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
);
235 set_extent_bits(&root
->fs_info
->freed_extents
[1],
236 start
, end
, EXTENT_UPTODATE
);
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
);
250 clear_extent_bits(&root
->fs_info
->freed_extents
[1],
251 start
, end
, EXTENT_UPTODATE
);
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 COWed 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
;
2046 * Avoid races with device replace and make sure our bbio has devices
2047 * associated to its stripes that don't go away while we are discarding.
2049 btrfs_bio_counter_inc_blocked(root
->fs_info
);
2050 /* Tell the block device(s) that the sectors can be discarded */
2051 ret
= btrfs_map_block(root
->fs_info
, REQ_DISCARD
,
2052 bytenr
, &num_bytes
, &bbio
, 0);
2053 /* Error condition is -ENOMEM */
2055 struct btrfs_bio_stripe
*stripe
= bbio
->stripes
;
2059 for (i
= 0; i
< bbio
->num_stripes
; i
++, stripe
++) {
2061 if (!stripe
->dev
->can_discard
)
2064 ret
= btrfs_issue_discard(stripe
->dev
->bdev
,
2069 discarded_bytes
+= bytes
;
2070 else if (ret
!= -EOPNOTSUPP
)
2071 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2074 * Just in case we get back EOPNOTSUPP for some reason,
2075 * just ignore the return value so we don't screw up
2076 * people calling discard_extent.
2080 btrfs_put_bbio(bbio
);
2082 btrfs_bio_counter_dec(root
->fs_info
);
2085 *actual_bytes
= discarded_bytes
;
2088 if (ret
== -EOPNOTSUPP
)
2093 /* Can return -ENOMEM */
2094 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
2095 struct btrfs_root
*root
,
2096 u64 bytenr
, u64 num_bytes
, u64 parent
,
2097 u64 root_objectid
, u64 owner
, u64 offset
)
2100 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2102 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
&&
2103 root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
2105 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
2106 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
2108 parent
, root_objectid
, (int)owner
,
2109 BTRFS_ADD_DELAYED_REF
, NULL
);
2111 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
2112 num_bytes
, parent
, root_objectid
,
2114 BTRFS_ADD_DELAYED_REF
, NULL
);
2119 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
2120 struct btrfs_root
*root
,
2121 struct btrfs_delayed_ref_node
*node
,
2122 u64 parent
, u64 root_objectid
,
2123 u64 owner
, u64 offset
, int refs_to_add
,
2124 struct btrfs_delayed_extent_op
*extent_op
)
2126 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2127 struct btrfs_path
*path
;
2128 struct extent_buffer
*leaf
;
2129 struct btrfs_extent_item
*item
;
2130 struct btrfs_key key
;
2131 u64 bytenr
= node
->bytenr
;
2132 u64 num_bytes
= node
->num_bytes
;
2136 path
= btrfs_alloc_path();
2140 path
->reada
= READA_FORWARD
;
2141 path
->leave_spinning
= 1;
2142 /* this will setup the path even if it fails to insert the back ref */
2143 ret
= insert_inline_extent_backref(trans
, fs_info
->extent_root
, path
,
2144 bytenr
, num_bytes
, parent
,
2145 root_objectid
, owner
, offset
,
2146 refs_to_add
, extent_op
);
2147 if ((ret
< 0 && ret
!= -EAGAIN
) || !ret
)
2151 * Ok we had -EAGAIN which means we didn't have space to insert and
2152 * inline extent ref, so just update the reference count and add a
2155 leaf
= path
->nodes
[0];
2156 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2157 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2158 refs
= btrfs_extent_refs(leaf
, item
);
2159 btrfs_set_extent_refs(leaf
, item
, refs
+ refs_to_add
);
2161 __run_delayed_extent_op(extent_op
, leaf
, item
);
2163 btrfs_mark_buffer_dirty(leaf
);
2164 btrfs_release_path(path
);
2166 path
->reada
= READA_FORWARD
;
2167 path
->leave_spinning
= 1;
2168 /* now insert the actual backref */
2169 ret
= insert_extent_backref(trans
, root
->fs_info
->extent_root
,
2170 path
, bytenr
, parent
, root_objectid
,
2171 owner
, offset
, refs_to_add
);
2173 btrfs_abort_transaction(trans
, root
, ret
);
2175 btrfs_free_path(path
);
2179 static int run_delayed_data_ref(struct btrfs_trans_handle
*trans
,
2180 struct btrfs_root
*root
,
2181 struct btrfs_delayed_ref_node
*node
,
2182 struct btrfs_delayed_extent_op
*extent_op
,
2183 int insert_reserved
)
2186 struct btrfs_delayed_data_ref
*ref
;
2187 struct btrfs_key ins
;
2192 ins
.objectid
= node
->bytenr
;
2193 ins
.offset
= node
->num_bytes
;
2194 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2196 ref
= btrfs_delayed_node_to_data_ref(node
);
2197 trace_run_delayed_data_ref(node
, ref
, node
->action
);
2199 if (node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2200 parent
= ref
->parent
;
2201 ref_root
= ref
->root
;
2203 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2205 flags
|= extent_op
->flags_to_set
;
2206 ret
= alloc_reserved_file_extent(trans
, root
,
2207 parent
, ref_root
, flags
,
2208 ref
->objectid
, ref
->offset
,
2209 &ins
, node
->ref_mod
);
2210 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2211 ret
= __btrfs_inc_extent_ref(trans
, root
, node
, parent
,
2212 ref_root
, ref
->objectid
,
2213 ref
->offset
, node
->ref_mod
,
2215 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2216 ret
= __btrfs_free_extent(trans
, root
, node
, parent
,
2217 ref_root
, ref
->objectid
,
2218 ref
->offset
, node
->ref_mod
,
2226 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
2227 struct extent_buffer
*leaf
,
2228 struct btrfs_extent_item
*ei
)
2230 u64 flags
= btrfs_extent_flags(leaf
, ei
);
2231 if (extent_op
->update_flags
) {
2232 flags
|= extent_op
->flags_to_set
;
2233 btrfs_set_extent_flags(leaf
, ei
, flags
);
2236 if (extent_op
->update_key
) {
2237 struct btrfs_tree_block_info
*bi
;
2238 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
));
2239 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
2240 btrfs_set_tree_block_key(leaf
, bi
, &extent_op
->key
);
2244 static int run_delayed_extent_op(struct btrfs_trans_handle
*trans
,
2245 struct btrfs_root
*root
,
2246 struct btrfs_delayed_ref_node
*node
,
2247 struct btrfs_delayed_extent_op
*extent_op
)
2249 struct btrfs_key key
;
2250 struct btrfs_path
*path
;
2251 struct btrfs_extent_item
*ei
;
2252 struct extent_buffer
*leaf
;
2256 int metadata
= !extent_op
->is_data
;
2261 if (metadata
&& !btrfs_fs_incompat(root
->fs_info
, SKINNY_METADATA
))
2264 path
= btrfs_alloc_path();
2268 key
.objectid
= node
->bytenr
;
2271 key
.type
= BTRFS_METADATA_ITEM_KEY
;
2272 key
.offset
= extent_op
->level
;
2274 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2275 key
.offset
= node
->num_bytes
;
2279 path
->reada
= READA_FORWARD
;
2280 path
->leave_spinning
= 1;
2281 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
,
2289 if (path
->slots
[0] > 0) {
2291 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
2293 if (key
.objectid
== node
->bytenr
&&
2294 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
2295 key
.offset
== node
->num_bytes
)
2299 btrfs_release_path(path
);
2302 key
.objectid
= node
->bytenr
;
2303 key
.offset
= node
->num_bytes
;
2304 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2313 leaf
= path
->nodes
[0];
2314 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2315 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2316 if (item_size
< sizeof(*ei
)) {
2317 ret
= convert_extent_item_v0(trans
, root
->fs_info
->extent_root
,
2323 leaf
= path
->nodes
[0];
2324 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2327 BUG_ON(item_size
< sizeof(*ei
));
2328 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2329 __run_delayed_extent_op(extent_op
, leaf
, ei
);
2331 btrfs_mark_buffer_dirty(leaf
);
2333 btrfs_free_path(path
);
2337 static int run_delayed_tree_ref(struct btrfs_trans_handle
*trans
,
2338 struct btrfs_root
*root
,
2339 struct btrfs_delayed_ref_node
*node
,
2340 struct btrfs_delayed_extent_op
*extent_op
,
2341 int insert_reserved
)
2344 struct btrfs_delayed_tree_ref
*ref
;
2345 struct btrfs_key ins
;
2348 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
2351 ref
= btrfs_delayed_node_to_tree_ref(node
);
2352 trace_run_delayed_tree_ref(node
, ref
, node
->action
);
2354 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2355 parent
= ref
->parent
;
2356 ref_root
= ref
->root
;
2358 ins
.objectid
= node
->bytenr
;
2359 if (skinny_metadata
) {
2360 ins
.offset
= ref
->level
;
2361 ins
.type
= BTRFS_METADATA_ITEM_KEY
;
2363 ins
.offset
= node
->num_bytes
;
2364 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2367 BUG_ON(node
->ref_mod
!= 1);
2368 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2369 BUG_ON(!extent_op
|| !extent_op
->update_flags
);
2370 ret
= alloc_reserved_tree_block(trans
, root
,
2372 extent_op
->flags_to_set
,
2375 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2376 ret
= __btrfs_inc_extent_ref(trans
, root
, node
,
2380 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2381 ret
= __btrfs_free_extent(trans
, root
, node
,
2383 ref
->level
, 0, 1, extent_op
);
2390 /* helper function to actually process a single delayed ref entry */
2391 static int run_one_delayed_ref(struct btrfs_trans_handle
*trans
,
2392 struct btrfs_root
*root
,
2393 struct btrfs_delayed_ref_node
*node
,
2394 struct btrfs_delayed_extent_op
*extent_op
,
2395 int insert_reserved
)
2399 if (trans
->aborted
) {
2400 if (insert_reserved
)
2401 btrfs_pin_extent(root
, node
->bytenr
,
2402 node
->num_bytes
, 1);
2406 if (btrfs_delayed_ref_is_head(node
)) {
2407 struct btrfs_delayed_ref_head
*head
;
2409 * we've hit the end of the chain and we were supposed
2410 * to insert this extent into the tree. But, it got
2411 * deleted before we ever needed to insert it, so all
2412 * we have to do is clean up the accounting
2415 head
= btrfs_delayed_node_to_head(node
);
2416 trace_run_delayed_ref_head(node
, head
, node
->action
);
2418 if (insert_reserved
) {
2419 btrfs_pin_extent(root
, node
->bytenr
,
2420 node
->num_bytes
, 1);
2421 if (head
->is_data
) {
2422 ret
= btrfs_del_csums(trans
, root
,
2428 /* Also free its reserved qgroup space */
2429 btrfs_qgroup_free_delayed_ref(root
->fs_info
,
2430 head
->qgroup_ref_root
,
2431 head
->qgroup_reserved
);
2435 if (node
->type
== BTRFS_TREE_BLOCK_REF_KEY
||
2436 node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2437 ret
= run_delayed_tree_ref(trans
, root
, node
, extent_op
,
2439 else if (node
->type
== BTRFS_EXTENT_DATA_REF_KEY
||
2440 node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2441 ret
= run_delayed_data_ref(trans
, root
, node
, extent_op
,
2448 static inline struct btrfs_delayed_ref_node
*
2449 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
2451 struct btrfs_delayed_ref_node
*ref
;
2453 if (list_empty(&head
->ref_list
))
2457 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2458 * This is to prevent a ref count from going down to zero, which deletes
2459 * the extent item from the extent tree, when there still are references
2460 * to add, which would fail because they would not find the extent item.
2462 list_for_each_entry(ref
, &head
->ref_list
, list
) {
2463 if (ref
->action
== BTRFS_ADD_DELAYED_REF
)
2467 return list_entry(head
->ref_list
.next
, struct btrfs_delayed_ref_node
,
2472 * Returns 0 on success or if called with an already aborted transaction.
2473 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2475 static noinline
int __btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2476 struct btrfs_root
*root
,
2479 struct btrfs_delayed_ref_root
*delayed_refs
;
2480 struct btrfs_delayed_ref_node
*ref
;
2481 struct btrfs_delayed_ref_head
*locked_ref
= NULL
;
2482 struct btrfs_delayed_extent_op
*extent_op
;
2483 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2484 ktime_t start
= ktime_get();
2486 unsigned long count
= 0;
2487 unsigned long actual_count
= 0;
2488 int must_insert_reserved
= 0;
2490 delayed_refs
= &trans
->transaction
->delayed_refs
;
2496 spin_lock(&delayed_refs
->lock
);
2497 locked_ref
= btrfs_select_ref_head(trans
);
2499 spin_unlock(&delayed_refs
->lock
);
2503 /* grab the lock that says we are going to process
2504 * all the refs for this head */
2505 ret
= btrfs_delayed_ref_lock(trans
, locked_ref
);
2506 spin_unlock(&delayed_refs
->lock
);
2508 * we may have dropped the spin lock to get the head
2509 * mutex lock, and that might have given someone else
2510 * time to free the head. If that's true, it has been
2511 * removed from our list and we can move on.
2513 if (ret
== -EAGAIN
) {
2521 * We need to try and merge add/drops of the same ref since we
2522 * can run into issues with relocate dropping the implicit ref
2523 * and then it being added back again before the drop can
2524 * finish. If we merged anything we need to re-loop so we can
2526 * Or we can get node references of the same type that weren't
2527 * merged when created due to bumps in the tree mod seq, and
2528 * we need to merge them to prevent adding an inline extent
2529 * backref before dropping it (triggering a BUG_ON at
2530 * insert_inline_extent_backref()).
2532 spin_lock(&locked_ref
->lock
);
2533 btrfs_merge_delayed_refs(trans
, fs_info
, delayed_refs
,
2537 * locked_ref is the head node, so we have to go one
2538 * node back for any delayed ref updates
2540 ref
= select_delayed_ref(locked_ref
);
2542 if (ref
&& ref
->seq
&&
2543 btrfs_check_delayed_seq(fs_info
, delayed_refs
, ref
->seq
)) {
2544 spin_unlock(&locked_ref
->lock
);
2545 btrfs_delayed_ref_unlock(locked_ref
);
2546 spin_lock(&delayed_refs
->lock
);
2547 locked_ref
->processing
= 0;
2548 delayed_refs
->num_heads_ready
++;
2549 spin_unlock(&delayed_refs
->lock
);
2557 * record the must insert reserved flag before we
2558 * drop the spin lock.
2560 must_insert_reserved
= locked_ref
->must_insert_reserved
;
2561 locked_ref
->must_insert_reserved
= 0;
2563 extent_op
= locked_ref
->extent_op
;
2564 locked_ref
->extent_op
= NULL
;
2569 /* All delayed refs have been processed, Go ahead
2570 * and send the head node to run_one_delayed_ref,
2571 * so that any accounting fixes can happen
2573 ref
= &locked_ref
->node
;
2575 if (extent_op
&& must_insert_reserved
) {
2576 btrfs_free_delayed_extent_op(extent_op
);
2581 spin_unlock(&locked_ref
->lock
);
2582 ret
= run_delayed_extent_op(trans
, root
,
2584 btrfs_free_delayed_extent_op(extent_op
);
2588 * Need to reset must_insert_reserved if
2589 * there was an error so the abort stuff
2590 * can cleanup the reserved space
2593 if (must_insert_reserved
)
2594 locked_ref
->must_insert_reserved
= 1;
2595 locked_ref
->processing
= 0;
2596 btrfs_debug(fs_info
, "run_delayed_extent_op returned %d", ret
);
2597 btrfs_delayed_ref_unlock(locked_ref
);
2604 * Need to drop our head ref lock and re-acquire the
2605 * delayed ref lock and then re-check to make sure
2608 spin_unlock(&locked_ref
->lock
);
2609 spin_lock(&delayed_refs
->lock
);
2610 spin_lock(&locked_ref
->lock
);
2611 if (!list_empty(&locked_ref
->ref_list
) ||
2612 locked_ref
->extent_op
) {
2613 spin_unlock(&locked_ref
->lock
);
2614 spin_unlock(&delayed_refs
->lock
);
2618 delayed_refs
->num_heads
--;
2619 rb_erase(&locked_ref
->href_node
,
2620 &delayed_refs
->href_root
);
2621 spin_unlock(&delayed_refs
->lock
);
2625 list_del(&ref
->list
);
2627 atomic_dec(&delayed_refs
->num_entries
);
2629 if (!btrfs_delayed_ref_is_head(ref
)) {
2631 * when we play the delayed ref, also correct the
2634 switch (ref
->action
) {
2635 case BTRFS_ADD_DELAYED_REF
:
2636 case BTRFS_ADD_DELAYED_EXTENT
:
2637 locked_ref
->node
.ref_mod
-= ref
->ref_mod
;
2639 case BTRFS_DROP_DELAYED_REF
:
2640 locked_ref
->node
.ref_mod
+= ref
->ref_mod
;
2646 spin_unlock(&locked_ref
->lock
);
2648 ret
= run_one_delayed_ref(trans
, root
, ref
, extent_op
,
2649 must_insert_reserved
);
2651 btrfs_free_delayed_extent_op(extent_op
);
2653 locked_ref
->processing
= 0;
2654 btrfs_delayed_ref_unlock(locked_ref
);
2655 btrfs_put_delayed_ref(ref
);
2656 btrfs_debug(fs_info
, "run_one_delayed_ref returned %d", ret
);
2661 * If this node is a head, that means all the refs in this head
2662 * have been dealt with, and we will pick the next head to deal
2663 * with, so we must unlock the head and drop it from the cluster
2664 * list before we release it.
2666 if (btrfs_delayed_ref_is_head(ref
)) {
2667 if (locked_ref
->is_data
&&
2668 locked_ref
->total_ref_mod
< 0) {
2669 spin_lock(&delayed_refs
->lock
);
2670 delayed_refs
->pending_csums
-= ref
->num_bytes
;
2671 spin_unlock(&delayed_refs
->lock
);
2673 btrfs_delayed_ref_unlock(locked_ref
);
2676 btrfs_put_delayed_ref(ref
);
2682 * We don't want to include ref heads since we can have empty ref heads
2683 * and those will drastically skew our runtime down since we just do
2684 * accounting, no actual extent tree updates.
2686 if (actual_count
> 0) {
2687 u64 runtime
= ktime_to_ns(ktime_sub(ktime_get(), start
));
2691 * We weigh the current average higher than our current runtime
2692 * to avoid large swings in the average.
2694 spin_lock(&delayed_refs
->lock
);
2695 avg
= fs_info
->avg_delayed_ref_runtime
* 3 + runtime
;
2696 fs_info
->avg_delayed_ref_runtime
= avg
>> 2; /* div by 4 */
2697 spin_unlock(&delayed_refs
->lock
);
2702 #ifdef SCRAMBLE_DELAYED_REFS
2704 * Normally delayed refs get processed in ascending bytenr order. This
2705 * correlates in most cases to the order added. To expose dependencies on this
2706 * order, we start to process the tree in the middle instead of the beginning
2708 static u64
find_middle(struct rb_root
*root
)
2710 struct rb_node
*n
= root
->rb_node
;
2711 struct btrfs_delayed_ref_node
*entry
;
2714 u64 first
= 0, last
= 0;
2718 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2719 first
= entry
->bytenr
;
2723 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2724 last
= entry
->bytenr
;
2729 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2730 WARN_ON(!entry
->in_tree
);
2732 middle
= entry
->bytenr
;
2745 static inline u64
heads_to_leaves(struct btrfs_root
*root
, u64 heads
)
2749 num_bytes
= heads
* (sizeof(struct btrfs_extent_item
) +
2750 sizeof(struct btrfs_extent_inline_ref
));
2751 if (!btrfs_fs_incompat(root
->fs_info
, SKINNY_METADATA
))
2752 num_bytes
+= heads
* sizeof(struct btrfs_tree_block_info
);
2755 * We don't ever fill up leaves all the way so multiply by 2 just to be
2756 * closer to what we're really going to want to use.
2758 return div_u64(num_bytes
, BTRFS_LEAF_DATA_SIZE(root
));
2762 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2763 * would require to store the csums for that many bytes.
2765 u64
btrfs_csum_bytes_to_leaves(struct btrfs_root
*root
, u64 csum_bytes
)
2768 u64 num_csums_per_leaf
;
2771 csum_size
= BTRFS_LEAF_DATA_SIZE(root
) - sizeof(struct btrfs_item
);
2772 num_csums_per_leaf
= div64_u64(csum_size
,
2773 (u64
)btrfs_super_csum_size(root
->fs_info
->super_copy
));
2774 num_csums
= div64_u64(csum_bytes
, root
->sectorsize
);
2775 num_csums
+= num_csums_per_leaf
- 1;
2776 num_csums
= div64_u64(num_csums
, num_csums_per_leaf
);
2780 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle
*trans
,
2781 struct btrfs_root
*root
)
2783 struct btrfs_block_rsv
*global_rsv
;
2784 u64 num_heads
= trans
->transaction
->delayed_refs
.num_heads_ready
;
2785 u64 csum_bytes
= trans
->transaction
->delayed_refs
.pending_csums
;
2786 u64 num_dirty_bgs
= trans
->transaction
->num_dirty_bgs
;
2787 u64 num_bytes
, num_dirty_bgs_bytes
;
2790 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
2791 num_heads
= heads_to_leaves(root
, num_heads
);
2793 num_bytes
+= (num_heads
- 1) * root
->nodesize
;
2795 num_bytes
+= btrfs_csum_bytes_to_leaves(root
, csum_bytes
) * root
->nodesize
;
2796 num_dirty_bgs_bytes
= btrfs_calc_trans_metadata_size(root
,
2798 global_rsv
= &root
->fs_info
->global_block_rsv
;
2801 * If we can't allocate any more chunks lets make sure we have _lots_ of
2802 * wiggle room since running delayed refs can create more delayed refs.
2804 if (global_rsv
->space_info
->full
) {
2805 num_dirty_bgs_bytes
<<= 1;
2809 spin_lock(&global_rsv
->lock
);
2810 if (global_rsv
->reserved
<= num_bytes
+ num_dirty_bgs_bytes
)
2812 spin_unlock(&global_rsv
->lock
);
2816 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle
*trans
,
2817 struct btrfs_root
*root
)
2819 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2821 atomic_read(&trans
->transaction
->delayed_refs
.num_entries
);
2826 avg_runtime
= fs_info
->avg_delayed_ref_runtime
;
2827 val
= num_entries
* avg_runtime
;
2828 if (num_entries
* avg_runtime
>= NSEC_PER_SEC
)
2830 if (val
>= NSEC_PER_SEC
/ 2)
2833 return btrfs_check_space_for_delayed_refs(trans
, root
);
2836 struct async_delayed_refs
{
2837 struct btrfs_root
*root
;
2842 struct completion wait
;
2843 struct btrfs_work work
;
2846 static void delayed_ref_async_start(struct btrfs_work
*work
)
2848 struct async_delayed_refs
*async
;
2849 struct btrfs_trans_handle
*trans
;
2852 async
= container_of(work
, struct async_delayed_refs
, work
);
2854 /* if the commit is already started, we don't need to wait here */
2855 if (btrfs_transaction_blocked(async
->root
->fs_info
))
2858 trans
= btrfs_join_transaction(async
->root
);
2859 if (IS_ERR(trans
)) {
2860 async
->error
= PTR_ERR(trans
);
2865 * trans->sync means that when we call end_transaction, we won't
2866 * wait on delayed refs
2870 /* Don't bother flushing if we got into a different transaction */
2871 if (trans
->transid
> async
->transid
)
2874 ret
= btrfs_run_delayed_refs(trans
, async
->root
, async
->count
);
2878 ret
= btrfs_end_transaction(trans
, async
->root
);
2879 if (ret
&& !async
->error
)
2883 complete(&async
->wait
);
2888 int btrfs_async_run_delayed_refs(struct btrfs_root
*root
,
2889 unsigned long count
, u64 transid
, int wait
)
2891 struct async_delayed_refs
*async
;
2894 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
2898 async
->root
= root
->fs_info
->tree_root
;
2899 async
->count
= count
;
2901 async
->transid
= transid
;
2906 init_completion(&async
->wait
);
2908 btrfs_init_work(&async
->work
, btrfs_extent_refs_helper
,
2909 delayed_ref_async_start
, NULL
, NULL
);
2911 btrfs_queue_work(root
->fs_info
->extent_workers
, &async
->work
);
2914 wait_for_completion(&async
->wait
);
2923 * this starts processing the delayed reference count updates and
2924 * extent insertions we have queued up so far. count can be
2925 * 0, which means to process everything in the tree at the start
2926 * of the run (but not newly added entries), or it can be some target
2927 * number you'd like to process.
2929 * Returns 0 on success or if called with an aborted transaction
2930 * Returns <0 on error and aborts the transaction
2932 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2933 struct btrfs_root
*root
, unsigned long count
)
2935 struct rb_node
*node
;
2936 struct btrfs_delayed_ref_root
*delayed_refs
;
2937 struct btrfs_delayed_ref_head
*head
;
2939 int run_all
= count
== (unsigned long)-1;
2940 bool can_flush_pending_bgs
= trans
->can_flush_pending_bgs
;
2942 /* We'll clean this up in btrfs_cleanup_transaction */
2946 if (root
->fs_info
->creating_free_space_tree
)
2949 if (root
== root
->fs_info
->extent_root
)
2950 root
= root
->fs_info
->tree_root
;
2952 delayed_refs
= &trans
->transaction
->delayed_refs
;
2954 count
= atomic_read(&delayed_refs
->num_entries
) * 2;
2957 #ifdef SCRAMBLE_DELAYED_REFS
2958 delayed_refs
->run_delayed_start
= find_middle(&delayed_refs
->root
);
2960 trans
->can_flush_pending_bgs
= false;
2961 ret
= __btrfs_run_delayed_refs(trans
, root
, count
);
2963 btrfs_abort_transaction(trans
, root
, ret
);
2968 if (!list_empty(&trans
->new_bgs
))
2969 btrfs_create_pending_block_groups(trans
, root
);
2971 spin_lock(&delayed_refs
->lock
);
2972 node
= rb_first(&delayed_refs
->href_root
);
2974 spin_unlock(&delayed_refs
->lock
);
2977 count
= (unsigned long)-1;
2980 head
= rb_entry(node
, struct btrfs_delayed_ref_head
,
2982 if (btrfs_delayed_ref_is_head(&head
->node
)) {
2983 struct btrfs_delayed_ref_node
*ref
;
2986 atomic_inc(&ref
->refs
);
2988 spin_unlock(&delayed_refs
->lock
);
2990 * Mutex was contended, block until it's
2991 * released and try again
2993 mutex_lock(&head
->mutex
);
2994 mutex_unlock(&head
->mutex
);
2996 btrfs_put_delayed_ref(ref
);
3002 node
= rb_next(node
);
3004 spin_unlock(&delayed_refs
->lock
);
3009 assert_qgroups_uptodate(trans
);
3010 trans
->can_flush_pending_bgs
= can_flush_pending_bgs
;
3014 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
3015 struct btrfs_root
*root
,
3016 u64 bytenr
, u64 num_bytes
, u64 flags
,
3017 int level
, int is_data
)
3019 struct btrfs_delayed_extent_op
*extent_op
;
3022 extent_op
= btrfs_alloc_delayed_extent_op();
3026 extent_op
->flags_to_set
= flags
;
3027 extent_op
->update_flags
= true;
3028 extent_op
->update_key
= false;
3029 extent_op
->is_data
= is_data
? true : false;
3030 extent_op
->level
= level
;
3032 ret
= btrfs_add_delayed_extent_op(root
->fs_info
, trans
, bytenr
,
3033 num_bytes
, extent_op
);
3035 btrfs_free_delayed_extent_op(extent_op
);
3039 static noinline
int check_delayed_ref(struct btrfs_trans_handle
*trans
,
3040 struct btrfs_root
*root
,
3041 struct btrfs_path
*path
,
3042 u64 objectid
, u64 offset
, u64 bytenr
)
3044 struct btrfs_delayed_ref_head
*head
;
3045 struct btrfs_delayed_ref_node
*ref
;
3046 struct btrfs_delayed_data_ref
*data_ref
;
3047 struct btrfs_delayed_ref_root
*delayed_refs
;
3050 delayed_refs
= &trans
->transaction
->delayed_refs
;
3051 spin_lock(&delayed_refs
->lock
);
3052 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
3054 spin_unlock(&delayed_refs
->lock
);
3058 if (!mutex_trylock(&head
->mutex
)) {
3059 atomic_inc(&head
->node
.refs
);
3060 spin_unlock(&delayed_refs
->lock
);
3062 btrfs_release_path(path
);
3065 * Mutex was contended, block until it's released and let
3068 mutex_lock(&head
->mutex
);
3069 mutex_unlock(&head
->mutex
);
3070 btrfs_put_delayed_ref(&head
->node
);
3073 spin_unlock(&delayed_refs
->lock
);
3075 spin_lock(&head
->lock
);
3076 list_for_each_entry(ref
, &head
->ref_list
, list
) {
3077 /* If it's a shared ref we know a cross reference exists */
3078 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
) {
3083 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
3086 * If our ref doesn't match the one we're currently looking at
3087 * then we have a cross reference.
3089 if (data_ref
->root
!= root
->root_key
.objectid
||
3090 data_ref
->objectid
!= objectid
||
3091 data_ref
->offset
!= offset
) {
3096 spin_unlock(&head
->lock
);
3097 mutex_unlock(&head
->mutex
);
3101 static noinline
int check_committed_ref(struct btrfs_trans_handle
*trans
,
3102 struct btrfs_root
*root
,
3103 struct btrfs_path
*path
,
3104 u64 objectid
, u64 offset
, u64 bytenr
)
3106 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
3107 struct extent_buffer
*leaf
;
3108 struct btrfs_extent_data_ref
*ref
;
3109 struct btrfs_extent_inline_ref
*iref
;
3110 struct btrfs_extent_item
*ei
;
3111 struct btrfs_key key
;
3115 key
.objectid
= bytenr
;
3116 key
.offset
= (u64
)-1;
3117 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
3119 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
3122 BUG_ON(ret
== 0); /* Corruption */
3125 if (path
->slots
[0] == 0)
3129 leaf
= path
->nodes
[0];
3130 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
3132 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
3136 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
3137 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3138 if (item_size
< sizeof(*ei
)) {
3139 WARN_ON(item_size
!= sizeof(struct btrfs_extent_item_v0
));
3143 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
3145 if (item_size
!= sizeof(*ei
) +
3146 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
3149 if (btrfs_extent_generation(leaf
, ei
) <=
3150 btrfs_root_last_snapshot(&root
->root_item
))
3153 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
3154 if (btrfs_extent_inline_ref_type(leaf
, iref
) !=
3155 BTRFS_EXTENT_DATA_REF_KEY
)
3158 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
3159 if (btrfs_extent_refs(leaf
, ei
) !=
3160 btrfs_extent_data_ref_count(leaf
, ref
) ||
3161 btrfs_extent_data_ref_root(leaf
, ref
) !=
3162 root
->root_key
.objectid
||
3163 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
3164 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
3172 int btrfs_cross_ref_exist(struct btrfs_trans_handle
*trans
,
3173 struct btrfs_root
*root
,
3174 u64 objectid
, u64 offset
, u64 bytenr
)
3176 struct btrfs_path
*path
;
3180 path
= btrfs_alloc_path();
3185 ret
= check_committed_ref(trans
, root
, path
, objectid
,
3187 if (ret
&& ret
!= -ENOENT
)
3190 ret2
= check_delayed_ref(trans
, root
, path
, objectid
,
3192 } while (ret2
== -EAGAIN
);
3194 if (ret2
&& ret2
!= -ENOENT
) {
3199 if (ret
!= -ENOENT
|| ret2
!= -ENOENT
)
3202 btrfs_free_path(path
);
3203 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
3208 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
3209 struct btrfs_root
*root
,
3210 struct extent_buffer
*buf
,
3211 int full_backref
, int inc
)
3218 struct btrfs_key key
;
3219 struct btrfs_file_extent_item
*fi
;
3223 int (*process_func
)(struct btrfs_trans_handle
*, struct btrfs_root
*,
3224 u64
, u64
, u64
, u64
, u64
, u64
);
3227 if (btrfs_test_is_dummy_root(root
))
3230 ref_root
= btrfs_header_owner(buf
);
3231 nritems
= btrfs_header_nritems(buf
);
3232 level
= btrfs_header_level(buf
);
3234 if (!test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) && level
== 0)
3238 process_func
= btrfs_inc_extent_ref
;
3240 process_func
= btrfs_free_extent
;
3243 parent
= buf
->start
;
3247 for (i
= 0; i
< nritems
; i
++) {
3249 btrfs_item_key_to_cpu(buf
, &key
, i
);
3250 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
3252 fi
= btrfs_item_ptr(buf
, i
,
3253 struct btrfs_file_extent_item
);
3254 if (btrfs_file_extent_type(buf
, fi
) ==
3255 BTRFS_FILE_EXTENT_INLINE
)
3257 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
3261 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
3262 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
3263 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
3264 parent
, ref_root
, key
.objectid
,
3269 bytenr
= btrfs_node_blockptr(buf
, i
);
3270 num_bytes
= root
->nodesize
;
3271 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
3272 parent
, ref_root
, level
- 1, 0);
3282 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3283 struct extent_buffer
*buf
, int full_backref
)
3285 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1);
3288 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3289 struct extent_buffer
*buf
, int full_backref
)
3291 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0);
3294 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
3295 struct btrfs_root
*root
,
3296 struct btrfs_path
*path
,
3297 struct btrfs_block_group_cache
*cache
)
3300 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
3302 struct extent_buffer
*leaf
;
3304 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
3311 leaf
= path
->nodes
[0];
3312 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
3313 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
3314 btrfs_mark_buffer_dirty(leaf
);
3316 btrfs_release_path(path
);
3321 static struct btrfs_block_group_cache
*
3322 next_block_group(struct btrfs_root
*root
,
3323 struct btrfs_block_group_cache
*cache
)
3325 struct rb_node
*node
;
3327 spin_lock(&root
->fs_info
->block_group_cache_lock
);
3329 /* If our block group was removed, we need a full search. */
3330 if (RB_EMPTY_NODE(&cache
->cache_node
)) {
3331 const u64 next_bytenr
= cache
->key
.objectid
+ cache
->key
.offset
;
3333 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
3334 btrfs_put_block_group(cache
);
3335 cache
= btrfs_lookup_first_block_group(root
->fs_info
,
3339 node
= rb_next(&cache
->cache_node
);
3340 btrfs_put_block_group(cache
);
3342 cache
= rb_entry(node
, struct btrfs_block_group_cache
,
3344 btrfs_get_block_group(cache
);
3347 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
3351 static int cache_save_setup(struct btrfs_block_group_cache
*block_group
,
3352 struct btrfs_trans_handle
*trans
,
3353 struct btrfs_path
*path
)
3355 struct btrfs_root
*root
= block_group
->fs_info
->tree_root
;
3356 struct inode
*inode
= NULL
;
3358 int dcs
= BTRFS_DC_ERROR
;
3364 * If this block group is smaller than 100 megs don't bother caching the
3367 if (block_group
->key
.offset
< (100 * SZ_1M
)) {
3368 spin_lock(&block_group
->lock
);
3369 block_group
->disk_cache_state
= BTRFS_DC_WRITTEN
;
3370 spin_unlock(&block_group
->lock
);
3377 inode
= lookup_free_space_inode(root
, block_group
, path
);
3378 if (IS_ERR(inode
) && PTR_ERR(inode
) != -ENOENT
) {
3379 ret
= PTR_ERR(inode
);
3380 btrfs_release_path(path
);
3384 if (IS_ERR(inode
)) {
3388 if (block_group
->ro
)
3391 ret
= create_free_space_inode(root
, trans
, block_group
, path
);
3397 /* We've already setup this transaction, go ahead and exit */
3398 if (block_group
->cache_generation
== trans
->transid
&&
3399 i_size_read(inode
)) {
3400 dcs
= BTRFS_DC_SETUP
;
3405 * We want to set the generation to 0, that way if anything goes wrong
3406 * from here on out we know not to trust this cache when we load up next
3409 BTRFS_I(inode
)->generation
= 0;
3410 ret
= btrfs_update_inode(trans
, root
, inode
);
3413 * So theoretically we could recover from this, simply set the
3414 * super cache generation to 0 so we know to invalidate the
3415 * cache, but then we'd have to keep track of the block groups
3416 * that fail this way so we know we _have_ to reset this cache
3417 * before the next commit or risk reading stale cache. So to
3418 * limit our exposure to horrible edge cases lets just abort the
3419 * transaction, this only happens in really bad situations
3422 btrfs_abort_transaction(trans
, root
, ret
);
3427 if (i_size_read(inode
) > 0) {
3428 ret
= btrfs_check_trunc_cache_free_space(root
,
3429 &root
->fs_info
->global_block_rsv
);
3433 ret
= btrfs_truncate_free_space_cache(root
, trans
, NULL
, inode
);
3438 spin_lock(&block_group
->lock
);
3439 if (block_group
->cached
!= BTRFS_CACHE_FINISHED
||
3440 !btrfs_test_opt(root
, SPACE_CACHE
)) {
3442 * don't bother trying to write stuff out _if_
3443 * a) we're not cached,
3444 * b) we're with nospace_cache mount option.
3446 dcs
= BTRFS_DC_WRITTEN
;
3447 spin_unlock(&block_group
->lock
);
3450 spin_unlock(&block_group
->lock
);
3453 * We hit an ENOSPC when setting up the cache in this transaction, just
3454 * skip doing the setup, we've already cleared the cache so we're safe.
3456 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC
, &trans
->transaction
->flags
)) {
3462 * Try to preallocate enough space based on how big the block group is.
3463 * Keep in mind this has to include any pinned space which could end up
3464 * taking up quite a bit since it's not folded into the other space
3467 num_pages
= div_u64(block_group
->key
.offset
, SZ_256M
);
3472 num_pages
*= PAGE_SIZE
;
3474 ret
= btrfs_check_data_free_space(inode
, 0, num_pages
);
3478 ret
= btrfs_prealloc_file_range_trans(inode
, trans
, 0, 0, num_pages
,
3479 num_pages
, num_pages
,
3482 * Our cache requires contiguous chunks so that we don't modify a bunch
3483 * of metadata or split extents when writing the cache out, which means
3484 * we can enospc if we are heavily fragmented in addition to just normal
3485 * out of space conditions. So if we hit this just skip setting up any
3486 * other block groups for this transaction, maybe we'll unpin enough
3487 * space the next time around.
3490 dcs
= BTRFS_DC_SETUP
;
3491 else if (ret
== -ENOSPC
)
3492 set_bit(BTRFS_TRANS_CACHE_ENOSPC
, &trans
->transaction
->flags
);
3493 btrfs_free_reserved_data_space(inode
, 0, num_pages
);
3498 btrfs_release_path(path
);
3500 spin_lock(&block_group
->lock
);
3501 if (!ret
&& dcs
== BTRFS_DC_SETUP
)
3502 block_group
->cache_generation
= trans
->transid
;
3503 block_group
->disk_cache_state
= dcs
;
3504 spin_unlock(&block_group
->lock
);
3509 int btrfs_setup_space_cache(struct btrfs_trans_handle
*trans
,
3510 struct btrfs_root
*root
)
3512 struct btrfs_block_group_cache
*cache
, *tmp
;
3513 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3514 struct btrfs_path
*path
;
3516 if (list_empty(&cur_trans
->dirty_bgs
) ||
3517 !btrfs_test_opt(root
, SPACE_CACHE
))
3520 path
= btrfs_alloc_path();
3524 /* Could add new block groups, use _safe just in case */
3525 list_for_each_entry_safe(cache
, tmp
, &cur_trans
->dirty_bgs
,
3527 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
)
3528 cache_save_setup(cache
, trans
, path
);
3531 btrfs_free_path(path
);
3536 * transaction commit does final block group cache writeback during a
3537 * critical section where nothing is allowed to change the FS. This is
3538 * required in order for the cache to actually match the block group,
3539 * but can introduce a lot of latency into the commit.
3541 * So, btrfs_start_dirty_block_groups is here to kick off block group
3542 * cache IO. There's a chance we'll have to redo some of it if the
3543 * block group changes again during the commit, but it greatly reduces
3544 * the commit latency by getting rid of the easy block groups while
3545 * we're still allowing others to join the commit.
3547 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle
*trans
,
3548 struct btrfs_root
*root
)
3550 struct btrfs_block_group_cache
*cache
;
3551 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3554 struct btrfs_path
*path
= NULL
;
3556 struct list_head
*io
= &cur_trans
->io_bgs
;
3557 int num_started
= 0;
3560 spin_lock(&cur_trans
->dirty_bgs_lock
);
3561 if (list_empty(&cur_trans
->dirty_bgs
)) {
3562 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3565 list_splice_init(&cur_trans
->dirty_bgs
, &dirty
);
3566 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3570 * make sure all the block groups on our dirty list actually
3573 btrfs_create_pending_block_groups(trans
, root
);
3576 path
= btrfs_alloc_path();
3582 * cache_write_mutex is here only to save us from balance or automatic
3583 * removal of empty block groups deleting this block group while we are
3584 * writing out the cache
3586 mutex_lock(&trans
->transaction
->cache_write_mutex
);
3587 while (!list_empty(&dirty
)) {
3588 cache
= list_first_entry(&dirty
,
3589 struct btrfs_block_group_cache
,
3592 * this can happen if something re-dirties a block
3593 * group that is already under IO. Just wait for it to
3594 * finish and then do it all again
3596 if (!list_empty(&cache
->io_list
)) {
3597 list_del_init(&cache
->io_list
);
3598 btrfs_wait_cache_io(root
, trans
, cache
,
3599 &cache
->io_ctl
, path
,
3600 cache
->key
.objectid
);
3601 btrfs_put_block_group(cache
);
3606 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3607 * if it should update the cache_state. Don't delete
3608 * until after we wait.
3610 * Since we're not running in the commit critical section
3611 * we need the dirty_bgs_lock to protect from update_block_group
3613 spin_lock(&cur_trans
->dirty_bgs_lock
);
3614 list_del_init(&cache
->dirty_list
);
3615 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3619 cache_save_setup(cache
, trans
, path
);
3621 if (cache
->disk_cache_state
== BTRFS_DC_SETUP
) {
3622 cache
->io_ctl
.inode
= NULL
;
3623 ret
= btrfs_write_out_cache(root
, trans
, cache
, path
);
3624 if (ret
== 0 && cache
->io_ctl
.inode
) {
3629 * the cache_write_mutex is protecting
3632 list_add_tail(&cache
->io_list
, io
);
3635 * if we failed to write the cache, the
3636 * generation will be bad and life goes on
3642 ret
= write_one_cache_group(trans
, root
, path
, cache
);
3644 * Our block group might still be attached to the list
3645 * of new block groups in the transaction handle of some
3646 * other task (struct btrfs_trans_handle->new_bgs). This
3647 * means its block group item isn't yet in the extent
3648 * tree. If this happens ignore the error, as we will
3649 * try again later in the critical section of the
3650 * transaction commit.
3652 if (ret
== -ENOENT
) {
3654 spin_lock(&cur_trans
->dirty_bgs_lock
);
3655 if (list_empty(&cache
->dirty_list
)) {
3656 list_add_tail(&cache
->dirty_list
,
3657 &cur_trans
->dirty_bgs
);
3658 btrfs_get_block_group(cache
);
3660 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3662 btrfs_abort_transaction(trans
, root
, ret
);
3666 /* if its not on the io list, we need to put the block group */
3668 btrfs_put_block_group(cache
);
3674 * Avoid blocking other tasks for too long. It might even save
3675 * us from writing caches for block groups that are going to be
3678 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
3679 mutex_lock(&trans
->transaction
->cache_write_mutex
);
3681 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
3684 * go through delayed refs for all the stuff we've just kicked off
3685 * and then loop back (just once)
3687 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
3688 if (!ret
&& loops
== 0) {
3690 spin_lock(&cur_trans
->dirty_bgs_lock
);
3691 list_splice_init(&cur_trans
->dirty_bgs
, &dirty
);
3693 * dirty_bgs_lock protects us from concurrent block group
3694 * deletes too (not just cache_write_mutex).
3696 if (!list_empty(&dirty
)) {
3697 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3700 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3703 btrfs_free_path(path
);
3707 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
3708 struct btrfs_root
*root
)
3710 struct btrfs_block_group_cache
*cache
;
3711 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3714 struct btrfs_path
*path
;
3715 struct list_head
*io
= &cur_trans
->io_bgs
;
3716 int num_started
= 0;
3718 path
= btrfs_alloc_path();
3723 * Even though we are in the critical section of the transaction commit,
3724 * we can still have concurrent tasks adding elements to this
3725 * transaction's list of dirty block groups. These tasks correspond to
3726 * endio free space workers started when writeback finishes for a
3727 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3728 * allocate new block groups as a result of COWing nodes of the root
3729 * tree when updating the free space inode. The writeback for the space
3730 * caches is triggered by an earlier call to
3731 * btrfs_start_dirty_block_groups() and iterations of the following
3733 * Also we want to do the cache_save_setup first and then run the
3734 * delayed refs to make sure we have the best chance at doing this all
3737 spin_lock(&cur_trans
->dirty_bgs_lock
);
3738 while (!list_empty(&cur_trans
->dirty_bgs
)) {
3739 cache
= list_first_entry(&cur_trans
->dirty_bgs
,
3740 struct btrfs_block_group_cache
,
3744 * this can happen if cache_save_setup re-dirties a block
3745 * group that is already under IO. Just wait for it to
3746 * finish and then do it all again
3748 if (!list_empty(&cache
->io_list
)) {
3749 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3750 list_del_init(&cache
->io_list
);
3751 btrfs_wait_cache_io(root
, trans
, cache
,
3752 &cache
->io_ctl
, path
,
3753 cache
->key
.objectid
);
3754 btrfs_put_block_group(cache
);
3755 spin_lock(&cur_trans
->dirty_bgs_lock
);
3759 * don't remove from the dirty list until after we've waited
3762 list_del_init(&cache
->dirty_list
);
3763 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3766 cache_save_setup(cache
, trans
, path
);
3769 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long) -1);
3771 if (!ret
&& cache
->disk_cache_state
== BTRFS_DC_SETUP
) {
3772 cache
->io_ctl
.inode
= NULL
;
3773 ret
= btrfs_write_out_cache(root
, trans
, cache
, path
);
3774 if (ret
== 0 && cache
->io_ctl
.inode
) {
3777 list_add_tail(&cache
->io_list
, io
);
3780 * if we failed to write the cache, the
3781 * generation will be bad and life goes on
3787 ret
= write_one_cache_group(trans
, root
, path
, cache
);
3789 * One of the free space endio workers might have
3790 * created a new block group while updating a free space
3791 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3792 * and hasn't released its transaction handle yet, in
3793 * which case the new block group is still attached to
3794 * its transaction handle and its creation has not
3795 * finished yet (no block group item in the extent tree
3796 * yet, etc). If this is the case, wait for all free
3797 * space endio workers to finish and retry. This is a
3798 * a very rare case so no need for a more efficient and
3801 if (ret
== -ENOENT
) {
3802 wait_event(cur_trans
->writer_wait
,
3803 atomic_read(&cur_trans
->num_writers
) == 1);
3804 ret
= write_one_cache_group(trans
, root
, path
,
3808 btrfs_abort_transaction(trans
, root
, ret
);
3811 /* if its not on the io list, we need to put the block group */
3813 btrfs_put_block_group(cache
);
3814 spin_lock(&cur_trans
->dirty_bgs_lock
);
3816 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3818 while (!list_empty(io
)) {
3819 cache
= list_first_entry(io
, struct btrfs_block_group_cache
,
3821 list_del_init(&cache
->io_list
);
3822 btrfs_wait_cache_io(root
, trans
, cache
,
3823 &cache
->io_ctl
, path
, cache
->key
.objectid
);
3824 btrfs_put_block_group(cache
);
3827 btrfs_free_path(path
);
3831 int btrfs_extent_readonly(struct btrfs_root
*root
, u64 bytenr
)
3833 struct btrfs_block_group_cache
*block_group
;
3836 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
3837 if (!block_group
|| block_group
->ro
)
3840 btrfs_put_block_group(block_group
);
3844 bool btrfs_inc_nocow_writers(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
3846 struct btrfs_block_group_cache
*bg
;
3849 bg
= btrfs_lookup_block_group(fs_info
, bytenr
);
3853 spin_lock(&bg
->lock
);
3857 atomic_inc(&bg
->nocow_writers
);
3858 spin_unlock(&bg
->lock
);
3860 /* no put on block group, done by btrfs_dec_nocow_writers */
3862 btrfs_put_block_group(bg
);
3868 void btrfs_dec_nocow_writers(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
3870 struct btrfs_block_group_cache
*bg
;
3872 bg
= btrfs_lookup_block_group(fs_info
, bytenr
);
3874 if (atomic_dec_and_test(&bg
->nocow_writers
))
3875 wake_up_atomic_t(&bg
->nocow_writers
);
3877 * Once for our lookup and once for the lookup done by a previous call
3878 * to btrfs_inc_nocow_writers()
3880 btrfs_put_block_group(bg
);
3881 btrfs_put_block_group(bg
);
3884 static int btrfs_wait_nocow_writers_atomic_t(atomic_t
*a
)
3890 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache
*bg
)
3892 wait_on_atomic_t(&bg
->nocow_writers
,
3893 btrfs_wait_nocow_writers_atomic_t
,
3894 TASK_UNINTERRUPTIBLE
);
3897 static const char *alloc_name(u64 flags
)
3900 case BTRFS_BLOCK_GROUP_METADATA
|BTRFS_BLOCK_GROUP_DATA
:
3902 case BTRFS_BLOCK_GROUP_METADATA
:
3904 case BTRFS_BLOCK_GROUP_DATA
:
3906 case BTRFS_BLOCK_GROUP_SYSTEM
:
3910 return "invalid-combination";
3914 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
3915 u64 total_bytes
, u64 bytes_used
,
3917 struct btrfs_space_info
**space_info
)
3919 struct btrfs_space_info
*found
;
3924 if (flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3925 BTRFS_BLOCK_GROUP_RAID10
))
3930 found
= __find_space_info(info
, flags
);
3932 spin_lock(&found
->lock
);
3933 found
->total_bytes
+= total_bytes
;
3934 found
->disk_total
+= total_bytes
* factor
;
3935 found
->bytes_used
+= bytes_used
;
3936 found
->disk_used
+= bytes_used
* factor
;
3937 found
->bytes_readonly
+= bytes_readonly
;
3938 if (total_bytes
> 0)
3940 spin_unlock(&found
->lock
);
3941 *space_info
= found
;
3944 found
= kzalloc(sizeof(*found
), GFP_NOFS
);
3948 ret
= percpu_counter_init(&found
->total_bytes_pinned
, 0, GFP_KERNEL
);
3954 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
3955 INIT_LIST_HEAD(&found
->block_groups
[i
]);
3956 init_rwsem(&found
->groups_sem
);
3957 spin_lock_init(&found
->lock
);
3958 found
->flags
= flags
& BTRFS_BLOCK_GROUP_TYPE_MASK
;
3959 found
->total_bytes
= total_bytes
;
3960 found
->disk_total
= total_bytes
* factor
;
3961 found
->bytes_used
= bytes_used
;
3962 found
->disk_used
= bytes_used
* factor
;
3963 found
->bytes_pinned
= 0;
3964 found
->bytes_reserved
= 0;
3965 found
->bytes_readonly
= bytes_readonly
;
3966 found
->bytes_may_use
= 0;
3968 found
->max_extent_size
= 0;
3969 found
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3970 found
->chunk_alloc
= 0;
3972 init_waitqueue_head(&found
->wait
);
3973 INIT_LIST_HEAD(&found
->ro_bgs
);
3975 ret
= kobject_init_and_add(&found
->kobj
, &space_info_ktype
,
3976 info
->space_info_kobj
, "%s",
3977 alloc_name(found
->flags
));
3983 *space_info
= found
;
3984 list_add_rcu(&found
->list
, &info
->space_info
);
3985 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3986 info
->data_sinfo
= found
;
3991 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
3993 u64 extra_flags
= chunk_to_extended(flags
) &
3994 BTRFS_EXTENDED_PROFILE_MASK
;
3996 write_seqlock(&fs_info
->profiles_lock
);
3997 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3998 fs_info
->avail_data_alloc_bits
|= extra_flags
;
3999 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
4000 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
4001 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
4002 fs_info
->avail_system_alloc_bits
|= extra_flags
;
4003 write_sequnlock(&fs_info
->profiles_lock
);
4007 * returns target flags in extended format or 0 if restripe for this
4008 * chunk_type is not in progress
4010 * should be called with either volume_mutex or balance_lock held
4012 static u64
get_restripe_target(struct btrfs_fs_info
*fs_info
, u64 flags
)
4014 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
4020 if (flags
& BTRFS_BLOCK_GROUP_DATA
&&
4021 bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
4022 target
= BTRFS_BLOCK_GROUP_DATA
| bctl
->data
.target
;
4023 } else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
&&
4024 bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
4025 target
= BTRFS_BLOCK_GROUP_SYSTEM
| bctl
->sys
.target
;
4026 } else if (flags
& BTRFS_BLOCK_GROUP_METADATA
&&
4027 bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
4028 target
= BTRFS_BLOCK_GROUP_METADATA
| bctl
->meta
.target
;
4035 * @flags: available profiles in extended format (see ctree.h)
4037 * Returns reduced profile in chunk format. If profile changing is in
4038 * progress (either running or paused) picks the target profile (if it's
4039 * already available), otherwise falls back to plain reducing.
4041 static u64
btrfs_reduce_alloc_profile(struct btrfs_root
*root
, u64 flags
)
4043 u64 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
4049 * see if restripe for this chunk_type is in progress, if so
4050 * try to reduce to the target profile
4052 spin_lock(&root
->fs_info
->balance_lock
);
4053 target
= get_restripe_target(root
->fs_info
, flags
);
4055 /* pick target profile only if it's already available */
4056 if ((flags
& target
) & BTRFS_EXTENDED_PROFILE_MASK
) {
4057 spin_unlock(&root
->fs_info
->balance_lock
);
4058 return extended_to_chunk(target
);
4061 spin_unlock(&root
->fs_info
->balance_lock
);
4063 /* First, mask out the RAID levels which aren't possible */
4064 for (raid_type
= 0; raid_type
< BTRFS_NR_RAID_TYPES
; raid_type
++) {
4065 if (num_devices
>= btrfs_raid_array
[raid_type
].devs_min
)
4066 allowed
|= btrfs_raid_group
[raid_type
];
4070 if (allowed
& BTRFS_BLOCK_GROUP_RAID6
)
4071 allowed
= BTRFS_BLOCK_GROUP_RAID6
;
4072 else if (allowed
& BTRFS_BLOCK_GROUP_RAID5
)
4073 allowed
= BTRFS_BLOCK_GROUP_RAID5
;
4074 else if (allowed
& BTRFS_BLOCK_GROUP_RAID10
)
4075 allowed
= BTRFS_BLOCK_GROUP_RAID10
;
4076 else if (allowed
& BTRFS_BLOCK_GROUP_RAID1
)
4077 allowed
= BTRFS_BLOCK_GROUP_RAID1
;
4078 else if (allowed
& BTRFS_BLOCK_GROUP_RAID0
)
4079 allowed
= BTRFS_BLOCK_GROUP_RAID0
;
4081 flags
&= ~BTRFS_BLOCK_GROUP_PROFILE_MASK
;
4083 return extended_to_chunk(flags
| allowed
);
4086 static u64
get_alloc_profile(struct btrfs_root
*root
, u64 orig_flags
)
4093 seq
= read_seqbegin(&root
->fs_info
->profiles_lock
);
4095 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
4096 flags
|= root
->fs_info
->avail_data_alloc_bits
;
4097 else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
4098 flags
|= root
->fs_info
->avail_system_alloc_bits
;
4099 else if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
4100 flags
|= root
->fs_info
->avail_metadata_alloc_bits
;
4101 } while (read_seqretry(&root
->fs_info
->profiles_lock
, seq
));
4103 return btrfs_reduce_alloc_profile(root
, flags
);
4106 u64
btrfs_get_alloc_profile(struct btrfs_root
*root
, int data
)
4112 flags
= BTRFS_BLOCK_GROUP_DATA
;
4113 else if (root
== root
->fs_info
->chunk_root
)
4114 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
4116 flags
= BTRFS_BLOCK_GROUP_METADATA
;
4118 ret
= get_alloc_profile(root
, flags
);
4122 int btrfs_alloc_data_chunk_ondemand(struct inode
*inode
, u64 bytes
)
4124 struct btrfs_space_info
*data_sinfo
;
4125 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4126 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4129 int need_commit
= 2;
4130 int have_pinned_space
;
4132 /* make sure bytes are sectorsize aligned */
4133 bytes
= ALIGN(bytes
, root
->sectorsize
);
4135 if (btrfs_is_free_space_inode(inode
)) {
4137 ASSERT(current
->journal_info
);
4140 data_sinfo
= fs_info
->data_sinfo
;
4145 /* make sure we have enough space to handle the data first */
4146 spin_lock(&data_sinfo
->lock
);
4147 used
= data_sinfo
->bytes_used
+ data_sinfo
->bytes_reserved
+
4148 data_sinfo
->bytes_pinned
+ data_sinfo
->bytes_readonly
+
4149 data_sinfo
->bytes_may_use
;
4151 if (used
+ bytes
> data_sinfo
->total_bytes
) {
4152 struct btrfs_trans_handle
*trans
;
4155 * if we don't have enough free bytes in this space then we need
4156 * to alloc a new chunk.
4158 if (!data_sinfo
->full
) {
4161 data_sinfo
->force_alloc
= CHUNK_ALLOC_FORCE
;
4162 spin_unlock(&data_sinfo
->lock
);
4164 alloc_target
= btrfs_get_alloc_profile(root
, 1);
4166 * It is ugly that we don't call nolock join
4167 * transaction for the free space inode case here.
4168 * But it is safe because we only do the data space
4169 * reservation for the free space cache in the
4170 * transaction context, the common join transaction
4171 * just increase the counter of the current transaction
4172 * handler, doesn't try to acquire the trans_lock of
4175 trans
= btrfs_join_transaction(root
);
4177 return PTR_ERR(trans
);
4179 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
4181 CHUNK_ALLOC_NO_FORCE
);
4182 btrfs_end_transaction(trans
, root
);
4187 have_pinned_space
= 1;
4193 data_sinfo
= fs_info
->data_sinfo
;
4199 * If we don't have enough pinned space to deal with this
4200 * allocation, and no removed chunk in current transaction,
4201 * don't bother committing the transaction.
4203 have_pinned_space
= percpu_counter_compare(
4204 &data_sinfo
->total_bytes_pinned
,
4205 used
+ bytes
- data_sinfo
->total_bytes
);
4206 spin_unlock(&data_sinfo
->lock
);
4208 /* commit the current transaction and try again */
4211 !atomic_read(&root
->fs_info
->open_ioctl_trans
)) {
4214 if (need_commit
> 0) {
4215 btrfs_start_delalloc_roots(fs_info
, 0, -1);
4216 btrfs_wait_ordered_roots(fs_info
, -1, 0, (u64
)-1);
4219 trans
= btrfs_join_transaction(root
);
4221 return PTR_ERR(trans
);
4222 if (have_pinned_space
>= 0 ||
4223 test_bit(BTRFS_TRANS_HAVE_FREE_BGS
,
4224 &trans
->transaction
->flags
) ||
4226 ret
= btrfs_commit_transaction(trans
, root
);
4230 * The cleaner kthread might still be doing iput
4231 * operations. Wait for it to finish so that
4232 * more space is released.
4234 mutex_lock(&root
->fs_info
->cleaner_delayed_iput_mutex
);
4235 mutex_unlock(&root
->fs_info
->cleaner_delayed_iput_mutex
);
4238 btrfs_end_transaction(trans
, root
);
4242 trace_btrfs_space_reservation(root
->fs_info
,
4243 "space_info:enospc",
4244 data_sinfo
->flags
, bytes
, 1);
4247 data_sinfo
->bytes_may_use
+= bytes
;
4248 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
4249 data_sinfo
->flags
, bytes
, 1);
4250 spin_unlock(&data_sinfo
->lock
);
4256 * New check_data_free_space() with ability for precious data reservation
4257 * Will replace old btrfs_check_data_free_space(), but for patch split,
4258 * add a new function first and then replace it.
4260 int btrfs_check_data_free_space(struct inode
*inode
, u64 start
, u64 len
)
4262 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4265 /* align the range */
4266 len
= round_up(start
+ len
, root
->sectorsize
) -
4267 round_down(start
, root
->sectorsize
);
4268 start
= round_down(start
, root
->sectorsize
);
4270 ret
= btrfs_alloc_data_chunk_ondemand(inode
, len
);
4275 * Use new btrfs_qgroup_reserve_data to reserve precious data space
4277 * TODO: Find a good method to avoid reserve data space for NOCOW
4278 * range, but don't impact performance on quota disable case.
4280 ret
= btrfs_qgroup_reserve_data(inode
, start
, len
);
4285 * Called if we need to clear a data reservation for this inode
4286 * Normally in a error case.
4288 * This one will *NOT* use accurate qgroup reserved space API, just for case
4289 * which we can't sleep and is sure it won't affect qgroup reserved space.
4290 * Like clear_bit_hook().
4292 void btrfs_free_reserved_data_space_noquota(struct inode
*inode
, u64 start
,
4295 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4296 struct btrfs_space_info
*data_sinfo
;
4298 /* Make sure the range is aligned to sectorsize */
4299 len
= round_up(start
+ len
, root
->sectorsize
) -
4300 round_down(start
, root
->sectorsize
);
4301 start
= round_down(start
, root
->sectorsize
);
4303 data_sinfo
= root
->fs_info
->data_sinfo
;
4304 spin_lock(&data_sinfo
->lock
);
4305 if (WARN_ON(data_sinfo
->bytes_may_use
< len
))
4306 data_sinfo
->bytes_may_use
= 0;
4308 data_sinfo
->bytes_may_use
-= len
;
4309 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
4310 data_sinfo
->flags
, len
, 0);
4311 spin_unlock(&data_sinfo
->lock
);
4315 * Called if we need to clear a data reservation for this inode
4316 * Normally in a error case.
4318 * This one will handle the per-inode data rsv map for accurate reserved
4321 void btrfs_free_reserved_data_space(struct inode
*inode
, u64 start
, u64 len
)
4323 btrfs_free_reserved_data_space_noquota(inode
, start
, len
);
4324 btrfs_qgroup_free_data(inode
, start
, len
);
4327 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
4329 struct list_head
*head
= &info
->space_info
;
4330 struct btrfs_space_info
*found
;
4333 list_for_each_entry_rcu(found
, head
, list
) {
4334 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
4335 found
->force_alloc
= CHUNK_ALLOC_FORCE
;
4340 static inline u64
calc_global_rsv_need_space(struct btrfs_block_rsv
*global
)
4342 return (global
->size
<< 1);
4345 static int should_alloc_chunk(struct btrfs_root
*root
,
4346 struct btrfs_space_info
*sinfo
, int force
)
4348 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
4349 u64 num_bytes
= sinfo
->total_bytes
- sinfo
->bytes_readonly
;
4350 u64 num_allocated
= sinfo
->bytes_used
+ sinfo
->bytes_reserved
;
4353 if (force
== CHUNK_ALLOC_FORCE
)
4357 * We need to take into account the global rsv because for all intents
4358 * and purposes it's used space. Don't worry about locking the
4359 * global_rsv, it doesn't change except when the transaction commits.
4361 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
4362 num_allocated
+= calc_global_rsv_need_space(global_rsv
);
4365 * in limited mode, we want to have some free space up to
4366 * about 1% of the FS size.
4368 if (force
== CHUNK_ALLOC_LIMITED
) {
4369 thresh
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
4370 thresh
= max_t(u64
, SZ_64M
, div_factor_fine(thresh
, 1));
4372 if (num_bytes
- num_allocated
< thresh
)
4376 if (num_allocated
+ SZ_2M
< div_factor(num_bytes
, 8))
4381 static u64
get_profile_num_devs(struct btrfs_root
*root
, u64 type
)
4385 if (type
& (BTRFS_BLOCK_GROUP_RAID10
|
4386 BTRFS_BLOCK_GROUP_RAID0
|
4387 BTRFS_BLOCK_GROUP_RAID5
|
4388 BTRFS_BLOCK_GROUP_RAID6
))
4389 num_dev
= root
->fs_info
->fs_devices
->rw_devices
;
4390 else if (type
& BTRFS_BLOCK_GROUP_RAID1
)
4393 num_dev
= 1; /* DUP or single */
4399 * If @is_allocation is true, reserve space in the system space info necessary
4400 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4403 void check_system_chunk(struct btrfs_trans_handle
*trans
,
4404 struct btrfs_root
*root
,
4407 struct btrfs_space_info
*info
;
4414 * Needed because we can end up allocating a system chunk and for an
4415 * atomic and race free space reservation in the chunk block reserve.
4417 ASSERT(mutex_is_locked(&root
->fs_info
->chunk_mutex
));
4419 info
= __find_space_info(root
->fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
4420 spin_lock(&info
->lock
);
4421 left
= info
->total_bytes
- info
->bytes_used
- info
->bytes_pinned
-
4422 info
->bytes_reserved
- info
->bytes_readonly
-
4423 info
->bytes_may_use
;
4424 spin_unlock(&info
->lock
);
4426 num_devs
= get_profile_num_devs(root
, type
);
4428 /* num_devs device items to update and 1 chunk item to add or remove */
4429 thresh
= btrfs_calc_trunc_metadata_size(root
, num_devs
) +
4430 btrfs_calc_trans_metadata_size(root
, 1);
4432 if (left
< thresh
&& btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
4433 btrfs_info(root
->fs_info
, "left=%llu, need=%llu, flags=%llu",
4434 left
, thresh
, type
);
4435 dump_space_info(info
, 0, 0);
4438 if (left
< thresh
) {
4441 flags
= btrfs_get_alloc_profile(root
->fs_info
->chunk_root
, 0);
4443 * Ignore failure to create system chunk. We might end up not
4444 * needing it, as we might not need to COW all nodes/leafs from
4445 * the paths we visit in the chunk tree (they were already COWed
4446 * or created in the current transaction for example).
4448 ret
= btrfs_alloc_chunk(trans
, root
, flags
);
4452 ret
= btrfs_block_rsv_add(root
->fs_info
->chunk_root
,
4453 &root
->fs_info
->chunk_block_rsv
,
4454 thresh
, BTRFS_RESERVE_NO_FLUSH
);
4456 trans
->chunk_bytes_reserved
+= thresh
;
4460 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
4461 struct btrfs_root
*extent_root
, u64 flags
, int force
)
4463 struct btrfs_space_info
*space_info
;
4464 struct btrfs_fs_info
*fs_info
= extent_root
->fs_info
;
4465 int wait_for_alloc
= 0;
4468 /* Don't re-enter if we're already allocating a chunk */
4469 if (trans
->allocating_chunk
)
4472 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
4474 ret
= update_space_info(extent_root
->fs_info
, flags
,
4475 0, 0, 0, &space_info
);
4476 BUG_ON(ret
); /* -ENOMEM */
4478 BUG_ON(!space_info
); /* Logic error */
4481 spin_lock(&space_info
->lock
);
4482 if (force
< space_info
->force_alloc
)
4483 force
= space_info
->force_alloc
;
4484 if (space_info
->full
) {
4485 if (should_alloc_chunk(extent_root
, space_info
, force
))
4489 spin_unlock(&space_info
->lock
);
4493 if (!should_alloc_chunk(extent_root
, space_info
, force
)) {
4494 spin_unlock(&space_info
->lock
);
4496 } else if (space_info
->chunk_alloc
) {
4499 space_info
->chunk_alloc
= 1;
4502 spin_unlock(&space_info
->lock
);
4504 mutex_lock(&fs_info
->chunk_mutex
);
4507 * The chunk_mutex is held throughout the entirety of a chunk
4508 * allocation, so once we've acquired the chunk_mutex we know that the
4509 * other guy is done and we need to recheck and see if we should
4512 if (wait_for_alloc
) {
4513 mutex_unlock(&fs_info
->chunk_mutex
);
4518 trans
->allocating_chunk
= true;
4521 * If we have mixed data/metadata chunks we want to make sure we keep
4522 * allocating mixed chunks instead of individual chunks.
4524 if (btrfs_mixed_space_info(space_info
))
4525 flags
|= (BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
);
4528 * if we're doing a data chunk, go ahead and make sure that
4529 * we keep a reasonable number of metadata chunks allocated in the
4532 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
4533 fs_info
->data_chunk_allocations
++;
4534 if (!(fs_info
->data_chunk_allocations
%
4535 fs_info
->metadata_ratio
))
4536 force_metadata_allocation(fs_info
);
4540 * Check if we have enough space in SYSTEM chunk because we may need
4541 * to update devices.
4543 check_system_chunk(trans
, extent_root
, flags
);
4545 ret
= btrfs_alloc_chunk(trans
, extent_root
, flags
);
4546 trans
->allocating_chunk
= false;
4548 spin_lock(&space_info
->lock
);
4549 if (ret
< 0 && ret
!= -ENOSPC
)
4552 space_info
->full
= 1;
4556 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
4558 space_info
->chunk_alloc
= 0;
4559 spin_unlock(&space_info
->lock
);
4560 mutex_unlock(&fs_info
->chunk_mutex
);
4562 * When we allocate a new chunk we reserve space in the chunk block
4563 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4564 * add new nodes/leafs to it if we end up needing to do it when
4565 * inserting the chunk item and updating device items as part of the
4566 * second phase of chunk allocation, performed by
4567 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4568 * large number of new block groups to create in our transaction
4569 * handle's new_bgs list to avoid exhausting the chunk block reserve
4570 * in extreme cases - like having a single transaction create many new
4571 * block groups when starting to write out the free space caches of all
4572 * the block groups that were made dirty during the lifetime of the
4575 if (trans
->can_flush_pending_bgs
&&
4576 trans
->chunk_bytes_reserved
>= (u64
)SZ_2M
) {
4577 btrfs_create_pending_block_groups(trans
, trans
->root
);
4578 btrfs_trans_release_chunk_metadata(trans
);
4583 static int can_overcommit(struct btrfs_root
*root
,
4584 struct btrfs_space_info
*space_info
, u64 bytes
,
4585 enum btrfs_reserve_flush_enum flush
)
4587 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
4588 u64 profile
= btrfs_get_alloc_profile(root
, 0);
4593 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4594 space_info
->bytes_pinned
+ space_info
->bytes_readonly
;
4597 * We only want to allow over committing if we have lots of actual space
4598 * free, but if we don't have enough space to handle the global reserve
4599 * space then we could end up having a real enospc problem when trying
4600 * to allocate a chunk or some other such important allocation.
4602 spin_lock(&global_rsv
->lock
);
4603 space_size
= calc_global_rsv_need_space(global_rsv
);
4604 spin_unlock(&global_rsv
->lock
);
4605 if (used
+ space_size
>= space_info
->total_bytes
)
4608 used
+= space_info
->bytes_may_use
;
4610 spin_lock(&root
->fs_info
->free_chunk_lock
);
4611 avail
= root
->fs_info
->free_chunk_space
;
4612 spin_unlock(&root
->fs_info
->free_chunk_lock
);
4615 * If we have dup, raid1 or raid10 then only half of the free
4616 * space is actually useable. For raid56, the space info used
4617 * doesn't include the parity drive, so we don't have to
4620 if (profile
& (BTRFS_BLOCK_GROUP_DUP
|
4621 BTRFS_BLOCK_GROUP_RAID1
|
4622 BTRFS_BLOCK_GROUP_RAID10
))
4626 * If we aren't flushing all things, let us overcommit up to
4627 * 1/2th of the space. If we can flush, don't let us overcommit
4628 * too much, let it overcommit up to 1/8 of the space.
4630 if (flush
== BTRFS_RESERVE_FLUSH_ALL
)
4635 if (used
+ bytes
< space_info
->total_bytes
+ avail
)
4640 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root
*root
,
4641 unsigned long nr_pages
, int nr_items
)
4643 struct super_block
*sb
= root
->fs_info
->sb
;
4645 if (down_read_trylock(&sb
->s_umount
)) {
4646 writeback_inodes_sb_nr(sb
, nr_pages
, WB_REASON_FS_FREE_SPACE
);
4647 up_read(&sb
->s_umount
);
4650 * We needn't worry the filesystem going from r/w to r/o though
4651 * we don't acquire ->s_umount mutex, because the filesystem
4652 * should guarantee the delalloc inodes list be empty after
4653 * the filesystem is readonly(all dirty pages are written to
4656 btrfs_start_delalloc_roots(root
->fs_info
, 0, nr_items
);
4657 if (!current
->journal_info
)
4658 btrfs_wait_ordered_roots(root
->fs_info
, nr_items
,
4663 static inline int calc_reclaim_items_nr(struct btrfs_root
*root
, u64 to_reclaim
)
4668 bytes
= btrfs_calc_trans_metadata_size(root
, 1);
4669 nr
= (int)div64_u64(to_reclaim
, bytes
);
4675 #define EXTENT_SIZE_PER_ITEM SZ_256K
4678 * shrink metadata reservation for delalloc
4680 static void shrink_delalloc(struct btrfs_root
*root
, u64 to_reclaim
, u64 orig
,
4683 struct btrfs_block_rsv
*block_rsv
;
4684 struct btrfs_space_info
*space_info
;
4685 struct btrfs_trans_handle
*trans
;
4689 unsigned long nr_pages
;
4692 enum btrfs_reserve_flush_enum flush
;
4694 /* Calc the number of the pages we need flush for space reservation */
4695 items
= calc_reclaim_items_nr(root
, to_reclaim
);
4696 to_reclaim
= (u64
)items
* EXTENT_SIZE_PER_ITEM
;
4698 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4699 block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
4700 space_info
= block_rsv
->space_info
;
4702 delalloc_bytes
= percpu_counter_sum_positive(
4703 &root
->fs_info
->delalloc_bytes
);
4704 if (delalloc_bytes
== 0) {
4708 btrfs_wait_ordered_roots(root
->fs_info
, items
,
4714 while (delalloc_bytes
&& loops
< 3) {
4715 max_reclaim
= min(delalloc_bytes
, to_reclaim
);
4716 nr_pages
= max_reclaim
>> PAGE_SHIFT
;
4717 btrfs_writeback_inodes_sb_nr(root
, nr_pages
, items
);
4719 * We need to wait for the async pages to actually start before
4722 max_reclaim
= atomic_read(&root
->fs_info
->async_delalloc_pages
);
4726 if (max_reclaim
<= nr_pages
)
4729 max_reclaim
-= nr_pages
;
4731 wait_event(root
->fs_info
->async_submit_wait
,
4732 atomic_read(&root
->fs_info
->async_delalloc_pages
) <=
4736 flush
= BTRFS_RESERVE_FLUSH_ALL
;
4738 flush
= BTRFS_RESERVE_NO_FLUSH
;
4739 spin_lock(&space_info
->lock
);
4740 if (can_overcommit(root
, space_info
, orig
, flush
)) {
4741 spin_unlock(&space_info
->lock
);
4744 spin_unlock(&space_info
->lock
);
4747 if (wait_ordered
&& !trans
) {
4748 btrfs_wait_ordered_roots(root
->fs_info
, items
,
4751 time_left
= schedule_timeout_killable(1);
4755 delalloc_bytes
= percpu_counter_sum_positive(
4756 &root
->fs_info
->delalloc_bytes
);
4761 * maybe_commit_transaction - possibly commit the transaction if its ok to
4762 * @root - the root we're allocating for
4763 * @bytes - the number of bytes we want to reserve
4764 * @force - force the commit
4766 * This will check to make sure that committing the transaction will actually
4767 * get us somewhere and then commit the transaction if it does. Otherwise it
4768 * will return -ENOSPC.
4770 static int may_commit_transaction(struct btrfs_root
*root
,
4771 struct btrfs_space_info
*space_info
,
4772 u64 bytes
, int force
)
4774 struct btrfs_block_rsv
*delayed_rsv
= &root
->fs_info
->delayed_block_rsv
;
4775 struct btrfs_trans_handle
*trans
;
4777 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4784 /* See if there is enough pinned space to make this reservation */
4785 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4790 * See if there is some space in the delayed insertion reservation for
4793 if (space_info
!= delayed_rsv
->space_info
)
4796 spin_lock(&delayed_rsv
->lock
);
4797 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4798 bytes
- delayed_rsv
->size
) >= 0) {
4799 spin_unlock(&delayed_rsv
->lock
);
4802 spin_unlock(&delayed_rsv
->lock
);
4805 trans
= btrfs_join_transaction(root
);
4809 return btrfs_commit_transaction(trans
, root
);
4813 FLUSH_DELAYED_ITEMS_NR
= 1,
4814 FLUSH_DELAYED_ITEMS
= 2,
4816 FLUSH_DELALLOC_WAIT
= 4,
4821 static int flush_space(struct btrfs_root
*root
,
4822 struct btrfs_space_info
*space_info
, u64 num_bytes
,
4823 u64 orig_bytes
, int state
)
4825 struct btrfs_trans_handle
*trans
;
4830 case FLUSH_DELAYED_ITEMS_NR
:
4831 case FLUSH_DELAYED_ITEMS
:
4832 if (state
== FLUSH_DELAYED_ITEMS_NR
)
4833 nr
= calc_reclaim_items_nr(root
, num_bytes
) * 2;
4837 trans
= btrfs_join_transaction(root
);
4838 if (IS_ERR(trans
)) {
4839 ret
= PTR_ERR(trans
);
4842 ret
= btrfs_run_delayed_items_nr(trans
, root
, nr
);
4843 btrfs_end_transaction(trans
, root
);
4845 case FLUSH_DELALLOC
:
4846 case FLUSH_DELALLOC_WAIT
:
4847 shrink_delalloc(root
, num_bytes
* 2, orig_bytes
,
4848 state
== FLUSH_DELALLOC_WAIT
);
4851 trans
= btrfs_join_transaction(root
);
4852 if (IS_ERR(trans
)) {
4853 ret
= PTR_ERR(trans
);
4856 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
4857 btrfs_get_alloc_profile(root
, 0),
4858 CHUNK_ALLOC_NO_FORCE
);
4859 btrfs_end_transaction(trans
, root
);
4864 ret
= may_commit_transaction(root
, space_info
, orig_bytes
, 0);
4875 btrfs_calc_reclaim_metadata_size(struct btrfs_root
*root
,
4876 struct btrfs_space_info
*space_info
)
4882 to_reclaim
= min_t(u64
, num_online_cpus() * SZ_1M
, SZ_16M
);
4883 spin_lock(&space_info
->lock
);
4884 if (can_overcommit(root
, space_info
, to_reclaim
,
4885 BTRFS_RESERVE_FLUSH_ALL
)) {
4890 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4891 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4892 space_info
->bytes_may_use
;
4893 if (can_overcommit(root
, space_info
, SZ_1M
, BTRFS_RESERVE_FLUSH_ALL
))
4894 expected
= div_factor_fine(space_info
->total_bytes
, 95);
4896 expected
= div_factor_fine(space_info
->total_bytes
, 90);
4898 if (used
> expected
)
4899 to_reclaim
= used
- expected
;
4902 to_reclaim
= min(to_reclaim
, space_info
->bytes_may_use
+
4903 space_info
->bytes_reserved
);
4905 spin_unlock(&space_info
->lock
);
4910 static inline int need_do_async_reclaim(struct btrfs_space_info
*space_info
,
4911 struct btrfs_fs_info
*fs_info
, u64 used
)
4913 u64 thresh
= div_factor_fine(space_info
->total_bytes
, 98);
4915 /* If we're just plain full then async reclaim just slows us down. */
4916 if ((space_info
->bytes_used
+ space_info
->bytes_reserved
) >= thresh
)
4919 return (used
>= thresh
&& !btrfs_fs_closing(fs_info
) &&
4920 !test_bit(BTRFS_FS_STATE_REMOUNTING
, &fs_info
->fs_state
));
4923 static int btrfs_need_do_async_reclaim(struct btrfs_space_info
*space_info
,
4924 struct btrfs_fs_info
*fs_info
,
4929 spin_lock(&space_info
->lock
);
4931 * We run out of space and have not got any free space via flush_space,
4932 * so don't bother doing async reclaim.
4934 if (flush_state
> COMMIT_TRANS
&& space_info
->full
) {
4935 spin_unlock(&space_info
->lock
);
4939 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4940 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4941 space_info
->bytes_may_use
;
4942 if (need_do_async_reclaim(space_info
, fs_info
, used
)) {
4943 spin_unlock(&space_info
->lock
);
4946 spin_unlock(&space_info
->lock
);
4951 static void btrfs_async_reclaim_metadata_space(struct work_struct
*work
)
4953 struct btrfs_fs_info
*fs_info
;
4954 struct btrfs_space_info
*space_info
;
4958 fs_info
= container_of(work
, struct btrfs_fs_info
, async_reclaim_work
);
4959 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4961 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
->fs_root
,
4966 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
4968 flush_space(fs_info
->fs_root
, space_info
, to_reclaim
,
4969 to_reclaim
, flush_state
);
4971 if (!btrfs_need_do_async_reclaim(space_info
, fs_info
,
4974 } while (flush_state
< COMMIT_TRANS
);
4977 void btrfs_init_async_reclaim_work(struct work_struct
*work
)
4979 INIT_WORK(work
, btrfs_async_reclaim_metadata_space
);
4983 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4984 * @root - the root we're allocating for
4985 * @block_rsv - the block_rsv we're allocating for
4986 * @orig_bytes - the number of bytes we want
4987 * @flush - whether or not we can flush to make our reservation
4989 * This will reserve orig_bytes number of bytes from the space info associated
4990 * with the block_rsv. If there is not enough space it will make an attempt to
4991 * flush out space to make room. It will do this by flushing delalloc if
4992 * possible or committing the transaction. If flush is 0 then no attempts to
4993 * regain reservations will be made and this will fail if there is not enough
4996 static int reserve_metadata_bytes(struct btrfs_root
*root
,
4997 struct btrfs_block_rsv
*block_rsv
,
4999 enum btrfs_reserve_flush_enum flush
)
5001 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
5003 u64 num_bytes
= orig_bytes
;
5004 int flush_state
= FLUSH_DELAYED_ITEMS_NR
;
5006 bool flushing
= false;
5010 spin_lock(&space_info
->lock
);
5012 * We only want to wait if somebody other than us is flushing and we
5013 * are actually allowed to flush all things.
5015 while (flush
== BTRFS_RESERVE_FLUSH_ALL
&& !flushing
&&
5016 space_info
->flush
) {
5017 spin_unlock(&space_info
->lock
);
5019 * If we have a trans handle we can't wait because the flusher
5020 * may have to commit the transaction, which would mean we would
5021 * deadlock since we are waiting for the flusher to finish, but
5022 * hold the current transaction open.
5024 if (current
->journal_info
)
5026 ret
= wait_event_killable(space_info
->wait
, !space_info
->flush
);
5027 /* Must have been killed, return */
5031 spin_lock(&space_info
->lock
);
5035 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
5036 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
5037 space_info
->bytes_may_use
;
5040 * The idea here is that we've not already over-reserved the block group
5041 * then we can go ahead and save our reservation first and then start
5042 * flushing if we need to. Otherwise if we've already overcommitted
5043 * lets start flushing stuff first and then come back and try to make
5046 if (used
<= space_info
->total_bytes
) {
5047 if (used
+ orig_bytes
<= space_info
->total_bytes
) {
5048 space_info
->bytes_may_use
+= orig_bytes
;
5049 trace_btrfs_space_reservation(root
->fs_info
,
5050 "space_info", space_info
->flags
, orig_bytes
, 1);
5054 * Ok set num_bytes to orig_bytes since we aren't
5055 * overocmmitted, this way we only try and reclaim what
5058 num_bytes
= orig_bytes
;
5062 * Ok we're over committed, set num_bytes to the overcommitted
5063 * amount plus the amount of bytes that we need for this
5066 num_bytes
= used
- space_info
->total_bytes
+
5070 if (ret
&& can_overcommit(root
, space_info
, orig_bytes
, flush
)) {
5071 space_info
->bytes_may_use
+= orig_bytes
;
5072 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
5073 space_info
->flags
, orig_bytes
,
5079 * Couldn't make our reservation, save our place so while we're trying
5080 * to reclaim space we can actually use it instead of somebody else
5081 * stealing it from us.
5083 * We make the other tasks wait for the flush only when we can flush
5086 if (ret
&& flush
!= BTRFS_RESERVE_NO_FLUSH
) {
5088 space_info
->flush
= 1;
5089 } else if (!ret
&& space_info
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
5092 * We will do the space reservation dance during log replay,
5093 * which means we won't have fs_info->fs_root set, so don't do
5094 * the async reclaim as we will panic.
5096 if (!root
->fs_info
->log_root_recovering
&&
5097 need_do_async_reclaim(space_info
, root
->fs_info
, used
) &&
5098 !work_busy(&root
->fs_info
->async_reclaim_work
))
5099 queue_work(system_unbound_wq
,
5100 &root
->fs_info
->async_reclaim_work
);
5102 spin_unlock(&space_info
->lock
);
5104 if (!ret
|| flush
== BTRFS_RESERVE_NO_FLUSH
)
5107 ret
= flush_space(root
, space_info
, num_bytes
, orig_bytes
,
5112 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
5113 * would happen. So skip delalloc flush.
5115 if (flush
== BTRFS_RESERVE_FLUSH_LIMIT
&&
5116 (flush_state
== FLUSH_DELALLOC
||
5117 flush_state
== FLUSH_DELALLOC_WAIT
))
5118 flush_state
= ALLOC_CHUNK
;
5122 else if (flush
== BTRFS_RESERVE_FLUSH_LIMIT
&&
5123 flush_state
< COMMIT_TRANS
)
5125 else if (flush
== BTRFS_RESERVE_FLUSH_ALL
&&
5126 flush_state
<= COMMIT_TRANS
)
5130 if (ret
== -ENOSPC
&&
5131 unlikely(root
->orphan_cleanup_state
== ORPHAN_CLEANUP_STARTED
)) {
5132 struct btrfs_block_rsv
*global_rsv
=
5133 &root
->fs_info
->global_block_rsv
;
5135 if (block_rsv
!= global_rsv
&&
5136 !block_rsv_use_bytes(global_rsv
, orig_bytes
))
5140 trace_btrfs_space_reservation(root
->fs_info
,
5141 "space_info:enospc",
5142 space_info
->flags
, orig_bytes
, 1);
5144 spin_lock(&space_info
->lock
);
5145 space_info
->flush
= 0;
5146 wake_up_all(&space_info
->wait
);
5147 spin_unlock(&space_info
->lock
);
5152 static struct btrfs_block_rsv
*get_block_rsv(
5153 const struct btrfs_trans_handle
*trans
,
5154 const struct btrfs_root
*root
)
5156 struct btrfs_block_rsv
*block_rsv
= NULL
;
5158 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
5159 (root
== root
->fs_info
->csum_root
&& trans
->adding_csums
) ||
5160 (root
== root
->fs_info
->uuid_root
))
5161 block_rsv
= trans
->block_rsv
;
5164 block_rsv
= root
->block_rsv
;
5167 block_rsv
= &root
->fs_info
->empty_block_rsv
;
5172 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
5176 spin_lock(&block_rsv
->lock
);
5177 if (block_rsv
->reserved
>= num_bytes
) {
5178 block_rsv
->reserved
-= num_bytes
;
5179 if (block_rsv
->reserved
< block_rsv
->size
)
5180 block_rsv
->full
= 0;
5183 spin_unlock(&block_rsv
->lock
);
5187 static void block_rsv_add_bytes(struct btrfs_block_rsv
*block_rsv
,
5188 u64 num_bytes
, int update_size
)
5190 spin_lock(&block_rsv
->lock
);
5191 block_rsv
->reserved
+= num_bytes
;
5193 block_rsv
->size
+= num_bytes
;
5194 else if (block_rsv
->reserved
>= block_rsv
->size
)
5195 block_rsv
->full
= 1;
5196 spin_unlock(&block_rsv
->lock
);
5199 int btrfs_cond_migrate_bytes(struct btrfs_fs_info
*fs_info
,
5200 struct btrfs_block_rsv
*dest
, u64 num_bytes
,
5203 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5206 if (global_rsv
->space_info
!= dest
->space_info
)
5209 spin_lock(&global_rsv
->lock
);
5210 min_bytes
= div_factor(global_rsv
->size
, min_factor
);
5211 if (global_rsv
->reserved
< min_bytes
+ num_bytes
) {
5212 spin_unlock(&global_rsv
->lock
);
5215 global_rsv
->reserved
-= num_bytes
;
5216 if (global_rsv
->reserved
< global_rsv
->size
)
5217 global_rsv
->full
= 0;
5218 spin_unlock(&global_rsv
->lock
);
5220 block_rsv_add_bytes(dest
, num_bytes
, 1);
5224 static void block_rsv_release_bytes(struct btrfs_fs_info
*fs_info
,
5225 struct btrfs_block_rsv
*block_rsv
,
5226 struct btrfs_block_rsv
*dest
, u64 num_bytes
)
5228 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
5230 spin_lock(&block_rsv
->lock
);
5231 if (num_bytes
== (u64
)-1)
5232 num_bytes
= block_rsv
->size
;
5233 block_rsv
->size
-= num_bytes
;
5234 if (block_rsv
->reserved
>= block_rsv
->size
) {
5235 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
5236 block_rsv
->reserved
= block_rsv
->size
;
5237 block_rsv
->full
= 1;
5241 spin_unlock(&block_rsv
->lock
);
5243 if (num_bytes
> 0) {
5245 spin_lock(&dest
->lock
);
5249 bytes_to_add
= dest
->size
- dest
->reserved
;
5250 bytes_to_add
= min(num_bytes
, bytes_to_add
);
5251 dest
->reserved
+= bytes_to_add
;
5252 if (dest
->reserved
>= dest
->size
)
5254 num_bytes
-= bytes_to_add
;
5256 spin_unlock(&dest
->lock
);
5259 spin_lock(&space_info
->lock
);
5260 space_info
->bytes_may_use
-= num_bytes
;
5261 trace_btrfs_space_reservation(fs_info
, "space_info",
5262 space_info
->flags
, num_bytes
, 0);
5263 spin_unlock(&space_info
->lock
);
5268 int btrfs_block_rsv_migrate(struct btrfs_block_rsv
*src
,
5269 struct btrfs_block_rsv
*dst
, u64 num_bytes
,
5274 ret
= block_rsv_use_bytes(src
, num_bytes
);
5278 block_rsv_add_bytes(dst
, num_bytes
, update_size
);
5282 void btrfs_init_block_rsv(struct btrfs_block_rsv
*rsv
, unsigned short type
)
5284 memset(rsv
, 0, sizeof(*rsv
));
5285 spin_lock_init(&rsv
->lock
);
5289 struct btrfs_block_rsv
*btrfs_alloc_block_rsv(struct btrfs_root
*root
,
5290 unsigned short type
)
5292 struct btrfs_block_rsv
*block_rsv
;
5293 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5295 block_rsv
= kmalloc(sizeof(*block_rsv
), GFP_NOFS
);
5299 btrfs_init_block_rsv(block_rsv
, type
);
5300 block_rsv
->space_info
= __find_space_info(fs_info
,
5301 BTRFS_BLOCK_GROUP_METADATA
);
5305 void btrfs_free_block_rsv(struct btrfs_root
*root
,
5306 struct btrfs_block_rsv
*rsv
)
5310 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
5314 void __btrfs_free_block_rsv(struct btrfs_block_rsv
*rsv
)
5319 int btrfs_block_rsv_add(struct btrfs_root
*root
,
5320 struct btrfs_block_rsv
*block_rsv
, u64 num_bytes
,
5321 enum btrfs_reserve_flush_enum flush
)
5328 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
5330 block_rsv_add_bytes(block_rsv
, num_bytes
, 1);
5337 int btrfs_block_rsv_check(struct btrfs_root
*root
,
5338 struct btrfs_block_rsv
*block_rsv
, int min_factor
)
5346 spin_lock(&block_rsv
->lock
);
5347 num_bytes
= div_factor(block_rsv
->size
, min_factor
);
5348 if (block_rsv
->reserved
>= num_bytes
)
5350 spin_unlock(&block_rsv
->lock
);
5355 int btrfs_block_rsv_refill(struct btrfs_root
*root
,
5356 struct btrfs_block_rsv
*block_rsv
, u64 min_reserved
,
5357 enum btrfs_reserve_flush_enum flush
)
5365 spin_lock(&block_rsv
->lock
);
5366 num_bytes
= min_reserved
;
5367 if (block_rsv
->reserved
>= num_bytes
)
5370 num_bytes
-= block_rsv
->reserved
;
5371 spin_unlock(&block_rsv
->lock
);
5376 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
5378 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
5385 void btrfs_block_rsv_release(struct btrfs_root
*root
,
5386 struct btrfs_block_rsv
*block_rsv
,
5389 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
5390 if (global_rsv
== block_rsv
||
5391 block_rsv
->space_info
!= global_rsv
->space_info
)
5393 block_rsv_release_bytes(root
->fs_info
, block_rsv
, global_rsv
,
5398 * helper to calculate size of global block reservation.
5399 * the desired value is sum of space used by extent tree,
5400 * checksum tree and root tree
5402 static u64
calc_global_metadata_size(struct btrfs_fs_info
*fs_info
)
5404 struct btrfs_space_info
*sinfo
;
5408 int csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
5410 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_DATA
);
5411 spin_lock(&sinfo
->lock
);
5412 data_used
= sinfo
->bytes_used
;
5413 spin_unlock(&sinfo
->lock
);
5415 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
5416 spin_lock(&sinfo
->lock
);
5417 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_DATA
)
5419 meta_used
= sinfo
->bytes_used
;
5420 spin_unlock(&sinfo
->lock
);
5422 num_bytes
= (data_used
>> fs_info
->sb
->s_blocksize_bits
) *
5424 num_bytes
+= div_u64(data_used
+ meta_used
, 50);
5426 if (num_bytes
* 3 > meta_used
)
5427 num_bytes
= div_u64(meta_used
, 3);
5429 return ALIGN(num_bytes
, fs_info
->extent_root
->nodesize
<< 10);
5432 static void update_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5434 struct btrfs_block_rsv
*block_rsv
= &fs_info
->global_block_rsv
;
5435 struct btrfs_space_info
*sinfo
= block_rsv
->space_info
;
5438 num_bytes
= calc_global_metadata_size(fs_info
);
5440 spin_lock(&sinfo
->lock
);
5441 spin_lock(&block_rsv
->lock
);
5443 block_rsv
->size
= min_t(u64
, num_bytes
, SZ_512M
);
5445 if (block_rsv
->reserved
< block_rsv
->size
) {
5446 num_bytes
= sinfo
->bytes_used
+ sinfo
->bytes_pinned
+
5447 sinfo
->bytes_reserved
+ sinfo
->bytes_readonly
+
5448 sinfo
->bytes_may_use
;
5449 if (sinfo
->total_bytes
> num_bytes
) {
5450 num_bytes
= sinfo
->total_bytes
- num_bytes
;
5451 num_bytes
= min(num_bytes
,
5452 block_rsv
->size
- block_rsv
->reserved
);
5453 block_rsv
->reserved
+= num_bytes
;
5454 sinfo
->bytes_may_use
+= num_bytes
;
5455 trace_btrfs_space_reservation(fs_info
, "space_info",
5456 sinfo
->flags
, num_bytes
,
5459 } else if (block_rsv
->reserved
> block_rsv
->size
) {
5460 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
5461 sinfo
->bytes_may_use
-= num_bytes
;
5462 trace_btrfs_space_reservation(fs_info
, "space_info",
5463 sinfo
->flags
, num_bytes
, 0);
5464 block_rsv
->reserved
= block_rsv
->size
;
5467 if (block_rsv
->reserved
== block_rsv
->size
)
5468 block_rsv
->full
= 1;
5470 block_rsv
->full
= 0;
5472 spin_unlock(&block_rsv
->lock
);
5473 spin_unlock(&sinfo
->lock
);
5476 static void init_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5478 struct btrfs_space_info
*space_info
;
5480 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
5481 fs_info
->chunk_block_rsv
.space_info
= space_info
;
5483 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
5484 fs_info
->global_block_rsv
.space_info
= space_info
;
5485 fs_info
->delalloc_block_rsv
.space_info
= space_info
;
5486 fs_info
->trans_block_rsv
.space_info
= space_info
;
5487 fs_info
->empty_block_rsv
.space_info
= space_info
;
5488 fs_info
->delayed_block_rsv
.space_info
= space_info
;
5490 fs_info
->extent_root
->block_rsv
= &fs_info
->global_block_rsv
;
5491 fs_info
->csum_root
->block_rsv
= &fs_info
->global_block_rsv
;
5492 fs_info
->dev_root
->block_rsv
= &fs_info
->global_block_rsv
;
5493 fs_info
->tree_root
->block_rsv
= &fs_info
->global_block_rsv
;
5494 if (fs_info
->quota_root
)
5495 fs_info
->quota_root
->block_rsv
= &fs_info
->global_block_rsv
;
5496 fs_info
->chunk_root
->block_rsv
= &fs_info
->chunk_block_rsv
;
5498 update_global_block_rsv(fs_info
);
5501 static void release_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5503 block_rsv_release_bytes(fs_info
, &fs_info
->global_block_rsv
, NULL
,
5505 WARN_ON(fs_info
->delalloc_block_rsv
.size
> 0);
5506 WARN_ON(fs_info
->delalloc_block_rsv
.reserved
> 0);
5507 WARN_ON(fs_info
->trans_block_rsv
.size
> 0);
5508 WARN_ON(fs_info
->trans_block_rsv
.reserved
> 0);
5509 WARN_ON(fs_info
->chunk_block_rsv
.size
> 0);
5510 WARN_ON(fs_info
->chunk_block_rsv
.reserved
> 0);
5511 WARN_ON(fs_info
->delayed_block_rsv
.size
> 0);
5512 WARN_ON(fs_info
->delayed_block_rsv
.reserved
> 0);
5515 void btrfs_trans_release_metadata(struct btrfs_trans_handle
*trans
,
5516 struct btrfs_root
*root
)
5518 if (!trans
->block_rsv
)
5521 if (!trans
->bytes_reserved
)
5524 trace_btrfs_space_reservation(root
->fs_info
, "transaction",
5525 trans
->transid
, trans
->bytes_reserved
, 0);
5526 btrfs_block_rsv_release(root
, trans
->block_rsv
, trans
->bytes_reserved
);
5527 trans
->bytes_reserved
= 0;
5531 * To be called after all the new block groups attached to the transaction
5532 * handle have been created (btrfs_create_pending_block_groups()).
5534 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle
*trans
)
5536 struct btrfs_fs_info
*fs_info
= trans
->root
->fs_info
;
5538 if (!trans
->chunk_bytes_reserved
)
5541 WARN_ON_ONCE(!list_empty(&trans
->new_bgs
));
5543 block_rsv_release_bytes(fs_info
, &fs_info
->chunk_block_rsv
, NULL
,
5544 trans
->chunk_bytes_reserved
);
5545 trans
->chunk_bytes_reserved
= 0;
5548 /* Can only return 0 or -ENOSPC */
5549 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle
*trans
,
5550 struct inode
*inode
)
5552 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5553 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
5554 struct btrfs_block_rsv
*dst_rsv
= root
->orphan_block_rsv
;
5557 * We need to hold space in order to delete our orphan item once we've
5558 * added it, so this takes the reservation so we can release it later
5559 * when we are truly done with the orphan item.
5561 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
5562 trace_btrfs_space_reservation(root
->fs_info
, "orphan",
5563 btrfs_ino(inode
), num_bytes
, 1);
5564 return btrfs_block_rsv_migrate(src_rsv
, dst_rsv
, num_bytes
, 1);
5567 void btrfs_orphan_release_metadata(struct inode
*inode
)
5569 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5570 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
5571 trace_btrfs_space_reservation(root
->fs_info
, "orphan",
5572 btrfs_ino(inode
), num_bytes
, 0);
5573 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
, num_bytes
);
5577 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5578 * root: the root of the parent directory
5579 * rsv: block reservation
5580 * items: the number of items that we need do reservation
5581 * qgroup_reserved: used to return the reserved size in qgroup
5583 * This function is used to reserve the space for snapshot/subvolume
5584 * creation and deletion. Those operations are different with the
5585 * common file/directory operations, they change two fs/file trees
5586 * and root tree, the number of items that the qgroup reserves is
5587 * different with the free space reservation. So we can not use
5588 * the space reservation mechanism in start_transaction().
5590 int btrfs_subvolume_reserve_metadata(struct btrfs_root
*root
,
5591 struct btrfs_block_rsv
*rsv
,
5593 u64
*qgroup_reserved
,
5594 bool use_global_rsv
)
5598 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
5600 if (root
->fs_info
->quota_enabled
) {
5601 /* One for parent inode, two for dir entries */
5602 num_bytes
= 3 * root
->nodesize
;
5603 ret
= btrfs_qgroup_reserve_meta(root
, num_bytes
);
5610 *qgroup_reserved
= num_bytes
;
5612 num_bytes
= btrfs_calc_trans_metadata_size(root
, items
);
5613 rsv
->space_info
= __find_space_info(root
->fs_info
,
5614 BTRFS_BLOCK_GROUP_METADATA
);
5615 ret
= btrfs_block_rsv_add(root
, rsv
, num_bytes
,
5616 BTRFS_RESERVE_FLUSH_ALL
);
5618 if (ret
== -ENOSPC
&& use_global_rsv
)
5619 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
, num_bytes
, 1);
5621 if (ret
&& *qgroup_reserved
)
5622 btrfs_qgroup_free_meta(root
, *qgroup_reserved
);
5627 void btrfs_subvolume_release_metadata(struct btrfs_root
*root
,
5628 struct btrfs_block_rsv
*rsv
,
5629 u64 qgroup_reserved
)
5631 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
5635 * drop_outstanding_extent - drop an outstanding extent
5636 * @inode: the inode we're dropping the extent for
5637 * @num_bytes: the number of bytes we're releasing.
5639 * This is called when we are freeing up an outstanding extent, either called
5640 * after an error or after an extent is written. This will return the number of
5641 * reserved extents that need to be freed. This must be called with
5642 * BTRFS_I(inode)->lock held.
5644 static unsigned drop_outstanding_extent(struct inode
*inode
, u64 num_bytes
)
5646 unsigned drop_inode_space
= 0;
5647 unsigned dropped_extents
= 0;
5648 unsigned num_extents
= 0;
5650 num_extents
= (unsigned)div64_u64(num_bytes
+
5651 BTRFS_MAX_EXTENT_SIZE
- 1,
5652 BTRFS_MAX_EXTENT_SIZE
);
5653 ASSERT(num_extents
);
5654 ASSERT(BTRFS_I(inode
)->outstanding_extents
>= num_extents
);
5655 BTRFS_I(inode
)->outstanding_extents
-= num_extents
;
5657 if (BTRFS_I(inode
)->outstanding_extents
== 0 &&
5658 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5659 &BTRFS_I(inode
)->runtime_flags
))
5660 drop_inode_space
= 1;
5663 * If we have more or the same amount of outstanding extents than we have
5664 * reserved then we need to leave the reserved extents count alone.
5666 if (BTRFS_I(inode
)->outstanding_extents
>=
5667 BTRFS_I(inode
)->reserved_extents
)
5668 return drop_inode_space
;
5670 dropped_extents
= BTRFS_I(inode
)->reserved_extents
-
5671 BTRFS_I(inode
)->outstanding_extents
;
5672 BTRFS_I(inode
)->reserved_extents
-= dropped_extents
;
5673 return dropped_extents
+ drop_inode_space
;
5677 * calc_csum_metadata_size - return the amount of metadata space that must be
5678 * reserved/freed for the given bytes.
5679 * @inode: the inode we're manipulating
5680 * @num_bytes: the number of bytes in question
5681 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5683 * This adjusts the number of csum_bytes in the inode and then returns the
5684 * correct amount of metadata that must either be reserved or freed. We
5685 * calculate how many checksums we can fit into one leaf and then divide the
5686 * number of bytes that will need to be checksumed by this value to figure out
5687 * how many checksums will be required. If we are adding bytes then the number
5688 * may go up and we will return the number of additional bytes that must be
5689 * reserved. If it is going down we will return the number of bytes that must
5692 * This must be called with BTRFS_I(inode)->lock held.
5694 static u64
calc_csum_metadata_size(struct inode
*inode
, u64 num_bytes
,
5697 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5698 u64 old_csums
, num_csums
;
5700 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
&&
5701 BTRFS_I(inode
)->csum_bytes
== 0)
5704 old_csums
= btrfs_csum_bytes_to_leaves(root
, BTRFS_I(inode
)->csum_bytes
);
5706 BTRFS_I(inode
)->csum_bytes
+= num_bytes
;
5708 BTRFS_I(inode
)->csum_bytes
-= num_bytes
;
5709 num_csums
= btrfs_csum_bytes_to_leaves(root
, BTRFS_I(inode
)->csum_bytes
);
5711 /* No change, no need to reserve more */
5712 if (old_csums
== num_csums
)
5716 return btrfs_calc_trans_metadata_size(root
,
5717 num_csums
- old_csums
);
5719 return btrfs_calc_trans_metadata_size(root
, old_csums
- num_csums
);
5722 int btrfs_delalloc_reserve_metadata(struct inode
*inode
, u64 num_bytes
)
5724 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5725 struct btrfs_block_rsv
*block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
5728 unsigned nr_extents
= 0;
5729 enum btrfs_reserve_flush_enum flush
= BTRFS_RESERVE_FLUSH_ALL
;
5731 bool delalloc_lock
= true;
5734 bool release_extra
= false;
5736 /* If we are a free space inode we need to not flush since we will be in
5737 * the middle of a transaction commit. We also don't need the delalloc
5738 * mutex since we won't race with anybody. We need this mostly to make
5739 * lockdep shut its filthy mouth.
5741 if (btrfs_is_free_space_inode(inode
)) {
5742 flush
= BTRFS_RESERVE_NO_FLUSH
;
5743 delalloc_lock
= false;
5746 if (flush
!= BTRFS_RESERVE_NO_FLUSH
&&
5747 btrfs_transaction_in_commit(root
->fs_info
))
5748 schedule_timeout(1);
5751 mutex_lock(&BTRFS_I(inode
)->delalloc_mutex
);
5753 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
5755 spin_lock(&BTRFS_I(inode
)->lock
);
5756 nr_extents
= (unsigned)div64_u64(num_bytes
+
5757 BTRFS_MAX_EXTENT_SIZE
- 1,
5758 BTRFS_MAX_EXTENT_SIZE
);
5759 BTRFS_I(inode
)->outstanding_extents
+= nr_extents
;
5762 if (BTRFS_I(inode
)->outstanding_extents
>
5763 BTRFS_I(inode
)->reserved_extents
)
5764 nr_extents
+= BTRFS_I(inode
)->outstanding_extents
-
5765 BTRFS_I(inode
)->reserved_extents
;
5767 /* We always want to reserve a slot for updating the inode. */
5768 to_reserve
= btrfs_calc_trans_metadata_size(root
, nr_extents
+ 1);
5769 to_reserve
+= calc_csum_metadata_size(inode
, num_bytes
, 1);
5770 csum_bytes
= BTRFS_I(inode
)->csum_bytes
;
5771 spin_unlock(&BTRFS_I(inode
)->lock
);
5773 if (root
->fs_info
->quota_enabled
) {
5774 ret
= btrfs_qgroup_reserve_meta(root
,
5775 nr_extents
* root
->nodesize
);
5780 ret
= btrfs_block_rsv_add(root
, block_rsv
, to_reserve
, flush
);
5781 if (unlikely(ret
)) {
5782 btrfs_qgroup_free_meta(root
, nr_extents
* root
->nodesize
);
5786 spin_lock(&BTRFS_I(inode
)->lock
);
5787 if (test_and_set_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5788 &BTRFS_I(inode
)->runtime_flags
))
5789 release_extra
= true;
5790 BTRFS_I(inode
)->reserved_extents
+= nr_extents
;
5791 spin_unlock(&BTRFS_I(inode
)->lock
);
5794 mutex_unlock(&BTRFS_I(inode
)->delalloc_mutex
);
5797 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5798 btrfs_ino(inode
), to_reserve
, 1);
5800 btrfs_block_rsv_release(root
, block_rsv
,
5801 btrfs_calc_trans_metadata_size(root
,
5806 spin_lock(&BTRFS_I(inode
)->lock
);
5807 dropped
= drop_outstanding_extent(inode
, num_bytes
);
5809 * If the inodes csum_bytes is the same as the original
5810 * csum_bytes then we know we haven't raced with any free()ers
5811 * so we can just reduce our inodes csum bytes and carry on.
5813 if (BTRFS_I(inode
)->csum_bytes
== csum_bytes
) {
5814 calc_csum_metadata_size(inode
, num_bytes
, 0);
5816 u64 orig_csum_bytes
= BTRFS_I(inode
)->csum_bytes
;
5820 * This is tricky, but first we need to figure out how much we
5821 * freed from any free-ers that occurred during this
5822 * reservation, so we reset ->csum_bytes to the csum_bytes
5823 * before we dropped our lock, and then call the free for the
5824 * number of bytes that were freed while we were trying our
5827 bytes
= csum_bytes
- BTRFS_I(inode
)->csum_bytes
;
5828 BTRFS_I(inode
)->csum_bytes
= csum_bytes
;
5829 to_free
= calc_csum_metadata_size(inode
, bytes
, 0);
5833 * Now we need to see how much we would have freed had we not
5834 * been making this reservation and our ->csum_bytes were not
5835 * artificially inflated.
5837 BTRFS_I(inode
)->csum_bytes
= csum_bytes
- num_bytes
;
5838 bytes
= csum_bytes
- orig_csum_bytes
;
5839 bytes
= calc_csum_metadata_size(inode
, bytes
, 0);
5842 * Now reset ->csum_bytes to what it should be. If bytes is
5843 * more than to_free then we would have freed more space had we
5844 * not had an artificially high ->csum_bytes, so we need to free
5845 * the remainder. If bytes is the same or less then we don't
5846 * need to do anything, the other free-ers did the correct
5849 BTRFS_I(inode
)->csum_bytes
= orig_csum_bytes
- num_bytes
;
5850 if (bytes
> to_free
)
5851 to_free
= bytes
- to_free
;
5855 spin_unlock(&BTRFS_I(inode
)->lock
);
5857 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
5860 btrfs_block_rsv_release(root
, block_rsv
, to_free
);
5861 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5862 btrfs_ino(inode
), to_free
, 0);
5865 mutex_unlock(&BTRFS_I(inode
)->delalloc_mutex
);
5870 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5871 * @inode: the inode to release the reservation for
5872 * @num_bytes: the number of bytes we're releasing
5874 * This will release the metadata reservation for an inode. This can be called
5875 * once we complete IO for a given set of bytes to release their metadata
5878 void btrfs_delalloc_release_metadata(struct inode
*inode
, u64 num_bytes
)
5880 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5884 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
5885 spin_lock(&BTRFS_I(inode
)->lock
);
5886 dropped
= drop_outstanding_extent(inode
, num_bytes
);
5889 to_free
= calc_csum_metadata_size(inode
, num_bytes
, 0);
5890 spin_unlock(&BTRFS_I(inode
)->lock
);
5892 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
5894 if (btrfs_test_is_dummy_root(root
))
5897 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5898 btrfs_ino(inode
), to_free
, 0);
5900 btrfs_block_rsv_release(root
, &root
->fs_info
->delalloc_block_rsv
,
5905 * btrfs_delalloc_reserve_space - reserve data and metadata space for
5907 * @inode: inode we're writing to
5908 * @start: start range we are writing to
5909 * @len: how long the range we are writing to
5911 * TODO: This function will finally replace old btrfs_delalloc_reserve_space()
5913 * This will do the following things
5915 * o reserve space in data space info for num bytes
5916 * and reserve precious corresponding qgroup space
5917 * (Done in check_data_free_space)
5919 * o reserve space for metadata space, based on the number of outstanding
5920 * extents and how much csums will be needed
5921 * also reserve metadata space in a per root over-reserve method.
5922 * o add to the inodes->delalloc_bytes
5923 * o add it to the fs_info's delalloc inodes list.
5924 * (Above 3 all done in delalloc_reserve_metadata)
5926 * Return 0 for success
5927 * Return <0 for error(-ENOSPC or -EQUOT)
5929 int btrfs_delalloc_reserve_space(struct inode
*inode
, u64 start
, u64 len
)
5933 ret
= btrfs_check_data_free_space(inode
, start
, len
);
5936 ret
= btrfs_delalloc_reserve_metadata(inode
, len
);
5938 btrfs_free_reserved_data_space(inode
, start
, len
);
5943 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5944 * @inode: inode we're releasing space for
5945 * @start: start position of the space already reserved
5946 * @len: the len of the space already reserved
5948 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5949 * called in the case that we don't need the metadata AND data reservations
5950 * anymore. So if there is an error or we insert an inline extent.
5952 * This function will release the metadata space that was not used and will
5953 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5954 * list if there are no delalloc bytes left.
5955 * Also it will handle the qgroup reserved space.
5957 void btrfs_delalloc_release_space(struct inode
*inode
, u64 start
, u64 len
)
5959 btrfs_delalloc_release_metadata(inode
, len
);
5960 btrfs_free_reserved_data_space(inode
, start
, len
);
5963 static int update_block_group(struct btrfs_trans_handle
*trans
,
5964 struct btrfs_root
*root
, u64 bytenr
,
5965 u64 num_bytes
, int alloc
)
5967 struct btrfs_block_group_cache
*cache
= NULL
;
5968 struct btrfs_fs_info
*info
= root
->fs_info
;
5969 u64 total
= num_bytes
;
5974 /* block accounting for super block */
5975 spin_lock(&info
->delalloc_root_lock
);
5976 old_val
= btrfs_super_bytes_used(info
->super_copy
);
5978 old_val
+= num_bytes
;
5980 old_val
-= num_bytes
;
5981 btrfs_set_super_bytes_used(info
->super_copy
, old_val
);
5982 spin_unlock(&info
->delalloc_root_lock
);
5985 cache
= btrfs_lookup_block_group(info
, bytenr
);
5988 if (cache
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
5989 BTRFS_BLOCK_GROUP_RAID1
|
5990 BTRFS_BLOCK_GROUP_RAID10
))
5995 * If this block group has free space cache written out, we
5996 * need to make sure to load it if we are removing space. This
5997 * is because we need the unpinning stage to actually add the
5998 * space back to the block group, otherwise we will leak space.
6000 if (!alloc
&& cache
->cached
== BTRFS_CACHE_NO
)
6001 cache_block_group(cache
, 1);
6003 byte_in_group
= bytenr
- cache
->key
.objectid
;
6004 WARN_ON(byte_in_group
> cache
->key
.offset
);
6006 spin_lock(&cache
->space_info
->lock
);
6007 spin_lock(&cache
->lock
);
6009 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
6010 cache
->disk_cache_state
< BTRFS_DC_CLEAR
)
6011 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
6013 old_val
= btrfs_block_group_used(&cache
->item
);
6014 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
6016 old_val
+= num_bytes
;
6017 btrfs_set_block_group_used(&cache
->item
, old_val
);
6018 cache
->reserved
-= num_bytes
;
6019 cache
->space_info
->bytes_reserved
-= num_bytes
;
6020 cache
->space_info
->bytes_used
+= num_bytes
;
6021 cache
->space_info
->disk_used
+= num_bytes
* factor
;
6022 spin_unlock(&cache
->lock
);
6023 spin_unlock(&cache
->space_info
->lock
);
6025 old_val
-= num_bytes
;
6026 btrfs_set_block_group_used(&cache
->item
, old_val
);
6027 cache
->pinned
+= num_bytes
;
6028 cache
->space_info
->bytes_pinned
+= num_bytes
;
6029 cache
->space_info
->bytes_used
-= num_bytes
;
6030 cache
->space_info
->disk_used
-= num_bytes
* factor
;
6031 spin_unlock(&cache
->lock
);
6032 spin_unlock(&cache
->space_info
->lock
);
6034 set_extent_dirty(info
->pinned_extents
,
6035 bytenr
, bytenr
+ num_bytes
- 1,
6036 GFP_NOFS
| __GFP_NOFAIL
);
6039 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
6040 if (list_empty(&cache
->dirty_list
)) {
6041 list_add_tail(&cache
->dirty_list
,
6042 &trans
->transaction
->dirty_bgs
);
6043 trans
->transaction
->num_dirty_bgs
++;
6044 btrfs_get_block_group(cache
);
6046 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
6049 * No longer have used bytes in this block group, queue it for
6050 * deletion. We do this after adding the block group to the
6051 * dirty list to avoid races between cleaner kthread and space
6054 if (!alloc
&& old_val
== 0) {
6055 spin_lock(&info
->unused_bgs_lock
);
6056 if (list_empty(&cache
->bg_list
)) {
6057 btrfs_get_block_group(cache
);
6058 list_add_tail(&cache
->bg_list
,
6061 spin_unlock(&info
->unused_bgs_lock
);
6064 btrfs_put_block_group(cache
);
6066 bytenr
+= num_bytes
;
6071 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
6073 struct btrfs_block_group_cache
*cache
;
6076 spin_lock(&root
->fs_info
->block_group_cache_lock
);
6077 bytenr
= root
->fs_info
->first_logical_byte
;
6078 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
6080 if (bytenr
< (u64
)-1)
6083 cache
= btrfs_lookup_first_block_group(root
->fs_info
, search_start
);
6087 bytenr
= cache
->key
.objectid
;
6088 btrfs_put_block_group(cache
);
6093 static int pin_down_extent(struct btrfs_root
*root
,
6094 struct btrfs_block_group_cache
*cache
,
6095 u64 bytenr
, u64 num_bytes
, int reserved
)
6097 spin_lock(&cache
->space_info
->lock
);
6098 spin_lock(&cache
->lock
);
6099 cache
->pinned
+= num_bytes
;
6100 cache
->space_info
->bytes_pinned
+= num_bytes
;
6102 cache
->reserved
-= num_bytes
;
6103 cache
->space_info
->bytes_reserved
-= num_bytes
;
6105 spin_unlock(&cache
->lock
);
6106 spin_unlock(&cache
->space_info
->lock
);
6108 set_extent_dirty(root
->fs_info
->pinned_extents
, bytenr
,
6109 bytenr
+ num_bytes
- 1, GFP_NOFS
| __GFP_NOFAIL
);
6111 trace_btrfs_reserved_extent_free(root
, bytenr
, num_bytes
);
6116 * this function must be called within transaction
6118 int btrfs_pin_extent(struct btrfs_root
*root
,
6119 u64 bytenr
, u64 num_bytes
, int reserved
)
6121 struct btrfs_block_group_cache
*cache
;
6123 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
6124 BUG_ON(!cache
); /* Logic error */
6126 pin_down_extent(root
, cache
, bytenr
, num_bytes
, reserved
);
6128 btrfs_put_block_group(cache
);
6133 * this function must be called within transaction
6135 int btrfs_pin_extent_for_log_replay(struct btrfs_root
*root
,
6136 u64 bytenr
, u64 num_bytes
)
6138 struct btrfs_block_group_cache
*cache
;
6141 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
6146 * pull in the free space cache (if any) so that our pin
6147 * removes the free space from the cache. We have load_only set
6148 * to one because the slow code to read in the free extents does check
6149 * the pinned extents.
6151 cache_block_group(cache
, 1);
6153 pin_down_extent(root
, cache
, bytenr
, num_bytes
, 0);
6155 /* remove us from the free space cache (if we're there at all) */
6156 ret
= btrfs_remove_free_space(cache
, bytenr
, num_bytes
);
6157 btrfs_put_block_group(cache
);
6161 static int __exclude_logged_extent(struct btrfs_root
*root
, u64 start
, u64 num_bytes
)
6164 struct btrfs_block_group_cache
*block_group
;
6165 struct btrfs_caching_control
*caching_ctl
;
6167 block_group
= btrfs_lookup_block_group(root
->fs_info
, start
);
6171 cache_block_group(block_group
, 0);
6172 caching_ctl
= get_caching_control(block_group
);
6176 BUG_ON(!block_group_cache_done(block_group
));
6177 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
6179 mutex_lock(&caching_ctl
->mutex
);
6181 if (start
>= caching_ctl
->progress
) {
6182 ret
= add_excluded_extent(root
, start
, num_bytes
);
6183 } else if (start
+ num_bytes
<= caching_ctl
->progress
) {
6184 ret
= btrfs_remove_free_space(block_group
,
6187 num_bytes
= caching_ctl
->progress
- start
;
6188 ret
= btrfs_remove_free_space(block_group
,
6193 num_bytes
= (start
+ num_bytes
) -
6194 caching_ctl
->progress
;
6195 start
= caching_ctl
->progress
;
6196 ret
= add_excluded_extent(root
, start
, num_bytes
);
6199 mutex_unlock(&caching_ctl
->mutex
);
6200 put_caching_control(caching_ctl
);
6202 btrfs_put_block_group(block_group
);
6206 int btrfs_exclude_logged_extents(struct btrfs_root
*log
,
6207 struct extent_buffer
*eb
)
6209 struct btrfs_file_extent_item
*item
;
6210 struct btrfs_key key
;
6214 if (!btrfs_fs_incompat(log
->fs_info
, MIXED_GROUPS
))
6217 for (i
= 0; i
< btrfs_header_nritems(eb
); i
++) {
6218 btrfs_item_key_to_cpu(eb
, &key
, i
);
6219 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
6221 item
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
6222 found_type
= btrfs_file_extent_type(eb
, item
);
6223 if (found_type
== BTRFS_FILE_EXTENT_INLINE
)
6225 if (btrfs_file_extent_disk_bytenr(eb
, item
) == 0)
6227 key
.objectid
= btrfs_file_extent_disk_bytenr(eb
, item
);
6228 key
.offset
= btrfs_file_extent_disk_num_bytes(eb
, item
);
6229 __exclude_logged_extent(log
, key
.objectid
, key
.offset
);
6236 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache
*bg
)
6238 atomic_inc(&bg
->reservations
);
6241 void btrfs_dec_block_group_reservations(struct btrfs_fs_info
*fs_info
,
6244 struct btrfs_block_group_cache
*bg
;
6246 bg
= btrfs_lookup_block_group(fs_info
, start
);
6248 if (atomic_dec_and_test(&bg
->reservations
))
6249 wake_up_atomic_t(&bg
->reservations
);
6250 btrfs_put_block_group(bg
);
6253 static int btrfs_wait_bg_reservations_atomic_t(atomic_t
*a
)
6259 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache
*bg
)
6261 struct btrfs_space_info
*space_info
= bg
->space_info
;
6265 if (!(bg
->flags
& BTRFS_BLOCK_GROUP_DATA
))
6269 * Our block group is read only but before we set it to read only,
6270 * some task might have had allocated an extent from it already, but it
6271 * has not yet created a respective ordered extent (and added it to a
6272 * root's list of ordered extents).
6273 * Therefore wait for any task currently allocating extents, since the
6274 * block group's reservations counter is incremented while a read lock
6275 * on the groups' semaphore is held and decremented after releasing
6276 * the read access on that semaphore and creating the ordered extent.
6278 down_write(&space_info
->groups_sem
);
6279 up_write(&space_info
->groups_sem
);
6281 wait_on_atomic_t(&bg
->reservations
,
6282 btrfs_wait_bg_reservations_atomic_t
,
6283 TASK_UNINTERRUPTIBLE
);
6287 * btrfs_update_reserved_bytes - update the block_group and space info counters
6288 * @cache: The cache we are manipulating
6289 * @num_bytes: The number of bytes in question
6290 * @reserve: One of the reservation enums
6291 * @delalloc: The blocks are allocated for the delalloc write
6293 * This is called by the allocator when it reserves space, or by somebody who is
6294 * freeing space that was never actually used on disk. For example if you
6295 * reserve some space for a new leaf in transaction A and before transaction A
6296 * commits you free that leaf, you call this with reserve set to 0 in order to
6297 * clear the reservation.
6299 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
6300 * ENOSPC accounting. For data we handle the reservation through clearing the
6301 * delalloc bits in the io_tree. We have to do this since we could end up
6302 * allocating less disk space for the amount of data we have reserved in the
6303 * case of compression.
6305 * If this is a reservation and the block group has become read only we cannot
6306 * make the reservation and return -EAGAIN, otherwise this function always
6309 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
6310 u64 num_bytes
, int reserve
, int delalloc
)
6312 struct btrfs_space_info
*space_info
= cache
->space_info
;
6315 spin_lock(&space_info
->lock
);
6316 spin_lock(&cache
->lock
);
6317 if (reserve
!= RESERVE_FREE
) {
6321 cache
->reserved
+= num_bytes
;
6322 space_info
->bytes_reserved
+= num_bytes
;
6323 if (reserve
== RESERVE_ALLOC
) {
6324 trace_btrfs_space_reservation(cache
->fs_info
,
6325 "space_info", space_info
->flags
,
6327 space_info
->bytes_may_use
-= num_bytes
;
6331 cache
->delalloc_bytes
+= num_bytes
;
6335 space_info
->bytes_readonly
+= num_bytes
;
6336 cache
->reserved
-= num_bytes
;
6337 space_info
->bytes_reserved
-= num_bytes
;
6340 cache
->delalloc_bytes
-= num_bytes
;
6342 spin_unlock(&cache
->lock
);
6343 spin_unlock(&space_info
->lock
);
6347 void btrfs_prepare_extent_commit(struct btrfs_trans_handle
*trans
,
6348 struct btrfs_root
*root
)
6350 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6351 struct btrfs_caching_control
*next
;
6352 struct btrfs_caching_control
*caching_ctl
;
6353 struct btrfs_block_group_cache
*cache
;
6355 down_write(&fs_info
->commit_root_sem
);
6357 list_for_each_entry_safe(caching_ctl
, next
,
6358 &fs_info
->caching_block_groups
, list
) {
6359 cache
= caching_ctl
->block_group
;
6360 if (block_group_cache_done(cache
)) {
6361 cache
->last_byte_to_unpin
= (u64
)-1;
6362 list_del_init(&caching_ctl
->list
);
6363 put_caching_control(caching_ctl
);
6365 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
6369 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
6370 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
6372 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
6374 up_write(&fs_info
->commit_root_sem
);
6376 update_global_block_rsv(fs_info
);
6380 * Returns the free cluster for the given space info and sets empty_cluster to
6381 * what it should be based on the mount options.
6383 static struct btrfs_free_cluster
*
6384 fetch_cluster_info(struct btrfs_root
*root
, struct btrfs_space_info
*space_info
,
6387 struct btrfs_free_cluster
*ret
= NULL
;
6388 bool ssd
= btrfs_test_opt(root
, SSD
);
6391 if (btrfs_mixed_space_info(space_info
))
6395 *empty_cluster
= SZ_2M
;
6396 if (space_info
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
6397 ret
= &root
->fs_info
->meta_alloc_cluster
;
6399 *empty_cluster
= SZ_64K
;
6400 } else if ((space_info
->flags
& BTRFS_BLOCK_GROUP_DATA
) && ssd
) {
6401 ret
= &root
->fs_info
->data_alloc_cluster
;
6407 static int unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
,
6408 const bool return_free_space
)
6410 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6411 struct btrfs_block_group_cache
*cache
= NULL
;
6412 struct btrfs_space_info
*space_info
;
6413 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
6414 struct btrfs_free_cluster
*cluster
= NULL
;
6416 u64 total_unpinned
= 0;
6417 u64 empty_cluster
= 0;
6420 while (start
<= end
) {
6423 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
6425 btrfs_put_block_group(cache
);
6427 cache
= btrfs_lookup_block_group(fs_info
, start
);
6428 BUG_ON(!cache
); /* Logic error */
6430 cluster
= fetch_cluster_info(root
,
6433 empty_cluster
<<= 1;
6436 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
6437 len
= min(len
, end
+ 1 - start
);
6439 if (start
< cache
->last_byte_to_unpin
) {
6440 len
= min(len
, cache
->last_byte_to_unpin
- start
);
6441 if (return_free_space
)
6442 btrfs_add_free_space(cache
, start
, len
);
6446 total_unpinned
+= len
;
6447 space_info
= cache
->space_info
;
6450 * If this space cluster has been marked as fragmented and we've
6451 * unpinned enough in this block group to potentially allow a
6452 * cluster to be created inside of it go ahead and clear the
6455 if (cluster
&& cluster
->fragmented
&&
6456 total_unpinned
> empty_cluster
) {
6457 spin_lock(&cluster
->lock
);
6458 cluster
->fragmented
= 0;
6459 spin_unlock(&cluster
->lock
);
6462 spin_lock(&space_info
->lock
);
6463 spin_lock(&cache
->lock
);
6464 cache
->pinned
-= len
;
6465 space_info
->bytes_pinned
-= len
;
6466 space_info
->max_extent_size
= 0;
6467 percpu_counter_add(&space_info
->total_bytes_pinned
, -len
);
6469 space_info
->bytes_readonly
+= len
;
6472 spin_unlock(&cache
->lock
);
6473 if (!readonly
&& global_rsv
->space_info
== space_info
) {
6474 spin_lock(&global_rsv
->lock
);
6475 if (!global_rsv
->full
) {
6476 len
= min(len
, global_rsv
->size
-
6477 global_rsv
->reserved
);
6478 global_rsv
->reserved
+= len
;
6479 space_info
->bytes_may_use
+= len
;
6480 if (global_rsv
->reserved
>= global_rsv
->size
)
6481 global_rsv
->full
= 1;
6483 spin_unlock(&global_rsv
->lock
);
6485 spin_unlock(&space_info
->lock
);
6489 btrfs_put_block_group(cache
);
6493 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
6494 struct btrfs_root
*root
)
6496 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6497 struct btrfs_block_group_cache
*block_group
, *tmp
;
6498 struct list_head
*deleted_bgs
;
6499 struct extent_io_tree
*unpin
;
6504 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
6505 unpin
= &fs_info
->freed_extents
[1];
6507 unpin
= &fs_info
->freed_extents
[0];
6509 while (!trans
->aborted
) {
6510 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
6511 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
6512 EXTENT_DIRTY
, NULL
);
6514 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
6518 if (btrfs_test_opt(root
, DISCARD
))
6519 ret
= btrfs_discard_extent(root
, start
,
6520 end
+ 1 - start
, NULL
);
6522 clear_extent_dirty(unpin
, start
, end
);
6523 unpin_extent_range(root
, start
, end
, true);
6524 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
6529 * Transaction is finished. We don't need the lock anymore. We
6530 * do need to clean up the block groups in case of a transaction
6533 deleted_bgs
= &trans
->transaction
->deleted_bgs
;
6534 list_for_each_entry_safe(block_group
, tmp
, deleted_bgs
, bg_list
) {
6538 if (!trans
->aborted
)
6539 ret
= btrfs_discard_extent(root
,
6540 block_group
->key
.objectid
,
6541 block_group
->key
.offset
,
6544 list_del_init(&block_group
->bg_list
);
6545 btrfs_put_block_group_trimming(block_group
);
6546 btrfs_put_block_group(block_group
);
6549 const char *errstr
= btrfs_decode_error(ret
);
6551 "Discard failed while removing blockgroup: errno=%d %s\n",
6559 static void add_pinned_bytes(struct btrfs_fs_info
*fs_info
, u64 num_bytes
,
6560 u64 owner
, u64 root_objectid
)
6562 struct btrfs_space_info
*space_info
;
6565 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
6566 if (root_objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
6567 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
6569 flags
= BTRFS_BLOCK_GROUP_METADATA
;
6571 flags
= BTRFS_BLOCK_GROUP_DATA
;
6574 space_info
= __find_space_info(fs_info
, flags
);
6575 BUG_ON(!space_info
); /* Logic bug */
6576 percpu_counter_add(&space_info
->total_bytes_pinned
, num_bytes
);
6580 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
6581 struct btrfs_root
*root
,
6582 struct btrfs_delayed_ref_node
*node
, u64 parent
,
6583 u64 root_objectid
, u64 owner_objectid
,
6584 u64 owner_offset
, int refs_to_drop
,
6585 struct btrfs_delayed_extent_op
*extent_op
)
6587 struct btrfs_key key
;
6588 struct btrfs_path
*path
;
6589 struct btrfs_fs_info
*info
= root
->fs_info
;
6590 struct btrfs_root
*extent_root
= info
->extent_root
;
6591 struct extent_buffer
*leaf
;
6592 struct btrfs_extent_item
*ei
;
6593 struct btrfs_extent_inline_ref
*iref
;
6596 int extent_slot
= 0;
6597 int found_extent
= 0;
6601 u64 bytenr
= node
->bytenr
;
6602 u64 num_bytes
= node
->num_bytes
;
6604 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
6607 path
= btrfs_alloc_path();
6611 path
->reada
= READA_FORWARD
;
6612 path
->leave_spinning
= 1;
6614 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
6615 BUG_ON(!is_data
&& refs_to_drop
!= 1);
6618 skinny_metadata
= 0;
6620 ret
= lookup_extent_backref(trans
, extent_root
, path
, &iref
,
6621 bytenr
, num_bytes
, parent
,
6622 root_objectid
, owner_objectid
,
6625 extent_slot
= path
->slots
[0];
6626 while (extent_slot
>= 0) {
6627 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
6629 if (key
.objectid
!= bytenr
)
6631 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
6632 key
.offset
== num_bytes
) {
6636 if (key
.type
== BTRFS_METADATA_ITEM_KEY
&&
6637 key
.offset
== owner_objectid
) {
6641 if (path
->slots
[0] - extent_slot
> 5)
6645 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6646 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
6647 if (found_extent
&& item_size
< sizeof(*ei
))
6650 if (!found_extent
) {
6652 ret
= remove_extent_backref(trans
, extent_root
, path
,
6654 is_data
, &last_ref
);
6656 btrfs_abort_transaction(trans
, extent_root
, ret
);
6659 btrfs_release_path(path
);
6660 path
->leave_spinning
= 1;
6662 key
.objectid
= bytenr
;
6663 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6664 key
.offset
= num_bytes
;
6666 if (!is_data
&& skinny_metadata
) {
6667 key
.type
= BTRFS_METADATA_ITEM_KEY
;
6668 key
.offset
= owner_objectid
;
6671 ret
= btrfs_search_slot(trans
, extent_root
,
6673 if (ret
> 0 && skinny_metadata
&& path
->slots
[0]) {
6675 * Couldn't find our skinny metadata item,
6676 * see if we have ye olde extent item.
6679 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
6681 if (key
.objectid
== bytenr
&&
6682 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
6683 key
.offset
== num_bytes
)
6687 if (ret
> 0 && skinny_metadata
) {
6688 skinny_metadata
= false;
6689 key
.objectid
= bytenr
;
6690 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6691 key
.offset
= num_bytes
;
6692 btrfs_release_path(path
);
6693 ret
= btrfs_search_slot(trans
, extent_root
,
6698 btrfs_err(info
, "umm, got %d back from search, was looking for %llu",
6701 btrfs_print_leaf(extent_root
,
6705 btrfs_abort_transaction(trans
, extent_root
, ret
);
6708 extent_slot
= path
->slots
[0];
6710 } else if (WARN_ON(ret
== -ENOENT
)) {
6711 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
6713 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6714 bytenr
, parent
, root_objectid
, owner_objectid
,
6716 btrfs_abort_transaction(trans
, extent_root
, ret
);
6719 btrfs_abort_transaction(trans
, extent_root
, ret
);
6723 leaf
= path
->nodes
[0];
6724 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
6725 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6726 if (item_size
< sizeof(*ei
)) {
6727 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
6728 ret
= convert_extent_item_v0(trans
, extent_root
, path
,
6731 btrfs_abort_transaction(trans
, extent_root
, ret
);
6735 btrfs_release_path(path
);
6736 path
->leave_spinning
= 1;
6738 key
.objectid
= bytenr
;
6739 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6740 key
.offset
= num_bytes
;
6742 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
6745 btrfs_err(info
, "umm, got %d back from search, was looking for %llu",
6747 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
6750 btrfs_abort_transaction(trans
, extent_root
, ret
);
6754 extent_slot
= path
->slots
[0];
6755 leaf
= path
->nodes
[0];
6756 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
6759 BUG_ON(item_size
< sizeof(*ei
));
6760 ei
= btrfs_item_ptr(leaf
, extent_slot
,
6761 struct btrfs_extent_item
);
6762 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
6763 key
.type
== BTRFS_EXTENT_ITEM_KEY
) {
6764 struct btrfs_tree_block_info
*bi
;
6765 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
6766 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
6767 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
6770 refs
= btrfs_extent_refs(leaf
, ei
);
6771 if (refs
< refs_to_drop
) {
6772 btrfs_err(info
, "trying to drop %d refs but we only have %Lu "
6773 "for bytenr %Lu", refs_to_drop
, refs
, bytenr
);
6775 btrfs_abort_transaction(trans
, extent_root
, ret
);
6778 refs
-= refs_to_drop
;
6782 __run_delayed_extent_op(extent_op
, leaf
, ei
);
6784 * In the case of inline back ref, reference count will
6785 * be updated by remove_extent_backref
6788 BUG_ON(!found_extent
);
6790 btrfs_set_extent_refs(leaf
, ei
, refs
);
6791 btrfs_mark_buffer_dirty(leaf
);
6794 ret
= remove_extent_backref(trans
, extent_root
, path
,
6796 is_data
, &last_ref
);
6798 btrfs_abort_transaction(trans
, extent_root
, ret
);
6802 add_pinned_bytes(root
->fs_info
, -num_bytes
, owner_objectid
,
6806 BUG_ON(is_data
&& refs_to_drop
!=
6807 extent_data_ref_count(path
, iref
));
6809 BUG_ON(path
->slots
[0] != extent_slot
);
6811 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
6812 path
->slots
[0] = extent_slot
;
6818 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
6821 btrfs_abort_transaction(trans
, extent_root
, ret
);
6824 btrfs_release_path(path
);
6827 ret
= btrfs_del_csums(trans
, root
, bytenr
, num_bytes
);
6829 btrfs_abort_transaction(trans
, extent_root
, ret
);
6834 ret
= add_to_free_space_tree(trans
, root
->fs_info
, bytenr
,
6837 btrfs_abort_transaction(trans
, extent_root
, ret
);
6841 ret
= update_block_group(trans
, root
, bytenr
, num_bytes
, 0);
6843 btrfs_abort_transaction(trans
, extent_root
, ret
);
6847 btrfs_release_path(path
);
6850 btrfs_free_path(path
);
6855 * when we free an block, it is possible (and likely) that we free the last
6856 * delayed ref for that extent as well. This searches the delayed ref tree for
6857 * a given extent, and if there are no other delayed refs to be processed, it
6858 * removes it from the tree.
6860 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
6861 struct btrfs_root
*root
, u64 bytenr
)
6863 struct btrfs_delayed_ref_head
*head
;
6864 struct btrfs_delayed_ref_root
*delayed_refs
;
6867 delayed_refs
= &trans
->transaction
->delayed_refs
;
6868 spin_lock(&delayed_refs
->lock
);
6869 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
6871 goto out_delayed_unlock
;
6873 spin_lock(&head
->lock
);
6874 if (!list_empty(&head
->ref_list
))
6877 if (head
->extent_op
) {
6878 if (!head
->must_insert_reserved
)
6880 btrfs_free_delayed_extent_op(head
->extent_op
);
6881 head
->extent_op
= NULL
;
6885 * waiting for the lock here would deadlock. If someone else has it
6886 * locked they are already in the process of dropping it anyway
6888 if (!mutex_trylock(&head
->mutex
))
6892 * at this point we have a head with no other entries. Go
6893 * ahead and process it.
6895 head
->node
.in_tree
= 0;
6896 rb_erase(&head
->href_node
, &delayed_refs
->href_root
);
6898 atomic_dec(&delayed_refs
->num_entries
);
6901 * we don't take a ref on the node because we're removing it from the
6902 * tree, so we just steal the ref the tree was holding.
6904 delayed_refs
->num_heads
--;
6905 if (head
->processing
== 0)
6906 delayed_refs
->num_heads_ready
--;
6907 head
->processing
= 0;
6908 spin_unlock(&head
->lock
);
6909 spin_unlock(&delayed_refs
->lock
);
6911 BUG_ON(head
->extent_op
);
6912 if (head
->must_insert_reserved
)
6915 mutex_unlock(&head
->mutex
);
6916 btrfs_put_delayed_ref(&head
->node
);
6919 spin_unlock(&head
->lock
);
6922 spin_unlock(&delayed_refs
->lock
);
6926 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
6927 struct btrfs_root
*root
,
6928 struct extent_buffer
*buf
,
6929 u64 parent
, int last_ref
)
6934 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
6935 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
6936 buf
->start
, buf
->len
,
6937 parent
, root
->root_key
.objectid
,
6938 btrfs_header_level(buf
),
6939 BTRFS_DROP_DELAYED_REF
, NULL
);
6940 BUG_ON(ret
); /* -ENOMEM */
6946 if (btrfs_header_generation(buf
) == trans
->transid
) {
6947 struct btrfs_block_group_cache
*cache
;
6949 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
6950 ret
= check_ref_cleanup(trans
, root
, buf
->start
);
6955 cache
= btrfs_lookup_block_group(root
->fs_info
, buf
->start
);
6957 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
6958 pin_down_extent(root
, cache
, buf
->start
, buf
->len
, 1);
6959 btrfs_put_block_group(cache
);
6963 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
6965 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
6966 btrfs_update_reserved_bytes(cache
, buf
->len
, RESERVE_FREE
, 0);
6967 btrfs_put_block_group(cache
);
6968 trace_btrfs_reserved_extent_free(root
, buf
->start
, buf
->len
);
6973 add_pinned_bytes(root
->fs_info
, buf
->len
,
6974 btrfs_header_level(buf
),
6975 root
->root_key
.objectid
);
6978 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6981 clear_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
);
6984 /* Can return -ENOMEM */
6985 int btrfs_free_extent(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
6986 u64 bytenr
, u64 num_bytes
, u64 parent
, u64 root_objectid
,
6987 u64 owner
, u64 offset
)
6990 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6992 if (btrfs_test_is_dummy_root(root
))
6995 add_pinned_bytes(root
->fs_info
, num_bytes
, owner
, root_objectid
);
6998 * tree log blocks never actually go into the extent allocation
6999 * tree, just update pinning info and exit early.
7001 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
7002 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
7003 /* unlocks the pinned mutex */
7004 btrfs_pin_extent(root
, bytenr
, num_bytes
, 1);
7006 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
7007 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
7009 parent
, root_objectid
, (int)owner
,
7010 BTRFS_DROP_DELAYED_REF
, NULL
);
7012 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
7014 parent
, root_objectid
, owner
,
7016 BTRFS_DROP_DELAYED_REF
, NULL
);
7022 * when we wait for progress in the block group caching, its because
7023 * our allocation attempt failed at least once. So, we must sleep
7024 * and let some progress happen before we try again.
7026 * This function will sleep at least once waiting for new free space to
7027 * show up, and then it will check the block group free space numbers
7028 * for our min num_bytes. Another option is to have it go ahead
7029 * and look in the rbtree for a free extent of a given size, but this
7032 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7033 * any of the information in this block group.
7035 static noinline
void
7036 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
7039 struct btrfs_caching_control
*caching_ctl
;
7041 caching_ctl
= get_caching_control(cache
);
7045 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
7046 (cache
->free_space_ctl
->free_space
>= num_bytes
));
7048 put_caching_control(caching_ctl
);
7052 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
7054 struct btrfs_caching_control
*caching_ctl
;
7057 caching_ctl
= get_caching_control(cache
);
7059 return (cache
->cached
== BTRFS_CACHE_ERROR
) ? -EIO
: 0;
7061 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
7062 if (cache
->cached
== BTRFS_CACHE_ERROR
)
7064 put_caching_control(caching_ctl
);
7068 int __get_raid_index(u64 flags
)
7070 if (flags
& BTRFS_BLOCK_GROUP_RAID10
)
7071 return BTRFS_RAID_RAID10
;
7072 else if (flags
& BTRFS_BLOCK_GROUP_RAID1
)
7073 return BTRFS_RAID_RAID1
;
7074 else if (flags
& BTRFS_BLOCK_GROUP_DUP
)
7075 return BTRFS_RAID_DUP
;
7076 else if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
7077 return BTRFS_RAID_RAID0
;
7078 else if (flags
& BTRFS_BLOCK_GROUP_RAID5
)
7079 return BTRFS_RAID_RAID5
;
7080 else if (flags
& BTRFS_BLOCK_GROUP_RAID6
)
7081 return BTRFS_RAID_RAID6
;
7083 return BTRFS_RAID_SINGLE
; /* BTRFS_BLOCK_GROUP_SINGLE */
7086 int get_block_group_index(struct btrfs_block_group_cache
*cache
)
7088 return __get_raid_index(cache
->flags
);
7091 static const char *btrfs_raid_type_names
[BTRFS_NR_RAID_TYPES
] = {
7092 [BTRFS_RAID_RAID10
] = "raid10",
7093 [BTRFS_RAID_RAID1
] = "raid1",
7094 [BTRFS_RAID_DUP
] = "dup",
7095 [BTRFS_RAID_RAID0
] = "raid0",
7096 [BTRFS_RAID_SINGLE
] = "single",
7097 [BTRFS_RAID_RAID5
] = "raid5",
7098 [BTRFS_RAID_RAID6
] = "raid6",
7101 static const char *get_raid_name(enum btrfs_raid_types type
)
7103 if (type
>= BTRFS_NR_RAID_TYPES
)
7106 return btrfs_raid_type_names
[type
];
7109 enum btrfs_loop_type
{
7110 LOOP_CACHING_NOWAIT
= 0,
7111 LOOP_CACHING_WAIT
= 1,
7112 LOOP_ALLOC_CHUNK
= 2,
7113 LOOP_NO_EMPTY_SIZE
= 3,
7117 btrfs_lock_block_group(struct btrfs_block_group_cache
*cache
,
7121 down_read(&cache
->data_rwsem
);
7125 btrfs_grab_block_group(struct btrfs_block_group_cache
*cache
,
7128 btrfs_get_block_group(cache
);
7130 down_read(&cache
->data_rwsem
);
7133 static struct btrfs_block_group_cache
*
7134 btrfs_lock_cluster(struct btrfs_block_group_cache
*block_group
,
7135 struct btrfs_free_cluster
*cluster
,
7138 struct btrfs_block_group_cache
*used_bg
= NULL
;
7140 spin_lock(&cluster
->refill_lock
);
7142 used_bg
= cluster
->block_group
;
7146 if (used_bg
== block_group
)
7149 btrfs_get_block_group(used_bg
);
7154 if (down_read_trylock(&used_bg
->data_rwsem
))
7157 spin_unlock(&cluster
->refill_lock
);
7159 down_read(&used_bg
->data_rwsem
);
7161 spin_lock(&cluster
->refill_lock
);
7162 if (used_bg
== cluster
->block_group
)
7165 up_read(&used_bg
->data_rwsem
);
7166 btrfs_put_block_group(used_bg
);
7171 btrfs_release_block_group(struct btrfs_block_group_cache
*cache
,
7175 up_read(&cache
->data_rwsem
);
7176 btrfs_put_block_group(cache
);
7180 * walks the btree of allocated extents and find a hole of a given size.
7181 * The key ins is changed to record the hole:
7182 * ins->objectid == start position
7183 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7184 * ins->offset == the size of the hole.
7185 * Any available blocks before search_start are skipped.
7187 * If there is no suitable free space, we will record the max size of
7188 * the free space extent currently.
7190 static noinline
int find_free_extent(struct btrfs_root
*orig_root
,
7191 u64 num_bytes
, u64 empty_size
,
7192 u64 hint_byte
, struct btrfs_key
*ins
,
7193 u64 flags
, int delalloc
)
7196 struct btrfs_root
*root
= orig_root
->fs_info
->extent_root
;
7197 struct btrfs_free_cluster
*last_ptr
= NULL
;
7198 struct btrfs_block_group_cache
*block_group
= NULL
;
7199 u64 search_start
= 0;
7200 u64 max_extent_size
= 0;
7201 u64 empty_cluster
= 0;
7202 struct btrfs_space_info
*space_info
;
7204 int index
= __get_raid_index(flags
);
7205 int alloc_type
= (flags
& BTRFS_BLOCK_GROUP_DATA
) ?
7206 RESERVE_ALLOC_NO_ACCOUNT
: RESERVE_ALLOC
;
7207 bool failed_cluster_refill
= false;
7208 bool failed_alloc
= false;
7209 bool use_cluster
= true;
7210 bool have_caching_bg
= false;
7211 bool orig_have_caching_bg
= false;
7212 bool full_search
= false;
7214 WARN_ON(num_bytes
< root
->sectorsize
);
7215 ins
->type
= BTRFS_EXTENT_ITEM_KEY
;
7219 trace_find_free_extent(orig_root
, num_bytes
, empty_size
, flags
);
7221 space_info
= __find_space_info(root
->fs_info
, flags
);
7223 btrfs_err(root
->fs_info
, "No space info for %llu", flags
);
7228 * If our free space is heavily fragmented we may not be able to make
7229 * big contiguous allocations, so instead of doing the expensive search
7230 * for free space, simply return ENOSPC with our max_extent_size so we
7231 * can go ahead and search for a more manageable chunk.
7233 * If our max_extent_size is large enough for our allocation simply
7234 * disable clustering since we will likely not be able to find enough
7235 * space to create a cluster and induce latency trying.
7237 if (unlikely(space_info
->max_extent_size
)) {
7238 spin_lock(&space_info
->lock
);
7239 if (space_info
->max_extent_size
&&
7240 num_bytes
> space_info
->max_extent_size
) {
7241 ins
->offset
= space_info
->max_extent_size
;
7242 spin_unlock(&space_info
->lock
);
7244 } else if (space_info
->max_extent_size
) {
7245 use_cluster
= false;
7247 spin_unlock(&space_info
->lock
);
7250 last_ptr
= fetch_cluster_info(orig_root
, space_info
, &empty_cluster
);
7252 spin_lock(&last_ptr
->lock
);
7253 if (last_ptr
->block_group
)
7254 hint_byte
= last_ptr
->window_start
;
7255 if (last_ptr
->fragmented
) {
7257 * We still set window_start so we can keep track of the
7258 * last place we found an allocation to try and save
7261 hint_byte
= last_ptr
->window_start
;
7262 use_cluster
= false;
7264 spin_unlock(&last_ptr
->lock
);
7267 search_start
= max(search_start
, first_logical_byte(root
, 0));
7268 search_start
= max(search_start
, hint_byte
);
7269 if (search_start
== hint_byte
) {
7270 block_group
= btrfs_lookup_block_group(root
->fs_info
,
7273 * we don't want to use the block group if it doesn't match our
7274 * allocation bits, or if its not cached.
7276 * However if we are re-searching with an ideal block group
7277 * picked out then we don't care that the block group is cached.
7279 if (block_group
&& block_group_bits(block_group
, flags
) &&
7280 block_group
->cached
!= BTRFS_CACHE_NO
) {
7281 down_read(&space_info
->groups_sem
);
7282 if (list_empty(&block_group
->list
) ||
7285 * someone is removing this block group,
7286 * we can't jump into the have_block_group
7287 * target because our list pointers are not
7290 btrfs_put_block_group(block_group
);
7291 up_read(&space_info
->groups_sem
);
7293 index
= get_block_group_index(block_group
);
7294 btrfs_lock_block_group(block_group
, delalloc
);
7295 goto have_block_group
;
7297 } else if (block_group
) {
7298 btrfs_put_block_group(block_group
);
7302 have_caching_bg
= false;
7303 if (index
== 0 || index
== __get_raid_index(flags
))
7305 down_read(&space_info
->groups_sem
);
7306 list_for_each_entry(block_group
, &space_info
->block_groups
[index
],
7311 btrfs_grab_block_group(block_group
, delalloc
);
7312 search_start
= block_group
->key
.objectid
;
7315 * this can happen if we end up cycling through all the
7316 * raid types, but we want to make sure we only allocate
7317 * for the proper type.
7319 if (!block_group_bits(block_group
, flags
)) {
7320 u64 extra
= BTRFS_BLOCK_GROUP_DUP
|
7321 BTRFS_BLOCK_GROUP_RAID1
|
7322 BTRFS_BLOCK_GROUP_RAID5
|
7323 BTRFS_BLOCK_GROUP_RAID6
|
7324 BTRFS_BLOCK_GROUP_RAID10
;
7327 * if they asked for extra copies and this block group
7328 * doesn't provide them, bail. This does allow us to
7329 * fill raid0 from raid1.
7331 if ((flags
& extra
) && !(block_group
->flags
& extra
))
7336 cached
= block_group_cache_done(block_group
);
7337 if (unlikely(!cached
)) {
7338 have_caching_bg
= true;
7339 ret
= cache_block_group(block_group
, 0);
7344 if (unlikely(block_group
->cached
== BTRFS_CACHE_ERROR
))
7346 if (unlikely(block_group
->ro
))
7350 * Ok we want to try and use the cluster allocator, so
7353 if (last_ptr
&& use_cluster
) {
7354 struct btrfs_block_group_cache
*used_block_group
;
7355 unsigned long aligned_cluster
;
7357 * the refill lock keeps out other
7358 * people trying to start a new cluster
7360 used_block_group
= btrfs_lock_cluster(block_group
,
7363 if (!used_block_group
)
7364 goto refill_cluster
;
7366 if (used_block_group
!= block_group
&&
7367 (used_block_group
->ro
||
7368 !block_group_bits(used_block_group
, flags
)))
7369 goto release_cluster
;
7371 offset
= btrfs_alloc_from_cluster(used_block_group
,
7374 used_block_group
->key
.objectid
,
7377 /* we have a block, we're done */
7378 spin_unlock(&last_ptr
->refill_lock
);
7379 trace_btrfs_reserve_extent_cluster(root
,
7381 search_start
, num_bytes
);
7382 if (used_block_group
!= block_group
) {
7383 btrfs_release_block_group(block_group
,
7385 block_group
= used_block_group
;
7390 WARN_ON(last_ptr
->block_group
!= used_block_group
);
7392 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7393 * set up a new clusters, so lets just skip it
7394 * and let the allocator find whatever block
7395 * it can find. If we reach this point, we
7396 * will have tried the cluster allocator
7397 * plenty of times and not have found
7398 * anything, so we are likely way too
7399 * fragmented for the clustering stuff to find
7402 * However, if the cluster is taken from the
7403 * current block group, release the cluster
7404 * first, so that we stand a better chance of
7405 * succeeding in the unclustered
7407 if (loop
>= LOOP_NO_EMPTY_SIZE
&&
7408 used_block_group
!= block_group
) {
7409 spin_unlock(&last_ptr
->refill_lock
);
7410 btrfs_release_block_group(used_block_group
,
7412 goto unclustered_alloc
;
7416 * this cluster didn't work out, free it and
7419 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
7421 if (used_block_group
!= block_group
)
7422 btrfs_release_block_group(used_block_group
,
7425 if (loop
>= LOOP_NO_EMPTY_SIZE
) {
7426 spin_unlock(&last_ptr
->refill_lock
);
7427 goto unclustered_alloc
;
7430 aligned_cluster
= max_t(unsigned long,
7431 empty_cluster
+ empty_size
,
7432 block_group
->full_stripe_len
);
7434 /* allocate a cluster in this block group */
7435 ret
= btrfs_find_space_cluster(root
, block_group
,
7436 last_ptr
, search_start
,
7441 * now pull our allocation out of this
7444 offset
= btrfs_alloc_from_cluster(block_group
,
7450 /* we found one, proceed */
7451 spin_unlock(&last_ptr
->refill_lock
);
7452 trace_btrfs_reserve_extent_cluster(root
,
7453 block_group
, search_start
,
7457 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
7458 && !failed_cluster_refill
) {
7459 spin_unlock(&last_ptr
->refill_lock
);
7461 failed_cluster_refill
= true;
7462 wait_block_group_cache_progress(block_group
,
7463 num_bytes
+ empty_cluster
+ empty_size
);
7464 goto have_block_group
;
7468 * at this point we either didn't find a cluster
7469 * or we weren't able to allocate a block from our
7470 * cluster. Free the cluster we've been trying
7471 * to use, and go to the next block group
7473 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
7474 spin_unlock(&last_ptr
->refill_lock
);
7480 * We are doing an unclustered alloc, set the fragmented flag so
7481 * we don't bother trying to setup a cluster again until we get
7484 if (unlikely(last_ptr
)) {
7485 spin_lock(&last_ptr
->lock
);
7486 last_ptr
->fragmented
= 1;
7487 spin_unlock(&last_ptr
->lock
);
7489 spin_lock(&block_group
->free_space_ctl
->tree_lock
);
7491 block_group
->free_space_ctl
->free_space
<
7492 num_bytes
+ empty_cluster
+ empty_size
) {
7493 if (block_group
->free_space_ctl
->free_space
>
7496 block_group
->free_space_ctl
->free_space
;
7497 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
7500 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
7502 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
7503 num_bytes
, empty_size
,
7506 * If we didn't find a chunk, and we haven't failed on this
7507 * block group before, and this block group is in the middle of
7508 * caching and we are ok with waiting, then go ahead and wait
7509 * for progress to be made, and set failed_alloc to true.
7511 * If failed_alloc is true then we've already waited on this
7512 * block group once and should move on to the next block group.
7514 if (!offset
&& !failed_alloc
&& !cached
&&
7515 loop
> LOOP_CACHING_NOWAIT
) {
7516 wait_block_group_cache_progress(block_group
,
7517 num_bytes
+ empty_size
);
7518 failed_alloc
= true;
7519 goto have_block_group
;
7520 } else if (!offset
) {
7524 search_start
= ALIGN(offset
, root
->stripesize
);
7526 /* move on to the next group */
7527 if (search_start
+ num_bytes
>
7528 block_group
->key
.objectid
+ block_group
->key
.offset
) {
7529 btrfs_add_free_space(block_group
, offset
, num_bytes
);
7533 if (offset
< search_start
)
7534 btrfs_add_free_space(block_group
, offset
,
7535 search_start
- offset
);
7536 BUG_ON(offset
> search_start
);
7538 ret
= btrfs_update_reserved_bytes(block_group
, num_bytes
,
7539 alloc_type
, delalloc
);
7540 if (ret
== -EAGAIN
) {
7541 btrfs_add_free_space(block_group
, offset
, num_bytes
);
7544 btrfs_inc_block_group_reservations(block_group
);
7546 /* we are all good, lets return */
7547 ins
->objectid
= search_start
;
7548 ins
->offset
= num_bytes
;
7550 trace_btrfs_reserve_extent(orig_root
, block_group
,
7551 search_start
, num_bytes
);
7552 btrfs_release_block_group(block_group
, delalloc
);
7555 failed_cluster_refill
= false;
7556 failed_alloc
= false;
7557 BUG_ON(index
!= get_block_group_index(block_group
));
7558 btrfs_release_block_group(block_group
, delalloc
);
7560 up_read(&space_info
->groups_sem
);
7562 if ((loop
== LOOP_CACHING_NOWAIT
) && have_caching_bg
7563 && !orig_have_caching_bg
)
7564 orig_have_caching_bg
= true;
7566 if (!ins
->objectid
&& loop
>= LOOP_CACHING_WAIT
&& have_caching_bg
)
7569 if (!ins
->objectid
&& ++index
< BTRFS_NR_RAID_TYPES
)
7573 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7574 * caching kthreads as we move along
7575 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7576 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7577 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7580 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
) {
7582 if (loop
== LOOP_CACHING_NOWAIT
) {
7584 * We want to skip the LOOP_CACHING_WAIT step if we
7585 * don't have any uncached bgs and we've already done a
7586 * full search through.
7588 if (orig_have_caching_bg
|| !full_search
)
7589 loop
= LOOP_CACHING_WAIT
;
7591 loop
= LOOP_ALLOC_CHUNK
;
7596 if (loop
== LOOP_ALLOC_CHUNK
) {
7597 struct btrfs_trans_handle
*trans
;
7600 trans
= current
->journal_info
;
7604 trans
= btrfs_join_transaction(root
);
7606 if (IS_ERR(trans
)) {
7607 ret
= PTR_ERR(trans
);
7611 ret
= do_chunk_alloc(trans
, root
, flags
,
7615 * If we can't allocate a new chunk we've already looped
7616 * through at least once, move on to the NO_EMPTY_SIZE
7620 loop
= LOOP_NO_EMPTY_SIZE
;
7623 * Do not bail out on ENOSPC since we
7624 * can do more things.
7626 if (ret
< 0 && ret
!= -ENOSPC
)
7627 btrfs_abort_transaction(trans
,
7632 btrfs_end_transaction(trans
, root
);
7637 if (loop
== LOOP_NO_EMPTY_SIZE
) {
7639 * Don't loop again if we already have no empty_size and
7642 if (empty_size
== 0 &&
7643 empty_cluster
== 0) {
7652 } else if (!ins
->objectid
) {
7654 } else if (ins
->objectid
) {
7655 if (!use_cluster
&& last_ptr
) {
7656 spin_lock(&last_ptr
->lock
);
7657 last_ptr
->window_start
= ins
->objectid
;
7658 spin_unlock(&last_ptr
->lock
);
7663 if (ret
== -ENOSPC
) {
7664 spin_lock(&space_info
->lock
);
7665 space_info
->max_extent_size
= max_extent_size
;
7666 spin_unlock(&space_info
->lock
);
7667 ins
->offset
= max_extent_size
;
7672 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
7673 int dump_block_groups
)
7675 struct btrfs_block_group_cache
*cache
;
7678 spin_lock(&info
->lock
);
7679 printk(KERN_INFO
"BTRFS: space_info %llu has %llu free, is %sfull\n",
7681 info
->total_bytes
- info
->bytes_used
- info
->bytes_pinned
-
7682 info
->bytes_reserved
- info
->bytes_readonly
,
7683 (info
->full
) ? "" : "not ");
7684 printk(KERN_INFO
"BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7685 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7686 info
->total_bytes
, info
->bytes_used
, info
->bytes_pinned
,
7687 info
->bytes_reserved
, info
->bytes_may_use
,
7688 info
->bytes_readonly
);
7689 spin_unlock(&info
->lock
);
7691 if (!dump_block_groups
)
7694 down_read(&info
->groups_sem
);
7696 list_for_each_entry(cache
, &info
->block_groups
[index
], list
) {
7697 spin_lock(&cache
->lock
);
7698 printk(KERN_INFO
"BTRFS: "
7699 "block group %llu has %llu bytes, "
7700 "%llu used %llu pinned %llu reserved %s\n",
7701 cache
->key
.objectid
, cache
->key
.offset
,
7702 btrfs_block_group_used(&cache
->item
), cache
->pinned
,
7703 cache
->reserved
, cache
->ro
? "[readonly]" : "");
7704 btrfs_dump_free_space(cache
, bytes
);
7705 spin_unlock(&cache
->lock
);
7707 if (++index
< BTRFS_NR_RAID_TYPES
)
7709 up_read(&info
->groups_sem
);
7712 int btrfs_reserve_extent(struct btrfs_root
*root
,
7713 u64 num_bytes
, u64 min_alloc_size
,
7714 u64 empty_size
, u64 hint_byte
,
7715 struct btrfs_key
*ins
, int is_data
, int delalloc
)
7717 bool final_tried
= num_bytes
== min_alloc_size
;
7721 flags
= btrfs_get_alloc_profile(root
, is_data
);
7723 WARN_ON(num_bytes
< root
->sectorsize
);
7724 ret
= find_free_extent(root
, num_bytes
, empty_size
, hint_byte
, ins
,
7726 if (!ret
&& !is_data
) {
7727 btrfs_dec_block_group_reservations(root
->fs_info
,
7729 } else if (ret
== -ENOSPC
) {
7730 if (!final_tried
&& ins
->offset
) {
7731 num_bytes
= min(num_bytes
>> 1, ins
->offset
);
7732 num_bytes
= round_down(num_bytes
, root
->sectorsize
);
7733 num_bytes
= max(num_bytes
, min_alloc_size
);
7734 if (num_bytes
== min_alloc_size
)
7737 } else if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
7738 struct btrfs_space_info
*sinfo
;
7740 sinfo
= __find_space_info(root
->fs_info
, flags
);
7741 btrfs_err(root
->fs_info
, "allocation failed flags %llu, wanted %llu",
7744 dump_space_info(sinfo
, num_bytes
, 1);
7751 static int __btrfs_free_reserved_extent(struct btrfs_root
*root
,
7753 int pin
, int delalloc
)
7755 struct btrfs_block_group_cache
*cache
;
7758 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
7760 btrfs_err(root
->fs_info
, "Unable to find block group for %llu",
7766 pin_down_extent(root
, cache
, start
, len
, 1);
7768 if (btrfs_test_opt(root
, DISCARD
))
7769 ret
= btrfs_discard_extent(root
, start
, len
, NULL
);
7770 btrfs_add_free_space(cache
, start
, len
);
7771 btrfs_update_reserved_bytes(cache
, len
, RESERVE_FREE
, delalloc
);
7774 btrfs_put_block_group(cache
);
7776 trace_btrfs_reserved_extent_free(root
, start
, len
);
7781 int btrfs_free_reserved_extent(struct btrfs_root
*root
,
7782 u64 start
, u64 len
, int delalloc
)
7784 return __btrfs_free_reserved_extent(root
, start
, len
, 0, delalloc
);
7787 int btrfs_free_and_pin_reserved_extent(struct btrfs_root
*root
,
7790 return __btrfs_free_reserved_extent(root
, start
, len
, 1, 0);
7793 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
7794 struct btrfs_root
*root
,
7795 u64 parent
, u64 root_objectid
,
7796 u64 flags
, u64 owner
, u64 offset
,
7797 struct btrfs_key
*ins
, int ref_mod
)
7800 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7801 struct btrfs_extent_item
*extent_item
;
7802 struct btrfs_extent_inline_ref
*iref
;
7803 struct btrfs_path
*path
;
7804 struct extent_buffer
*leaf
;
7809 type
= BTRFS_SHARED_DATA_REF_KEY
;
7811 type
= BTRFS_EXTENT_DATA_REF_KEY
;
7813 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
7815 path
= btrfs_alloc_path();
7819 path
->leave_spinning
= 1;
7820 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
7823 btrfs_free_path(path
);
7827 leaf
= path
->nodes
[0];
7828 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
7829 struct btrfs_extent_item
);
7830 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
7831 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
7832 btrfs_set_extent_flags(leaf
, extent_item
,
7833 flags
| BTRFS_EXTENT_FLAG_DATA
);
7835 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
7836 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
7838 struct btrfs_shared_data_ref
*ref
;
7839 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
7840 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
7841 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
7843 struct btrfs_extent_data_ref
*ref
;
7844 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
7845 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
7846 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
7847 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
7848 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
7851 btrfs_mark_buffer_dirty(path
->nodes
[0]);
7852 btrfs_free_path(path
);
7854 ret
= remove_from_free_space_tree(trans
, fs_info
, ins
->objectid
,
7859 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1);
7860 if (ret
) { /* -ENOENT, logic error */
7861 btrfs_err(fs_info
, "update block group failed for %llu %llu",
7862 ins
->objectid
, ins
->offset
);
7865 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, ins
->offset
);
7869 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
7870 struct btrfs_root
*root
,
7871 u64 parent
, u64 root_objectid
,
7872 u64 flags
, struct btrfs_disk_key
*key
,
7873 int level
, struct btrfs_key
*ins
)
7876 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7877 struct btrfs_extent_item
*extent_item
;
7878 struct btrfs_tree_block_info
*block_info
;
7879 struct btrfs_extent_inline_ref
*iref
;
7880 struct btrfs_path
*path
;
7881 struct extent_buffer
*leaf
;
7882 u32 size
= sizeof(*extent_item
) + sizeof(*iref
);
7883 u64 num_bytes
= ins
->offset
;
7884 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
7887 if (!skinny_metadata
)
7888 size
+= sizeof(*block_info
);
7890 path
= btrfs_alloc_path();
7892 btrfs_free_and_pin_reserved_extent(root
, ins
->objectid
,
7897 path
->leave_spinning
= 1;
7898 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
7901 btrfs_free_path(path
);
7902 btrfs_free_and_pin_reserved_extent(root
, ins
->objectid
,
7907 leaf
= path
->nodes
[0];
7908 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
7909 struct btrfs_extent_item
);
7910 btrfs_set_extent_refs(leaf
, extent_item
, 1);
7911 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
7912 btrfs_set_extent_flags(leaf
, extent_item
,
7913 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
7915 if (skinny_metadata
) {
7916 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
7917 num_bytes
= root
->nodesize
;
7919 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
7920 btrfs_set_tree_block_key(leaf
, block_info
, key
);
7921 btrfs_set_tree_block_level(leaf
, block_info
, level
);
7922 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
7926 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
7927 btrfs_set_extent_inline_ref_type(leaf
, iref
,
7928 BTRFS_SHARED_BLOCK_REF_KEY
);
7929 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
7931 btrfs_set_extent_inline_ref_type(leaf
, iref
,
7932 BTRFS_TREE_BLOCK_REF_KEY
);
7933 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
7936 btrfs_mark_buffer_dirty(leaf
);
7937 btrfs_free_path(path
);
7939 ret
= remove_from_free_space_tree(trans
, fs_info
, ins
->objectid
,
7944 ret
= update_block_group(trans
, root
, ins
->objectid
, root
->nodesize
,
7946 if (ret
) { /* -ENOENT, logic error */
7947 btrfs_err(fs_info
, "update block group failed for %llu %llu",
7948 ins
->objectid
, ins
->offset
);
7952 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, root
->nodesize
);
7956 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
7957 struct btrfs_root
*root
,
7958 u64 root_objectid
, u64 owner
,
7959 u64 offset
, u64 ram_bytes
,
7960 struct btrfs_key
*ins
)
7964 BUG_ON(root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
7966 ret
= btrfs_add_delayed_data_ref(root
->fs_info
, trans
, ins
->objectid
,
7968 root_objectid
, owner
, offset
,
7969 ram_bytes
, BTRFS_ADD_DELAYED_EXTENT
,
7975 * this is used by the tree logging recovery code. It records that
7976 * an extent has been allocated and makes sure to clear the free
7977 * space cache bits as well
7979 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
7980 struct btrfs_root
*root
,
7981 u64 root_objectid
, u64 owner
, u64 offset
,
7982 struct btrfs_key
*ins
)
7985 struct btrfs_block_group_cache
*block_group
;
7988 * Mixed block groups will exclude before processing the log so we only
7989 * need to do the exclude dance if this fs isn't mixed.
7991 if (!btrfs_fs_incompat(root
->fs_info
, MIXED_GROUPS
)) {
7992 ret
= __exclude_logged_extent(root
, ins
->objectid
, ins
->offset
);
7997 block_group
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
8001 ret
= btrfs_update_reserved_bytes(block_group
, ins
->offset
,
8002 RESERVE_ALLOC_NO_ACCOUNT
, 0);
8003 BUG_ON(ret
); /* logic error */
8004 ret
= alloc_reserved_file_extent(trans
, root
, 0, root_objectid
,
8005 0, owner
, offset
, ins
, 1);
8006 btrfs_put_block_group(block_group
);
8010 static struct extent_buffer
*
8011 btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
8012 u64 bytenr
, int level
)
8014 struct extent_buffer
*buf
;
8016 buf
= btrfs_find_create_tree_block(root
, bytenr
);
8020 btrfs_set_header_generation(buf
, trans
->transid
);
8021 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, buf
, level
);
8022 btrfs_tree_lock(buf
);
8023 clean_tree_block(trans
, root
->fs_info
, buf
);
8024 clear_bit(EXTENT_BUFFER_STALE
, &buf
->bflags
);
8026 btrfs_set_lock_blocking(buf
);
8027 set_extent_buffer_uptodate(buf
);
8029 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
8030 buf
->log_index
= root
->log_transid
% 2;
8032 * we allow two log transactions at a time, use different
8033 * EXENT bit to differentiate dirty pages.
8035 if (buf
->log_index
== 0)
8036 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
8037 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
8039 set_extent_new(&root
->dirty_log_pages
, buf
->start
,
8040 buf
->start
+ buf
->len
- 1);
8042 buf
->log_index
= -1;
8043 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
8044 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
8046 trans
->dirty
= true;
8047 /* this returns a buffer locked for blocking */
8051 static struct btrfs_block_rsv
*
8052 use_block_rsv(struct btrfs_trans_handle
*trans
,
8053 struct btrfs_root
*root
, u32 blocksize
)
8055 struct btrfs_block_rsv
*block_rsv
;
8056 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
8058 bool global_updated
= false;
8060 block_rsv
= get_block_rsv(trans
, root
);
8062 if (unlikely(block_rsv
->size
== 0))
8065 ret
= block_rsv_use_bytes(block_rsv
, blocksize
);
8069 if (block_rsv
->failfast
)
8070 return ERR_PTR(ret
);
8072 if (block_rsv
->type
== BTRFS_BLOCK_RSV_GLOBAL
&& !global_updated
) {
8073 global_updated
= true;
8074 update_global_block_rsv(root
->fs_info
);
8078 if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
8079 static DEFINE_RATELIMIT_STATE(_rs
,
8080 DEFAULT_RATELIMIT_INTERVAL
* 10,
8081 /*DEFAULT_RATELIMIT_BURST*/ 1);
8082 if (__ratelimit(&_rs
))
8084 "BTRFS: block rsv returned %d\n", ret
);
8087 ret
= reserve_metadata_bytes(root
, block_rsv
, blocksize
,
8088 BTRFS_RESERVE_NO_FLUSH
);
8092 * If we couldn't reserve metadata bytes try and use some from
8093 * the global reserve if its space type is the same as the global
8096 if (block_rsv
->type
!= BTRFS_BLOCK_RSV_GLOBAL
&&
8097 block_rsv
->space_info
== global_rsv
->space_info
) {
8098 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
8102 return ERR_PTR(ret
);
8105 static void unuse_block_rsv(struct btrfs_fs_info
*fs_info
,
8106 struct btrfs_block_rsv
*block_rsv
, u32 blocksize
)
8108 block_rsv_add_bytes(block_rsv
, blocksize
, 0);
8109 block_rsv_release_bytes(fs_info
, block_rsv
, NULL
, 0);
8113 * finds a free extent and does all the dirty work required for allocation
8114 * returns the tree buffer or an ERR_PTR on error.
8116 struct extent_buffer
*btrfs_alloc_tree_block(struct btrfs_trans_handle
*trans
,
8117 struct btrfs_root
*root
,
8118 u64 parent
, u64 root_objectid
,
8119 struct btrfs_disk_key
*key
, int level
,
8120 u64 hint
, u64 empty_size
)
8122 struct btrfs_key ins
;
8123 struct btrfs_block_rsv
*block_rsv
;
8124 struct extent_buffer
*buf
;
8125 struct btrfs_delayed_extent_op
*extent_op
;
8128 u32 blocksize
= root
->nodesize
;
8129 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
8132 if (btrfs_test_is_dummy_root(root
)) {
8133 buf
= btrfs_init_new_buffer(trans
, root
, root
->alloc_bytenr
,
8136 root
->alloc_bytenr
+= blocksize
;
8140 block_rsv
= use_block_rsv(trans
, root
, blocksize
);
8141 if (IS_ERR(block_rsv
))
8142 return ERR_CAST(block_rsv
);
8144 ret
= btrfs_reserve_extent(root
, blocksize
, blocksize
,
8145 empty_size
, hint
, &ins
, 0, 0);
8149 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
, level
);
8152 goto out_free_reserved
;
8155 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
8157 parent
= ins
.objectid
;
8158 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
8162 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
8163 extent_op
= btrfs_alloc_delayed_extent_op();
8169 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
8171 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
8172 extent_op
->flags_to_set
= flags
;
8173 extent_op
->update_key
= skinny_metadata
? false : true;
8174 extent_op
->update_flags
= true;
8175 extent_op
->is_data
= false;
8176 extent_op
->level
= level
;
8178 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
8179 ins
.objectid
, ins
.offset
,
8180 parent
, root_objectid
, level
,
8181 BTRFS_ADD_DELAYED_EXTENT
,
8184 goto out_free_delayed
;
8189 btrfs_free_delayed_extent_op(extent_op
);
8191 free_extent_buffer(buf
);
8193 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
, 0);
8195 unuse_block_rsv(root
->fs_info
, block_rsv
, blocksize
);
8196 return ERR_PTR(ret
);
8199 struct walk_control
{
8200 u64 refs
[BTRFS_MAX_LEVEL
];
8201 u64 flags
[BTRFS_MAX_LEVEL
];
8202 struct btrfs_key update_progress
;
8213 #define DROP_REFERENCE 1
8214 #define UPDATE_BACKREF 2
8216 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
8217 struct btrfs_root
*root
,
8218 struct walk_control
*wc
,
8219 struct btrfs_path
*path
)
8227 struct btrfs_key key
;
8228 struct extent_buffer
*eb
;
8233 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
8234 wc
->reada_count
= wc
->reada_count
* 2 / 3;
8235 wc
->reada_count
= max(wc
->reada_count
, 2);
8237 wc
->reada_count
= wc
->reada_count
* 3 / 2;
8238 wc
->reada_count
= min_t(int, wc
->reada_count
,
8239 BTRFS_NODEPTRS_PER_BLOCK(root
));
8242 eb
= path
->nodes
[wc
->level
];
8243 nritems
= btrfs_header_nritems(eb
);
8244 blocksize
= root
->nodesize
;
8246 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
8247 if (nread
>= wc
->reada_count
)
8251 bytenr
= btrfs_node_blockptr(eb
, slot
);
8252 generation
= btrfs_node_ptr_generation(eb
, slot
);
8254 if (slot
== path
->slots
[wc
->level
])
8257 if (wc
->stage
== UPDATE_BACKREF
&&
8258 generation
<= root
->root_key
.offset
)
8261 /* We don't lock the tree block, it's OK to be racy here */
8262 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
,
8263 wc
->level
- 1, 1, &refs
,
8265 /* We don't care about errors in readahead. */
8270 if (wc
->stage
== DROP_REFERENCE
) {
8274 if (wc
->level
== 1 &&
8275 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8277 if (!wc
->update_ref
||
8278 generation
<= root
->root_key
.offset
)
8280 btrfs_node_key_to_cpu(eb
, &key
, slot
);
8281 ret
= btrfs_comp_cpu_keys(&key
,
8282 &wc
->update_progress
);
8286 if (wc
->level
== 1 &&
8287 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8291 readahead_tree_block(root
, bytenr
);
8294 wc
->reada_slot
= slot
;
8298 * These may not be seen by the usual inc/dec ref code so we have to
8301 static int record_one_subtree_extent(struct btrfs_trans_handle
*trans
,
8302 struct btrfs_root
*root
, u64 bytenr
,
8305 struct btrfs_qgroup_extent_record
*qrecord
;
8306 struct btrfs_delayed_ref_root
*delayed_refs
;
8308 qrecord
= kmalloc(sizeof(*qrecord
), GFP_NOFS
);
8312 qrecord
->bytenr
= bytenr
;
8313 qrecord
->num_bytes
= num_bytes
;
8314 qrecord
->old_roots
= NULL
;
8316 delayed_refs
= &trans
->transaction
->delayed_refs
;
8317 spin_lock(&delayed_refs
->lock
);
8318 if (btrfs_qgroup_insert_dirty_extent(delayed_refs
, qrecord
))
8320 spin_unlock(&delayed_refs
->lock
);
8325 static int account_leaf_items(struct btrfs_trans_handle
*trans
,
8326 struct btrfs_root
*root
,
8327 struct extent_buffer
*eb
)
8329 int nr
= btrfs_header_nritems(eb
);
8330 int i
, extent_type
, ret
;
8331 struct btrfs_key key
;
8332 struct btrfs_file_extent_item
*fi
;
8333 u64 bytenr
, num_bytes
;
8335 /* We can be called directly from walk_up_proc() */
8336 if (!root
->fs_info
->quota_enabled
)
8339 for (i
= 0; i
< nr
; i
++) {
8340 btrfs_item_key_to_cpu(eb
, &key
, i
);
8342 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
8345 fi
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
8346 /* filter out non qgroup-accountable extents */
8347 extent_type
= btrfs_file_extent_type(eb
, fi
);
8349 if (extent_type
== BTRFS_FILE_EXTENT_INLINE
)
8352 bytenr
= btrfs_file_extent_disk_bytenr(eb
, fi
);
8356 num_bytes
= btrfs_file_extent_disk_num_bytes(eb
, fi
);
8358 ret
= record_one_subtree_extent(trans
, root
, bytenr
, num_bytes
);
8366 * Walk up the tree from the bottom, freeing leaves and any interior
8367 * nodes which have had all slots visited. If a node (leaf or
8368 * interior) is freed, the node above it will have it's slot
8369 * incremented. The root node will never be freed.
8371 * At the end of this function, we should have a path which has all
8372 * slots incremented to the next position for a search. If we need to
8373 * read a new node it will be NULL and the node above it will have the
8374 * correct slot selected for a later read.
8376 * If we increment the root nodes slot counter past the number of
8377 * elements, 1 is returned to signal completion of the search.
8379 static int adjust_slots_upwards(struct btrfs_root
*root
,
8380 struct btrfs_path
*path
, int root_level
)
8384 struct extent_buffer
*eb
;
8386 if (root_level
== 0)
8389 while (level
<= root_level
) {
8390 eb
= path
->nodes
[level
];
8391 nr
= btrfs_header_nritems(eb
);
8392 path
->slots
[level
]++;
8393 slot
= path
->slots
[level
];
8394 if (slot
>= nr
|| level
== 0) {
8396 * Don't free the root - we will detect this
8397 * condition after our loop and return a
8398 * positive value for caller to stop walking the tree.
8400 if (level
!= root_level
) {
8401 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8402 path
->locks
[level
] = 0;
8404 free_extent_buffer(eb
);
8405 path
->nodes
[level
] = NULL
;
8406 path
->slots
[level
] = 0;
8410 * We have a valid slot to walk back down
8411 * from. Stop here so caller can process these
8420 eb
= path
->nodes
[root_level
];
8421 if (path
->slots
[root_level
] >= btrfs_header_nritems(eb
))
8428 * root_eb is the subtree root and is locked before this function is called.
8430 static int account_shared_subtree(struct btrfs_trans_handle
*trans
,
8431 struct btrfs_root
*root
,
8432 struct extent_buffer
*root_eb
,
8438 struct extent_buffer
*eb
= root_eb
;
8439 struct btrfs_path
*path
= NULL
;
8441 BUG_ON(root_level
< 0 || root_level
> BTRFS_MAX_LEVEL
);
8442 BUG_ON(root_eb
== NULL
);
8444 if (!root
->fs_info
->quota_enabled
)
8447 if (!extent_buffer_uptodate(root_eb
)) {
8448 ret
= btrfs_read_buffer(root_eb
, root_gen
);
8453 if (root_level
== 0) {
8454 ret
= account_leaf_items(trans
, root
, root_eb
);
8458 path
= btrfs_alloc_path();
8463 * Walk down the tree. Missing extent blocks are filled in as
8464 * we go. Metadata is accounted every time we read a new
8467 * When we reach a leaf, we account for file extent items in it,
8468 * walk back up the tree (adjusting slot pointers as we go)
8469 * and restart the search process.
8471 extent_buffer_get(root_eb
); /* For path */
8472 path
->nodes
[root_level
] = root_eb
;
8473 path
->slots
[root_level
] = 0;
8474 path
->locks
[root_level
] = 0; /* so release_path doesn't try to unlock */
8477 while (level
>= 0) {
8478 if (path
->nodes
[level
] == NULL
) {
8483 /* We need to get child blockptr/gen from
8484 * parent before we can read it. */
8485 eb
= path
->nodes
[level
+ 1];
8486 parent_slot
= path
->slots
[level
+ 1];
8487 child_bytenr
= btrfs_node_blockptr(eb
, parent_slot
);
8488 child_gen
= btrfs_node_ptr_generation(eb
, parent_slot
);
8490 eb
= read_tree_block(root
, child_bytenr
, child_gen
);
8494 } else if (!extent_buffer_uptodate(eb
)) {
8495 free_extent_buffer(eb
);
8500 path
->nodes
[level
] = eb
;
8501 path
->slots
[level
] = 0;
8503 btrfs_tree_read_lock(eb
);
8504 btrfs_set_lock_blocking_rw(eb
, BTRFS_READ_LOCK
);
8505 path
->locks
[level
] = BTRFS_READ_LOCK_BLOCKING
;
8507 ret
= record_one_subtree_extent(trans
, root
, child_bytenr
,
8514 ret
= account_leaf_items(trans
, root
, path
->nodes
[level
]);
8518 /* Nonzero return here means we completed our search */
8519 ret
= adjust_slots_upwards(root
, path
, root_level
);
8523 /* Restart search with new slots */
8532 btrfs_free_path(path
);
8538 * helper to process tree block while walking down the tree.
8540 * when wc->stage == UPDATE_BACKREF, this function updates
8541 * back refs for pointers in the block.
8543 * NOTE: return value 1 means we should stop walking down.
8545 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
8546 struct btrfs_root
*root
,
8547 struct btrfs_path
*path
,
8548 struct walk_control
*wc
, int lookup_info
)
8550 int level
= wc
->level
;
8551 struct extent_buffer
*eb
= path
->nodes
[level
];
8552 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
8555 if (wc
->stage
== UPDATE_BACKREF
&&
8556 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
8560 * when reference count of tree block is 1, it won't increase
8561 * again. once full backref flag is set, we never clear it.
8564 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
8565 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
8566 BUG_ON(!path
->locks
[level
]);
8567 ret
= btrfs_lookup_extent_info(trans
, root
,
8568 eb
->start
, level
, 1,
8571 BUG_ON(ret
== -ENOMEM
);
8574 BUG_ON(wc
->refs
[level
] == 0);
8577 if (wc
->stage
== DROP_REFERENCE
) {
8578 if (wc
->refs
[level
] > 1)
8581 if (path
->locks
[level
] && !wc
->keep_locks
) {
8582 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8583 path
->locks
[level
] = 0;
8588 /* wc->stage == UPDATE_BACKREF */
8589 if (!(wc
->flags
[level
] & flag
)) {
8590 BUG_ON(!path
->locks
[level
]);
8591 ret
= btrfs_inc_ref(trans
, root
, eb
, 1);
8592 BUG_ON(ret
); /* -ENOMEM */
8593 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
8594 BUG_ON(ret
); /* -ENOMEM */
8595 ret
= btrfs_set_disk_extent_flags(trans
, root
, eb
->start
,
8597 btrfs_header_level(eb
), 0);
8598 BUG_ON(ret
); /* -ENOMEM */
8599 wc
->flags
[level
] |= flag
;
8603 * the block is shared by multiple trees, so it's not good to
8604 * keep the tree lock
8606 if (path
->locks
[level
] && level
> 0) {
8607 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8608 path
->locks
[level
] = 0;
8614 * helper to process tree block pointer.
8616 * when wc->stage == DROP_REFERENCE, this function checks
8617 * reference count of the block pointed to. if the block
8618 * is shared and we need update back refs for the subtree
8619 * rooted at the block, this function changes wc->stage to
8620 * UPDATE_BACKREF. if the block is shared and there is no
8621 * need to update back, this function drops the reference
8624 * NOTE: return value 1 means we should stop walking down.
8626 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
8627 struct btrfs_root
*root
,
8628 struct btrfs_path
*path
,
8629 struct walk_control
*wc
, int *lookup_info
)
8635 struct btrfs_key key
;
8636 struct extent_buffer
*next
;
8637 int level
= wc
->level
;
8640 bool need_account
= false;
8642 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
8643 path
->slots
[level
]);
8645 * if the lower level block was created before the snapshot
8646 * was created, we know there is no need to update back refs
8649 if (wc
->stage
== UPDATE_BACKREF
&&
8650 generation
<= root
->root_key
.offset
) {
8655 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
8656 blocksize
= root
->nodesize
;
8658 next
= btrfs_find_tree_block(root
->fs_info
, bytenr
);
8660 next
= btrfs_find_create_tree_block(root
, bytenr
);
8662 return PTR_ERR(next
);
8664 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, next
,
8668 btrfs_tree_lock(next
);
8669 btrfs_set_lock_blocking(next
);
8671 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, level
- 1, 1,
8672 &wc
->refs
[level
- 1],
8673 &wc
->flags
[level
- 1]);
8675 btrfs_tree_unlock(next
);
8679 if (unlikely(wc
->refs
[level
- 1] == 0)) {
8680 btrfs_err(root
->fs_info
, "Missing references.");
8685 if (wc
->stage
== DROP_REFERENCE
) {
8686 if (wc
->refs
[level
- 1] > 1) {
8687 need_account
= true;
8689 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8692 if (!wc
->update_ref
||
8693 generation
<= root
->root_key
.offset
)
8696 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
8697 path
->slots
[level
]);
8698 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
8702 wc
->stage
= UPDATE_BACKREF
;
8703 wc
->shared_level
= level
- 1;
8707 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8711 if (!btrfs_buffer_uptodate(next
, generation
, 0)) {
8712 btrfs_tree_unlock(next
);
8713 free_extent_buffer(next
);
8719 if (reada
&& level
== 1)
8720 reada_walk_down(trans
, root
, wc
, path
);
8721 next
= read_tree_block(root
, bytenr
, generation
);
8723 return PTR_ERR(next
);
8724 } else if (!extent_buffer_uptodate(next
)) {
8725 free_extent_buffer(next
);
8728 btrfs_tree_lock(next
);
8729 btrfs_set_lock_blocking(next
);
8733 BUG_ON(level
!= btrfs_header_level(next
));
8734 path
->nodes
[level
] = next
;
8735 path
->slots
[level
] = 0;
8736 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8742 wc
->refs
[level
- 1] = 0;
8743 wc
->flags
[level
- 1] = 0;
8744 if (wc
->stage
== DROP_REFERENCE
) {
8745 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
8746 parent
= path
->nodes
[level
]->start
;
8748 BUG_ON(root
->root_key
.objectid
!=
8749 btrfs_header_owner(path
->nodes
[level
]));
8754 ret
= account_shared_subtree(trans
, root
, next
,
8755 generation
, level
- 1);
8757 btrfs_err_rl(root
->fs_info
,
8759 "%d accounting shared subtree. Quota "
8760 "is out of sync, rescan required.",
8764 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
, parent
,
8765 root
->root_key
.objectid
, level
- 1, 0);
8766 BUG_ON(ret
); /* -ENOMEM */
8768 btrfs_tree_unlock(next
);
8769 free_extent_buffer(next
);
8775 * helper to process tree block while walking up the tree.
8777 * when wc->stage == DROP_REFERENCE, this function drops
8778 * reference count on the block.
8780 * when wc->stage == UPDATE_BACKREF, this function changes
8781 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8782 * to UPDATE_BACKREF previously while processing the block.
8784 * NOTE: return value 1 means we should stop walking up.
8786 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
8787 struct btrfs_root
*root
,
8788 struct btrfs_path
*path
,
8789 struct walk_control
*wc
)
8792 int level
= wc
->level
;
8793 struct extent_buffer
*eb
= path
->nodes
[level
];
8796 if (wc
->stage
== UPDATE_BACKREF
) {
8797 BUG_ON(wc
->shared_level
< level
);
8798 if (level
< wc
->shared_level
)
8801 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
8805 wc
->stage
= DROP_REFERENCE
;
8806 wc
->shared_level
= -1;
8807 path
->slots
[level
] = 0;
8810 * check reference count again if the block isn't locked.
8811 * we should start walking down the tree again if reference
8814 if (!path
->locks
[level
]) {
8816 btrfs_tree_lock(eb
);
8817 btrfs_set_lock_blocking(eb
);
8818 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8820 ret
= btrfs_lookup_extent_info(trans
, root
,
8821 eb
->start
, level
, 1,
8825 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8826 path
->locks
[level
] = 0;
8829 BUG_ON(wc
->refs
[level
] == 0);
8830 if (wc
->refs
[level
] == 1) {
8831 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8832 path
->locks
[level
] = 0;
8838 /* wc->stage == DROP_REFERENCE */
8839 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
8841 if (wc
->refs
[level
] == 1) {
8843 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8844 ret
= btrfs_dec_ref(trans
, root
, eb
, 1);
8846 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
8847 BUG_ON(ret
); /* -ENOMEM */
8848 ret
= account_leaf_items(trans
, root
, eb
);
8850 btrfs_err_rl(root
->fs_info
,
8852 "%d accounting leaf items. Quota "
8853 "is out of sync, rescan required.",
8857 /* make block locked assertion in clean_tree_block happy */
8858 if (!path
->locks
[level
] &&
8859 btrfs_header_generation(eb
) == trans
->transid
) {
8860 btrfs_tree_lock(eb
);
8861 btrfs_set_lock_blocking(eb
);
8862 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8864 clean_tree_block(trans
, root
->fs_info
, eb
);
8867 if (eb
== root
->node
) {
8868 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8871 BUG_ON(root
->root_key
.objectid
!=
8872 btrfs_header_owner(eb
));
8874 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8875 parent
= path
->nodes
[level
+ 1]->start
;
8877 BUG_ON(root
->root_key
.objectid
!=
8878 btrfs_header_owner(path
->nodes
[level
+ 1]));
8881 btrfs_free_tree_block(trans
, root
, eb
, parent
, wc
->refs
[level
] == 1);
8883 wc
->refs
[level
] = 0;
8884 wc
->flags
[level
] = 0;
8888 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
8889 struct btrfs_root
*root
,
8890 struct btrfs_path
*path
,
8891 struct walk_control
*wc
)
8893 int level
= wc
->level
;
8894 int lookup_info
= 1;
8897 while (level
>= 0) {
8898 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
8905 if (path
->slots
[level
] >=
8906 btrfs_header_nritems(path
->nodes
[level
]))
8909 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
8911 path
->slots
[level
]++;
8920 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
8921 struct btrfs_root
*root
,
8922 struct btrfs_path
*path
,
8923 struct walk_control
*wc
, int max_level
)
8925 int level
= wc
->level
;
8928 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
8929 while (level
< max_level
&& path
->nodes
[level
]) {
8931 if (path
->slots
[level
] + 1 <
8932 btrfs_header_nritems(path
->nodes
[level
])) {
8933 path
->slots
[level
]++;
8936 ret
= walk_up_proc(trans
, root
, path
, wc
);
8940 if (path
->locks
[level
]) {
8941 btrfs_tree_unlock_rw(path
->nodes
[level
],
8942 path
->locks
[level
]);
8943 path
->locks
[level
] = 0;
8945 free_extent_buffer(path
->nodes
[level
]);
8946 path
->nodes
[level
] = NULL
;
8954 * drop a subvolume tree.
8956 * this function traverses the tree freeing any blocks that only
8957 * referenced by the tree.
8959 * when a shared tree block is found. this function decreases its
8960 * reference count by one. if update_ref is true, this function
8961 * also make sure backrefs for the shared block and all lower level
8962 * blocks are properly updated.
8964 * If called with for_reloc == 0, may exit early with -EAGAIN
8966 int btrfs_drop_snapshot(struct btrfs_root
*root
,
8967 struct btrfs_block_rsv
*block_rsv
, int update_ref
,
8970 struct btrfs_path
*path
;
8971 struct btrfs_trans_handle
*trans
;
8972 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
8973 struct btrfs_root_item
*root_item
= &root
->root_item
;
8974 struct walk_control
*wc
;
8975 struct btrfs_key key
;
8979 bool root_dropped
= false;
8981 btrfs_debug(root
->fs_info
, "Drop subvolume %llu", root
->objectid
);
8983 path
= btrfs_alloc_path();
8989 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
8991 btrfs_free_path(path
);
8996 trans
= btrfs_start_transaction(tree_root
, 0);
8997 if (IS_ERR(trans
)) {
8998 err
= PTR_ERR(trans
);
9003 trans
->block_rsv
= block_rsv
;
9005 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
9006 level
= btrfs_header_level(root
->node
);
9007 path
->nodes
[level
] = btrfs_lock_root_node(root
);
9008 btrfs_set_lock_blocking(path
->nodes
[level
]);
9009 path
->slots
[level
] = 0;
9010 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
9011 memset(&wc
->update_progress
, 0,
9012 sizeof(wc
->update_progress
));
9014 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
9015 memcpy(&wc
->update_progress
, &key
,
9016 sizeof(wc
->update_progress
));
9018 level
= root_item
->drop_level
;
9020 path
->lowest_level
= level
;
9021 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
9022 path
->lowest_level
= 0;
9030 * unlock our path, this is safe because only this
9031 * function is allowed to delete this snapshot
9033 btrfs_unlock_up_safe(path
, 0);
9035 level
= btrfs_header_level(root
->node
);
9037 btrfs_tree_lock(path
->nodes
[level
]);
9038 btrfs_set_lock_blocking(path
->nodes
[level
]);
9039 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
9041 ret
= btrfs_lookup_extent_info(trans
, root
,
9042 path
->nodes
[level
]->start
,
9043 level
, 1, &wc
->refs
[level
],
9049 BUG_ON(wc
->refs
[level
] == 0);
9051 if (level
== root_item
->drop_level
)
9054 btrfs_tree_unlock(path
->nodes
[level
]);
9055 path
->locks
[level
] = 0;
9056 WARN_ON(wc
->refs
[level
] != 1);
9062 wc
->shared_level
= -1;
9063 wc
->stage
= DROP_REFERENCE
;
9064 wc
->update_ref
= update_ref
;
9066 wc
->for_reloc
= for_reloc
;
9067 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
9071 ret
= walk_down_tree(trans
, root
, path
, wc
);
9077 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
9084 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
9088 if (wc
->stage
== DROP_REFERENCE
) {
9090 btrfs_node_key(path
->nodes
[level
],
9091 &root_item
->drop_progress
,
9092 path
->slots
[level
]);
9093 root_item
->drop_level
= level
;
9096 BUG_ON(wc
->level
== 0);
9097 if (btrfs_should_end_transaction(trans
, tree_root
) ||
9098 (!for_reloc
&& btrfs_need_cleaner_sleep(root
))) {
9099 ret
= btrfs_update_root(trans
, tree_root
,
9103 btrfs_abort_transaction(trans
, tree_root
, ret
);
9108 btrfs_end_transaction_throttle(trans
, tree_root
);
9109 if (!for_reloc
&& btrfs_need_cleaner_sleep(root
)) {
9110 pr_debug("BTRFS: drop snapshot early exit\n");
9115 trans
= btrfs_start_transaction(tree_root
, 0);
9116 if (IS_ERR(trans
)) {
9117 err
= PTR_ERR(trans
);
9121 trans
->block_rsv
= block_rsv
;
9124 btrfs_release_path(path
);
9128 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
9130 btrfs_abort_transaction(trans
, tree_root
, ret
);
9134 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
9135 ret
= btrfs_find_root(tree_root
, &root
->root_key
, path
,
9138 btrfs_abort_transaction(trans
, tree_root
, ret
);
9141 } else if (ret
> 0) {
9142 /* if we fail to delete the orphan item this time
9143 * around, it'll get picked up the next time.
9145 * The most common failure here is just -ENOENT.
9147 btrfs_del_orphan_item(trans
, tree_root
,
9148 root
->root_key
.objectid
);
9152 if (test_bit(BTRFS_ROOT_IN_RADIX
, &root
->state
)) {
9153 btrfs_add_dropped_root(trans
, root
);
9155 free_extent_buffer(root
->node
);
9156 free_extent_buffer(root
->commit_root
);
9157 btrfs_put_fs_root(root
);
9159 root_dropped
= true;
9161 btrfs_end_transaction_throttle(trans
, tree_root
);
9164 btrfs_free_path(path
);
9167 * So if we need to stop dropping the snapshot for whatever reason we
9168 * need to make sure to add it back to the dead root list so that we
9169 * keep trying to do the work later. This also cleans up roots if we
9170 * don't have it in the radix (like when we recover after a power fail
9171 * or unmount) so we don't leak memory.
9173 if (!for_reloc
&& root_dropped
== false)
9174 btrfs_add_dead_root(root
);
9175 if (err
&& err
!= -EAGAIN
)
9176 btrfs_handle_fs_error(root
->fs_info
, err
, NULL
);
9181 * drop subtree rooted at tree block 'node'.
9183 * NOTE: this function will unlock and release tree block 'node'
9184 * only used by relocation code
9186 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
9187 struct btrfs_root
*root
,
9188 struct extent_buffer
*node
,
9189 struct extent_buffer
*parent
)
9191 struct btrfs_path
*path
;
9192 struct walk_control
*wc
;
9198 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
9200 path
= btrfs_alloc_path();
9204 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
9206 btrfs_free_path(path
);
9210 btrfs_assert_tree_locked(parent
);
9211 parent_level
= btrfs_header_level(parent
);
9212 extent_buffer_get(parent
);
9213 path
->nodes
[parent_level
] = parent
;
9214 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
9216 btrfs_assert_tree_locked(node
);
9217 level
= btrfs_header_level(node
);
9218 path
->nodes
[level
] = node
;
9219 path
->slots
[level
] = 0;
9220 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
9222 wc
->refs
[parent_level
] = 1;
9223 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
9225 wc
->shared_level
= -1;
9226 wc
->stage
= DROP_REFERENCE
;
9230 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
9233 wret
= walk_down_tree(trans
, root
, path
, wc
);
9239 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
9247 btrfs_free_path(path
);
9251 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
9257 * if restripe for this chunk_type is on pick target profile and
9258 * return, otherwise do the usual balance
9260 stripped
= get_restripe_target(root
->fs_info
, flags
);
9262 return extended_to_chunk(stripped
);
9264 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
9266 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
9267 BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
|
9268 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
9270 if (num_devices
== 1) {
9271 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
9272 stripped
= flags
& ~stripped
;
9274 /* turn raid0 into single device chunks */
9275 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
9278 /* turn mirroring into duplication */
9279 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
9280 BTRFS_BLOCK_GROUP_RAID10
))
9281 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
9283 /* they already had raid on here, just return */
9284 if (flags
& stripped
)
9287 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
9288 stripped
= flags
& ~stripped
;
9290 /* switch duplicated blocks with raid1 */
9291 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
9292 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
9294 /* this is drive concat, leave it alone */
9300 static int inc_block_group_ro(struct btrfs_block_group_cache
*cache
, int force
)
9302 struct btrfs_space_info
*sinfo
= cache
->space_info
;
9304 u64 min_allocable_bytes
;
9308 * We need some metadata space and system metadata space for
9309 * allocating chunks in some corner cases until we force to set
9310 * it to be readonly.
9313 (BTRFS_BLOCK_GROUP_SYSTEM
| BTRFS_BLOCK_GROUP_METADATA
)) &&
9315 min_allocable_bytes
= SZ_1M
;
9317 min_allocable_bytes
= 0;
9319 spin_lock(&sinfo
->lock
);
9320 spin_lock(&cache
->lock
);
9328 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
9329 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
9331 if (sinfo
->bytes_used
+ sinfo
->bytes_reserved
+ sinfo
->bytes_pinned
+
9332 sinfo
->bytes_may_use
+ sinfo
->bytes_readonly
+ num_bytes
+
9333 min_allocable_bytes
<= sinfo
->total_bytes
) {
9334 sinfo
->bytes_readonly
+= num_bytes
;
9336 list_add_tail(&cache
->ro_list
, &sinfo
->ro_bgs
);
9340 spin_unlock(&cache
->lock
);
9341 spin_unlock(&sinfo
->lock
);
9345 int btrfs_inc_block_group_ro(struct btrfs_root
*root
,
9346 struct btrfs_block_group_cache
*cache
)
9349 struct btrfs_trans_handle
*trans
;
9354 trans
= btrfs_join_transaction(root
);
9356 return PTR_ERR(trans
);
9359 * we're not allowed to set block groups readonly after the dirty
9360 * block groups cache has started writing. If it already started,
9361 * back off and let this transaction commit
9363 mutex_lock(&root
->fs_info
->ro_block_group_mutex
);
9364 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN
, &trans
->transaction
->flags
)) {
9365 u64 transid
= trans
->transid
;
9367 mutex_unlock(&root
->fs_info
->ro_block_group_mutex
);
9368 btrfs_end_transaction(trans
, root
);
9370 ret
= btrfs_wait_for_commit(root
, transid
);
9377 * if we are changing raid levels, try to allocate a corresponding
9378 * block group with the new raid level.
9380 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
9381 if (alloc_flags
!= cache
->flags
) {
9382 ret
= do_chunk_alloc(trans
, root
, alloc_flags
,
9385 * ENOSPC is allowed here, we may have enough space
9386 * already allocated at the new raid level to
9395 ret
= inc_block_group_ro(cache
, 0);
9398 alloc_flags
= get_alloc_profile(root
, cache
->space_info
->flags
);
9399 ret
= do_chunk_alloc(trans
, root
, alloc_flags
,
9403 ret
= inc_block_group_ro(cache
, 0);
9405 if (cache
->flags
& BTRFS_BLOCK_GROUP_SYSTEM
) {
9406 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
9407 lock_chunks(root
->fs_info
->chunk_root
);
9408 check_system_chunk(trans
, root
, alloc_flags
);
9409 unlock_chunks(root
->fs_info
->chunk_root
);
9411 mutex_unlock(&root
->fs_info
->ro_block_group_mutex
);
9413 btrfs_end_transaction(trans
, root
);
9417 int btrfs_force_chunk_alloc(struct btrfs_trans_handle
*trans
,
9418 struct btrfs_root
*root
, u64 type
)
9420 u64 alloc_flags
= get_alloc_profile(root
, type
);
9421 return do_chunk_alloc(trans
, root
, alloc_flags
,
9426 * helper to account the unused space of all the readonly block group in the
9427 * space_info. takes mirrors into account.
9429 u64
btrfs_account_ro_block_groups_free_space(struct btrfs_space_info
*sinfo
)
9431 struct btrfs_block_group_cache
*block_group
;
9435 /* It's df, we don't care if it's racy */
9436 if (list_empty(&sinfo
->ro_bgs
))
9439 spin_lock(&sinfo
->lock
);
9440 list_for_each_entry(block_group
, &sinfo
->ro_bgs
, ro_list
) {
9441 spin_lock(&block_group
->lock
);
9443 if (!block_group
->ro
) {
9444 spin_unlock(&block_group
->lock
);
9448 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_RAID1
|
9449 BTRFS_BLOCK_GROUP_RAID10
|
9450 BTRFS_BLOCK_GROUP_DUP
))
9455 free_bytes
+= (block_group
->key
.offset
-
9456 btrfs_block_group_used(&block_group
->item
)) *
9459 spin_unlock(&block_group
->lock
);
9461 spin_unlock(&sinfo
->lock
);
9466 void btrfs_dec_block_group_ro(struct btrfs_root
*root
,
9467 struct btrfs_block_group_cache
*cache
)
9469 struct btrfs_space_info
*sinfo
= cache
->space_info
;
9474 spin_lock(&sinfo
->lock
);
9475 spin_lock(&cache
->lock
);
9477 num_bytes
= cache
->key
.offset
- cache
->reserved
-
9478 cache
->pinned
- cache
->bytes_super
-
9479 btrfs_block_group_used(&cache
->item
);
9480 sinfo
->bytes_readonly
-= num_bytes
;
9481 list_del_init(&cache
->ro_list
);
9483 spin_unlock(&cache
->lock
);
9484 spin_unlock(&sinfo
->lock
);
9488 * checks to see if its even possible to relocate this block group.
9490 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9491 * ok to go ahead and try.
9493 int btrfs_can_relocate(struct btrfs_root
*root
, u64 bytenr
)
9495 struct btrfs_block_group_cache
*block_group
;
9496 struct btrfs_space_info
*space_info
;
9497 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
9498 struct btrfs_device
*device
;
9499 struct btrfs_trans_handle
*trans
;
9509 debug
= btrfs_test_opt(root
, ENOSPC_DEBUG
);
9511 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
9513 /* odd, couldn't find the block group, leave it alone */
9516 btrfs_warn(root
->fs_info
,
9517 "can't find block group for bytenr %llu",
9522 min_free
= btrfs_block_group_used(&block_group
->item
);
9524 /* no bytes used, we're good */
9528 space_info
= block_group
->space_info
;
9529 spin_lock(&space_info
->lock
);
9531 full
= space_info
->full
;
9534 * if this is the last block group we have in this space, we can't
9535 * relocate it unless we're able to allocate a new chunk below.
9537 * Otherwise, we need to make sure we have room in the space to handle
9538 * all of the extents from this block group. If we can, we're good
9540 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
9541 (space_info
->bytes_used
+ space_info
->bytes_reserved
+
9542 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
9543 min_free
< space_info
->total_bytes
)) {
9544 spin_unlock(&space_info
->lock
);
9547 spin_unlock(&space_info
->lock
);
9550 * ok we don't have enough space, but maybe we have free space on our
9551 * devices to allocate new chunks for relocation, so loop through our
9552 * alloc devices and guess if we have enough space. if this block
9553 * group is going to be restriped, run checks against the target
9554 * profile instead of the current one.
9566 target
= get_restripe_target(root
->fs_info
, block_group
->flags
);
9568 index
= __get_raid_index(extended_to_chunk(target
));
9571 * this is just a balance, so if we were marked as full
9572 * we know there is no space for a new chunk
9576 btrfs_warn(root
->fs_info
,
9577 "no space to alloc new chunk for block group %llu",
9578 block_group
->key
.objectid
);
9582 index
= get_block_group_index(block_group
);
9585 if (index
== BTRFS_RAID_RAID10
) {
9589 } else if (index
== BTRFS_RAID_RAID1
) {
9591 } else if (index
== BTRFS_RAID_DUP
) {
9594 } else if (index
== BTRFS_RAID_RAID0
) {
9595 dev_min
= fs_devices
->rw_devices
;
9596 min_free
= div64_u64(min_free
, dev_min
);
9599 /* We need to do this so that we can look at pending chunks */
9600 trans
= btrfs_join_transaction(root
);
9601 if (IS_ERR(trans
)) {
9602 ret
= PTR_ERR(trans
);
9606 mutex_lock(&root
->fs_info
->chunk_mutex
);
9607 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
9611 * check to make sure we can actually find a chunk with enough
9612 * space to fit our block group in.
9614 if (device
->total_bytes
> device
->bytes_used
+ min_free
&&
9615 !device
->is_tgtdev_for_dev_replace
) {
9616 ret
= find_free_dev_extent(trans
, device
, min_free
,
9621 if (dev_nr
>= dev_min
)
9627 if (debug
&& ret
== -1)
9628 btrfs_warn(root
->fs_info
,
9629 "no space to allocate a new chunk for block group %llu",
9630 block_group
->key
.objectid
);
9631 mutex_unlock(&root
->fs_info
->chunk_mutex
);
9632 btrfs_end_transaction(trans
, root
);
9634 btrfs_put_block_group(block_group
);
9638 static int find_first_block_group(struct btrfs_root
*root
,
9639 struct btrfs_path
*path
, struct btrfs_key
*key
)
9642 struct btrfs_key found_key
;
9643 struct extent_buffer
*leaf
;
9646 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
9651 slot
= path
->slots
[0];
9652 leaf
= path
->nodes
[0];
9653 if (slot
>= btrfs_header_nritems(leaf
)) {
9654 ret
= btrfs_next_leaf(root
, path
);
9661 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
9663 if (found_key
.objectid
>= key
->objectid
&&
9664 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
9674 void btrfs_put_block_group_cache(struct btrfs_fs_info
*info
)
9676 struct btrfs_block_group_cache
*block_group
;
9680 struct inode
*inode
;
9682 block_group
= btrfs_lookup_first_block_group(info
, last
);
9683 while (block_group
) {
9684 spin_lock(&block_group
->lock
);
9685 if (block_group
->iref
)
9687 spin_unlock(&block_group
->lock
);
9688 block_group
= next_block_group(info
->tree_root
,
9698 inode
= block_group
->inode
;
9699 block_group
->iref
= 0;
9700 block_group
->inode
= NULL
;
9701 spin_unlock(&block_group
->lock
);
9703 last
= block_group
->key
.objectid
+ block_group
->key
.offset
;
9704 btrfs_put_block_group(block_group
);
9708 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
9710 struct btrfs_block_group_cache
*block_group
;
9711 struct btrfs_space_info
*space_info
;
9712 struct btrfs_caching_control
*caching_ctl
;
9715 down_write(&info
->commit_root_sem
);
9716 while (!list_empty(&info
->caching_block_groups
)) {
9717 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
9718 struct btrfs_caching_control
, list
);
9719 list_del(&caching_ctl
->list
);
9720 put_caching_control(caching_ctl
);
9722 up_write(&info
->commit_root_sem
);
9724 spin_lock(&info
->unused_bgs_lock
);
9725 while (!list_empty(&info
->unused_bgs
)) {
9726 block_group
= list_first_entry(&info
->unused_bgs
,
9727 struct btrfs_block_group_cache
,
9729 list_del_init(&block_group
->bg_list
);
9730 btrfs_put_block_group(block_group
);
9732 spin_unlock(&info
->unused_bgs_lock
);
9734 spin_lock(&info
->block_group_cache_lock
);
9735 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
9736 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
9738 rb_erase(&block_group
->cache_node
,
9739 &info
->block_group_cache_tree
);
9740 RB_CLEAR_NODE(&block_group
->cache_node
);
9741 spin_unlock(&info
->block_group_cache_lock
);
9743 down_write(&block_group
->space_info
->groups_sem
);
9744 list_del(&block_group
->list
);
9745 up_write(&block_group
->space_info
->groups_sem
);
9747 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
9748 wait_block_group_cache_done(block_group
);
9751 * We haven't cached this block group, which means we could
9752 * possibly have excluded extents on this block group.
9754 if (block_group
->cached
== BTRFS_CACHE_NO
||
9755 block_group
->cached
== BTRFS_CACHE_ERROR
)
9756 free_excluded_extents(info
->extent_root
, block_group
);
9758 btrfs_remove_free_space_cache(block_group
);
9759 btrfs_put_block_group(block_group
);
9761 spin_lock(&info
->block_group_cache_lock
);
9763 spin_unlock(&info
->block_group_cache_lock
);
9765 /* now that all the block groups are freed, go through and
9766 * free all the space_info structs. This is only called during
9767 * the final stages of unmount, and so we know nobody is
9768 * using them. We call synchronize_rcu() once before we start,
9769 * just to be on the safe side.
9773 release_global_block_rsv(info
);
9775 while (!list_empty(&info
->space_info
)) {
9778 space_info
= list_entry(info
->space_info
.next
,
9779 struct btrfs_space_info
,
9783 * Do not hide this behind enospc_debug, this is actually
9784 * important and indicates a real bug if this happens.
9786 if (WARN_ON(space_info
->bytes_pinned
> 0 ||
9787 space_info
->bytes_reserved
> 0 ||
9788 space_info
->bytes_may_use
> 0))
9789 dump_space_info(space_info
, 0, 0);
9790 list_del(&space_info
->list
);
9791 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++) {
9792 struct kobject
*kobj
;
9793 kobj
= space_info
->block_group_kobjs
[i
];
9794 space_info
->block_group_kobjs
[i
] = NULL
;
9800 kobject_del(&space_info
->kobj
);
9801 kobject_put(&space_info
->kobj
);
9806 static void __link_block_group(struct btrfs_space_info
*space_info
,
9807 struct btrfs_block_group_cache
*cache
)
9809 int index
= get_block_group_index(cache
);
9812 down_write(&space_info
->groups_sem
);
9813 if (list_empty(&space_info
->block_groups
[index
]))
9815 list_add_tail(&cache
->list
, &space_info
->block_groups
[index
]);
9816 up_write(&space_info
->groups_sem
);
9819 struct raid_kobject
*rkobj
;
9822 rkobj
= kzalloc(sizeof(*rkobj
), GFP_NOFS
);
9825 rkobj
->raid_type
= index
;
9826 kobject_init(&rkobj
->kobj
, &btrfs_raid_ktype
);
9827 ret
= kobject_add(&rkobj
->kobj
, &space_info
->kobj
,
9828 "%s", get_raid_name(index
));
9830 kobject_put(&rkobj
->kobj
);
9833 space_info
->block_group_kobjs
[index
] = &rkobj
->kobj
;
9838 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
9841 static struct btrfs_block_group_cache
*
9842 btrfs_create_block_group_cache(struct btrfs_root
*root
, u64 start
, u64 size
)
9844 struct btrfs_block_group_cache
*cache
;
9846 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
9850 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
9852 if (!cache
->free_space_ctl
) {
9857 cache
->key
.objectid
= start
;
9858 cache
->key
.offset
= size
;
9859 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
9861 cache
->sectorsize
= root
->sectorsize
;
9862 cache
->fs_info
= root
->fs_info
;
9863 cache
->full_stripe_len
= btrfs_full_stripe_len(root
,
9864 &root
->fs_info
->mapping_tree
,
9866 set_free_space_tree_thresholds(cache
);
9868 atomic_set(&cache
->count
, 1);
9869 spin_lock_init(&cache
->lock
);
9870 init_rwsem(&cache
->data_rwsem
);
9871 INIT_LIST_HEAD(&cache
->list
);
9872 INIT_LIST_HEAD(&cache
->cluster_list
);
9873 INIT_LIST_HEAD(&cache
->bg_list
);
9874 INIT_LIST_HEAD(&cache
->ro_list
);
9875 INIT_LIST_HEAD(&cache
->dirty_list
);
9876 INIT_LIST_HEAD(&cache
->io_list
);
9877 btrfs_init_free_space_ctl(cache
);
9878 atomic_set(&cache
->trimming
, 0);
9879 mutex_init(&cache
->free_space_lock
);
9884 int btrfs_read_block_groups(struct btrfs_root
*root
)
9886 struct btrfs_path
*path
;
9888 struct btrfs_block_group_cache
*cache
;
9889 struct btrfs_fs_info
*info
= root
->fs_info
;
9890 struct btrfs_space_info
*space_info
;
9891 struct btrfs_key key
;
9892 struct btrfs_key found_key
;
9893 struct extent_buffer
*leaf
;
9897 root
= info
->extent_root
;
9900 key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
9901 path
= btrfs_alloc_path();
9904 path
->reada
= READA_FORWARD
;
9906 cache_gen
= btrfs_super_cache_generation(root
->fs_info
->super_copy
);
9907 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
9908 btrfs_super_generation(root
->fs_info
->super_copy
) != cache_gen
)
9910 if (btrfs_test_opt(root
, CLEAR_CACHE
))
9914 ret
= find_first_block_group(root
, path
, &key
);
9920 leaf
= path
->nodes
[0];
9921 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
9923 cache
= btrfs_create_block_group_cache(root
, found_key
.objectid
,
9932 * When we mount with old space cache, we need to
9933 * set BTRFS_DC_CLEAR and set dirty flag.
9935 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9936 * truncate the old free space cache inode and
9938 * b) Setting 'dirty flag' makes sure that we flush
9939 * the new space cache info onto disk.
9941 if (btrfs_test_opt(root
, SPACE_CACHE
))
9942 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
9945 read_extent_buffer(leaf
, &cache
->item
,
9946 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
9947 sizeof(cache
->item
));
9948 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
9950 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
9951 btrfs_release_path(path
);
9954 * We need to exclude the super stripes now so that the space
9955 * info has super bytes accounted for, otherwise we'll think
9956 * we have more space than we actually do.
9958 ret
= exclude_super_stripes(root
, cache
);
9961 * We may have excluded something, so call this just in
9964 free_excluded_extents(root
, cache
);
9965 btrfs_put_block_group(cache
);
9970 * check for two cases, either we are full, and therefore
9971 * don't need to bother with the caching work since we won't
9972 * find any space, or we are empty, and we can just add all
9973 * the space in and be done with it. This saves us _alot_ of
9974 * time, particularly in the full case.
9976 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
9977 cache
->last_byte_to_unpin
= (u64
)-1;
9978 cache
->cached
= BTRFS_CACHE_FINISHED
;
9979 free_excluded_extents(root
, cache
);
9980 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
9981 cache
->last_byte_to_unpin
= (u64
)-1;
9982 cache
->cached
= BTRFS_CACHE_FINISHED
;
9983 add_new_free_space(cache
, root
->fs_info
,
9985 found_key
.objectid
+
9987 free_excluded_extents(root
, cache
);
9990 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
9992 btrfs_remove_free_space_cache(cache
);
9993 btrfs_put_block_group(cache
);
9997 trace_btrfs_add_block_group(root
->fs_info
, cache
, 0);
9998 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
9999 btrfs_block_group_used(&cache
->item
),
10000 cache
->bytes_super
, &space_info
);
10002 btrfs_remove_free_space_cache(cache
);
10003 spin_lock(&info
->block_group_cache_lock
);
10004 rb_erase(&cache
->cache_node
,
10005 &info
->block_group_cache_tree
);
10006 RB_CLEAR_NODE(&cache
->cache_node
);
10007 spin_unlock(&info
->block_group_cache_lock
);
10008 btrfs_put_block_group(cache
);
10012 cache
->space_info
= space_info
;
10014 __link_block_group(space_info
, cache
);
10016 set_avail_alloc_bits(root
->fs_info
, cache
->flags
);
10017 if (btrfs_chunk_readonly(root
, cache
->key
.objectid
)) {
10018 inc_block_group_ro(cache
, 1);
10019 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
10020 spin_lock(&info
->unused_bgs_lock
);
10021 /* Should always be true but just in case. */
10022 if (list_empty(&cache
->bg_list
)) {
10023 btrfs_get_block_group(cache
);
10024 list_add_tail(&cache
->bg_list
,
10025 &info
->unused_bgs
);
10027 spin_unlock(&info
->unused_bgs_lock
);
10031 list_for_each_entry_rcu(space_info
, &root
->fs_info
->space_info
, list
) {
10032 if (!(get_alloc_profile(root
, space_info
->flags
) &
10033 (BTRFS_BLOCK_GROUP_RAID10
|
10034 BTRFS_BLOCK_GROUP_RAID1
|
10035 BTRFS_BLOCK_GROUP_RAID5
|
10036 BTRFS_BLOCK_GROUP_RAID6
|
10037 BTRFS_BLOCK_GROUP_DUP
)))
10040 * avoid allocating from un-mirrored block group if there are
10041 * mirrored block groups.
10043 list_for_each_entry(cache
,
10044 &space_info
->block_groups
[BTRFS_RAID_RAID0
],
10046 inc_block_group_ro(cache
, 1);
10047 list_for_each_entry(cache
,
10048 &space_info
->block_groups
[BTRFS_RAID_SINGLE
],
10050 inc_block_group_ro(cache
, 1);
10053 init_global_block_rsv(info
);
10056 btrfs_free_path(path
);
10060 void btrfs_create_pending_block_groups(struct btrfs_trans_handle
*trans
,
10061 struct btrfs_root
*root
)
10063 struct btrfs_block_group_cache
*block_group
, *tmp
;
10064 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
10065 struct btrfs_block_group_item item
;
10066 struct btrfs_key key
;
10068 bool can_flush_pending_bgs
= trans
->can_flush_pending_bgs
;
10070 trans
->can_flush_pending_bgs
= false;
10071 list_for_each_entry_safe(block_group
, tmp
, &trans
->new_bgs
, bg_list
) {
10075 spin_lock(&block_group
->lock
);
10076 memcpy(&item
, &block_group
->item
, sizeof(item
));
10077 memcpy(&key
, &block_group
->key
, sizeof(key
));
10078 spin_unlock(&block_group
->lock
);
10080 ret
= btrfs_insert_item(trans
, extent_root
, &key
, &item
,
10083 btrfs_abort_transaction(trans
, extent_root
, ret
);
10084 ret
= btrfs_finish_chunk_alloc(trans
, extent_root
,
10085 key
.objectid
, key
.offset
);
10087 btrfs_abort_transaction(trans
, extent_root
, ret
);
10088 add_block_group_free_space(trans
, root
->fs_info
, block_group
);
10089 /* already aborted the transaction if it failed. */
10091 list_del_init(&block_group
->bg_list
);
10093 trans
->can_flush_pending_bgs
= can_flush_pending_bgs
;
10096 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
10097 struct btrfs_root
*root
, u64 bytes_used
,
10098 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
10102 struct btrfs_root
*extent_root
;
10103 struct btrfs_block_group_cache
*cache
;
10104 extent_root
= root
->fs_info
->extent_root
;
10106 btrfs_set_log_full_commit(root
->fs_info
, trans
);
10108 cache
= btrfs_create_block_group_cache(root
, chunk_offset
, size
);
10112 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
10113 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
10114 btrfs_set_block_group_flags(&cache
->item
, type
);
10116 cache
->flags
= type
;
10117 cache
->last_byte_to_unpin
= (u64
)-1;
10118 cache
->cached
= BTRFS_CACHE_FINISHED
;
10119 cache
->needs_free_space
= 1;
10120 ret
= exclude_super_stripes(root
, cache
);
10123 * We may have excluded something, so call this just in
10126 free_excluded_extents(root
, cache
);
10127 btrfs_put_block_group(cache
);
10131 add_new_free_space(cache
, root
->fs_info
, chunk_offset
,
10132 chunk_offset
+ size
);
10134 free_excluded_extents(root
, cache
);
10136 #ifdef CONFIG_BTRFS_DEBUG
10137 if (btrfs_should_fragment_free_space(root
, cache
)) {
10138 u64 new_bytes_used
= size
- bytes_used
;
10140 bytes_used
+= new_bytes_used
>> 1;
10141 fragment_free_space(root
, cache
);
10145 * Call to ensure the corresponding space_info object is created and
10146 * assigned to our block group, but don't update its counters just yet.
10147 * We want our bg to be added to the rbtree with its ->space_info set.
10149 ret
= update_space_info(root
->fs_info
, cache
->flags
, 0, 0, 0,
10150 &cache
->space_info
);
10152 btrfs_remove_free_space_cache(cache
);
10153 btrfs_put_block_group(cache
);
10157 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
10159 btrfs_remove_free_space_cache(cache
);
10160 btrfs_put_block_group(cache
);
10165 * Now that our block group has its ->space_info set and is inserted in
10166 * the rbtree, update the space info's counters.
10168 trace_btrfs_add_block_group(root
->fs_info
, cache
, 1);
10169 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
10170 cache
->bytes_super
, &cache
->space_info
);
10172 btrfs_remove_free_space_cache(cache
);
10173 spin_lock(&root
->fs_info
->block_group_cache_lock
);
10174 rb_erase(&cache
->cache_node
,
10175 &root
->fs_info
->block_group_cache_tree
);
10176 RB_CLEAR_NODE(&cache
->cache_node
);
10177 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
10178 btrfs_put_block_group(cache
);
10181 update_global_block_rsv(root
->fs_info
);
10183 __link_block_group(cache
->space_info
, cache
);
10185 list_add_tail(&cache
->bg_list
, &trans
->new_bgs
);
10187 set_avail_alloc_bits(extent_root
->fs_info
, type
);
10191 static void clear_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
10193 u64 extra_flags
= chunk_to_extended(flags
) &
10194 BTRFS_EXTENDED_PROFILE_MASK
;
10196 write_seqlock(&fs_info
->profiles_lock
);
10197 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
10198 fs_info
->avail_data_alloc_bits
&= ~extra_flags
;
10199 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
10200 fs_info
->avail_metadata_alloc_bits
&= ~extra_flags
;
10201 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
10202 fs_info
->avail_system_alloc_bits
&= ~extra_flags
;
10203 write_sequnlock(&fs_info
->profiles_lock
);
10206 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
10207 struct btrfs_root
*root
, u64 group_start
,
10208 struct extent_map
*em
)
10210 struct btrfs_path
*path
;
10211 struct btrfs_block_group_cache
*block_group
;
10212 struct btrfs_free_cluster
*cluster
;
10213 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
10214 struct btrfs_key key
;
10215 struct inode
*inode
;
10216 struct kobject
*kobj
= NULL
;
10220 struct btrfs_caching_control
*caching_ctl
= NULL
;
10223 root
= root
->fs_info
->extent_root
;
10225 block_group
= btrfs_lookup_block_group(root
->fs_info
, group_start
);
10226 BUG_ON(!block_group
);
10227 BUG_ON(!block_group
->ro
);
10230 * Free the reserved super bytes from this block group before
10233 free_excluded_extents(root
, block_group
);
10235 memcpy(&key
, &block_group
->key
, sizeof(key
));
10236 index
= get_block_group_index(block_group
);
10237 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
10238 BTRFS_BLOCK_GROUP_RAID1
|
10239 BTRFS_BLOCK_GROUP_RAID10
))
10244 /* make sure this block group isn't part of an allocation cluster */
10245 cluster
= &root
->fs_info
->data_alloc_cluster
;
10246 spin_lock(&cluster
->refill_lock
);
10247 btrfs_return_cluster_to_free_space(block_group
, cluster
);
10248 spin_unlock(&cluster
->refill_lock
);
10251 * make sure this block group isn't part of a metadata
10252 * allocation cluster
10254 cluster
= &root
->fs_info
->meta_alloc_cluster
;
10255 spin_lock(&cluster
->refill_lock
);
10256 btrfs_return_cluster_to_free_space(block_group
, cluster
);
10257 spin_unlock(&cluster
->refill_lock
);
10259 path
= btrfs_alloc_path();
10266 * get the inode first so any iput calls done for the io_list
10267 * aren't the final iput (no unlinks allowed now)
10269 inode
= lookup_free_space_inode(tree_root
, block_group
, path
);
10271 mutex_lock(&trans
->transaction
->cache_write_mutex
);
10273 * make sure our free spache cache IO is done before remove the
10276 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
10277 if (!list_empty(&block_group
->io_list
)) {
10278 list_del_init(&block_group
->io_list
);
10280 WARN_ON(!IS_ERR(inode
) && inode
!= block_group
->io_ctl
.inode
);
10282 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
10283 btrfs_wait_cache_io(root
, trans
, block_group
,
10284 &block_group
->io_ctl
, path
,
10285 block_group
->key
.objectid
);
10286 btrfs_put_block_group(block_group
);
10287 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
10290 if (!list_empty(&block_group
->dirty_list
)) {
10291 list_del_init(&block_group
->dirty_list
);
10292 btrfs_put_block_group(block_group
);
10294 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
10295 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
10297 if (!IS_ERR(inode
)) {
10298 ret
= btrfs_orphan_add(trans
, inode
);
10300 btrfs_add_delayed_iput(inode
);
10303 clear_nlink(inode
);
10304 /* One for the block groups ref */
10305 spin_lock(&block_group
->lock
);
10306 if (block_group
->iref
) {
10307 block_group
->iref
= 0;
10308 block_group
->inode
= NULL
;
10309 spin_unlock(&block_group
->lock
);
10312 spin_unlock(&block_group
->lock
);
10314 /* One for our lookup ref */
10315 btrfs_add_delayed_iput(inode
);
10318 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
10319 key
.offset
= block_group
->key
.objectid
;
10322 ret
= btrfs_search_slot(trans
, tree_root
, &key
, path
, -1, 1);
10326 btrfs_release_path(path
);
10328 ret
= btrfs_del_item(trans
, tree_root
, path
);
10331 btrfs_release_path(path
);
10334 spin_lock(&root
->fs_info
->block_group_cache_lock
);
10335 rb_erase(&block_group
->cache_node
,
10336 &root
->fs_info
->block_group_cache_tree
);
10337 RB_CLEAR_NODE(&block_group
->cache_node
);
10339 if (root
->fs_info
->first_logical_byte
== block_group
->key
.objectid
)
10340 root
->fs_info
->first_logical_byte
= (u64
)-1;
10341 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
10343 down_write(&block_group
->space_info
->groups_sem
);
10345 * we must use list_del_init so people can check to see if they
10346 * are still on the list after taking the semaphore
10348 list_del_init(&block_group
->list
);
10349 if (list_empty(&block_group
->space_info
->block_groups
[index
])) {
10350 kobj
= block_group
->space_info
->block_group_kobjs
[index
];
10351 block_group
->space_info
->block_group_kobjs
[index
] = NULL
;
10352 clear_avail_alloc_bits(root
->fs_info
, block_group
->flags
);
10354 up_write(&block_group
->space_info
->groups_sem
);
10360 if (block_group
->has_caching_ctl
)
10361 caching_ctl
= get_caching_control(block_group
);
10362 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
10363 wait_block_group_cache_done(block_group
);
10364 if (block_group
->has_caching_ctl
) {
10365 down_write(&root
->fs_info
->commit_root_sem
);
10366 if (!caching_ctl
) {
10367 struct btrfs_caching_control
*ctl
;
10369 list_for_each_entry(ctl
,
10370 &root
->fs_info
->caching_block_groups
, list
)
10371 if (ctl
->block_group
== block_group
) {
10373 atomic_inc(&caching_ctl
->count
);
10378 list_del_init(&caching_ctl
->list
);
10379 up_write(&root
->fs_info
->commit_root_sem
);
10381 /* Once for the caching bgs list and once for us. */
10382 put_caching_control(caching_ctl
);
10383 put_caching_control(caching_ctl
);
10387 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
10388 if (!list_empty(&block_group
->dirty_list
)) {
10391 if (!list_empty(&block_group
->io_list
)) {
10394 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
10395 btrfs_remove_free_space_cache(block_group
);
10397 spin_lock(&block_group
->space_info
->lock
);
10398 list_del_init(&block_group
->ro_list
);
10400 if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
10401 WARN_ON(block_group
->space_info
->total_bytes
10402 < block_group
->key
.offset
);
10403 WARN_ON(block_group
->space_info
->bytes_readonly
10404 < block_group
->key
.offset
);
10405 WARN_ON(block_group
->space_info
->disk_total
10406 < block_group
->key
.offset
* factor
);
10408 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
10409 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
10410 block_group
->space_info
->disk_total
-= block_group
->key
.offset
* factor
;
10412 spin_unlock(&block_group
->space_info
->lock
);
10414 memcpy(&key
, &block_group
->key
, sizeof(key
));
10417 if (!list_empty(&em
->list
)) {
10418 /* We're in the transaction->pending_chunks list. */
10419 free_extent_map(em
);
10421 spin_lock(&block_group
->lock
);
10422 block_group
->removed
= 1;
10424 * At this point trimming can't start on this block group, because we
10425 * removed the block group from the tree fs_info->block_group_cache_tree
10426 * so no one can't find it anymore and even if someone already got this
10427 * block group before we removed it from the rbtree, they have already
10428 * incremented block_group->trimming - if they didn't, they won't find
10429 * any free space entries because we already removed them all when we
10430 * called btrfs_remove_free_space_cache().
10432 * And we must not remove the extent map from the fs_info->mapping_tree
10433 * to prevent the same logical address range and physical device space
10434 * ranges from being reused for a new block group. This is because our
10435 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10436 * completely transactionless, so while it is trimming a range the
10437 * currently running transaction might finish and a new one start,
10438 * allowing for new block groups to be created that can reuse the same
10439 * physical device locations unless we take this special care.
10441 * There may also be an implicit trim operation if the file system
10442 * is mounted with -odiscard. The same protections must remain
10443 * in place until the extents have been discarded completely when
10444 * the transaction commit has completed.
10446 remove_em
= (atomic_read(&block_group
->trimming
) == 0);
10448 * Make sure a trimmer task always sees the em in the pinned_chunks list
10449 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10450 * before checking block_group->removed).
10454 * Our em might be in trans->transaction->pending_chunks which
10455 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10456 * and so is the fs_info->pinned_chunks list.
10458 * So at this point we must be holding the chunk_mutex to avoid
10459 * any races with chunk allocation (more specifically at
10460 * volumes.c:contains_pending_extent()), to ensure it always
10461 * sees the em, either in the pending_chunks list or in the
10462 * pinned_chunks list.
10464 list_move_tail(&em
->list
, &root
->fs_info
->pinned_chunks
);
10466 spin_unlock(&block_group
->lock
);
10469 struct extent_map_tree
*em_tree
;
10471 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
10472 write_lock(&em_tree
->lock
);
10474 * The em might be in the pending_chunks list, so make sure the
10475 * chunk mutex is locked, since remove_extent_mapping() will
10476 * delete us from that list.
10478 remove_extent_mapping(em_tree
, em
);
10479 write_unlock(&em_tree
->lock
);
10480 /* once for the tree */
10481 free_extent_map(em
);
10484 unlock_chunks(root
);
10486 ret
= remove_block_group_free_space(trans
, root
->fs_info
, block_group
);
10490 btrfs_put_block_group(block_group
);
10491 btrfs_put_block_group(block_group
);
10493 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
10499 ret
= btrfs_del_item(trans
, root
, path
);
10501 btrfs_free_path(path
);
10505 struct btrfs_trans_handle
*
10506 btrfs_start_trans_remove_block_group(struct btrfs_fs_info
*fs_info
,
10507 const u64 chunk_offset
)
10509 struct extent_map_tree
*em_tree
= &fs_info
->mapping_tree
.map_tree
;
10510 struct extent_map
*em
;
10511 struct map_lookup
*map
;
10512 unsigned int num_items
;
10514 read_lock(&em_tree
->lock
);
10515 em
= lookup_extent_mapping(em_tree
, chunk_offset
, 1);
10516 read_unlock(&em_tree
->lock
);
10517 ASSERT(em
&& em
->start
== chunk_offset
);
10520 * We need to reserve 3 + N units from the metadata space info in order
10521 * to remove a block group (done at btrfs_remove_chunk() and at
10522 * btrfs_remove_block_group()), which are used for:
10524 * 1 unit for adding the free space inode's orphan (located in the tree
10526 * 1 unit for deleting the block group item (located in the extent
10528 * 1 unit for deleting the free space item (located in tree of tree
10530 * N units for deleting N device extent items corresponding to each
10531 * stripe (located in the device tree).
10533 * In order to remove a block group we also need to reserve units in the
10534 * system space info in order to update the chunk tree (update one or
10535 * more device items and remove one chunk item), but this is done at
10536 * btrfs_remove_chunk() through a call to check_system_chunk().
10538 map
= em
->map_lookup
;
10539 num_items
= 3 + map
->num_stripes
;
10540 free_extent_map(em
);
10542 return btrfs_start_transaction_fallback_global_rsv(fs_info
->extent_root
,
10547 * Process the unused_bgs list and remove any that don't have any allocated
10548 * space inside of them.
10550 void btrfs_delete_unused_bgs(struct btrfs_fs_info
*fs_info
)
10552 struct btrfs_block_group_cache
*block_group
;
10553 struct btrfs_space_info
*space_info
;
10554 struct btrfs_root
*root
= fs_info
->extent_root
;
10555 struct btrfs_trans_handle
*trans
;
10558 if (!fs_info
->open
)
10561 spin_lock(&fs_info
->unused_bgs_lock
);
10562 while (!list_empty(&fs_info
->unused_bgs
)) {
10566 block_group
= list_first_entry(&fs_info
->unused_bgs
,
10567 struct btrfs_block_group_cache
,
10569 list_del_init(&block_group
->bg_list
);
10571 space_info
= block_group
->space_info
;
10573 if (ret
|| btrfs_mixed_space_info(space_info
)) {
10574 btrfs_put_block_group(block_group
);
10577 spin_unlock(&fs_info
->unused_bgs_lock
);
10579 mutex_lock(&fs_info
->delete_unused_bgs_mutex
);
10581 /* Don't want to race with allocators so take the groups_sem */
10582 down_write(&space_info
->groups_sem
);
10583 spin_lock(&block_group
->lock
);
10584 if (block_group
->reserved
||
10585 btrfs_block_group_used(&block_group
->item
) ||
10587 list_is_singular(&block_group
->list
)) {
10589 * We want to bail if we made new allocations or have
10590 * outstanding allocations in this block group. We do
10591 * the ro check in case balance is currently acting on
10592 * this block group.
10594 spin_unlock(&block_group
->lock
);
10595 up_write(&space_info
->groups_sem
);
10598 spin_unlock(&block_group
->lock
);
10600 /* We don't want to force the issue, only flip if it's ok. */
10601 ret
= inc_block_group_ro(block_group
, 0);
10602 up_write(&space_info
->groups_sem
);
10609 * Want to do this before we do anything else so we can recover
10610 * properly if we fail to join the transaction.
10612 trans
= btrfs_start_trans_remove_block_group(fs_info
,
10613 block_group
->key
.objectid
);
10614 if (IS_ERR(trans
)) {
10615 btrfs_dec_block_group_ro(root
, block_group
);
10616 ret
= PTR_ERR(trans
);
10621 * We could have pending pinned extents for this block group,
10622 * just delete them, we don't care about them anymore.
10624 start
= block_group
->key
.objectid
;
10625 end
= start
+ block_group
->key
.offset
- 1;
10627 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10628 * btrfs_finish_extent_commit(). If we are at transaction N,
10629 * another task might be running finish_extent_commit() for the
10630 * previous transaction N - 1, and have seen a range belonging
10631 * to the block group in freed_extents[] before we were able to
10632 * clear the whole block group range from freed_extents[]. This
10633 * means that task can lookup for the block group after we
10634 * unpinned it from freed_extents[] and removed it, leading to
10635 * a BUG_ON() at btrfs_unpin_extent_range().
10637 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
10638 ret
= clear_extent_bits(&fs_info
->freed_extents
[0], start
, end
,
10641 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10642 btrfs_dec_block_group_ro(root
, block_group
);
10645 ret
= clear_extent_bits(&fs_info
->freed_extents
[1], start
, end
,
10648 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10649 btrfs_dec_block_group_ro(root
, block_group
);
10652 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10654 /* Reset pinned so btrfs_put_block_group doesn't complain */
10655 spin_lock(&space_info
->lock
);
10656 spin_lock(&block_group
->lock
);
10658 space_info
->bytes_pinned
-= block_group
->pinned
;
10659 space_info
->bytes_readonly
+= block_group
->pinned
;
10660 percpu_counter_add(&space_info
->total_bytes_pinned
,
10661 -block_group
->pinned
);
10662 block_group
->pinned
= 0;
10664 spin_unlock(&block_group
->lock
);
10665 spin_unlock(&space_info
->lock
);
10667 /* DISCARD can flip during remount */
10668 trimming
= btrfs_test_opt(root
, DISCARD
);
10670 /* Implicit trim during transaction commit. */
10672 btrfs_get_block_group_trimming(block_group
);
10675 * Btrfs_remove_chunk will abort the transaction if things go
10678 ret
= btrfs_remove_chunk(trans
, root
,
10679 block_group
->key
.objectid
);
10683 btrfs_put_block_group_trimming(block_group
);
10688 * If we're not mounted with -odiscard, we can just forget
10689 * about this block group. Otherwise we'll need to wait
10690 * until transaction commit to do the actual discard.
10693 spin_lock(&fs_info
->unused_bgs_lock
);
10695 * A concurrent scrub might have added us to the list
10696 * fs_info->unused_bgs, so use a list_move operation
10697 * to add the block group to the deleted_bgs list.
10699 list_move(&block_group
->bg_list
,
10700 &trans
->transaction
->deleted_bgs
);
10701 spin_unlock(&fs_info
->unused_bgs_lock
);
10702 btrfs_get_block_group(block_group
);
10705 btrfs_end_transaction(trans
, root
);
10707 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
10708 btrfs_put_block_group(block_group
);
10709 spin_lock(&fs_info
->unused_bgs_lock
);
10711 spin_unlock(&fs_info
->unused_bgs_lock
);
10714 int btrfs_init_space_info(struct btrfs_fs_info
*fs_info
)
10716 struct btrfs_space_info
*space_info
;
10717 struct btrfs_super_block
*disk_super
;
10723 disk_super
= fs_info
->super_copy
;
10724 if (!btrfs_super_root(disk_super
))
10727 features
= btrfs_super_incompat_flags(disk_super
);
10728 if (features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
10731 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
10732 ret
= update_space_info(fs_info
, flags
, 0, 0, 0, &space_info
);
10737 flags
= BTRFS_BLOCK_GROUP_METADATA
| BTRFS_BLOCK_GROUP_DATA
;
10738 ret
= update_space_info(fs_info
, flags
, 0, 0, 0, &space_info
);
10740 flags
= BTRFS_BLOCK_GROUP_METADATA
;
10741 ret
= update_space_info(fs_info
, flags
, 0, 0, 0, &space_info
);
10745 flags
= BTRFS_BLOCK_GROUP_DATA
;
10746 ret
= update_space_info(fs_info
, flags
, 0, 0, 0, &space_info
);
10752 int btrfs_error_unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
10754 return unpin_extent_range(root
, start
, end
, false);
10758 * It used to be that old block groups would be left around forever.
10759 * Iterating over them would be enough to trim unused space. Since we
10760 * now automatically remove them, we also need to iterate over unallocated
10763 * We don't want a transaction for this since the discard may take a
10764 * substantial amount of time. We don't require that a transaction be
10765 * running, but we do need to take a running transaction into account
10766 * to ensure that we're not discarding chunks that were released in
10767 * the current transaction.
10769 * Holding the chunks lock will prevent other threads from allocating
10770 * or releasing chunks, but it won't prevent a running transaction
10771 * from committing and releasing the memory that the pending chunks
10772 * list head uses. For that, we need to take a reference to the
10775 static int btrfs_trim_free_extents(struct btrfs_device
*device
,
10776 u64 minlen
, u64
*trimmed
)
10778 u64 start
= 0, len
= 0;
10783 /* Not writeable = nothing to do. */
10784 if (!device
->writeable
)
10787 /* No free space = nothing to do. */
10788 if (device
->total_bytes
<= device
->bytes_used
)
10794 struct btrfs_fs_info
*fs_info
= device
->dev_root
->fs_info
;
10795 struct btrfs_transaction
*trans
;
10798 ret
= mutex_lock_interruptible(&fs_info
->chunk_mutex
);
10802 down_read(&fs_info
->commit_root_sem
);
10804 spin_lock(&fs_info
->trans_lock
);
10805 trans
= fs_info
->running_transaction
;
10807 atomic_inc(&trans
->use_count
);
10808 spin_unlock(&fs_info
->trans_lock
);
10810 ret
= find_free_dev_extent_start(trans
, device
, minlen
, start
,
10813 btrfs_put_transaction(trans
);
10816 up_read(&fs_info
->commit_root_sem
);
10817 mutex_unlock(&fs_info
->chunk_mutex
);
10818 if (ret
== -ENOSPC
)
10823 ret
= btrfs_issue_discard(device
->bdev
, start
, len
, &bytes
);
10824 up_read(&fs_info
->commit_root_sem
);
10825 mutex_unlock(&fs_info
->chunk_mutex
);
10833 if (fatal_signal_pending(current
)) {
10834 ret
= -ERESTARTSYS
;
10844 int btrfs_trim_fs(struct btrfs_root
*root
, struct fstrim_range
*range
)
10846 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
10847 struct btrfs_block_group_cache
*cache
= NULL
;
10848 struct btrfs_device
*device
;
10849 struct list_head
*devices
;
10854 u64 total_bytes
= btrfs_super_total_bytes(fs_info
->super_copy
);
10858 * try to trim all FS space, our block group may start from non-zero.
10860 if (range
->len
== total_bytes
)
10861 cache
= btrfs_lookup_first_block_group(fs_info
, range
->start
);
10863 cache
= btrfs_lookup_block_group(fs_info
, range
->start
);
10866 if (cache
->key
.objectid
>= (range
->start
+ range
->len
)) {
10867 btrfs_put_block_group(cache
);
10871 start
= max(range
->start
, cache
->key
.objectid
);
10872 end
= min(range
->start
+ range
->len
,
10873 cache
->key
.objectid
+ cache
->key
.offset
);
10875 if (end
- start
>= range
->minlen
) {
10876 if (!block_group_cache_done(cache
)) {
10877 ret
= cache_block_group(cache
, 0);
10879 btrfs_put_block_group(cache
);
10882 ret
= wait_block_group_cache_done(cache
);
10884 btrfs_put_block_group(cache
);
10888 ret
= btrfs_trim_block_group(cache
,
10894 trimmed
+= group_trimmed
;
10896 btrfs_put_block_group(cache
);
10901 cache
= next_block_group(fs_info
->tree_root
, cache
);
10904 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
10905 devices
= &root
->fs_info
->fs_devices
->alloc_list
;
10906 list_for_each_entry(device
, devices
, dev_alloc_list
) {
10907 ret
= btrfs_trim_free_extents(device
, range
->minlen
,
10912 trimmed
+= group_trimmed
;
10914 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
10916 range
->len
= trimmed
;
10921 * btrfs_{start,end}_write_no_snapshoting() are similar to
10922 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10923 * data into the page cache through nocow before the subvolume is snapshoted,
10924 * but flush the data into disk after the snapshot creation, or to prevent
10925 * operations while snapshoting is ongoing and that cause the snapshot to be
10926 * inconsistent (writes followed by expanding truncates for example).
10928 void btrfs_end_write_no_snapshoting(struct btrfs_root
*root
)
10930 percpu_counter_dec(&root
->subv_writers
->counter
);
10932 * Make sure counter is updated before we wake up waiters.
10935 if (waitqueue_active(&root
->subv_writers
->wait
))
10936 wake_up(&root
->subv_writers
->wait
);
10939 int btrfs_start_write_no_snapshoting(struct btrfs_root
*root
)
10941 if (atomic_read(&root
->will_be_snapshoted
))
10944 percpu_counter_inc(&root
->subv_writers
->counter
);
10946 * Make sure counter is updated before we check for snapshot creation.
10949 if (atomic_read(&root
->will_be_snapshoted
)) {
10950 btrfs_end_write_no_snapshoting(root
);
10956 static int wait_snapshoting_atomic_t(atomic_t
*a
)
10962 void btrfs_wait_for_snapshot_creation(struct btrfs_root
*root
)
10967 ret
= btrfs_start_write_no_snapshoting(root
);
10970 wait_on_atomic_t(&root
->will_be_snapshoted
,
10971 wait_snapshoting_atomic_t
,
10972 TASK_UNINTERRUPTIBLE
);