 BUG_ON(ret
); /* -ENOMEM or logic error */
399 static int load_extent_tree_free(struct btrfs_caching_control
*caching_ctl
)
401 struct btrfs_block_group_cache
*block_group
;
402 struct btrfs_fs_info
*fs_info
;
403 struct btrfs_root
*extent_root
;
404 struct btrfs_path
*path
;
405 struct extent_buffer
*leaf
;
406 struct btrfs_key key
;
413 block_group
= caching_ctl
->block_group
;
414 fs_info
= block_group
->fs_info
;
415 extent_root
= fs_info
->extent_root
;
417 path
= btrfs_alloc_path();
421 last
= max_t(u64
, block_group
->key
.objectid
, BTRFS_SUPER_INFO_OFFSET
);
423 #ifdef CONFIG_BTRFS_DEBUG
425 * If we're fragmenting we don't want to make anybody think we can
426 * allocate from this block group until we've had a chance to fragment
429 if (btrfs_should_fragment_free_space(extent_root
, block_group
))
433 * We don't want to deadlock with somebody trying to allocate a new
434 * extent for the extent root while also trying to search the extent
435 * root to add free space. So we skip locking and search the commit
436 * root, since its read-only
438 path
->skip_locking
= 1;
439 path
->search_commit_root
= 1;
440 path
->reada
= READA_FORWARD
;
444 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
447 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
451 leaf
= path
->nodes
[0];
452 nritems
= btrfs_header_nritems(leaf
);
455 if (btrfs_fs_closing(fs_info
) > 1) {
460 if (path
->slots
[0] < nritems
) {
461 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
463 ret
= find_next_key(path
, 0, &key
);
467 if (need_resched() ||
468 rwsem_is_contended(&fs_info
->commit_root_sem
)) {
470 caching_ctl
->progress
= last
;
471 btrfs_release_path(path
);
472 up_read(&fs_info
->commit_root_sem
);
473 mutex_unlock(&caching_ctl
->mutex
);
475 mutex_lock(&caching_ctl
->mutex
);
476 down_read(&fs_info
->commit_root_sem
);
480 ret
= btrfs_next_leaf(extent_root
, path
);
485 leaf
= path
->nodes
[0];
486 nritems
= btrfs_header_nritems(leaf
);
490 if (key
.objectid
< last
) {
493 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
496 caching_ctl
->progress
= last
;
497 btrfs_release_path(path
);
501 if (key
.objectid
< block_group
->key
.objectid
) {
506 if (key
.objectid
>= block_group
->key
.objectid
+
507 block_group
->key
.offset
)
510 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
||
511 key
.type
== BTRFS_METADATA_ITEM_KEY
) {
512 total_found
+= add_new_free_space(block_group
,
515 if (key
.type
== BTRFS_METADATA_ITEM_KEY
)
516 last
= key
.objectid
+
517 fs_info
->tree_root
->nodesize
;
519 last
= key
.objectid
+ key
.offset
;
521 if (total_found
> CACHING_CTL_WAKE_UP
) {
524 wake_up(&caching_ctl
->wait
);
531 total_found
+= add_new_free_space(block_group
, fs_info
, last
,
532 block_group
->key
.objectid
+
533 block_group
->key
.offset
);
534 caching_ctl
->progress
= (u64
)-1;
537 btrfs_free_path(path
);
541 static noinline
void caching_thread(struct btrfs_work
*work
)
543 struct btrfs_block_group_cache
*block_group
;
544 struct btrfs_fs_info
*fs_info
;
545 struct btrfs_caching_control
*caching_ctl
;
546 struct btrfs_root
*extent_root
;
549 caching_ctl
= container_of(work
, struct btrfs_caching_control
, work
);
550 block_group
= caching_ctl
->block_group
;
551 fs_info
= block_group
->fs_info
;
552 extent_root
= fs_info
->extent_root
;
554 mutex_lock(&caching_ctl
->mutex
);
555 down_read(&fs_info
->commit_root_sem
);
557 if (btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
))
558 ret
= load_free_space_tree(caching_ctl
);
560 ret
= load_extent_tree_free(caching_ctl
);
562 spin_lock(&block_group
->lock
);
563 block_group
->caching_ctl
= NULL
;
564 block_group
->cached
= ret
? BTRFS_CACHE_ERROR
: BTRFS_CACHE_FINISHED
;
565 spin_unlock(&block_group
->lock
);
567 #ifdef CONFIG_BTRFS_DEBUG
568 if (btrfs_should_fragment_free_space(extent_root
, block_group
)) {
571 spin_lock(&block_group
->space_info
->lock
);
572 spin_lock(&block_group
->lock
);
573 bytes_used
= block_group
->key
.offset
-
574 btrfs_block_group_used(&block_group
->item
);
575 block_group
->space_info
->bytes_used
+= bytes_used
>> 1;
576 spin_unlock(&block_group
->lock
);
577 spin_unlock(&block_group
->space_info
->lock
);
578 fragment_free_space(extent_root
, block_group
);
582 caching_ctl
->progress
= (u64
)-1;
584 up_read(&fs_info
->commit_root_sem
);
585 free_excluded_extents(fs_info
->extent_root
, block_group
);
586 mutex_unlock(&caching_ctl
->mutex
);
588 wake_up(&caching_ctl
->wait
);
590 put_caching_control(caching_ctl
);
591 btrfs_put_block_group(block_group
);
594 static int cache_block_group(struct btrfs_block_group_cache
*cache
,
598 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
599 struct btrfs_caching_control
*caching_ctl
;
602 caching_ctl
= kzalloc(sizeof(*caching_ctl
), GFP_NOFS
);
606 INIT_LIST_HEAD(&caching_ctl
->list
);
607 mutex_init(&caching_ctl
->mutex
);
608 init_waitqueue_head(&caching_ctl
->wait
);
609 caching_ctl
->block_group
= cache
;
610 caching_ctl
->progress
= cache
->key
.objectid
;
611 atomic_set(&caching_ctl
->count
, 1);
612 btrfs_init_work(&caching_ctl
->work
, btrfs_cache_helper
,
613 caching_thread
, NULL
, NULL
);
615 spin_lock(&cache
->lock
);
617 * This should be a rare occasion, but this could happen I think in the
618 * case where one thread starts to load the space cache info, and then
619 * some other thread starts a transaction commit which tries to do an
620 * allocation while the other thread is still loading the space cache
621 * info. The previous loop should have kept us from choosing this block
622 * group, but if we've moved to the state where we will wait on caching
623 * block groups we need to first check if we're doing a fast load here,
624 * so we can wait for it to finish, otherwise we could end up allocating
625 * from a block group who's cache gets evicted for one reason or
628 while (cache
->cached
== BTRFS_CACHE_FAST
) {
629 struct btrfs_caching_control
*ctl
;
631 ctl
= cache
->caching_ctl
;
632 atomic_inc(&ctl
->count
);
633 prepare_to_wait(&ctl
->wait
, &wait
, TASK_UNINTERRUPTIBLE
);
634 spin_unlock(&cache
->lock
);
638 finish_wait(&ctl
->wait
, &wait
);
639 put_caching_control(ctl
);
640 spin_lock(&cache
->lock
);
643 if (cache
->cached
!= BTRFS_CACHE_NO
) {
644 spin_unlock(&cache
->lock
);
648 WARN_ON(cache
->caching_ctl
);
649 cache
->caching_ctl
= caching_ctl
;
650 cache
->cached
= BTRFS_CACHE_FAST
;
651 spin_unlock(&cache
->lock
);
653 if (fs_info
->mount_opt
& BTRFS_MOUNT_SPACE_CACHE
) {
654 mutex_lock(&caching_ctl
->mutex
);
655 ret
= load_free_space_cache(fs_info
, cache
);
657 spin_lock(&cache
->lock
);
659 cache
->caching_ctl
= NULL
;
660 cache
->cached
= BTRFS_CACHE_FINISHED
;
661 cache
->last_byte_to_unpin
= (u64
)-1;
662 caching_ctl
->progress
= (u64
)-1;
664 if (load_cache_only
) {
665 cache
->caching_ctl
= NULL
;
666 cache
->cached
= BTRFS_CACHE_NO
;
668 cache
->cached
= BTRFS_CACHE_STARTED
;
669 cache
->has_caching_ctl
= 1;
672 spin_unlock(&cache
->lock
);
673 #ifdef CONFIG_BTRFS_DEBUG
675 btrfs_should_fragment_free_space(fs_info
->extent_root
,
679 spin_lock(&cache
->space_info
->lock
);
680 spin_lock(&cache
->lock
);
681 bytes_used
= cache
->key
.offset
-
682 btrfs_block_group_used(&cache
->item
);
683 cache
->space_info
->bytes_used
+= bytes_used
>> 1;
684 spin_unlock(&cache
->lock
);
685 spin_unlock(&cache
->space_info
->lock
);
686 fragment_free_space(fs_info
->extent_root
, cache
);
689 mutex_unlock(&caching_ctl
->mutex
);
691 wake_up(&caching_ctl
->wait
);
693 put_caching_control(caching_ctl
);
694 free_excluded_extents(fs_info
->extent_root
, cache
);
699 * We're either using the free space tree or no caching at all.
700 * Set cached to the appropriate value and wakeup any waiters.
702 spin_lock(&cache
->lock
);
703 if (load_cache_only
) {
704 cache
->caching_ctl
= NULL
;
705 cache
->cached
= BTRFS_CACHE_NO
;
707 cache
->cached
= BTRFS_CACHE_STARTED
;
708 cache
->has_caching_ctl
= 1;
710 spin_unlock(&cache
->lock
);
711 wake_up(&caching_ctl
->wait
);
714 if (load_cache_only
) {
715 put_caching_control(caching_ctl
);
719 down_write(&fs_info
->commit_root_sem
);
720 atomic_inc(&caching_ctl
->count
);
721 list_add_tail(&caching_ctl
->list
, &fs_info
->caching_block_groups
);
722 up_write(&fs_info
->commit_root_sem
);
724 btrfs_get_block_group(cache
);
726 btrfs_queue_work(fs_info
->caching_workers
, &caching_ctl
->work
);
732 * return the block group that starts at or after bytenr
734 static struct btrfs_block_group_cache
*
735 btrfs_lookup_first_block_group(struct btrfs_fs_info
*info
, u64 bytenr
)
737 struct btrfs_block_group_cache
*cache
;
739 cache
= block_group_cache_tree_search(info
, bytenr
, 0);
745 * return the block group that contains the given bytenr
747 struct btrfs_block_group_cache
*btrfs_lookup_block_group(
748 struct btrfs_fs_info
*info
,
751 struct btrfs_block_group_cache
*cache
;
753 cache
= block_group_cache_tree_search(info
, bytenr
, 1);
758 static struct btrfs_space_info
*__find_space_info(struct btrfs_fs_info
*info
,
761 struct list_head
*head
= &info
->space_info
;
762 struct btrfs_space_info
*found
;
764 flags
&= BTRFS_BLOCK_GROUP_TYPE_MASK
;
767 list_for_each_entry_rcu(found
, head
, list
) {
768 if (found
->flags
& flags
) {
778 * after adding space to the filesystem, we need to clear the full flags
779 * on all the space infos.
781 void btrfs_clear_space_info_full(struct btrfs_fs_info
*info
)
783 struct list_head
*head
= &info
->space_info
;
784 struct btrfs_space_info
*found
;
787 list_for_each_entry_rcu(found
, head
, list
)
792 /* simple helper to search for an existing data extent at a given offset */
793 int btrfs_lookup_data_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
796 struct btrfs_key key
;
797 struct btrfs_path
*path
;
799 path
= btrfs_alloc_path();
803 key
.objectid
= start
;
805 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
806 ret
= btrfs_search_slot(NULL
, root
->fs_info
->extent_root
, &key
, path
,
808 btrfs_free_path(path
);
813 * helper function to lookup reference count and flags of a tree block.
815 * the head node for delayed ref is used to store the sum of all the
816 * reference count modifications queued up in the rbtree. the head
817 * node may also store the extent flags to set. This way you can check
818 * to see what the reference count and extent flags would be if all of
819 * the delayed refs are not processed.
821 int btrfs_lookup_extent_info(struct btrfs_trans_handle
*trans
,
822 struct btrfs_root
*root
, u64 bytenr
,
823 u64 offset
, int metadata
, u64
*refs
, u64
*flags
)
825 struct btrfs_delayed_ref_head
*head
;
826 struct btrfs_delayed_ref_root
*delayed_refs
;
827 struct btrfs_path
*path
;
828 struct btrfs_extent_item
*ei
;
829 struct extent_buffer
*leaf
;
830 struct btrfs_key key
;
837 * If we don't have skinny metadata, don't bother doing anything
840 if (metadata
&& !btrfs_fs_incompat(root
->fs_info
, SKINNY_METADATA
)) {
841 offset
= root
->nodesize
;
845 path
= btrfs_alloc_path();
850 path
->skip_locking
= 1;
851 path
->search_commit_root
= 1;
855 key
.objectid
= bytenr
;
858 key
.type
= BTRFS_METADATA_ITEM_KEY
;
860 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
862 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
,
867 if (ret
> 0 && metadata
&& key
.type
== BTRFS_METADATA_ITEM_KEY
) {
868 if (path
->slots
[0]) {
870 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
872 if (key
.objectid
== bytenr
&&
873 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
874 key
.offset
== root
->nodesize
)
880 leaf
= path
->nodes
[0];
881 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
882 if (item_size
>= sizeof(*ei
)) {
883 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
884 struct btrfs_extent_item
);
885 num_refs
= btrfs_extent_refs(leaf
, ei
);
886 extent_flags
= btrfs_extent_flags(leaf
, ei
);
888 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
889 struct btrfs_extent_item_v0
*ei0
;
890 BUG_ON(item_size
!= sizeof(*ei0
));
891 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
892 struct btrfs_extent_item_v0
);
893 num_refs
= btrfs_extent_refs_v0(leaf
, ei0
);
894 /* FIXME: this isn't correct for data */
895 extent_flags
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
900 BUG_ON(num_refs
== 0);
910 delayed_refs
= &trans
->transaction
->delayed_refs
;
911 spin_lock(&delayed_refs
->lock
);
912 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
914 if (!mutex_trylock(&head
->mutex
)) {
915 atomic_inc(&head
->node
.refs
);
916 spin_unlock(&delayed_refs
->lock
);
918 btrfs_release_path(path
);
921 * Mutex was contended, block until it's released and try
924 mutex_lock(&head
->mutex
);
925 mutex_unlock(&head
->mutex
);
926 btrfs_put_delayed_ref(&head
->node
);
929 spin_lock(&head
->lock
);
930 if (head
->extent_op
&& head
->extent_op
->update_flags
)
931 extent_flags
|= head
->extent_op
->flags_to_set
;
933 BUG_ON(num_refs
== 0);
935 num_refs
+= head
->node
.ref_mod
;
936 spin_unlock(&head
->lock
);
937 mutex_unlock(&head
->mutex
);
939 spin_unlock(&delayed_refs
->lock
);
941 WARN_ON(num_refs
== 0);
945 *flags
= extent_flags
;
947 btrfs_free_path(path
);
952 * Back reference rules. Back refs have three main goals:
954 * 1) differentiate between all holders of references to an extent so that
955 * when a reference is dropped we can make sure it was a valid reference
956 * before freeing the extent.
958 * 2) Provide enough information to quickly find the holders of an extent
959 * if we notice a given block is corrupted or bad.
961 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
962 * maintenance. This is actually the same as #2, but with a slightly
963 * different use case.
965 * There are two kinds of back refs. The implicit back refs is optimized
966 * for pointers in non-shared tree blocks. For a given pointer in a block,
967 * back refs of this kind provide information about the block's owner tree
968 * and the pointer's key. These information allow us to find the block by
969 * b-tree searching. The full back refs is for pointers in tree blocks not
970 * referenced by their owner trees. The location of tree block is recorded
971 * in the back refs. Actually the full back refs is generic, and can be
972 * used in all cases the implicit back refs is used. The major shortcoming
973 * of the full back refs is its overhead. Every time a tree block gets
974 * COWed, we have to update back refs entry for all pointers in it.
976 * For a newly allocated tree block, we use implicit back refs for
977 * pointers in it. This means most tree related operations only involve
978 * implicit back refs. For a tree block created in old transaction, the
979 * only way to drop a reference to it is COW it. So we can detect the
980 * event that tree block loses its owner tree's reference and do the
981 * back refs conversion.
983 * When a tree block is COW'd through a tree, there are four cases:
985 * The reference count of the block is one and the tree is the block's
986 * owner tree. Nothing to do in this case.
988 * The reference count of the block is one and the tree is not the
989 * block's owner tree. In this case, full back refs is used for pointers
990 * in the block. Remove these full back refs, add implicit back refs for
991 * every pointers in the new block.
993 * The reference count of the block is greater than one and the tree is
994 * the block's owner tree. In this case, implicit back refs is used for
995 * pointers in the block. Add full back refs for every pointers in the
996 * block, increase lower level extents' reference counts. The original
997 * implicit back refs are entailed to the new block.
999 * The reference count of the block is greater than one and the tree is
1000 * not the block's owner tree. Add implicit back refs for every pointer in
1001 * the new block, increase lower level extents' reference count.
1003 * Back Reference Key composing:
1005 * The key objectid corresponds to the first byte in the extent,
1006 * The key type is used to differentiate between types of back refs.
1007 * There are different meanings of the key offset for different types
1010 * File extents can be referenced by:
1012 * - multiple snapshots, subvolumes, or different generations in one subvol
1013 * - different files inside a single subvolume
1014 * - different offsets inside a file (bookend extents in file.c)
1016 * The extent ref structure for the implicit back refs has fields for:
1018 * - Objectid of the subvolume root
1019 * - objectid of the file holding the reference
1020 * - original offset in the file
1021 * - how many bookend extents
1023 * The key offset for the implicit back refs is hash of the first
1026 * The extent ref structure for the full back refs has field for:
1028 * - number of pointers in the tree leaf
1030 * The key offset for the implicit back refs is the first byte of
1033 * When a file extent is allocated, The implicit back refs is used.
1034 * the fields are filled in:
1036 * (root_key.objectid, inode objectid, offset in file, 1)
1038 * When a file extent is removed file truncation, we find the
1039 * corresponding implicit back refs and check the following fields:
1041 * (btrfs_header_owner(leaf), inode objectid, offset in file)
1043 * Btree extents can be referenced by:
1045 * - Different subvolumes
1047 * Both the implicit back refs and the full back refs for tree blocks
1048 * only consist of key. The key offset for the implicit back refs is
1049 * objectid of block's owner tree. The key offset for the full back refs
1050 * is the first byte of parent block.
1052 * When implicit back refs is used, information about the lowest key and
1053 * level of the tree block are required. These information are stored in
1054 * tree block info structure.
1057 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1058 static int convert_extent_item_v0(struct btrfs_trans_handle
*trans
,
1059 struct btrfs_root
*root
,
1060 struct btrfs_path
*path
,
1061 u64 owner
, u32 extra_size
)
1063 struct btrfs_extent_item
*item
;
1064 struct btrfs_extent_item_v0
*ei0
;
1065 struct btrfs_extent_ref_v0
*ref0
;
1066 struct btrfs_tree_block_info
*bi
;
1067 struct extent_buffer
*leaf
;
1068 struct btrfs_key key
;
1069 struct btrfs_key found_key
;
1070 u32 new_size
= sizeof(*item
);
1074 leaf
= path
->nodes
[0];
1075 BUG_ON(btrfs_item_size_nr(leaf
, path
->slots
[0]) != sizeof(*ei0
));
1077 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1078 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1079 struct btrfs_extent_item_v0
);
1080 refs
= btrfs_extent_refs_v0(leaf
, ei0
);
1082 if (owner
== (u64
)-1) {
1084 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1085 ret
= btrfs_next_leaf(root
, path
);
1088 BUG_ON(ret
> 0); /* Corruption */
1089 leaf
= path
->nodes
[0];
1091 btrfs_item_key_to_cpu(leaf
, &found_key
,
1093 BUG_ON(key
.objectid
!= found_key
.objectid
);
1094 if (found_key
.type
!= BTRFS_EXTENT_REF_V0_KEY
) {
1098 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1099 struct btrfs_extent_ref_v0
);
1100 owner
= btrfs_ref_objectid_v0(leaf
, ref0
);
1104 btrfs_release_path(path
);
1106 if (owner
< BTRFS_FIRST_FREE_OBJECTID
)
1107 new_size
+= sizeof(*bi
);
1109 new_size
-= sizeof(*ei0
);
1110 ret
= btrfs_search_slot(trans
, root
, &key
, path
,
1111 new_size
+ extra_size
, 1);
1114 BUG_ON(ret
); /* Corruption */
1116 btrfs_extend_item(root
, path
, new_size
);
1118 leaf
= path
->nodes
[0];
1119 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1120 btrfs_set_extent_refs(leaf
, item
, refs
);
1121 /* FIXME: get real generation */
1122 btrfs_set_extent_generation(leaf
, item
, 0);
1123 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1124 btrfs_set_extent_flags(leaf
, item
,
1125 BTRFS_EXTENT_FLAG_TREE_BLOCK
|
1126 BTRFS_BLOCK_FLAG_FULL_BACKREF
);
1127 bi
= (struct btrfs_tree_block_info
*)(item
+ 1);
1128 /* FIXME: get first key of the block */
1129 memset_extent_buffer(leaf
, 0, (unsigned long)bi
, sizeof(*bi
));
1130 btrfs_set_tree_block_level(leaf
, bi
, (int)owner
);
1132 btrfs_set_extent_flags(leaf
, item
, BTRFS_EXTENT_FLAG_DATA
);
1134 btrfs_mark_buffer_dirty(leaf
);
1139 static u64
hash_extent_data_ref(u64 root_objectid
, u64 owner
, u64 offset
)
1141 u32 high_crc
= ~(u32
)0;
1142 u32 low_crc
= ~(u32
)0;
1145 lenum
= cpu_to_le64(root_objectid
);
1146 high_crc
= btrfs_crc32c(high_crc
, &lenum
, sizeof(lenum
));
1147 lenum
= cpu_to_le64(owner
);
1148 low_crc
= btrfs_crc32c(low_crc
, &lenum
, sizeof(lenum
));
1149 lenum
= cpu_to_le64(offset
);
1150 low_crc
= btrfs_crc32c(low_crc
, &lenum
, sizeof(lenum
));
1152 return ((u64
)high_crc
<< 31) ^ (u64
)low_crc
;
1155 static u64
hash_extent_data_ref_item(struct extent_buffer
*leaf
,
1156 struct btrfs_extent_data_ref
*ref
)
1158 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf
, ref
),
1159 btrfs_extent_data_ref_objectid(leaf
, ref
),
1160 btrfs_extent_data_ref_offset(leaf
, ref
));
1163 static int match_extent_data_ref(struct extent_buffer
*leaf
,
1164 struct btrfs_extent_data_ref
*ref
,
1165 u64 root_objectid
, u64 owner
, u64 offset
)
1167 if (btrfs_extent_data_ref_root(leaf
, ref
) != root_objectid
||
1168 btrfs_extent_data_ref_objectid(leaf
, ref
) != owner
||
1169 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
1174 static noinline
int lookup_extent_data_ref(struct btrfs_trans_handle
*trans
,
1175 struct btrfs_root
*root
,
1176 struct btrfs_path
*path
,
1177 u64 bytenr
, u64 parent
,
1179 u64 owner
, u64 offset
)
1181 struct btrfs_key key
;
1182 struct btrfs_extent_data_ref
*ref
;
1183 struct extent_buffer
*leaf
;
1189 key
.objectid
= bytenr
;
1191 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1192 key
.offset
= parent
;
1194 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1195 key
.offset
= hash_extent_data_ref(root_objectid
,
1200 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1209 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1210 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1211 btrfs_release_path(path
);
1212 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1223 leaf
= path
->nodes
[0];
1224 nritems
= btrfs_header_nritems(leaf
);
1226 if (path
->slots
[0] >= nritems
) {
1227 ret
= btrfs_next_leaf(root
, path
);
1233 leaf
= path
->nodes
[0];
1234 nritems
= btrfs_header_nritems(leaf
);
1238 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1239 if (key
.objectid
!= bytenr
||
1240 key
.type
!= BTRFS_EXTENT_DATA_REF_KEY
)
1243 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1244 struct btrfs_extent_data_ref
);
1246 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1249 btrfs_release_path(path
);
1261 static noinline
int insert_extent_data_ref(struct btrfs_trans_handle
*trans
,
1262 struct btrfs_root
*root
,
1263 struct btrfs_path
*path
,
1264 u64 bytenr
, u64 parent
,
1265 u64 root_objectid
, u64 owner
,
1266 u64 offset
, int refs_to_add
)
1268 struct btrfs_key key
;
1269 struct extent_buffer
*leaf
;
1274 key
.objectid
= bytenr
;
1276 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1277 key
.offset
= parent
;
1278 size
= sizeof(struct btrfs_shared_data_ref
);
1280 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1281 key
.offset
= hash_extent_data_ref(root_objectid
,
1283 size
= sizeof(struct btrfs_extent_data_ref
);
1286 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, size
);
1287 if (ret
&& ret
!= -EEXIST
)
1290 leaf
= path
->nodes
[0];
1292 struct btrfs_shared_data_ref
*ref
;
1293 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1294 struct btrfs_shared_data_ref
);
1296 btrfs_set_shared_data_ref_count(leaf
, ref
, refs_to_add
);
1298 num_refs
= btrfs_shared_data_ref_count(leaf
, ref
);
1299 num_refs
+= refs_to_add
;
1300 btrfs_set_shared_data_ref_count(leaf
, ref
, num_refs
);
1303 struct btrfs_extent_data_ref
*ref
;
1304 while (ret
== -EEXIST
) {
1305 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1306 struct btrfs_extent_data_ref
);
1307 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1310 btrfs_release_path(path
);
1312 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1314 if (ret
&& ret
!= -EEXIST
)
1317 leaf
= path
->nodes
[0];
1319 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1320 struct btrfs_extent_data_ref
);
1322 btrfs_set_extent_data_ref_root(leaf
, ref
,
1324 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
1325 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
1326 btrfs_set_extent_data_ref_count(leaf
, ref
, refs_to_add
);
1328 num_refs
= btrfs_extent_data_ref_count(leaf
, ref
);
1329 num_refs
+= refs_to_add
;
1330 btrfs_set_extent_data_ref_count(leaf
, ref
, num_refs
);
1333 btrfs_mark_buffer_dirty(leaf
);
1336 btrfs_release_path(path
);
1340 static noinline
int remove_extent_data_ref(struct btrfs_trans_handle
*trans
,
1341 struct btrfs_root
*root
,
1342 struct btrfs_path
*path
,
1343 int refs_to_drop
, int *last_ref
)
1345 struct btrfs_key key
;
1346 struct btrfs_extent_data_ref
*ref1
= NULL
;
1347 struct btrfs_shared_data_ref
*ref2
= NULL
;
1348 struct extent_buffer
*leaf
;
1352 leaf
= path
->nodes
[0];
1353 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1355 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1356 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1357 struct btrfs_extent_data_ref
);
1358 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1359 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1360 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1361 struct btrfs_shared_data_ref
);
1362 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1363 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1364 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1365 struct btrfs_extent_ref_v0
*ref0
;
1366 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1367 struct btrfs_extent_ref_v0
);
1368 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1374 BUG_ON(num_refs
< refs_to_drop
);
1375 num_refs
-= refs_to_drop
;
1377 if (num_refs
== 0) {
1378 ret
= btrfs_del_item(trans
, root
, path
);
1381 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
)
1382 btrfs_set_extent_data_ref_count(leaf
, ref1
, num_refs
);
1383 else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
)
1384 btrfs_set_shared_data_ref_count(leaf
, ref2
, num_refs
);
1385 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1387 struct btrfs_extent_ref_v0
*ref0
;
1388 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1389 struct btrfs_extent_ref_v0
);
1390 btrfs_set_ref_count_v0(leaf
, ref0
, num_refs
);
1393 btrfs_mark_buffer_dirty(leaf
);
1398 static noinline u32
extent_data_ref_count(struct btrfs_path
*path
,
1399 struct btrfs_extent_inline_ref
*iref
)
1401 struct btrfs_key key
;
1402 struct extent_buffer
*leaf
;
1403 struct btrfs_extent_data_ref
*ref1
;
1404 struct btrfs_shared_data_ref
*ref2
;
1407 leaf
= path
->nodes
[0];
1408 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1410 if (btrfs_extent_inline_ref_type(leaf
, iref
) ==
1411 BTRFS_EXTENT_DATA_REF_KEY
) {
1412 ref1
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1413 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1415 ref2
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1416 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1418 } else if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1419 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1420 struct btrfs_extent_data_ref
);
1421 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1422 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1423 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1424 struct btrfs_shared_data_ref
);
1425 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1426 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1427 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1428 struct btrfs_extent_ref_v0
*ref0
;
1429 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1430 struct btrfs_extent_ref_v0
);
1431 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1439 static noinline
int lookup_tree_block_ref(struct btrfs_trans_handle
*trans
,
1440 struct btrfs_root
*root
,
1441 struct btrfs_path
*path
,
1442 u64 bytenr
, u64 parent
,
1445 struct btrfs_key key
;
1448 key
.objectid
= bytenr
;
1450 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1451 key
.offset
= parent
;
1453 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1454 key
.offset
= root_objectid
;
1457 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1460 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1461 if (ret
== -ENOENT
&& parent
) {
1462 btrfs_release_path(path
);
1463 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1464 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1472 static noinline
int insert_tree_block_ref(struct btrfs_trans_handle
*trans
,
1473 struct btrfs_root
*root
,
1474 struct btrfs_path
*path
,
1475 u64 bytenr
, u64 parent
,
1478 struct btrfs_key key
;
1481 key
.objectid
= bytenr
;
1483 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1484 key
.offset
= parent
;
1486 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1487 key
.offset
= root_objectid
;
1490 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
1491 btrfs_release_path(path
);
1495 static inline int extent_ref_type(u64 parent
, u64 owner
)
1498 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1500 type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1502 type
= BTRFS_TREE_BLOCK_REF_KEY
;
1505 type
= BTRFS_SHARED_DATA_REF_KEY
;
1507 type
= BTRFS_EXTENT_DATA_REF_KEY
;
1512 static int find_next_key(struct btrfs_path
*path
, int level
,
1513 struct btrfs_key
*key
)
1516 for (; level
< BTRFS_MAX_LEVEL
; level
++) {
1517 if (!path
->nodes
[level
])
1519 if (path
->slots
[level
] + 1 >=
1520 btrfs_header_nritems(path
->nodes
[level
]))
1523 btrfs_item_key_to_cpu(path
->nodes
[level
], key
,
1524 path
->slots
[level
] + 1);
1526 btrfs_node_key_to_cpu(path
->nodes
[level
], key
,
1527 path
->slots
[level
] + 1);
1534 * look for inline back ref. if back ref is found, *ref_ret is set
1535 * to the address of inline back ref, and 0 is returned.
1537 * if back ref isn't found, *ref_ret is set to the address where it
1538 * should be inserted, and -ENOENT is returned.
1540 * if insert is true and there are too many inline back refs, the path
1541 * points to the extent item, and -EAGAIN is returned.
1543 * NOTE: inline back refs are ordered in the same way that back ref
1544 * items in the tree are ordered.
1546 static noinline_for_stack
1547 int lookup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1548 struct btrfs_root
*root
,
1549 struct btrfs_path
*path
,
1550 struct btrfs_extent_inline_ref
**ref_ret
,
1551 u64 bytenr
, u64 num_bytes
,
1552 u64 parent
, u64 root_objectid
,
1553 u64 owner
, u64 offset
, int insert
)
1555 struct btrfs_key key
;
1556 struct extent_buffer
*leaf
;
1557 struct btrfs_extent_item
*ei
;
1558 struct btrfs_extent_inline_ref
*iref
;
1568 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
1571 key
.objectid
= bytenr
;
1572 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1573 key
.offset
= num_bytes
;
1575 want
= extent_ref_type(parent
, owner
);
1577 extra_size
= btrfs_extent_inline_ref_size(want
);
1578 path
->keep_locks
= 1;
1583 * Owner is our parent level, so we can just add one to get the level
1584 * for the block we are interested in.
1586 if (skinny_metadata
&& owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1587 key
.type
= BTRFS_METADATA_ITEM_KEY
;
1592 ret
= btrfs_search_slot(trans
, root
, &key
, path
, extra_size
, 1);
1599 * We may be a newly converted file system which still has the old fat
1600 * extent entries for metadata, so try and see if we have one of those.
1602 if (ret
> 0 && skinny_metadata
) {
1603 skinny_metadata
= false;
1604 if (path
->slots
[0]) {
1606 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
1608 if (key
.objectid
== bytenr
&&
1609 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
1610 key
.offset
== num_bytes
)
1614 key
.objectid
= bytenr
;
1615 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1616 key
.offset
= num_bytes
;
1617 btrfs_release_path(path
);
1622 if (ret
&& !insert
) {
1625 } else if (WARN_ON(ret
)) {
1630 leaf
= path
->nodes
[0];
1631 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1632 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1633 if (item_size
< sizeof(*ei
)) {
1638 ret
= convert_extent_item_v0(trans
, root
, path
, owner
,
1644 leaf
= path
->nodes
[0];
1645 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1648 BUG_ON(item_size
< sizeof(*ei
));
1650 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1651 flags
= btrfs_extent_flags(leaf
, ei
);
1653 ptr
= (unsigned long)(ei
+ 1);
1654 end
= (unsigned long)ei
+ item_size
;
1656 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
&& !skinny_metadata
) {
1657 ptr
+= sizeof(struct btrfs_tree_block_info
);
1667 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1668 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1672 ptr
+= btrfs_extent_inline_ref_size(type
);
1676 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1677 struct btrfs_extent_data_ref
*dref
;
1678 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1679 if (match_extent_data_ref(leaf
, dref
, root_objectid
,
1684 if (hash_extent_data_ref_item(leaf
, dref
) <
1685 hash_extent_data_ref(root_objectid
, owner
, offset
))
1689 ref_offset
= btrfs_extent_inline_ref_offset(leaf
, iref
);
1691 if (parent
== ref_offset
) {
1695 if (ref_offset
< parent
)
1698 if (root_objectid
== ref_offset
) {
1702 if (ref_offset
< root_objectid
)
1706 ptr
+= btrfs_extent_inline_ref_size(type
);
1708 if (err
== -ENOENT
&& insert
) {
1709 if (item_size
+ extra_size
>=
1710 BTRFS_MAX_EXTENT_ITEM_SIZE(root
)) {
1715 * To add new inline back ref, we have to make sure
1716 * there is no corresponding back ref item.
1717 * For simplicity, we just do not add new inline back
1718 * ref if there is any kind of item for this block
1720 if (find_next_key(path
, 0, &key
) == 0 &&
1721 key
.objectid
== bytenr
&&
1722 key
.type
< BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1727 *ref_ret
= (struct btrfs_extent_inline_ref
*)ptr
;
1730 path
->keep_locks
= 0;
1731 btrfs_unlock_up_safe(path
, 1);
1737 * helper to add new inline back ref
1739 static noinline_for_stack
1740 void setup_inline_extent_backref(struct btrfs_root
*root
,
1741 struct btrfs_path
*path
,
1742 struct btrfs_extent_inline_ref
*iref
,
1743 u64 parent
, u64 root_objectid
,
1744 u64 owner
, u64 offset
, int refs_to_add
,
1745 struct btrfs_delayed_extent_op
*extent_op
)
1747 struct extent_buffer
*leaf
;
1748 struct btrfs_extent_item
*ei
;
1751 unsigned long item_offset
;
1756 leaf
= path
->nodes
[0];
1757 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1758 item_offset
= (unsigned long)iref
- (unsigned long)ei
;
1760 type
= extent_ref_type(parent
, owner
);
1761 size
= btrfs_extent_inline_ref_size(type
);
1763 btrfs_extend_item(root
, path
, size
);
1765 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1766 refs
= btrfs_extent_refs(leaf
, ei
);
1767 refs
+= refs_to_add
;
1768 btrfs_set_extent_refs(leaf
, ei
, refs
);
1770 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1772 ptr
= (unsigned long)ei
+ item_offset
;
1773 end
= (unsigned long)ei
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
1774 if (ptr
< end
- size
)
1775 memmove_extent_buffer(leaf
, ptr
+ size
, ptr
,
1778 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1779 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
1780 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1781 struct btrfs_extent_data_ref
*dref
;
1782 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1783 btrfs_set_extent_data_ref_root(leaf
, dref
, root_objectid
);
1784 btrfs_set_extent_data_ref_objectid(leaf
, dref
, owner
);
1785 btrfs_set_extent_data_ref_offset(leaf
, dref
, offset
);
1786 btrfs_set_extent_data_ref_count(leaf
, dref
, refs_to_add
);
1787 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1788 struct btrfs_shared_data_ref
*sref
;
1789 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1790 btrfs_set_shared_data_ref_count(leaf
, sref
, refs_to_add
);
1791 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1792 } else if (type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
1793 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1795 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
1797 btrfs_mark_buffer_dirty(leaf
);
1800 static int lookup_extent_backref(struct btrfs_trans_handle
*trans
,
1801 struct btrfs_root
*root
,
1802 struct btrfs_path
*path
,
1803 struct btrfs_extent_inline_ref
**ref_ret
,
1804 u64 bytenr
, u64 num_bytes
, u64 parent
,
1805 u64 root_objectid
, u64 owner
, u64 offset
)
1809 ret
= lookup_inline_extent_backref(trans
, root
, path
, ref_ret
,
1810 bytenr
, num_bytes
, parent
,
1811 root_objectid
, owner
, offset
, 0);
1815 btrfs_release_path(path
);
1818 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1819 ret
= lookup_tree_block_ref(trans
, root
, path
, bytenr
, parent
,
1822 ret
= lookup_extent_data_ref(trans
, root
, path
, bytenr
, parent
,
1823 root_objectid
, owner
, offset
);
1829 * helper to update/remove inline back ref
1831 static noinline_for_stack
1832 void update_inline_extent_backref(struct btrfs_root
*root
,
1833 struct btrfs_path
*path
,
1834 struct btrfs_extent_inline_ref
*iref
,
1836 struct btrfs_delayed_extent_op
*extent_op
,
1839 struct extent_buffer
*leaf
;
1840 struct btrfs_extent_item
*ei
;
1841 struct btrfs_extent_data_ref
*dref
= NULL
;
1842 struct btrfs_shared_data_ref
*sref
= NULL
;
1850 leaf
= path
->nodes
[0];
1851 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1852 refs
= btrfs_extent_refs(leaf
, ei
);
1853 WARN_ON(refs_to_mod
< 0 && refs
+ refs_to_mod
<= 0);
1854 refs
+= refs_to_mod
;
1855 btrfs_set_extent_refs(leaf
, ei
, refs
);
1857 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1859 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1861 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1862 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1863 refs
= btrfs_extent_data_ref_count(leaf
, dref
);
1864 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1865 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1866 refs
= btrfs_shared_data_ref_count(leaf
, sref
);
1869 BUG_ON(refs_to_mod
!= -1);
1872 BUG_ON(refs_to_mod
< 0 && refs
< -refs_to_mod
);
1873 refs
+= refs_to_mod
;
1876 if (type
== BTRFS_EXTENT_DATA_REF_KEY
)
1877 btrfs_set_extent_data_ref_count(leaf
, dref
, refs
);
1879 btrfs_set_shared_data_ref_count(leaf
, sref
, refs
);
1882 size
= btrfs_extent_inline_ref_size(type
);
1883 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1884 ptr
= (unsigned long)iref
;
1885 end
= (unsigned long)ei
+ item_size
;
1886 if (ptr
+ size
< end
)
1887 memmove_extent_buffer(leaf
, ptr
, ptr
+ size
,
1890 btrfs_truncate_item(root
, path
, item_size
, 1);
1892 btrfs_mark_buffer_dirty(leaf
);
1895 static noinline_for_stack
1896 int insert_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1897 struct btrfs_root
*root
,
1898 struct btrfs_path
*path
,
1899 u64 bytenr
, u64 num_bytes
, u64 parent
,
1900 u64 root_objectid
, u64 owner
,
1901 u64 offset
, int refs_to_add
,
1902 struct btrfs_delayed_extent_op
*extent_op
)
1904 struct btrfs_extent_inline_ref
*iref
;
1907 ret
= lookup_inline_extent_backref(trans
, root
, path
, &iref
,
1908 bytenr
, num_bytes
, parent
,
1909 root_objectid
, owner
, offset
, 1);
1911 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
);
1912 update_inline_extent_backref(root
, path
, iref
,
1913 refs_to_add
, extent_op
, NULL
);
1914 } else if (ret
== -ENOENT
) {
1915 setup_inline_extent_backref(root
, path
, iref
, parent
,
1916 root_objectid
, owner
, offset
,
1917 refs_to_add
, extent_op
);
1923 static int insert_extent_backref(struct btrfs_trans_handle
*trans
,
1924 struct btrfs_root
*root
,
1925 struct btrfs_path
*path
,
1926 u64 bytenr
, u64 parent
, u64 root_objectid
,
1927 u64 owner
, u64 offset
, int refs_to_add
)
1930 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1931 BUG_ON(refs_to_add
!= 1);
1932 ret
= insert_tree_block_ref(trans
, root
, path
, bytenr
,
1933 parent
, root_objectid
);
1935 ret
= insert_extent_data_ref(trans
, root
, path
, bytenr
,
1936 parent
, root_objectid
,
1937 owner
, offset
, refs_to_add
);
1942 static int remove_extent_backref(struct btrfs_trans_handle
*trans
,
1943 struct btrfs_root
*root
,
1944 struct btrfs_path
*path
,
1945 struct btrfs_extent_inline_ref
*iref
,
1946 int refs_to_drop
, int is_data
, int *last_ref
)
1950 BUG_ON(!is_data
&& refs_to_drop
!= 1);
1952 update_inline_extent_backref(root
, path
, iref
,
1953 -refs_to_drop
, NULL
, last_ref
);
1954 } else if (is_data
) {
1955 ret
= remove_extent_data_ref(trans
, root
, path
, refs_to_drop
,
1959 ret
= btrfs_del_item(trans
, root
, path
);
1964 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
1965 static int btrfs_issue_discard(struct block_device
*bdev
, u64 start
, u64 len
,
1966 u64
*discarded_bytes
)
1969 u64 bytes_left
, end
;
1970 u64 aligned_start
= ALIGN(start
, 1 << 9);
1972 if (WARN_ON(start
!= aligned_start
)) {
1973 len
-= aligned_start
- start
;
1974 len
= round_down(len
, 1 << 9);
1975 start
= aligned_start
;
1978 *discarded_bytes
= 0;
1986 /* Skip any superblocks on this device. */
1987 for (j
= 0; j
< BTRFS_SUPER_MIRROR_MAX
; j
++) {
1988 u64 sb_start
= btrfs_sb_offset(j
);
1989 u64 sb_end
= sb_start
+ BTRFS_SUPER_INFO_SIZE
;
1990 u64 size
= sb_start
- start
;
1992 if (!in_range(sb_start
, start
, bytes_left
) &&
1993 !in_range(sb_end
, start
, bytes_left
) &&
1994 !in_range(start
, sb_start
, BTRFS_SUPER_INFO_SIZE
))
1998 * Superblock spans beginning of range. Adjust start and
2001 if (sb_start
<= start
) {
2002 start
+= sb_end
- start
;
2007 bytes_left
= end
- start
;
2012 ret
= blkdev_issue_discard(bdev
, start
>> 9, size
>> 9,
2015 *discarded_bytes
+= size
;
2016 else if (ret
!= -EOPNOTSUPP
)
2025 bytes_left
= end
- start
;
2029 ret
= blkdev_issue_discard(bdev
, start
>> 9, bytes_left
>> 9,
2032 *discarded_bytes
+= bytes_left
;
2037 int btrfs_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
2038 u64 num_bytes
, u64
*actual_bytes
)
2041 u64 discarded_bytes
= 0;
2042 struct btrfs_bio
*bbio
= NULL
;
2045 /* Tell the block device(s) that the sectors can be discarded */
2046 ret
= btrfs_map_block(root
->fs_info
, REQ_DISCARD
,
2047 bytenr
, &num_bytes
, &bbio
, 0);
2048 /* Error condition is -ENOMEM */
2050 struct btrfs_bio_stripe
*stripe
= bbio
->stripes
;
2054 for (i
= 0; i
< bbio
->num_stripes
; i
++, stripe
++) {
2056 if (!stripe
->dev
->can_discard
)
2059 ret
= btrfs_issue_discard(stripe
->dev
->bdev
,
2064 discarded_bytes
+= bytes
;
2065 else if (ret
!= -EOPNOTSUPP
)
2066 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2069 * Just in case we get back EOPNOTSUPP for some reason,
2070 * just ignore the return value so we don't screw up
2071 * people calling discard_extent.
2075 btrfs_put_bbio(bbio
);
2079 *actual_bytes
= discarded_bytes
;
2082 if (ret
== -EOPNOTSUPP
)
2087 /* Can return -ENOMEM */
2088 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
2089 struct btrfs_root
*root
,
2090 u64 bytenr
, u64 num_bytes
, u64 parent
,
2091 u64 root_objectid
, u64 owner
, u64 offset
)
2094 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2096 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
&&
2097 root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
2099 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
2100 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
2102 parent
, root_objectid
, (int)owner
,
2103 BTRFS_ADD_DELAYED_REF
, NULL
);
2105 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
2106 num_bytes
, parent
, root_objectid
,
2108 BTRFS_ADD_DELAYED_REF
, NULL
);
2113 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
2114 struct btrfs_root
*root
,
2115 struct btrfs_delayed_ref_node
*node
,
2116 u64 parent
, u64 root_objectid
,
2117 u64 owner
, u64 offset
, int refs_to_add
,
2118 struct btrfs_delayed_extent_op
*extent_op
)
2120 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2121 struct btrfs_path
*path
;
2122 struct extent_buffer
*leaf
;
2123 struct btrfs_extent_item
*item
;
2124 struct btrfs_key key
;
2125 u64 bytenr
= node
->bytenr
;
2126 u64 num_bytes
= node
->num_bytes
;
2130 path
= btrfs_alloc_path();
2134 path
->reada
= READA_FORWARD
;
2135 path
->leave_spinning
= 1;
2136 /* this will setup the path even if it fails to insert the back ref */
2137 ret
= insert_inline_extent_backref(trans
, fs_info
->extent_root
, path
,
2138 bytenr
, num_bytes
, parent
,
2139 root_objectid
, owner
, offset
,
2140 refs_to_add
, extent_op
);
2141 if ((ret
< 0 && ret
!= -EAGAIN
) || !ret
)
2145 * Ok we had -EAGAIN which means we didn't have space to insert and
2146 * inline extent ref, so just update the reference count and add a
2149 leaf
= path
->nodes
[0];
2150 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2151 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2152 refs
= btrfs_extent_refs(leaf
, item
);
2153 btrfs_set_extent_refs(leaf
, item
, refs
+ refs_to_add
);
2155 __run_delayed_extent_op(extent_op
, leaf
, item
);
2157 btrfs_mark_buffer_dirty(leaf
);
2158 btrfs_release_path(path
);
2160 path
->reada
= READA_FORWARD
;
2161 path
->leave_spinning
= 1;
2162 /* now insert the actual backref */
2163 ret
= insert_extent_backref(trans
, root
->fs_info
->extent_root
,
2164 path
, bytenr
, parent
, root_objectid
,
2165 owner
, offset
, refs_to_add
);
2167 btrfs_abort_transaction(trans
, root
, ret
);
2169 btrfs_free_path(path
);
2173 static int run_delayed_data_ref(struct btrfs_trans_handle
*trans
,
2174 struct btrfs_root
*root
,
2175 struct btrfs_delayed_ref_node
*node
,
2176 struct btrfs_delayed_extent_op
*extent_op
,
2177 int insert_reserved
)
2180 struct btrfs_delayed_data_ref
*ref
;
2181 struct btrfs_key ins
;
2186 ins
.objectid
= node
->bytenr
;
2187 ins
.offset
= node
->num_bytes
;
2188 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2190 ref
= btrfs_delayed_node_to_data_ref(node
);
2191 trace_run_delayed_data_ref(node
, ref
, node
->action
);
2193 if (node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2194 parent
= ref
->parent
;
2195 ref_root
= ref
->root
;
2197 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2199 flags
|= extent_op
->flags_to_set
;
2200 ret
= alloc_reserved_file_extent(trans
, root
,
2201 parent
, ref_root
, flags
,
2202 ref
->objectid
, ref
->offset
,
2203 &ins
, node
->ref_mod
);
2204 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2205 ret
= __btrfs_inc_extent_ref(trans
, root
, node
, parent
,
2206 ref_root
, ref
->objectid
,
2207 ref
->offset
, node
->ref_mod
,
2209 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2210 ret
= __btrfs_free_extent(trans
, root
, node
, parent
,
2211 ref_root
, ref
->objectid
,
2212 ref
->offset
, node
->ref_mod
,
2220 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
2221 struct extent_buffer
*leaf
,
2222 struct btrfs_extent_item
*ei
)
2224 u64 flags
= btrfs_extent_flags(leaf
, ei
);
2225 if (extent_op
->update_flags
) {
2226 flags
|= extent_op
->flags_to_set
;
2227 btrfs_set_extent_flags(leaf
, ei
, flags
);
2230 if (extent_op
->update_key
) {
2231 struct btrfs_tree_block_info
*bi
;
2232 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
));
2233 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
2234 btrfs_set_tree_block_key(leaf
, bi
, &extent_op
->key
);
2238 static int run_delayed_extent_op(struct btrfs_trans_handle
*trans
,
2239 struct btrfs_root
*root
,
2240 struct btrfs_delayed_ref_node
*node
,
2241 struct btrfs_delayed_extent_op
*extent_op
)
2243 struct btrfs_key key
;
2244 struct btrfs_path
*path
;
2245 struct btrfs_extent_item
*ei
;
2246 struct extent_buffer
*leaf
;
2250 int metadata
= !extent_op
->is_data
;
2255 if (metadata
&& !btrfs_fs_incompat(root
->fs_info
, SKINNY_METADATA
))
2258 path
= btrfs_alloc_path();
2262 key
.objectid
= node
->bytenr
;
2265 key
.type
= BTRFS_METADATA_ITEM_KEY
;
2266 key
.offset
= extent_op
->level
;
2268 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2269 key
.offset
= node
->num_bytes
;
2273 path
->reada
= READA_FORWARD
;
2274 path
->leave_spinning
= 1;
2275 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
,
2283 if (path
->slots
[0] > 0) {
2285 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
2287 if (key
.objectid
== node
->bytenr
&&
2288 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
2289 key
.offset
== node
->num_bytes
)
2293 btrfs_release_path(path
);
2296 key
.objectid
= node
->bytenr
;
2297 key
.offset
= node
->num_bytes
;
2298 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2307 leaf
= path
->nodes
[0];
2308 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2309 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2310 if (item_size
< sizeof(*ei
)) {
2311 ret
= convert_extent_item_v0(trans
, root
->fs_info
->extent_root
,
2317 leaf
= path
->nodes
[0];
2318 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2321 BUG_ON(item_size
< sizeof(*ei
));
2322 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2323 __run_delayed_extent_op(extent_op
, leaf
, ei
);
2325 btrfs_mark_buffer_dirty(leaf
);
2327 btrfs_free_path(path
);
2331 static int run_delayed_tree_ref(struct btrfs_trans_handle
*trans
,
2332 struct btrfs_root
*root
,
2333 struct btrfs_delayed_ref_node
*node
,
2334 struct btrfs_delayed_extent_op
*extent_op
,
2335 int insert_reserved
)
2338 struct btrfs_delayed_tree_ref
*ref
;
2339 struct btrfs_key ins
;
2342 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
2345 ref
= btrfs_delayed_node_to_tree_ref(node
);
2346 trace_run_delayed_tree_ref(node
, ref
, node
->action
);
2348 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2349 parent
= ref
->parent
;
2350 ref_root
= ref
->root
;
2352 ins
.objectid
= node
->bytenr
;
2353 if (skinny_metadata
) {
2354 ins
.offset
= ref
->level
;
2355 ins
.type
= BTRFS_METADATA_ITEM_KEY
;
2357 ins
.offset
= node
->num_bytes
;
2358 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2361 BUG_ON(node
->ref_mod
!= 1);
2362 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2363 BUG_ON(!extent_op
|| !extent_op
->update_flags
);
2364 ret
= alloc_reserved_tree_block(trans
, root
,
2366 extent_op
->flags_to_set
,
2369 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2370 ret
= __btrfs_inc_extent_ref(trans
, root
, node
,
2374 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2375 ret
= __btrfs_free_extent(trans
, root
, node
,
2377 ref
->level
, 0, 1, extent_op
);
2384 /* helper function to actually process a single delayed ref entry */
2385 static int run_one_delayed_ref(struct btrfs_trans_handle
*trans
,
2386 struct btrfs_root
*root
,
2387 struct btrfs_delayed_ref_node
*node
,
2388 struct btrfs_delayed_extent_op
*extent_op
,
2389 int insert_reserved
)
2393 if (trans
->aborted
) {
2394 if (insert_reserved
)
2395 btrfs_pin_extent(root
, node
->bytenr
,
2396 node
->num_bytes
, 1);
2400 if (btrfs_delayed_ref_is_head(node
)) {
2401 struct btrfs_delayed_ref_head
*head
;
2403 * we've hit the end of the chain and we were supposed
2404 * to insert this extent into the tree. But, it got
2405 * deleted before we ever needed to insert it, so all
2406 * we have to do is clean up the accounting
2409 head
= btrfs_delayed_node_to_head(node
);
2410 trace_run_delayed_ref_head(node
, head
, node
->action
);
2412 if (insert_reserved
) {
2413 btrfs_pin_extent(root
, node
->bytenr
,
2414 node
->num_bytes
, 1);
2415 if (head
->is_data
) {
2416 ret
= btrfs_del_csums(trans
, root
,
2422 /* Also free its reserved qgroup space */
2423 btrfs_qgroup_free_delayed_ref(root
->fs_info
,
2424 head
->qgroup_ref_root
,
2425 head
->qgroup_reserved
);
2429 if (node
->type
== BTRFS_TREE_BLOCK_REF_KEY
||
2430 node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2431 ret
= run_delayed_tree_ref(trans
, root
, node
, extent_op
,
2433 else if (node
->type
== BTRFS_EXTENT_DATA_REF_KEY
||
2434 node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2435 ret
= run_delayed_data_ref(trans
, root
, node
, extent_op
,
2442 static inline struct btrfs_delayed_ref_node
*
2443 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
2445 struct btrfs_delayed_ref_node
*ref
;
2447 if (list_empty(&head
->ref_list
))
2451 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2452 * This is to prevent a ref count from going down to zero, which deletes
2453 * the extent item from the extent tree, when there still are references
2454 * to add, which would fail because they would not find the extent item.
2456 list_for_each_entry(ref
, &head
->ref_list
, list
) {
2457 if (ref
->action
== BTRFS_ADD_DELAYED_REF
)
2461 return list_entry(head
->ref_list
.next
, struct btrfs_delayed_ref_node
,
2466 * Returns 0 on success or if called with an already aborted transaction.
2467 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2469 static noinline
int __btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2470 struct btrfs_root
*root
,
2473 struct btrfs_delayed_ref_root
*delayed_refs
;
2474 struct btrfs_delayed_ref_node
*ref
;
2475 struct btrfs_delayed_ref_head
*locked_ref
= NULL
;
2476 struct btrfs_delayed_extent_op
*extent_op
;
2477 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2478 ktime_t start
= ktime_get();
2480 unsigned long count
= 0;
2481 unsigned long actual_count
= 0;
2482 int must_insert_reserved
= 0;
2484 delayed_refs
= &trans
->transaction
->delayed_refs
;
2490 spin_lock(&delayed_refs
->lock
);
2491 locked_ref
= btrfs_select_ref_head(trans
);
2493 spin_unlock(&delayed_refs
->lock
);
2497 /* grab the lock that says we are going to process
2498 * all the refs for this head */
2499 ret
= btrfs_delayed_ref_lock(trans
, locked_ref
);
2500 spin_unlock(&delayed_refs
->lock
);
2502 * we may have dropped the spin lock to get the head
2503 * mutex lock, and that might have given someone else
2504 * time to free the head. If that's true, it has been
2505 * removed from our list and we can move on.
2507 if (ret
== -EAGAIN
) {
2515 * We need to try and merge add/drops of the same ref since we
2516 * can run into issues with relocate dropping the implicit ref
2517 * and then it being added back again before the drop can
2518 * finish. If we merged anything we need to re-loop so we can
2520 * Or we can get node references of the same type that weren't
2521 * merged when created due to bumps in the tree mod seq, and
2522 * we need to merge them to prevent adding an inline extent
2523 * backref before dropping it (triggering a BUG_ON at
2524 * insert_inline_extent_backref()).
2526 spin_lock(&locked_ref
->lock
);
2527 btrfs_merge_delayed_refs(trans
, fs_info
, delayed_refs
,
2531 * locked_ref is the head node, so we have to go one
2532 * node back for any delayed ref updates
2534 ref
= select_delayed_ref(locked_ref
);
2536 if (ref
&& ref
->seq
&&
2537 btrfs_check_delayed_seq(fs_info
, delayed_refs
, ref
->seq
)) {
2538 spin_unlock(&locked_ref
->lock
);
2539 btrfs_delayed_ref_unlock(locked_ref
);
2540 spin_lock(&delayed_refs
->lock
);
2541 locked_ref
->processing
= 0;
2542 delayed_refs
->num_heads_ready
++;
2543 spin_unlock(&delayed_refs
->lock
);
2551 * record the must insert reserved flag before we
2552 * drop the spin lock.
2554 must_insert_reserved
= locked_ref
->must_insert_reserved
;
2555 locked_ref
->must_insert_reserved
= 0;
2557 extent_op
= locked_ref
->extent_op
;
2558 locked_ref
->extent_op
= NULL
;
2563 /* All delayed refs have been processed, Go ahead
2564 * and send the head node to run_one_delayed_ref,
2565 * so that any accounting fixes can happen
2567 ref
= &locked_ref
->node
;
2569 if (extent_op
&& must_insert_reserved
) {
2570 btrfs_free_delayed_extent_op(extent_op
);
2575 spin_unlock(&locked_ref
->lock
);
2576 ret
= run_delayed_extent_op(trans
, root
,
2578 btrfs_free_delayed_extent_op(extent_op
);
2582 * Need to reset must_insert_reserved if
2583 * there was an error so the abort stuff
2584 * can cleanup the reserved space
2587 if (must_insert_reserved
)
2588 locked_ref
->must_insert_reserved
= 1;
2589 locked_ref
->processing
= 0;
2590 btrfs_debug(fs_info
, "run_delayed_extent_op returned %d", ret
);
2591 btrfs_delayed_ref_unlock(locked_ref
);
2598 * Need to drop our head ref lock and re-aqcuire the
2599 * delayed ref lock and then re-check to make sure
2602 spin_unlock(&locked_ref
->lock
);
2603 spin_lock(&delayed_refs
->lock
);
2604 spin_lock(&locked_ref
->lock
);
2605 if (!list_empty(&locked_ref
->ref_list
) ||
2606 locked_ref
->extent_op
) {
2607 spin_unlock(&locked_ref
->lock
);
2608 spin_unlock(&delayed_refs
->lock
);
2612 delayed_refs
->num_heads
--;
2613 rb_erase(&locked_ref
->href_node
,
2614 &delayed_refs
->href_root
);
2615 spin_unlock(&delayed_refs
->lock
);
2619 list_del(&ref
->list
);
2621 atomic_dec(&delayed_refs
->num_entries
);
2623 if (!btrfs_delayed_ref_is_head(ref
)) {
2625 * when we play the delayed ref, also correct the
2628 switch (ref
->action
) {
2629 case BTRFS_ADD_DELAYED_REF
:
2630 case BTRFS_ADD_DELAYED_EXTENT
:
2631 locked_ref
->node
.ref_mod
-= ref
->ref_mod
;
2633 case BTRFS_DROP_DELAYED_REF
:
2634 locked_ref
->node
.ref_mod
+= ref
->ref_mod
;
2640 spin_unlock(&locked_ref
->lock
);
2642 ret
= run_one_delayed_ref(trans
, root
, ref
, extent_op
,
2643 must_insert_reserved
);
2645 btrfs_free_delayed_extent_op(extent_op
);
2647 locked_ref
->processing
= 0;
2648 btrfs_delayed_ref_unlock(locked_ref
);
2649 btrfs_put_delayed_ref(ref
);
2650 btrfs_debug(fs_info
, "run_one_delayed_ref returned %d", ret
);
2655 * If this node is a head, that means all the refs in this head
2656 * have been dealt with, and we will pick the next head to deal
2657 * with, so we must unlock the head and drop it from the cluster
2658 * list before we release it.
2660 if (btrfs_delayed_ref_is_head(ref
)) {
2661 if (locked_ref
->is_data
&&
2662 locked_ref
->total_ref_mod
< 0) {
2663 spin_lock(&delayed_refs
->lock
);
2664 delayed_refs
->pending_csums
-= ref
->num_bytes
;
2665 spin_unlock(&delayed_refs
->lock
);
2667 btrfs_delayed_ref_unlock(locked_ref
);
2670 btrfs_put_delayed_ref(ref
);
2676 * We don't want to include ref heads since we can have empty ref heads
2677 * and those will drastically skew our runtime down since we just do
2678 * accounting, no actual extent tree updates.
2680 if (actual_count
> 0) {
2681 u64 runtime
= ktime_to_ns(ktime_sub(ktime_get(), start
));
2685 * We weigh the current average higher than our current runtime
2686 * to avoid large swings in the average.
2688 spin_lock(&delayed_refs
->lock
);
2689 avg
= fs_info
->avg_delayed_ref_runtime
* 3 + runtime
;
2690 fs_info
->avg_delayed_ref_runtime
= avg
>> 2; /* div by 4 */
2691 spin_unlock(&delayed_refs
->lock
);
2696 #ifdef SCRAMBLE_DELAYED_REFS
2698 * Normally delayed refs get processed in ascending bytenr order. This
2699 * correlates in most cases to the order added. To expose dependencies on this
2700 * order, we start to process the tree in the middle instead of the beginning
2702 static u64
find_middle(struct rb_root
*root
)
2704 struct rb_node
*n
= root
->rb_node
;
2705 struct btrfs_delayed_ref_node
*entry
;
2708 u64 first
= 0, last
= 0;
2712 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2713 first
= entry
->bytenr
;
2717 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2718 last
= entry
->bytenr
;
2723 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2724 WARN_ON(!entry
->in_tree
);
2726 middle
= entry
->bytenr
;
2739 static inline u64
heads_to_leaves(struct btrfs_root
*root
, u64 heads
)
2743 num_bytes
= heads
* (sizeof(struct btrfs_extent_item
) +
2744 sizeof(struct btrfs_extent_inline_ref
));
2745 if (!btrfs_fs_incompat(root
->fs_info
, SKINNY_METADATA
))
2746 num_bytes
+= heads
* sizeof(struct btrfs_tree_block_info
);
2749 * We don't ever fill up leaves all the way so multiply by 2 just to be
2750 * closer to what we're really going to want to ouse.
2752 return div_u64(num_bytes
, BTRFS_LEAF_DATA_SIZE(root
));
2756 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2757 * would require to store the csums for that many bytes.
2759 u64
btrfs_csum_bytes_to_leaves(struct btrfs_root
*root
, u64 csum_bytes
)
2762 u64 num_csums_per_leaf
;
2765 csum_size
= BTRFS_LEAF_DATA_SIZE(root
) - sizeof(struct btrfs_item
);
2766 num_csums_per_leaf
= div64_u64(csum_size
,
2767 (u64
)btrfs_super_csum_size(root
->fs_info
->super_copy
));
2768 num_csums
= div64_u64(csum_bytes
, root
->sectorsize
);
2769 num_csums
+= num_csums_per_leaf
- 1;
2770 num_csums
= div64_u64(num_csums
, num_csums_per_leaf
);
2774 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle
*trans
,
2775 struct btrfs_root
*root
)
2777 struct btrfs_block_rsv
*global_rsv
;
2778 u64 num_heads
= trans
->transaction
->delayed_refs
.num_heads_ready
;
2779 u64 csum_bytes
= trans
->transaction
->delayed_refs
.pending_csums
;
2780 u64 num_dirty_bgs
= trans
->transaction
->num_dirty_bgs
;
2781 u64 num_bytes
, num_dirty_bgs_bytes
;
2784 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
2785 num_heads
= heads_to_leaves(root
, num_heads
);
2787 num_bytes
+= (num_heads
- 1) * root
->nodesize
;
2789 num_bytes
+= btrfs_csum_bytes_to_leaves(root
, csum_bytes
) * root
->nodesize
;
2790 num_dirty_bgs_bytes
= btrfs_calc_trans_metadata_size(root
,
2792 global_rsv
= &root
->fs_info
->global_block_rsv
;
2795 * If we can't allocate any more chunks lets make sure we have _lots_ of
2796 * wiggle room since running delayed refs can create more delayed refs.
2798 if (global_rsv
->space_info
->full
) {
2799 num_dirty_bgs_bytes
<<= 1;
2803 spin_lock(&global_rsv
->lock
);
2804 if (global_rsv
->reserved
<= num_bytes
+ num_dirty_bgs_bytes
)
2806 spin_unlock(&global_rsv
->lock
);
2810 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle
*trans
,
2811 struct btrfs_root
*root
)
2813 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2815 atomic_read(&trans
->transaction
->delayed_refs
.num_entries
);
2820 avg_runtime
= fs_info
->avg_delayed_ref_runtime
;
2821 val
= num_entries
* avg_runtime
;
2822 if (num_entries
* avg_runtime
>= NSEC_PER_SEC
)
2824 if (val
>= NSEC_PER_SEC
/ 2)
2827 return btrfs_check_space_for_delayed_refs(trans
, root
);
2830 struct async_delayed_refs
{
2831 struct btrfs_root
*root
;
2835 struct completion wait
;
2836 struct btrfs_work work
;
2839 static void delayed_ref_async_start(struct btrfs_work
*work
)
2841 struct async_delayed_refs
*async
;
2842 struct btrfs_trans_handle
*trans
;
2845 async
= container_of(work
, struct async_delayed_refs
, work
);
2847 trans
= btrfs_join_transaction(async
->root
);
2848 if (IS_ERR(trans
)) {
2849 async
->error
= PTR_ERR(trans
);
2854 * trans->sync means that when we call end_transaciton, we won't
2855 * wait on delayed refs
2858 ret
= btrfs_run_delayed_refs(trans
, async
->root
, async
->count
);
2862 ret
= btrfs_end_transaction(trans
, async
->root
);
2863 if (ret
&& !async
->error
)
2867 complete(&async
->wait
);
2872 int btrfs_async_run_delayed_refs(struct btrfs_root
*root
,
2873 unsigned long count
, int wait
)
2875 struct async_delayed_refs
*async
;
2878 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
2882 async
->root
= root
->fs_info
->tree_root
;
2883 async
->count
= count
;
2889 init_completion(&async
->wait
);
2891 btrfs_init_work(&async
->work
, btrfs_extent_refs_helper
,
2892 delayed_ref_async_start
, NULL
, NULL
);
2894 btrfs_queue_work(root
->fs_info
->extent_workers
, &async
->work
);
2897 wait_for_completion(&async
->wait
);
2906 * this starts processing the delayed reference count updates and
2907 * extent insertions we have queued up so far. count can be
2908 * 0, which means to process everything in the tree at the start
2909 * of the run (but not newly added entries), or it can be some target
2910 * number you'd like to process.
2912 * Returns 0 on success or if called with an aborted transaction
2913 * Returns <0 on error and aborts the transaction
2915 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2916 struct btrfs_root
*root
, unsigned long count
)
2918 struct rb_node
*node
;
2919 struct btrfs_delayed_ref_root
*delayed_refs
;
2920 struct btrfs_delayed_ref_head
*head
;
2922 int run_all
= count
== (unsigned long)-1;
2923 bool can_flush_pending_bgs
= trans
->can_flush_pending_bgs
;
2925 /* We'll clean this up in btrfs_cleanup_transaction */
2929 if (root
->fs_info
->creating_free_space_tree
)
2932 if (root
== root
->fs_info
->extent_root
)
2933 root
= root
->fs_info
->tree_root
;
2935 delayed_refs
= &trans
->transaction
->delayed_refs
;
2937 count
= atomic_read(&delayed_refs
->num_entries
) * 2;
2940 #ifdef SCRAMBLE_DELAYED_REFS
2941 delayed_refs
->run_delayed_start
= find_middle(&delayed_refs
->root
);
2943 trans
->can_flush_pending_bgs
= false;
2944 ret
= __btrfs_run_delayed_refs(trans
, root
, count
);
2946 btrfs_abort_transaction(trans
, root
, ret
);
2951 if (!list_empty(&trans
->new_bgs
))
2952 btrfs_create_pending_block_groups(trans
, root
);
2954 spin_lock(&delayed_refs
->lock
);
2955 node
= rb_first(&delayed_refs
->href_root
);
2957 spin_unlock(&delayed_refs
->lock
);
2960 count
= (unsigned long)-1;
2963 head
= rb_entry(node
, struct btrfs_delayed_ref_head
,
2965 if (btrfs_delayed_ref_is_head(&head
->node
)) {
2966 struct btrfs_delayed_ref_node
*ref
;
2969 atomic_inc(&ref
->refs
);
2971 spin_unlock(&delayed_refs
->lock
);
2973 * Mutex was contended, block until it's
2974 * released and try again
2976 mutex_lock(&head
->mutex
);
2977 mutex_unlock(&head
->mutex
);
2979 btrfs_put_delayed_ref(ref
);
2985 node
= rb_next(node
);
2987 spin_unlock(&delayed_refs
->lock
);
2992 assert_qgroups_uptodate(trans
);
2993 trans
->can_flush_pending_bgs
= can_flush_pending_bgs
;
2997 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
2998 struct btrfs_root
*root
,
2999 u64 bytenr
, u64 num_bytes
, u64 flags
,
3000 int level
, int is_data
)
3002 struct btrfs_delayed_extent_op
*extent_op
;
3005 extent_op
= btrfs_alloc_delayed_extent_op();
3009 extent_op
->flags_to_set
= flags
;
3010 extent_op
->update_flags
= true;
3011 extent_op
->update_key
= false;
3012 extent_op
->is_data
= is_data
? true : false;
3013 extent_op
->level
= level
;
3015 ret
= btrfs_add_delayed_extent_op(root
->fs_info
, trans
, bytenr
,
3016 num_bytes
, extent_op
);
3018 btrfs_free_delayed_extent_op(extent_op
);
3022 static noinline
int check_delayed_ref(struct btrfs_trans_handle
*trans
,
3023 struct btrfs_root
*root
,
3024 struct btrfs_path
*path
,
3025 u64 objectid
, u64 offset
, u64 bytenr
)
3027 struct btrfs_delayed_ref_head
*head
;
3028 struct btrfs_delayed_ref_node
*ref
;
3029 struct btrfs_delayed_data_ref
*data_ref
;
3030 struct btrfs_delayed_ref_root
*delayed_refs
;
3033 delayed_refs
= &trans
->transaction
->delayed_refs
;
3034 spin_lock(&delayed_refs
->lock
);
3035 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
3037 spin_unlock(&delayed_refs
->lock
);
3041 if (!mutex_trylock(&head
->mutex
)) {
3042 atomic_inc(&head
->node
.refs
);
3043 spin_unlock(&delayed_refs
->lock
);
3045 btrfs_release_path(path
);
3048 * Mutex was contended, block until it's released and let
3051 mutex_lock(&head
->mutex
);
3052 mutex_unlock(&head
->mutex
);
3053 btrfs_put_delayed_ref(&head
->node
);
3056 spin_unlock(&delayed_refs
->lock
);
3058 spin_lock(&head
->lock
);
3059 list_for_each_entry(ref
, &head
->ref_list
, list
) {
3060 /* If it's a shared ref we know a cross reference exists */
3061 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
) {
3066 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
3069 * If our ref doesn't match the one we're currently looking at
3070 * then we have a cross reference.
3072 if (data_ref
->root
!= root
->root_key
.objectid
||
3073 data_ref
->objectid
!= objectid
||
3074 data_ref
->offset
!= offset
) {
3079 spin_unlock(&head
->lock
);
3080 mutex_unlock(&head
->mutex
);
3084 static noinline
int check_committed_ref(struct btrfs_trans_handle
*trans
,
3085 struct btrfs_root
*root
,
3086 struct btrfs_path
*path
,
3087 u64 objectid
, u64 offset
, u64 bytenr
)
3089 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
3090 struct extent_buffer
*leaf
;
3091 struct btrfs_extent_data_ref
*ref
;
3092 struct btrfs_extent_inline_ref
*iref
;
3093 struct btrfs_extent_item
*ei
;
3094 struct btrfs_key key
;
3098 key
.objectid
= bytenr
;
3099 key
.offset
= (u64
)-1;
3100 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
3102 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
3105 BUG_ON(ret
== 0); /* Corruption */
3108 if (path
->slots
[0] == 0)
3112 leaf
= path
->nodes
[0];
3113 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
3115 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
3119 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
3120 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3121 if (item_size
< sizeof(*ei
)) {
3122 WARN_ON(item_size
!= sizeof(struct btrfs_extent_item_v0
));
3126 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
3128 if (item_size
!= sizeof(*ei
) +
3129 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
3132 if (btrfs_extent_generation(leaf
, ei
) <=
3133 btrfs_root_last_snapshot(&root
->root_item
))
3136 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
3137 if (btrfs_extent_inline_ref_type(leaf
, iref
) !=
3138 BTRFS_EXTENT_DATA_REF_KEY
)
3141 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
3142 if (btrfs_extent_refs(leaf
, ei
) !=
3143 btrfs_extent_data_ref_count(leaf
, ref
) ||
3144 btrfs_extent_data_ref_root(leaf
, ref
) !=
3145 root
->root_key
.objectid
||
3146 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
3147 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
3155 int btrfs_cross_ref_exist(struct btrfs_trans_handle
*trans
,
3156 struct btrfs_root
*root
,
3157 u64 objectid
, u64 offset
, u64 bytenr
)
3159 struct btrfs_path
*path
;
3163 path
= btrfs_alloc_path();
3168 ret
= check_committed_ref(trans
, root
, path
, objectid
,
3170 if (ret
&& ret
!= -ENOENT
)
3173 ret2
= check_delayed_ref(trans
, root
, path
, objectid
,
3175 } while (ret2
== -EAGAIN
);
3177 if (ret2
&& ret2
!= -ENOENT
) {
3182 if (ret
!= -ENOENT
|| ret2
!= -ENOENT
)
3185 btrfs_free_path(path
);
3186 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
3191 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
3192 struct btrfs_root
*root
,
3193 struct extent_buffer
*buf
,
3194 int full_backref
, int inc
)
3201 struct btrfs_key key
;
3202 struct btrfs_file_extent_item
*fi
;
3206 int (*process_func
)(struct btrfs_trans_handle
*, struct btrfs_root
*,
3207 u64
, u64
, u64
, u64
, u64
, u64
);
3210 if (btrfs_test_is_dummy_root(root
))
3213 ref_root
= btrfs_header_owner(buf
);
3214 nritems
= btrfs_header_nritems(buf
);
3215 level
= btrfs_header_level(buf
);
3217 if (!test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) && level
== 0)
3221 process_func
= btrfs_inc_extent_ref
;
3223 process_func
= btrfs_free_extent
;
3226 parent
= buf
->start
;
3230 for (i
= 0; i
< nritems
; i
++) {
3232 btrfs_item_key_to_cpu(buf
, &key
, i
);
3233 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
3235 fi
= btrfs_item_ptr(buf
, i
,
3236 struct btrfs_file_extent_item
);
3237 if (btrfs_file_extent_type(buf
, fi
) ==
3238 BTRFS_FILE_EXTENT_INLINE
)
3240 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
3244 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
3245 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
3246 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
3247 parent
, ref_root
, key
.objectid
,
3252 bytenr
= btrfs_node_blockptr(buf
, i
);
3253 num_bytes
= root
->nodesize
;
3254 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
3255 parent
, ref_root
, level
- 1, 0);
3265 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3266 struct extent_buffer
*buf
, int full_backref
)
3268 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1);
3271 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3272 struct extent_buffer
*buf
, int full_backref
)
3274 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0);
3277 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
3278 struct btrfs_root
*root
,
3279 struct btrfs_path
*path
,
3280 struct btrfs_block_group_cache
*cache
)
3283 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
3285 struct extent_buffer
*leaf
;
3287 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
3294 leaf
= path
->nodes
[0];
3295 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
3296 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
3297 btrfs_mark_buffer_dirty(leaf
);
3299 btrfs_release_path(path
);
3304 static struct btrfs_block_group_cache
*
3305 next_block_group(struct btrfs_root
*root
,
3306 struct btrfs_block_group_cache
*cache
)
3308 struct rb_node
*node
;
3310 spin_lock(&root
->fs_info
->block_group_cache_lock
);
3312 /* If our block group was removed, we need a full search. */
3313 if (RB_EMPTY_NODE(&cache
->cache_node
)) {
3314 const u64 next_bytenr
= cache
->key
.objectid
+ cache
->key
.offset
;
3316 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
3317 btrfs_put_block_group(cache
);
3318 cache
= btrfs_lookup_first_block_group(root
->fs_info
,
3322 node
= rb_next(&cache
->cache_node
);
3323 btrfs_put_block_group(cache
);
3325 cache
= rb_entry(node
, struct btrfs_block_group_cache
,
3327 btrfs_get_block_group(cache
);
3330 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
3334 static int cache_save_setup(struct btrfs_block_group_cache
*block_group
,
3335 struct btrfs_trans_handle
*trans
,
3336 struct btrfs_path
*path
)
3338 struct btrfs_root
*root
= block_group
->fs_info
->tree_root
;
3339 struct inode
*inode
= NULL
;
3341 int dcs
= BTRFS_DC_ERROR
;
3347 * If this block group is smaller than 100 megs don't bother caching the
3350 if (block_group
->key
.offset
< (100 * SZ_1M
)) {
3351 spin_lock(&block_group
->lock
);
3352 block_group
->disk_cache_state
= BTRFS_DC_WRITTEN
;
3353 spin_unlock(&block_group
->lock
);
3360 inode
= lookup_free_space_inode(root
, block_group
, path
);
3361 if (IS_ERR(inode
) && PTR_ERR(inode
) != -ENOENT
) {
3362 ret
= PTR_ERR(inode
);
3363 btrfs_release_path(path
);
3367 if (IS_ERR(inode
)) {
3371 if (block_group
->ro
)
3374 ret
= create_free_space_inode(root
, trans
, block_group
, path
);
3380 /* We've already setup this transaction, go ahead and exit */
3381 if (block_group
->cache_generation
== trans
->transid
&&
3382 i_size_read(inode
)) {
3383 dcs
= BTRFS_DC_SETUP
;
3388 * We want to set the generation to 0, that way if anything goes wrong
3389 * from here on out we know not to trust this cache when we load up next
3392 BTRFS_I(inode
)->generation
= 0;
3393 ret
= btrfs_update_inode(trans
, root
, inode
);
3396 * So theoretically we could recover from this, simply set the
3397 * super cache generation to 0 so we know to invalidate the
3398 * cache, but then we'd have to keep track of the block groups
3399 * that fail this way so we know we _have_ to reset this cache
3400 * before the next commit or risk reading stale cache. So to
3401 * limit our exposure to horrible edge cases lets just abort the
3402 * transaction, this only happens in really bad situations
3405 btrfs_abort_transaction(trans
, root
, ret
);
3410 if (i_size_read(inode
) > 0) {
3411 ret
= btrfs_check_trunc_cache_free_space(root
,
3412 &root
->fs_info
->global_block_rsv
);
3416 ret
= btrfs_truncate_free_space_cache(root
, trans
, NULL
, inode
);
3421 spin_lock(&block_group
->lock
);
3422 if (block_group
->cached
!= BTRFS_CACHE_FINISHED
||
3423 !btrfs_test_opt(root
, SPACE_CACHE
)) {
3425 * don't bother trying to write stuff out _if_
3426 * a) we're not cached,
3427 * b) we're with nospace_cache mount option.
3429 dcs
= BTRFS_DC_WRITTEN
;
3430 spin_unlock(&block_group
->lock
);
3433 spin_unlock(&block_group
->lock
);
3436 * We hit an ENOSPC when setting up the cache in this transaction, just
3437 * skip doing the setup, we've already cleared the cache so we're safe.
3439 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC
, &trans
->transaction
->flags
)) {
3445 * Try to preallocate enough space based on how big the block group is.
3446 * Keep in mind this has to include any pinned space which could end up
3447 * taking up quite a bit since it's not folded into the other space
3450 num_pages
= div_u64(block_group
->key
.offset
, SZ_256M
);
3455 num_pages
*= PAGE_SIZE
;
3457 ret
= btrfs_check_data_free_space(inode
, 0, num_pages
);
3461 ret
= btrfs_prealloc_file_range_trans(inode
, trans
, 0, 0, num_pages
,
3462 num_pages
, num_pages
,
3465 * Our cache requires contiguous chunks so that we don't modify a bunch
3466 * of metadata or split extents when writing the cache out, which means
3467 * we can enospc if we are heavily fragmented in addition to just normal
3468 * out of space conditions. So if we hit this just skip setting up any
3469 * other block groups for this transaction, maybe we'll unpin enough
3470 * space the next time around.
3473 dcs
= BTRFS_DC_SETUP
;
3474 else if (ret
== -ENOSPC
)
3475 set_bit(BTRFS_TRANS_CACHE_ENOSPC
, &trans
->transaction
->flags
);
3476 btrfs_free_reserved_data_space(inode
, 0, num_pages
);
3481 btrfs_release_path(path
);
3483 spin_lock(&block_group
->lock
);
3484 if (!ret
&& dcs
== BTRFS_DC_SETUP
)
3485 block_group
->cache_generation
= trans
->transid
;
3486 block_group
->disk_cache_state
= dcs
;
3487 spin_unlock(&block_group
->lock
);
3492 int btrfs_setup_space_cache(struct btrfs_trans_handle
*trans
,
3493 struct btrfs_root
*root
)
3495 struct btrfs_block_group_cache
*cache
, *tmp
;
3496 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3497 struct btrfs_path
*path
;
3499 if (list_empty(&cur_trans
->dirty_bgs
) ||
3500 !btrfs_test_opt(root
, SPACE_CACHE
))
3503 path
= btrfs_alloc_path();
3507 /* Could add new block groups, use _safe just in case */
3508 list_for_each_entry_safe(cache
, tmp
, &cur_trans
->dirty_bgs
,
3510 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
)
3511 cache_save_setup(cache
, trans
, path
);
3514 btrfs_free_path(path
);
3519 * transaction commit does final block group cache writeback during a
3520 * critical section where nothing is allowed to change the FS. This is
3521 * required in order for the cache to actually match the block group,
3522 * but can introduce a lot of latency into the commit.
3524 * So, btrfs_start_dirty_block_groups is here to kick off block group
3525 * cache IO. There's a chance we'll have to redo some of it if the
3526 * block group changes again during the commit, but it greatly reduces
3527 * the commit latency by getting rid of the easy block groups while
3528 * we're still allowing others to join the commit.
3530 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle
*trans
,
3531 struct btrfs_root
*root
)
3533 struct btrfs_block_group_cache
*cache
;
3534 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3537 struct btrfs_path
*path
= NULL
;
3539 struct list_head
*io
= &cur_trans
->io_bgs
;
3540 int num_started
= 0;
3543 spin_lock(&cur_trans
->dirty_bgs_lock
);
3544 if (list_empty(&cur_trans
->dirty_bgs
)) {
3545 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3548 list_splice_init(&cur_trans
->dirty_bgs
, &dirty
);
3549 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3553 * make sure all the block groups on our dirty list actually
3556 btrfs_create_pending_block_groups(trans
, root
);
3559 path
= btrfs_alloc_path();
3565 * cache_write_mutex is here only to save us from balance or automatic
3566 * removal of empty block groups deleting this block group while we are
3567 * writing out the cache
3569 mutex_lock(&trans
->transaction
->cache_write_mutex
);
3570 while (!list_empty(&dirty
)) {
3571 cache
= list_first_entry(&dirty
,
3572 struct btrfs_block_group_cache
,
3575 * this can happen if something re-dirties a block
3576 * group that is already under IO. Just wait for it to
3577 * finish and then do it all again
3579 if (!list_empty(&cache
->io_list
)) {
3580 list_del_init(&cache
->io_list
);
3581 btrfs_wait_cache_io(root
, trans
, cache
,
3582 &cache
->io_ctl
, path
,
3583 cache
->key
.objectid
);
3584 btrfs_put_block_group(cache
);
3589 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3590 * if it should update the cache_state. Don't delete
3591 * until after we wait.
3593 * Since we're not running in the commit critical section
3594 * we need the dirty_bgs_lock to protect from update_block_group
3596 spin_lock(&cur_trans
->dirty_bgs_lock
);
3597 list_del_init(&cache
->dirty_list
);
3598 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3602 cache_save_setup(cache
, trans
, path
);
3604 if (cache
->disk_cache_state
== BTRFS_DC_SETUP
) {
3605 cache
->io_ctl
.inode
= NULL
;
3606 ret
= btrfs_write_out_cache(root
, trans
, cache
, path
);
3607 if (ret
== 0 && cache
->io_ctl
.inode
) {
3612 * the cache_write_mutex is protecting
3615 list_add_tail(&cache
->io_list
, io
);
3618 * if we failed to write the cache, the
3619 * generation will be bad and life goes on
3625 ret
= write_one_cache_group(trans
, root
, path
, cache
);
3627 * Our block group might still be attached to the list
3628 * of new block groups in the transaction handle of some
3629 * other task (struct btrfs_trans_handle->new_bgs). This
3630 * means its block group item isn't yet in the extent
3631 * tree. If this happens ignore the error, as we will
3632 * try again later in the critical section of the
3633 * transaction commit.
3635 if (ret
== -ENOENT
) {
3637 spin_lock(&cur_trans
->dirty_bgs_lock
);
3638 if (list_empty(&cache
->dirty_list
)) {
3639 list_add_tail(&cache
->dirty_list
,
3640 &cur_trans
->dirty_bgs
);
3641 btrfs_get_block_group(cache
);
3643 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3645 btrfs_abort_transaction(trans
, root
, ret
);
3649 /* if its not on the io list, we need to put the block group */
3651 btrfs_put_block_group(cache
);
3657 * Avoid blocking other tasks for too long. It might even save
3658 * us from writing caches for block groups that are going to be
3661 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
3662 mutex_lock(&trans
->transaction
->cache_write_mutex
);
3664 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
3667 * go through delayed refs for all the stuff we've just kicked off
3668 * and then loop back (just once)
3670 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
3671 if (!ret
&& loops
== 0) {
3673 spin_lock(&cur_trans
->dirty_bgs_lock
);
3674 list_splice_init(&cur_trans
->dirty_bgs
, &dirty
);
3676 * dirty_bgs_lock protects us from concurrent block group
3677 * deletes too (not just cache_write_mutex).
3679 if (!list_empty(&dirty
)) {
3680 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3683 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3686 btrfs_free_path(path
);
3690 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
3691 struct btrfs_root
*root
)
3693 struct btrfs_block_group_cache
*cache
;
3694 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3697 struct btrfs_path
*path
;
3698 struct list_head
*io
= &cur_trans
->io_bgs
;
3699 int num_started
= 0;
3701 path
= btrfs_alloc_path();
3706 * Even though we are in the critical section of the transaction commit,
3707 * we can still have concurrent tasks adding elements to this
3708 * transaction's list of dirty block groups. These tasks correspond to
3709 * endio free space workers started when writeback finishes for a
3710 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3711 * allocate new block groups as a result of COWing nodes of the root
3712 * tree when updating the free space inode. The writeback for the space
3713 * caches is triggered by an earlier call to
3714 * btrfs_start_dirty_block_groups() and iterations of the following
3716 * Also we want to do the cache_save_setup first and then run the
3717 * delayed refs to make sure we have the best chance at doing this all
3720 spin_lock(&cur_trans
->dirty_bgs_lock
);
3721 while (!list_empty(&cur_trans
->dirty_bgs
)) {
3722 cache
= list_first_entry(&cur_trans
->dirty_bgs
,
3723 struct btrfs_block_group_cache
,
3727 * this can happen if cache_save_setup re-dirties a block
3728 * group that is already under IO. Just wait for it to
3729 * finish and then do it all again
3731 if (!list_empty(&cache
->io_list
)) {
3732 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3733 list_del_init(&cache
->io_list
);
3734 btrfs_wait_cache_io(root
, trans
, cache
,
3735 &cache
->io_ctl
, path
,
3736 cache
->key
.objectid
);
3737 btrfs_put_block_group(cache
);
3738 spin_lock(&cur_trans
->dirty_bgs_lock
);
3742 * don't remove from the dirty list until after we've waited
3745 list_del_init(&cache
->dirty_list
);
3746 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3749 cache_save_setup(cache
, trans
, path
);
3752 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long) -1);
3754 if (!ret
&& cache
->disk_cache_state
== BTRFS_DC_SETUP
) {
3755 cache
->io_ctl
.inode
= NULL
;
3756 ret
= btrfs_write_out_cache(root
, trans
, cache
, path
);
3757 if (ret
== 0 && cache
->io_ctl
.inode
) {
3760 list_add_tail(&cache
->io_list
, io
);
3763 * if we failed to write the cache, the
3764 * generation will be bad and life goes on
3770 ret
= write_one_cache_group(trans
, root
, path
, cache
);
3772 * One of the free space endio workers might have
3773 * created a new block group while updating a free space
3774 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3775 * and hasn't released its transaction handle yet, in
3776 * which case the new block group is still attached to
3777 * its transaction handle and its creation has not
3778 * finished yet (no block group item in the extent tree
3779 * yet, etc). If this is the case, wait for all free
3780 * space endio workers to finish and retry. This is a
3781 * a very rare case so no need for a more efficient and
3784 if (ret
== -ENOENT
) {
3785 wait_event(cur_trans
->writer_wait
,
3786 atomic_read(&cur_trans
->num_writers
) == 1);
3787 ret
= write_one_cache_group(trans
, root
, path
,
3791 btrfs_abort_transaction(trans
, root
, ret
);
3794 /* if its not on the io list, we need to put the block group */
3796 btrfs_put_block_group(cache
);
3797 spin_lock(&cur_trans
->dirty_bgs_lock
);
3799 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3801 while (!list_empty(io
)) {
3802 cache
= list_first_entry(io
, struct btrfs_block_group_cache
,
3804 list_del_init(&cache
->io_list
);
3805 btrfs_wait_cache_io(root
, trans
, cache
,
3806 &cache
->io_ctl
, path
, cache
->key
.objectid
);
3807 btrfs_put_block_group(cache
);
3810 btrfs_free_path(path
);
3814 int btrfs_extent_readonly(struct btrfs_root
*root
, u64 bytenr
)
3816 struct btrfs_block_group_cache
*block_group
;
3819 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
3820 if (!block_group
|| block_group
->ro
)
3823 btrfs_put_block_group(block_group
);
3827 static const char *alloc_name(u64 flags
)
3830 case BTRFS_BLOCK_GROUP_METADATA
|BTRFS_BLOCK_GROUP_DATA
:
3832 case BTRFS_BLOCK_GROUP_METADATA
:
3834 case BTRFS_BLOCK_GROUP_DATA
:
3836 case BTRFS_BLOCK_GROUP_SYSTEM
:
3840 return "invalid-combination";
3844 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
3845 u64 total_bytes
, u64 bytes_used
,
3846 struct btrfs_space_info
**space_info
)
3848 struct btrfs_space_info
*found
;
3853 if (flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3854 BTRFS_BLOCK_GROUP_RAID10
))
3859 found
= __find_space_info(info
, flags
);
3861 spin_lock(&found
->lock
);
3862 found
->total_bytes
+= total_bytes
;
3863 found
->disk_total
+= total_bytes
* factor
;
3864 found
->bytes_used
+= bytes_used
;
3865 found
->disk_used
+= bytes_used
* factor
;
3866 if (total_bytes
> 0)
3868 spin_unlock(&found
->lock
);
3869 *space_info
= found
;
3872 found
= kzalloc(sizeof(*found
), GFP_NOFS
);
3876 ret
= percpu_counter_init(&found
->total_bytes_pinned
, 0, GFP_KERNEL
);
3882 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
3883 INIT_LIST_HEAD(&found
->block_groups
[i
]);
3884 init_rwsem(&found
->groups_sem
);
3885 spin_lock_init(&found
->lock
);
3886 found
->flags
= flags
& BTRFS_BLOCK_GROUP_TYPE_MASK
;
3887 found
->total_bytes
= total_bytes
;
3888 found
->disk_total
= total_bytes
* factor
;
3889 found
->bytes_used
= bytes_used
;
3890 found
->disk_used
= bytes_used
* factor
;
3891 found
->bytes_pinned
= 0;
3892 found
->bytes_reserved
= 0;
3893 found
->bytes_readonly
= 0;
3894 found
->bytes_may_use
= 0;
3896 found
->max_extent_size
= 0;
3897 found
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3898 found
->chunk_alloc
= 0;
3900 init_waitqueue_head(&found
->wait
);
3901 INIT_LIST_HEAD(&found
->ro_bgs
);
3903 ret
= kobject_init_and_add(&found
->kobj
, &space_info_ktype
,
3904 info
->space_info_kobj
, "%s",
3905 alloc_name(found
->flags
));
3911 *space_info
= found
;
3912 list_add_rcu(&found
->list
, &info
->space_info
);
3913 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3914 info
->data_sinfo
= found
;
3919 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
3921 u64 extra_flags
= chunk_to_extended(flags
) &
3922 BTRFS_EXTENDED_PROFILE_MASK
;
3924 write_seqlock(&fs_info
->profiles_lock
);
3925 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3926 fs_info
->avail_data_alloc_bits
|= extra_flags
;
3927 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3928 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
3929 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3930 fs_info
->avail_system_alloc_bits
|= extra_flags
;
3931 write_sequnlock(&fs_info
->profiles_lock
);
3935 * returns target flags in extended format or 0 if restripe for this
3936 * chunk_type is not in progress
3938 * should be called with either volume_mutex or balance_lock held
3940 static u64
get_restripe_target(struct btrfs_fs_info
*fs_info
, u64 flags
)
3942 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3948 if (flags
& BTRFS_BLOCK_GROUP_DATA
&&
3949 bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3950 target
= BTRFS_BLOCK_GROUP_DATA
| bctl
->data
.target
;
3951 } else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
&&
3952 bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3953 target
= BTRFS_BLOCK_GROUP_SYSTEM
| bctl
->sys
.target
;
3954 } else if (flags
& BTRFS_BLOCK_GROUP_METADATA
&&
3955 bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3956 target
= BTRFS_BLOCK_GROUP_METADATA
| bctl
->meta
.target
;
3963 * @flags: available profiles in extended format (see ctree.h)
3965 * Returns reduced profile in chunk format. If profile changing is in
3966 * progress (either running or paused) picks the target profile (if it's
3967 * already available), otherwise falls back to plain reducing.
3969 static u64
btrfs_reduce_alloc_profile(struct btrfs_root
*root
, u64 flags
)
3971 u64 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
3977 * see if restripe for this chunk_type is in progress, if so
3978 * try to reduce to the target profile
3980 spin_lock(&root
->fs_info
->balance_lock
);
3981 target
= get_restripe_target(root
->fs_info
, flags
);
3983 /* pick target profile only if it's already available */
3984 if ((flags
& target
) & BTRFS_EXTENDED_PROFILE_MASK
) {
3985 spin_unlock(&root
->fs_info
->balance_lock
);
3986 return extended_to_chunk(target
);
3989 spin_unlock(&root
->fs_info
->balance_lock
);
3991 /* First, mask out the RAID levels which aren't possible */
3992 for (raid_type
= 0; raid_type
< BTRFS_NR_RAID_TYPES
; raid_type
++) {
3993 if (num_devices
>= btrfs_raid_array
[raid_type
].devs_min
)
3994 allowed
|= btrfs_raid_group
[raid_type
];
3998 if (allowed
& BTRFS_BLOCK_GROUP_RAID6
)
3999 allowed
= BTRFS_BLOCK_GROUP_RAID6
;
4000 else if (allowed
& BTRFS_BLOCK_GROUP_RAID5
)
4001 allowed
= BTRFS_BLOCK_GROUP_RAID5
;
4002 else if (allowed
& BTRFS_BLOCK_GROUP_RAID10
)
4003 allowed
= BTRFS_BLOCK_GROUP_RAID10
;
4004 else if (allowed
& BTRFS_BLOCK_GROUP_RAID1
)
4005 allowed
= BTRFS_BLOCK_GROUP_RAID1
;
4006 else if (allowed
& BTRFS_BLOCK_GROUP_RAID0
)
4007 allowed
= BTRFS_BLOCK_GROUP_RAID0
;
4009 flags
&= ~BTRFS_BLOCK_GROUP_PROFILE_MASK
;
4011 return extended_to_chunk(flags
| allowed
);
4014 static u64
get_alloc_profile(struct btrfs_root
*root
, u64 orig_flags
)
4021 seq
= read_seqbegin(&root
->fs_info
->profiles_lock
);
4023 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
4024 flags
|= root
->fs_info
->avail_data_alloc_bits
;
4025 else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
4026 flags
|= root
->fs_info
->avail_system_alloc_bits
;
4027 else if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
4028 flags
|= root
->fs_info
->avail_metadata_alloc_bits
;
4029 } while (read_seqretry(&root
->fs_info
->profiles_lock
, seq
));
4031 return btrfs_reduce_alloc_profile(root
, flags
);
4034 u64
btrfs_get_alloc_profile(struct btrfs_root
*root
, int data
)
4040 flags
= BTRFS_BLOCK_GROUP_DATA
;
4041 else if (root
== root
->fs_info
->chunk_root
)
4042 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
4044 flags
= BTRFS_BLOCK_GROUP_METADATA
;
4046 ret
= get_alloc_profile(root
, flags
);
4050 int btrfs_alloc_data_chunk_ondemand(struct inode
*inode
, u64 bytes
)
4052 struct btrfs_space_info
*data_sinfo
;
4053 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4054 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4057 int need_commit
= 2;
4058 int have_pinned_space
;
4060 /* make sure bytes are sectorsize aligned */
4061 bytes
= ALIGN(bytes
, root
->sectorsize
);
4063 if (btrfs_is_free_space_inode(inode
)) {
4065 ASSERT(current
->journal_info
);
4068 data_sinfo
= fs_info
->data_sinfo
;
4073 /* make sure we have enough space to handle the data first */
4074 spin_lock(&data_sinfo
->lock
);
4075 used
= data_sinfo
->bytes_used
+ data_sinfo
->bytes_reserved
+
4076 data_sinfo
->bytes_pinned
+ data_sinfo
->bytes_readonly
+
4077 data_sinfo
->bytes_may_use
;
4079 if (used
+ bytes
> data_sinfo
->total_bytes
) {
4080 struct btrfs_trans_handle
*trans
;
4083 * if we don't have enough free bytes in this space then we need
4084 * to alloc a new chunk.
4086 if (!data_sinfo
->full
) {
4089 data_sinfo
->force_alloc
= CHUNK_ALLOC_FORCE
;
4090 spin_unlock(&data_sinfo
->lock
);
4092 alloc_target
= btrfs_get_alloc_profile(root
, 1);
4094 * It is ugly that we don't call nolock join
4095 * transaction for the free space inode case here.
4096 * But it is safe because we only do the data space
4097 * reservation for the free space cache in the
4098 * transaction context, the common join transaction
4099 * just increase the counter of the current transaction
4100 * handler, doesn't try to acquire the trans_lock of
4103 trans
= btrfs_join_transaction(root
);
4105 return PTR_ERR(trans
);
4107 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
4109 CHUNK_ALLOC_NO_FORCE
);
4110 btrfs_end_transaction(trans
, root
);
4115 have_pinned_space
= 1;
4121 data_sinfo
= fs_info
->data_sinfo
;
4127 * If we don't have enough pinned space to deal with this
4128 * allocation, and no removed chunk in current transaction,
4129 * don't bother committing the transaction.
4131 have_pinned_space
= percpu_counter_compare(
4132 &data_sinfo
->total_bytes_pinned
,
4133 used
+ bytes
- data_sinfo
->total_bytes
);
4134 spin_unlock(&data_sinfo
->lock
);
4136 /* commit the current transaction and try again */
4139 !atomic_read(&root
->fs_info
->open_ioctl_trans
)) {
4142 if (need_commit
> 0) {
4143 btrfs_start_delalloc_roots(fs_info
, 0, -1);
4144 btrfs_wait_ordered_roots(fs_info
, -1, 0, (u64
)-1);
4147 trans
= btrfs_join_transaction(root
);
4149 return PTR_ERR(trans
);
4150 if (have_pinned_space
>= 0 ||
4151 test_bit(BTRFS_TRANS_HAVE_FREE_BGS
,
4152 &trans
->transaction
->flags
) ||
4154 ret
= btrfs_commit_transaction(trans
, root
);
4158 * The cleaner kthread might still be doing iput
4159 * operations. Wait for it to finish so that
4160 * more space is released.
4162 mutex_lock(&root
->fs_info
->cleaner_delayed_iput_mutex
);
4163 mutex_unlock(&root
->fs_info
->cleaner_delayed_iput_mutex
);
4166 btrfs_end_transaction(trans
, root
);
4170 trace_btrfs_space_reservation(root
->fs_info
,
4171 "space_info:enospc",
4172 data_sinfo
->flags
, bytes
, 1);
4175 data_sinfo
->bytes_may_use
+= bytes
;
4176 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
4177 data_sinfo
->flags
, bytes
, 1);
4178 spin_unlock(&data_sinfo
->lock
);
4184 * New check_data_free_space() with ability for precious data reservation
4185 * Will replace old btrfs_check_data_free_space(), but for patch split,
4186 * add a new function first and then replace it.
4188 int btrfs_check_data_free_space(struct inode
*inode
, u64 start
, u64 len
)
4190 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4193 /* align the range */
4194 len
= round_up(start
+ len
, root
->sectorsize
) -
4195 round_down(start
, root
->sectorsize
);
4196 start
= round_down(start
, root
->sectorsize
);
4198 ret
= btrfs_alloc_data_chunk_ondemand(inode
, len
);
4203 * Use new btrfs_qgroup_reserve_data to reserve precious data space
4205 * TODO: Find a good method to avoid reserve data space for NOCOW
4206 * range, but don't impact performance on quota disable case.
4208 ret
= btrfs_qgroup_reserve_data(inode
, start
, len
);
4213 * Called if we need to clear a data reservation for this inode
4214 * Normally in a error case.
4216 * This one will *NOT* use accurate qgroup reserved space API, just for case
4217 * which we can't sleep and is sure it won't affect qgroup reserved space.
4218 * Like clear_bit_hook().
4220 void btrfs_free_reserved_data_space_noquota(struct inode
*inode
, u64 start
,
4223 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4224 struct btrfs_space_info
*data_sinfo
;
4226 /* Make sure the range is aligned to sectorsize */
4227 len
= round_up(start
+ len
, root
->sectorsize
) -
4228 round_down(start
, root
->sectorsize
);
4229 start
= round_down(start
, root
->sectorsize
);
4231 data_sinfo
= root
->fs_info
->data_sinfo
;
4232 spin_lock(&data_sinfo
->lock
);
4233 if (WARN_ON(data_sinfo
->bytes_may_use
< len
))
4234 data_sinfo
->bytes_may_use
= 0;
4236 data_sinfo
->bytes_may_use
-= len
;
4237 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
4238 data_sinfo
->flags
, len
, 0);
4239 spin_unlock(&data_sinfo
->lock
);
4243 * Called if we need to clear a data reservation for this inode
4244 * Normally in a error case.
4246 * This one will handle the per-indoe data rsv map for accurate reserved
4249 void btrfs_free_reserved_data_space(struct inode
*inode
, u64 start
, u64 len
)
4251 btrfs_free_reserved_data_space_noquota(inode
, start
, len
);
4252 btrfs_qgroup_free_data(inode
, start
, len
);
4255 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
4257 struct list_head
*head
= &info
->space_info
;
4258 struct btrfs_space_info
*found
;
4261 list_for_each_entry_rcu(found
, head
, list
) {
4262 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
4263 found
->force_alloc
= CHUNK_ALLOC_FORCE
;
4268 static inline u64
calc_global_rsv_need_space(struct btrfs_block_rsv
*global
)
4270 return (global
->size
<< 1);
4273 static int should_alloc_chunk(struct btrfs_root
*root
,
4274 struct btrfs_space_info
*sinfo
, int force
)
4276 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
4277 u64 num_bytes
= sinfo
->total_bytes
- sinfo
->bytes_readonly
;
4278 u64 num_allocated
= sinfo
->bytes_used
+ sinfo
->bytes_reserved
;
4281 if (force
== CHUNK_ALLOC_FORCE
)
4285 * We need to take into account the global rsv because for all intents
4286 * and purposes it's used space. Don't worry about locking the
4287 * global_rsv, it doesn't change except when the transaction commits.
4289 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
4290 num_allocated
+= calc_global_rsv_need_space(global_rsv
);
4293 * in limited mode, we want to have some free space up to
4294 * about 1% of the FS size.
4296 if (force
== CHUNK_ALLOC_LIMITED
) {
4297 thresh
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
4298 thresh
= max_t(u64
, SZ_64M
, div_factor_fine(thresh
, 1));
4300 if (num_bytes
- num_allocated
< thresh
)
4304 if (num_allocated
+ SZ_2M
< div_factor(num_bytes
, 8))
4309 static u64
get_profile_num_devs(struct btrfs_root
*root
, u64 type
)
4313 if (type
& (BTRFS_BLOCK_GROUP_RAID10
|
4314 BTRFS_BLOCK_GROUP_RAID0
|
4315 BTRFS_BLOCK_GROUP_RAID5
|
4316 BTRFS_BLOCK_GROUP_RAID6
))
4317 num_dev
= root
->fs_info
->fs_devices
->rw_devices
;
4318 else if (type
& BTRFS_BLOCK_GROUP_RAID1
)
4321 num_dev
= 1; /* DUP or single */
4327 * If @is_allocation is true, reserve space in the system space info necessary
4328 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4331 void check_system_chunk(struct btrfs_trans_handle
*trans
,
4332 struct btrfs_root
*root
,
4335 struct btrfs_space_info
*info
;
4342 * Needed because we can end up allocating a system chunk and for an
4343 * atomic and race free space reservation in the chunk block reserve.
4345 ASSERT(mutex_is_locked(&root
->fs_info
->chunk_mutex
));
4347 info
= __find_space_info(root
->fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
4348 spin_lock(&info
->lock
);
4349 left
= info
->total_bytes
- info
->bytes_used
- info
->bytes_pinned
-
4350 info
->bytes_reserved
- info
->bytes_readonly
-
4351 info
->bytes_may_use
;
4352 spin_unlock(&info
->lock
);
4354 num_devs
= get_profile_num_devs(root
, type
);
4356 /* num_devs device items to update and 1 chunk item to add or remove */
4357 thresh
= btrfs_calc_trunc_metadata_size(root
, num_devs
) +
4358 btrfs_calc_trans_metadata_size(root
, 1);
4360 if (left
< thresh
&& btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
4361 btrfs_info(root
->fs_info
, "left=%llu, need=%llu, flags=%llu",
4362 left
, thresh
, type
);
4363 dump_space_info(info
, 0, 0);
4366 if (left
< thresh
) {
4369 flags
= btrfs_get_alloc_profile(root
->fs_info
->chunk_root
, 0);
4371 * Ignore failure to create system chunk. We might end up not
4372 * needing it, as we might not need to COW all nodes/leafs from
4373 * the paths we visit in the chunk tree (they were already COWed
4374 * or created in the current transaction for example).
4376 ret
= btrfs_alloc_chunk(trans
, root
, flags
);
4380 ret
= btrfs_block_rsv_add(root
->fs_info
->chunk_root
,
4381 &root
->fs_info
->chunk_block_rsv
,
4382 thresh
, BTRFS_RESERVE_NO_FLUSH
);
4384 trans
->chunk_bytes_reserved
+= thresh
;
4388 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
4389 struct btrfs_root
*extent_root
, u64 flags
, int force
)
4391 struct btrfs_space_info
*space_info
;
4392 struct btrfs_fs_info
*fs_info
= extent_root
->fs_info
;
4393 int wait_for_alloc
= 0;
4396 /* Don't re-enter if we're already allocating a chunk */
4397 if (trans
->allocating_chunk
)
4400 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
4402 ret
= update_space_info(extent_root
->fs_info
, flags
,
4404 BUG_ON(ret
); /* -ENOMEM */
4406 BUG_ON(!space_info
); /* Logic error */
4409 spin_lock(&space_info
->lock
);
4410 if (force
< space_info
->force_alloc
)
4411 force
= space_info
->force_alloc
;
4412 if (space_info
->full
) {
4413 if (should_alloc_chunk(extent_root
, space_info
, force
))
4417 spin_unlock(&space_info
->lock
);
4421 if (!should_alloc_chunk(extent_root
, space_info
, force
)) {
4422 spin_unlock(&space_info
->lock
);
4424 } else if (space_info
->chunk_alloc
) {
4427 space_info
->chunk_alloc
= 1;
4430 spin_unlock(&space_info
->lock
);
4432 mutex_lock(&fs_info
->chunk_mutex
);
4435 * The chunk_mutex is held throughout the entirety of a chunk
4436 * allocation, so once we've acquired the chunk_mutex we know that the
4437 * other guy is done and we need to recheck and see if we should
4440 if (wait_for_alloc
) {
4441 mutex_unlock(&fs_info
->chunk_mutex
);
4446 trans
->allocating_chunk
= true;
4449 * If we have mixed data/metadata chunks we want to make sure we keep
4450 * allocating mixed chunks instead of individual chunks.
4452 if (btrfs_mixed_space_info(space_info
))
4453 flags
|= (BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
);
4456 * if we're doing a data chunk, go ahead and make sure that
4457 * we keep a reasonable number of metadata chunks allocated in the
4460 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
4461 fs_info
->data_chunk_allocations
++;
4462 if (!(fs_info
->data_chunk_allocations
%
4463 fs_info
->metadata_ratio
))
4464 force_metadata_allocation(fs_info
);
4468 * Check if we have enough space in SYSTEM chunk because we may need
4469 * to update devices.
4471 check_system_chunk(trans
, extent_root
, flags
);
4473 ret
= btrfs_alloc_chunk(trans
, extent_root
, flags
);
4474 trans
->allocating_chunk
= false;
4476 spin_lock(&space_info
->lock
);
4477 if (ret
< 0 && ret
!= -ENOSPC
)
4480 space_info
->full
= 1;
4484 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
4486 space_info
->chunk_alloc
= 0;
4487 spin_unlock(&space_info
->lock
);
4488 mutex_unlock(&fs_info
->chunk_mutex
);
4490 * When we allocate a new chunk we reserve space in the chunk block
4491 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4492 * add new nodes/leafs to it if we end up needing to do it when
4493 * inserting the chunk item and updating device items as part of the
4494 * second phase of chunk allocation, performed by
4495 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4496 * large number of new block groups to create in our transaction
4497 * handle's new_bgs list to avoid exhausting the chunk block reserve
4498 * in extreme cases - like having a single transaction create many new
4499 * block groups when starting to write out the free space caches of all
4500 * the block groups that were made dirty during the lifetime of the
4503 if (trans
->can_flush_pending_bgs
&&
4504 trans
->chunk_bytes_reserved
>= (u64
)SZ_2M
) {
4505 btrfs_create_pending_block_groups(trans
, trans
->root
);
4506 btrfs_trans_release_chunk_metadata(trans
);
4511 static int can_overcommit(struct btrfs_root
*root
,
4512 struct btrfs_space_info
*space_info
, u64 bytes
,
4513 enum btrfs_reserve_flush_enum flush
)
4515 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
4516 u64 profile
= btrfs_get_alloc_profile(root
, 0);
4521 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4522 space_info
->bytes_pinned
+ space_info
->bytes_readonly
;
4525 * We only want to allow over committing if we have lots of actual space
4526 * free, but if we don't have enough space to handle the global reserve
4527 * space then we could end up having a real enospc problem when trying
4528 * to allocate a chunk or some other such important allocation.
4530 spin_lock(&global_rsv
->lock
);
4531 space_size
= calc_global_rsv_need_space(global_rsv
);
4532 spin_unlock(&global_rsv
->lock
);
4533 if (used
+ space_size
>= space_info
->total_bytes
)
4536 used
+= space_info
->bytes_may_use
;
4538 spin_lock(&root
->fs_info
->free_chunk_lock
);
4539 avail
= root
->fs_info
->free_chunk_space
;
4540 spin_unlock(&root
->fs_info
->free_chunk_lock
);
4543 * If we have dup, raid1 or raid10 then only half of the free
4544 * space is actually useable. For raid56, the space info used
4545 * doesn't include the parity drive, so we don't have to
4548 if (profile
& (BTRFS_BLOCK_GROUP_DUP
|
4549 BTRFS_BLOCK_GROUP_RAID1
|
4550 BTRFS_BLOCK_GROUP_RAID10
))
4554 * If we aren't flushing all things, let us overcommit up to
4555 * 1/2th of the space. If we can flush, don't let us overcommit
4556 * too much, let it overcommit up to 1/8 of the space.
4558 if (flush
== BTRFS_RESERVE_FLUSH_ALL
)
4563 if (used
+ bytes
< space_info
->total_bytes
+ avail
)
4568 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root
*root
,
4569 unsigned long nr_pages
, int nr_items
)
4571 struct super_block
*sb
= root
->fs_info
->sb
;
4573 if (down_read_trylock(&sb
->s_umount
)) {
4574 writeback_inodes_sb_nr(sb
, nr_pages
, WB_REASON_FS_FREE_SPACE
);
4575 up_read(&sb
->s_umount
);
4578 * We needn't worry the filesystem going from r/w to r/o though
4579 * we don't acquire ->s_umount mutex, because the filesystem
4580 * should guarantee the delalloc inodes list be empty after
4581 * the filesystem is readonly(all dirty pages are written to
4584 btrfs_start_delalloc_roots(root
->fs_info
, 0, nr_items
);
4585 if (!current
->journal_info
)
4586 btrfs_wait_ordered_roots(root
->fs_info
, nr_items
,
4591 static inline int calc_reclaim_items_nr(struct btrfs_root
*root
, u64 to_reclaim
)
4596 bytes
= btrfs_calc_trans_metadata_size(root
, 1);
4597 nr
= (int)div64_u64(to_reclaim
, bytes
);
4603 #define EXTENT_SIZE_PER_ITEM SZ_256K
4606 * shrink metadata reservation for delalloc
4608 static void shrink_delalloc(struct btrfs_root
*root
, u64 to_reclaim
, u64 orig
,
4611 struct btrfs_block_rsv
*block_rsv
;
4612 struct btrfs_space_info
*space_info
;
4613 struct btrfs_trans_handle
*trans
;
4617 unsigned long nr_pages
;
4620 enum btrfs_reserve_flush_enum flush
;
4622 /* Calc the number of the pages we need flush for space reservation */
4623 items
= calc_reclaim_items_nr(root
, to_reclaim
);
4624 to_reclaim
= items
* EXTENT_SIZE_PER_ITEM
;
4626 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4627 block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
4628 space_info
= block_rsv
->space_info
;
4630 delalloc_bytes
= percpu_counter_sum_positive(
4631 &root
->fs_info
->delalloc_bytes
);
4632 if (delalloc_bytes
== 0) {
4636 btrfs_wait_ordered_roots(root
->fs_info
, items
,
4642 while (delalloc_bytes
&& loops
< 3) {
4643 max_reclaim
= min(delalloc_bytes
, to_reclaim
);
4644 nr_pages
= max_reclaim
>> PAGE_SHIFT
;
4645 btrfs_writeback_inodes_sb_nr(root
, nr_pages
, items
);
4647 * We need to wait for the async pages to actually start before
4650 max_reclaim
= atomic_read(&root
->fs_info
->async_delalloc_pages
);
4654 if (max_reclaim
<= nr_pages
)
4657 max_reclaim
-= nr_pages
;
4659 wait_event(root
->fs_info
->async_submit_wait
,
4660 atomic_read(&root
->fs_info
->async_delalloc_pages
) <=
4664 flush
= BTRFS_RESERVE_FLUSH_ALL
;
4666 flush
= BTRFS_RESERVE_NO_FLUSH
;
4667 spin_lock(&space_info
->lock
);
4668 if (can_overcommit(root
, space_info
, orig
, flush
)) {
4669 spin_unlock(&space_info
->lock
);
4672 spin_unlock(&space_info
->lock
);
4675 if (wait_ordered
&& !trans
) {
4676 btrfs_wait_ordered_roots(root
->fs_info
, items
,
4679 time_left
= schedule_timeout_killable(1);
4683 delalloc_bytes
= percpu_counter_sum_positive(
4684 &root
->fs_info
->delalloc_bytes
);
4689 * maybe_commit_transaction - possibly commit the transaction if its ok to
4690 * @root - the root we're allocating for
4691 * @bytes - the number of bytes we want to reserve
4692 * @force - force the commit
4694 * This will check to make sure that committing the transaction will actually
4695 * get us somewhere and then commit the transaction if it does. Otherwise it
4696 * will return -ENOSPC.
4698 static int may_commit_transaction(struct btrfs_root
*root
,
4699 struct btrfs_space_info
*space_info
,
4700 u64 bytes
, int force
)
4702 struct btrfs_block_rsv
*delayed_rsv
= &root
->fs_info
->delayed_block_rsv
;
4703 struct btrfs_trans_handle
*trans
;
4705 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4712 /* See if there is enough pinned space to make this reservation */
4713 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4718 * See if there is some space in the delayed insertion reservation for
4721 if (space_info
!= delayed_rsv
->space_info
)
4724 spin_lock(&delayed_rsv
->lock
);
4725 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4726 bytes
- delayed_rsv
->size
) >= 0) {
4727 spin_unlock(&delayed_rsv
->lock
);
4730 spin_unlock(&delayed_rsv
->lock
);
4733 trans
= btrfs_join_transaction(root
);
4737 return btrfs_commit_transaction(trans
, root
);
4741 FLUSH_DELAYED_ITEMS_NR
= 1,
4742 FLUSH_DELAYED_ITEMS
= 2,
4744 FLUSH_DELALLOC_WAIT
= 4,
4749 static int flush_space(struct btrfs_root
*root
,
4750 struct btrfs_space_info
*space_info
, u64 num_bytes
,
4751 u64 orig_bytes
, int state
)
4753 struct btrfs_trans_handle
*trans
;
4758 case FLUSH_DELAYED_ITEMS_NR
:
4759 case FLUSH_DELAYED_ITEMS
:
4760 if (state
== FLUSH_DELAYED_ITEMS_NR
)
4761 nr
= calc_reclaim_items_nr(root
, num_bytes
) * 2;
4765 trans
= btrfs_join_transaction(root
);
4766 if (IS_ERR(trans
)) {
4767 ret
= PTR_ERR(trans
);
4770 ret
= btrfs_run_delayed_items_nr(trans
, root
, nr
);
4771 btrfs_end_transaction(trans
, root
);
4773 case FLUSH_DELALLOC
:
4774 case FLUSH_DELALLOC_WAIT
:
4775 shrink_delalloc(root
, num_bytes
* 2, orig_bytes
,
4776 state
== FLUSH_DELALLOC_WAIT
);
4779 trans
= btrfs_join_transaction(root
);
4780 if (IS_ERR(trans
)) {
4781 ret
= PTR_ERR(trans
);
4784 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
4785 btrfs_get_alloc_profile(root
, 0),
4786 CHUNK_ALLOC_NO_FORCE
);
4787 btrfs_end_transaction(trans
, root
);
4792 ret
= may_commit_transaction(root
, space_info
, orig_bytes
, 0);
4803 btrfs_calc_reclaim_metadata_size(struct btrfs_root
*root
,
4804 struct btrfs_space_info
*space_info
)
4810 to_reclaim
= min_t(u64
, num_online_cpus() * SZ_1M
, SZ_16M
);
4811 spin_lock(&space_info
->lock
);
4812 if (can_overcommit(root
, space_info
, to_reclaim
,
4813 BTRFS_RESERVE_FLUSH_ALL
)) {
4818 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4819 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4820 space_info
->bytes_may_use
;
4821 if (can_overcommit(root
, space_info
, SZ_1M
, BTRFS_RESERVE_FLUSH_ALL
))
4822 expected
= div_factor_fine(space_info
->total_bytes
, 95);
4824 expected
= div_factor_fine(space_info
->total_bytes
, 90);
4826 if (used
> expected
)
4827 to_reclaim
= used
- expected
;
4830 to_reclaim
= min(to_reclaim
, space_info
->bytes_may_use
+
4831 space_info
->bytes_reserved
);
4833 spin_unlock(&space_info
->lock
);
4838 static inline int need_do_async_reclaim(struct btrfs_space_info
*space_info
,
4839 struct btrfs_fs_info
*fs_info
, u64 used
)
4841 u64 thresh
= div_factor_fine(space_info
->total_bytes
, 98);
4843 /* If we're just plain full then async reclaim just slows us down. */
4844 if ((space_info
->bytes_used
+ space_info
->bytes_reserved
) >= thresh
)
4847 return (used
>= thresh
&& !btrfs_fs_closing(fs_info
) &&
4848 !test_bit(BTRFS_FS_STATE_REMOUNTING
, &fs_info
->fs_state
));
4851 static int btrfs_need_do_async_reclaim(struct btrfs_space_info
*space_info
,
4852 struct btrfs_fs_info
*fs_info
,
4857 spin_lock(&space_info
->lock
);
4859 * We run out of space and have not got any free space via flush_space,
4860 * so don't bother doing async reclaim.
4862 if (flush_state
> COMMIT_TRANS
&& space_info
->full
) {
4863 spin_unlock(&space_info
->lock
);
4867 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4868 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4869 space_info
->bytes_may_use
;
4870 if (need_do_async_reclaim(space_info
, fs_info
, used
)) {
4871 spin_unlock(&space_info
->lock
);
4874 spin_unlock(&space_info
->lock
);
4879 static void btrfs_async_reclaim_metadata_space(struct work_struct
*work
)
4881 struct btrfs_fs_info
*fs_info
;
4882 struct btrfs_space_info
*space_info
;
4886 fs_info
= container_of(work
, struct btrfs_fs_info
, async_reclaim_work
);
4887 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4889 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
->fs_root
,
4894 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
4896 flush_space(fs_info
->fs_root
, space_info
, to_reclaim
,
4897 to_reclaim
, flush_state
);
4899 if (!btrfs_need_do_async_reclaim(space_info
, fs_info
,
4902 } while (flush_state
< COMMIT_TRANS
);
4905 void btrfs_init_async_reclaim_work(struct work_struct
*work
)
4907 INIT_WORK(work
, btrfs_async_reclaim_metadata_space
);
4911 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4912 * @root - the root we're allocating for
4913 * @block_rsv - the block_rsv we're allocating for
4914 * @orig_bytes - the number of bytes we want
4915 * @flush - whether or not we can flush to make our reservation
4917 * This will reserve orgi_bytes number of bytes from the space info associated
4918 * with the block_rsv. If there is not enough space it will make an attempt to
4919 * flush out space to make room. It will do this by flushing delalloc if
4920 * possible or committing the transaction. If flush is 0 then no attempts to
4921 * regain reservations will be made and this will fail if there is not enough
4924 static int reserve_metadata_bytes(struct btrfs_root
*root
,
4925 struct btrfs_block_rsv
*block_rsv
,
4927 enum btrfs_reserve_flush_enum flush
)
4929 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
4931 u64 num_bytes
= orig_bytes
;
4932 int flush_state
= FLUSH_DELAYED_ITEMS_NR
;
4934 bool flushing
= false;
4938 spin_lock(&space_info
->lock
);
4940 * We only want to wait if somebody other than us is flushing and we
4941 * are actually allowed to flush all things.
4943 while (flush
== BTRFS_RESERVE_FLUSH_ALL
&& !flushing
&&
4944 space_info
->flush
) {
4945 spin_unlock(&space_info
->lock
);
4947 * If we have a trans handle we can't wait because the flusher
4948 * may have to commit the transaction, which would mean we would
4949 * deadlock since we are waiting for the flusher to finish, but
4950 * hold the current transaction open.
4952 if (current
->journal_info
)
4954 ret
= wait_event_killable(space_info
->wait
, !space_info
->flush
);
4955 /* Must have been killed, return */
4959 spin_lock(&space_info
->lock
);
4963 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4964 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4965 space_info
->bytes_may_use
;
4968 * The idea here is that we've not already over-reserved the block group
4969 * then we can go ahead and save our reservation first and then start
4970 * flushing if we need to. Otherwise if we've already overcommitted
4971 * lets start flushing stuff first and then come back and try to make
4974 if (used
<= space_info
->total_bytes
) {
4975 if (used
+ orig_bytes
<= space_info
->total_bytes
) {
4976 space_info
->bytes_may_use
+= orig_bytes
;
4977 trace_btrfs_space_reservation(root
->fs_info
,
4978 "space_info", space_info
->flags
, orig_bytes
, 1);
4982 * Ok set num_bytes to orig_bytes since we aren't
4983 * overocmmitted, this way we only try and reclaim what
4986 num_bytes
= orig_bytes
;
4990 * Ok we're over committed, set num_bytes to the overcommitted
4991 * amount plus the amount of bytes that we need for this
4994 num_bytes
= used
- space_info
->total_bytes
+
4998 if (ret
&& can_overcommit(root
, space_info
, orig_bytes
, flush
)) {
4999 space_info
->bytes_may_use
+= orig_bytes
;
5000 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
5001 space_info
->flags
, orig_bytes
,
5007 * Couldn't make our reservation, save our place so while we're trying
5008 * to reclaim space we can actually use it instead of somebody else
5009 * stealing it from us.
5011 * We make the other tasks wait for the flush only when we can flush
5014 if (ret
&& flush
!= BTRFS_RESERVE_NO_FLUSH
) {
5016 space_info
->flush
= 1;
5017 } else if (!ret
&& space_info
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
5020 * We will do the space reservation dance during log replay,
5021 * which means we won't have fs_info->fs_root set, so don't do
5022 * the async reclaim as we will panic.
5024 if (!root
->fs_info
->log_root_recovering
&&
5025 need_do_async_reclaim(space_info
, root
->fs_info
, used
) &&
5026 !work_busy(&root
->fs_info
->async_reclaim_work
))
5027 queue_work(system_unbound_wq
,
5028 &root
->fs_info
->async_reclaim_work
);
5030 spin_unlock(&space_info
->lock
);
5032 if (!ret
|| flush
== BTRFS_RESERVE_NO_FLUSH
)
5035 ret
= flush_space(root
, space_info
, num_bytes
, orig_bytes
,
5040 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
5041 * would happen. So skip delalloc flush.
5043 if (flush
== BTRFS_RESERVE_FLUSH_LIMIT
&&
5044 (flush_state
== FLUSH_DELALLOC
||
5045 flush_state
== FLUSH_DELALLOC_WAIT
))
5046 flush_state
= ALLOC_CHUNK
;
5050 else if (flush
== BTRFS_RESERVE_FLUSH_LIMIT
&&
5051 flush_state
< COMMIT_TRANS
)
5053 else if (flush
== BTRFS_RESERVE_FLUSH_ALL
&&
5054 flush_state
<= COMMIT_TRANS
)
5058 if (ret
== -ENOSPC
&&
5059 unlikely(root
->orphan_cleanup_state
== ORPHAN_CLEANUP_STARTED
)) {
5060 struct btrfs_block_rsv
*global_rsv
=
5061 &root
->fs_info
->global_block_rsv
;
5063 if (block_rsv
!= global_rsv
&&
5064 !block_rsv_use_bytes(global_rsv
, orig_bytes
))
5068 trace_btrfs_space_reservation(root
->fs_info
,
5069 "space_info:enospc",
5070 space_info
->flags
, orig_bytes
, 1);
5072 spin_lock(&space_info
->lock
);
5073 space_info
->flush
= 0;
5074 wake_up_all(&space_info
->wait
);
5075 spin_unlock(&space_info
->lock
);
5080 static struct btrfs_block_rsv
*get_block_rsv(
5081 const struct btrfs_trans_handle
*trans
,
5082 const struct btrfs_root
*root
)
5084 struct btrfs_block_rsv
*block_rsv
= NULL
;
5086 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
5087 (root
== root
->fs_info
->csum_root
&& trans
->adding_csums
) ||
5088 (root
== root
->fs_info
->uuid_root
))
5089 block_rsv
= trans
->block_rsv
;
5092 block_rsv
= root
->block_rsv
;
5095 block_rsv
= &root
->fs_info
->empty_block_rsv
;
5100 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
5104 spin_lock(&block_rsv
->lock
);
5105 if (block_rsv
->reserved
>= num_bytes
) {
5106 block_rsv
->reserved
-= num_bytes
;
5107 if (block_rsv
->reserved
< block_rsv
->size
)
5108 block_rsv
->full
= 0;
5111 spin_unlock(&block_rsv
->lock
);
5115 static void block_rsv_add_bytes(struct btrfs_block_rsv
*block_rsv
,
5116 u64 num_bytes
, int update_size
)
5118 spin_lock(&block_rsv
->lock
);
5119 block_rsv
->reserved
+= num_bytes
;
5121 block_rsv
->size
+= num_bytes
;
5122 else if (block_rsv
->reserved
>= block_rsv
->size
)
5123 block_rsv
->full
= 1;
5124 spin_unlock(&block_rsv
->lock
);
5127 int btrfs_cond_migrate_bytes(struct btrfs_fs_info
*fs_info
,
5128 struct btrfs_block_rsv
*dest
, u64 num_bytes
,
5131 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5134 if (global_rsv
->space_info
!= dest
->space_info
)
5137 spin_lock(&global_rsv
->lock
);
5138 min_bytes
= div_factor(global_rsv
->size
, min_factor
);
5139 if (global_rsv
->reserved
< min_bytes
+ num_bytes
) {
5140 spin_unlock(&global_rsv
->lock
);
5143 global_rsv
->reserved
-= num_bytes
;
5144 if (global_rsv
->reserved
< global_rsv
->size
)
5145 global_rsv
->full
= 0;
5146 spin_unlock(&global_rsv
->lock
);
5148 block_rsv_add_bytes(dest
, num_bytes
, 1);
5152 static void block_rsv_release_bytes(struct btrfs_fs_info
*fs_info
,
5153 struct btrfs_block_rsv
*block_rsv
,
5154 struct btrfs_block_rsv
*dest
, u64 num_bytes
)
5156 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
5158 spin_lock(&block_rsv
->lock
);
5159 if (num_bytes
== (u64
)-1)
5160 num_bytes
= block_rsv
->size
;
5161 block_rsv
->size
-= num_bytes
;
5162 if (block_rsv
->reserved
>= block_rsv
->size
) {
5163 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
5164 block_rsv
->reserved
= block_rsv
->size
;
5165 block_rsv
->full
= 1;
5169 spin_unlock(&block_rsv
->lock
);
5171 if (num_bytes
> 0) {
5173 spin_lock(&dest
->lock
);
5177 bytes_to_add
= dest
->size
- dest
->reserved
;
5178 bytes_to_add
= min(num_bytes
, bytes_to_add
);
5179 dest
->reserved
+= bytes_to_add
;
5180 if (dest
->reserved
>= dest
->size
)
5182 num_bytes
-= bytes_to_add
;
5184 spin_unlock(&dest
->lock
);
5187 spin_lock(&space_info
->lock
);
5188 space_info
->bytes_may_use
-= num_bytes
;
5189 trace_btrfs_space_reservation(fs_info
, "space_info",
5190 space_info
->flags
, num_bytes
, 0);
5191 spin_unlock(&space_info
->lock
);
5196 static int block_rsv_migrate_bytes(struct btrfs_block_rsv
*src
,
5197 struct btrfs_block_rsv
*dst
, u64 num_bytes
)
5201 ret
= block_rsv_use_bytes(src
, num_bytes
);
5205 block_rsv_add_bytes(dst
, num_bytes
, 1);
5209 void btrfs_init_block_rsv(struct btrfs_block_rsv
*rsv
, unsigned short type
)
5211 memset(rsv
, 0, sizeof(*rsv
));
5212 spin_lock_init(&rsv
->lock
);
5216 struct btrfs_block_rsv
*btrfs_alloc_block_rsv(struct btrfs_root
*root
,
5217 unsigned short type
)
5219 struct btrfs_block_rsv
*block_rsv
;
5220 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5222 block_rsv
= kmalloc(sizeof(*block_rsv
), GFP_NOFS
);
5226 btrfs_init_block_rsv(block_rsv
, type
);
5227 block_rsv
->space_info
= __find_space_info(fs_info
,
5228 BTRFS_BLOCK_GROUP_METADATA
);
5232 void btrfs_free_block_rsv(struct btrfs_root
*root
,
5233 struct btrfs_block_rsv
*rsv
)
5237 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
5241 void __btrfs_free_block_rsv(struct btrfs_block_rsv
*rsv
)
5246 int btrfs_block_rsv_add(struct btrfs_root
*root
,
5247 struct btrfs_block_rsv
*block_rsv
, u64 num_bytes
,
5248 enum btrfs_reserve_flush_enum flush
)
5255 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
5257 block_rsv_add_bytes(block_rsv
, num_bytes
, 1);
5264 int btrfs_block_rsv_check(struct btrfs_root
*root
,
5265 struct btrfs_block_rsv
*block_rsv
, int min_factor
)
5273 spin_lock(&block_rsv
->lock
);
5274 num_bytes
= div_factor(block_rsv
->size
, min_factor
);
5275 if (block_rsv
->reserved
>= num_bytes
)
5277 spin_unlock(&block_rsv
->lock
);
5282 int btrfs_block_rsv_refill(struct btrfs_root
*root
,
5283 struct btrfs_block_rsv
*block_rsv
, u64 min_reserved
,
5284 enum btrfs_reserve_flush_enum flush
)
5292 spin_lock(&block_rsv
->lock
);
5293 num_bytes
= min_reserved
;
5294 if (block_rsv
->reserved
>= num_bytes
)
5297 num_bytes
-= block_rsv
->reserved
;
5298 spin_unlock(&block_rsv
->lock
);
5303 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
5305 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
5312 int btrfs_block_rsv_migrate(struct btrfs_block_rsv
*src_rsv
,
5313 struct btrfs_block_rsv
*dst_rsv
,
5316 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
5319 void btrfs_block_rsv_release(struct btrfs_root
*root
,
5320 struct btrfs_block_rsv
*block_rsv
,
5323 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
5324 if (global_rsv
== block_rsv
||
5325 block_rsv
->space_info
!= global_rsv
->space_info
)
5327 block_rsv_release_bytes(root
->fs_info
, block_rsv
, global_rsv
,
5332 * helper to calculate size of global block reservation.
5333 * the desired value is sum of space used by extent tree,
5334 * checksum tree and root tree
5336 static u64
calc_global_metadata_size(struct btrfs_fs_info
*fs_info
)
5338 struct btrfs_space_info
*sinfo
;
5342 int csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
5344 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_DATA
);
5345 spin_lock(&sinfo
->lock
);
5346 data_used
= sinfo
->bytes_used
;
5347 spin_unlock(&sinfo
->lock
);
5349 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
5350 spin_lock(&sinfo
->lock
);
5351 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_DATA
)
5353 meta_used
= sinfo
->bytes_used
;
5354 spin_unlock(&sinfo
->lock
);
5356 num_bytes
= (data_used
>> fs_info
->sb
->s_blocksize_bits
) *
5358 num_bytes
+= div_u64(data_used
+ meta_used
, 50);
5360 if (num_bytes
* 3 > meta_used
)
5361 num_bytes
= div_u64(meta_used
, 3);
5363 return ALIGN(num_bytes
, fs_info
->extent_root
->nodesize
<< 10);
5366 static void update_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5368 struct btrfs_block_rsv
*block_rsv
= &fs_info
->global_block_rsv
;
5369 struct btrfs_space_info
*sinfo
= block_rsv
->space_info
;
5372 num_bytes
= calc_global_metadata_size(fs_info
);
5374 spin_lock(&sinfo
->lock
);
5375 spin_lock(&block_rsv
->lock
);
5377 block_rsv
->size
= min_t(u64
, num_bytes
, SZ_512M
);
5379 if (block_rsv
->reserved
< block_rsv
->size
) {
5380 num_bytes
= sinfo
->bytes_used
+ sinfo
->bytes_pinned
+
5381 sinfo
->bytes_reserved
+ sinfo
->bytes_readonly
+
5382 sinfo
->bytes_may_use
;
5383 if (sinfo
->total_bytes
> num_bytes
) {
5384 num_bytes
= sinfo
->total_bytes
- num_bytes
;
5385 num_bytes
= min(num_bytes
,
5386 block_rsv
->size
- block_rsv
->reserved
);
5387 block_rsv
->reserved
+= num_bytes
;
5388 sinfo
->bytes_may_use
+= num_bytes
;
5389 trace_btrfs_space_reservation(fs_info
, "space_info",
5390 sinfo
->flags
, num_bytes
,
5393 } else if (block_rsv
->reserved
> block_rsv
->size
) {
5394 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
5395 sinfo
->bytes_may_use
-= num_bytes
;
5396 trace_btrfs_space_reservation(fs_info
, "space_info",
5397 sinfo
->flags
, num_bytes
, 0);
5398 block_rsv
->reserved
= block_rsv
->size
;
5401 if (block_rsv
->reserved
== block_rsv
->size
)
5402 block_rsv
->full
= 1;
5404 block_rsv
->full
= 0;
5406 spin_unlock(&block_rsv
->lock
);
5407 spin_unlock(&sinfo
->lock
);
5410 static void init_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5412 struct btrfs_space_info
*space_info
;
5414 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
5415 fs_info
->chunk_block_rsv
.space_info
= space_info
;
5417 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
5418 fs_info
->global_block_rsv
.space_info
= space_info
;
5419 fs_info
->delalloc_block_rsv
.space_info
= space_info
;
5420 fs_info
->trans_block_rsv
.space_info
= space_info
;
5421 fs_info
->empty_block_rsv
.space_info
= space_info
;
5422 fs_info
->delayed_block_rsv
.space_info
= space_info
;
5424 fs_info
->extent_root
->block_rsv
= &fs_info
->global_block_rsv
;
5425 fs_info
->csum_root
->block_rsv
= &fs_info
->global_block_rsv
;
5426 fs_info
->dev_root
->block_rsv
= &fs_info
->global_block_rsv
;
5427 fs_info
->tree_root
->block_rsv
= &fs_info
->global_block_rsv
;
5428 if (fs_info
->quota_root
)
5429 fs_info
->quota_root
->block_rsv
= &fs_info
->global_block_rsv
;
5430 fs_info
->chunk_root
->block_rsv
= &fs_info
->chunk_block_rsv
;
5432 update_global_block_rsv(fs_info
);
5435 static void release_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5437 block_rsv_release_bytes(fs_info
, &fs_info
->global_block_rsv
, NULL
,
5439 WARN_ON(fs_info
->delalloc_block_rsv
.size
> 0);
5440 WARN_ON(fs_info
->delalloc_block_rsv
.reserved
> 0);
5441 WARN_ON(fs_info
->trans_block_rsv
.size
> 0);
5442 WARN_ON(fs_info
->trans_block_rsv
.reserved
> 0);
5443 WARN_ON(fs_info
->chunk_block_rsv
.size
> 0);
5444 WARN_ON(fs_info
->chunk_block_rsv
.reserved
> 0);
5445 WARN_ON(fs_info
->delayed_block_rsv
.size
> 0);
5446 WARN_ON(fs_info
->delayed_block_rsv
.reserved
> 0);
5449 void btrfs_trans_release_metadata(struct btrfs_trans_handle
*trans
,
5450 struct btrfs_root
*root
)
5452 if (!trans
->block_rsv
)
5455 if (!trans
->bytes_reserved
)
5458 trace_btrfs_space_reservation(root
->fs_info
, "transaction",
5459 trans
->transid
, trans
->bytes_reserved
, 0);
5460 btrfs_block_rsv_release(root
, trans
->block_rsv
, trans
->bytes_reserved
);
5461 trans
->bytes_reserved
= 0;
5465 * To be called after all the new block groups attached to the transaction
5466 * handle have been created (btrfs_create_pending_block_groups()).
5468 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle
*trans
)
5470 struct btrfs_fs_info
*fs_info
= trans
->root
->fs_info
;
5472 if (!trans
->chunk_bytes_reserved
)
5475 WARN_ON_ONCE(!list_empty(&trans
->new_bgs
));
5477 block_rsv_release_bytes(fs_info
, &fs_info
->chunk_block_rsv
, NULL
,
5478 trans
->chunk_bytes_reserved
);
5479 trans
->chunk_bytes_reserved
= 0;
5482 /* Can only return 0 or -ENOSPC */
5483 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle
*trans
,
5484 struct inode
*inode
)
5486 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5487 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
5488 struct btrfs_block_rsv
*dst_rsv
= root
->orphan_block_rsv
;
5491 * We need to hold space in order to delete our orphan item once we've
5492 * added it, so this takes the reservation so we can release it later
5493 * when we are truly done with the orphan item.
5495 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
5496 trace_btrfs_space_reservation(root
->fs_info
, "orphan",
5497 btrfs_ino(inode
), num_bytes
, 1);
5498 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
5501 void btrfs_orphan_release_metadata(struct inode
*inode
)
5503 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5504 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
5505 trace_btrfs_space_reservation(root
->fs_info
, "orphan",
5506 btrfs_ino(inode
), num_bytes
, 0);
5507 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
, num_bytes
);
5511 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5512 * root: the root of the parent directory
5513 * rsv: block reservation
5514 * items: the number of items that we need do reservation
5515 * qgroup_reserved: used to return the reserved size in qgroup
5517 * This function is used to reserve the space for snapshot/subvolume
5518 * creation and deletion. Those operations are different with the
5519 * common file/directory operations, they change two fs/file trees
5520 * and root tree, the number of items that the qgroup reserves is
5521 * different with the free space reservation. So we can not use
5522 * the space reseravtion mechanism in start_transaction().
5524 int btrfs_subvolume_reserve_metadata(struct btrfs_root
*root
,
5525 struct btrfs_block_rsv
*rsv
,
5527 u64
*qgroup_reserved
,
5528 bool use_global_rsv
)
5532 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
5534 if (root
->fs_info
->quota_enabled
) {
5535 /* One for parent inode, two for dir entries */
5536 num_bytes
= 3 * root
->nodesize
;
5537 ret
= btrfs_qgroup_reserve_meta(root
, num_bytes
);
5544 *qgroup_reserved
= num_bytes
;
5546 num_bytes
= btrfs_calc_trans_metadata_size(root
, items
);
5547 rsv
->space_info
= __find_space_info(root
->fs_info
,
5548 BTRFS_BLOCK_GROUP_METADATA
);
5549 ret
= btrfs_block_rsv_add(root
, rsv
, num_bytes
,
5550 BTRFS_RESERVE_FLUSH_ALL
);
5552 if (ret
== -ENOSPC
&& use_global_rsv
)
5553 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
, num_bytes
);
5555 if (ret
&& *qgroup_reserved
)
5556 btrfs_qgroup_free_meta(root
, *qgroup_reserved
);
5561 void btrfs_subvolume_release_metadata(struct btrfs_root
*root
,
5562 struct btrfs_block_rsv
*rsv
,
5563 u64 qgroup_reserved
)
5565 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
5569 * drop_outstanding_extent - drop an outstanding extent
5570 * @inode: the inode we're dropping the extent for
5571 * @num_bytes: the number of bytes we're relaseing.
5573 * This is called when we are freeing up an outstanding extent, either called
5574 * after an error or after an extent is written. This will return the number of
5575 * reserved extents that need to be freed. This must be called with
5576 * BTRFS_I(inode)->lock held.
5578 static unsigned drop_outstanding_extent(struct inode
*inode
, u64 num_bytes
)
5580 unsigned drop_inode_space
= 0;
5581 unsigned dropped_extents
= 0;
5582 unsigned num_extents
= 0;
5584 num_extents
= (unsigned)div64_u64(num_bytes
+
5585 BTRFS_MAX_EXTENT_SIZE
- 1,
5586 BTRFS_MAX_EXTENT_SIZE
);
5587 ASSERT(num_extents
);
5588 ASSERT(BTRFS_I(inode
)->outstanding_extents
>= num_extents
);
5589 BTRFS_I(inode
)->outstanding_extents
-= num_extents
;
5591 if (BTRFS_I(inode
)->outstanding_extents
== 0 &&
5592 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5593 &BTRFS_I(inode
)->runtime_flags
))
5594 drop_inode_space
= 1;
5597 * If we have more or the same amount of outsanding extents than we have
5598 * reserved then we need to leave the reserved extents count alone.
5600 if (BTRFS_I(inode
)->outstanding_extents
>=
5601 BTRFS_I(inode
)->reserved_extents
)
5602 return drop_inode_space
;
5604 dropped_extents
= BTRFS_I(inode
)->reserved_extents
-
5605 BTRFS_I(inode
)->outstanding_extents
;
5606 BTRFS_I(inode
)->reserved_extents
-= dropped_extents
;
5607 return dropped_extents
+ drop_inode_space
;
5611 * calc_csum_metadata_size - return the amount of metada space that must be
5612 * reserved/free'd for the given bytes.
5613 * @inode: the inode we're manipulating
5614 * @num_bytes: the number of bytes in question
5615 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5617 * This adjusts the number of csum_bytes in the inode and then returns the
5618 * correct amount of metadata that must either be reserved or freed. We
5619 * calculate how many checksums we can fit into one leaf and then divide the
5620 * number of bytes that will need to be checksumed by this value to figure out
5621 * how many checksums will be required. If we are adding bytes then the number
5622 * may go up and we will return the number of additional bytes that must be
5623 * reserved. If it is going down we will return the number of bytes that must
5626 * This must be called with BTRFS_I(inode)->lock held.
5628 static u64
calc_csum_metadata_size(struct inode
*inode
, u64 num_bytes
,
5631 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5632 u64 old_csums
, num_csums
;
5634 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
&&
5635 BTRFS_I(inode
)->csum_bytes
== 0)
5638 old_csums
= btrfs_csum_bytes_to_leaves(root
, BTRFS_I(inode
)->csum_bytes
);
5640 BTRFS_I(inode
)->csum_bytes
+= num_bytes
;
5642 BTRFS_I(inode
)->csum_bytes
-= num_bytes
;
5643 num_csums
= btrfs_csum_bytes_to_leaves(root
, BTRFS_I(inode
)->csum_bytes
);
5645 /* No change, no need to reserve more */
5646 if (old_csums
== num_csums
)
5650 return btrfs_calc_trans_metadata_size(root
,
5651 num_csums
- old_csums
);
5653 return btrfs_calc_trans_metadata_size(root
, old_csums
- num_csums
);
5656 int btrfs_delalloc_reserve_metadata(struct inode
*inode
, u64 num_bytes
)
5658 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5659 struct btrfs_block_rsv
*block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
5662 unsigned nr_extents
= 0;
5663 int extra_reserve
= 0;
5664 enum btrfs_reserve_flush_enum flush
= BTRFS_RESERVE_FLUSH_ALL
;
5666 bool delalloc_lock
= true;
5670 /* If we are a free space inode we need to not flush since we will be in
5671 * the middle of a transaction commit. We also don't need the delalloc
5672 * mutex since we won't race with anybody. We need this mostly to make
5673 * lockdep shut its filthy mouth.
5675 if (btrfs_is_free_space_inode(inode
)) {
5676 flush
= BTRFS_RESERVE_NO_FLUSH
;
5677 delalloc_lock
= false;
5680 if (flush
!= BTRFS_RESERVE_NO_FLUSH
&&
5681 btrfs_transaction_in_commit(root
->fs_info
))
5682 schedule_timeout(1);
5685 mutex_lock(&BTRFS_I(inode
)->delalloc_mutex
);
5687 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
5689 spin_lock(&BTRFS_I(inode
)->lock
);
5690 nr_extents
= (unsigned)div64_u64(num_bytes
+
5691 BTRFS_MAX_EXTENT_SIZE
- 1,
5692 BTRFS_MAX_EXTENT_SIZE
);
5693 BTRFS_I(inode
)->outstanding_extents
+= nr_extents
;
5696 if (BTRFS_I(inode
)->outstanding_extents
>
5697 BTRFS_I(inode
)->reserved_extents
)
5698 nr_extents
= BTRFS_I(inode
)->outstanding_extents
-
5699 BTRFS_I(inode
)->reserved_extents
;
5702 * Add an item to reserve for updating the inode when we complete the
5705 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5706 &BTRFS_I(inode
)->runtime_flags
)) {
5711 to_reserve
= btrfs_calc_trans_metadata_size(root
, nr_extents
);
5712 to_reserve
+= calc_csum_metadata_size(inode
, num_bytes
, 1);
5713 csum_bytes
= BTRFS_I(inode
)->csum_bytes
;
5714 spin_unlock(&BTRFS_I(inode
)->lock
);
5716 if (root
->fs_info
->quota_enabled
) {
5717 ret
= btrfs_qgroup_reserve_meta(root
,
5718 nr_extents
* root
->nodesize
);
5723 ret
= reserve_metadata_bytes(root
, block_rsv
, to_reserve
, flush
);
5724 if (unlikely(ret
)) {
5725 btrfs_qgroup_free_meta(root
, nr_extents
* root
->nodesize
);
5729 spin_lock(&BTRFS_I(inode
)->lock
);
5730 if (extra_reserve
) {
5731 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5732 &BTRFS_I(inode
)->runtime_flags
);
5735 BTRFS_I(inode
)->reserved_extents
+= nr_extents
;
5736 spin_unlock(&BTRFS_I(inode
)->lock
);
5739 mutex_unlock(&BTRFS_I(inode
)->delalloc_mutex
);
5742 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5743 btrfs_ino(inode
), to_reserve
, 1);
5744 block_rsv_add_bytes(block_rsv
, to_reserve
, 1);
5749 spin_lock(&BTRFS_I(inode
)->lock
);
5750 dropped
= drop_outstanding_extent(inode
, num_bytes
);
5752 * If the inodes csum_bytes is the same as the original
5753 * csum_bytes then we know we haven't raced with any free()ers
5754 * so we can just reduce our inodes csum bytes and carry on.
5756 if (BTRFS_I(inode
)->csum_bytes
== csum_bytes
) {
5757 calc_csum_metadata_size(inode
, num_bytes
, 0);
5759 u64 orig_csum_bytes
= BTRFS_I(inode
)->csum_bytes
;
5763 * This is tricky, but first we need to figure out how much we
5764 * free'd from any free-ers that occurred during this
5765 * reservation, so we reset ->csum_bytes to the csum_bytes
5766 * before we dropped our lock, and then call the free for the
5767 * number of bytes that were freed while we were trying our
5770 bytes
= csum_bytes
- BTRFS_I(inode
)->csum_bytes
;
5771 BTRFS_I(inode
)->csum_bytes
= csum_bytes
;
5772 to_free
= calc_csum_metadata_size(inode
, bytes
, 0);
5776 * Now we need to see how much we would have freed had we not
5777 * been making this reservation and our ->csum_bytes were not
5778 * artificially inflated.
5780 BTRFS_I(inode
)->csum_bytes
= csum_bytes
- num_bytes
;
5781 bytes
= csum_bytes
- orig_csum_bytes
;
5782 bytes
= calc_csum_metadata_size(inode
, bytes
, 0);
5785 * Now reset ->csum_bytes to what it should be. If bytes is
5786 * more than to_free then we would have free'd more space had we
5787 * not had an artificially high ->csum_bytes, so we need to free
5788 * the remainder. If bytes is the same or less then we don't
5789 * need to do anything, the other free-ers did the correct
5792 BTRFS_I(inode
)->csum_bytes
= orig_csum_bytes
- num_bytes
;
5793 if (bytes
> to_free
)
5794 to_free
= bytes
- to_free
;
5798 spin_unlock(&BTRFS_I(inode
)->lock
);
5800 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
5803 btrfs_block_rsv_release(root
, block_rsv
, to_free
);
5804 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5805 btrfs_ino(inode
), to_free
, 0);
5808 mutex_unlock(&BTRFS_I(inode
)->delalloc_mutex
);
5813 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5814 * @inode: the inode to release the reservation for
5815 * @num_bytes: the number of bytes we're releasing
5817 * This will release the metadata reservation for an inode. This can be called
5818 * once we complete IO for a given set of bytes to release their metadata
5821 void btrfs_delalloc_release_metadata(struct inode
*inode
, u64 num_bytes
)
5823 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5827 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
5828 spin_lock(&BTRFS_I(inode
)->lock
);
5829 dropped
= drop_outstanding_extent(inode
, num_bytes
);
5832 to_free
= calc_csum_metadata_size(inode
, num_bytes
, 0);
5833 spin_unlock(&BTRFS_I(inode
)->lock
);
5835 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
5837 if (btrfs_test_is_dummy_root(root
))
5840 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5841 btrfs_ino(inode
), to_free
, 0);
5843 btrfs_block_rsv_release(root
, &root
->fs_info
->delalloc_block_rsv
,
5848 * btrfs_delalloc_reserve_space - reserve data and metadata space for
5850 * @inode: inode we're writing to
5851 * @start: start range we are writing to
5852 * @len: how long the range we are writing to
5854 * TODO: This function will finally replace old btrfs_delalloc_reserve_space()
5856 * This will do the following things
5858 * o reserve space in data space info for num bytes
5859 * and reserve precious corresponding qgroup space
5860 * (Done in check_data_free_space)
5862 * o reserve space for metadata space, based on the number of outstanding
5863 * extents and how much csums will be needed
5864 * also reserve metadata space in a per root over-reserve method.
5865 * o add to the inodes->delalloc_bytes
5866 * o add it to the fs_info's delalloc inodes list.
5867 * (Above 3 all done in delalloc_reserve_metadata)
5869 * Return 0 for success
5870 * Return <0 for error(-ENOSPC or -EQUOT)
5872 int btrfs_delalloc_reserve_space(struct inode
*inode
, u64 start
, u64 len
)
5876 ret
= btrfs_check_data_free_space(inode
, start
, len
);
5879 ret
= btrfs_delalloc_reserve_metadata(inode
, len
);
5881 btrfs_free_reserved_data_space(inode
, start
, len
);
5886 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5887 * @inode: inode we're releasing space for
5888 * @start: start position of the space already reserved
5889 * @len: the len of the space already reserved
5891 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5892 * called in the case that we don't need the metadata AND data reservations
5893 * anymore. So if there is an error or we insert an inline extent.
5895 * This function will release the metadata space that was not used and will
5896 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5897 * list if there are no delalloc bytes left.
5898 * Also it will handle the qgroup reserved space.
5900 void btrfs_delalloc_release_space(struct inode
*inode
, u64 start
, u64 len
)
5902 btrfs_delalloc_release_metadata(inode
, len
);
5903 btrfs_free_reserved_data_space(inode
, start
, len
);
5906 static int update_block_group(struct btrfs_trans_handle
*trans
,
5907 struct btrfs_root
*root
, u64 bytenr
,
5908 u64 num_bytes
, int alloc
)
5910 struct btrfs_block_group_cache
*cache
= NULL
;
5911 struct btrfs_fs_info
*info
= root
->fs_info
;
5912 u64 total
= num_bytes
;
5917 /* block accounting for super block */
5918 spin_lock(&info
->delalloc_root_lock
);
5919 old_val
= btrfs_super_bytes_used(info
->super_copy
);
5921 old_val
+= num_bytes
;
5923 old_val
-= num_bytes
;
5924 btrfs_set_super_bytes_used(info
->super_copy
, old_val
);
5925 spin_unlock(&info
->delalloc_root_lock
);
5928 cache
= btrfs_lookup_block_group(info
, bytenr
);
5931 if (cache
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
5932 BTRFS_BLOCK_GROUP_RAID1
|
5933 BTRFS_BLOCK_GROUP_RAID10
))
5938 * If this block group has free space cache written out, we
5939 * need to make sure to load it if we are removing space. This
5940 * is because we need the unpinning stage to actually add the
5941 * space back to the block group, otherwise we will leak space.
5943 if (!alloc
&& cache
->cached
== BTRFS_CACHE_NO
)
5944 cache_block_group(cache
, 1);
5946 byte_in_group
= bytenr
- cache
->key
.objectid
;
5947 WARN_ON(byte_in_group
> cache
->key
.offset
);
5949 spin_lock(&cache
->space_info
->lock
);
5950 spin_lock(&cache
->lock
);
5952 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
5953 cache
->disk_cache_state
< BTRFS_DC_CLEAR
)
5954 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
5956 old_val
= btrfs_block_group_used(&cache
->item
);
5957 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
5959 old_val
+= num_bytes
;
5960 btrfs_set_block_group_used(&cache
->item
, old_val
);
5961 cache
->reserved
-= num_bytes
;
5962 cache
->space_info
->bytes_reserved
-= num_bytes
;
5963 cache
->space_info
->bytes_used
+= num_bytes
;
5964 cache
->space_info
->disk_used
+= num_bytes
* factor
;
5965 spin_unlock(&cache
->lock
);
5966 spin_unlock(&cache
->space_info
->lock
);
5968 old_val
-= num_bytes
;
5969 btrfs_set_block_group_used(&cache
->item
, old_val
);
5970 cache
->pinned
+= num_bytes
;
5971 cache
->space_info
->bytes_pinned
+= num_bytes
;
5972 cache
->space_info
->bytes_used
-= num_bytes
;
5973 cache
->space_info
->disk_used
-= num_bytes
* factor
;
5974 spin_unlock(&cache
->lock
);
5975 spin_unlock(&cache
->space_info
->lock
);
5977 set_extent_dirty(info
->pinned_extents
,
5978 bytenr
, bytenr
+ num_bytes
- 1,
5979 GFP_NOFS
| __GFP_NOFAIL
);
5982 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
5983 if (list_empty(&cache
->dirty_list
)) {
5984 list_add_tail(&cache
->dirty_list
,
5985 &trans
->transaction
->dirty_bgs
);
5986 trans
->transaction
->num_dirty_bgs
++;
5987 btrfs_get_block_group(cache
);
5989 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
5992 * No longer have used bytes in this block group, queue it for
5993 * deletion. We do this after adding the block group to the
5994 * dirty list to avoid races between cleaner kthread and space
5997 if (!alloc
&& old_val
== 0) {
5998 spin_lock(&info
->unused_bgs_lock
);
5999 if (list_empty(&cache
->bg_list
)) {
6000 btrfs_get_block_group(cache
);
6001 list_add_tail(&cache
->bg_list
,
6004 spin_unlock(&info
->unused_bgs_lock
);
6007 btrfs_put_block_group(cache
);
6009 bytenr
+= num_bytes
;
6014 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
6016 struct btrfs_block_group_cache
*cache
;
6019 spin_lock(&root
->fs_info
->block_group_cache_lock
);
6020 bytenr
= root
->fs_info
->first_logical_byte
;
6021 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
6023 if (bytenr
< (u64
)-1)
6026 cache
= btrfs_lookup_first_block_group(root
->fs_info
, search_start
);
6030 bytenr
= cache
->key
.objectid
;
6031 btrfs_put_block_group(cache
);
6036 static int pin_down_extent(struct btrfs_root
*root
,
6037 struct btrfs_block_group_cache
*cache
,
6038 u64 bytenr
, u64 num_bytes
, int reserved
)
6040 spin_lock(&cache
->space_info
->lock
);
6041 spin_lock(&cache
->lock
);
6042 cache
->pinned
+= num_bytes
;
6043 cache
->space_info
->bytes_pinned
+= num_bytes
;
6045 cache
->reserved
-= num_bytes
;
6046 cache
->space_info
->bytes_reserved
-= num_bytes
;
6048 spin_unlock(&cache
->lock
);
6049 spin_unlock(&cache
->space_info
->lock
);
6051 set_extent_dirty(root
->fs_info
->pinned_extents
, bytenr
,
6052 bytenr
+ num_bytes
- 1, GFP_NOFS
| __GFP_NOFAIL
);
6054 trace_btrfs_reserved_extent_free(root
, bytenr
, num_bytes
);
6059 * this function must be called within transaction
6061 int btrfs_pin_extent(struct btrfs_root
*root
,
6062 u64 bytenr
, u64 num_bytes
, int reserved
)
6064 struct btrfs_block_group_cache
*cache
;
6066 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
6067 BUG_ON(!cache
); /* Logic error */
6069 pin_down_extent(root
, cache
, bytenr
, num_bytes
, reserved
);
6071 btrfs_put_block_group(cache
);
6076 * this function must be called within transaction
6078 int btrfs_pin_extent_for_log_replay(struct btrfs_root
*root
,
6079 u64 bytenr
, u64 num_bytes
)
6081 struct btrfs_block_group_cache
*cache
;
6084 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
6089 * pull in the free space cache (if any) so that our pin
6090 * removes the free space from the cache. We have load_only set
6091 * to one because the slow code to read in the free extents does check
6092 * the pinned extents.
6094 cache_block_group(cache
, 1);
6096 pin_down_extent(root
, cache
, bytenr
, num_bytes
, 0);
6098 /* remove us from the free space cache (if we're there at all) */
6099 ret
= btrfs_remove_free_space(cache
, bytenr
, num_bytes
);
6100 btrfs_put_block_group(cache
);
6104 static int __exclude_logged_extent(struct btrfs_root
*root
, u64 start
, u64 num_bytes
)
6107 struct btrfs_block_group_cache
*block_group
;
6108 struct btrfs_caching_control
*caching_ctl
;
6110 block_group
= btrfs_lookup_block_group(root
->fs_info
, start
);
6114 cache_block_group(block_group
, 0);
6115 caching_ctl
= get_caching_control(block_group
);
6119 BUG_ON(!block_group_cache_done(block_group
));
6120 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
6122 mutex_lock(&caching_ctl
->mutex
);
6124 if (start
>= caching_ctl
->progress
) {
6125 ret
= add_excluded_extent(root
, start
, num_bytes
);
6126 } else if (start
+ num_bytes
<= caching_ctl
->progress
) {
6127 ret
= btrfs_remove_free_space(block_group
,
6130 num_bytes
= caching_ctl
->progress
- start
;
6131 ret
= btrfs_remove_free_space(block_group
,
6136 num_bytes
= (start
+ num_bytes
) -
6137 caching_ctl
->progress
;
6138 start
= caching_ctl
->progress
;
6139 ret
= add_excluded_extent(root
, start
, num_bytes
);
6142 mutex_unlock(&caching_ctl
->mutex
);
6143 put_caching_control(caching_ctl
);
6145 btrfs_put_block_group(block_group
);
6149 int btrfs_exclude_logged_extents(struct btrfs_root
*log
,
6150 struct extent_buffer
*eb
)
6152 struct btrfs_file_extent_item
*item
;
6153 struct btrfs_key key
;
6157 if (!btrfs_fs_incompat(log
->fs_info
, MIXED_GROUPS
))
6160 for (i
= 0; i
< btrfs_header_nritems(eb
); i
++) {
6161 btrfs_item_key_to_cpu(eb
, &key
, i
);
6162 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
6164 item
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
6165 found_type
= btrfs_file_extent_type(eb
, item
);
6166 if (found_type
== BTRFS_FILE_EXTENT_INLINE
)
6168 if (btrfs_file_extent_disk_bytenr(eb
, item
) == 0)
6170 key
.objectid
= btrfs_file_extent_disk_bytenr(eb
, item
);
6171 key
.offset
= btrfs_file_extent_disk_num_bytes(eb
, item
);
6172 __exclude_logged_extent(log
, key
.objectid
, key
.offset
);
6179 * btrfs_update_reserved_bytes - update the block_group and space info counters
6180 * @cache: The cache we are manipulating
6181 * @num_bytes: The number of bytes in question
6182 * @reserve: One of the reservation enums
6183 * @delalloc: The blocks are allocated for the delalloc write
6185 * This is called by the allocator when it reserves space, or by somebody who is
6186 * freeing space that was never actually used on disk. For example if you
6187 * reserve some space for a new leaf in transaction A and before transaction A
6188 * commits you free that leaf, you call this with reserve set to 0 in order to
6189 * clear the reservation.
6191 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
6192 * ENOSPC accounting. For data we handle the reservation through clearing the
6193 * delalloc bits in the io_tree. We have to do this since we could end up
6194 * allocating less disk space for the amount of data we have reserved in the
6195 * case of compression.
6197 * If this is a reservation and the block group has become read only we cannot
6198 * make the reservation and return -EAGAIN, otherwise this function always
6201 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
6202 u64 num_bytes
, int reserve
, int delalloc
)
6204 struct btrfs_space_info
*space_info
= cache
->space_info
;
6207 spin_lock(&space_info
->lock
);
6208 spin_lock(&cache
->lock
);
6209 if (reserve
!= RESERVE_FREE
) {
6213 cache
->reserved
+= num_bytes
;
6214 space_info
->bytes_reserved
+= num_bytes
;
6215 if (reserve
== RESERVE_ALLOC
) {
6216 trace_btrfs_space_reservation(cache
->fs_info
,
6217 "space_info", space_info
->flags
,
6219 space_info
->bytes_may_use
-= num_bytes
;
6223 cache
->delalloc_bytes
+= num_bytes
;
6227 space_info
->bytes_readonly
+= num_bytes
;
6228 cache
->reserved
-= num_bytes
;
6229 space_info
->bytes_reserved
-= num_bytes
;
6232 cache
->delalloc_bytes
-= num_bytes
;
6234 spin_unlock(&cache
->lock
);
6235 spin_unlock(&space_info
->lock
);
6239 void btrfs_prepare_extent_commit(struct btrfs_trans_handle
*trans
,
6240 struct btrfs_root
*root
)
6242 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6243 struct btrfs_caching_control
*next
;
6244 struct btrfs_caching_control
*caching_ctl
;
6245 struct btrfs_block_group_cache
*cache
;
6247 down_write(&fs_info
->commit_root_sem
);
6249 list_for_each_entry_safe(caching_ctl
, next
,
6250 &fs_info
->caching_block_groups
, list
) {
6251 cache
= caching_ctl
->block_group
;
6252 if (block_group_cache_done(cache
)) {
6253 cache
->last_byte_to_unpin
= (u64
)-1;
6254 list_del_init(&caching_ctl
->list
);
6255 put_caching_control(caching_ctl
);
6257 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
6261 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
6262 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
6264 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
6266 up_write(&fs_info
->commit_root_sem
);
6268 update_global_block_rsv(fs_info
);
6272 * Returns the free cluster for the given space info and sets empty_cluster to
6273 * what it should be based on the mount options.
6275 static struct btrfs_free_cluster
*
6276 fetch_cluster_info(struct btrfs_root
*root
, struct btrfs_space_info
*space_info
,
6279 struct btrfs_free_cluster
*ret
= NULL
;
6280 bool ssd
= btrfs_test_opt(root
, SSD
);
6283 if (btrfs_mixed_space_info(space_info
))
6287 *empty_cluster
= SZ_2M
;
6288 if (space_info
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
6289 ret
= &root
->fs_info
->meta_alloc_cluster
;
6291 *empty_cluster
= SZ_64K
;
6292 } else if ((space_info
->flags
& BTRFS_BLOCK_GROUP_DATA
) && ssd
) {
6293 ret
= &root
->fs_info
->data_alloc_cluster
;
6299 static int unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
,
6300 const bool return_free_space
)
6302 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6303 struct btrfs_block_group_cache
*cache
= NULL
;
6304 struct btrfs_space_info
*space_info
;
6305 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
6306 struct btrfs_free_cluster
*cluster
= NULL
;
6308 u64 total_unpinned
= 0;
6309 u64 empty_cluster
= 0;
6312 while (start
<= end
) {
6315 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
6317 btrfs_put_block_group(cache
);
6319 cache
= btrfs_lookup_block_group(fs_info
, start
);
6320 BUG_ON(!cache
); /* Logic error */
6322 cluster
= fetch_cluster_info(root
,
6325 empty_cluster
<<= 1;
6328 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
6329 len
= min(len
, end
+ 1 - start
);
6331 if (start
< cache
->last_byte_to_unpin
) {
6332 len
= min(len
, cache
->last_byte_to_unpin
- start
);
6333 if (return_free_space
)
6334 btrfs_add_free_space(cache
, start
, len
);
6338 total_unpinned
+= len
;
6339 space_info
= cache
->space_info
;
6342 * If this space cluster has been marked as fragmented and we've
6343 * unpinned enough in this block group to potentially allow a
6344 * cluster to be created inside of it go ahead and clear the
6347 if (cluster
&& cluster
->fragmented
&&
6348 total_unpinned
> empty_cluster
) {
6349 spin_lock(&cluster
->lock
);
6350 cluster
->fragmented
= 0;
6351 spin_unlock(&cluster
->lock
);
6354 spin_lock(&space_info
->lock
);
6355 spin_lock(&cache
->lock
);
6356 cache
->pinned
-= len
;
6357 space_info
->bytes_pinned
-= len
;
6358 space_info
->max_extent_size
= 0;
6359 percpu_counter_add(&space_info
->total_bytes_pinned
, -len
);
6361 space_info
->bytes_readonly
+= len
;
6364 spin_unlock(&cache
->lock
);
6365 if (!readonly
&& global_rsv
->space_info
== space_info
) {
6366 spin_lock(&global_rsv
->lock
);
6367 if (!global_rsv
->full
) {
6368 len
= min(len
, global_rsv
->size
-
6369 global_rsv
->reserved
);
6370 global_rsv
->reserved
+= len
;
6371 space_info
->bytes_may_use
+= len
;
6372 if (global_rsv
->reserved
>= global_rsv
->size
)
6373 global_rsv
->full
= 1;
6375 spin_unlock(&global_rsv
->lock
);
6377 spin_unlock(&space_info
->lock
);
6381 btrfs_put_block_group(cache
);
6385 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
6386 struct btrfs_root
*root
)
6388 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6389 struct btrfs_block_group_cache
*block_group
, *tmp
;
6390 struct list_head
*deleted_bgs
;
6391 struct extent_io_tree
*unpin
;
6396 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
6397 unpin
= &fs_info
->freed_extents
[1];
6399 unpin
= &fs_info
->freed_extents
[0];
6401 while (!trans
->aborted
) {
6402 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
6403 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
6404 EXTENT_DIRTY
, NULL
);
6406 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
6410 if (btrfs_test_opt(root
, DISCARD
))
6411 ret
= btrfs_discard_extent(root
, start
,
6412 end
+ 1 - start
, NULL
);
6414 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
6415 unpin_extent_range(root
, start
, end
, true);
6416 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
6421 * Transaction is finished. We don't need the lock anymore. We
6422 * do need to clean up the block groups in case of a transaction
6425 deleted_bgs
= &trans
->transaction
->deleted_bgs
;
6426 list_for_each_entry_safe(block_group
, tmp
, deleted_bgs
, bg_list
) {
6430 if (!trans
->aborted
)
6431 ret
= btrfs_discard_extent(root
,
6432 block_group
->key
.objectid
,
6433 block_group
->key
.offset
,
6436 list_del_init(&block_group
->bg_list
);
6437 btrfs_put_block_group_trimming(block_group
);
6438 btrfs_put_block_group(block_group
);
6441 const char *errstr
= btrfs_decode_error(ret
);
6443 "Discard failed while removing blockgroup: errno=%d %s\n",
6451 static void add_pinned_bytes(struct btrfs_fs_info
*fs_info
, u64 num_bytes
,
6452 u64 owner
, u64 root_objectid
)
6454 struct btrfs_space_info
*space_info
;
6457 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
6458 if (root_objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
6459 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
6461 flags
= BTRFS_BLOCK_GROUP_METADATA
;
6463 flags
= BTRFS_BLOCK_GROUP_DATA
;
6466 space_info
= __find_space_info(fs_info
, flags
);
6467 BUG_ON(!space_info
); /* Logic bug */
6468 percpu_counter_add(&space_info
->total_bytes_pinned
, num_bytes
);
6472 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
6473 struct btrfs_root
*root
,
6474 struct btrfs_delayed_ref_node
*node
, u64 parent
,
6475 u64 root_objectid
, u64 owner_objectid
,
6476 u64 owner_offset
, int refs_to_drop
,
6477 struct btrfs_delayed_extent_op
*extent_op
)
6479 struct btrfs_key key
;
6480 struct btrfs_path
*path
;
6481 struct btrfs_fs_info
*info
= root
->fs_info
;
6482 struct btrfs_root
*extent_root
= info
->extent_root
;
6483 struct extent_buffer
*leaf
;
6484 struct btrfs_extent_item
*ei
;
6485 struct btrfs_extent_inline_ref
*iref
;
6488 int extent_slot
= 0;
6489 int found_extent
= 0;
6493 u64 bytenr
= node
->bytenr
;
6494 u64 num_bytes
= node
->num_bytes
;
6496 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
6499 path
= btrfs_alloc_path();
6503 path
->reada
= READA_FORWARD
;
6504 path
->leave_spinning
= 1;
6506 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
6507 BUG_ON(!is_data
&& refs_to_drop
!= 1);
6510 skinny_metadata
= 0;
6512 ret
= lookup_extent_backref(trans
, extent_root
, path
, &iref
,
6513 bytenr
, num_bytes
, parent
,
6514 root_objectid
, owner_objectid
,
6517 extent_slot
= path
->slots
[0];
6518 while (extent_slot
>= 0) {
6519 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
6521 if (key
.objectid
!= bytenr
)
6523 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
6524 key
.offset
== num_bytes
) {
6528 if (key
.type
== BTRFS_METADATA_ITEM_KEY
&&
6529 key
.offset
== owner_objectid
) {
6533 if (path
->slots
[0] - extent_slot
> 5)
6537 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6538 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
6539 if (found_extent
&& item_size
< sizeof(*ei
))
6542 if (!found_extent
) {
6544 ret
= remove_extent_backref(trans
, extent_root
, path
,
6546 is_data
, &last_ref
);
6548 btrfs_abort_transaction(trans
, extent_root
, ret
);
6551 btrfs_release_path(path
);
6552 path
->leave_spinning
= 1;
6554 key
.objectid
= bytenr
;
6555 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6556 key
.offset
= num_bytes
;
6558 if (!is_data
&& skinny_metadata
) {
6559 key
.type
= BTRFS_METADATA_ITEM_KEY
;
6560 key
.offset
= owner_objectid
;
6563 ret
= btrfs_search_slot(trans
, extent_root
,
6565 if (ret
> 0 && skinny_metadata
&& path
->slots
[0]) {
6567 * Couldn't find our skinny metadata item,
6568 * see if we have ye olde extent item.
6571 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
6573 if (key
.objectid
== bytenr
&&
6574 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
6575 key
.offset
== num_bytes
)
6579 if (ret
> 0 && skinny_metadata
) {
6580 skinny_metadata
= false;
6581 key
.objectid
= bytenr
;
6582 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6583 key
.offset
= num_bytes
;
6584 btrfs_release_path(path
);
6585 ret
= btrfs_search_slot(trans
, extent_root
,
6590 btrfs_err(info
, "umm, got %d back from search, was looking for %llu",
6593 btrfs_print_leaf(extent_root
,
6597 btrfs_abort_transaction(trans
, extent_root
, ret
);
6600 extent_slot
= path
->slots
[0];
6602 } else if (WARN_ON(ret
== -ENOENT
)) {
6603 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
6605 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6606 bytenr
, parent
, root_objectid
, owner_objectid
,
6608 btrfs_abort_transaction(trans
, extent_root
, ret
);
6611 btrfs_abort_transaction(trans
, extent_root
, ret
);
6615 leaf
= path
->nodes
[0];
6616 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
6617 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6618 if (item_size
< sizeof(*ei
)) {
6619 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
6620 ret
= convert_extent_item_v0(trans
, extent_root
, path
,
6623 btrfs_abort_transaction(trans
, extent_root
, ret
);
6627 btrfs_release_path(path
);
6628 path
->leave_spinning
= 1;
6630 key
.objectid
= bytenr
;
6631 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6632 key
.offset
= num_bytes
;
6634 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
6637 btrfs_err(info
, "umm, got %d back from search, was looking for %llu",
6639 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
6642 btrfs_abort_transaction(trans
, extent_root
, ret
);
6646 extent_slot
= path
->slots
[0];
6647 leaf
= path
->nodes
[0];
6648 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
6651 BUG_ON(item_size
< sizeof(*ei
));
6652 ei
= btrfs_item_ptr(leaf
, extent_slot
,
6653 struct btrfs_extent_item
);
6654 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
6655 key
.type
== BTRFS_EXTENT_ITEM_KEY
) {
6656 struct btrfs_tree_block_info
*bi
;
6657 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
6658 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
6659 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
6662 refs
= btrfs_extent_refs(leaf
, ei
);
6663 if (refs
< refs_to_drop
) {
6664 btrfs_err(info
, "trying to drop %d refs but we only have %Lu "
6665 "for bytenr %Lu", refs_to_drop
, refs
, bytenr
);
6667 btrfs_abort_transaction(trans
, extent_root
, ret
);
6670 refs
-= refs_to_drop
;
6674 __run_delayed_extent_op(extent_op
, leaf
, ei
);
6676 * In the case of inline back ref, reference count will
6677 * be updated by remove_extent_backref
6680 BUG_ON(!found_extent
);
6682 btrfs_set_extent_refs(leaf
, ei
, refs
);
6683 btrfs_mark_buffer_dirty(leaf
);
6686 ret
= remove_extent_backref(trans
, extent_root
, path
,
6688 is_data
, &last_ref
);
6690 btrfs_abort_transaction(trans
, extent_root
, ret
);
6694 add_pinned_bytes(root
->fs_info
, -num_bytes
, owner_objectid
,
6698 BUG_ON(is_data
&& refs_to_drop
!=
6699 extent_data_ref_count(path
, iref
));
6701 BUG_ON(path
->slots
[0] != extent_slot
);
6703 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
6704 path
->slots
[0] = extent_slot
;
6710 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
6713 btrfs_abort_transaction(trans
, extent_root
, ret
);
6716 btrfs_release_path(path
);
6719 ret
= btrfs_del_csums(trans
, root
, bytenr
, num_bytes
);
6721 btrfs_abort_transaction(trans
, extent_root
, ret
);
6726 ret
= add_to_free_space_tree(trans
, root
->fs_info
, bytenr
,
6729 btrfs_abort_transaction(trans
, extent_root
, ret
);
6733 ret
= update_block_group(trans
, root
, bytenr
, num_bytes
, 0);
6735 btrfs_abort_transaction(trans
, extent_root
, ret
);
6739 btrfs_release_path(path
);
6742 btrfs_free_path(path
);
6747 * when we free an block, it is possible (and likely) that we free the last
6748 * delayed ref for that extent as well. This searches the delayed ref tree for
6749 * a given extent, and if there are no other delayed refs to be processed, it
6750 * removes it from the tree.
6752 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
6753 struct btrfs_root
*root
, u64 bytenr
)
6755 struct btrfs_delayed_ref_head
*head
;
6756 struct btrfs_delayed_ref_root
*delayed_refs
;
6759 delayed_refs
= &trans
->transaction
->delayed_refs
;
6760 spin_lock(&delayed_refs
->lock
);
6761 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
6763 goto out_delayed_unlock
;
6765 spin_lock(&head
->lock
);
6766 if (!list_empty(&head
->ref_list
))
6769 if (head
->extent_op
) {
6770 if (!head
->must_insert_reserved
)
6772 btrfs_free_delayed_extent_op(head
->extent_op
);
6773 head
->extent_op
= NULL
;
6777 * waiting for the lock here would deadlock. If someone else has it
6778 * locked they are already in the process of dropping it anyway
6780 if (!mutex_trylock(&head
->mutex
))
6784 * at this point we have a head with no other entries. Go
6785 * ahead and process it.
6787 head
->node
.in_tree
= 0;
6788 rb_erase(&head
->href_node
, &delayed_refs
->href_root
);
6790 atomic_dec(&delayed_refs
->num_entries
);
6793 * we don't take a ref on the node because we're removing it from the
6794 * tree, so we just steal the ref the tree was holding.
6796 delayed_refs
->num_heads
--;
6797 if (head
->processing
== 0)
6798 delayed_refs
->num_heads_ready
--;
6799 head
->processing
= 0;
6800 spin_unlock(&head
->lock
);
6801 spin_unlock(&delayed_refs
->lock
);
6803 BUG_ON(head
->extent_op
);
6804 if (head
->must_insert_reserved
)
6807 mutex_unlock(&head
->mutex
);
6808 btrfs_put_delayed_ref(&head
->node
);
6811 spin_unlock(&head
->lock
);
6814 spin_unlock(&delayed_refs
->lock
);
6818 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
6819 struct btrfs_root
*root
,
6820 struct extent_buffer
*buf
,
6821 u64 parent
, int last_ref
)
6826 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
6827 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
6828 buf
->start
, buf
->len
,
6829 parent
, root
->root_key
.objectid
,
6830 btrfs_header_level(buf
),
6831 BTRFS_DROP_DELAYED_REF
, NULL
);
6832 BUG_ON(ret
); /* -ENOMEM */
6838 if (btrfs_header_generation(buf
) == trans
->transid
) {
6839 struct btrfs_block_group_cache
*cache
;
6841 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
6842 ret
= check_ref_cleanup(trans
, root
, buf
->start
);
6847 cache
= btrfs_lookup_block_group(root
->fs_info
, buf
->start
);
6849 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
6850 pin_down_extent(root
, cache
, buf
->start
, buf
->len
, 1);
6851 btrfs_put_block_group(cache
);
6855 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
6857 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
6858 btrfs_update_reserved_bytes(cache
, buf
->len
, RESERVE_FREE
, 0);
6859 btrfs_put_block_group(cache
);
6860 trace_btrfs_reserved_extent_free(root
, buf
->start
, buf
->len
);
6865 add_pinned_bytes(root
->fs_info
, buf
->len
,
6866 btrfs_header_level(buf
),
6867 root
->root_key
.objectid
);
6870 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6873 clear_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
);
6876 /* Can return -ENOMEM */
6877 int btrfs_free_extent(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
6878 u64 bytenr
, u64 num_bytes
, u64 parent
, u64 root_objectid
,
6879 u64 owner
, u64 offset
)
6882 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6884 if (btrfs_test_is_dummy_root(root
))
6887 add_pinned_bytes(root
->fs_info
, num_bytes
, owner
, root_objectid
);
6890 * tree log blocks never actually go into the extent allocation
6891 * tree, just update pinning info and exit early.
6893 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
6894 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
6895 /* unlocks the pinned mutex */
6896 btrfs_pin_extent(root
, bytenr
, num_bytes
, 1);
6898 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
6899 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
6901 parent
, root_objectid
, (int)owner
,
6902 BTRFS_DROP_DELAYED_REF
, NULL
);
6904 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
6906 parent
, root_objectid
, owner
,
6908 BTRFS_DROP_DELAYED_REF
, NULL
);
6914 * when we wait for progress in the block group caching, its because
6915 * our allocation attempt failed at least once. So, we must sleep
6916 * and let some progress happen before we try again.
6918 * This function will sleep at least once waiting for new free space to
6919 * show up, and then it will check the block group free space numbers
6920 * for our min num_bytes. Another option is to have it go ahead
6921 * and look in the rbtree for a free extent of a given size, but this
6924 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6925 * any of the information in this block group.
6927 static noinline
void
6928 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
6931 struct btrfs_caching_control
*caching_ctl
;
6933 caching_ctl
= get_caching_control(cache
);
6937 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
6938 (cache
->free_space_ctl
->free_space
>= num_bytes
));
6940 put_caching_control(caching_ctl
);
6944 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
6946 struct btrfs_caching_control
*caching_ctl
;
6949 caching_ctl
= get_caching_control(cache
);
6951 return (cache
->cached
== BTRFS_CACHE_ERROR
) ? -EIO
: 0;
6953 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
6954 if (cache
->cached
== BTRFS_CACHE_ERROR
)
6956 put_caching_control(caching_ctl
);
6960 int __get_raid_index(u64 flags
)
6962 if (flags
& BTRFS_BLOCK_GROUP_RAID10
)
6963 return BTRFS_RAID_RAID10
;
6964 else if (flags
& BTRFS_BLOCK_GROUP_RAID1
)
6965 return BTRFS_RAID_RAID1
;
6966 else if (flags
& BTRFS_BLOCK_GROUP_DUP
)
6967 return BTRFS_RAID_DUP
;
6968 else if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
6969 return BTRFS_RAID_RAID0
;
6970 else if (flags
& BTRFS_BLOCK_GROUP_RAID5
)
6971 return BTRFS_RAID_RAID5
;
6972 else if (flags
& BTRFS_BLOCK_GROUP_RAID6
)
6973 return BTRFS_RAID_RAID6
;
6975 return BTRFS_RAID_SINGLE
; /* BTRFS_BLOCK_GROUP_SINGLE */
6978 int get_block_group_index(struct btrfs_block_group_cache
*cache
)
6980 return __get_raid_index(cache
->flags
);
6983 static const char *btrfs_raid_type_names
[BTRFS_NR_RAID_TYPES
] = {
6984 [BTRFS_RAID_RAID10
] = "raid10",
6985 [BTRFS_RAID_RAID1
] = "raid1",
6986 [BTRFS_RAID_DUP
] = "dup",
6987 [BTRFS_RAID_RAID0
] = "raid0",
6988 [BTRFS_RAID_SINGLE
] = "single",
6989 [BTRFS_RAID_RAID5
] = "raid5",
6990 [BTRFS_RAID_RAID6
] = "raid6",
6993 static const char *get_raid_name(enum btrfs_raid_types type
)
6995 if (type
>= BTRFS_NR_RAID_TYPES
)
6998 return btrfs_raid_type_names
[type
];
7001 enum btrfs_loop_type
{
7002 LOOP_CACHING_NOWAIT
= 0,
7003 LOOP_CACHING_WAIT
= 1,
7004 LOOP_ALLOC_CHUNK
= 2,
7005 LOOP_NO_EMPTY_SIZE
= 3,
7009 btrfs_lock_block_group(struct btrfs_block_group_cache
*cache
,
7013 down_read(&cache
->data_rwsem
);
7017 btrfs_grab_block_group(struct btrfs_block_group_cache
*cache
,
7020 btrfs_get_block_group(cache
);
7022 down_read(&cache
->data_rwsem
);
7025 static struct btrfs_block_group_cache
*
7026 btrfs_lock_cluster(struct btrfs_block_group_cache
*block_group
,
7027 struct btrfs_free_cluster
*cluster
,
7030 struct btrfs_block_group_cache
*used_bg
= NULL
;
7031 bool locked
= false;
7033 spin_lock(&cluster
->refill_lock
);
7035 if (used_bg
== cluster
->block_group
)
7038 up_read(&used_bg
->data_rwsem
);
7039 btrfs_put_block_group(used_bg
);
7042 used_bg
= cluster
->block_group
;
7046 if (used_bg
== block_group
)
7049 btrfs_get_block_group(used_bg
);
7054 if (down_read_trylock(&used_bg
->data_rwsem
))
7057 spin_unlock(&cluster
->refill_lock
);
7058 down_read(&used_bg
->data_rwsem
);
7064 btrfs_release_block_group(struct btrfs_block_group_cache
*cache
,
7068 up_read(&cache
->data_rwsem
);
7069 btrfs_put_block_group(cache
);
7073 * walks the btree of allocated extents and find a hole of a given size.
7074 * The key ins is changed to record the hole:
7075 * ins->objectid == start position
7076 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7077 * ins->offset == the size of the hole.
7078 * Any available blocks before search_start are skipped.
7080 * If there is no suitable free space, we will record the max size of
7081 * the free space extent currently.
7083 static noinline
int find_free_extent(struct btrfs_root
*orig_root
,
7084 u64 num_bytes
, u64 empty_size
,
7085 u64 hint_byte
, struct btrfs_key
*ins
,
7086 u64 flags
, int delalloc
)
7089 struct btrfs_root
*root
= orig_root
->fs_info
->extent_root
;
7090 struct btrfs_free_cluster
*last_ptr
= NULL
;
7091 struct btrfs_block_group_cache
*block_group
= NULL
;
7092 u64 search_start
= 0;
7093 u64 max_extent_size
= 0;
7094 u64 empty_cluster
= 0;
7095 struct btrfs_space_info
*space_info
;
7097 int index
= __get_raid_index(flags
);
7098 int alloc_type
= (flags
& BTRFS_BLOCK_GROUP_DATA
) ?
7099 RESERVE_ALLOC_NO_ACCOUNT
: RESERVE_ALLOC
;
7100 bool failed_cluster_refill
= false;
7101 bool failed_alloc
= false;
7102 bool use_cluster
= true;
7103 bool have_caching_bg
= false;
7104 bool orig_have_caching_bg
= false;
7105 bool full_search
= false;
7107 WARN_ON(num_bytes
< root
->sectorsize
);
7108 ins
->type
= BTRFS_EXTENT_ITEM_KEY
;
7112 trace_find_free_extent(orig_root
, num_bytes
, empty_size
, flags
);
7114 space_info
= __find_space_info(root
->fs_info
, flags
);
7116 btrfs_err(root
->fs_info
, "No space info for %llu", flags
);
7121 * If our free space is heavily fragmented we may not be able to make
7122 * big contiguous allocations, so instead of doing the expensive search
7123 * for free space, simply return ENOSPC with our max_extent_size so we
7124 * can go ahead and search for a more manageable chunk.
7126 * If our max_extent_size is large enough for our allocation simply
7127 * disable clustering since we will likely not be able to find enough
7128 * space to create a cluster and induce latency trying.
7130 if (unlikely(space_info
->max_extent_size
)) {
7131 spin_lock(&space_info
->lock
);
7132 if (space_info
->max_extent_size
&&
7133 num_bytes
> space_info
->max_extent_size
) {
7134 ins
->offset
= space_info
->max_extent_size
;
7135 spin_unlock(&space_info
->lock
);
7137 } else if (space_info
->max_extent_size
) {
7138 use_cluster
= false;
7140 spin_unlock(&space_info
->lock
);
7143 last_ptr
= fetch_cluster_info(orig_root
, space_info
, &empty_cluster
);
7145 spin_lock(&last_ptr
->lock
);
7146 if (last_ptr
->block_group
)
7147 hint_byte
= last_ptr
->window_start
;
7148 if (last_ptr
->fragmented
) {
7150 * We still set window_start so we can keep track of the
7151 * last place we found an allocation to try and save
7154 hint_byte
= last_ptr
->window_start
;
7155 use_cluster
= false;
7157 spin_unlock(&last_ptr
->lock
);
7160 search_start
= max(search_start
, first_logical_byte(root
, 0));
7161 search_start
= max(search_start
, hint_byte
);
7162 if (search_start
== hint_byte
) {
7163 block_group
= btrfs_lookup_block_group(root
->fs_info
,
7166 * we don't want to use the block group if it doesn't match our
7167 * allocation bits, or if its not cached.
7169 * However if we are re-searching with an ideal block group
7170 * picked out then we don't care that the block group is cached.
7172 if (block_group
&& block_group_bits(block_group
, flags
) &&
7173 block_group
->cached
!= BTRFS_CACHE_NO
) {
7174 down_read(&space_info
->groups_sem
);
7175 if (list_empty(&block_group
->list
) ||
7178 * someone is removing this block group,
7179 * we can't jump into the have_block_group
7180 * target because our list pointers are not
7183 btrfs_put_block_group(block_group
);
7184 up_read(&space_info
->groups_sem
);
7186 index
= get_block_group_index(block_group
);
7187 btrfs_lock_block_group(block_group
, delalloc
);
7188 goto have_block_group
;
7190 } else if (block_group
) {
7191 btrfs_put_block_group(block_group
);
7195 have_caching_bg
= false;
7196 if (index
== 0 || index
== __get_raid_index(flags
))
7198 down_read(&space_info
->groups_sem
);
7199 list_for_each_entry(block_group
, &space_info
->block_groups
[index
],
7204 btrfs_grab_block_group(block_group
, delalloc
);
7205 search_start
= block_group
->key
.objectid
;
7208 * this can happen if we end up cycling through all the
7209 * raid types, but we want to make sure we only allocate
7210 * for the proper type.
7212 if (!block_group_bits(block_group
, flags
)) {
7213 u64 extra
= BTRFS_BLOCK_GROUP_DUP
|
7214 BTRFS_BLOCK_GROUP_RAID1
|
7215 BTRFS_BLOCK_GROUP_RAID5
|
7216 BTRFS_BLOCK_GROUP_RAID6
|
7217 BTRFS_BLOCK_GROUP_RAID10
;
7220 * if they asked for extra copies and this block group
7221 * doesn't provide them, bail. This does allow us to
7222 * fill raid0 from raid1.
7224 if ((flags
& extra
) && !(block_group
->flags
& extra
))
7229 cached
= block_group_cache_done(block_group
);
7230 if (unlikely(!cached
)) {
7231 have_caching_bg
= true;
7232 ret
= cache_block_group(block_group
, 0);
7237 if (unlikely(block_group
->cached
== BTRFS_CACHE_ERROR
))
7239 if (unlikely(block_group
->ro
))
7243 * Ok we want to try and use the cluster allocator, so
7246 if (last_ptr
&& use_cluster
) {
7247 struct btrfs_block_group_cache
*used_block_group
;
7248 unsigned long aligned_cluster
;
7250 * the refill lock keeps out other
7251 * people trying to start a new cluster
7253 used_block_group
= btrfs_lock_cluster(block_group
,
7256 if (!used_block_group
)
7257 goto refill_cluster
;
7259 if (used_block_group
!= block_group
&&
7260 (used_block_group
->ro
||
7261 !block_group_bits(used_block_group
, flags
)))
7262 goto release_cluster
;
7264 offset
= btrfs_alloc_from_cluster(used_block_group
,
7267 used_block_group
->key
.objectid
,
7270 /* we have a block, we're done */
7271 spin_unlock(&last_ptr
->refill_lock
);
7272 trace_btrfs_reserve_extent_cluster(root
,
7274 search_start
, num_bytes
);
7275 if (used_block_group
!= block_group
) {
7276 btrfs_release_block_group(block_group
,
7278 block_group
= used_block_group
;
7283 WARN_ON(last_ptr
->block_group
!= used_block_group
);
7285 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7286 * set up a new clusters, so lets just skip it
7287 * and let the allocator find whatever block
7288 * it can find. If we reach this point, we
7289 * will have tried the cluster allocator
7290 * plenty of times and not have found
7291 * anything, so we are likely way too
7292 * fragmented for the clustering stuff to find
7295 * However, if the cluster is taken from the
7296 * current block group, release the cluster
7297 * first, so that we stand a better chance of
7298 * succeeding in the unclustered
7300 if (loop
>= LOOP_NO_EMPTY_SIZE
&&
7301 used_block_group
!= block_group
) {
7302 spin_unlock(&last_ptr
->refill_lock
);
7303 btrfs_release_block_group(used_block_group
,
7305 goto unclustered_alloc
;
7309 * this cluster didn't work out, free it and
7312 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
7314 if (used_block_group
!= block_group
)
7315 btrfs_release_block_group(used_block_group
,
7318 if (loop
>= LOOP_NO_EMPTY_SIZE
) {
7319 spin_unlock(&last_ptr
->refill_lock
);
7320 goto unclustered_alloc
;
7323 aligned_cluster
= max_t(unsigned long,
7324 empty_cluster
+ empty_size
,
7325 block_group
->full_stripe_len
);
7327 /* allocate a cluster in this block group */
7328 ret
= btrfs_find_space_cluster(root
, block_group
,
7329 last_ptr
, search_start
,
7334 * now pull our allocation out of this
7337 offset
= btrfs_alloc_from_cluster(block_group
,
7343 /* we found one, proceed */
7344 spin_unlock(&last_ptr
->refill_lock
);
7345 trace_btrfs_reserve_extent_cluster(root
,
7346 block_group
, search_start
,
7350 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
7351 && !failed_cluster_refill
) {
7352 spin_unlock(&last_ptr
->refill_lock
);
7354 failed_cluster_refill
= true;
7355 wait_block_group_cache_progress(block_group
,
7356 num_bytes
+ empty_cluster
+ empty_size
);
7357 goto have_block_group
;
7361 * at this point we either didn't find a cluster
7362 * or we weren't able to allocate a block from our
7363 * cluster. Free the cluster we've been trying
7364 * to use, and go to the next block group
7366 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
7367 spin_unlock(&last_ptr
->refill_lock
);
7373 * We are doing an unclustered alloc, set the fragmented flag so
7374 * we don't bother trying to setup a cluster again until we get
7377 if (unlikely(last_ptr
)) {
7378 spin_lock(&last_ptr
->lock
);
7379 last_ptr
->fragmented
= 1;
7380 spin_unlock(&last_ptr
->lock
);
7382 spin_lock(&block_group
->free_space_ctl
->tree_lock
);
7384 block_group
->free_space_ctl
->free_space
<
7385 num_bytes
+ empty_cluster
+ empty_size
) {
7386 if (block_group
->free_space_ctl
->free_space
>
7389 block_group
->free_space_ctl
->free_space
;
7390 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
7393 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
7395 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
7396 num_bytes
, empty_size
,
7399 * If we didn't find a chunk, and we haven't failed on this
7400 * block group before, and this block group is in the middle of
7401 * caching and we are ok with waiting, then go ahead and wait
7402 * for progress to be made, and set failed_alloc to true.
7404 * If failed_alloc is true then we've already waited on this
7405 * block group once and should move on to the next block group.
7407 if (!offset
&& !failed_alloc
&& !cached
&&
7408 loop
> LOOP_CACHING_NOWAIT
) {
7409 wait_block_group_cache_progress(block_group
,
7410 num_bytes
+ empty_size
);
7411 failed_alloc
= true;
7412 goto have_block_group
;
7413 } else if (!offset
) {
7417 search_start
= ALIGN(offset
, root
->stripesize
);
7419 /* move on to the next group */
7420 if (search_start
+ num_bytes
>
7421 block_group
->key
.objectid
+ block_group
->key
.offset
) {
7422 btrfs_add_free_space(block_group
, offset
, num_bytes
);
7426 if (offset
< search_start
)
7427 btrfs_add_free_space(block_group
, offset
,
7428 search_start
- offset
);
7429 BUG_ON(offset
> search_start
);
7431 ret
= btrfs_update_reserved_bytes(block_group
, num_bytes
,
7432 alloc_type
, delalloc
);
7433 if (ret
== -EAGAIN
) {
7434 btrfs_add_free_space(block_group
, offset
, num_bytes
);
7438 /* we are all good, lets return */
7439 ins
->objectid
= search_start
;
7440 ins
->offset
= num_bytes
;
7442 trace_btrfs_reserve_extent(orig_root
, block_group
,
7443 search_start
, num_bytes
);
7444 btrfs_release_block_group(block_group
, delalloc
);
7447 failed_cluster_refill
= false;
7448 failed_alloc
= false;
7449 BUG_ON(index
!= get_block_group_index(block_group
));
7450 btrfs_release_block_group(block_group
, delalloc
);
7452 up_read(&space_info
->groups_sem
);
7454 if ((loop
== LOOP_CACHING_NOWAIT
) && have_caching_bg
7455 && !orig_have_caching_bg
)
7456 orig_have_caching_bg
= true;
7458 if (!ins
->objectid
&& loop
>= LOOP_CACHING_WAIT
&& have_caching_bg
)
7461 if (!ins
->objectid
&& ++index
< BTRFS_NR_RAID_TYPES
)
7465 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7466 * caching kthreads as we move along
7467 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7468 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7469 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7472 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
) {
7474 if (loop
== LOOP_CACHING_NOWAIT
) {
7476 * We want to skip the LOOP_CACHING_WAIT step if we
7477 * don't have any unached bgs and we've alrelady done a
7478 * full search through.
7480 if (orig_have_caching_bg
|| !full_search
)
7481 loop
= LOOP_CACHING_WAIT
;
7483 loop
= LOOP_ALLOC_CHUNK
;
7488 if (loop
== LOOP_ALLOC_CHUNK
) {
7489 struct btrfs_trans_handle
*trans
;
7492 trans
= current
->journal_info
;
7496 trans
= btrfs_join_transaction(root
);
7498 if (IS_ERR(trans
)) {
7499 ret
= PTR_ERR(trans
);
7503 ret
= do_chunk_alloc(trans
, root
, flags
,
7507 * If we can't allocate a new chunk we've already looped
7508 * through at least once, move on to the NO_EMPTY_SIZE
7512 loop
= LOOP_NO_EMPTY_SIZE
;
7515 * Do not bail out on ENOSPC since we
7516 * can do more things.
7518 if (ret
< 0 && ret
!= -ENOSPC
)
7519 btrfs_abort_transaction(trans
,
7524 btrfs_end_transaction(trans
, root
);
7529 if (loop
== LOOP_NO_EMPTY_SIZE
) {
7531 * Don't loop again if we already have no empty_size and
7534 if (empty_size
== 0 &&
7535 empty_cluster
== 0) {
7544 } else if (!ins
->objectid
) {
7546 } else if (ins
->objectid
) {
7547 if (!use_cluster
&& last_ptr
) {
7548 spin_lock(&last_ptr
->lock
);
7549 last_ptr
->window_start
= ins
->objectid
;
7550 spin_unlock(&last_ptr
->lock
);
7555 if (ret
== -ENOSPC
) {
7556 spin_lock(&space_info
->lock
);
7557 space_info
->max_extent_size
= max_extent_size
;
7558 spin_unlock(&space_info
->lock
);
7559 ins
->offset
= max_extent_size
;
7564 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
7565 int dump_block_groups
)
7567 struct btrfs_block_group_cache
*cache
;
7570 spin_lock(&info
->lock
);
7571 printk(KERN_INFO
"BTRFS: space_info %llu has %llu free, is %sfull\n",
7573 info
->total_bytes
- info
->bytes_used
- info
->bytes_pinned
-
7574 info
->bytes_reserved
- info
->bytes_readonly
,
7575 (info
->full
) ? "" : "not ");
7576 printk(KERN_INFO
"BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7577 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7578 info
->total_bytes
, info
->bytes_used
, info
->bytes_pinned
,
7579 info
->bytes_reserved
, info
->bytes_may_use
,
7580 info
->bytes_readonly
);
7581 spin_unlock(&info
->lock
);
7583 if (!dump_block_groups
)
7586 down_read(&info
->groups_sem
);
7588 list_for_each_entry(cache
, &info
->block_groups
[index
], list
) {
7589 spin_lock(&cache
->lock
);
7590 printk(KERN_INFO
"BTRFS: "
7591 "block group %llu has %llu bytes, "
7592 "%llu used %llu pinned %llu reserved %s\n",
7593 cache
->key
.objectid
, cache
->key
.offset
,
7594 btrfs_block_group_used(&cache
->item
), cache
->pinned
,
7595 cache
->reserved
, cache
->ro
? "[readonly]" : "");
7596 btrfs_dump_free_space(cache
, bytes
);
7597 spin_unlock(&cache
->lock
);
7599 if (++index
< BTRFS_NR_RAID_TYPES
)
7601 up_read(&info
->groups_sem
);
7604 int btrfs_reserve_extent(struct btrfs_root
*root
,
7605 u64 num_bytes
, u64 min_alloc_size
,
7606 u64 empty_size
, u64 hint_byte
,
7607 struct btrfs_key
*ins
, int is_data
, int delalloc
)
7609 bool final_tried
= num_bytes
== min_alloc_size
;
7613 flags
= btrfs_get_alloc_profile(root
, is_data
);
7615 WARN_ON(num_bytes
< root
->sectorsize
);
7616 ret
= find_free_extent(root
, num_bytes
, empty_size
, hint_byte
, ins
,
7619 if (ret
== -ENOSPC
) {
7620 if (!final_tried
&& ins
->offset
) {
7621 num_bytes
= min(num_bytes
>> 1, ins
->offset
);
7622 num_bytes
= round_down(num_bytes
, root
->sectorsize
);
7623 num_bytes
= max(num_bytes
, min_alloc_size
);
7624 if (num_bytes
== min_alloc_size
)
7627 } else if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
7628 struct btrfs_space_info
*sinfo
;
7630 sinfo
= __find_space_info(root
->fs_info
, flags
);
7631 btrfs_err(root
->fs_info
, "allocation failed flags %llu, wanted %llu",
7634 dump_space_info(sinfo
, num_bytes
, 1);
7641 static int __btrfs_free_reserved_extent(struct btrfs_root
*root
,
7643 int pin
, int delalloc
)
7645 struct btrfs_block_group_cache
*cache
;
7648 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
7650 btrfs_err(root
->fs_info
, "Unable to find block group for %llu",
7656 pin_down_extent(root
, cache
, start
, len
, 1);
7658 if (btrfs_test_opt(root
, DISCARD
))
7659 ret
= btrfs_discard_extent(root
, start
, len
, NULL
);
7660 btrfs_add_free_space(cache
, start
, len
);
7661 btrfs_update_reserved_bytes(cache
, len
, RESERVE_FREE
, delalloc
);
7664 btrfs_put_block_group(cache
);
7666 trace_btrfs_reserved_extent_free(root
, start
, len
);
7671 int btrfs_free_reserved_extent(struct btrfs_root
*root
,
7672 u64 start
, u64 len
, int delalloc
)
7674 return __btrfs_free_reserved_extent(root
, start
, len
, 0, delalloc
);
7677 int btrfs_free_and_pin_reserved_extent(struct btrfs_root
*root
,
7680 return __btrfs_free_reserved_extent(root
, start
, len
, 1, 0);
7683 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
7684 struct btrfs_root
*root
,
7685 u64 parent
, u64 root_objectid
,
7686 u64 flags
, u64 owner
, u64 offset
,
7687 struct btrfs_key
*ins
, int ref_mod
)
7690 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7691 struct btrfs_extent_item
*extent_item
;
7692 struct btrfs_extent_inline_ref
*iref
;
7693 struct btrfs_path
*path
;
7694 struct extent_buffer
*leaf
;
7699 type
= BTRFS_SHARED_DATA_REF_KEY
;
7701 type
= BTRFS_EXTENT_DATA_REF_KEY
;
7703 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
7705 path
= btrfs_alloc_path();
7709 path
->leave_spinning
= 1;
7710 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
7713 btrfs_free_path(path
);
7717 leaf
= path
->nodes
[0];
7718 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
7719 struct btrfs_extent_item
);
7720 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
7721 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
7722 btrfs_set_extent_flags(leaf
, extent_item
,
7723 flags
| BTRFS_EXTENT_FLAG_DATA
);
7725 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
7726 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
7728 struct btrfs_shared_data_ref
*ref
;
7729 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
7730 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
7731 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
7733 struct btrfs_extent_data_ref
*ref
;
7734 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
7735 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
7736 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
7737 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
7738 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
7741 btrfs_mark_buffer_dirty(path
->nodes
[0]);
7742 btrfs_free_path(path
);
7744 ret
= remove_from_free_space_tree(trans
, fs_info
, ins
->objectid
,
7749 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1);
7750 if (ret
) { /* -ENOENT, logic error */
7751 btrfs_err(fs_info
, "update block group failed for %llu %llu",
7752 ins
->objectid
, ins
->offset
);
7755 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, ins
->offset
);
7759 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
7760 struct btrfs_root
*root
,
7761 u64 parent
, u64 root_objectid
,
7762 u64 flags
, struct btrfs_disk_key
*key
,
7763 int level
, struct btrfs_key
*ins
)
7766 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7767 struct btrfs_extent_item
*extent_item
;
7768 struct btrfs_tree_block_info
*block_info
;
7769 struct btrfs_extent_inline_ref
*iref
;
7770 struct btrfs_path
*path
;
7771 struct extent_buffer
*leaf
;
7772 u32 size
= sizeof(*extent_item
) + sizeof(*iref
);
7773 u64 num_bytes
= ins
->offset
;
7774 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
7777 if (!skinny_metadata
)
7778 size
+= sizeof(*block_info
);
7780 path
= btrfs_alloc_path();
7782 btrfs_free_and_pin_reserved_extent(root
, ins
->objectid
,
7787 path
->leave_spinning
= 1;
7788 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
7791 btrfs_free_path(path
);
7792 btrfs_free_and_pin_reserved_extent(root
, ins
->objectid
,
7797 leaf
= path
->nodes
[0];
7798 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
7799 struct btrfs_extent_item
);
7800 btrfs_set_extent_refs(leaf
, extent_item
, 1);
7801 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
7802 btrfs_set_extent_flags(leaf
, extent_item
,
7803 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
7805 if (skinny_metadata
) {
7806 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
7807 num_bytes
= root
->nodesize
;
7809 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
7810 btrfs_set_tree_block_key(leaf
, block_info
, key
);
7811 btrfs_set_tree_block_level(leaf
, block_info
, level
);
7812 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
7816 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
7817 btrfs_set_extent_inline_ref_type(leaf
, iref
,
7818 BTRFS_SHARED_BLOCK_REF_KEY
);
7819 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
7821 btrfs_set_extent_inline_ref_type(leaf
, iref
,
7822 BTRFS_TREE_BLOCK_REF_KEY
);
7823 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
7826 btrfs_mark_buffer_dirty(leaf
);
7827 btrfs_free_path(path
);
7829 ret
= remove_from_free_space_tree(trans
, fs_info
, ins
->objectid
,
7834 ret
= update_block_group(trans
, root
, ins
->objectid
, root
->nodesize
,
7836 if (ret
) { /* -ENOENT, logic error */
7837 btrfs_err(fs_info
, "update block group failed for %llu %llu",
7838 ins
->objectid
, ins
->offset
);
7842 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, root
->nodesize
);
7846 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
7847 struct btrfs_root
*root
,
7848 u64 root_objectid
, u64 owner
,
7849 u64 offset
, u64 ram_bytes
,
7850 struct btrfs_key
*ins
)
7854 BUG_ON(root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
7856 ret
= btrfs_add_delayed_data_ref(root
->fs_info
, trans
, ins
->objectid
,
7858 root_objectid
, owner
, offset
,
7859 ram_bytes
, BTRFS_ADD_DELAYED_EXTENT
,
7865 * this is used by the tree logging recovery code. It records that
7866 * an extent has been allocated and makes sure to clear the free
7867 * space cache bits as well
7869 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
7870 struct btrfs_root
*root
,
7871 u64 root_objectid
, u64 owner
, u64 offset
,
7872 struct btrfs_key
*ins
)
7875 struct btrfs_block_group_cache
*block_group
;
7878 * Mixed block groups will exclude before processing the log so we only
7879 * need to do the exlude dance if this fs isn't mixed.
7881 if (!btrfs_fs_incompat(root
->fs_info
, MIXED_GROUPS
)) {
7882 ret
= __exclude_logged_extent(root
, ins
->objectid
, ins
->offset
);
7887 block_group
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
7891 ret
= btrfs_update_reserved_bytes(block_group
, ins
->offset
,
7892 RESERVE_ALLOC_NO_ACCOUNT
, 0);
7893 BUG_ON(ret
); /* logic error */
7894 ret
= alloc_reserved_file_extent(trans
, root
, 0, root_objectid
,
7895 0, owner
, offset
, ins
, 1);
7896 btrfs_put_block_group(block_group
);
7900 static struct extent_buffer
*
7901 btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
7902 u64 bytenr
, int level
)
7904 struct extent_buffer
*buf
;
7906 buf
= btrfs_find_create_tree_block(root
, bytenr
);
7908 return ERR_PTR(-ENOMEM
);
7909 btrfs_set_header_generation(buf
, trans
->transid
);
7910 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, buf
, level
);
7911 btrfs_tree_lock(buf
);
7912 clean_tree_block(trans
, root
->fs_info
, buf
);
7913 clear_bit(EXTENT_BUFFER_STALE
, &buf
->bflags
);
7915 btrfs_set_lock_blocking(buf
);
7916 set_extent_buffer_uptodate(buf
);
7918 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
7919 buf
->log_index
= root
->log_transid
% 2;
7921 * we allow two log transactions at a time, use different
7922 * EXENT bit to differentiate dirty pages.
7924 if (buf
->log_index
== 0)
7925 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
7926 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
7928 set_extent_new(&root
->dirty_log_pages
, buf
->start
,
7929 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
7931 buf
->log_index
= -1;
7932 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
7933 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
7935 trans
->blocks_used
++;
7936 /* this returns a buffer locked for blocking */
7940 static struct btrfs_block_rsv
*
7941 use_block_rsv(struct btrfs_trans_handle
*trans
,
7942 struct btrfs_root
*root
, u32 blocksize
)
7944 struct btrfs_block_rsv
*block_rsv
;
7945 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
7947 bool global_updated
= false;
7949 block_rsv
= get_block_rsv(trans
, root
);
7951 if (unlikely(block_rsv
->size
== 0))
7954 ret
= block_rsv_use_bytes(block_rsv
, blocksize
);
7958 if (block_rsv
->failfast
)
7959 return ERR_PTR(ret
);
7961 if (block_rsv
->type
== BTRFS_BLOCK_RSV_GLOBAL
&& !global_updated
) {
7962 global_updated
= true;
7963 update_global_block_rsv(root
->fs_info
);
7967 if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
7968 static DEFINE_RATELIMIT_STATE(_rs
,
7969 DEFAULT_RATELIMIT_INTERVAL
* 10,
7970 /*DEFAULT_RATELIMIT_BURST*/ 1);
7971 if (__ratelimit(&_rs
))
7973 "BTRFS: block rsv returned %d\n", ret
);
7976 ret
= reserve_metadata_bytes(root
, block_rsv
, blocksize
,
7977 BTRFS_RESERVE_NO_FLUSH
);
7981 * If we couldn't reserve metadata bytes try and use some from
7982 * the global reserve if its space type is the same as the global
7985 if (block_rsv
->type
!= BTRFS_BLOCK_RSV_GLOBAL
&&
7986 block_rsv
->space_info
== global_rsv
->space_info
) {
7987 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
7991 return ERR_PTR(ret
);
7994 static void unuse_block_rsv(struct btrfs_fs_info
*fs_info
,
7995 struct btrfs_block_rsv
*block_rsv
, u32 blocksize
)
7997 block_rsv_add_bytes(block_rsv
, blocksize
, 0);
7998 block_rsv_release_bytes(fs_info
, block_rsv
, NULL
, 0);
8002 * finds a free extent and does all the dirty work required for allocation
8003 * returns the tree buffer or an ERR_PTR on error.
8005 struct extent_buffer
*btrfs_alloc_tree_block(struct btrfs_trans_handle
*trans
,
8006 struct btrfs_root
*root
,
8007 u64 parent
, u64 root_objectid
,
8008 struct btrfs_disk_key
*key
, int level
,
8009 u64 hint
, u64 empty_size
)
8011 struct btrfs_key ins
;
8012 struct btrfs_block_rsv
*block_rsv
;
8013 struct extent_buffer
*buf
;
8014 struct btrfs_delayed_extent_op
*extent_op
;
8017 u32 blocksize
= root
->nodesize
;
8018 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
8021 if (btrfs_test_is_dummy_root(root
)) {
8022 buf
= btrfs_init_new_buffer(trans
, root
, root
->alloc_bytenr
,
8025 root
->alloc_bytenr
+= blocksize
;
8029 block_rsv
= use_block_rsv(trans
, root
, blocksize
);
8030 if (IS_ERR(block_rsv
))
8031 return ERR_CAST(block_rsv
);
8033 ret
= btrfs_reserve_extent(root
, blocksize
, blocksize
,
8034 empty_size
, hint
, &ins
, 0, 0);
8038 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
, level
);
8041 goto out_free_reserved
;
8044 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
8046 parent
= ins
.objectid
;
8047 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
8051 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
8052 extent_op
= btrfs_alloc_delayed_extent_op();
8058 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
8060 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
8061 extent_op
->flags_to_set
= flags
;
8062 extent_op
->update_key
= skinny_metadata
? false : true;
8063 extent_op
->update_flags
= true;
8064 extent_op
->is_data
= false;
8065 extent_op
->level
= level
;
8067 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
8068 ins
.objectid
, ins
.offset
,
8069 parent
, root_objectid
, level
,
8070 BTRFS_ADD_DELAYED_EXTENT
,
8073 goto out_free_delayed
;
8078 btrfs_free_delayed_extent_op(extent_op
);
8080 free_extent_buffer(buf
);
8082 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
, 0);
8084 unuse_block_rsv(root
->fs_info
, block_rsv
, blocksize
);
8085 return ERR_PTR(ret
);
8088 struct walk_control
{
8089 u64 refs
[BTRFS_MAX_LEVEL
];
8090 u64 flags
[BTRFS_MAX_LEVEL
];
8091 struct btrfs_key update_progress
;
8102 #define DROP_REFERENCE 1
8103 #define UPDATE_BACKREF 2
8105 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
8106 struct btrfs_root
*root
,
8107 struct walk_control
*wc
,
8108 struct btrfs_path
*path
)
8116 struct btrfs_key key
;
8117 struct extent_buffer
*eb
;
8122 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
8123 wc
->reada_count
= wc
->reada_count
* 2 / 3;
8124 wc
->reada_count
= max(wc
->reada_count
, 2);
8126 wc
->reada_count
= wc
->reada_count
* 3 / 2;
8127 wc
->reada_count
= min_t(int, wc
->reada_count
,
8128 BTRFS_NODEPTRS_PER_BLOCK(root
));
8131 eb
= path
->nodes
[wc
->level
];
8132 nritems
= btrfs_header_nritems(eb
);
8133 blocksize
= root
->nodesize
;
8135 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
8136 if (nread
>= wc
->reada_count
)
8140 bytenr
= btrfs_node_blockptr(eb
, slot
);
8141 generation
= btrfs_node_ptr_generation(eb
, slot
);
8143 if (slot
== path
->slots
[wc
->level
])
8146 if (wc
->stage
== UPDATE_BACKREF
&&
8147 generation
<= root
->root_key
.offset
)
8150 /* We don't lock the tree block, it's OK to be racy here */
8151 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
,
8152 wc
->level
- 1, 1, &refs
,
8154 /* We don't care about errors in readahead. */
8159 if (wc
->stage
== DROP_REFERENCE
) {
8163 if (wc
->level
== 1 &&
8164 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8166 if (!wc
->update_ref
||
8167 generation
<= root
->root_key
.offset
)
8169 btrfs_node_key_to_cpu(eb
, &key
, slot
);
8170 ret
= btrfs_comp_cpu_keys(&key
,
8171 &wc
->update_progress
);
8175 if (wc
->level
== 1 &&
8176 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8180 readahead_tree_block(root
, bytenr
);
8183 wc
->reada_slot
= slot
;
8187 * These may not be seen by the usual inc/dec ref code so we have to
8190 static int record_one_subtree_extent(struct btrfs_trans_handle
*trans
,
8191 struct btrfs_root
*root
, u64 bytenr
,
8194 struct btrfs_qgroup_extent_record
*qrecord
;
8195 struct btrfs_delayed_ref_root
*delayed_refs
;
8197 qrecord
= kmalloc(sizeof(*qrecord
), GFP_NOFS
);
8201 qrecord
->bytenr
= bytenr
;
8202 qrecord
->num_bytes
= num_bytes
;
8203 qrecord
->old_roots
= NULL
;
8205 delayed_refs
= &trans
->transaction
->delayed_refs
;
8206 spin_lock(&delayed_refs
->lock
);
8207 if (btrfs_qgroup_insert_dirty_extent(delayed_refs
, qrecord
))
8209 spin_unlock(&delayed_refs
->lock
);
8214 static int account_leaf_items(struct btrfs_trans_handle
*trans
,
8215 struct btrfs_root
*root
,
8216 struct extent_buffer
*eb
)
8218 int nr
= btrfs_header_nritems(eb
);
8219 int i
, extent_type
, ret
;
8220 struct btrfs_key key
;
8221 struct btrfs_file_extent_item
*fi
;
8222 u64 bytenr
, num_bytes
;
8224 /* We can be called directly from walk_up_proc() */
8225 if (!root
->fs_info
->quota_enabled
)
8228 for (i
= 0; i
< nr
; i
++) {
8229 btrfs_item_key_to_cpu(eb
, &key
, i
);
8231 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
8234 fi
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
8235 /* filter out non qgroup-accountable extents */
8236 extent_type
= btrfs_file_extent_type(eb
, fi
);
8238 if (extent_type
== BTRFS_FILE_EXTENT_INLINE
)
8241 bytenr
= btrfs_file_extent_disk_bytenr(eb
, fi
);
8245 num_bytes
= btrfs_file_extent_disk_num_bytes(eb
, fi
);
8247 ret
= record_one_subtree_extent(trans
, root
, bytenr
, num_bytes
);
8255 * Walk up the tree from the bottom, freeing leaves and any interior
8256 * nodes which have had all slots visited. If a node (leaf or
8257 * interior) is freed, the node above it will have it's slot
8258 * incremented. The root node will never be freed.
8260 * At the end of this function, we should have a path which has all
8261 * slots incremented to the next position for a search. If we need to
8262 * read a new node it will be NULL and the node above it will have the
8263 * correct slot selected for a later read.
8265 * If we increment the root nodes slot counter past the number of
8266 * elements, 1 is returned to signal completion of the search.
8268 static int adjust_slots_upwards(struct btrfs_root
*root
,
8269 struct btrfs_path
*path
, int root_level
)
8273 struct extent_buffer
*eb
;
8275 if (root_level
== 0)
8278 while (level
<= root_level
) {
8279 eb
= path
->nodes
[level
];
8280 nr
= btrfs_header_nritems(eb
);
8281 path
->slots
[level
]++;
8282 slot
= path
->slots
[level
];
8283 if (slot
>= nr
|| level
== 0) {
8285 * Don't free the root - we will detect this
8286 * condition after our loop and return a
8287 * positive value for caller to stop walking the tree.
8289 if (level
!= root_level
) {
8290 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8291 path
->locks
[level
] = 0;
8293 free_extent_buffer(eb
);
8294 path
->nodes
[level
] = NULL
;
8295 path
->slots
[level
] = 0;
8299 * We have a valid slot to walk back down
8300 * from. Stop here so caller can process these
8309 eb
= path
->nodes
[root_level
];
8310 if (path
->slots
[root_level
] >= btrfs_header_nritems(eb
))
8317 * root_eb is the subtree root and is locked before this function is called.
8319 static int account_shared_subtree(struct btrfs_trans_handle
*trans
,
8320 struct btrfs_root
*root
,
8321 struct extent_buffer
*root_eb
,
8327 struct extent_buffer
*eb
= root_eb
;
8328 struct btrfs_path
*path
= NULL
;
8330 BUG_ON(root_level
< 0 || root_level
> BTRFS_MAX_LEVEL
);
8331 BUG_ON(root_eb
== NULL
);
8333 if (!root
->fs_info
->quota_enabled
)
8336 if (!extent_buffer_uptodate(root_eb
)) {
8337 ret
= btrfs_read_buffer(root_eb
, root_gen
);
8342 if (root_level
== 0) {
8343 ret
= account_leaf_items(trans
, root
, root_eb
);
8347 path
= btrfs_alloc_path();
8352 * Walk down the tree. Missing extent blocks are filled in as
8353 * we go. Metadata is accounted every time we read a new
8356 * When we reach a leaf, we account for file extent items in it,
8357 * walk back up the tree (adjusting slot pointers as we go)
8358 * and restart the search process.
8360 extent_buffer_get(root_eb
); /* For path */
8361 path
->nodes
[root_level
] = root_eb
;
8362 path
->slots
[root_level
] = 0;
8363 path
->locks
[root_level
] = 0; /* so release_path doesn't try to unlock */
8366 while (level
>= 0) {
8367 if (path
->nodes
[level
] == NULL
) {
8372 /* We need to get child blockptr/gen from
8373 * parent before we can read it. */
8374 eb
= path
->nodes
[level
+ 1];
8375 parent_slot
= path
->slots
[level
+ 1];
8376 child_bytenr
= btrfs_node_blockptr(eb
, parent_slot
);
8377 child_gen
= btrfs_node_ptr_generation(eb
, parent_slot
);
8379 eb
= read_tree_block(root
, child_bytenr
, child_gen
);
8383 } else if (!extent_buffer_uptodate(eb
)) {
8384 free_extent_buffer(eb
);
8389 path
->nodes
[level
] = eb
;
8390 path
->slots
[level
] = 0;
8392 btrfs_tree_read_lock(eb
);
8393 btrfs_set_lock_blocking_rw(eb
, BTRFS_READ_LOCK
);
8394 path
->locks
[level
] = BTRFS_READ_LOCK_BLOCKING
;
8396 ret
= record_one_subtree_extent(trans
, root
, child_bytenr
,
8403 ret
= account_leaf_items(trans
, root
, path
->nodes
[level
]);
8407 /* Nonzero return here means we completed our search */
8408 ret
= adjust_slots_upwards(root
, path
, root_level
);
8412 /* Restart search with new slots */
8421 btrfs_free_path(path
);
8427 * helper to process tree block while walking down the tree.
8429 * when wc->stage == UPDATE_BACKREF, this function updates
8430 * back refs for pointers in the block.
8432 * NOTE: return value 1 means we should stop walking down.
8434 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
8435 struct btrfs_root
*root
,
8436 struct btrfs_path
*path
,
8437 struct walk_control
*wc
, int lookup_info
)
8439 int level
= wc
->level
;
8440 struct extent_buffer
*eb
= path
->nodes
[level
];
8441 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
8444 if (wc
->stage
== UPDATE_BACKREF
&&
8445 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
8449 * when reference count of tree block is 1, it won't increase
8450 * again. once full backref flag is set, we never clear it.
8453 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
8454 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
8455 BUG_ON(!path
->locks
[level
]);
8456 ret
= btrfs_lookup_extent_info(trans
, root
,
8457 eb
->start
, level
, 1,
8460 BUG_ON(ret
== -ENOMEM
);
8463 BUG_ON(wc
->refs
[level
] == 0);
8466 if (wc
->stage
== DROP_REFERENCE
) {
8467 if (wc
->refs
[level
] > 1)
8470 if (path
->locks
[level
] && !wc
->keep_locks
) {
8471 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8472 path
->locks
[level
] = 0;
8477 /* wc->stage == UPDATE_BACKREF */
8478 if (!(wc
->flags
[level
] & flag
)) {
8479 BUG_ON(!path
->locks
[level
]);
8480 ret
= btrfs_inc_ref(trans
, root
, eb
, 1);
8481 BUG_ON(ret
); /* -ENOMEM */
8482 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
8483 BUG_ON(ret
); /* -ENOMEM */
8484 ret
= btrfs_set_disk_extent_flags(trans
, root
, eb
->start
,
8486 btrfs_header_level(eb
), 0);
8487 BUG_ON(ret
); /* -ENOMEM */
8488 wc
->flags
[level
] |= flag
;
8492 * the block is shared by multiple trees, so it's not good to
8493 * keep the tree lock
8495 if (path
->locks
[level
] && level
> 0) {
8496 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8497 path
->locks
[level
] = 0;
8503 * helper to process tree block pointer.
8505 * when wc->stage == DROP_REFERENCE, this function checks
8506 * reference count of the block pointed to. if the block
8507 * is shared and we need update back refs for the subtree
8508 * rooted at the block, this function changes wc->stage to
8509 * UPDATE_BACKREF. if the block is shared and there is no
8510 * need to update back, this function drops the reference
8513 * NOTE: return value 1 means we should stop walking down.
8515 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
8516 struct btrfs_root
*root
,
8517 struct btrfs_path
*path
,
8518 struct walk_control
*wc
, int *lookup_info
)
8524 struct btrfs_key key
;
8525 struct extent_buffer
*next
;
8526 int level
= wc
->level
;
8529 bool need_account
= false;
8531 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
8532 path
->slots
[level
]);
8534 * if the lower level block was created before the snapshot
8535 * was created, we know there is no need to update back refs
8538 if (wc
->stage
== UPDATE_BACKREF
&&
8539 generation
<= root
->root_key
.offset
) {
8544 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
8545 blocksize
= root
->nodesize
;
8547 next
= btrfs_find_tree_block(root
->fs_info
, bytenr
);
8549 next
= btrfs_find_create_tree_block(root
, bytenr
);
8552 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, next
,
8556 btrfs_tree_lock(next
);
8557 btrfs_set_lock_blocking(next
);
8559 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, level
- 1, 1,
8560 &wc
->refs
[level
- 1],
8561 &wc
->flags
[level
- 1]);
8563 btrfs_tree_unlock(next
);
8567 if (unlikely(wc
->refs
[level
- 1] == 0)) {
8568 btrfs_err(root
->fs_info
, "Missing references.");
8573 if (wc
->stage
== DROP_REFERENCE
) {
8574 if (wc
->refs
[level
- 1] > 1) {
8575 need_account
= true;
8577 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8580 if (!wc
->update_ref
||
8581 generation
<= root
->root_key
.offset
)
8584 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
8585 path
->slots
[level
]);
8586 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
8590 wc
->stage
= UPDATE_BACKREF
;
8591 wc
->shared_level
= level
- 1;
8595 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8599 if (!btrfs_buffer_uptodate(next
, generation
, 0)) {
8600 btrfs_tree_unlock(next
);
8601 free_extent_buffer(next
);
8607 if (reada
&& level
== 1)
8608 reada_walk_down(trans
, root
, wc
, path
);
8609 next
= read_tree_block(root
, bytenr
, generation
);
8611 return PTR_ERR(next
);
8612 } else if (!extent_buffer_uptodate(next
)) {
8613 free_extent_buffer(next
);
8616 btrfs_tree_lock(next
);
8617 btrfs_set_lock_blocking(next
);
8621 BUG_ON(level
!= btrfs_header_level(next
));
8622 path
->nodes
[level
] = next
;
8623 path
->slots
[level
] = 0;
8624 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8630 wc
->refs
[level
- 1] = 0;
8631 wc
->flags
[level
- 1] = 0;
8632 if (wc
->stage
== DROP_REFERENCE
) {
8633 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
8634 parent
= path
->nodes
[level
]->start
;
8636 BUG_ON(root
->root_key
.objectid
!=
8637 btrfs_header_owner(path
->nodes
[level
]));
8642 ret
= account_shared_subtree(trans
, root
, next
,
8643 generation
, level
- 1);
8645 btrfs_err_rl(root
->fs_info
,
8647 "%d accounting shared subtree. Quota "
8648 "is out of sync, rescan required.",
8652 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
, parent
,
8653 root
->root_key
.objectid
, level
- 1, 0);
8654 BUG_ON(ret
); /* -ENOMEM */
8656 btrfs_tree_unlock(next
);
8657 free_extent_buffer(next
);
8663 * helper to process tree block while walking up the tree.
8665 * when wc->stage == DROP_REFERENCE, this function drops
8666 * reference count on the block.
8668 * when wc->stage == UPDATE_BACKREF, this function changes
8669 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8670 * to UPDATE_BACKREF previously while processing the block.
8672 * NOTE: return value 1 means we should stop walking up.
8674 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
8675 struct btrfs_root
*root
,
8676 struct btrfs_path
*path
,
8677 struct walk_control
*wc
)
8680 int level
= wc
->level
;
8681 struct extent_buffer
*eb
= path
->nodes
[level
];
8684 if (wc
->stage
== UPDATE_BACKREF
) {
8685 BUG_ON(wc
->shared_level
< level
);
8686 if (level
< wc
->shared_level
)
8689 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
8693 wc
->stage
= DROP_REFERENCE
;
8694 wc
->shared_level
= -1;
8695 path
->slots
[level
] = 0;
8698 * check reference count again if the block isn't locked.
8699 * we should start walking down the tree again if reference
8702 if (!path
->locks
[level
]) {
8704 btrfs_tree_lock(eb
);
8705 btrfs_set_lock_blocking(eb
);
8706 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8708 ret
= btrfs_lookup_extent_info(trans
, root
,
8709 eb
->start
, level
, 1,
8713 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8714 path
->locks
[level
] = 0;
8717 BUG_ON(wc
->refs
[level
] == 0);
8718 if (wc
->refs
[level
] == 1) {
8719 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8720 path
->locks
[level
] = 0;
8726 /* wc->stage == DROP_REFERENCE */
8727 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
8729 if (wc
->refs
[level
] == 1) {
8731 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8732 ret
= btrfs_dec_ref(trans
, root
, eb
, 1);
8734 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
8735 BUG_ON(ret
); /* -ENOMEM */
8736 ret
= account_leaf_items(trans
, root
, eb
);
8738 btrfs_err_rl(root
->fs_info
,
8740 "%d accounting leaf items. Quota "
8741 "is out of sync, rescan required.",
8745 /* make block locked assertion in clean_tree_block happy */
8746 if (!path
->locks
[level
] &&
8747 btrfs_header_generation(eb
) == trans
->transid
) {
8748 btrfs_tree_lock(eb
);
8749 btrfs_set_lock_blocking(eb
);
8750 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8752 clean_tree_block(trans
, root
->fs_info
, eb
);
8755 if (eb
== root
->node
) {
8756 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8759 BUG_ON(root
->root_key
.objectid
!=
8760 btrfs_header_owner(eb
));
8762 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8763 parent
= path
->nodes
[level
+ 1]->start
;
8765 BUG_ON(root
->root_key
.objectid
!=
8766 btrfs_header_owner(path
->nodes
[level
+ 1]));
8769 btrfs_free_tree_block(trans
, root
, eb
, parent
, wc
->refs
[level
] == 1);
8771 wc
->refs
[level
] = 0;
8772 wc
->flags
[level
] = 0;
8776 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
8777 struct btrfs_root
*root
,
8778 struct btrfs_path
*path
,
8779 struct walk_control
*wc
)
8781 int level
= wc
->level
;
8782 int lookup_info
= 1;
8785 while (level
>= 0) {
8786 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
8793 if (path
->slots
[level
] >=
8794 btrfs_header_nritems(path
->nodes
[level
]))
8797 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
8799 path
->slots
[level
]++;
8808 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
8809 struct btrfs_root
*root
,
8810 struct btrfs_path
*path
,
8811 struct walk_control
*wc
, int max_level
)
8813 int level
= wc
->level
;
8816 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
8817 while (level
< max_level
&& path
->nodes
[level
]) {
8819 if (path
->slots
[level
] + 1 <
8820 btrfs_header_nritems(path
->nodes
[level
])) {
8821 path
->slots
[level
]++;
8824 ret
= walk_up_proc(trans
, root
, path
, wc
);
8828 if (path
->locks
[level
]) {
8829 btrfs_tree_unlock_rw(path
->nodes
[level
],
8830 path
->locks
[level
]);
8831 path
->locks
[level
] = 0;
8833 free_extent_buffer(path
->nodes
[level
]);
8834 path
->nodes
[level
] = NULL
;
8842 * drop a subvolume tree.
8844 * this function traverses the tree freeing any blocks that only
8845 * referenced by the tree.
8847 * when a shared tree block is found. this function decreases its
8848 * reference count by one. if update_ref is true, this function
8849 * also make sure backrefs for the shared block and all lower level
8850 * blocks are properly updated.
8852 * If called with for_reloc == 0, may exit early with -EAGAIN
8854 int btrfs_drop_snapshot(struct btrfs_root
*root
,
8855 struct btrfs_block_rsv
*block_rsv
, int update_ref
,
8858 struct btrfs_path
*path
;
8859 struct btrfs_trans_handle
*trans
;
8860 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
8861 struct btrfs_root_item
*root_item
= &root
->root_item
;
8862 struct walk_control
*wc
;
8863 struct btrfs_key key
;
8867 bool root_dropped
= false;
8869 btrfs_debug(root
->fs_info
, "Drop subvolume %llu", root
->objectid
);
8871 path
= btrfs_alloc_path();
8877 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
8879 btrfs_free_path(path
);
8884 trans
= btrfs_start_transaction(tree_root
, 0);
8885 if (IS_ERR(trans
)) {
8886 err
= PTR_ERR(trans
);
8891 trans
->block_rsv
= block_rsv
;
8893 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
8894 level
= btrfs_header_level(root
->node
);
8895 path
->nodes
[level
] = btrfs_lock_root_node(root
);
8896 btrfs_set_lock_blocking(path
->nodes
[level
]);
8897 path
->slots
[level
] = 0;
8898 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8899 memset(&wc
->update_progress
, 0,
8900 sizeof(wc
->update_progress
));
8902 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
8903 memcpy(&wc
->update_progress
, &key
,
8904 sizeof(wc
->update_progress
));
8906 level
= root_item
->drop_level
;
8908 path
->lowest_level
= level
;
8909 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
8910 path
->lowest_level
= 0;
8918 * unlock our path, this is safe because only this
8919 * function is allowed to delete this snapshot
8921 btrfs_unlock_up_safe(path
, 0);
8923 level
= btrfs_header_level(root
->node
);
8925 btrfs_tree_lock(path
->nodes
[level
]);
8926 btrfs_set_lock_blocking(path
->nodes
[level
]);
8927 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8929 ret
= btrfs_lookup_extent_info(trans
, root
,
8930 path
->nodes
[level
]->start
,
8931 level
, 1, &wc
->refs
[level
],
8937 BUG_ON(wc
->refs
[level
] == 0);
8939 if (level
== root_item
->drop_level
)
8942 btrfs_tree_unlock(path
->nodes
[level
]);
8943 path
->locks
[level
] = 0;
8944 WARN_ON(wc
->refs
[level
] != 1);
8950 wc
->shared_level
= -1;
8951 wc
->stage
= DROP_REFERENCE
;
8952 wc
->update_ref
= update_ref
;
8954 wc
->for_reloc
= for_reloc
;
8955 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
8959 ret
= walk_down_tree(trans
, root
, path
, wc
);
8965 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
8972 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
8976 if (wc
->stage
== DROP_REFERENCE
) {
8978 btrfs_node_key(path
->nodes
[level
],
8979 &root_item
->drop_progress
,
8980 path
->slots
[level
]);
8981 root_item
->drop_level
= level
;
8984 BUG_ON(wc
->level
== 0);
8985 if (btrfs_should_end_transaction(trans
, tree_root
) ||
8986 (!for_reloc
&& btrfs_need_cleaner_sleep(root
))) {
8987 ret
= btrfs_update_root(trans
, tree_root
,
8991 btrfs_abort_transaction(trans
, tree_root
, ret
);
8996 btrfs_end_transaction_throttle(trans
, tree_root
);
8997 if (!for_reloc
&& btrfs_need_cleaner_sleep(root
)) {
8998 pr_debug("BTRFS: drop snapshot early exit\n");
9003 trans
= btrfs_start_transaction(tree_root
, 0);
9004 if (IS_ERR(trans
)) {
9005 err
= PTR_ERR(trans
);
9009 trans
->block_rsv
= block_rsv
;
9012 btrfs_release_path(path
);
9016 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
9018 btrfs_abort_transaction(trans
, tree_root
, ret
);
9022 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
9023 ret
= btrfs_find_root(tree_root
, &root
->root_key
, path
,
9026 btrfs_abort_transaction(trans
, tree_root
, ret
);
9029 } else if (ret
> 0) {
9030 /* if we fail to delete the orphan item this time
9031 * around, it'll get picked up the next time.
9033 * The most common failure here is just -ENOENT.
9035 btrfs_del_orphan_item(trans
, tree_root
,
9036 root
->root_key
.objectid
);
9040 if (test_bit(BTRFS_ROOT_IN_RADIX
, &root
->state
)) {
9041 btrfs_add_dropped_root(trans
, root
);
9043 free_extent_buffer(root
->node
);
9044 free_extent_buffer(root
->commit_root
);
9045 btrfs_put_fs_root(root
);
9047 root_dropped
= true;
9049 btrfs_end_transaction_throttle(trans
, tree_root
);
9052 btrfs_free_path(path
);
9055 * So if we need to stop dropping the snapshot for whatever reason we
9056 * need to make sure to add it back to the dead root list so that we
9057 * keep trying to do the work later. This also cleans up roots if we
9058 * don't have it in the radix (like when we recover after a power fail
9059 * or unmount) so we don't leak memory.
9061 if (!for_reloc
&& root_dropped
== false)
9062 btrfs_add_dead_root(root
);
9063 if (err
&& err
!= -EAGAIN
)
9064 btrfs_std_error(root
->fs_info
, err
, NULL
);
9069 * drop subtree rooted at tree block 'node'.
9071 * NOTE: this function will unlock and release tree block 'node'
9072 * only used by relocation code
9074 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
9075 struct btrfs_root
*root
,
9076 struct extent_buffer
*node
,
9077 struct extent_buffer
*parent
)
9079 struct btrfs_path
*path
;
9080 struct walk_control
*wc
;
9086 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
9088 path
= btrfs_alloc_path();
9092 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
9094 btrfs_free_path(path
);
9098 btrfs_assert_tree_locked(parent
);
9099 parent_level
= btrfs_header_level(parent
);
9100 extent_buffer_get(parent
);
9101 path
->nodes
[parent_level
] = parent
;
9102 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
9104 btrfs_assert_tree_locked(node
);
9105 level
= btrfs_header_level(node
);
9106 path
->nodes
[level
] = node
;
9107 path
->slots
[level
] = 0;
9108 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
9110 wc
->refs
[parent_level
] = 1;
9111 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
9113 wc
->shared_level
= -1;
9114 wc
->stage
= DROP_REFERENCE
;
9118 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
9121 wret
= walk_down_tree(trans
, root
, path
, wc
);
9127 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
9135 btrfs_free_path(path
);
9139 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
9145 * if restripe for this chunk_type is on pick target profile and
9146 * return, otherwise do the usual balance
9148 stripped
= get_restripe_target(root
->fs_info
, flags
);
9150 return extended_to_chunk(stripped
);
9152 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
9154 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
9155 BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
|
9156 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
9158 if (num_devices
== 1) {
9159 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
9160 stripped
= flags
& ~stripped
;
9162 /* turn raid0 into single device chunks */
9163 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
9166 /* turn mirroring into duplication */
9167 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
9168 BTRFS_BLOCK_GROUP_RAID10
))
9169 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
9171 /* they already had raid on here, just return */
9172 if (flags
& stripped
)
9175 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
9176 stripped
= flags
& ~stripped
;
9178 /* switch duplicated blocks with raid1 */
9179 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
9180 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
9182 /* this is drive concat, leave it alone */
9188 static int inc_block_group_ro(struct btrfs_block_group_cache
*cache
, int force
)
9190 struct btrfs_space_info
*sinfo
= cache
->space_info
;
9192 u64 min_allocable_bytes
;
9196 * We need some metadata space and system metadata space for
9197 * allocating chunks in some corner cases until we force to set
9198 * it to be readonly.
9201 (BTRFS_BLOCK_GROUP_SYSTEM
| BTRFS_BLOCK_GROUP_METADATA
)) &&
9203 min_allocable_bytes
= SZ_1M
;
9205 min_allocable_bytes
= 0;
9207 spin_lock(&sinfo
->lock
);
9208 spin_lock(&cache
->lock
);
9216 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
9217 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
9219 if (sinfo
->bytes_used
+ sinfo
->bytes_reserved
+ sinfo
->bytes_pinned
+
9220 sinfo
->bytes_may_use
+ sinfo
->bytes_readonly
+ num_bytes
+
9221 min_allocable_bytes
<= sinfo
->total_bytes
) {
9222 sinfo
->bytes_readonly
+= num_bytes
;
9224 list_add_tail(&cache
->ro_list
, &sinfo
->ro_bgs
);
9228 spin_unlock(&cache
->lock
);
9229 spin_unlock(&sinfo
->lock
);
9233 int btrfs_inc_block_group_ro(struct btrfs_root
*root
,
9234 struct btrfs_block_group_cache
*cache
)
9237 struct btrfs_trans_handle
*trans
;
9242 trans
= btrfs_join_transaction(root
);
9244 return PTR_ERR(trans
);
9247 * we're not allowed to set block groups readonly after the dirty
9248 * block groups cache has started writing. If it already started,
9249 * back off and let this transaction commit
9251 mutex_lock(&root
->fs_info
->ro_block_group_mutex
);
9252 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN
, &trans
->transaction
->flags
)) {
9253 u64 transid
= trans
->transid
;
9255 mutex_unlock(&root
->fs_info
->ro_block_group_mutex
);
9256 btrfs_end_transaction(trans
, root
);
9258 ret
= btrfs_wait_for_commit(root
, transid
);
9265 * if we are changing raid levels, try to allocate a corresponding
9266 * block group with the new raid level.
9268 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
9269 if (alloc_flags
!= cache
->flags
) {
9270 ret
= do_chunk_alloc(trans
, root
, alloc_flags
,
9273 * ENOSPC is allowed here, we may have enough space
9274 * already allocated at the new raid level to
9283 ret
= inc_block_group_ro(cache
, 0);
9286 alloc_flags
= get_alloc_profile(root
, cache
->space_info
->flags
);
9287 ret
= do_chunk_alloc(trans
, root
, alloc_flags
,
9291 ret
= inc_block_group_ro(cache
, 0);
9293 if (cache
->flags
& BTRFS_BLOCK_GROUP_SYSTEM
) {
9294 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
9295 lock_chunks(root
->fs_info
->chunk_root
);
9296 check_system_chunk(trans
, root
, alloc_flags
);
9297 unlock_chunks(root
->fs_info
->chunk_root
);
9299 mutex_unlock(&root
->fs_info
->ro_block_group_mutex
);
9301 btrfs_end_transaction(trans
, root
);
9305 int btrfs_force_chunk_alloc(struct btrfs_trans_handle
*trans
,
9306 struct btrfs_root
*root
, u64 type
)
9308 u64 alloc_flags
= get_alloc_profile(root
, type
);
9309 return do_chunk_alloc(trans
, root
, alloc_flags
,
9314 * helper to account the unused space of all the readonly block group in the
9315 * space_info. takes mirrors into account.
9317 u64
btrfs_account_ro_block_groups_free_space(struct btrfs_space_info
*sinfo
)
9319 struct btrfs_block_group_cache
*block_group
;
9323 /* It's df, we don't care if it's racey */
9324 if (list_empty(&sinfo
->ro_bgs
))
9327 spin_lock(&sinfo
->lock
);
9328 list_for_each_entry(block_group
, &sinfo
->ro_bgs
, ro_list
) {
9329 spin_lock(&block_group
->lock
);
9331 if (!block_group
->ro
) {
9332 spin_unlock(&block_group
->lock
);
9336 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_RAID1
|
9337 BTRFS_BLOCK_GROUP_RAID10
|
9338 BTRFS_BLOCK_GROUP_DUP
))
9343 free_bytes
+= (block_group
->key
.offset
-
9344 btrfs_block_group_used(&block_group
->item
)) *
9347 spin_unlock(&block_group
->lock
);
9349 spin_unlock(&sinfo
->lock
);
9354 void btrfs_dec_block_group_ro(struct btrfs_root
*root
,
9355 struct btrfs_block_group_cache
*cache
)
9357 struct btrfs_space_info
*sinfo
= cache
->space_info
;
9362 spin_lock(&sinfo
->lock
);
9363 spin_lock(&cache
->lock
);
9365 num_bytes
= cache
->key
.offset
- cache
->reserved
-
9366 cache
->pinned
- cache
->bytes_super
-
9367 btrfs_block_group_used(&cache
->item
);
9368 sinfo
->bytes_readonly
-= num_bytes
;
9369 list_del_init(&cache
->ro_list
);
9371 spin_unlock(&cache
->lock
);
9372 spin_unlock(&sinfo
->lock
);
9376 * checks to see if its even possible to relocate this block group.
9378 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9379 * ok to go ahead and try.
9381 int btrfs_can_relocate(struct btrfs_root
*root
, u64 bytenr
)
9383 struct btrfs_block_group_cache
*block_group
;
9384 struct btrfs_space_info
*space_info
;
9385 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
9386 struct btrfs_device
*device
;
9387 struct btrfs_trans_handle
*trans
;
9397 debug
= btrfs_test_opt(root
, ENOSPC_DEBUG
);
9399 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
9401 /* odd, couldn't find the block group, leave it alone */
9404 btrfs_warn(root
->fs_info
,
9405 "can't find block group for bytenr %llu",
9410 min_free
= btrfs_block_group_used(&block_group
->item
);
9412 /* no bytes used, we're good */
9416 space_info
= block_group
->space_info
;
9417 spin_lock(&space_info
->lock
);
9419 full
= space_info
->full
;
9422 * if this is the last block group we have in this space, we can't
9423 * relocate it unless we're able to allocate a new chunk below.
9425 * Otherwise, we need to make sure we have room in the space to handle
9426 * all of the extents from this block group. If we can, we're good
9428 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
9429 (space_info
->bytes_used
+ space_info
->bytes_reserved
+
9430 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
9431 min_free
< space_info
->total_bytes
)) {
9432 spin_unlock(&space_info
->lock
);
9435 spin_unlock(&space_info
->lock
);
9438 * ok we don't have enough space, but maybe we have free space on our
9439 * devices to allocate new chunks for relocation, so loop through our
9440 * alloc devices and guess if we have enough space. if this block
9441 * group is going to be restriped, run checks against the target
9442 * profile instead of the current one.
9454 target
= get_restripe_target(root
->fs_info
, block_group
->flags
);
9456 index
= __get_raid_index(extended_to_chunk(target
));
9459 * this is just a balance, so if we were marked as full
9460 * we know there is no space for a new chunk
9464 btrfs_warn(root
->fs_info
,
9465 "no space to alloc new chunk for block group %llu",
9466 block_group
->key
.objectid
);
9470 index
= get_block_group_index(block_group
);
9473 if (index
== BTRFS_RAID_RAID10
) {
9477 } else if (index
== BTRFS_RAID_RAID1
) {
9479 } else if (index
== BTRFS_RAID_DUP
) {
9482 } else if (index
== BTRFS_RAID_RAID0
) {
9483 dev_min
= fs_devices
->rw_devices
;
9484 min_free
= div64_u64(min_free
, dev_min
);
9487 /* We need to do this so that we can look at pending chunks */
9488 trans
= btrfs_join_transaction(root
);
9489 if (IS_ERR(trans
)) {
9490 ret
= PTR_ERR(trans
);
9494 mutex_lock(&root
->fs_info
->chunk_mutex
);
9495 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
9499 * check to make sure we can actually find a chunk with enough
9500 * space to fit our block group in.
9502 if (device
->total_bytes
> device
->bytes_used
+ min_free
&&
9503 !device
->is_tgtdev_for_dev_replace
) {
9504 ret
= find_free_dev_extent(trans
, device
, min_free
,
9509 if (dev_nr
>= dev_min
)
9515 if (debug
&& ret
== -1)
9516 btrfs_warn(root
->fs_info
,
9517 "no space to allocate a new chunk for block group %llu",
9518 block_group
->key
.objectid
);
9519 mutex_unlock(&root
->fs_info
->chunk_mutex
);
9520 btrfs_end_transaction(trans
, root
);
9522 btrfs_put_block_group(block_group
);
9526 static int find_first_block_group(struct btrfs_root
*root
,
9527 struct btrfs_path
*path
, struct btrfs_key
*key
)
9530 struct btrfs_key found_key
;
9531 struct extent_buffer
*leaf
;
9534 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
9539 slot
= path
->slots
[0];
9540 leaf
= path
->nodes
[0];
9541 if (slot
>= btrfs_header_nritems(leaf
)) {
9542 ret
= btrfs_next_leaf(root
, path
);
9549 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
9551 if (found_key
.objectid
>= key
->objectid
&&
9552 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
9562 void btrfs_put_block_group_cache(struct btrfs_fs_info
*info
)
9564 struct btrfs_block_group_cache
*block_group
;
9568 struct inode
*inode
;
9570 block_group
= btrfs_lookup_first_block_group(info
, last
);
9571 while (block_group
) {
9572 spin_lock(&block_group
->lock
);
9573 if (block_group
->iref
)
9575 spin_unlock(&block_group
->lock
);
9576 block_group
= next_block_group(info
->tree_root
,
9586 inode
= block_group
->inode
;
9587 block_group
->iref
= 0;
9588 block_group
->inode
= NULL
;
9589 spin_unlock(&block_group
->lock
);
9591 last
= block_group
->key
.objectid
+ block_group
->key
.offset
;
9592 btrfs_put_block_group(block_group
);
9596 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
9598 struct btrfs_block_group_cache
*block_group
;
9599 struct btrfs_space_info
*space_info
;
9600 struct btrfs_caching_control
*caching_ctl
;
9603 down_write(&info
->commit_root_sem
);
9604 while (!list_empty(&info
->caching_block_groups
)) {
9605 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
9606 struct btrfs_caching_control
, list
);
9607 list_del(&caching_ctl
->list
);
9608 put_caching_control(caching_ctl
);
9610 up_write(&info
->commit_root_sem
);
9612 spin_lock(&info
->unused_bgs_lock
);
9613 while (!list_empty(&info
->unused_bgs
)) {
9614 block_group
= list_first_entry(&info
->unused_bgs
,
9615 struct btrfs_block_group_cache
,
9617 list_del_init(&block_group
->bg_list
);
9618 btrfs_put_block_group(block_group
);
9620 spin_unlock(&info
->unused_bgs_lock
);
9622 spin_lock(&info
->block_group_cache_lock
);
9623 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
9624 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
9626 rb_erase(&block_group
->cache_node
,
9627 &info
->block_group_cache_tree
);
9628 RB_CLEAR_NODE(&block_group
->cache_node
);
9629 spin_unlock(&info
->block_group_cache_lock
);
9631 down_write(&block_group
->space_info
->groups_sem
);
9632 list_del(&block_group
->list
);
9633 up_write(&block_group
->space_info
->groups_sem
);
9635 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
9636 wait_block_group_cache_done(block_group
);
9639 * We haven't cached this block group, which means we could
9640 * possibly have excluded extents on this block group.
9642 if (block_group
->cached
== BTRFS_CACHE_NO
||
9643 block_group
->cached
== BTRFS_CACHE_ERROR
)
9644 free_excluded_extents(info
->extent_root
, block_group
);
9646 btrfs_remove_free_space_cache(block_group
);
9647 btrfs_put_block_group(block_group
);
9649 spin_lock(&info
->block_group_cache_lock
);
9651 spin_unlock(&info
->block_group_cache_lock
);
9653 /* now that all the block groups are freed, go through and
9654 * free all the space_info structs. This is only called during
9655 * the final stages of unmount, and so we know nobody is
9656 * using them. We call synchronize_rcu() once before we start,
9657 * just to be on the safe side.
9661 release_global_block_rsv(info
);
9663 while (!list_empty(&info
->space_info
)) {
9666 space_info
= list_entry(info
->space_info
.next
,
9667 struct btrfs_space_info
,
9669 if (btrfs_test_opt(info
->tree_root
, ENOSPC_DEBUG
)) {
9670 if (WARN_ON(space_info
->bytes_pinned
> 0 ||
9671 space_info
->bytes_reserved
> 0 ||
9672 space_info
->bytes_may_use
> 0)) {
9673 dump_space_info(space_info
, 0, 0);
9676 list_del(&space_info
->list
);
9677 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++) {
9678 struct kobject
*kobj
;
9679 kobj
= space_info
->block_group_kobjs
[i
];
9680 space_info
->block_group_kobjs
[i
] = NULL
;
9686 kobject_del(&space_info
->kobj
);
9687 kobject_put(&space_info
->kobj
);
9692 static void __link_block_group(struct btrfs_space_info
*space_info
,
9693 struct btrfs_block_group_cache
*cache
)
9695 int index
= get_block_group_index(cache
);
9698 down_write(&space_info
->groups_sem
);
9699 if (list_empty(&space_info
->block_groups
[index
]))
9701 list_add_tail(&cache
->list
, &space_info
->block_groups
[index
]);
9702 up_write(&space_info
->groups_sem
);
9705 struct raid_kobject
*rkobj
;
9708 rkobj
= kzalloc(sizeof(*rkobj
), GFP_NOFS
);
9711 rkobj
->raid_type
= index
;
9712 kobject_init(&rkobj
->kobj
, &btrfs_raid_ktype
);
9713 ret
= kobject_add(&rkobj
->kobj
, &space_info
->kobj
,
9714 "%s", get_raid_name(index
));
9716 kobject_put(&rkobj
->kobj
);
9719 space_info
->block_group_kobjs
[index
] = &rkobj
->kobj
;
9724 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
9727 static struct btrfs_block_group_cache
*
9728 btrfs_create_block_group_cache(struct btrfs_root
*root
, u64 start
, u64 size
)
9730 struct btrfs_block_group_cache
*cache
;
9732 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
9736 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
9738 if (!cache
->free_space_ctl
) {
9743 cache
->key
.objectid
= start
;
9744 cache
->key
.offset
= size
;
9745 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
9747 cache
->sectorsize
= root
->sectorsize
;
9748 cache
->fs_info
= root
->fs_info
;
9749 cache
->full_stripe_len
= btrfs_full_stripe_len(root
,
9750 &root
->fs_info
->mapping_tree
,
9752 set_free_space_tree_thresholds(cache
);
9754 atomic_set(&cache
->count
, 1);
9755 spin_lock_init(&cache
->lock
);
9756 init_rwsem(&cache
->data_rwsem
);
9757 INIT_LIST_HEAD(&cache
->list
);
9758 INIT_LIST_HEAD(&cache
->cluster_list
);
9759 INIT_LIST_HEAD(&cache
->bg_list
);
9760 INIT_LIST_HEAD(&cache
->ro_list
);
9761 INIT_LIST_HEAD(&cache
->dirty_list
);
9762 INIT_LIST_HEAD(&cache
->io_list
);
9763 btrfs_init_free_space_ctl(cache
);
9764 atomic_set(&cache
->trimming
, 0);
9765 mutex_init(&cache
->free_space_lock
);
9770 int btrfs_read_block_groups(struct btrfs_root
*root
)
9772 struct btrfs_path
*path
;
9774 struct btrfs_block_group_cache
*cache
;
9775 struct btrfs_fs_info
*info
= root
->fs_info
;
9776 struct btrfs_space_info
*space_info
;
9777 struct btrfs_key key
;
9778 struct btrfs_key found_key
;
9779 struct extent_buffer
*leaf
;
9783 root
= info
->extent_root
;
9786 key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
9787 path
= btrfs_alloc_path();
9790 path
->reada
= READA_FORWARD
;
9792 cache_gen
= btrfs_super_cache_generation(root
->fs_info
->super_copy
);
9793 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
9794 btrfs_super_generation(root
->fs_info
->super_copy
) != cache_gen
)
9796 if (btrfs_test_opt(root
, CLEAR_CACHE
))
9800 ret
= find_first_block_group(root
, path
, &key
);
9806 leaf
= path
->nodes
[0];
9807 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
9809 cache
= btrfs_create_block_group_cache(root
, found_key
.objectid
,
9818 * When we mount with old space cache, we need to
9819 * set BTRFS_DC_CLEAR and set dirty flag.
9821 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9822 * truncate the old free space cache inode and
9824 * b) Setting 'dirty flag' makes sure that we flush
9825 * the new space cache info onto disk.
9827 if (btrfs_test_opt(root
, SPACE_CACHE
))
9828 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
9831 read_extent_buffer(leaf
, &cache
->item
,
9832 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
9833 sizeof(cache
->item
));
9834 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
9836 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
9837 btrfs_release_path(path
);
9840 * We need to exclude the super stripes now so that the space
9841 * info has super bytes accounted for, otherwise we'll think
9842 * we have more space than we actually do.
9844 ret
= exclude_super_stripes(root
, cache
);
9847 * We may have excluded something, so call this just in
9850 free_excluded_extents(root
, cache
);
9851 btrfs_put_block_group(cache
);
9856 * check for two cases, either we are full, and therefore
9857 * don't need to bother with the caching work since we won't
9858 * find any space, or we are empty, and we can just add all
9859 * the space in and be done with it. This saves us _alot_ of
9860 * time, particularly in the full case.
9862 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
9863 cache
->last_byte_to_unpin
= (u64
)-1;
9864 cache
->cached
= BTRFS_CACHE_FINISHED
;
9865 free_excluded_extents(root
, cache
);
9866 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
9867 cache
->last_byte_to_unpin
= (u64
)-1;
9868 cache
->cached
= BTRFS_CACHE_FINISHED
;
9869 add_new_free_space(cache
, root
->fs_info
,
9871 found_key
.objectid
+
9873 free_excluded_extents(root
, cache
);
9876 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
9878 btrfs_remove_free_space_cache(cache
);
9879 btrfs_put_block_group(cache
);
9883 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
9884 btrfs_block_group_used(&cache
->item
),
9887 btrfs_remove_free_space_cache(cache
);
9888 spin_lock(&info
->block_group_cache_lock
);
9889 rb_erase(&cache
->cache_node
,
9890 &info
->block_group_cache_tree
);
9891 RB_CLEAR_NODE(&cache
->cache_node
);
9892 spin_unlock(&info
->block_group_cache_lock
);
9893 btrfs_put_block_group(cache
);
9897 cache
->space_info
= space_info
;
9898 spin_lock(&cache
->space_info
->lock
);
9899 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
9900 spin_unlock(&cache
->space_info
->lock
);
9902 __link_block_group(space_info
, cache
);
9904 set_avail_alloc_bits(root
->fs_info
, cache
->flags
);
9905 if (btrfs_chunk_readonly(root
, cache
->key
.objectid
)) {
9906 inc_block_group_ro(cache
, 1);
9907 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
9908 spin_lock(&info
->unused_bgs_lock
);
9909 /* Should always be true but just in case. */
9910 if (list_empty(&cache
->bg_list
)) {
9911 btrfs_get_block_group(cache
);
9912 list_add_tail(&cache
->bg_list
,
9915 spin_unlock(&info
->unused_bgs_lock
);
9919 list_for_each_entry_rcu(space_info
, &root
->fs_info
->space_info
, list
) {
9920 if (!(get_alloc_profile(root
, space_info
->flags
) &
9921 (BTRFS_BLOCK_GROUP_RAID10
|
9922 BTRFS_BLOCK_GROUP_RAID1
|
9923 BTRFS_BLOCK_GROUP_RAID5
|
9924 BTRFS_BLOCK_GROUP_RAID6
|
9925 BTRFS_BLOCK_GROUP_DUP
)))
9928 * avoid allocating from un-mirrored block group if there are
9929 * mirrored block groups.
9931 list_for_each_entry(cache
,
9932 &space_info
->block_groups
[BTRFS_RAID_RAID0
],
9934 inc_block_group_ro(cache
, 1);
9935 list_for_each_entry(cache
,
9936 &space_info
->block_groups
[BTRFS_RAID_SINGLE
],
9938 inc_block_group_ro(cache
, 1);
9941 init_global_block_rsv(info
);
9944 btrfs_free_path(path
);
9948 void btrfs_create_pending_block_groups(struct btrfs_trans_handle
*trans
,
9949 struct btrfs_root
*root
)
9951 struct btrfs_block_group_cache
*block_group
, *tmp
;
9952 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
9953 struct btrfs_block_group_item item
;
9954 struct btrfs_key key
;
9956 bool can_flush_pending_bgs
= trans
->can_flush_pending_bgs
;
9958 trans
->can_flush_pending_bgs
= false;
9959 list_for_each_entry_safe(block_group
, tmp
, &trans
->new_bgs
, bg_list
) {
9963 spin_lock(&block_group
->lock
);
9964 memcpy(&item
, &block_group
->item
, sizeof(item
));
9965 memcpy(&key
, &block_group
->key
, sizeof(key
));
9966 spin_unlock(&block_group
->lock
);
9968 ret
= btrfs_insert_item(trans
, extent_root
, &key
, &item
,
9971 btrfs_abort_transaction(trans
, extent_root
, ret
);
9972 ret
= btrfs_finish_chunk_alloc(trans
, extent_root
,
9973 key
.objectid
, key
.offset
);
9975 btrfs_abort_transaction(trans
, extent_root
, ret
);
9976 add_block_group_free_space(trans
, root
->fs_info
, block_group
);
9977 /* already aborted the transaction if it failed. */
9979 list_del_init(&block_group
->bg_list
);
9981 trans
->can_flush_pending_bgs
= can_flush_pending_bgs
;
9984 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
9985 struct btrfs_root
*root
, u64 bytes_used
,
9986 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
9990 struct btrfs_root
*extent_root
;
9991 struct btrfs_block_group_cache
*cache
;
9993 extent_root
= root
->fs_info
->extent_root
;
9995 btrfs_set_log_full_commit(root
->fs_info
, trans
);
9997 cache
= btrfs_create_block_group_cache(root
, chunk_offset
, size
);
10001 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
10002 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
10003 btrfs_set_block_group_flags(&cache
->item
, type
);
10005 cache
->flags
= type
;
10006 cache
->last_byte_to_unpin
= (u64
)-1;
10007 cache
->cached
= BTRFS_CACHE_FINISHED
;
10008 cache
->needs_free_space
= 1;
10009 ret
= exclude_super_stripes(root
, cache
);
10012 * We may have excluded something, so call this just in
10015 free_excluded_extents(root
, cache
);
10016 btrfs_put_block_group(cache
);
10020 add_new_free_space(cache
, root
->fs_info
, chunk_offset
,
10021 chunk_offset
+ size
);
10023 free_excluded_extents(root
, cache
);
10025 #ifdef CONFIG_BTRFS_DEBUG
10026 if (btrfs_should_fragment_free_space(root
, cache
)) {
10027 u64 new_bytes_used
= size
- bytes_used
;
10029 bytes_used
+= new_bytes_used
>> 1;
10030 fragment_free_space(root
, cache
);
10034 * Call to ensure the corresponding space_info object is created and
10035 * assigned to our block group, but don't update its counters just yet.
10036 * We want our bg to be added to the rbtree with its ->space_info set.
10038 ret
= update_space_info(root
->fs_info
, cache
->flags
, 0, 0,
10039 &cache
->space_info
);
10041 btrfs_remove_free_space_cache(cache
);
10042 btrfs_put_block_group(cache
);
10046 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
10048 btrfs_remove_free_space_cache(cache
);
10049 btrfs_put_block_group(cache
);
10054 * Now that our block group has its ->space_info set and is inserted in
10055 * the rbtree, update the space info's counters.
10057 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
10058 &cache
->space_info
);
10060 btrfs_remove_free_space_cache(cache
);
10061 spin_lock(&root
->fs_info
->block_group_cache_lock
);
10062 rb_erase(&cache
->cache_node
,
10063 &root
->fs_info
->block_group_cache_tree
);
10064 RB_CLEAR_NODE(&cache
->cache_node
);
10065 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
10066 btrfs_put_block_group(cache
);
10069 update_global_block_rsv(root
->fs_info
);
10071 spin_lock(&cache
->space_info
->lock
);
10072 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
10073 spin_unlock(&cache
->space_info
->lock
);
10075 __link_block_group(cache
->space_info
, cache
);
10077 list_add_tail(&cache
->bg_list
, &trans
->new_bgs
);
10079 set_avail_alloc_bits(extent_root
->fs_info
, type
);
10084 static void clear_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
10086 u64 extra_flags
= chunk_to_extended(flags
) &
10087 BTRFS_EXTENDED_PROFILE_MASK
;
10089 write_seqlock(&fs_info
->profiles_lock
);
10090 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
10091 fs_info
->avail_data_alloc_bits
&= ~extra_flags
;
10092 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
10093 fs_info
->avail_metadata_alloc_bits
&= ~extra_flags
;
10094 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
10095 fs_info
->avail_system_alloc_bits
&= ~extra_flags
;
10096 write_sequnlock(&fs_info
->profiles_lock
);
10099 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
10100 struct btrfs_root
*root
, u64 group_start
,
10101 struct extent_map
*em
)
10103 struct btrfs_path
*path
;
10104 struct btrfs_block_group_cache
*block_group
;
10105 struct btrfs_free_cluster
*cluster
;
10106 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
10107 struct btrfs_key key
;
10108 struct inode
*inode
;
10109 struct kobject
*kobj
= NULL
;
10113 struct btrfs_caching_control
*caching_ctl
= NULL
;
10116 root
= root
->fs_info
->extent_root
;
10118 block_group
= btrfs_lookup_block_group(root
->fs_info
, group_start
);
10119 BUG_ON(!block_group
);
10120 BUG_ON(!block_group
->ro
);
10123 * Free the reserved super bytes from this block group before
10126 free_excluded_extents(root
, block_group
);
10128 memcpy(&key
, &block_group
->key
, sizeof(key
));
10129 index
= get_block_group_index(block_group
);
10130 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
10131 BTRFS_BLOCK_GROUP_RAID1
|
10132 BTRFS_BLOCK_GROUP_RAID10
))
10137 /* make sure this block group isn't part of an allocation cluster */
10138 cluster
= &root
->fs_info
->data_alloc_cluster
;
10139 spin_lock(&cluster
->refill_lock
);
10140 btrfs_return_cluster_to_free_space(block_group
, cluster
);
10141 spin_unlock(&cluster
->refill_lock
);
10144 * make sure this block group isn't part of a metadata
10145 * allocation cluster
10147 cluster
= &root
->fs_info
->meta_alloc_cluster
;
10148 spin_lock(&cluster
->refill_lock
);
10149 btrfs_return_cluster_to_free_space(block_group
, cluster
);
10150 spin_unlock(&cluster
->refill_lock
);
10152 path
= btrfs_alloc_path();
10159 * get the inode first so any iput calls done for the io_list
10160 * aren't the final iput (no unlinks allowed now)
10162 inode
= lookup_free_space_inode(tree_root
, block_group
, path
);
10164 mutex_lock(&trans
->transaction
->cache_write_mutex
);
10166 * make sure our free spache cache IO is done before remove the
10169 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
10170 if (!list_empty(&block_group
->io_list
)) {
10171 list_del_init(&block_group
->io_list
);
10173 WARN_ON(!IS_ERR(inode
) && inode
!= block_group
->io_ctl
.inode
);
10175 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
10176 btrfs_wait_cache_io(root
, trans
, block_group
,
10177 &block_group
->io_ctl
, path
,
10178 block_group
->key
.objectid
);
10179 btrfs_put_block_group(block_group
);
10180 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
10183 if (!list_empty(&block_group
->dirty_list
)) {
10184 list_del_init(&block_group
->dirty_list
);
10185 btrfs_put_block_group(block_group
);
10187 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
10188 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
10190 if (!IS_ERR(inode
)) {
10191 ret
= btrfs_orphan_add(trans
, inode
);
10193 btrfs_add_delayed_iput(inode
);
10196 clear_nlink(inode
);
10197 /* One for the block groups ref */
10198 spin_lock(&block_group
->lock
);
10199 if (block_group
->iref
) {
10200 block_group
->iref
= 0;
10201 block_group
->inode
= NULL
;
10202 spin_unlock(&block_group
->lock
);
10205 spin_unlock(&block_group
->lock
);
10207 /* One for our lookup ref */
10208 btrfs_add_delayed_iput(inode
);
10211 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
10212 key
.offset
= block_group
->key
.objectid
;
10215 ret
= btrfs_search_slot(trans
, tree_root
, &key
, path
, -1, 1);
10219 btrfs_release_path(path
);
10221 ret
= btrfs_del_item(trans
, tree_root
, path
);
10224 btrfs_release_path(path
);
10227 spin_lock(&root
->fs_info
->block_group_cache_lock
);
10228 rb_erase(&block_group
->cache_node
,
10229 &root
->fs_info
->block_group_cache_tree
);
10230 RB_CLEAR_NODE(&block_group
->cache_node
);
10232 if (root
->fs_info
->first_logical_byte
== block_group
->key
.objectid
)
10233 root
->fs_info
->first_logical_byte
= (u64
)-1;
10234 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
10236 down_write(&block_group
->space_info
->groups_sem
);
10238 * we must use list_del_init so people can check to see if they
10239 * are still on the list after taking the semaphore
10241 list_del_init(&block_group
->list
);
10242 if (list_empty(&block_group
->space_info
->block_groups
[index
])) {
10243 kobj
= block_group
->space_info
->block_group_kobjs
[index
];
10244 block_group
->space_info
->block_group_kobjs
[index
] = NULL
;
10245 clear_avail_alloc_bits(root
->fs_info
, block_group
->flags
);
10247 up_write(&block_group
->space_info
->groups_sem
);
10253 if (block_group
->has_caching_ctl
)
10254 caching_ctl
= get_caching_control(block_group
);
10255 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
10256 wait_block_group_cache_done(block_group
);
10257 if (block_group
->has_caching_ctl
) {
10258 down_write(&root
->fs_info
->commit_root_sem
);
10259 if (!caching_ctl
) {
10260 struct btrfs_caching_control
*ctl
;
10262 list_for_each_entry(ctl
,
10263 &root
->fs_info
->caching_block_groups
, list
)
10264 if (ctl
->block_group
== block_group
) {
10266 atomic_inc(&caching_ctl
->count
);
10271 list_del_init(&caching_ctl
->list
);
10272 up_write(&root
->fs_info
->commit_root_sem
);
10274 /* Once for the caching bgs list and once for us. */
10275 put_caching_control(caching_ctl
);
10276 put_caching_control(caching_ctl
);
10280 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
10281 if (!list_empty(&block_group
->dirty_list
)) {
10284 if (!list_empty(&block_group
->io_list
)) {
10287 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
10288 btrfs_remove_free_space_cache(block_group
);
10290 spin_lock(&block_group
->space_info
->lock
);
10291 list_del_init(&block_group
->ro_list
);
10293 if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
10294 WARN_ON(block_group
->space_info
->total_bytes
10295 < block_group
->key
.offset
);
10296 WARN_ON(block_group
->space_info
->bytes_readonly
10297 < block_group
->key
.offset
);
10298 WARN_ON(block_group
->space_info
->disk_total
10299 < block_group
->key
.offset
* factor
);
10301 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
10302 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
10303 block_group
->space_info
->disk_total
-= block_group
->key
.offset
* factor
;
10305 spin_unlock(&block_group
->space_info
->lock
);
10307 memcpy(&key
, &block_group
->key
, sizeof(key
));
10310 if (!list_empty(&em
->list
)) {
10311 /* We're in the transaction->pending_chunks list. */
10312 free_extent_map(em
);
10314 spin_lock(&block_group
->lock
);
10315 block_group
->removed
= 1;
10317 * At this point trimming can't start on this block group, because we
10318 * removed the block group from the tree fs_info->block_group_cache_tree
10319 * so no one can't find it anymore and even if someone already got this
10320 * block group before we removed it from the rbtree, they have already
10321 * incremented block_group->trimming - if they didn't, they won't find
10322 * any free space entries because we already removed them all when we
10323 * called btrfs_remove_free_space_cache().
10325 * And we must not remove the extent map from the fs_info->mapping_tree
10326 * to prevent the same logical address range and physical device space
10327 * ranges from being reused for a new block group. This is because our
10328 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10329 * completely transactionless, so while it is trimming a range the
10330 * currently running transaction might finish and a new one start,
10331 * allowing for new block groups to be created that can reuse the same
10332 * physical device locations unless we take this special care.
10334 * There may also be an implicit trim operation if the file system
10335 * is mounted with -odiscard. The same protections must remain
10336 * in place until the extents have been discarded completely when
10337 * the transaction commit has completed.
10339 remove_em
= (atomic_read(&block_group
->trimming
) == 0);
10341 * Make sure a trimmer task always sees the em in the pinned_chunks list
10342 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10343 * before checking block_group->removed).
10347 * Our em might be in trans->transaction->pending_chunks which
10348 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10349 * and so is the fs_info->pinned_chunks list.
10351 * So at this point we must be holding the chunk_mutex to avoid
10352 * any races with chunk allocation (more specifically at
10353 * volumes.c:contains_pending_extent()), to ensure it always
10354 * sees the em, either in the pending_chunks list or in the
10355 * pinned_chunks list.
10357 list_move_tail(&em
->list
, &root
->fs_info
->pinned_chunks
);
10359 spin_unlock(&block_group
->lock
);
10362 struct extent_map_tree
*em_tree
;
10364 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
10365 write_lock(&em_tree
->lock
);
10367 * The em might be in the pending_chunks list, so make sure the
10368 * chunk mutex is locked, since remove_extent_mapping() will
10369 * delete us from that list.
10371 remove_extent_mapping(em_tree
, em
);
10372 write_unlock(&em_tree
->lock
);
10373 /* once for the tree */
10374 free_extent_map(em
);
10377 unlock_chunks(root
);
10379 ret
= remove_block_group_free_space(trans
, root
->fs_info
, block_group
);
10383 btrfs_put_block_group(block_group
);
10384 btrfs_put_block_group(block_group
);
10386 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
10392 ret
= btrfs_del_item(trans
, root
, path
);
10394 btrfs_free_path(path
);
10398 struct btrfs_trans_handle
*
10399 btrfs_start_trans_remove_block_group(struct btrfs_fs_info
*fs_info
,
10400 const u64 chunk_offset
)
10402 struct extent_map_tree
*em_tree
= &fs_info
->mapping_tree
.map_tree
;
10403 struct extent_map
*em
;
10404 struct map_lookup
*map
;
10405 unsigned int num_items
;
10407 read_lock(&em_tree
->lock
);
10408 em
= lookup_extent_mapping(em_tree
, chunk_offset
, 1);
10409 read_unlock(&em_tree
->lock
);
10410 ASSERT(em
&& em
->start
== chunk_offset
);
10413 * We need to reserve 3 + N units from the metadata space info in order
10414 * to remove a block group (done at btrfs_remove_chunk() and at
10415 * btrfs_remove_block_group()), which are used for:
10417 * 1 unit for adding the free space inode's orphan (located in the tree
10419 * 1 unit for deleting the block group item (located in the extent
10421 * 1 unit for deleting the free space item (located in tree of tree
10423 * N units for deleting N device extent items corresponding to each
10424 * stripe (located in the device tree).
10426 * In order to remove a block group we also need to reserve units in the
10427 * system space info in order to update the chunk tree (update one or
10428 * more device items and remove one chunk item), but this is done at
10429 * btrfs_remove_chunk() through a call to check_system_chunk().
10431 map
= em
->map_lookup
;
10432 num_items
= 3 + map
->num_stripes
;
10433 free_extent_map(em
);
10435 return btrfs_start_transaction_fallback_global_rsv(fs_info
->extent_root
,
10440 * Process the unused_bgs list and remove any that don't have any allocated
10441 * space inside of them.
10443 void btrfs_delete_unused_bgs(struct btrfs_fs_info
*fs_info
)
10445 struct btrfs_block_group_cache
*block_group
;
10446 struct btrfs_space_info
*space_info
;
10447 struct btrfs_root
*root
= fs_info
->extent_root
;
10448 struct btrfs_trans_handle
*trans
;
10451 if (!fs_info
->open
)
10454 spin_lock(&fs_info
->unused_bgs_lock
);
10455 while (!list_empty(&fs_info
->unused_bgs
)) {
10459 block_group
= list_first_entry(&fs_info
->unused_bgs
,
10460 struct btrfs_block_group_cache
,
10462 list_del_init(&block_group
->bg_list
);
10464 space_info
= block_group
->space_info
;
10466 if (ret
|| btrfs_mixed_space_info(space_info
)) {
10467 btrfs_put_block_group(block_group
);
10470 spin_unlock(&fs_info
->unused_bgs_lock
);
10472 mutex_lock(&fs_info
->delete_unused_bgs_mutex
);
10474 /* Don't want to race with allocators so take the groups_sem */
10475 down_write(&space_info
->groups_sem
);
10476 spin_lock(&block_group
->lock
);
10477 if (block_group
->reserved
||
10478 btrfs_block_group_used(&block_group
->item
) ||
10480 list_is_singular(&block_group
->list
)) {
10482 * We want to bail if we made new allocations or have
10483 * outstanding allocations in this block group. We do
10484 * the ro check in case balance is currently acting on
10485 * this block group.
10487 spin_unlock(&block_group
->lock
);
10488 up_write(&space_info
->groups_sem
);
10491 spin_unlock(&block_group
->lock
);
10493 /* We don't want to force the issue, only flip if it's ok. */
10494 ret
= inc_block_group_ro(block_group
, 0);
10495 up_write(&space_info
->groups_sem
);
10502 * Want to do this before we do anything else so we can recover
10503 * properly if we fail to join the transaction.
10505 trans
= btrfs_start_trans_remove_block_group(fs_info
,
10506 block_group
->key
.objectid
);
10507 if (IS_ERR(trans
)) {
10508 btrfs_dec_block_group_ro(root
, block_group
);
10509 ret
= PTR_ERR(trans
);
10514 * We could have pending pinned extents for this block group,
10515 * just delete them, we don't care about them anymore.
10517 start
= block_group
->key
.objectid
;
10518 end
= start
+ block_group
->key
.offset
- 1;
10520 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10521 * btrfs_finish_extent_commit(). If we are at transaction N,
10522 * another task might be running finish_extent_commit() for the
10523 * previous transaction N - 1, and have seen a range belonging
10524 * to the block group in freed_extents[] before we were able to
10525 * clear the whole block group range from freed_extents[]. This
10526 * means that task can lookup for the block group after we
10527 * unpinned it from freed_extents[] and removed it, leading to
10528 * a BUG_ON() at btrfs_unpin_extent_range().
10530 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
10531 ret
= clear_extent_bits(&fs_info
->freed_extents
[0], start
, end
,
10532 EXTENT_DIRTY
, GFP_NOFS
);
10534 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10535 btrfs_dec_block_group_ro(root
, block_group
);
10538 ret
= clear_extent_bits(&fs_info
->freed_extents
[1], start
, end
,
10539 EXTENT_DIRTY
, GFP_NOFS
);
10541 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10542 btrfs_dec_block_group_ro(root
, block_group
);
10545 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10547 /* Reset pinned so btrfs_put_block_group doesn't complain */
10548 spin_lock(&space_info
->lock
);
10549 spin_lock(&block_group
->lock
);
10551 space_info
->bytes_pinned
-= block_group
->pinned
;
10552 space_info
->bytes_readonly
+= block_group
->pinned
;
10553 percpu_counter_add(&space_info
->total_bytes_pinned
,
10554 -block_group
->pinned
);
10555 block_group
->pinned
= 0;
10557 spin_unlock(&block_group
->lock
);
10558 spin_unlock(&space_info
->lock
);
10560 /* DISCARD can flip during remount */
10561 trimming
= btrfs_test_opt(root
, DISCARD
);
10563 /* Implicit trim during transaction commit. */
10565 btrfs_get_block_group_trimming(block_group
);
10568 * Btrfs_remove_chunk will abort the transaction if things go
10571 ret
= btrfs_remove_chunk(trans
, root
,
10572 block_group
->key
.objectid
);
10576 btrfs_put_block_group_trimming(block_group
);
10581 * If we're not mounted with -odiscard, we can just forget
10582 * about this block group. Otherwise we'll need to wait
10583 * until transaction commit to do the actual discard.
10586 spin_lock(&fs_info
->unused_bgs_lock
);
10588 * A concurrent scrub might have added us to the list
10589 * fs_info->unused_bgs, so use a list_move operation
10590 * to add the block group to the deleted_bgs list.
10592 list_move(&block_group
->bg_list
,
10593 &trans
->transaction
->deleted_bgs
);
10594 spin_unlock(&fs_info
->unused_bgs_lock
);
10595 btrfs_get_block_group(block_group
);
10598 btrfs_end_transaction(trans
, root
);
10600 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
10601 btrfs_put_block_group(block_group
);
10602 spin_lock(&fs_info
->unused_bgs_lock
);
10604 spin_unlock(&fs_info
->unused_bgs_lock
);
10607 int btrfs_init_space_info(struct btrfs_fs_info
*fs_info
)
10609 struct btrfs_space_info
*space_info
;
10610 struct btrfs_super_block
*disk_super
;
10616 disk_super
= fs_info
->super_copy
;
10617 if (!btrfs_super_root(disk_super
))
10620 features
= btrfs_super_incompat_flags(disk_super
);
10621 if (features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
10624 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
10625 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
10630 flags
= BTRFS_BLOCK_GROUP_METADATA
| BTRFS_BLOCK_GROUP_DATA
;
10631 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
10633 flags
= BTRFS_BLOCK_GROUP_METADATA
;
10634 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
10638 flags
= BTRFS_BLOCK_GROUP_DATA
;
10639 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
10645 int btrfs_error_unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
10647 return unpin_extent_range(root
, start
, end
, false);
10651 * It used to be that old block groups would be left around forever.
10652 * Iterating over them would be enough to trim unused space. Since we
10653 * now automatically remove them, we also need to iterate over unallocated
10656 * We don't want a transaction for this since the discard may take a
10657 * substantial amount of time. We don't require that a transaction be
10658 * running, but we do need to take a running transaction into account
10659 * to ensure that we're not discarding chunks that were released in
10660 * the current transaction.
10662 * Holding the chunks lock will prevent other threads from allocating
10663 * or releasing chunks, but it won't prevent a running transaction
10664 * from committing and releasing the memory that the pending chunks
10665 * list head uses. For that, we need to take a reference to the
10668 static int btrfs_trim_free_extents(struct btrfs_device
*device
,
10669 u64 minlen
, u64
*trimmed
)
10671 u64 start
= 0, len
= 0;
10676 /* Not writeable = nothing to do. */
10677 if (!device
->writeable
)
10680 /* No free space = nothing to do. */
10681 if (device
->total_bytes
<= device
->bytes_used
)
10687 struct btrfs_fs_info
*fs_info
= device
->dev_root
->fs_info
;
10688 struct btrfs_transaction
*trans
;
10691 ret
= mutex_lock_interruptible(&fs_info
->chunk_mutex
);
10695 down_read(&fs_info
->commit_root_sem
);
10697 spin_lock(&fs_info
->trans_lock
);
10698 trans
= fs_info
->running_transaction
;
10700 atomic_inc(&trans
->use_count
);
10701 spin_unlock(&fs_info
->trans_lock
);
10703 ret
= find_free_dev_extent_start(trans
, device
, minlen
, start
,
10706 btrfs_put_transaction(trans
);
10709 up_read(&fs_info
->commit_root_sem
);
10710 mutex_unlock(&fs_info
->chunk_mutex
);
10711 if (ret
== -ENOSPC
)
10716 ret
= btrfs_issue_discard(device
->bdev
, start
, len
, &bytes
);
10717 up_read(&fs_info
->commit_root_sem
);
10718 mutex_unlock(&fs_info
->chunk_mutex
);
10726 if (fatal_signal_pending(current
)) {
10727 ret
= -ERESTARTSYS
;
10737 int btrfs_trim_fs(struct btrfs_root
*root
, struct fstrim_range
*range
)
10739 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
10740 struct btrfs_block_group_cache
*cache
= NULL
;
10741 struct btrfs_device
*device
;
10742 struct list_head
*devices
;
10747 u64 total_bytes
= btrfs_super_total_bytes(fs_info
->super_copy
);
10751 * try to trim all FS space, our block group may start from non-zero.
10753 if (range
->len
== total_bytes
)
10754 cache
= btrfs_lookup_first_block_group(fs_info
, range
->start
);
10756 cache
= btrfs_lookup_block_group(fs_info
, range
->start
);
10759 if (cache
->key
.objectid
>= (range
->start
+ range
->len
)) {
10760 btrfs_put_block_group(cache
);
10764 start
= max(range
->start
, cache
->key
.objectid
);
10765 end
= min(range
->start
+ range
->len
,
10766 cache
->key
.objectid
+ cache
->key
.offset
);
10768 if (end
- start
>= range
->minlen
) {
10769 if (!block_group_cache_done(cache
)) {
10770 ret
= cache_block_group(cache
, 0);
10772 btrfs_put_block_group(cache
);
10775 ret
= wait_block_group_cache_done(cache
);
10777 btrfs_put_block_group(cache
);
10781 ret
= btrfs_trim_block_group(cache
,
10787 trimmed
+= group_trimmed
;
10789 btrfs_put_block_group(cache
);
10794 cache
= next_block_group(fs_info
->tree_root
, cache
);
10797 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
10798 devices
= &root
->fs_info
->fs_devices
->alloc_list
;
10799 list_for_each_entry(device
, devices
, dev_alloc_list
) {
10800 ret
= btrfs_trim_free_extents(device
, range
->minlen
,
10805 trimmed
+= group_trimmed
;
10807 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
10809 range
->len
= trimmed
;
10814 * btrfs_{start,end}_write_no_snapshoting() are similar to
10815 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10816 * data into the page cache through nocow before the subvolume is snapshoted,
10817 * but flush the data into disk after the snapshot creation, or to prevent
10818 * operations while snapshoting is ongoing and that cause the snapshot to be
10819 * inconsistent (writes followed by expanding truncates for example).
10821 void btrfs_end_write_no_snapshoting(struct btrfs_root
*root
)
10823 percpu_counter_dec(&root
->subv_writers
->counter
);
10825 * Make sure counter is updated before we wake up waiters.
10828 if (waitqueue_active(&root
->subv_writers
->wait
))
10829 wake_up(&root
->subv_writers
->wait
);
10832 int btrfs_start_write_no_snapshoting(struct btrfs_root
*root
)
10834 if (atomic_read(&root
->will_be_snapshoted
))
10837 percpu_counter_inc(&root
->subv_writers
->counter
);
10839 * Make sure counter is updated before we check for snapshot creation.
10842 if (atomic_read(&root
->will_be_snapshoted
)) {
10843 btrfs_end_write_no_snapshoting(root
);
10849 static int wait_snapshoting_atomic_t(atomic_t
*a
)
10855 void btrfs_wait_for_snapshot_creation(struct btrfs_root
*root
)
10860 ret
= btrfs_start_write_no_snapshoting(root
);
10863 wait_on_atomic_t(&root
->will_be_snapshoted
,
10864 wait_snapshoting_atomic_t
,
10865 TASK_UNINTERRUPTIBLE
);