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"
40 #undef SCRAMBLE_DELAYED_REFS
43 * control flags for do_chunk_alloc's force field
44 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
45 * if we really need one.
47 * CHUNK_ALLOC_LIMITED means to only try and allocate one
48 * if we have very few chunks already allocated. This is
49 * used as part of the clustering code to help make sure
50 * we have a good pool of storage to cluster in, without
51 * filling the FS with empty chunks
53 * CHUNK_ALLOC_FORCE means it must try to allocate one
57 CHUNK_ALLOC_NO_FORCE
= 0,
58 CHUNK_ALLOC_LIMITED
= 1,
59 CHUNK_ALLOC_FORCE
= 2,
63 * Control how reservations are dealt with.
65 * RESERVE_FREE - freeing a reservation.
66 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
68 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
69 * bytes_may_use as the ENOSPC accounting is done elsewhere
74 RESERVE_ALLOC_NO_ACCOUNT
= 2,
77 static int update_block_group(struct btrfs_root
*root
,
78 u64 bytenr
, u64 num_bytes
, int alloc
);
79 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
80 struct btrfs_root
*root
,
81 u64 bytenr
, u64 num_bytes
, u64 parent
,
82 u64 root_objectid
, u64 owner_objectid
,
83 u64 owner_offset
, int refs_to_drop
,
84 struct btrfs_delayed_extent_op
*extra_op
,
86 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
87 struct extent_buffer
*leaf
,
88 struct btrfs_extent_item
*ei
);
89 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
90 struct btrfs_root
*root
,
91 u64 parent
, u64 root_objectid
,
92 u64 flags
, u64 owner
, u64 offset
,
93 struct btrfs_key
*ins
, int ref_mod
);
94 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
95 struct btrfs_root
*root
,
96 u64 parent
, u64 root_objectid
,
97 u64 flags
, struct btrfs_disk_key
*key
,
98 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 static void btrfs_get_block_group(struct btrfs_block_group_cache
*cache
)
130 atomic_inc(&cache
->count
);
133 void btrfs_put_block_group(struct btrfs_block_group_cache
*cache
)
135 if (atomic_dec_and_test(&cache
->count
)) {
136 WARN_ON(cache
->pinned
> 0);
137 WARN_ON(cache
->reserved
> 0);
138 kfree(cache
->free_space_ctl
);
144 * this adds the block group to the fs_info rb tree for the block group
147 static int btrfs_add_block_group_cache(struct btrfs_fs_info
*info
,
148 struct btrfs_block_group_cache
*block_group
)
151 struct rb_node
*parent
= NULL
;
152 struct btrfs_block_group_cache
*cache
;
154 spin_lock(&info
->block_group_cache_lock
);
155 p
= &info
->block_group_cache_tree
.rb_node
;
159 cache
= rb_entry(parent
, struct btrfs_block_group_cache
,
161 if (block_group
->key
.objectid
< cache
->key
.objectid
) {
163 } else if (block_group
->key
.objectid
> cache
->key
.objectid
) {
166 spin_unlock(&info
->block_group_cache_lock
);
171 rb_link_node(&block_group
->cache_node
, parent
, p
);
172 rb_insert_color(&block_group
->cache_node
,
173 &info
->block_group_cache_tree
);
175 if (info
->first_logical_byte
> block_group
->key
.objectid
)
176 info
->first_logical_byte
= block_group
->key
.objectid
;
178 spin_unlock(&info
->block_group_cache_lock
);
184 * This will return the block group at or after bytenr if contains is 0, else
185 * it will return the block group that contains the bytenr
187 static struct btrfs_block_group_cache
*
188 block_group_cache_tree_search(struct btrfs_fs_info
*info
, u64 bytenr
,
191 struct btrfs_block_group_cache
*cache
, *ret
= NULL
;
195 spin_lock(&info
->block_group_cache_lock
);
196 n
= info
->block_group_cache_tree
.rb_node
;
199 cache
= rb_entry(n
, struct btrfs_block_group_cache
,
201 end
= cache
->key
.objectid
+ cache
->key
.offset
- 1;
202 start
= cache
->key
.objectid
;
204 if (bytenr
< start
) {
205 if (!contains
&& (!ret
|| start
< ret
->key
.objectid
))
208 } else if (bytenr
> start
) {
209 if (contains
&& bytenr
<= end
) {
220 btrfs_get_block_group(ret
);
221 if (bytenr
== 0 && info
->first_logical_byte
> ret
->key
.objectid
)
222 info
->first_logical_byte
= ret
->key
.objectid
;
224 spin_unlock(&info
->block_group_cache_lock
);
229 static int add_excluded_extent(struct btrfs_root
*root
,
230 u64 start
, u64 num_bytes
)
232 u64 end
= start
+ num_bytes
- 1;
233 set_extent_bits(&root
->fs_info
->freed_extents
[0],
234 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
235 set_extent_bits(&root
->fs_info
->freed_extents
[1],
236 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
240 static void free_excluded_extents(struct btrfs_root
*root
,
241 struct btrfs_block_group_cache
*cache
)
245 start
= cache
->key
.objectid
;
246 end
= start
+ cache
->key
.offset
- 1;
248 clear_extent_bits(&root
->fs_info
->freed_extents
[0],
249 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
250 clear_extent_bits(&root
->fs_info
->freed_extents
[1],
251 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
254 static int exclude_super_stripes(struct btrfs_root
*root
,
255 struct btrfs_block_group_cache
*cache
)
262 if (cache
->key
.objectid
< BTRFS_SUPER_INFO_OFFSET
) {
263 stripe_len
= BTRFS_SUPER_INFO_OFFSET
- cache
->key
.objectid
;
264 cache
->bytes_super
+= stripe_len
;
265 ret
= add_excluded_extent(root
, cache
->key
.objectid
,
271 for (i
= 0; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
272 bytenr
= btrfs_sb_offset(i
);
273 ret
= btrfs_rmap_block(&root
->fs_info
->mapping_tree
,
274 cache
->key
.objectid
, bytenr
,
275 0, &logical
, &nr
, &stripe_len
);
282 if (logical
[nr
] > cache
->key
.objectid
+
286 if (logical
[nr
] + stripe_len
<= cache
->key
.objectid
)
290 if (start
< cache
->key
.objectid
) {
291 start
= cache
->key
.objectid
;
292 len
= (logical
[nr
] + stripe_len
) - start
;
294 len
= min_t(u64
, stripe_len
,
295 cache
->key
.objectid
+
296 cache
->key
.offset
- start
);
299 cache
->bytes_super
+= len
;
300 ret
= add_excluded_extent(root
, start
, len
);
312 static struct btrfs_caching_control
*
313 get_caching_control(struct btrfs_block_group_cache
*cache
)
315 struct btrfs_caching_control
*ctl
;
317 spin_lock(&cache
->lock
);
318 if (cache
->cached
!= BTRFS_CACHE_STARTED
) {
319 spin_unlock(&cache
->lock
);
323 /* We're loading it the fast way, so we don't have a caching_ctl. */
324 if (!cache
->caching_ctl
) {
325 spin_unlock(&cache
->lock
);
329 ctl
= cache
->caching_ctl
;
330 atomic_inc(&ctl
->count
);
331 spin_unlock(&cache
->lock
);
335 static void put_caching_control(struct btrfs_caching_control
*ctl
)
337 if (atomic_dec_and_test(&ctl
->count
))
342 * this is only called by cache_block_group, since we could have freed extents
343 * we need to check the pinned_extents for any extents that can't be used yet
344 * since their free space will be released as soon as the transaction commits.
346 static u64
add_new_free_space(struct btrfs_block_group_cache
*block_group
,
347 struct btrfs_fs_info
*info
, u64 start
, u64 end
)
349 u64 extent_start
, extent_end
, size
, total_added
= 0;
352 while (start
< end
) {
353 ret
= find_first_extent_bit(info
->pinned_extents
, start
,
354 &extent_start
, &extent_end
,
355 EXTENT_DIRTY
| EXTENT_UPTODATE
,
360 if (extent_start
<= start
) {
361 start
= extent_end
+ 1;
362 } else if (extent_start
> start
&& extent_start
< end
) {
363 size
= extent_start
- start
;
365 ret
= btrfs_add_free_space(block_group
, start
,
367 BUG_ON(ret
); /* -ENOMEM or logic error */
368 start
= extent_end
+ 1;
377 ret
= btrfs_add_free_space(block_group
, start
, size
);
378 BUG_ON(ret
); /* -ENOMEM or logic error */
384 static noinline
void caching_thread(struct btrfs_work
*work
)
386 struct btrfs_block_group_cache
*block_group
;
387 struct btrfs_fs_info
*fs_info
;
388 struct btrfs_caching_control
*caching_ctl
;
389 struct btrfs_root
*extent_root
;
390 struct btrfs_path
*path
;
391 struct extent_buffer
*leaf
;
392 struct btrfs_key key
;
398 caching_ctl
= container_of(work
, struct btrfs_caching_control
, work
);
399 block_group
= caching_ctl
->block_group
;
400 fs_info
= block_group
->fs_info
;
401 extent_root
= fs_info
->extent_root
;
403 path
= btrfs_alloc_path();
407 last
= max_t(u64
, block_group
->key
.objectid
, BTRFS_SUPER_INFO_OFFSET
);
410 * We don't want to deadlock with somebody trying to allocate a new
411 * extent for the extent root while also trying to search the extent
412 * root to add free space. So we skip locking and search the commit
413 * root, since its read-only
415 path
->skip_locking
= 1;
416 path
->search_commit_root
= 1;
421 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
423 mutex_lock(&caching_ctl
->mutex
);
424 /* need to make sure the commit_root doesn't disappear */
425 down_read(&fs_info
->commit_root_sem
);
428 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
432 leaf
= path
->nodes
[0];
433 nritems
= btrfs_header_nritems(leaf
);
436 if (btrfs_fs_closing(fs_info
) > 1) {
441 if (path
->slots
[0] < nritems
) {
442 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
444 ret
= find_next_key(path
, 0, &key
);
448 if (need_resched() ||
449 rwsem_is_contended(&fs_info
->commit_root_sem
)) {
450 caching_ctl
->progress
= last
;
451 btrfs_release_path(path
);
452 up_read(&fs_info
->commit_root_sem
);
453 mutex_unlock(&caching_ctl
->mutex
);
458 ret
= btrfs_next_leaf(extent_root
, path
);
463 leaf
= path
->nodes
[0];
464 nritems
= btrfs_header_nritems(leaf
);
468 if (key
.objectid
< last
) {
471 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
473 caching_ctl
->progress
= last
;
474 btrfs_release_path(path
);
478 if (key
.objectid
< block_group
->key
.objectid
) {
483 if (key
.objectid
>= block_group
->key
.objectid
+
484 block_group
->key
.offset
)
487 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
||
488 key
.type
== BTRFS_METADATA_ITEM_KEY
) {
489 total_found
+= add_new_free_space(block_group
,
492 if (key
.type
== BTRFS_METADATA_ITEM_KEY
)
493 last
= key
.objectid
+
494 fs_info
->tree_root
->leafsize
;
496 last
= key
.objectid
+ key
.offset
;
498 if (total_found
> (1024 * 1024 * 2)) {
500 wake_up(&caching_ctl
->wait
);
507 total_found
+= add_new_free_space(block_group
, fs_info
, last
,
508 block_group
->key
.objectid
+
509 block_group
->key
.offset
);
510 caching_ctl
->progress
= (u64
)-1;
512 spin_lock(&block_group
->lock
);
513 block_group
->caching_ctl
= NULL
;
514 block_group
->cached
= BTRFS_CACHE_FINISHED
;
515 spin_unlock(&block_group
->lock
);
518 btrfs_free_path(path
);
519 up_read(&fs_info
->commit_root_sem
);
521 free_excluded_extents(extent_root
, block_group
);
523 mutex_unlock(&caching_ctl
->mutex
);
526 spin_lock(&block_group
->lock
);
527 block_group
->caching_ctl
= NULL
;
528 block_group
->cached
= BTRFS_CACHE_ERROR
;
529 spin_unlock(&block_group
->lock
);
531 wake_up(&caching_ctl
->wait
);
533 put_caching_control(caching_ctl
);
534 btrfs_put_block_group(block_group
);
537 static int cache_block_group(struct btrfs_block_group_cache
*cache
,
541 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
542 struct btrfs_caching_control
*caching_ctl
;
545 caching_ctl
= kzalloc(sizeof(*caching_ctl
), GFP_NOFS
);
549 INIT_LIST_HEAD(&caching_ctl
->list
);
550 mutex_init(&caching_ctl
->mutex
);
551 init_waitqueue_head(&caching_ctl
->wait
);
552 caching_ctl
->block_group
= cache
;
553 caching_ctl
->progress
= cache
->key
.objectid
;
554 atomic_set(&caching_ctl
->count
, 1);
555 btrfs_init_work(&caching_ctl
->work
, btrfs_cache_helper
,
556 caching_thread
, NULL
, NULL
);
558 spin_lock(&cache
->lock
);
560 * This should be a rare occasion, but this could happen I think in the
561 * case where one thread starts to load the space cache info, and then
562 * some other thread starts a transaction commit which tries to do an
563 * allocation while the other thread is still loading the space cache
564 * info. The previous loop should have kept us from choosing this block
565 * group, but if we've moved to the state where we will wait on caching
566 * block groups we need to first check if we're doing a fast load here,
567 * so we can wait for it to finish, otherwise we could end up allocating
568 * from a block group who's cache gets evicted for one reason or
571 while (cache
->cached
== BTRFS_CACHE_FAST
) {
572 struct btrfs_caching_control
*ctl
;
574 ctl
= cache
->caching_ctl
;
575 atomic_inc(&ctl
->count
);
576 prepare_to_wait(&ctl
->wait
, &wait
, TASK_UNINTERRUPTIBLE
);
577 spin_unlock(&cache
->lock
);
581 finish_wait(&ctl
->wait
, &wait
);
582 put_caching_control(ctl
);
583 spin_lock(&cache
->lock
);
586 if (cache
->cached
!= BTRFS_CACHE_NO
) {
587 spin_unlock(&cache
->lock
);
591 WARN_ON(cache
->caching_ctl
);
592 cache
->caching_ctl
= caching_ctl
;
593 cache
->cached
= BTRFS_CACHE_FAST
;
594 spin_unlock(&cache
->lock
);
596 if (fs_info
->mount_opt
& BTRFS_MOUNT_SPACE_CACHE
) {
597 ret
= load_free_space_cache(fs_info
, cache
);
599 spin_lock(&cache
->lock
);
601 cache
->caching_ctl
= NULL
;
602 cache
->cached
= BTRFS_CACHE_FINISHED
;
603 cache
->last_byte_to_unpin
= (u64
)-1;
605 if (load_cache_only
) {
606 cache
->caching_ctl
= NULL
;
607 cache
->cached
= BTRFS_CACHE_NO
;
609 cache
->cached
= BTRFS_CACHE_STARTED
;
612 spin_unlock(&cache
->lock
);
613 wake_up(&caching_ctl
->wait
);
615 put_caching_control(caching_ctl
);
616 free_excluded_extents(fs_info
->extent_root
, cache
);
621 * We are not going to do the fast caching, set cached to the
622 * appropriate value and wakeup any waiters.
624 spin_lock(&cache
->lock
);
625 if (load_cache_only
) {
626 cache
->caching_ctl
= NULL
;
627 cache
->cached
= BTRFS_CACHE_NO
;
629 cache
->cached
= BTRFS_CACHE_STARTED
;
631 spin_unlock(&cache
->lock
);
632 wake_up(&caching_ctl
->wait
);
635 if (load_cache_only
) {
636 put_caching_control(caching_ctl
);
640 down_write(&fs_info
->commit_root_sem
);
641 atomic_inc(&caching_ctl
->count
);
642 list_add_tail(&caching_ctl
->list
, &fs_info
->caching_block_groups
);
643 up_write(&fs_info
->commit_root_sem
);
645 btrfs_get_block_group(cache
);
647 btrfs_queue_work(fs_info
->caching_workers
, &caching_ctl
->work
);
653 * return the block group that starts at or after bytenr
655 static struct btrfs_block_group_cache
*
656 btrfs_lookup_first_block_group(struct btrfs_fs_info
*info
, u64 bytenr
)
658 struct btrfs_block_group_cache
*cache
;
660 cache
= block_group_cache_tree_search(info
, bytenr
, 0);
666 * return the block group that contains the given bytenr
668 struct btrfs_block_group_cache
*btrfs_lookup_block_group(
669 struct btrfs_fs_info
*info
,
672 struct btrfs_block_group_cache
*cache
;
674 cache
= block_group_cache_tree_search(info
, bytenr
, 1);
679 static struct btrfs_space_info
*__find_space_info(struct btrfs_fs_info
*info
,
682 struct list_head
*head
= &info
->space_info
;
683 struct btrfs_space_info
*found
;
685 flags
&= BTRFS_BLOCK_GROUP_TYPE_MASK
;
688 list_for_each_entry_rcu(found
, head
, list
) {
689 if (found
->flags
& flags
) {
699 * after adding space to the filesystem, we need to clear the full flags
700 * on all the space infos.
702 void btrfs_clear_space_info_full(struct btrfs_fs_info
*info
)
704 struct list_head
*head
= &info
->space_info
;
705 struct btrfs_space_info
*found
;
708 list_for_each_entry_rcu(found
, head
, list
)
713 /* simple helper to search for an existing extent at a given offset */
714 int btrfs_lookup_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
717 struct btrfs_key key
;
718 struct btrfs_path
*path
;
720 path
= btrfs_alloc_path();
724 key
.objectid
= start
;
726 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
727 ret
= btrfs_search_slot(NULL
, root
->fs_info
->extent_root
, &key
, path
,
730 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
731 if (key
.objectid
== start
&&
732 key
.type
== BTRFS_METADATA_ITEM_KEY
)
735 btrfs_free_path(path
);
740 * helper function to lookup reference count and flags of a tree block.
742 * the head node for delayed ref is used to store the sum of all the
743 * reference count modifications queued up in the rbtree. the head
744 * node may also store the extent flags to set. This way you can check
745 * to see what the reference count and extent flags would be if all of
746 * the delayed refs are not processed.
748 int btrfs_lookup_extent_info(struct btrfs_trans_handle
*trans
,
749 struct btrfs_root
*root
, u64 bytenr
,
750 u64 offset
, int metadata
, u64
*refs
, u64
*flags
)
752 struct btrfs_delayed_ref_head
*head
;
753 struct btrfs_delayed_ref_root
*delayed_refs
;
754 struct btrfs_path
*path
;
755 struct btrfs_extent_item
*ei
;
756 struct extent_buffer
*leaf
;
757 struct btrfs_key key
;
764 * If we don't have skinny metadata, don't bother doing anything
767 if (metadata
&& !btrfs_fs_incompat(root
->fs_info
, SKINNY_METADATA
)) {
768 offset
= root
->leafsize
;
772 path
= btrfs_alloc_path();
777 path
->skip_locking
= 1;
778 path
->search_commit_root
= 1;
782 key
.objectid
= bytenr
;
785 key
.type
= BTRFS_METADATA_ITEM_KEY
;
787 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
790 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
,
795 if (ret
> 0 && metadata
&& key
.type
== BTRFS_METADATA_ITEM_KEY
) {
796 if (path
->slots
[0]) {
798 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
800 if (key
.objectid
== bytenr
&&
801 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
802 key
.offset
== root
->leafsize
)
806 key
.objectid
= bytenr
;
807 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
808 key
.offset
= root
->leafsize
;
809 btrfs_release_path(path
);
815 leaf
= path
->nodes
[0];
816 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
817 if (item_size
>= sizeof(*ei
)) {
818 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
819 struct btrfs_extent_item
);
820 num_refs
= btrfs_extent_refs(leaf
, ei
);
821 extent_flags
= btrfs_extent_flags(leaf
, ei
);
823 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
824 struct btrfs_extent_item_v0
*ei0
;
825 BUG_ON(item_size
!= sizeof(*ei0
));
826 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
827 struct btrfs_extent_item_v0
);
828 num_refs
= btrfs_extent_refs_v0(leaf
, ei0
);
829 /* FIXME: this isn't correct for data */
830 extent_flags
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
835 BUG_ON(num_refs
== 0);
845 delayed_refs
= &trans
->transaction
->delayed_refs
;
846 spin_lock(&delayed_refs
->lock
);
847 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
849 if (!mutex_trylock(&head
->mutex
)) {
850 atomic_inc(&head
->node
.refs
);
851 spin_unlock(&delayed_refs
->lock
);
853 btrfs_release_path(path
);
856 * Mutex was contended, block until it's released and try
859 mutex_lock(&head
->mutex
);
860 mutex_unlock(&head
->mutex
);
861 btrfs_put_delayed_ref(&head
->node
);
864 spin_lock(&head
->lock
);
865 if (head
->extent_op
&& head
->extent_op
->update_flags
)
866 extent_flags
|= head
->extent_op
->flags_to_set
;
868 BUG_ON(num_refs
== 0);
870 num_refs
+= head
->node
.ref_mod
;
871 spin_unlock(&head
->lock
);
872 mutex_unlock(&head
->mutex
);
874 spin_unlock(&delayed_refs
->lock
);
876 WARN_ON(num_refs
== 0);
880 *flags
= extent_flags
;
882 btrfs_free_path(path
);
887 * Back reference rules. Back refs have three main goals:
889 * 1) differentiate between all holders of references to an extent so that
890 * when a reference is dropped we can make sure it was a valid reference
891 * before freeing the extent.
893 * 2) Provide enough information to quickly find the holders of an extent
894 * if we notice a given block is corrupted or bad.
896 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
897 * maintenance. This is actually the same as #2, but with a slightly
898 * different use case.
900 * There are two kinds of back refs. The implicit back refs is optimized
901 * for pointers in non-shared tree blocks. For a given pointer in a block,
902 * back refs of this kind provide information about the block's owner tree
903 * and the pointer's key. These information allow us to find the block by
904 * b-tree searching. The full back refs is for pointers in tree blocks not
905 * referenced by their owner trees. The location of tree block is recorded
906 * in the back refs. Actually the full back refs is generic, and can be
907 * used in all cases the implicit back refs is used. The major shortcoming
908 * of the full back refs is its overhead. Every time a tree block gets
909 * COWed, we have to update back refs entry for all pointers in it.
911 * For a newly allocated tree block, we use implicit back refs for
912 * pointers in it. This means most tree related operations only involve
913 * implicit back refs. For a tree block created in old transaction, the
914 * only way to drop a reference to it is COW it. So we can detect the
915 * event that tree block loses its owner tree's reference and do the
916 * back refs conversion.
918 * When a tree block is COW'd through a tree, there are four cases:
920 * The reference count of the block is one and the tree is the block's
921 * owner tree. Nothing to do in this case.
923 * The reference count of the block is one and the tree is not the
924 * block's owner tree. In this case, full back refs is used for pointers
925 * in the block. Remove these full back refs, add implicit back refs for
926 * every pointers in the new block.
928 * The reference count of the block is greater than one and the tree is
929 * the block's owner tree. In this case, implicit back refs is used for
930 * pointers in the block. Add full back refs for every pointers in the
931 * block, increase lower level extents' reference counts. The original
932 * implicit back refs are entailed to the new block.
934 * The reference count of the block is greater than one and the tree is
935 * not the block's owner tree. Add implicit back refs for every pointer in
936 * the new block, increase lower level extents' reference count.
938 * Back Reference Key composing:
940 * The key objectid corresponds to the first byte in the extent,
941 * The key type is used to differentiate between types of back refs.
942 * There are different meanings of the key offset for different types
945 * File extents can be referenced by:
947 * - multiple snapshots, subvolumes, or different generations in one subvol
948 * - different files inside a single subvolume
949 * - different offsets inside a file (bookend extents in file.c)
951 * The extent ref structure for the implicit back refs has fields for:
953 * - Objectid of the subvolume root
954 * - objectid of the file holding the reference
955 * - original offset in the file
956 * - how many bookend extents
958 * The key offset for the implicit back refs is hash of the first
961 * The extent ref structure for the full back refs has field for:
963 * - number of pointers in the tree leaf
965 * The key offset for the implicit back refs is the first byte of
968 * When a file extent is allocated, The implicit back refs is used.
969 * the fields are filled in:
971 * (root_key.objectid, inode objectid, offset in file, 1)
973 * When a file extent is removed file truncation, we find the
974 * corresponding implicit back refs and check the following fields:
976 * (btrfs_header_owner(leaf), inode objectid, offset in file)
978 * Btree extents can be referenced by:
980 * - Different subvolumes
982 * Both the implicit back refs and the full back refs for tree blocks
983 * only consist of key. The key offset for the implicit back refs is
984 * objectid of block's owner tree. The key offset for the full back refs
985 * is the first byte of parent block.
987 * When implicit back refs is used, information about the lowest key and
988 * level of the tree block are required. These information are stored in
989 * tree block info structure.
992 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
993 static int convert_extent_item_v0(struct btrfs_trans_handle
*trans
,
994 struct btrfs_root
*root
,
995 struct btrfs_path
*path
,
996 u64 owner
, u32 extra_size
)
998 struct btrfs_extent_item
*item
;
999 struct btrfs_extent_item_v0
*ei0
;
1000 struct btrfs_extent_ref_v0
*ref0
;
1001 struct btrfs_tree_block_info
*bi
;
1002 struct extent_buffer
*leaf
;
1003 struct btrfs_key key
;
1004 struct btrfs_key found_key
;
1005 u32 new_size
= sizeof(*item
);
1009 leaf
= path
->nodes
[0];
1010 BUG_ON(btrfs_item_size_nr(leaf
, path
->slots
[0]) != sizeof(*ei0
));
1012 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1013 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1014 struct btrfs_extent_item_v0
);
1015 refs
= btrfs_extent_refs_v0(leaf
, ei0
);
1017 if (owner
== (u64
)-1) {
1019 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1020 ret
= btrfs_next_leaf(root
, path
);
1023 BUG_ON(ret
> 0); /* Corruption */
1024 leaf
= path
->nodes
[0];
1026 btrfs_item_key_to_cpu(leaf
, &found_key
,
1028 BUG_ON(key
.objectid
!= found_key
.objectid
);
1029 if (found_key
.type
!= BTRFS_EXTENT_REF_V0_KEY
) {
1033 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1034 struct btrfs_extent_ref_v0
);
1035 owner
= btrfs_ref_objectid_v0(leaf
, ref0
);
1039 btrfs_release_path(path
);
1041 if (owner
< BTRFS_FIRST_FREE_OBJECTID
)
1042 new_size
+= sizeof(*bi
);
1044 new_size
-= sizeof(*ei0
);
1045 ret
= btrfs_search_slot(trans
, root
, &key
, path
,
1046 new_size
+ extra_size
, 1);
1049 BUG_ON(ret
); /* Corruption */
1051 btrfs_extend_item(root
, path
, new_size
);
1053 leaf
= path
->nodes
[0];
1054 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1055 btrfs_set_extent_refs(leaf
, item
, refs
);
1056 /* FIXME: get real generation */
1057 btrfs_set_extent_generation(leaf
, item
, 0);
1058 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1059 btrfs_set_extent_flags(leaf
, item
,
1060 BTRFS_EXTENT_FLAG_TREE_BLOCK
|
1061 BTRFS_BLOCK_FLAG_FULL_BACKREF
);
1062 bi
= (struct btrfs_tree_block_info
*)(item
+ 1);
1063 /* FIXME: get first key of the block */
1064 memset_extent_buffer(leaf
, 0, (unsigned long)bi
, sizeof(*bi
));
1065 btrfs_set_tree_block_level(leaf
, bi
, (int)owner
);
1067 btrfs_set_extent_flags(leaf
, item
, BTRFS_EXTENT_FLAG_DATA
);
1069 btrfs_mark_buffer_dirty(leaf
);
1074 static u64
hash_extent_data_ref(u64 root_objectid
, u64 owner
, u64 offset
)
1076 u32 high_crc
= ~(u32
)0;
1077 u32 low_crc
= ~(u32
)0;
1080 lenum
= cpu_to_le64(root_objectid
);
1081 high_crc
= btrfs_crc32c(high_crc
, &lenum
, sizeof(lenum
));
1082 lenum
= cpu_to_le64(owner
);
1083 low_crc
= btrfs_crc32c(low_crc
, &lenum
, sizeof(lenum
));
1084 lenum
= cpu_to_le64(offset
);
1085 low_crc
= btrfs_crc32c(low_crc
, &lenum
, sizeof(lenum
));
1087 return ((u64
)high_crc
<< 31) ^ (u64
)low_crc
;
1090 static u64
hash_extent_data_ref_item(struct extent_buffer
*leaf
,
1091 struct btrfs_extent_data_ref
*ref
)
1093 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf
, ref
),
1094 btrfs_extent_data_ref_objectid(leaf
, ref
),
1095 btrfs_extent_data_ref_offset(leaf
, ref
));
1098 static int match_extent_data_ref(struct extent_buffer
*leaf
,
1099 struct btrfs_extent_data_ref
*ref
,
1100 u64 root_objectid
, u64 owner
, u64 offset
)
1102 if (btrfs_extent_data_ref_root(leaf
, ref
) != root_objectid
||
1103 btrfs_extent_data_ref_objectid(leaf
, ref
) != owner
||
1104 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
1109 static noinline
int lookup_extent_data_ref(struct btrfs_trans_handle
*trans
,
1110 struct btrfs_root
*root
,
1111 struct btrfs_path
*path
,
1112 u64 bytenr
, u64 parent
,
1114 u64 owner
, u64 offset
)
1116 struct btrfs_key key
;
1117 struct btrfs_extent_data_ref
*ref
;
1118 struct extent_buffer
*leaf
;
1124 key
.objectid
= bytenr
;
1126 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1127 key
.offset
= parent
;
1129 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1130 key
.offset
= hash_extent_data_ref(root_objectid
,
1135 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1144 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1145 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1146 btrfs_release_path(path
);
1147 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1158 leaf
= path
->nodes
[0];
1159 nritems
= btrfs_header_nritems(leaf
);
1161 if (path
->slots
[0] >= nritems
) {
1162 ret
= btrfs_next_leaf(root
, path
);
1168 leaf
= path
->nodes
[0];
1169 nritems
= btrfs_header_nritems(leaf
);
1173 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1174 if (key
.objectid
!= bytenr
||
1175 key
.type
!= BTRFS_EXTENT_DATA_REF_KEY
)
1178 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1179 struct btrfs_extent_data_ref
);
1181 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1184 btrfs_release_path(path
);
1196 static noinline
int insert_extent_data_ref(struct btrfs_trans_handle
*trans
,
1197 struct btrfs_root
*root
,
1198 struct btrfs_path
*path
,
1199 u64 bytenr
, u64 parent
,
1200 u64 root_objectid
, u64 owner
,
1201 u64 offset
, int refs_to_add
)
1203 struct btrfs_key key
;
1204 struct extent_buffer
*leaf
;
1209 key
.objectid
= bytenr
;
1211 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1212 key
.offset
= parent
;
1213 size
= sizeof(struct btrfs_shared_data_ref
);
1215 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1216 key
.offset
= hash_extent_data_ref(root_objectid
,
1218 size
= sizeof(struct btrfs_extent_data_ref
);
1221 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, size
);
1222 if (ret
&& ret
!= -EEXIST
)
1225 leaf
= path
->nodes
[0];
1227 struct btrfs_shared_data_ref
*ref
;
1228 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1229 struct btrfs_shared_data_ref
);
1231 btrfs_set_shared_data_ref_count(leaf
, ref
, refs_to_add
);
1233 num_refs
= btrfs_shared_data_ref_count(leaf
, ref
);
1234 num_refs
+= refs_to_add
;
1235 btrfs_set_shared_data_ref_count(leaf
, ref
, num_refs
);
1238 struct btrfs_extent_data_ref
*ref
;
1239 while (ret
== -EEXIST
) {
1240 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1241 struct btrfs_extent_data_ref
);
1242 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1245 btrfs_release_path(path
);
1247 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1249 if (ret
&& ret
!= -EEXIST
)
1252 leaf
= path
->nodes
[0];
1254 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1255 struct btrfs_extent_data_ref
);
1257 btrfs_set_extent_data_ref_root(leaf
, ref
,
1259 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
1260 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
1261 btrfs_set_extent_data_ref_count(leaf
, ref
, refs_to_add
);
1263 num_refs
= btrfs_extent_data_ref_count(leaf
, ref
);
1264 num_refs
+= refs_to_add
;
1265 btrfs_set_extent_data_ref_count(leaf
, ref
, num_refs
);
1268 btrfs_mark_buffer_dirty(leaf
);
1271 btrfs_release_path(path
);
1275 static noinline
int remove_extent_data_ref(struct btrfs_trans_handle
*trans
,
1276 struct btrfs_root
*root
,
1277 struct btrfs_path
*path
,
1278 int refs_to_drop
, int *last_ref
)
1280 struct btrfs_key key
;
1281 struct btrfs_extent_data_ref
*ref1
= NULL
;
1282 struct btrfs_shared_data_ref
*ref2
= NULL
;
1283 struct extent_buffer
*leaf
;
1287 leaf
= path
->nodes
[0];
1288 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1290 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1291 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1292 struct btrfs_extent_data_ref
);
1293 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1294 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1295 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1296 struct btrfs_shared_data_ref
);
1297 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1298 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1299 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1300 struct btrfs_extent_ref_v0
*ref0
;
1301 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1302 struct btrfs_extent_ref_v0
);
1303 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1309 BUG_ON(num_refs
< refs_to_drop
);
1310 num_refs
-= refs_to_drop
;
1312 if (num_refs
== 0) {
1313 ret
= btrfs_del_item(trans
, root
, path
);
1316 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
)
1317 btrfs_set_extent_data_ref_count(leaf
, ref1
, num_refs
);
1318 else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
)
1319 btrfs_set_shared_data_ref_count(leaf
, ref2
, num_refs
);
1320 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1322 struct btrfs_extent_ref_v0
*ref0
;
1323 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1324 struct btrfs_extent_ref_v0
);
1325 btrfs_set_ref_count_v0(leaf
, ref0
, num_refs
);
1328 btrfs_mark_buffer_dirty(leaf
);
1333 static noinline u32
extent_data_ref_count(struct btrfs_root
*root
,
1334 struct btrfs_path
*path
,
1335 struct btrfs_extent_inline_ref
*iref
)
1337 struct btrfs_key key
;
1338 struct extent_buffer
*leaf
;
1339 struct btrfs_extent_data_ref
*ref1
;
1340 struct btrfs_shared_data_ref
*ref2
;
1343 leaf
= path
->nodes
[0];
1344 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1346 if (btrfs_extent_inline_ref_type(leaf
, iref
) ==
1347 BTRFS_EXTENT_DATA_REF_KEY
) {
1348 ref1
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1349 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1351 ref2
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1352 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1354 } else if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1355 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1356 struct btrfs_extent_data_ref
);
1357 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1358 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1359 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1360 struct btrfs_shared_data_ref
);
1361 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1362 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1363 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1364 struct btrfs_extent_ref_v0
*ref0
;
1365 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1366 struct btrfs_extent_ref_v0
);
1367 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1375 static noinline
int lookup_tree_block_ref(struct btrfs_trans_handle
*trans
,
1376 struct btrfs_root
*root
,
1377 struct btrfs_path
*path
,
1378 u64 bytenr
, u64 parent
,
1381 struct btrfs_key key
;
1384 key
.objectid
= bytenr
;
1386 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1387 key
.offset
= parent
;
1389 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1390 key
.offset
= root_objectid
;
1393 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1396 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1397 if (ret
== -ENOENT
&& parent
) {
1398 btrfs_release_path(path
);
1399 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1400 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1408 static noinline
int insert_tree_block_ref(struct btrfs_trans_handle
*trans
,
1409 struct btrfs_root
*root
,
1410 struct btrfs_path
*path
,
1411 u64 bytenr
, u64 parent
,
1414 struct btrfs_key key
;
1417 key
.objectid
= bytenr
;
1419 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1420 key
.offset
= parent
;
1422 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1423 key
.offset
= root_objectid
;
1426 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
1427 btrfs_release_path(path
);
1431 static inline int extent_ref_type(u64 parent
, u64 owner
)
1434 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1436 type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1438 type
= BTRFS_TREE_BLOCK_REF_KEY
;
1441 type
= BTRFS_SHARED_DATA_REF_KEY
;
1443 type
= BTRFS_EXTENT_DATA_REF_KEY
;
1448 static int find_next_key(struct btrfs_path
*path
, int level
,
1449 struct btrfs_key
*key
)
1452 for (; level
< BTRFS_MAX_LEVEL
; level
++) {
1453 if (!path
->nodes
[level
])
1455 if (path
->slots
[level
] + 1 >=
1456 btrfs_header_nritems(path
->nodes
[level
]))
1459 btrfs_item_key_to_cpu(path
->nodes
[level
], key
,
1460 path
->slots
[level
] + 1);
1462 btrfs_node_key_to_cpu(path
->nodes
[level
], key
,
1463 path
->slots
[level
] + 1);
1470 * look for inline back ref. if back ref is found, *ref_ret is set
1471 * to the address of inline back ref, and 0 is returned.
1473 * if back ref isn't found, *ref_ret is set to the address where it
1474 * should be inserted, and -ENOENT is returned.
1476 * if insert is true and there are too many inline back refs, the path
1477 * points to the extent item, and -EAGAIN is returned.
1479 * NOTE: inline back refs are ordered in the same way that back ref
1480 * items in the tree are ordered.
1482 static noinline_for_stack
1483 int lookup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1484 struct btrfs_root
*root
,
1485 struct btrfs_path
*path
,
1486 struct btrfs_extent_inline_ref
**ref_ret
,
1487 u64 bytenr
, u64 num_bytes
,
1488 u64 parent
, u64 root_objectid
,
1489 u64 owner
, u64 offset
, int insert
)
1491 struct btrfs_key key
;
1492 struct extent_buffer
*leaf
;
1493 struct btrfs_extent_item
*ei
;
1494 struct btrfs_extent_inline_ref
*iref
;
1504 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
1507 key
.objectid
= bytenr
;
1508 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1509 key
.offset
= num_bytes
;
1511 want
= extent_ref_type(parent
, owner
);
1513 extra_size
= btrfs_extent_inline_ref_size(want
);
1514 path
->keep_locks
= 1;
1519 * Owner is our parent level, so we can just add one to get the level
1520 * for the block we are interested in.
1522 if (skinny_metadata
&& owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1523 key
.type
= BTRFS_METADATA_ITEM_KEY
;
1528 ret
= btrfs_search_slot(trans
, root
, &key
, path
, extra_size
, 1);
1535 * We may be a newly converted file system which still has the old fat
1536 * extent entries for metadata, so try and see if we have one of those.
1538 if (ret
> 0 && skinny_metadata
) {
1539 skinny_metadata
= false;
1540 if (path
->slots
[0]) {
1542 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
1544 if (key
.objectid
== bytenr
&&
1545 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
1546 key
.offset
== num_bytes
)
1550 key
.objectid
= bytenr
;
1551 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1552 key
.offset
= num_bytes
;
1553 btrfs_release_path(path
);
1558 if (ret
&& !insert
) {
1561 } else if (WARN_ON(ret
)) {
1566 leaf
= path
->nodes
[0];
1567 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1568 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1569 if (item_size
< sizeof(*ei
)) {
1574 ret
= convert_extent_item_v0(trans
, root
, path
, owner
,
1580 leaf
= path
->nodes
[0];
1581 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1584 BUG_ON(item_size
< sizeof(*ei
));
1586 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1587 flags
= btrfs_extent_flags(leaf
, ei
);
1589 ptr
= (unsigned long)(ei
+ 1);
1590 end
= (unsigned long)ei
+ item_size
;
1592 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
&& !skinny_metadata
) {
1593 ptr
+= sizeof(struct btrfs_tree_block_info
);
1603 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1604 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1608 ptr
+= btrfs_extent_inline_ref_size(type
);
1612 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1613 struct btrfs_extent_data_ref
*dref
;
1614 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1615 if (match_extent_data_ref(leaf
, dref
, root_objectid
,
1620 if (hash_extent_data_ref_item(leaf
, dref
) <
1621 hash_extent_data_ref(root_objectid
, owner
, offset
))
1625 ref_offset
= btrfs_extent_inline_ref_offset(leaf
, iref
);
1627 if (parent
== ref_offset
) {
1631 if (ref_offset
< parent
)
1634 if (root_objectid
== ref_offset
) {
1638 if (ref_offset
< root_objectid
)
1642 ptr
+= btrfs_extent_inline_ref_size(type
);
1644 if (err
== -ENOENT
&& insert
) {
1645 if (item_size
+ extra_size
>=
1646 BTRFS_MAX_EXTENT_ITEM_SIZE(root
)) {
1651 * To add new inline back ref, we have to make sure
1652 * there is no corresponding back ref item.
1653 * For simplicity, we just do not add new inline back
1654 * ref if there is any kind of item for this block
1656 if (find_next_key(path
, 0, &key
) == 0 &&
1657 key
.objectid
== bytenr
&&
1658 key
.type
< BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1663 *ref_ret
= (struct btrfs_extent_inline_ref
*)ptr
;
1666 path
->keep_locks
= 0;
1667 btrfs_unlock_up_safe(path
, 1);
1673 * helper to add new inline back ref
1675 static noinline_for_stack
1676 void setup_inline_extent_backref(struct btrfs_root
*root
,
1677 struct btrfs_path
*path
,
1678 struct btrfs_extent_inline_ref
*iref
,
1679 u64 parent
, u64 root_objectid
,
1680 u64 owner
, u64 offset
, int refs_to_add
,
1681 struct btrfs_delayed_extent_op
*extent_op
)
1683 struct extent_buffer
*leaf
;
1684 struct btrfs_extent_item
*ei
;
1687 unsigned long item_offset
;
1692 leaf
= path
->nodes
[0];
1693 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1694 item_offset
= (unsigned long)iref
- (unsigned long)ei
;
1696 type
= extent_ref_type(parent
, owner
);
1697 size
= btrfs_extent_inline_ref_size(type
);
1699 btrfs_extend_item(root
, path
, size
);
1701 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1702 refs
= btrfs_extent_refs(leaf
, ei
);
1703 refs
+= refs_to_add
;
1704 btrfs_set_extent_refs(leaf
, ei
, refs
);
1706 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1708 ptr
= (unsigned long)ei
+ item_offset
;
1709 end
= (unsigned long)ei
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
1710 if (ptr
< end
- size
)
1711 memmove_extent_buffer(leaf
, ptr
+ size
, ptr
,
1714 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1715 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
1716 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1717 struct btrfs_extent_data_ref
*dref
;
1718 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1719 btrfs_set_extent_data_ref_root(leaf
, dref
, root_objectid
);
1720 btrfs_set_extent_data_ref_objectid(leaf
, dref
, owner
);
1721 btrfs_set_extent_data_ref_offset(leaf
, dref
, offset
);
1722 btrfs_set_extent_data_ref_count(leaf
, dref
, refs_to_add
);
1723 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1724 struct btrfs_shared_data_ref
*sref
;
1725 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1726 btrfs_set_shared_data_ref_count(leaf
, sref
, refs_to_add
);
1727 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1728 } else if (type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
1729 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1731 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
1733 btrfs_mark_buffer_dirty(leaf
);
1736 static int lookup_extent_backref(struct btrfs_trans_handle
*trans
,
1737 struct btrfs_root
*root
,
1738 struct btrfs_path
*path
,
1739 struct btrfs_extent_inline_ref
**ref_ret
,
1740 u64 bytenr
, u64 num_bytes
, u64 parent
,
1741 u64 root_objectid
, u64 owner
, u64 offset
)
1745 ret
= lookup_inline_extent_backref(trans
, root
, path
, ref_ret
,
1746 bytenr
, num_bytes
, parent
,
1747 root_objectid
, owner
, offset
, 0);
1751 btrfs_release_path(path
);
1754 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1755 ret
= lookup_tree_block_ref(trans
, root
, path
, bytenr
, parent
,
1758 ret
= lookup_extent_data_ref(trans
, root
, path
, bytenr
, parent
,
1759 root_objectid
, owner
, offset
);
1765 * helper to update/remove inline back ref
1767 static noinline_for_stack
1768 void update_inline_extent_backref(struct btrfs_root
*root
,
1769 struct btrfs_path
*path
,
1770 struct btrfs_extent_inline_ref
*iref
,
1772 struct btrfs_delayed_extent_op
*extent_op
,
1775 struct extent_buffer
*leaf
;
1776 struct btrfs_extent_item
*ei
;
1777 struct btrfs_extent_data_ref
*dref
= NULL
;
1778 struct btrfs_shared_data_ref
*sref
= NULL
;
1786 leaf
= path
->nodes
[0];
1787 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1788 refs
= btrfs_extent_refs(leaf
, ei
);
1789 WARN_ON(refs_to_mod
< 0 && refs
+ refs_to_mod
<= 0);
1790 refs
+= refs_to_mod
;
1791 btrfs_set_extent_refs(leaf
, ei
, refs
);
1793 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1795 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1797 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1798 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1799 refs
= btrfs_extent_data_ref_count(leaf
, dref
);
1800 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1801 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1802 refs
= btrfs_shared_data_ref_count(leaf
, sref
);
1805 BUG_ON(refs_to_mod
!= -1);
1808 BUG_ON(refs_to_mod
< 0 && refs
< -refs_to_mod
);
1809 refs
+= refs_to_mod
;
1812 if (type
== BTRFS_EXTENT_DATA_REF_KEY
)
1813 btrfs_set_extent_data_ref_count(leaf
, dref
, refs
);
1815 btrfs_set_shared_data_ref_count(leaf
, sref
, refs
);
1818 size
= btrfs_extent_inline_ref_size(type
);
1819 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1820 ptr
= (unsigned long)iref
;
1821 end
= (unsigned long)ei
+ item_size
;
1822 if (ptr
+ size
< end
)
1823 memmove_extent_buffer(leaf
, ptr
, ptr
+ size
,
1826 btrfs_truncate_item(root
, path
, item_size
, 1);
1828 btrfs_mark_buffer_dirty(leaf
);
1831 static noinline_for_stack
1832 int insert_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1833 struct btrfs_root
*root
,
1834 struct btrfs_path
*path
,
1835 u64 bytenr
, u64 num_bytes
, u64 parent
,
1836 u64 root_objectid
, u64 owner
,
1837 u64 offset
, int refs_to_add
,
1838 struct btrfs_delayed_extent_op
*extent_op
)
1840 struct btrfs_extent_inline_ref
*iref
;
1843 ret
= lookup_inline_extent_backref(trans
, root
, path
, &iref
,
1844 bytenr
, num_bytes
, parent
,
1845 root_objectid
, owner
, offset
, 1);
1847 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
);
1848 update_inline_extent_backref(root
, path
, iref
,
1849 refs_to_add
, extent_op
, NULL
);
1850 } else if (ret
== -ENOENT
) {
1851 setup_inline_extent_backref(root
, path
, iref
, parent
,
1852 root_objectid
, owner
, offset
,
1853 refs_to_add
, extent_op
);
1859 static int insert_extent_backref(struct btrfs_trans_handle
*trans
,
1860 struct btrfs_root
*root
,
1861 struct btrfs_path
*path
,
1862 u64 bytenr
, u64 parent
, u64 root_objectid
,
1863 u64 owner
, u64 offset
, int refs_to_add
)
1866 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1867 BUG_ON(refs_to_add
!= 1);
1868 ret
= insert_tree_block_ref(trans
, root
, path
, bytenr
,
1869 parent
, root_objectid
);
1871 ret
= insert_extent_data_ref(trans
, root
, path
, bytenr
,
1872 parent
, root_objectid
,
1873 owner
, offset
, refs_to_add
);
1878 static int remove_extent_backref(struct btrfs_trans_handle
*trans
,
1879 struct btrfs_root
*root
,
1880 struct btrfs_path
*path
,
1881 struct btrfs_extent_inline_ref
*iref
,
1882 int refs_to_drop
, int is_data
, int *last_ref
)
1886 BUG_ON(!is_data
&& refs_to_drop
!= 1);
1888 update_inline_extent_backref(root
, path
, iref
,
1889 -refs_to_drop
, NULL
, last_ref
);
1890 } else if (is_data
) {
1891 ret
= remove_extent_data_ref(trans
, root
, path
, refs_to_drop
,
1895 ret
= btrfs_del_item(trans
, root
, path
);
1900 static int btrfs_issue_discard(struct block_device
*bdev
,
1903 return blkdev_issue_discard(bdev
, start
>> 9, len
>> 9, GFP_NOFS
, 0);
1906 static int btrfs_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
1907 u64 num_bytes
, u64
*actual_bytes
)
1910 u64 discarded_bytes
= 0;
1911 struct btrfs_bio
*bbio
= NULL
;
1914 /* Tell the block device(s) that the sectors can be discarded */
1915 ret
= btrfs_map_block(root
->fs_info
, REQ_DISCARD
,
1916 bytenr
, &num_bytes
, &bbio
, 0);
1917 /* Error condition is -ENOMEM */
1919 struct btrfs_bio_stripe
*stripe
= bbio
->stripes
;
1923 for (i
= 0; i
< bbio
->num_stripes
; i
++, stripe
++) {
1924 if (!stripe
->dev
->can_discard
)
1927 ret
= btrfs_issue_discard(stripe
->dev
->bdev
,
1931 discarded_bytes
+= stripe
->length
;
1932 else if (ret
!= -EOPNOTSUPP
)
1933 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1936 * Just in case we get back EOPNOTSUPP for some reason,
1937 * just ignore the return value so we don't screw up
1938 * people calling discard_extent.
1946 *actual_bytes
= discarded_bytes
;
1949 if (ret
== -EOPNOTSUPP
)
1954 /* Can return -ENOMEM */
1955 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1956 struct btrfs_root
*root
,
1957 u64 bytenr
, u64 num_bytes
, u64 parent
,
1958 u64 root_objectid
, u64 owner
, u64 offset
,
1962 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1964 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
&&
1965 root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
1967 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1968 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
1970 parent
, root_objectid
, (int)owner
,
1971 BTRFS_ADD_DELAYED_REF
, NULL
, no_quota
);
1973 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
1975 parent
, root_objectid
, owner
, offset
,
1976 BTRFS_ADD_DELAYED_REF
, NULL
, no_quota
);
1981 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1982 struct btrfs_root
*root
,
1983 u64 bytenr
, u64 num_bytes
,
1984 u64 parent
, u64 root_objectid
,
1985 u64 owner
, u64 offset
, int refs_to_add
,
1987 struct btrfs_delayed_extent_op
*extent_op
)
1989 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1990 struct btrfs_path
*path
;
1991 struct extent_buffer
*leaf
;
1992 struct btrfs_extent_item
*item
;
1993 struct btrfs_key key
;
1996 enum btrfs_qgroup_operation_type type
= BTRFS_QGROUP_OPER_ADD_EXCL
;
1998 path
= btrfs_alloc_path();
2002 if (!is_fstree(root_objectid
) || !root
->fs_info
->quota_enabled
)
2006 path
->leave_spinning
= 1;
2007 /* this will setup the path even if it fails to insert the back ref */
2008 ret
= insert_inline_extent_backref(trans
, fs_info
->extent_root
, path
,
2009 bytenr
, num_bytes
, parent
,
2010 root_objectid
, owner
, offset
,
2011 refs_to_add
, extent_op
);
2012 if ((ret
< 0 && ret
!= -EAGAIN
) || (!ret
&& no_quota
))
2015 * Ok we were able to insert an inline extent and it appears to be a new
2016 * reference, deal with the qgroup accounting.
2018 if (!ret
&& !no_quota
) {
2019 ASSERT(root
->fs_info
->quota_enabled
);
2020 leaf
= path
->nodes
[0];
2021 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2022 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2023 struct btrfs_extent_item
);
2024 if (btrfs_extent_refs(leaf
, item
) > (u64
)refs_to_add
)
2025 type
= BTRFS_QGROUP_OPER_ADD_SHARED
;
2026 btrfs_release_path(path
);
2028 ret
= btrfs_qgroup_record_ref(trans
, fs_info
, root_objectid
,
2029 bytenr
, num_bytes
, type
, 0);
2034 * Ok we had -EAGAIN which means we didn't have space to insert and
2035 * inline extent ref, so just update the reference count and add a
2038 leaf
= path
->nodes
[0];
2039 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2040 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2041 refs
= btrfs_extent_refs(leaf
, item
);
2043 type
= BTRFS_QGROUP_OPER_ADD_SHARED
;
2044 btrfs_set_extent_refs(leaf
, item
, refs
+ refs_to_add
);
2046 __run_delayed_extent_op(extent_op
, leaf
, item
);
2048 btrfs_mark_buffer_dirty(leaf
);
2049 btrfs_release_path(path
);
2052 ret
= btrfs_qgroup_record_ref(trans
, fs_info
, root_objectid
,
2053 bytenr
, num_bytes
, type
, 0);
2059 path
->leave_spinning
= 1;
2060 /* now insert the actual backref */
2061 ret
= insert_extent_backref(trans
, root
->fs_info
->extent_root
,
2062 path
, bytenr
, parent
, root_objectid
,
2063 owner
, offset
, refs_to_add
);
2065 btrfs_abort_transaction(trans
, root
, ret
);
2067 btrfs_free_path(path
);
2071 static int run_delayed_data_ref(struct btrfs_trans_handle
*trans
,
2072 struct btrfs_root
*root
,
2073 struct btrfs_delayed_ref_node
*node
,
2074 struct btrfs_delayed_extent_op
*extent_op
,
2075 int insert_reserved
)
2078 struct btrfs_delayed_data_ref
*ref
;
2079 struct btrfs_key ins
;
2084 ins
.objectid
= node
->bytenr
;
2085 ins
.offset
= node
->num_bytes
;
2086 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2088 ref
= btrfs_delayed_node_to_data_ref(node
);
2089 trace_run_delayed_data_ref(node
, ref
, node
->action
);
2091 if (node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2092 parent
= ref
->parent
;
2093 ref_root
= ref
->root
;
2095 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2097 flags
|= extent_op
->flags_to_set
;
2098 ret
= alloc_reserved_file_extent(trans
, root
,
2099 parent
, ref_root
, flags
,
2100 ref
->objectid
, ref
->offset
,
2101 &ins
, node
->ref_mod
);
2102 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2103 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
2104 node
->num_bytes
, parent
,
2105 ref_root
, ref
->objectid
,
2106 ref
->offset
, node
->ref_mod
,
2107 node
->no_quota
, extent_op
);
2108 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2109 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
2110 node
->num_bytes
, parent
,
2111 ref_root
, ref
->objectid
,
2112 ref
->offset
, node
->ref_mod
,
2113 extent_op
, node
->no_quota
);
2120 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
2121 struct extent_buffer
*leaf
,
2122 struct btrfs_extent_item
*ei
)
2124 u64 flags
= btrfs_extent_flags(leaf
, ei
);
2125 if (extent_op
->update_flags
) {
2126 flags
|= extent_op
->flags_to_set
;
2127 btrfs_set_extent_flags(leaf
, ei
, flags
);
2130 if (extent_op
->update_key
) {
2131 struct btrfs_tree_block_info
*bi
;
2132 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
));
2133 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
2134 btrfs_set_tree_block_key(leaf
, bi
, &extent_op
->key
);
2138 static int run_delayed_extent_op(struct btrfs_trans_handle
*trans
,
2139 struct btrfs_root
*root
,
2140 struct btrfs_delayed_ref_node
*node
,
2141 struct btrfs_delayed_extent_op
*extent_op
)
2143 struct btrfs_key key
;
2144 struct btrfs_path
*path
;
2145 struct btrfs_extent_item
*ei
;
2146 struct extent_buffer
*leaf
;
2150 int metadata
= !extent_op
->is_data
;
2155 if (metadata
&& !btrfs_fs_incompat(root
->fs_info
, SKINNY_METADATA
))
2158 path
= btrfs_alloc_path();
2162 key
.objectid
= node
->bytenr
;
2165 key
.type
= BTRFS_METADATA_ITEM_KEY
;
2166 key
.offset
= extent_op
->level
;
2168 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2169 key
.offset
= node
->num_bytes
;
2174 path
->leave_spinning
= 1;
2175 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
,
2183 if (path
->slots
[0] > 0) {
2185 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
2187 if (key
.objectid
== node
->bytenr
&&
2188 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
2189 key
.offset
== node
->num_bytes
)
2193 btrfs_release_path(path
);
2196 key
.objectid
= node
->bytenr
;
2197 key
.offset
= node
->num_bytes
;
2198 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2207 leaf
= path
->nodes
[0];
2208 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2209 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2210 if (item_size
< sizeof(*ei
)) {
2211 ret
= convert_extent_item_v0(trans
, root
->fs_info
->extent_root
,
2217 leaf
= path
->nodes
[0];
2218 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2221 BUG_ON(item_size
< sizeof(*ei
));
2222 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2223 __run_delayed_extent_op(extent_op
, leaf
, ei
);
2225 btrfs_mark_buffer_dirty(leaf
);
2227 btrfs_free_path(path
);
2231 static int run_delayed_tree_ref(struct btrfs_trans_handle
*trans
,
2232 struct btrfs_root
*root
,
2233 struct btrfs_delayed_ref_node
*node
,
2234 struct btrfs_delayed_extent_op
*extent_op
,
2235 int insert_reserved
)
2238 struct btrfs_delayed_tree_ref
*ref
;
2239 struct btrfs_key ins
;
2242 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
2245 ref
= btrfs_delayed_node_to_tree_ref(node
);
2246 trace_run_delayed_tree_ref(node
, ref
, node
->action
);
2248 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2249 parent
= ref
->parent
;
2250 ref_root
= ref
->root
;
2252 ins
.objectid
= node
->bytenr
;
2253 if (skinny_metadata
) {
2254 ins
.offset
= ref
->level
;
2255 ins
.type
= BTRFS_METADATA_ITEM_KEY
;
2257 ins
.offset
= node
->num_bytes
;
2258 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2261 BUG_ON(node
->ref_mod
!= 1);
2262 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2263 BUG_ON(!extent_op
|| !extent_op
->update_flags
);
2264 ret
= alloc_reserved_tree_block(trans
, root
,
2266 extent_op
->flags_to_set
,
2270 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2271 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
2272 node
->num_bytes
, parent
, ref_root
,
2273 ref
->level
, 0, 1, node
->no_quota
,
2275 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2276 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
2277 node
->num_bytes
, parent
, ref_root
,
2278 ref
->level
, 0, 1, extent_op
,
2286 /* helper function to actually process a single delayed ref entry */
2287 static int run_one_delayed_ref(struct btrfs_trans_handle
*trans
,
2288 struct btrfs_root
*root
,
2289 struct btrfs_delayed_ref_node
*node
,
2290 struct btrfs_delayed_extent_op
*extent_op
,
2291 int insert_reserved
)
2295 if (trans
->aborted
) {
2296 if (insert_reserved
)
2297 btrfs_pin_extent(root
, node
->bytenr
,
2298 node
->num_bytes
, 1);
2302 if (btrfs_delayed_ref_is_head(node
)) {
2303 struct btrfs_delayed_ref_head
*head
;
2305 * we've hit the end of the chain and we were supposed
2306 * to insert this extent into the tree. But, it got
2307 * deleted before we ever needed to insert it, so all
2308 * we have to do is clean up the accounting
2311 head
= btrfs_delayed_node_to_head(node
);
2312 trace_run_delayed_ref_head(node
, head
, node
->action
);
2314 if (insert_reserved
) {
2315 btrfs_pin_extent(root
, node
->bytenr
,
2316 node
->num_bytes
, 1);
2317 if (head
->is_data
) {
2318 ret
= btrfs_del_csums(trans
, root
,
2326 if (node
->type
== BTRFS_TREE_BLOCK_REF_KEY
||
2327 node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2328 ret
= run_delayed_tree_ref(trans
, root
, node
, extent_op
,
2330 else if (node
->type
== BTRFS_EXTENT_DATA_REF_KEY
||
2331 node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2332 ret
= run_delayed_data_ref(trans
, root
, node
, extent_op
,
2339 static noinline
struct btrfs_delayed_ref_node
*
2340 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
2342 struct rb_node
*node
;
2343 struct btrfs_delayed_ref_node
*ref
, *last
= NULL
;;
2346 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2347 * this prevents ref count from going down to zero when
2348 * there still are pending delayed ref.
2350 node
= rb_first(&head
->ref_root
);
2352 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
2354 if (ref
->action
== BTRFS_ADD_DELAYED_REF
)
2356 else if (last
== NULL
)
2358 node
= rb_next(node
);
2364 * Returns 0 on success or if called with an already aborted transaction.
2365 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2367 static noinline
int __btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2368 struct btrfs_root
*root
,
2371 struct btrfs_delayed_ref_root
*delayed_refs
;
2372 struct btrfs_delayed_ref_node
*ref
;
2373 struct btrfs_delayed_ref_head
*locked_ref
= NULL
;
2374 struct btrfs_delayed_extent_op
*extent_op
;
2375 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2376 ktime_t start
= ktime_get();
2378 unsigned long count
= 0;
2379 unsigned long actual_count
= 0;
2380 int must_insert_reserved
= 0;
2382 delayed_refs
= &trans
->transaction
->delayed_refs
;
2388 spin_lock(&delayed_refs
->lock
);
2389 locked_ref
= btrfs_select_ref_head(trans
);
2391 spin_unlock(&delayed_refs
->lock
);
2395 /* grab the lock that says we are going to process
2396 * all the refs for this head */
2397 ret
= btrfs_delayed_ref_lock(trans
, locked_ref
);
2398 spin_unlock(&delayed_refs
->lock
);
2400 * we may have dropped the spin lock to get the head
2401 * mutex lock, and that might have given someone else
2402 * time to free the head. If that's true, it has been
2403 * removed from our list and we can move on.
2405 if (ret
== -EAGAIN
) {
2413 * We need to try and merge add/drops of the same ref since we
2414 * can run into issues with relocate dropping the implicit ref
2415 * and then it being added back again before the drop can
2416 * finish. If we merged anything we need to re-loop so we can
2419 spin_lock(&locked_ref
->lock
);
2420 btrfs_merge_delayed_refs(trans
, fs_info
, delayed_refs
,
2424 * locked_ref is the head node, so we have to go one
2425 * node back for any delayed ref updates
2427 ref
= select_delayed_ref(locked_ref
);
2429 if (ref
&& ref
->seq
&&
2430 btrfs_check_delayed_seq(fs_info
, delayed_refs
, ref
->seq
)) {
2431 spin_unlock(&locked_ref
->lock
);
2432 btrfs_delayed_ref_unlock(locked_ref
);
2433 spin_lock(&delayed_refs
->lock
);
2434 locked_ref
->processing
= 0;
2435 delayed_refs
->num_heads_ready
++;
2436 spin_unlock(&delayed_refs
->lock
);
2444 * record the must insert reserved flag before we
2445 * drop the spin lock.
2447 must_insert_reserved
= locked_ref
->must_insert_reserved
;
2448 locked_ref
->must_insert_reserved
= 0;
2450 extent_op
= locked_ref
->extent_op
;
2451 locked_ref
->extent_op
= NULL
;
2456 /* All delayed refs have been processed, Go ahead
2457 * and send the head node to run_one_delayed_ref,
2458 * so that any accounting fixes can happen
2460 ref
= &locked_ref
->node
;
2462 if (extent_op
&& must_insert_reserved
) {
2463 btrfs_free_delayed_extent_op(extent_op
);
2468 spin_unlock(&locked_ref
->lock
);
2469 ret
= run_delayed_extent_op(trans
, root
,
2471 btrfs_free_delayed_extent_op(extent_op
);
2475 * Need to reset must_insert_reserved if
2476 * there was an error so the abort stuff
2477 * can cleanup the reserved space
2480 if (must_insert_reserved
)
2481 locked_ref
->must_insert_reserved
= 1;
2482 locked_ref
->processing
= 0;
2483 btrfs_debug(fs_info
, "run_delayed_extent_op returned %d", ret
);
2484 btrfs_delayed_ref_unlock(locked_ref
);
2491 * Need to drop our head ref lock and re-aqcuire the
2492 * delayed ref lock and then re-check to make sure
2495 spin_unlock(&locked_ref
->lock
);
2496 spin_lock(&delayed_refs
->lock
);
2497 spin_lock(&locked_ref
->lock
);
2498 if (rb_first(&locked_ref
->ref_root
) ||
2499 locked_ref
->extent_op
) {
2500 spin_unlock(&locked_ref
->lock
);
2501 spin_unlock(&delayed_refs
->lock
);
2505 delayed_refs
->num_heads
--;
2506 rb_erase(&locked_ref
->href_node
,
2507 &delayed_refs
->href_root
);
2508 spin_unlock(&delayed_refs
->lock
);
2512 rb_erase(&ref
->rb_node
, &locked_ref
->ref_root
);
2514 atomic_dec(&delayed_refs
->num_entries
);
2516 if (!btrfs_delayed_ref_is_head(ref
)) {
2518 * when we play the delayed ref, also correct the
2521 switch (ref
->action
) {
2522 case BTRFS_ADD_DELAYED_REF
:
2523 case BTRFS_ADD_DELAYED_EXTENT
:
2524 locked_ref
->node
.ref_mod
-= ref
->ref_mod
;
2526 case BTRFS_DROP_DELAYED_REF
:
2527 locked_ref
->node
.ref_mod
+= ref
->ref_mod
;
2533 spin_unlock(&locked_ref
->lock
);
2535 ret
= run_one_delayed_ref(trans
, root
, ref
, extent_op
,
2536 must_insert_reserved
);
2538 btrfs_free_delayed_extent_op(extent_op
);
2540 locked_ref
->processing
= 0;
2541 btrfs_delayed_ref_unlock(locked_ref
);
2542 btrfs_put_delayed_ref(ref
);
2543 btrfs_debug(fs_info
, "run_one_delayed_ref returned %d", ret
);
2548 * If this node is a head, that means all the refs in this head
2549 * have been dealt with, and we will pick the next head to deal
2550 * with, so we must unlock the head and drop it from the cluster
2551 * list before we release it.
2553 if (btrfs_delayed_ref_is_head(ref
)) {
2554 btrfs_delayed_ref_unlock(locked_ref
);
2557 btrfs_put_delayed_ref(ref
);
2563 * We don't want to include ref heads since we can have empty ref heads
2564 * and those will drastically skew our runtime down since we just do
2565 * accounting, no actual extent tree updates.
2567 if (actual_count
> 0) {
2568 u64 runtime
= ktime_to_ns(ktime_sub(ktime_get(), start
));
2572 * We weigh the current average higher than our current runtime
2573 * to avoid large swings in the average.
2575 spin_lock(&delayed_refs
->lock
);
2576 avg
= fs_info
->avg_delayed_ref_runtime
* 3 + runtime
;
2577 avg
= div64_u64(avg
, 4);
2578 fs_info
->avg_delayed_ref_runtime
= avg
;
2579 spin_unlock(&delayed_refs
->lock
);
2584 #ifdef SCRAMBLE_DELAYED_REFS
2586 * Normally delayed refs get processed in ascending bytenr order. This
2587 * correlates in most cases to the order added. To expose dependencies on this
2588 * order, we start to process the tree in the middle instead of the beginning
2590 static u64
find_middle(struct rb_root
*root
)
2592 struct rb_node
*n
= root
->rb_node
;
2593 struct btrfs_delayed_ref_node
*entry
;
2596 u64 first
= 0, last
= 0;
2600 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2601 first
= entry
->bytenr
;
2605 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2606 last
= entry
->bytenr
;
2611 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2612 WARN_ON(!entry
->in_tree
);
2614 middle
= entry
->bytenr
;
2627 static inline u64
heads_to_leaves(struct btrfs_root
*root
, u64 heads
)
2631 num_bytes
= heads
* (sizeof(struct btrfs_extent_item
) +
2632 sizeof(struct btrfs_extent_inline_ref
));
2633 if (!btrfs_fs_incompat(root
->fs_info
, SKINNY_METADATA
))
2634 num_bytes
+= heads
* sizeof(struct btrfs_tree_block_info
);
2637 * We don't ever fill up leaves all the way so multiply by 2 just to be
2638 * closer to what we're really going to want to ouse.
2640 return div64_u64(num_bytes
, BTRFS_LEAF_DATA_SIZE(root
));
2643 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle
*trans
,
2644 struct btrfs_root
*root
)
2646 struct btrfs_block_rsv
*global_rsv
;
2647 u64 num_heads
= trans
->transaction
->delayed_refs
.num_heads_ready
;
2651 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
2652 num_heads
= heads_to_leaves(root
, num_heads
);
2654 num_bytes
+= (num_heads
- 1) * root
->leafsize
;
2656 global_rsv
= &root
->fs_info
->global_block_rsv
;
2659 * If we can't allocate any more chunks lets make sure we have _lots_ of
2660 * wiggle room since running delayed refs can create more delayed refs.
2662 if (global_rsv
->space_info
->full
)
2665 spin_lock(&global_rsv
->lock
);
2666 if (global_rsv
->reserved
<= num_bytes
)
2668 spin_unlock(&global_rsv
->lock
);
2672 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle
*trans
,
2673 struct btrfs_root
*root
)
2675 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2677 atomic_read(&trans
->transaction
->delayed_refs
.num_entries
);
2682 avg_runtime
= fs_info
->avg_delayed_ref_runtime
;
2683 val
= num_entries
* avg_runtime
;
2684 if (num_entries
* avg_runtime
>= NSEC_PER_SEC
)
2686 if (val
>= NSEC_PER_SEC
/ 2)
2689 return btrfs_check_space_for_delayed_refs(trans
, root
);
2692 struct async_delayed_refs
{
2693 struct btrfs_root
*root
;
2697 struct completion wait
;
2698 struct btrfs_work work
;
2701 static void delayed_ref_async_start(struct btrfs_work
*work
)
2703 struct async_delayed_refs
*async
;
2704 struct btrfs_trans_handle
*trans
;
2707 async
= container_of(work
, struct async_delayed_refs
, work
);
2709 trans
= btrfs_join_transaction(async
->root
);
2710 if (IS_ERR(trans
)) {
2711 async
->error
= PTR_ERR(trans
);
2716 * trans->sync means that when we call end_transaciton, we won't
2717 * wait on delayed refs
2720 ret
= btrfs_run_delayed_refs(trans
, async
->root
, async
->count
);
2724 ret
= btrfs_end_transaction(trans
, async
->root
);
2725 if (ret
&& !async
->error
)
2729 complete(&async
->wait
);
2734 int btrfs_async_run_delayed_refs(struct btrfs_root
*root
,
2735 unsigned long count
, int wait
)
2737 struct async_delayed_refs
*async
;
2740 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
2744 async
->root
= root
->fs_info
->tree_root
;
2745 async
->count
= count
;
2751 init_completion(&async
->wait
);
2753 btrfs_init_work(&async
->work
, btrfs_extent_refs_helper
,
2754 delayed_ref_async_start
, NULL
, NULL
);
2756 btrfs_queue_work(root
->fs_info
->extent_workers
, &async
->work
);
2759 wait_for_completion(&async
->wait
);
2768 * this starts processing the delayed reference count updates and
2769 * extent insertions we have queued up so far. count can be
2770 * 0, which means to process everything in the tree at the start
2771 * of the run (but not newly added entries), or it can be some target
2772 * number you'd like to process.
2774 * Returns 0 on success or if called with an aborted transaction
2775 * Returns <0 on error and aborts the transaction
2777 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2778 struct btrfs_root
*root
, unsigned long count
)
2780 struct rb_node
*node
;
2781 struct btrfs_delayed_ref_root
*delayed_refs
;
2782 struct btrfs_delayed_ref_head
*head
;
2784 int run_all
= count
== (unsigned long)-1;
2787 /* We'll clean this up in btrfs_cleanup_transaction */
2791 if (root
== root
->fs_info
->extent_root
)
2792 root
= root
->fs_info
->tree_root
;
2794 delayed_refs
= &trans
->transaction
->delayed_refs
;
2796 count
= atomic_read(&delayed_refs
->num_entries
) * 2;
2801 #ifdef SCRAMBLE_DELAYED_REFS
2802 delayed_refs
->run_delayed_start
= find_middle(&delayed_refs
->root
);
2804 ret
= __btrfs_run_delayed_refs(trans
, root
, count
);
2806 btrfs_abort_transaction(trans
, root
, ret
);
2811 if (!list_empty(&trans
->new_bgs
))
2812 btrfs_create_pending_block_groups(trans
, root
);
2814 spin_lock(&delayed_refs
->lock
);
2815 node
= rb_first(&delayed_refs
->href_root
);
2817 spin_unlock(&delayed_refs
->lock
);
2820 count
= (unsigned long)-1;
2823 head
= rb_entry(node
, struct btrfs_delayed_ref_head
,
2825 if (btrfs_delayed_ref_is_head(&head
->node
)) {
2826 struct btrfs_delayed_ref_node
*ref
;
2829 atomic_inc(&ref
->refs
);
2831 spin_unlock(&delayed_refs
->lock
);
2833 * Mutex was contended, block until it's
2834 * released and try again
2836 mutex_lock(&head
->mutex
);
2837 mutex_unlock(&head
->mutex
);
2839 btrfs_put_delayed_ref(ref
);
2845 node
= rb_next(node
);
2847 spin_unlock(&delayed_refs
->lock
);
2852 ret
= btrfs_delayed_qgroup_accounting(trans
, root
->fs_info
);
2855 assert_qgroups_uptodate(trans
);
2859 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
2860 struct btrfs_root
*root
,
2861 u64 bytenr
, u64 num_bytes
, u64 flags
,
2862 int level
, int is_data
)
2864 struct btrfs_delayed_extent_op
*extent_op
;
2867 extent_op
= btrfs_alloc_delayed_extent_op();
2871 extent_op
->flags_to_set
= flags
;
2872 extent_op
->update_flags
= 1;
2873 extent_op
->update_key
= 0;
2874 extent_op
->is_data
= is_data
? 1 : 0;
2875 extent_op
->level
= level
;
2877 ret
= btrfs_add_delayed_extent_op(root
->fs_info
, trans
, bytenr
,
2878 num_bytes
, extent_op
);
2880 btrfs_free_delayed_extent_op(extent_op
);
2884 static noinline
int check_delayed_ref(struct btrfs_trans_handle
*trans
,
2885 struct btrfs_root
*root
,
2886 struct btrfs_path
*path
,
2887 u64 objectid
, u64 offset
, u64 bytenr
)
2889 struct btrfs_delayed_ref_head
*head
;
2890 struct btrfs_delayed_ref_node
*ref
;
2891 struct btrfs_delayed_data_ref
*data_ref
;
2892 struct btrfs_delayed_ref_root
*delayed_refs
;
2893 struct rb_node
*node
;
2896 delayed_refs
= &trans
->transaction
->delayed_refs
;
2897 spin_lock(&delayed_refs
->lock
);
2898 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
2900 spin_unlock(&delayed_refs
->lock
);
2904 if (!mutex_trylock(&head
->mutex
)) {
2905 atomic_inc(&head
->node
.refs
);
2906 spin_unlock(&delayed_refs
->lock
);
2908 btrfs_release_path(path
);
2911 * Mutex was contended, block until it's released and let
2914 mutex_lock(&head
->mutex
);
2915 mutex_unlock(&head
->mutex
);
2916 btrfs_put_delayed_ref(&head
->node
);
2919 spin_unlock(&delayed_refs
->lock
);
2921 spin_lock(&head
->lock
);
2922 node
= rb_first(&head
->ref_root
);
2924 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2925 node
= rb_next(node
);
2927 /* If it's a shared ref we know a cross reference exists */
2928 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
) {
2933 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
2936 * If our ref doesn't match the one we're currently looking at
2937 * then we have a cross reference.
2939 if (data_ref
->root
!= root
->root_key
.objectid
||
2940 data_ref
->objectid
!= objectid
||
2941 data_ref
->offset
!= offset
) {
2946 spin_unlock(&head
->lock
);
2947 mutex_unlock(&head
->mutex
);
2951 static noinline
int check_committed_ref(struct btrfs_trans_handle
*trans
,
2952 struct btrfs_root
*root
,
2953 struct btrfs_path
*path
,
2954 u64 objectid
, u64 offset
, u64 bytenr
)
2956 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2957 struct extent_buffer
*leaf
;
2958 struct btrfs_extent_data_ref
*ref
;
2959 struct btrfs_extent_inline_ref
*iref
;
2960 struct btrfs_extent_item
*ei
;
2961 struct btrfs_key key
;
2965 key
.objectid
= bytenr
;
2966 key
.offset
= (u64
)-1;
2967 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2969 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
2972 BUG_ON(ret
== 0); /* Corruption */
2975 if (path
->slots
[0] == 0)
2979 leaf
= path
->nodes
[0];
2980 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2982 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
2986 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2987 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2988 if (item_size
< sizeof(*ei
)) {
2989 WARN_ON(item_size
!= sizeof(struct btrfs_extent_item_v0
));
2993 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2995 if (item_size
!= sizeof(*ei
) +
2996 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
2999 if (btrfs_extent_generation(leaf
, ei
) <=
3000 btrfs_root_last_snapshot(&root
->root_item
))
3003 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
3004 if (btrfs_extent_inline_ref_type(leaf
, iref
) !=
3005 BTRFS_EXTENT_DATA_REF_KEY
)
3008 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
3009 if (btrfs_extent_refs(leaf
, ei
) !=
3010 btrfs_extent_data_ref_count(leaf
, ref
) ||
3011 btrfs_extent_data_ref_root(leaf
, ref
) !=
3012 root
->root_key
.objectid
||
3013 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
3014 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
3022 int btrfs_cross_ref_exist(struct btrfs_trans_handle
*trans
,
3023 struct btrfs_root
*root
,
3024 u64 objectid
, u64 offset
, u64 bytenr
)
3026 struct btrfs_path
*path
;
3030 path
= btrfs_alloc_path();
3035 ret
= check_committed_ref(trans
, root
, path
, objectid
,
3037 if (ret
&& ret
!= -ENOENT
)
3040 ret2
= check_delayed_ref(trans
, root
, path
, objectid
,
3042 } while (ret2
== -EAGAIN
);
3044 if (ret2
&& ret2
!= -ENOENT
) {
3049 if (ret
!= -ENOENT
|| ret2
!= -ENOENT
)
3052 btrfs_free_path(path
);
3053 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
3058 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
3059 struct btrfs_root
*root
,
3060 struct extent_buffer
*buf
,
3061 int full_backref
, int inc
)
3068 struct btrfs_key key
;
3069 struct btrfs_file_extent_item
*fi
;
3073 int (*process_func
)(struct btrfs_trans_handle
*, struct btrfs_root
*,
3074 u64
, u64
, u64
, u64
, u64
, u64
, int);
3076 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3077 if (unlikely(test_bit(BTRFS_ROOT_DUMMY_ROOT
, &root
->state
)))
3080 ref_root
= btrfs_header_owner(buf
);
3081 nritems
= btrfs_header_nritems(buf
);
3082 level
= btrfs_header_level(buf
);
3084 if (!test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) && level
== 0)
3088 process_func
= btrfs_inc_extent_ref
;
3090 process_func
= btrfs_free_extent
;
3093 parent
= buf
->start
;
3097 for (i
= 0; i
< nritems
; i
++) {
3099 btrfs_item_key_to_cpu(buf
, &key
, i
);
3100 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
3102 fi
= btrfs_item_ptr(buf
, i
,
3103 struct btrfs_file_extent_item
);
3104 if (btrfs_file_extent_type(buf
, fi
) ==
3105 BTRFS_FILE_EXTENT_INLINE
)
3107 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
3111 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
3112 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
3113 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
3114 parent
, ref_root
, key
.objectid
,
3119 bytenr
= btrfs_node_blockptr(buf
, i
);
3120 num_bytes
= btrfs_level_size(root
, level
- 1);
3121 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
3122 parent
, ref_root
, level
- 1, 0,
3133 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3134 struct extent_buffer
*buf
, int full_backref
)
3136 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1);
3139 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3140 struct extent_buffer
*buf
, int full_backref
)
3142 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0);
3145 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
3146 struct btrfs_root
*root
,
3147 struct btrfs_path
*path
,
3148 struct btrfs_block_group_cache
*cache
)
3151 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
3153 struct extent_buffer
*leaf
;
3155 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
3158 BUG_ON(ret
); /* Corruption */
3160 leaf
= path
->nodes
[0];
3161 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
3162 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
3163 btrfs_mark_buffer_dirty(leaf
);
3164 btrfs_release_path(path
);
3167 btrfs_abort_transaction(trans
, root
, ret
);
3174 static struct btrfs_block_group_cache
*
3175 next_block_group(struct btrfs_root
*root
,
3176 struct btrfs_block_group_cache
*cache
)
3178 struct rb_node
*node
;
3179 spin_lock(&root
->fs_info
->block_group_cache_lock
);
3180 node
= rb_next(&cache
->cache_node
);
3181 btrfs_put_block_group(cache
);
3183 cache
= rb_entry(node
, struct btrfs_block_group_cache
,
3185 btrfs_get_block_group(cache
);
3188 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
3192 static int cache_save_setup(struct btrfs_block_group_cache
*block_group
,
3193 struct btrfs_trans_handle
*trans
,
3194 struct btrfs_path
*path
)
3196 struct btrfs_root
*root
= block_group
->fs_info
->tree_root
;
3197 struct inode
*inode
= NULL
;
3199 int dcs
= BTRFS_DC_ERROR
;
3205 * If this block group is smaller than 100 megs don't bother caching the
3208 if (block_group
->key
.offset
< (100 * 1024 * 1024)) {
3209 spin_lock(&block_group
->lock
);
3210 block_group
->disk_cache_state
= BTRFS_DC_WRITTEN
;
3211 spin_unlock(&block_group
->lock
);
3216 inode
= lookup_free_space_inode(root
, block_group
, path
);
3217 if (IS_ERR(inode
) && PTR_ERR(inode
) != -ENOENT
) {
3218 ret
= PTR_ERR(inode
);
3219 btrfs_release_path(path
);
3223 if (IS_ERR(inode
)) {
3227 if (block_group
->ro
)
3230 ret
= create_free_space_inode(root
, trans
, block_group
, path
);
3236 /* We've already setup this transaction, go ahead and exit */
3237 if (block_group
->cache_generation
== trans
->transid
&&
3238 i_size_read(inode
)) {
3239 dcs
= BTRFS_DC_SETUP
;
3244 * We want to set the generation to 0, that way if anything goes wrong
3245 * from here on out we know not to trust this cache when we load up next
3248 BTRFS_I(inode
)->generation
= 0;
3249 ret
= btrfs_update_inode(trans
, root
, inode
);
3252 if (i_size_read(inode
) > 0) {
3253 ret
= btrfs_check_trunc_cache_free_space(root
,
3254 &root
->fs_info
->global_block_rsv
);
3258 ret
= btrfs_truncate_free_space_cache(root
, trans
, inode
);
3263 spin_lock(&block_group
->lock
);
3264 if (block_group
->cached
!= BTRFS_CACHE_FINISHED
||
3265 !btrfs_test_opt(root
, SPACE_CACHE
) ||
3266 block_group
->delalloc_bytes
) {
3268 * don't bother trying to write stuff out _if_
3269 * a) we're not cached,
3270 * b) we're with nospace_cache mount option.
3272 dcs
= BTRFS_DC_WRITTEN
;
3273 spin_unlock(&block_group
->lock
);
3276 spin_unlock(&block_group
->lock
);
3279 * Try to preallocate enough space based on how big the block group is.
3280 * Keep in mind this has to include any pinned space which could end up
3281 * taking up quite a bit since it's not folded into the other space
3284 num_pages
= (int)div64_u64(block_group
->key
.offset
, 256 * 1024 * 1024);
3289 num_pages
*= PAGE_CACHE_SIZE
;
3291 ret
= btrfs_check_data_free_space(inode
, num_pages
);
3295 ret
= btrfs_prealloc_file_range_trans(inode
, trans
, 0, 0, num_pages
,
3296 num_pages
, num_pages
,
3299 dcs
= BTRFS_DC_SETUP
;
3300 btrfs_free_reserved_data_space(inode
, num_pages
);
3305 btrfs_release_path(path
);
3307 spin_lock(&block_group
->lock
);
3308 if (!ret
&& dcs
== BTRFS_DC_SETUP
)
3309 block_group
->cache_generation
= trans
->transid
;
3310 block_group
->disk_cache_state
= dcs
;
3311 spin_unlock(&block_group
->lock
);
3316 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
3317 struct btrfs_root
*root
)
3319 struct btrfs_block_group_cache
*cache
;
3321 struct btrfs_path
*path
;
3324 path
= btrfs_alloc_path();
3330 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
3332 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
)
3334 cache
= next_block_group(root
, cache
);
3342 err
= cache_save_setup(cache
, trans
, path
);
3343 last
= cache
->key
.objectid
+ cache
->key
.offset
;
3344 btrfs_put_block_group(cache
);
3349 err
= btrfs_run_delayed_refs(trans
, root
,
3351 if (err
) /* File system offline */
3355 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
3357 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
) {
3358 btrfs_put_block_group(cache
);
3364 cache
= next_block_group(root
, cache
);
3373 if (cache
->disk_cache_state
== BTRFS_DC_SETUP
)
3374 cache
->disk_cache_state
= BTRFS_DC_NEED_WRITE
;
3376 last
= cache
->key
.objectid
+ cache
->key
.offset
;
3378 err
= write_one_cache_group(trans
, root
, path
, cache
);
3379 btrfs_put_block_group(cache
);
3380 if (err
) /* File system offline */
3386 * I don't think this is needed since we're just marking our
3387 * preallocated extent as written, but just in case it can't
3391 err
= btrfs_run_delayed_refs(trans
, root
,
3393 if (err
) /* File system offline */
3397 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
3400 * Really this shouldn't happen, but it could if we
3401 * couldn't write the entire preallocated extent and
3402 * splitting the extent resulted in a new block.
3405 btrfs_put_block_group(cache
);
3408 if (cache
->disk_cache_state
== BTRFS_DC_NEED_WRITE
)
3410 cache
= next_block_group(root
, cache
);
3419 err
= btrfs_write_out_cache(root
, trans
, cache
, path
);
3422 * If we didn't have an error then the cache state is still
3423 * NEED_WRITE, so we can set it to WRITTEN.
3425 if (!err
&& cache
->disk_cache_state
== BTRFS_DC_NEED_WRITE
)
3426 cache
->disk_cache_state
= BTRFS_DC_WRITTEN
;
3427 last
= cache
->key
.objectid
+ cache
->key
.offset
;
3428 btrfs_put_block_group(cache
);
3432 btrfs_free_path(path
);
3436 int btrfs_extent_readonly(struct btrfs_root
*root
, u64 bytenr
)
3438 struct btrfs_block_group_cache
*block_group
;
3441 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
3442 if (!block_group
|| block_group
->ro
)
3445 btrfs_put_block_group(block_group
);
3449 static const char *alloc_name(u64 flags
)
3452 case BTRFS_BLOCK_GROUP_METADATA
|BTRFS_BLOCK_GROUP_DATA
:
3454 case BTRFS_BLOCK_GROUP_METADATA
:
3456 case BTRFS_BLOCK_GROUP_DATA
:
3458 case BTRFS_BLOCK_GROUP_SYSTEM
:
3462 return "invalid-combination";
3466 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
3467 u64 total_bytes
, u64 bytes_used
,
3468 struct btrfs_space_info
**space_info
)
3470 struct btrfs_space_info
*found
;
3475 if (flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3476 BTRFS_BLOCK_GROUP_RAID10
))
3481 found
= __find_space_info(info
, flags
);
3483 spin_lock(&found
->lock
);
3484 found
->total_bytes
+= total_bytes
;
3485 found
->disk_total
+= total_bytes
* factor
;
3486 found
->bytes_used
+= bytes_used
;
3487 found
->disk_used
+= bytes_used
* factor
;
3489 spin_unlock(&found
->lock
);
3490 *space_info
= found
;
3493 found
= kzalloc(sizeof(*found
), GFP_NOFS
);
3497 ret
= percpu_counter_init(&found
->total_bytes_pinned
, 0);
3503 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
3504 INIT_LIST_HEAD(&found
->block_groups
[i
]);
3505 init_rwsem(&found
->groups_sem
);
3506 spin_lock_init(&found
->lock
);
3507 found
->flags
= flags
& BTRFS_BLOCK_GROUP_TYPE_MASK
;
3508 found
->total_bytes
= total_bytes
;
3509 found
->disk_total
= total_bytes
* factor
;
3510 found
->bytes_used
= bytes_used
;
3511 found
->disk_used
= bytes_used
* factor
;
3512 found
->bytes_pinned
= 0;
3513 found
->bytes_reserved
= 0;
3514 found
->bytes_readonly
= 0;
3515 found
->bytes_may_use
= 0;
3517 found
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3518 found
->chunk_alloc
= 0;
3520 init_waitqueue_head(&found
->wait
);
3522 ret
= kobject_init_and_add(&found
->kobj
, &space_info_ktype
,
3523 info
->space_info_kobj
, "%s",
3524 alloc_name(found
->flags
));
3530 *space_info
= found
;
3531 list_add_rcu(&found
->list
, &info
->space_info
);
3532 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3533 info
->data_sinfo
= found
;
3538 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
3540 u64 extra_flags
= chunk_to_extended(flags
) &
3541 BTRFS_EXTENDED_PROFILE_MASK
;
3543 write_seqlock(&fs_info
->profiles_lock
);
3544 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3545 fs_info
->avail_data_alloc_bits
|= extra_flags
;
3546 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3547 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
3548 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3549 fs_info
->avail_system_alloc_bits
|= extra_flags
;
3550 write_sequnlock(&fs_info
->profiles_lock
);
3554 * returns target flags in extended format or 0 if restripe for this
3555 * chunk_type is not in progress
3557 * should be called with either volume_mutex or balance_lock held
3559 static u64
get_restripe_target(struct btrfs_fs_info
*fs_info
, u64 flags
)
3561 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3567 if (flags
& BTRFS_BLOCK_GROUP_DATA
&&
3568 bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3569 target
= BTRFS_BLOCK_GROUP_DATA
| bctl
->data
.target
;
3570 } else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
&&
3571 bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3572 target
= BTRFS_BLOCK_GROUP_SYSTEM
| bctl
->sys
.target
;
3573 } else if (flags
& BTRFS_BLOCK_GROUP_METADATA
&&
3574 bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3575 target
= BTRFS_BLOCK_GROUP_METADATA
| bctl
->meta
.target
;
3582 * @flags: available profiles in extended format (see ctree.h)
3584 * Returns reduced profile in chunk format. If profile changing is in
3585 * progress (either running or paused) picks the target profile (if it's
3586 * already available), otherwise falls back to plain reducing.
3588 static u64
btrfs_reduce_alloc_profile(struct btrfs_root
*root
, u64 flags
)
3590 u64 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
3595 * see if restripe for this chunk_type is in progress, if so
3596 * try to reduce to the target profile
3598 spin_lock(&root
->fs_info
->balance_lock
);
3599 target
= get_restripe_target(root
->fs_info
, flags
);
3601 /* pick target profile only if it's already available */
3602 if ((flags
& target
) & BTRFS_EXTENDED_PROFILE_MASK
) {
3603 spin_unlock(&root
->fs_info
->balance_lock
);
3604 return extended_to_chunk(target
);
3607 spin_unlock(&root
->fs_info
->balance_lock
);
3609 /* First, mask out the RAID levels which aren't possible */
3610 if (num_devices
== 1)
3611 flags
&= ~(BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID0
|
3612 BTRFS_BLOCK_GROUP_RAID5
);
3613 if (num_devices
< 3)
3614 flags
&= ~BTRFS_BLOCK_GROUP_RAID6
;
3615 if (num_devices
< 4)
3616 flags
&= ~BTRFS_BLOCK_GROUP_RAID10
;
3618 tmp
= flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID0
|
3619 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID5
|
3620 BTRFS_BLOCK_GROUP_RAID6
| BTRFS_BLOCK_GROUP_RAID10
);
3623 if (tmp
& BTRFS_BLOCK_GROUP_RAID6
)
3624 tmp
= BTRFS_BLOCK_GROUP_RAID6
;
3625 else if (tmp
& BTRFS_BLOCK_GROUP_RAID5
)
3626 tmp
= BTRFS_BLOCK_GROUP_RAID5
;
3627 else if (tmp
& BTRFS_BLOCK_GROUP_RAID10
)
3628 tmp
= BTRFS_BLOCK_GROUP_RAID10
;
3629 else if (tmp
& BTRFS_BLOCK_GROUP_RAID1
)
3630 tmp
= BTRFS_BLOCK_GROUP_RAID1
;
3631 else if (tmp
& BTRFS_BLOCK_GROUP_RAID0
)
3632 tmp
= BTRFS_BLOCK_GROUP_RAID0
;
3634 return extended_to_chunk(flags
| tmp
);
3637 static u64
get_alloc_profile(struct btrfs_root
*root
, u64 orig_flags
)
3644 seq
= read_seqbegin(&root
->fs_info
->profiles_lock
);
3646 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3647 flags
|= root
->fs_info
->avail_data_alloc_bits
;
3648 else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3649 flags
|= root
->fs_info
->avail_system_alloc_bits
;
3650 else if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3651 flags
|= root
->fs_info
->avail_metadata_alloc_bits
;
3652 } while (read_seqretry(&root
->fs_info
->profiles_lock
, seq
));
3654 return btrfs_reduce_alloc_profile(root
, flags
);
3657 u64
btrfs_get_alloc_profile(struct btrfs_root
*root
, int data
)
3663 flags
= BTRFS_BLOCK_GROUP_DATA
;
3664 else if (root
== root
->fs_info
->chunk_root
)
3665 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
3667 flags
= BTRFS_BLOCK_GROUP_METADATA
;
3669 ret
= get_alloc_profile(root
, flags
);
3674 * This will check the space that the inode allocates from to make sure we have
3675 * enough space for bytes.
3677 int btrfs_check_data_free_space(struct inode
*inode
, u64 bytes
)
3679 struct btrfs_space_info
*data_sinfo
;
3680 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3681 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3683 int ret
= 0, committed
= 0, alloc_chunk
= 1;
3685 /* make sure bytes are sectorsize aligned */
3686 bytes
= ALIGN(bytes
, root
->sectorsize
);
3688 if (btrfs_is_free_space_inode(inode
)) {
3690 ASSERT(current
->journal_info
);
3693 data_sinfo
= fs_info
->data_sinfo
;
3698 /* make sure we have enough space to handle the data first */
3699 spin_lock(&data_sinfo
->lock
);
3700 used
= data_sinfo
->bytes_used
+ data_sinfo
->bytes_reserved
+
3701 data_sinfo
->bytes_pinned
+ data_sinfo
->bytes_readonly
+
3702 data_sinfo
->bytes_may_use
;
3704 if (used
+ bytes
> data_sinfo
->total_bytes
) {
3705 struct btrfs_trans_handle
*trans
;
3708 * if we don't have enough free bytes in this space then we need
3709 * to alloc a new chunk.
3711 if (!data_sinfo
->full
&& alloc_chunk
) {
3714 data_sinfo
->force_alloc
= CHUNK_ALLOC_FORCE
;
3715 spin_unlock(&data_sinfo
->lock
);
3717 alloc_target
= btrfs_get_alloc_profile(root
, 1);
3719 * It is ugly that we don't call nolock join
3720 * transaction for the free space inode case here.
3721 * But it is safe because we only do the data space
3722 * reservation for the free space cache in the
3723 * transaction context, the common join transaction
3724 * just increase the counter of the current transaction
3725 * handler, doesn't try to acquire the trans_lock of
3728 trans
= btrfs_join_transaction(root
);
3730 return PTR_ERR(trans
);
3732 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
3734 CHUNK_ALLOC_NO_FORCE
);
3735 btrfs_end_transaction(trans
, root
);
3744 data_sinfo
= fs_info
->data_sinfo
;
3750 * If we don't have enough pinned space to deal with this
3751 * allocation don't bother committing the transaction.
3753 if (percpu_counter_compare(&data_sinfo
->total_bytes_pinned
,
3756 spin_unlock(&data_sinfo
->lock
);
3758 /* commit the current transaction and try again */
3761 !atomic_read(&root
->fs_info
->open_ioctl_trans
)) {
3764 trans
= btrfs_join_transaction(root
);
3766 return PTR_ERR(trans
);
3767 ret
= btrfs_commit_transaction(trans
, root
);
3773 trace_btrfs_space_reservation(root
->fs_info
,
3774 "space_info:enospc",
3775 data_sinfo
->flags
, bytes
, 1);
3778 data_sinfo
->bytes_may_use
+= bytes
;
3779 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
3780 data_sinfo
->flags
, bytes
, 1);
3781 spin_unlock(&data_sinfo
->lock
);
3787 * Called if we need to clear a data reservation for this inode.
3789 void btrfs_free_reserved_data_space(struct inode
*inode
, u64 bytes
)
3791 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3792 struct btrfs_space_info
*data_sinfo
;
3794 /* make sure bytes are sectorsize aligned */
3795 bytes
= ALIGN(bytes
, root
->sectorsize
);
3797 data_sinfo
= root
->fs_info
->data_sinfo
;
3798 spin_lock(&data_sinfo
->lock
);
3799 WARN_ON(data_sinfo
->bytes_may_use
< bytes
);
3800 data_sinfo
->bytes_may_use
-= bytes
;
3801 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
3802 data_sinfo
->flags
, bytes
, 0);
3803 spin_unlock(&data_sinfo
->lock
);
3806 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
3808 struct list_head
*head
= &info
->space_info
;
3809 struct btrfs_space_info
*found
;
3812 list_for_each_entry_rcu(found
, head
, list
) {
3813 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
3814 found
->force_alloc
= CHUNK_ALLOC_FORCE
;
3819 static inline u64
calc_global_rsv_need_space(struct btrfs_block_rsv
*global
)
3821 return (global
->size
<< 1);
3824 static int should_alloc_chunk(struct btrfs_root
*root
,
3825 struct btrfs_space_info
*sinfo
, int force
)
3827 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
3828 u64 num_bytes
= sinfo
->total_bytes
- sinfo
->bytes_readonly
;
3829 u64 num_allocated
= sinfo
->bytes_used
+ sinfo
->bytes_reserved
;
3832 if (force
== CHUNK_ALLOC_FORCE
)
3836 * We need to take into account the global rsv because for all intents
3837 * and purposes it's used space. Don't worry about locking the
3838 * global_rsv, it doesn't change except when the transaction commits.
3840 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
3841 num_allocated
+= calc_global_rsv_need_space(global_rsv
);
3844 * in limited mode, we want to have some free space up to
3845 * about 1% of the FS size.
3847 if (force
== CHUNK_ALLOC_LIMITED
) {
3848 thresh
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
3849 thresh
= max_t(u64
, 64 * 1024 * 1024,
3850 div_factor_fine(thresh
, 1));
3852 if (num_bytes
- num_allocated
< thresh
)
3856 if (num_allocated
+ 2 * 1024 * 1024 < div_factor(num_bytes
, 8))
3861 static u64
get_system_chunk_thresh(struct btrfs_root
*root
, u64 type
)
3865 if (type
& (BTRFS_BLOCK_GROUP_RAID10
|
3866 BTRFS_BLOCK_GROUP_RAID0
|
3867 BTRFS_BLOCK_GROUP_RAID5
|
3868 BTRFS_BLOCK_GROUP_RAID6
))
3869 num_dev
= root
->fs_info
->fs_devices
->rw_devices
;
3870 else if (type
& BTRFS_BLOCK_GROUP_RAID1
)
3873 num_dev
= 1; /* DUP or single */
3875 /* metadata for updaing devices and chunk tree */
3876 return btrfs_calc_trans_metadata_size(root
, num_dev
+ 1);
3879 static void check_system_chunk(struct btrfs_trans_handle
*trans
,
3880 struct btrfs_root
*root
, u64 type
)
3882 struct btrfs_space_info
*info
;
3886 info
= __find_space_info(root
->fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
3887 spin_lock(&info
->lock
);
3888 left
= info
->total_bytes
- info
->bytes_used
- info
->bytes_pinned
-
3889 info
->bytes_reserved
- info
->bytes_readonly
;
3890 spin_unlock(&info
->lock
);
3892 thresh
= get_system_chunk_thresh(root
, type
);
3893 if (left
< thresh
&& btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
3894 btrfs_info(root
->fs_info
, "left=%llu, need=%llu, flags=%llu",
3895 left
, thresh
, type
);
3896 dump_space_info(info
, 0, 0);
3899 if (left
< thresh
) {
3902 flags
= btrfs_get_alloc_profile(root
->fs_info
->chunk_root
, 0);
3903 btrfs_alloc_chunk(trans
, root
, flags
);
3907 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
3908 struct btrfs_root
*extent_root
, u64 flags
, int force
)
3910 struct btrfs_space_info
*space_info
;
3911 struct btrfs_fs_info
*fs_info
= extent_root
->fs_info
;
3912 int wait_for_alloc
= 0;
3915 /* Don't re-enter if we're already allocating a chunk */
3916 if (trans
->allocating_chunk
)
3919 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
3921 ret
= update_space_info(extent_root
->fs_info
, flags
,
3923 BUG_ON(ret
); /* -ENOMEM */
3925 BUG_ON(!space_info
); /* Logic error */
3928 spin_lock(&space_info
->lock
);
3929 if (force
< space_info
->force_alloc
)
3930 force
= space_info
->force_alloc
;
3931 if (space_info
->full
) {
3932 if (should_alloc_chunk(extent_root
, space_info
, force
))
3936 spin_unlock(&space_info
->lock
);
3940 if (!should_alloc_chunk(extent_root
, space_info
, force
)) {
3941 spin_unlock(&space_info
->lock
);
3943 } else if (space_info
->chunk_alloc
) {
3946 space_info
->chunk_alloc
= 1;
3949 spin_unlock(&space_info
->lock
);
3951 mutex_lock(&fs_info
->chunk_mutex
);
3954 * The chunk_mutex is held throughout the entirety of a chunk
3955 * allocation, so once we've acquired the chunk_mutex we know that the
3956 * other guy is done and we need to recheck and see if we should
3959 if (wait_for_alloc
) {
3960 mutex_unlock(&fs_info
->chunk_mutex
);
3965 trans
->allocating_chunk
= true;
3968 * If we have mixed data/metadata chunks we want to make sure we keep
3969 * allocating mixed chunks instead of individual chunks.
3971 if (btrfs_mixed_space_info(space_info
))
3972 flags
|= (BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
);
3975 * if we're doing a data chunk, go ahead and make sure that
3976 * we keep a reasonable number of metadata chunks allocated in the
3979 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
3980 fs_info
->data_chunk_allocations
++;
3981 if (!(fs_info
->data_chunk_allocations
%
3982 fs_info
->metadata_ratio
))
3983 force_metadata_allocation(fs_info
);
3987 * Check if we have enough space in SYSTEM chunk because we may need
3988 * to update devices.
3990 check_system_chunk(trans
, extent_root
, flags
);
3992 ret
= btrfs_alloc_chunk(trans
, extent_root
, flags
);
3993 trans
->allocating_chunk
= false;
3995 spin_lock(&space_info
->lock
);
3996 if (ret
< 0 && ret
!= -ENOSPC
)
3999 space_info
->full
= 1;
4003 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
4005 space_info
->chunk_alloc
= 0;
4006 spin_unlock(&space_info
->lock
);
4007 mutex_unlock(&fs_info
->chunk_mutex
);
4011 static int can_overcommit(struct btrfs_root
*root
,
4012 struct btrfs_space_info
*space_info
, u64 bytes
,
4013 enum btrfs_reserve_flush_enum flush
)
4015 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
4016 u64 profile
= btrfs_get_alloc_profile(root
, 0);
4021 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4022 space_info
->bytes_pinned
+ space_info
->bytes_readonly
;
4025 * We only want to allow over committing if we have lots of actual space
4026 * free, but if we don't have enough space to handle the global reserve
4027 * space then we could end up having a real enospc problem when trying
4028 * to allocate a chunk or some other such important allocation.
4030 spin_lock(&global_rsv
->lock
);
4031 space_size
= calc_global_rsv_need_space(global_rsv
);
4032 spin_unlock(&global_rsv
->lock
);
4033 if (used
+ space_size
>= space_info
->total_bytes
)
4036 used
+= space_info
->bytes_may_use
;
4038 spin_lock(&root
->fs_info
->free_chunk_lock
);
4039 avail
= root
->fs_info
->free_chunk_space
;
4040 spin_unlock(&root
->fs_info
->free_chunk_lock
);
4043 * If we have dup, raid1 or raid10 then only half of the free
4044 * space is actually useable. For raid56, the space info used
4045 * doesn't include the parity drive, so we don't have to
4048 if (profile
& (BTRFS_BLOCK_GROUP_DUP
|
4049 BTRFS_BLOCK_GROUP_RAID1
|
4050 BTRFS_BLOCK_GROUP_RAID10
))
4054 * If we aren't flushing all things, let us overcommit up to
4055 * 1/2th of the space. If we can flush, don't let us overcommit
4056 * too much, let it overcommit up to 1/8 of the space.
4058 if (flush
== BTRFS_RESERVE_FLUSH_ALL
)
4063 if (used
+ bytes
< space_info
->total_bytes
+ avail
)
4068 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root
*root
,
4069 unsigned long nr_pages
, int nr_items
)
4071 struct super_block
*sb
= root
->fs_info
->sb
;
4073 if (down_read_trylock(&sb
->s_umount
)) {
4074 writeback_inodes_sb_nr(sb
, nr_pages
, WB_REASON_FS_FREE_SPACE
);
4075 up_read(&sb
->s_umount
);
4078 * We needn't worry the filesystem going from r/w to r/o though
4079 * we don't acquire ->s_umount mutex, because the filesystem
4080 * should guarantee the delalloc inodes list be empty after
4081 * the filesystem is readonly(all dirty pages are written to
4084 btrfs_start_delalloc_roots(root
->fs_info
, 0, nr_items
);
4085 if (!current
->journal_info
)
4086 btrfs_wait_ordered_roots(root
->fs_info
, nr_items
);
4090 static inline int calc_reclaim_items_nr(struct btrfs_root
*root
, u64 to_reclaim
)
4095 bytes
= btrfs_calc_trans_metadata_size(root
, 1);
4096 nr
= (int)div64_u64(to_reclaim
, bytes
);
4102 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4105 * shrink metadata reservation for delalloc
4107 static void shrink_delalloc(struct btrfs_root
*root
, u64 to_reclaim
, u64 orig
,
4110 struct btrfs_block_rsv
*block_rsv
;
4111 struct btrfs_space_info
*space_info
;
4112 struct btrfs_trans_handle
*trans
;
4116 unsigned long nr_pages
;
4119 enum btrfs_reserve_flush_enum flush
;
4121 /* Calc the number of the pages we need flush for space reservation */
4122 items
= calc_reclaim_items_nr(root
, to_reclaim
);
4123 to_reclaim
= items
* EXTENT_SIZE_PER_ITEM
;
4125 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4126 block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
4127 space_info
= block_rsv
->space_info
;
4129 delalloc_bytes
= percpu_counter_sum_positive(
4130 &root
->fs_info
->delalloc_bytes
);
4131 if (delalloc_bytes
== 0) {
4135 btrfs_wait_ordered_roots(root
->fs_info
, items
);
4140 while (delalloc_bytes
&& loops
< 3) {
4141 max_reclaim
= min(delalloc_bytes
, to_reclaim
);
4142 nr_pages
= max_reclaim
>> PAGE_CACHE_SHIFT
;
4143 btrfs_writeback_inodes_sb_nr(root
, nr_pages
, items
);
4145 * We need to wait for the async pages to actually start before
4148 max_reclaim
= atomic_read(&root
->fs_info
->async_delalloc_pages
);
4152 if (max_reclaim
<= nr_pages
)
4155 max_reclaim
-= nr_pages
;
4157 wait_event(root
->fs_info
->async_submit_wait
,
4158 atomic_read(&root
->fs_info
->async_delalloc_pages
) <=
4162 flush
= BTRFS_RESERVE_FLUSH_ALL
;
4164 flush
= BTRFS_RESERVE_NO_FLUSH
;
4165 spin_lock(&space_info
->lock
);
4166 if (can_overcommit(root
, space_info
, orig
, flush
)) {
4167 spin_unlock(&space_info
->lock
);
4170 spin_unlock(&space_info
->lock
);
4173 if (wait_ordered
&& !trans
) {
4174 btrfs_wait_ordered_roots(root
->fs_info
, items
);
4176 time_left
= schedule_timeout_killable(1);
4180 delalloc_bytes
= percpu_counter_sum_positive(
4181 &root
->fs_info
->delalloc_bytes
);
4186 * maybe_commit_transaction - possibly commit the transaction if its ok to
4187 * @root - the root we're allocating for
4188 * @bytes - the number of bytes we want to reserve
4189 * @force - force the commit
4191 * This will check to make sure that committing the transaction will actually
4192 * get us somewhere and then commit the transaction if it does. Otherwise it
4193 * will return -ENOSPC.
4195 static int may_commit_transaction(struct btrfs_root
*root
,
4196 struct btrfs_space_info
*space_info
,
4197 u64 bytes
, int force
)
4199 struct btrfs_block_rsv
*delayed_rsv
= &root
->fs_info
->delayed_block_rsv
;
4200 struct btrfs_trans_handle
*trans
;
4202 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4209 /* See if there is enough pinned space to make this reservation */
4210 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4215 * See if there is some space in the delayed insertion reservation for
4218 if (space_info
!= delayed_rsv
->space_info
)
4221 spin_lock(&delayed_rsv
->lock
);
4222 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4223 bytes
- delayed_rsv
->size
) >= 0) {
4224 spin_unlock(&delayed_rsv
->lock
);
4227 spin_unlock(&delayed_rsv
->lock
);
4230 trans
= btrfs_join_transaction(root
);
4234 return btrfs_commit_transaction(trans
, root
);
4238 FLUSH_DELAYED_ITEMS_NR
= 1,
4239 FLUSH_DELAYED_ITEMS
= 2,
4241 FLUSH_DELALLOC_WAIT
= 4,
4246 static int flush_space(struct btrfs_root
*root
,
4247 struct btrfs_space_info
*space_info
, u64 num_bytes
,
4248 u64 orig_bytes
, int state
)
4250 struct btrfs_trans_handle
*trans
;
4255 case FLUSH_DELAYED_ITEMS_NR
:
4256 case FLUSH_DELAYED_ITEMS
:
4257 if (state
== FLUSH_DELAYED_ITEMS_NR
)
4258 nr
= calc_reclaim_items_nr(root
, num_bytes
) * 2;
4262 trans
= btrfs_join_transaction(root
);
4263 if (IS_ERR(trans
)) {
4264 ret
= PTR_ERR(trans
);
4267 ret
= btrfs_run_delayed_items_nr(trans
, root
, nr
);
4268 btrfs_end_transaction(trans
, root
);
4270 case FLUSH_DELALLOC
:
4271 case FLUSH_DELALLOC_WAIT
:
4272 shrink_delalloc(root
, num_bytes
* 2, orig_bytes
,
4273 state
== FLUSH_DELALLOC_WAIT
);
4276 trans
= btrfs_join_transaction(root
);
4277 if (IS_ERR(trans
)) {
4278 ret
= PTR_ERR(trans
);
4281 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
4282 btrfs_get_alloc_profile(root
, 0),
4283 CHUNK_ALLOC_NO_FORCE
);
4284 btrfs_end_transaction(trans
, root
);
4289 ret
= may_commit_transaction(root
, space_info
, orig_bytes
, 0);
4300 btrfs_calc_reclaim_metadata_size(struct btrfs_root
*root
,
4301 struct btrfs_space_info
*space_info
)
4307 to_reclaim
= min_t(u64
, num_online_cpus() * 1024 * 1024,
4309 spin_lock(&space_info
->lock
);
4310 if (can_overcommit(root
, space_info
, to_reclaim
,
4311 BTRFS_RESERVE_FLUSH_ALL
)) {
4316 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4317 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4318 space_info
->bytes_may_use
;
4319 if (can_overcommit(root
, space_info
, 1024 * 1024,
4320 BTRFS_RESERVE_FLUSH_ALL
))
4321 expected
= div_factor_fine(space_info
->total_bytes
, 95);
4323 expected
= div_factor_fine(space_info
->total_bytes
, 90);
4325 if (used
> expected
)
4326 to_reclaim
= used
- expected
;
4329 to_reclaim
= min(to_reclaim
, space_info
->bytes_may_use
+
4330 space_info
->bytes_reserved
);
4332 spin_unlock(&space_info
->lock
);
4337 static inline int need_do_async_reclaim(struct btrfs_space_info
*space_info
,
4338 struct btrfs_fs_info
*fs_info
, u64 used
)
4340 return (used
>= div_factor_fine(space_info
->total_bytes
, 98) &&
4341 !btrfs_fs_closing(fs_info
) &&
4342 !test_bit(BTRFS_FS_STATE_REMOUNTING
, &fs_info
->fs_state
));
4345 static int btrfs_need_do_async_reclaim(struct btrfs_space_info
*space_info
,
4346 struct btrfs_fs_info
*fs_info
)
4350 spin_lock(&space_info
->lock
);
4351 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4352 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4353 space_info
->bytes_may_use
;
4354 if (need_do_async_reclaim(space_info
, fs_info
, used
)) {
4355 spin_unlock(&space_info
->lock
);
4358 spin_unlock(&space_info
->lock
);
4363 static void btrfs_async_reclaim_metadata_space(struct work_struct
*work
)
4365 struct btrfs_fs_info
*fs_info
;
4366 struct btrfs_space_info
*space_info
;
4370 fs_info
= container_of(work
, struct btrfs_fs_info
, async_reclaim_work
);
4371 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4373 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
->fs_root
,
4378 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
4380 flush_space(fs_info
->fs_root
, space_info
, to_reclaim
,
4381 to_reclaim
, flush_state
);
4383 if (!btrfs_need_do_async_reclaim(space_info
, fs_info
))
4385 } while (flush_state
<= COMMIT_TRANS
);
4387 if (btrfs_need_do_async_reclaim(space_info
, fs_info
))
4388 queue_work(system_unbound_wq
, work
);
4391 void btrfs_init_async_reclaim_work(struct work_struct
*work
)
4393 INIT_WORK(work
, btrfs_async_reclaim_metadata_space
);
4397 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4398 * @root - the root we're allocating for
4399 * @block_rsv - the block_rsv we're allocating for
4400 * @orig_bytes - the number of bytes we want
4401 * @flush - whether or not we can flush to make our reservation
4403 * This will reserve orgi_bytes number of bytes from the space info associated
4404 * with the block_rsv. If there is not enough space it will make an attempt to
4405 * flush out space to make room. It will do this by flushing delalloc if
4406 * possible or committing the transaction. If flush is 0 then no attempts to
4407 * regain reservations will be made and this will fail if there is not enough
4410 static int reserve_metadata_bytes(struct btrfs_root
*root
,
4411 struct btrfs_block_rsv
*block_rsv
,
4413 enum btrfs_reserve_flush_enum flush
)
4415 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
4417 u64 num_bytes
= orig_bytes
;
4418 int flush_state
= FLUSH_DELAYED_ITEMS_NR
;
4420 bool flushing
= false;
4424 spin_lock(&space_info
->lock
);
4426 * We only want to wait if somebody other than us is flushing and we
4427 * are actually allowed to flush all things.
4429 while (flush
== BTRFS_RESERVE_FLUSH_ALL
&& !flushing
&&
4430 space_info
->flush
) {
4431 spin_unlock(&space_info
->lock
);
4433 * If we have a trans handle we can't wait because the flusher
4434 * may have to commit the transaction, which would mean we would
4435 * deadlock since we are waiting for the flusher to finish, but
4436 * hold the current transaction open.
4438 if (current
->journal_info
)
4440 ret
= wait_event_killable(space_info
->wait
, !space_info
->flush
);
4441 /* Must have been killed, return */
4445 spin_lock(&space_info
->lock
);
4449 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4450 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4451 space_info
->bytes_may_use
;
4454 * The idea here is that we've not already over-reserved the block group
4455 * then we can go ahead and save our reservation first and then start
4456 * flushing if we need to. Otherwise if we've already overcommitted
4457 * lets start flushing stuff first and then come back and try to make
4460 if (used
<= space_info
->total_bytes
) {
4461 if (used
+ orig_bytes
<= space_info
->total_bytes
) {
4462 space_info
->bytes_may_use
+= orig_bytes
;
4463 trace_btrfs_space_reservation(root
->fs_info
,
4464 "space_info", space_info
->flags
, orig_bytes
, 1);
4468 * Ok set num_bytes to orig_bytes since we aren't
4469 * overocmmitted, this way we only try and reclaim what
4472 num_bytes
= orig_bytes
;
4476 * Ok we're over committed, set num_bytes to the overcommitted
4477 * amount plus the amount of bytes that we need for this
4480 num_bytes
= used
- space_info
->total_bytes
+
4484 if (ret
&& can_overcommit(root
, space_info
, orig_bytes
, flush
)) {
4485 space_info
->bytes_may_use
+= orig_bytes
;
4486 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
4487 space_info
->flags
, orig_bytes
,
4493 * Couldn't make our reservation, save our place so while we're trying
4494 * to reclaim space we can actually use it instead of somebody else
4495 * stealing it from us.
4497 * We make the other tasks wait for the flush only when we can flush
4500 if (ret
&& flush
!= BTRFS_RESERVE_NO_FLUSH
) {
4502 space_info
->flush
= 1;
4503 } else if (!ret
&& space_info
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
4505 if (need_do_async_reclaim(space_info
, root
->fs_info
, used
) &&
4506 !work_busy(&root
->fs_info
->async_reclaim_work
))
4507 queue_work(system_unbound_wq
,
4508 &root
->fs_info
->async_reclaim_work
);
4510 spin_unlock(&space_info
->lock
);
4512 if (!ret
|| flush
== BTRFS_RESERVE_NO_FLUSH
)
4515 ret
= flush_space(root
, space_info
, num_bytes
, orig_bytes
,
4520 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4521 * would happen. So skip delalloc flush.
4523 if (flush
== BTRFS_RESERVE_FLUSH_LIMIT
&&
4524 (flush_state
== FLUSH_DELALLOC
||
4525 flush_state
== FLUSH_DELALLOC_WAIT
))
4526 flush_state
= ALLOC_CHUNK
;
4530 else if (flush
== BTRFS_RESERVE_FLUSH_LIMIT
&&
4531 flush_state
< COMMIT_TRANS
)
4533 else if (flush
== BTRFS_RESERVE_FLUSH_ALL
&&
4534 flush_state
<= COMMIT_TRANS
)
4538 if (ret
== -ENOSPC
&&
4539 unlikely(root
->orphan_cleanup_state
== ORPHAN_CLEANUP_STARTED
)) {
4540 struct btrfs_block_rsv
*global_rsv
=
4541 &root
->fs_info
->global_block_rsv
;
4543 if (block_rsv
!= global_rsv
&&
4544 !block_rsv_use_bytes(global_rsv
, orig_bytes
))
4548 trace_btrfs_space_reservation(root
->fs_info
,
4549 "space_info:enospc",
4550 space_info
->flags
, orig_bytes
, 1);
4552 spin_lock(&space_info
->lock
);
4553 space_info
->flush
= 0;
4554 wake_up_all(&space_info
->wait
);
4555 spin_unlock(&space_info
->lock
);
4560 static struct btrfs_block_rsv
*get_block_rsv(
4561 const struct btrfs_trans_handle
*trans
,
4562 const struct btrfs_root
*root
)
4564 struct btrfs_block_rsv
*block_rsv
= NULL
;
4566 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
))
4567 block_rsv
= trans
->block_rsv
;
4569 if (root
== root
->fs_info
->csum_root
&& trans
->adding_csums
)
4570 block_rsv
= trans
->block_rsv
;
4572 if (root
== root
->fs_info
->uuid_root
)
4573 block_rsv
= trans
->block_rsv
;
4576 block_rsv
= root
->block_rsv
;
4579 block_rsv
= &root
->fs_info
->empty_block_rsv
;
4584 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
4588 spin_lock(&block_rsv
->lock
);
4589 if (block_rsv
->reserved
>= num_bytes
) {
4590 block_rsv
->reserved
-= num_bytes
;
4591 if (block_rsv
->reserved
< block_rsv
->size
)
4592 block_rsv
->full
= 0;
4595 spin_unlock(&block_rsv
->lock
);
4599 static void block_rsv_add_bytes(struct btrfs_block_rsv
*block_rsv
,
4600 u64 num_bytes
, int update_size
)
4602 spin_lock(&block_rsv
->lock
);
4603 block_rsv
->reserved
+= num_bytes
;
4605 block_rsv
->size
+= num_bytes
;
4606 else if (block_rsv
->reserved
>= block_rsv
->size
)
4607 block_rsv
->full
= 1;
4608 spin_unlock(&block_rsv
->lock
);
4611 int btrfs_cond_migrate_bytes(struct btrfs_fs_info
*fs_info
,
4612 struct btrfs_block_rsv
*dest
, u64 num_bytes
,
4615 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
4618 if (global_rsv
->space_info
!= dest
->space_info
)
4621 spin_lock(&global_rsv
->lock
);
4622 min_bytes
= div_factor(global_rsv
->size
, min_factor
);
4623 if (global_rsv
->reserved
< min_bytes
+ num_bytes
) {
4624 spin_unlock(&global_rsv
->lock
);
4627 global_rsv
->reserved
-= num_bytes
;
4628 if (global_rsv
->reserved
< global_rsv
->size
)
4629 global_rsv
->full
= 0;
4630 spin_unlock(&global_rsv
->lock
);
4632 block_rsv_add_bytes(dest
, num_bytes
, 1);
4636 static void block_rsv_release_bytes(struct btrfs_fs_info
*fs_info
,
4637 struct btrfs_block_rsv
*block_rsv
,
4638 struct btrfs_block_rsv
*dest
, u64 num_bytes
)
4640 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
4642 spin_lock(&block_rsv
->lock
);
4643 if (num_bytes
== (u64
)-1)
4644 num_bytes
= block_rsv
->size
;
4645 block_rsv
->size
-= num_bytes
;
4646 if (block_rsv
->reserved
>= block_rsv
->size
) {
4647 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
4648 block_rsv
->reserved
= block_rsv
->size
;
4649 block_rsv
->full
= 1;
4653 spin_unlock(&block_rsv
->lock
);
4655 if (num_bytes
> 0) {
4657 spin_lock(&dest
->lock
);
4661 bytes_to_add
= dest
->size
- dest
->reserved
;
4662 bytes_to_add
= min(num_bytes
, bytes_to_add
);
4663 dest
->reserved
+= bytes_to_add
;
4664 if (dest
->reserved
>= dest
->size
)
4666 num_bytes
-= bytes_to_add
;
4668 spin_unlock(&dest
->lock
);
4671 spin_lock(&space_info
->lock
);
4672 space_info
->bytes_may_use
-= num_bytes
;
4673 trace_btrfs_space_reservation(fs_info
, "space_info",
4674 space_info
->flags
, num_bytes
, 0);
4675 spin_unlock(&space_info
->lock
);
4680 static int block_rsv_migrate_bytes(struct btrfs_block_rsv
*src
,
4681 struct btrfs_block_rsv
*dst
, u64 num_bytes
)
4685 ret
= block_rsv_use_bytes(src
, num_bytes
);
4689 block_rsv_add_bytes(dst
, num_bytes
, 1);
4693 void btrfs_init_block_rsv(struct btrfs_block_rsv
*rsv
, unsigned short type
)
4695 memset(rsv
, 0, sizeof(*rsv
));
4696 spin_lock_init(&rsv
->lock
);
4700 struct btrfs_block_rsv
*btrfs_alloc_block_rsv(struct btrfs_root
*root
,
4701 unsigned short type
)
4703 struct btrfs_block_rsv
*block_rsv
;
4704 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4706 block_rsv
= kmalloc(sizeof(*block_rsv
), GFP_NOFS
);
4710 btrfs_init_block_rsv(block_rsv
, type
);
4711 block_rsv
->space_info
= __find_space_info(fs_info
,
4712 BTRFS_BLOCK_GROUP_METADATA
);
4716 void btrfs_free_block_rsv(struct btrfs_root
*root
,
4717 struct btrfs_block_rsv
*rsv
)
4721 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
4725 int btrfs_block_rsv_add(struct btrfs_root
*root
,
4726 struct btrfs_block_rsv
*block_rsv
, u64 num_bytes
,
4727 enum btrfs_reserve_flush_enum flush
)
4734 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
4736 block_rsv_add_bytes(block_rsv
, num_bytes
, 1);
4743 int btrfs_block_rsv_check(struct btrfs_root
*root
,
4744 struct btrfs_block_rsv
*block_rsv
, int min_factor
)
4752 spin_lock(&block_rsv
->lock
);
4753 num_bytes
= div_factor(block_rsv
->size
, min_factor
);
4754 if (block_rsv
->reserved
>= num_bytes
)
4756 spin_unlock(&block_rsv
->lock
);
4761 int btrfs_block_rsv_refill(struct btrfs_root
*root
,
4762 struct btrfs_block_rsv
*block_rsv
, u64 min_reserved
,
4763 enum btrfs_reserve_flush_enum flush
)
4771 spin_lock(&block_rsv
->lock
);
4772 num_bytes
= min_reserved
;
4773 if (block_rsv
->reserved
>= num_bytes
)
4776 num_bytes
-= block_rsv
->reserved
;
4777 spin_unlock(&block_rsv
->lock
);
4782 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
4784 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
4791 int btrfs_block_rsv_migrate(struct btrfs_block_rsv
*src_rsv
,
4792 struct btrfs_block_rsv
*dst_rsv
,
4795 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
4798 void btrfs_block_rsv_release(struct btrfs_root
*root
,
4799 struct btrfs_block_rsv
*block_rsv
,
4802 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
4803 if (global_rsv
== block_rsv
||
4804 block_rsv
->space_info
!= global_rsv
->space_info
)
4806 block_rsv_release_bytes(root
->fs_info
, block_rsv
, global_rsv
,
4811 * helper to calculate size of global block reservation.
4812 * the desired value is sum of space used by extent tree,
4813 * checksum tree and root tree
4815 static u64
calc_global_metadata_size(struct btrfs_fs_info
*fs_info
)
4817 struct btrfs_space_info
*sinfo
;
4821 int csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
4823 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_DATA
);
4824 spin_lock(&sinfo
->lock
);
4825 data_used
= sinfo
->bytes_used
;
4826 spin_unlock(&sinfo
->lock
);
4828 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4829 spin_lock(&sinfo
->lock
);
4830 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_DATA
)
4832 meta_used
= sinfo
->bytes_used
;
4833 spin_unlock(&sinfo
->lock
);
4835 num_bytes
= (data_used
>> fs_info
->sb
->s_blocksize_bits
) *
4837 num_bytes
+= div64_u64(data_used
+ meta_used
, 50);
4839 if (num_bytes
* 3 > meta_used
)
4840 num_bytes
= div64_u64(meta_used
, 3);
4842 return ALIGN(num_bytes
, fs_info
->extent_root
->leafsize
<< 10);
4845 static void update_global_block_rsv(struct btrfs_fs_info
*fs_info
)
4847 struct btrfs_block_rsv
*block_rsv
= &fs_info
->global_block_rsv
;
4848 struct btrfs_space_info
*sinfo
= block_rsv
->space_info
;
4851 num_bytes
= calc_global_metadata_size(fs_info
);
4853 spin_lock(&sinfo
->lock
);
4854 spin_lock(&block_rsv
->lock
);
4856 block_rsv
->size
= min_t(u64
, num_bytes
, 512 * 1024 * 1024);
4858 num_bytes
= sinfo
->bytes_used
+ sinfo
->bytes_pinned
+
4859 sinfo
->bytes_reserved
+ sinfo
->bytes_readonly
+
4860 sinfo
->bytes_may_use
;
4862 if (sinfo
->total_bytes
> num_bytes
) {
4863 num_bytes
= sinfo
->total_bytes
- num_bytes
;
4864 block_rsv
->reserved
+= num_bytes
;
4865 sinfo
->bytes_may_use
+= num_bytes
;
4866 trace_btrfs_space_reservation(fs_info
, "space_info",
4867 sinfo
->flags
, num_bytes
, 1);
4870 if (block_rsv
->reserved
>= block_rsv
->size
) {
4871 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
4872 sinfo
->bytes_may_use
-= num_bytes
;
4873 trace_btrfs_space_reservation(fs_info
, "space_info",
4874 sinfo
->flags
, num_bytes
, 0);
4875 block_rsv
->reserved
= block_rsv
->size
;
4876 block_rsv
->full
= 1;
4879 spin_unlock(&block_rsv
->lock
);
4880 spin_unlock(&sinfo
->lock
);
4883 static void init_global_block_rsv(struct btrfs_fs_info
*fs_info
)
4885 struct btrfs_space_info
*space_info
;
4887 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
4888 fs_info
->chunk_block_rsv
.space_info
= space_info
;
4890 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4891 fs_info
->global_block_rsv
.space_info
= space_info
;
4892 fs_info
->delalloc_block_rsv
.space_info
= space_info
;
4893 fs_info
->trans_block_rsv
.space_info
= space_info
;
4894 fs_info
->empty_block_rsv
.space_info
= space_info
;
4895 fs_info
->delayed_block_rsv
.space_info
= space_info
;
4897 fs_info
->extent_root
->block_rsv
= &fs_info
->global_block_rsv
;
4898 fs_info
->csum_root
->block_rsv
= &fs_info
->global_block_rsv
;
4899 fs_info
->dev_root
->block_rsv
= &fs_info
->global_block_rsv
;
4900 fs_info
->tree_root
->block_rsv
= &fs_info
->global_block_rsv
;
4901 if (fs_info
->quota_root
)
4902 fs_info
->quota_root
->block_rsv
= &fs_info
->global_block_rsv
;
4903 fs_info
->chunk_root
->block_rsv
= &fs_info
->chunk_block_rsv
;
4905 update_global_block_rsv(fs_info
);
4908 static void release_global_block_rsv(struct btrfs_fs_info
*fs_info
)
4910 block_rsv_release_bytes(fs_info
, &fs_info
->global_block_rsv
, NULL
,
4912 WARN_ON(fs_info
->delalloc_block_rsv
.size
> 0);
4913 WARN_ON(fs_info
->delalloc_block_rsv
.reserved
> 0);
4914 WARN_ON(fs_info
->trans_block_rsv
.size
> 0);
4915 WARN_ON(fs_info
->trans_block_rsv
.reserved
> 0);
4916 WARN_ON(fs_info
->chunk_block_rsv
.size
> 0);
4917 WARN_ON(fs_info
->chunk_block_rsv
.reserved
> 0);
4918 WARN_ON(fs_info
->delayed_block_rsv
.size
> 0);
4919 WARN_ON(fs_info
->delayed_block_rsv
.reserved
> 0);
4922 void btrfs_trans_release_metadata(struct btrfs_trans_handle
*trans
,
4923 struct btrfs_root
*root
)
4925 if (!trans
->block_rsv
)
4928 if (!trans
->bytes_reserved
)
4931 trace_btrfs_space_reservation(root
->fs_info
, "transaction",
4932 trans
->transid
, trans
->bytes_reserved
, 0);
4933 btrfs_block_rsv_release(root
, trans
->block_rsv
, trans
->bytes_reserved
);
4934 trans
->bytes_reserved
= 0;
4937 /* Can only return 0 or -ENOSPC */
4938 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle
*trans
,
4939 struct inode
*inode
)
4941 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4942 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
4943 struct btrfs_block_rsv
*dst_rsv
= root
->orphan_block_rsv
;
4946 * We need to hold space in order to delete our orphan item once we've
4947 * added it, so this takes the reservation so we can release it later
4948 * when we are truly done with the orphan item.
4950 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
4951 trace_btrfs_space_reservation(root
->fs_info
, "orphan",
4952 btrfs_ino(inode
), num_bytes
, 1);
4953 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
4956 void btrfs_orphan_release_metadata(struct inode
*inode
)
4958 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4959 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
4960 trace_btrfs_space_reservation(root
->fs_info
, "orphan",
4961 btrfs_ino(inode
), num_bytes
, 0);
4962 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
, num_bytes
);
4966 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4967 * root: the root of the parent directory
4968 * rsv: block reservation
4969 * items: the number of items that we need do reservation
4970 * qgroup_reserved: used to return the reserved size in qgroup
4972 * This function is used to reserve the space for snapshot/subvolume
4973 * creation and deletion. Those operations are different with the
4974 * common file/directory operations, they change two fs/file trees
4975 * and root tree, the number of items that the qgroup reserves is
4976 * different with the free space reservation. So we can not use
4977 * the space reseravtion mechanism in start_transaction().
4979 int btrfs_subvolume_reserve_metadata(struct btrfs_root
*root
,
4980 struct btrfs_block_rsv
*rsv
,
4982 u64
*qgroup_reserved
,
4983 bool use_global_rsv
)
4987 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
4989 if (root
->fs_info
->quota_enabled
) {
4990 /* One for parent inode, two for dir entries */
4991 num_bytes
= 3 * root
->leafsize
;
4992 ret
= btrfs_qgroup_reserve(root
, num_bytes
);
4999 *qgroup_reserved
= num_bytes
;
5001 num_bytes
= btrfs_calc_trans_metadata_size(root
, items
);
5002 rsv
->space_info
= __find_space_info(root
->fs_info
,
5003 BTRFS_BLOCK_GROUP_METADATA
);
5004 ret
= btrfs_block_rsv_add(root
, rsv
, num_bytes
,
5005 BTRFS_RESERVE_FLUSH_ALL
);
5007 if (ret
== -ENOSPC
&& use_global_rsv
)
5008 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
, num_bytes
);
5011 if (*qgroup_reserved
)
5012 btrfs_qgroup_free(root
, *qgroup_reserved
);
5018 void btrfs_subvolume_release_metadata(struct btrfs_root
*root
,
5019 struct btrfs_block_rsv
*rsv
,
5020 u64 qgroup_reserved
)
5022 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
5023 if (qgroup_reserved
)
5024 btrfs_qgroup_free(root
, qgroup_reserved
);
5028 * drop_outstanding_extent - drop an outstanding extent
5029 * @inode: the inode we're dropping the extent for
5031 * This is called when we are freeing up an outstanding extent, either called
5032 * after an error or after an extent is written. This will return the number of
5033 * reserved extents that need to be freed. This must be called with
5034 * BTRFS_I(inode)->lock held.
5036 static unsigned drop_outstanding_extent(struct inode
*inode
)
5038 unsigned drop_inode_space
= 0;
5039 unsigned dropped_extents
= 0;
5041 BUG_ON(!BTRFS_I(inode
)->outstanding_extents
);
5042 BTRFS_I(inode
)->outstanding_extents
--;
5044 if (BTRFS_I(inode
)->outstanding_extents
== 0 &&
5045 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5046 &BTRFS_I(inode
)->runtime_flags
))
5047 drop_inode_space
= 1;
5050 * If we have more or the same amount of outsanding extents than we have
5051 * reserved then we need to leave the reserved extents count alone.
5053 if (BTRFS_I(inode
)->outstanding_extents
>=
5054 BTRFS_I(inode
)->reserved_extents
)
5055 return drop_inode_space
;
5057 dropped_extents
= BTRFS_I(inode
)->reserved_extents
-
5058 BTRFS_I(inode
)->outstanding_extents
;
5059 BTRFS_I(inode
)->reserved_extents
-= dropped_extents
;
5060 return dropped_extents
+ drop_inode_space
;
5064 * calc_csum_metadata_size - return the amount of metada space that must be
5065 * reserved/free'd for the given bytes.
5066 * @inode: the inode we're manipulating
5067 * @num_bytes: the number of bytes in question
5068 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5070 * This adjusts the number of csum_bytes in the inode and then returns the
5071 * correct amount of metadata that must either be reserved or freed. We
5072 * calculate how many checksums we can fit into one leaf and then divide the
5073 * number of bytes that will need to be checksumed by this value to figure out
5074 * how many checksums will be required. If we are adding bytes then the number
5075 * may go up and we will return the number of additional bytes that must be
5076 * reserved. If it is going down we will return the number of bytes that must
5079 * This must be called with BTRFS_I(inode)->lock held.
5081 static u64
calc_csum_metadata_size(struct inode
*inode
, u64 num_bytes
,
5084 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5086 int num_csums_per_leaf
;
5090 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
&&
5091 BTRFS_I(inode
)->csum_bytes
== 0)
5094 old_csums
= (int)div64_u64(BTRFS_I(inode
)->csum_bytes
, root
->sectorsize
);
5096 BTRFS_I(inode
)->csum_bytes
+= num_bytes
;
5098 BTRFS_I(inode
)->csum_bytes
-= num_bytes
;
5099 csum_size
= BTRFS_LEAF_DATA_SIZE(root
) - sizeof(struct btrfs_item
);
5100 num_csums_per_leaf
= (int)div64_u64(csum_size
,
5101 sizeof(struct btrfs_csum_item
) +
5102 sizeof(struct btrfs_disk_key
));
5103 num_csums
= (int)div64_u64(BTRFS_I(inode
)->csum_bytes
, root
->sectorsize
);
5104 num_csums
= num_csums
+ num_csums_per_leaf
- 1;
5105 num_csums
= num_csums
/ num_csums_per_leaf
;
5107 old_csums
= old_csums
+ num_csums_per_leaf
- 1;
5108 old_csums
= old_csums
/ num_csums_per_leaf
;
5110 /* No change, no need to reserve more */
5111 if (old_csums
== num_csums
)
5115 return btrfs_calc_trans_metadata_size(root
,
5116 num_csums
- old_csums
);
5118 return btrfs_calc_trans_metadata_size(root
, old_csums
- num_csums
);
5121 int btrfs_delalloc_reserve_metadata(struct inode
*inode
, u64 num_bytes
)
5123 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5124 struct btrfs_block_rsv
*block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
5127 unsigned nr_extents
= 0;
5128 int extra_reserve
= 0;
5129 enum btrfs_reserve_flush_enum flush
= BTRFS_RESERVE_FLUSH_ALL
;
5131 bool delalloc_lock
= true;
5135 /* If we are a free space inode we need to not flush since we will be in
5136 * the middle of a transaction commit. We also don't need the delalloc
5137 * mutex since we won't race with anybody. We need this mostly to make
5138 * lockdep shut its filthy mouth.
5140 if (btrfs_is_free_space_inode(inode
)) {
5141 flush
= BTRFS_RESERVE_NO_FLUSH
;
5142 delalloc_lock
= false;
5145 if (flush
!= BTRFS_RESERVE_NO_FLUSH
&&
5146 btrfs_transaction_in_commit(root
->fs_info
))
5147 schedule_timeout(1);
5150 mutex_lock(&BTRFS_I(inode
)->delalloc_mutex
);
5152 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
5154 spin_lock(&BTRFS_I(inode
)->lock
);
5155 BTRFS_I(inode
)->outstanding_extents
++;
5157 if (BTRFS_I(inode
)->outstanding_extents
>
5158 BTRFS_I(inode
)->reserved_extents
)
5159 nr_extents
= BTRFS_I(inode
)->outstanding_extents
-
5160 BTRFS_I(inode
)->reserved_extents
;
5163 * Add an item to reserve for updating the inode when we complete the
5166 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5167 &BTRFS_I(inode
)->runtime_flags
)) {
5172 to_reserve
= btrfs_calc_trans_metadata_size(root
, nr_extents
);
5173 to_reserve
+= calc_csum_metadata_size(inode
, num_bytes
, 1);
5174 csum_bytes
= BTRFS_I(inode
)->csum_bytes
;
5175 spin_unlock(&BTRFS_I(inode
)->lock
);
5177 if (root
->fs_info
->quota_enabled
) {
5178 ret
= btrfs_qgroup_reserve(root
, num_bytes
+
5179 nr_extents
* root
->leafsize
);
5184 ret
= reserve_metadata_bytes(root
, block_rsv
, to_reserve
, flush
);
5185 if (unlikely(ret
)) {
5186 if (root
->fs_info
->quota_enabled
)
5187 btrfs_qgroup_free(root
, num_bytes
+
5188 nr_extents
* root
->leafsize
);
5192 spin_lock(&BTRFS_I(inode
)->lock
);
5193 if (extra_reserve
) {
5194 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5195 &BTRFS_I(inode
)->runtime_flags
);
5198 BTRFS_I(inode
)->reserved_extents
+= nr_extents
;
5199 spin_unlock(&BTRFS_I(inode
)->lock
);
5202 mutex_unlock(&BTRFS_I(inode
)->delalloc_mutex
);
5205 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5206 btrfs_ino(inode
), to_reserve
, 1);
5207 block_rsv_add_bytes(block_rsv
, to_reserve
, 1);
5212 spin_lock(&BTRFS_I(inode
)->lock
);
5213 dropped
= drop_outstanding_extent(inode
);
5215 * If the inodes csum_bytes is the same as the original
5216 * csum_bytes then we know we haven't raced with any free()ers
5217 * so we can just reduce our inodes csum bytes and carry on.
5219 if (BTRFS_I(inode
)->csum_bytes
== csum_bytes
) {
5220 calc_csum_metadata_size(inode
, num_bytes
, 0);
5222 u64 orig_csum_bytes
= BTRFS_I(inode
)->csum_bytes
;
5226 * This is tricky, but first we need to figure out how much we
5227 * free'd from any free-ers that occured during this
5228 * reservation, so we reset ->csum_bytes to the csum_bytes
5229 * before we dropped our lock, and then call the free for the
5230 * number of bytes that were freed while we were trying our
5233 bytes
= csum_bytes
- BTRFS_I(inode
)->csum_bytes
;
5234 BTRFS_I(inode
)->csum_bytes
= csum_bytes
;
5235 to_free
= calc_csum_metadata_size(inode
, bytes
, 0);
5239 * Now we need to see how much we would have freed had we not
5240 * been making this reservation and our ->csum_bytes were not
5241 * artificially inflated.
5243 BTRFS_I(inode
)->csum_bytes
= csum_bytes
- num_bytes
;
5244 bytes
= csum_bytes
- orig_csum_bytes
;
5245 bytes
= calc_csum_metadata_size(inode
, bytes
, 0);
5248 * Now reset ->csum_bytes to what it should be. If bytes is
5249 * more than to_free then we would have free'd more space had we
5250 * not had an artificially high ->csum_bytes, so we need to free
5251 * the remainder. If bytes is the same or less then we don't
5252 * need to do anything, the other free-ers did the correct
5255 BTRFS_I(inode
)->csum_bytes
= orig_csum_bytes
- num_bytes
;
5256 if (bytes
> to_free
)
5257 to_free
= bytes
- to_free
;
5261 spin_unlock(&BTRFS_I(inode
)->lock
);
5263 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
5266 btrfs_block_rsv_release(root
, block_rsv
, to_free
);
5267 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5268 btrfs_ino(inode
), to_free
, 0);
5271 mutex_unlock(&BTRFS_I(inode
)->delalloc_mutex
);
5276 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5277 * @inode: the inode to release the reservation for
5278 * @num_bytes: the number of bytes we're releasing
5280 * This will release the metadata reservation for an inode. This can be called
5281 * once we complete IO for a given set of bytes to release their metadata
5284 void btrfs_delalloc_release_metadata(struct inode
*inode
, u64 num_bytes
)
5286 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5290 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
5291 spin_lock(&BTRFS_I(inode
)->lock
);
5292 dropped
= drop_outstanding_extent(inode
);
5295 to_free
= calc_csum_metadata_size(inode
, num_bytes
, 0);
5296 spin_unlock(&BTRFS_I(inode
)->lock
);
5298 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
5300 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5301 btrfs_ino(inode
), to_free
, 0);
5302 if (root
->fs_info
->quota_enabled
) {
5303 btrfs_qgroup_free(root
, num_bytes
+
5304 dropped
* root
->leafsize
);
5307 btrfs_block_rsv_release(root
, &root
->fs_info
->delalloc_block_rsv
,
5312 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5313 * @inode: inode we're writing to
5314 * @num_bytes: the number of bytes we want to allocate
5316 * This will do the following things
5318 * o reserve space in the data space info for num_bytes
5319 * o reserve space in the metadata space info based on number of outstanding
5320 * extents and how much csums will be needed
5321 * o add to the inodes ->delalloc_bytes
5322 * o add it to the fs_info's delalloc inodes list.
5324 * This will return 0 for success and -ENOSPC if there is no space left.
5326 int btrfs_delalloc_reserve_space(struct inode
*inode
, u64 num_bytes
)
5330 ret
= btrfs_check_data_free_space(inode
, num_bytes
);
5334 ret
= btrfs_delalloc_reserve_metadata(inode
, num_bytes
);
5336 btrfs_free_reserved_data_space(inode
, num_bytes
);
5344 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5345 * @inode: inode we're releasing space for
5346 * @num_bytes: the number of bytes we want to free up
5348 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5349 * called in the case that we don't need the metadata AND data reservations
5350 * anymore. So if there is an error or we insert an inline extent.
5352 * This function will release the metadata space that was not used and will
5353 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5354 * list if there are no delalloc bytes left.
5356 void btrfs_delalloc_release_space(struct inode
*inode
, u64 num_bytes
)
5358 btrfs_delalloc_release_metadata(inode
, num_bytes
);
5359 btrfs_free_reserved_data_space(inode
, num_bytes
);
5362 static int update_block_group(struct btrfs_root
*root
,
5363 u64 bytenr
, u64 num_bytes
, int alloc
)
5365 struct btrfs_block_group_cache
*cache
= NULL
;
5366 struct btrfs_fs_info
*info
= root
->fs_info
;
5367 u64 total
= num_bytes
;
5372 /* block accounting for super block */
5373 spin_lock(&info
->delalloc_root_lock
);
5374 old_val
= btrfs_super_bytes_used(info
->super_copy
);
5376 old_val
+= num_bytes
;
5378 old_val
-= num_bytes
;
5379 btrfs_set_super_bytes_used(info
->super_copy
, old_val
);
5380 spin_unlock(&info
->delalloc_root_lock
);
5383 cache
= btrfs_lookup_block_group(info
, bytenr
);
5386 if (cache
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
5387 BTRFS_BLOCK_GROUP_RAID1
|
5388 BTRFS_BLOCK_GROUP_RAID10
))
5393 * If this block group has free space cache written out, we
5394 * need to make sure to load it if we are removing space. This
5395 * is because we need the unpinning stage to actually add the
5396 * space back to the block group, otherwise we will leak space.
5398 if (!alloc
&& cache
->cached
== BTRFS_CACHE_NO
)
5399 cache_block_group(cache
, 1);
5401 byte_in_group
= bytenr
- cache
->key
.objectid
;
5402 WARN_ON(byte_in_group
> cache
->key
.offset
);
5404 spin_lock(&cache
->space_info
->lock
);
5405 spin_lock(&cache
->lock
);
5407 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
5408 cache
->disk_cache_state
< BTRFS_DC_CLEAR
)
5409 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
5412 old_val
= btrfs_block_group_used(&cache
->item
);
5413 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
5415 old_val
+= num_bytes
;
5416 btrfs_set_block_group_used(&cache
->item
, old_val
);
5417 cache
->reserved
-= num_bytes
;
5418 cache
->space_info
->bytes_reserved
-= num_bytes
;
5419 cache
->space_info
->bytes_used
+= num_bytes
;
5420 cache
->space_info
->disk_used
+= num_bytes
* factor
;
5421 spin_unlock(&cache
->lock
);
5422 spin_unlock(&cache
->space_info
->lock
);
5424 old_val
-= num_bytes
;
5425 btrfs_set_block_group_used(&cache
->item
, old_val
);
5426 cache
->pinned
+= num_bytes
;
5427 cache
->space_info
->bytes_pinned
+= num_bytes
;
5428 cache
->space_info
->bytes_used
-= num_bytes
;
5429 cache
->space_info
->disk_used
-= num_bytes
* factor
;
5430 spin_unlock(&cache
->lock
);
5431 spin_unlock(&cache
->space_info
->lock
);
5433 set_extent_dirty(info
->pinned_extents
,
5434 bytenr
, bytenr
+ num_bytes
- 1,
5435 GFP_NOFS
| __GFP_NOFAIL
);
5437 btrfs_put_block_group(cache
);
5439 bytenr
+= num_bytes
;
5444 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
5446 struct btrfs_block_group_cache
*cache
;
5449 spin_lock(&root
->fs_info
->block_group_cache_lock
);
5450 bytenr
= root
->fs_info
->first_logical_byte
;
5451 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
5453 if (bytenr
< (u64
)-1)
5456 cache
= btrfs_lookup_first_block_group(root
->fs_info
, search_start
);
5460 bytenr
= cache
->key
.objectid
;
5461 btrfs_put_block_group(cache
);
5466 static int pin_down_extent(struct btrfs_root
*root
,
5467 struct btrfs_block_group_cache
*cache
,
5468 u64 bytenr
, u64 num_bytes
, int reserved
)
5470 spin_lock(&cache
->space_info
->lock
);
5471 spin_lock(&cache
->lock
);
5472 cache
->pinned
+= num_bytes
;
5473 cache
->space_info
->bytes_pinned
+= num_bytes
;
5475 cache
->reserved
-= num_bytes
;
5476 cache
->space_info
->bytes_reserved
-= num_bytes
;
5478 spin_unlock(&cache
->lock
);
5479 spin_unlock(&cache
->space_info
->lock
);
5481 set_extent_dirty(root
->fs_info
->pinned_extents
, bytenr
,
5482 bytenr
+ num_bytes
- 1, GFP_NOFS
| __GFP_NOFAIL
);
5484 trace_btrfs_reserved_extent_free(root
, bytenr
, num_bytes
);
5489 * this function must be called within transaction
5491 int btrfs_pin_extent(struct btrfs_root
*root
,
5492 u64 bytenr
, u64 num_bytes
, int reserved
)
5494 struct btrfs_block_group_cache
*cache
;
5496 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
5497 BUG_ON(!cache
); /* Logic error */
5499 pin_down_extent(root
, cache
, bytenr
, num_bytes
, reserved
);
5501 btrfs_put_block_group(cache
);
5506 * this function must be called within transaction
5508 int btrfs_pin_extent_for_log_replay(struct btrfs_root
*root
,
5509 u64 bytenr
, u64 num_bytes
)
5511 struct btrfs_block_group_cache
*cache
;
5514 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
5519 * pull in the free space cache (if any) so that our pin
5520 * removes the free space from the cache. We have load_only set
5521 * to one because the slow code to read in the free extents does check
5522 * the pinned extents.
5524 cache_block_group(cache
, 1);
5526 pin_down_extent(root
, cache
, bytenr
, num_bytes
, 0);
5528 /* remove us from the free space cache (if we're there at all) */
5529 ret
= btrfs_remove_free_space(cache
, bytenr
, num_bytes
);
5530 btrfs_put_block_group(cache
);
5534 static int __exclude_logged_extent(struct btrfs_root
*root
, u64 start
, u64 num_bytes
)
5537 struct btrfs_block_group_cache
*block_group
;
5538 struct btrfs_caching_control
*caching_ctl
;
5540 block_group
= btrfs_lookup_block_group(root
->fs_info
, start
);
5544 cache_block_group(block_group
, 0);
5545 caching_ctl
= get_caching_control(block_group
);
5549 BUG_ON(!block_group_cache_done(block_group
));
5550 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
5552 mutex_lock(&caching_ctl
->mutex
);
5554 if (start
>= caching_ctl
->progress
) {
5555 ret
= add_excluded_extent(root
, start
, num_bytes
);
5556 } else if (start
+ num_bytes
<= caching_ctl
->progress
) {
5557 ret
= btrfs_remove_free_space(block_group
,
5560 num_bytes
= caching_ctl
->progress
- start
;
5561 ret
= btrfs_remove_free_space(block_group
,
5566 num_bytes
= (start
+ num_bytes
) -
5567 caching_ctl
->progress
;
5568 start
= caching_ctl
->progress
;
5569 ret
= add_excluded_extent(root
, start
, num_bytes
);
5572 mutex_unlock(&caching_ctl
->mutex
);
5573 put_caching_control(caching_ctl
);
5575 btrfs_put_block_group(block_group
);
5579 int btrfs_exclude_logged_extents(struct btrfs_root
*log
,
5580 struct extent_buffer
*eb
)
5582 struct btrfs_file_extent_item
*item
;
5583 struct btrfs_key key
;
5587 if (!btrfs_fs_incompat(log
->fs_info
, MIXED_GROUPS
))
5590 for (i
= 0; i
< btrfs_header_nritems(eb
); i
++) {
5591 btrfs_item_key_to_cpu(eb
, &key
, i
);
5592 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
5594 item
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
5595 found_type
= btrfs_file_extent_type(eb
, item
);
5596 if (found_type
== BTRFS_FILE_EXTENT_INLINE
)
5598 if (btrfs_file_extent_disk_bytenr(eb
, item
) == 0)
5600 key
.objectid
= btrfs_file_extent_disk_bytenr(eb
, item
);
5601 key
.offset
= btrfs_file_extent_disk_num_bytes(eb
, item
);
5602 __exclude_logged_extent(log
, key
.objectid
, key
.offset
);
5609 * btrfs_update_reserved_bytes - update the block_group and space info counters
5610 * @cache: The cache we are manipulating
5611 * @num_bytes: The number of bytes in question
5612 * @reserve: One of the reservation enums
5613 * @delalloc: The blocks are allocated for the delalloc write
5615 * This is called by the allocator when it reserves space, or by somebody who is
5616 * freeing space that was never actually used on disk. For example if you
5617 * reserve some space for a new leaf in transaction A and before transaction A
5618 * commits you free that leaf, you call this with reserve set to 0 in order to
5619 * clear the reservation.
5621 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5622 * ENOSPC accounting. For data we handle the reservation through clearing the
5623 * delalloc bits in the io_tree. We have to do this since we could end up
5624 * allocating less disk space for the amount of data we have reserved in the
5625 * case of compression.
5627 * If this is a reservation and the block group has become read only we cannot
5628 * make the reservation and return -EAGAIN, otherwise this function always
5631 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
5632 u64 num_bytes
, int reserve
, int delalloc
)
5634 struct btrfs_space_info
*space_info
= cache
->space_info
;
5637 spin_lock(&space_info
->lock
);
5638 spin_lock(&cache
->lock
);
5639 if (reserve
!= RESERVE_FREE
) {
5643 cache
->reserved
+= num_bytes
;
5644 space_info
->bytes_reserved
+= num_bytes
;
5645 if (reserve
== RESERVE_ALLOC
) {
5646 trace_btrfs_space_reservation(cache
->fs_info
,
5647 "space_info", space_info
->flags
,
5649 space_info
->bytes_may_use
-= num_bytes
;
5653 cache
->delalloc_bytes
+= num_bytes
;
5657 space_info
->bytes_readonly
+= num_bytes
;
5658 cache
->reserved
-= num_bytes
;
5659 space_info
->bytes_reserved
-= num_bytes
;
5662 cache
->delalloc_bytes
-= num_bytes
;
5664 spin_unlock(&cache
->lock
);
5665 spin_unlock(&space_info
->lock
);
5669 void btrfs_prepare_extent_commit(struct btrfs_trans_handle
*trans
,
5670 struct btrfs_root
*root
)
5672 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5673 struct btrfs_caching_control
*next
;
5674 struct btrfs_caching_control
*caching_ctl
;
5675 struct btrfs_block_group_cache
*cache
;
5677 down_write(&fs_info
->commit_root_sem
);
5679 list_for_each_entry_safe(caching_ctl
, next
,
5680 &fs_info
->caching_block_groups
, list
) {
5681 cache
= caching_ctl
->block_group
;
5682 if (block_group_cache_done(cache
)) {
5683 cache
->last_byte_to_unpin
= (u64
)-1;
5684 list_del_init(&caching_ctl
->list
);
5685 put_caching_control(caching_ctl
);
5687 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
5691 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
5692 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
5694 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
5696 up_write(&fs_info
->commit_root_sem
);
5698 update_global_block_rsv(fs_info
);
5701 static int unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
5703 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5704 struct btrfs_block_group_cache
*cache
= NULL
;
5705 struct btrfs_space_info
*space_info
;
5706 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5710 while (start
<= end
) {
5713 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
5715 btrfs_put_block_group(cache
);
5716 cache
= btrfs_lookup_block_group(fs_info
, start
);
5717 BUG_ON(!cache
); /* Logic error */
5720 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
5721 len
= min(len
, end
+ 1 - start
);
5723 if (start
< cache
->last_byte_to_unpin
) {
5724 len
= min(len
, cache
->last_byte_to_unpin
- start
);
5725 btrfs_add_free_space(cache
, start
, len
);
5729 space_info
= cache
->space_info
;
5731 spin_lock(&space_info
->lock
);
5732 spin_lock(&cache
->lock
);
5733 cache
->pinned
-= len
;
5734 space_info
->bytes_pinned
-= len
;
5735 percpu_counter_add(&space_info
->total_bytes_pinned
, -len
);
5737 space_info
->bytes_readonly
+= len
;
5740 spin_unlock(&cache
->lock
);
5741 if (!readonly
&& global_rsv
->space_info
== space_info
) {
5742 spin_lock(&global_rsv
->lock
);
5743 if (!global_rsv
->full
) {
5744 len
= min(len
, global_rsv
->size
-
5745 global_rsv
->reserved
);
5746 global_rsv
->reserved
+= len
;
5747 space_info
->bytes_may_use
+= len
;
5748 if (global_rsv
->reserved
>= global_rsv
->size
)
5749 global_rsv
->full
= 1;
5751 spin_unlock(&global_rsv
->lock
);
5753 spin_unlock(&space_info
->lock
);
5757 btrfs_put_block_group(cache
);
5761 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
5762 struct btrfs_root
*root
)
5764 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5765 struct extent_io_tree
*unpin
;
5773 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
5774 unpin
= &fs_info
->freed_extents
[1];
5776 unpin
= &fs_info
->freed_extents
[0];
5779 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
5780 EXTENT_DIRTY
, NULL
);
5784 if (btrfs_test_opt(root
, DISCARD
))
5785 ret
= btrfs_discard_extent(root
, start
,
5786 end
+ 1 - start
, NULL
);
5788 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
5789 unpin_extent_range(root
, start
, end
);
5796 static void add_pinned_bytes(struct btrfs_fs_info
*fs_info
, u64 num_bytes
,
5797 u64 owner
, u64 root_objectid
)
5799 struct btrfs_space_info
*space_info
;
5802 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
5803 if (root_objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
5804 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
5806 flags
= BTRFS_BLOCK_GROUP_METADATA
;
5808 flags
= BTRFS_BLOCK_GROUP_DATA
;
5811 space_info
= __find_space_info(fs_info
, flags
);
5812 BUG_ON(!space_info
); /* Logic bug */
5813 percpu_counter_add(&space_info
->total_bytes_pinned
, num_bytes
);
5817 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
5818 struct btrfs_root
*root
,
5819 u64 bytenr
, u64 num_bytes
, u64 parent
,
5820 u64 root_objectid
, u64 owner_objectid
,
5821 u64 owner_offset
, int refs_to_drop
,
5822 struct btrfs_delayed_extent_op
*extent_op
,
5825 struct btrfs_key key
;
5826 struct btrfs_path
*path
;
5827 struct btrfs_fs_info
*info
= root
->fs_info
;
5828 struct btrfs_root
*extent_root
= info
->extent_root
;
5829 struct extent_buffer
*leaf
;
5830 struct btrfs_extent_item
*ei
;
5831 struct btrfs_extent_inline_ref
*iref
;
5834 int extent_slot
= 0;
5835 int found_extent
= 0;
5840 enum btrfs_qgroup_operation_type type
= BTRFS_QGROUP_OPER_SUB_EXCL
;
5841 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
5844 if (!info
->quota_enabled
|| !is_fstree(root_objectid
))
5847 path
= btrfs_alloc_path();
5852 path
->leave_spinning
= 1;
5854 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
5855 BUG_ON(!is_data
&& refs_to_drop
!= 1);
5858 skinny_metadata
= 0;
5860 ret
= lookup_extent_backref(trans
, extent_root
, path
, &iref
,
5861 bytenr
, num_bytes
, parent
,
5862 root_objectid
, owner_objectid
,
5865 extent_slot
= path
->slots
[0];
5866 while (extent_slot
>= 0) {
5867 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
5869 if (key
.objectid
!= bytenr
)
5871 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
5872 key
.offset
== num_bytes
) {
5876 if (key
.type
== BTRFS_METADATA_ITEM_KEY
&&
5877 key
.offset
== owner_objectid
) {
5881 if (path
->slots
[0] - extent_slot
> 5)
5885 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5886 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
5887 if (found_extent
&& item_size
< sizeof(*ei
))
5890 if (!found_extent
) {
5892 ret
= remove_extent_backref(trans
, extent_root
, path
,
5894 is_data
, &last_ref
);
5896 btrfs_abort_transaction(trans
, extent_root
, ret
);
5899 btrfs_release_path(path
);
5900 path
->leave_spinning
= 1;
5902 key
.objectid
= bytenr
;
5903 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
5904 key
.offset
= num_bytes
;
5906 if (!is_data
&& skinny_metadata
) {
5907 key
.type
= BTRFS_METADATA_ITEM_KEY
;
5908 key
.offset
= owner_objectid
;
5911 ret
= btrfs_search_slot(trans
, extent_root
,
5913 if (ret
> 0 && skinny_metadata
&& path
->slots
[0]) {
5915 * Couldn't find our skinny metadata item,
5916 * see if we have ye olde extent item.
5919 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
5921 if (key
.objectid
== bytenr
&&
5922 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
5923 key
.offset
== num_bytes
)
5927 if (ret
> 0 && skinny_metadata
) {
5928 skinny_metadata
= false;
5929 key
.objectid
= bytenr
;
5930 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
5931 key
.offset
= num_bytes
;
5932 btrfs_release_path(path
);
5933 ret
= btrfs_search_slot(trans
, extent_root
,
5938 btrfs_err(info
, "umm, got %d back from search, was looking for %llu",
5941 btrfs_print_leaf(extent_root
,
5945 btrfs_abort_transaction(trans
, extent_root
, ret
);
5948 extent_slot
= path
->slots
[0];
5950 } else if (WARN_ON(ret
== -ENOENT
)) {
5951 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
5953 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
5954 bytenr
, parent
, root_objectid
, owner_objectid
,
5956 btrfs_abort_transaction(trans
, extent_root
, ret
);
5959 btrfs_abort_transaction(trans
, extent_root
, ret
);
5963 leaf
= path
->nodes
[0];
5964 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
5965 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5966 if (item_size
< sizeof(*ei
)) {
5967 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
5968 ret
= convert_extent_item_v0(trans
, extent_root
, path
,
5971 btrfs_abort_transaction(trans
, extent_root
, ret
);
5975 btrfs_release_path(path
);
5976 path
->leave_spinning
= 1;
5978 key
.objectid
= bytenr
;
5979 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
5980 key
.offset
= num_bytes
;
5982 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
5985 btrfs_err(info
, "umm, got %d back from search, was looking for %llu",
5987 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
5990 btrfs_abort_transaction(trans
, extent_root
, ret
);
5994 extent_slot
= path
->slots
[0];
5995 leaf
= path
->nodes
[0];
5996 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
5999 BUG_ON(item_size
< sizeof(*ei
));
6000 ei
= btrfs_item_ptr(leaf
, extent_slot
,
6001 struct btrfs_extent_item
);
6002 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
6003 key
.type
== BTRFS_EXTENT_ITEM_KEY
) {
6004 struct btrfs_tree_block_info
*bi
;
6005 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
6006 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
6007 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
6010 refs
= btrfs_extent_refs(leaf
, ei
);
6011 if (refs
< refs_to_drop
) {
6012 btrfs_err(info
, "trying to drop %d refs but we only have %Lu "
6013 "for bytenr %Lu", refs_to_drop
, refs
, bytenr
);
6015 btrfs_abort_transaction(trans
, extent_root
, ret
);
6018 refs
-= refs_to_drop
;
6021 type
= BTRFS_QGROUP_OPER_SUB_SHARED
;
6023 __run_delayed_extent_op(extent_op
, leaf
, ei
);
6025 * In the case of inline back ref, reference count will
6026 * be updated by remove_extent_backref
6029 BUG_ON(!found_extent
);
6031 btrfs_set_extent_refs(leaf
, ei
, refs
);
6032 btrfs_mark_buffer_dirty(leaf
);
6035 ret
= remove_extent_backref(trans
, extent_root
, path
,
6037 is_data
, &last_ref
);
6039 btrfs_abort_transaction(trans
, extent_root
, ret
);
6043 add_pinned_bytes(root
->fs_info
, -num_bytes
, owner_objectid
,
6047 BUG_ON(is_data
&& refs_to_drop
!=
6048 extent_data_ref_count(root
, path
, iref
));
6050 BUG_ON(path
->slots
[0] != extent_slot
);
6052 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
6053 path
->slots
[0] = extent_slot
;
6059 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
6062 btrfs_abort_transaction(trans
, extent_root
, ret
);
6065 btrfs_release_path(path
);
6068 ret
= btrfs_del_csums(trans
, root
, bytenr
, num_bytes
);
6070 btrfs_abort_transaction(trans
, extent_root
, ret
);
6075 ret
= update_block_group(root
, bytenr
, num_bytes
, 0);
6077 btrfs_abort_transaction(trans
, extent_root
, ret
);
6081 btrfs_release_path(path
);
6083 /* Deal with the quota accounting */
6084 if (!ret
&& last_ref
&& !no_quota
) {
6087 if (owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
&&
6088 type
== BTRFS_QGROUP_OPER_SUB_SHARED
)
6091 ret
= btrfs_qgroup_record_ref(trans
, info
, root_objectid
,
6092 bytenr
, num_bytes
, type
,
6096 btrfs_free_path(path
);
6101 * when we free an block, it is possible (and likely) that we free the last
6102 * delayed ref for that extent as well. This searches the delayed ref tree for
6103 * a given extent, and if there are no other delayed refs to be processed, it
6104 * removes it from the tree.
6106 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
6107 struct btrfs_root
*root
, u64 bytenr
)
6109 struct btrfs_delayed_ref_head
*head
;
6110 struct btrfs_delayed_ref_root
*delayed_refs
;
6113 delayed_refs
= &trans
->transaction
->delayed_refs
;
6114 spin_lock(&delayed_refs
->lock
);
6115 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
6117 goto out_delayed_unlock
;
6119 spin_lock(&head
->lock
);
6120 if (rb_first(&head
->ref_root
))
6123 if (head
->extent_op
) {
6124 if (!head
->must_insert_reserved
)
6126 btrfs_free_delayed_extent_op(head
->extent_op
);
6127 head
->extent_op
= NULL
;
6131 * waiting for the lock here would deadlock. If someone else has it
6132 * locked they are already in the process of dropping it anyway
6134 if (!mutex_trylock(&head
->mutex
))
6138 * at this point we have a head with no other entries. Go
6139 * ahead and process it.
6141 head
->node
.in_tree
= 0;
6142 rb_erase(&head
->href_node
, &delayed_refs
->href_root
);
6144 atomic_dec(&delayed_refs
->num_entries
);
6147 * we don't take a ref on the node because we're removing it from the
6148 * tree, so we just steal the ref the tree was holding.
6150 delayed_refs
->num_heads
--;
6151 if (head
->processing
== 0)
6152 delayed_refs
->num_heads_ready
--;
6153 head
->processing
= 0;
6154 spin_unlock(&head
->lock
);
6155 spin_unlock(&delayed_refs
->lock
);
6157 BUG_ON(head
->extent_op
);
6158 if (head
->must_insert_reserved
)
6161 mutex_unlock(&head
->mutex
);
6162 btrfs_put_delayed_ref(&head
->node
);
6165 spin_unlock(&head
->lock
);
6168 spin_unlock(&delayed_refs
->lock
);
6172 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
6173 struct btrfs_root
*root
,
6174 struct extent_buffer
*buf
,
6175 u64 parent
, int last_ref
)
6177 struct btrfs_block_group_cache
*cache
= NULL
;
6181 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
6182 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
6183 buf
->start
, buf
->len
,
6184 parent
, root
->root_key
.objectid
,
6185 btrfs_header_level(buf
),
6186 BTRFS_DROP_DELAYED_REF
, NULL
, 0);
6187 BUG_ON(ret
); /* -ENOMEM */
6193 cache
= btrfs_lookup_block_group(root
->fs_info
, buf
->start
);
6195 if (btrfs_header_generation(buf
) == trans
->transid
) {
6196 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
6197 ret
= check_ref_cleanup(trans
, root
, buf
->start
);
6202 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
6203 pin_down_extent(root
, cache
, buf
->start
, buf
->len
, 1);
6207 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
6209 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
6210 btrfs_update_reserved_bytes(cache
, buf
->len
, RESERVE_FREE
, 0);
6211 trace_btrfs_reserved_extent_free(root
, buf
->start
, buf
->len
);
6216 add_pinned_bytes(root
->fs_info
, buf
->len
,
6217 btrfs_header_level(buf
),
6218 root
->root_key
.objectid
);
6221 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6224 clear_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
);
6225 btrfs_put_block_group(cache
);
6228 /* Can return -ENOMEM */
6229 int btrfs_free_extent(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
6230 u64 bytenr
, u64 num_bytes
, u64 parent
, u64 root_objectid
,
6231 u64 owner
, u64 offset
, int no_quota
)
6234 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6236 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
6237 if (unlikely(test_bit(BTRFS_ROOT_DUMMY_ROOT
, &root
->state
)))
6240 add_pinned_bytes(root
->fs_info
, num_bytes
, owner
, root_objectid
);
6243 * tree log blocks never actually go into the extent allocation
6244 * tree, just update pinning info and exit early.
6246 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
6247 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
6248 /* unlocks the pinned mutex */
6249 btrfs_pin_extent(root
, bytenr
, num_bytes
, 1);
6251 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
6252 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
6254 parent
, root_objectid
, (int)owner
,
6255 BTRFS_DROP_DELAYED_REF
, NULL
, no_quota
);
6257 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
6259 parent
, root_objectid
, owner
,
6260 offset
, BTRFS_DROP_DELAYED_REF
,
6266 static u64
stripe_align(struct btrfs_root
*root
,
6267 struct btrfs_block_group_cache
*cache
,
6268 u64 val
, u64 num_bytes
)
6270 u64 ret
= ALIGN(val
, root
->stripesize
);
6275 * when we wait for progress in the block group caching, its because
6276 * our allocation attempt failed at least once. So, we must sleep
6277 * and let some progress happen before we try again.
6279 * This function will sleep at least once waiting for new free space to
6280 * show up, and then it will check the block group free space numbers
6281 * for our min num_bytes. Another option is to have it go ahead
6282 * and look in the rbtree for a free extent of a given size, but this
6285 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6286 * any of the information in this block group.
6288 static noinline
void
6289 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
6292 struct btrfs_caching_control
*caching_ctl
;
6294 caching_ctl
= get_caching_control(cache
);
6298 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
6299 (cache
->free_space_ctl
->free_space
>= num_bytes
));
6301 put_caching_control(caching_ctl
);
6305 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
6307 struct btrfs_caching_control
*caching_ctl
;
6310 caching_ctl
= get_caching_control(cache
);
6312 return (cache
->cached
== BTRFS_CACHE_ERROR
) ? -EIO
: 0;
6314 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
6315 if (cache
->cached
== BTRFS_CACHE_ERROR
)
6317 put_caching_control(caching_ctl
);
6321 int __get_raid_index(u64 flags
)
6323 if (flags
& BTRFS_BLOCK_GROUP_RAID10
)
6324 return BTRFS_RAID_RAID10
;
6325 else if (flags
& BTRFS_BLOCK_GROUP_RAID1
)
6326 return BTRFS_RAID_RAID1
;
6327 else if (flags
& BTRFS_BLOCK_GROUP_DUP
)
6328 return BTRFS_RAID_DUP
;
6329 else if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
6330 return BTRFS_RAID_RAID0
;
6331 else if (flags
& BTRFS_BLOCK_GROUP_RAID5
)
6332 return BTRFS_RAID_RAID5
;
6333 else if (flags
& BTRFS_BLOCK_GROUP_RAID6
)
6334 return BTRFS_RAID_RAID6
;
6336 return BTRFS_RAID_SINGLE
; /* BTRFS_BLOCK_GROUP_SINGLE */
6339 int get_block_group_index(struct btrfs_block_group_cache
*cache
)
6341 return __get_raid_index(cache
->flags
);
6344 static const char *btrfs_raid_type_names
[BTRFS_NR_RAID_TYPES
] = {
6345 [BTRFS_RAID_RAID10
] = "raid10",
6346 [BTRFS_RAID_RAID1
] = "raid1",
6347 [BTRFS_RAID_DUP
] = "dup",
6348 [BTRFS_RAID_RAID0
] = "raid0",
6349 [BTRFS_RAID_SINGLE
] = "single",
6350 [BTRFS_RAID_RAID5
] = "raid5",
6351 [BTRFS_RAID_RAID6
] = "raid6",
6354 static const char *get_raid_name(enum btrfs_raid_types type
)
6356 if (type
>= BTRFS_NR_RAID_TYPES
)
6359 return btrfs_raid_type_names
[type
];
6362 enum btrfs_loop_type
{
6363 LOOP_CACHING_NOWAIT
= 0,
6364 LOOP_CACHING_WAIT
= 1,
6365 LOOP_ALLOC_CHUNK
= 2,
6366 LOOP_NO_EMPTY_SIZE
= 3,
6370 btrfs_lock_block_group(struct btrfs_block_group_cache
*cache
,
6374 down_read(&cache
->data_rwsem
);
6378 btrfs_grab_block_group(struct btrfs_block_group_cache
*cache
,
6381 btrfs_get_block_group(cache
);
6383 down_read(&cache
->data_rwsem
);
6386 static struct btrfs_block_group_cache
*
6387 btrfs_lock_cluster(struct btrfs_block_group_cache
*block_group
,
6388 struct btrfs_free_cluster
*cluster
,
6391 struct btrfs_block_group_cache
*used_bg
;
6392 bool locked
= false;
6394 spin_lock(&cluster
->refill_lock
);
6396 if (used_bg
== cluster
->block_group
)
6399 up_read(&used_bg
->data_rwsem
);
6400 btrfs_put_block_group(used_bg
);
6403 used_bg
= cluster
->block_group
;
6407 if (used_bg
== block_group
)
6410 btrfs_get_block_group(used_bg
);
6415 if (down_read_trylock(&used_bg
->data_rwsem
))
6418 spin_unlock(&cluster
->refill_lock
);
6419 down_read(&used_bg
->data_rwsem
);
6425 btrfs_release_block_group(struct btrfs_block_group_cache
*cache
,
6429 up_read(&cache
->data_rwsem
);
6430 btrfs_put_block_group(cache
);
6434 * walks the btree of allocated extents and find a hole of a given size.
6435 * The key ins is changed to record the hole:
6436 * ins->objectid == start position
6437 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6438 * ins->offset == the size of the hole.
6439 * Any available blocks before search_start are skipped.
6441 * If there is no suitable free space, we will record the max size of
6442 * the free space extent currently.
6444 static noinline
int find_free_extent(struct btrfs_root
*orig_root
,
6445 u64 num_bytes
, u64 empty_size
,
6446 u64 hint_byte
, struct btrfs_key
*ins
,
6447 u64 flags
, int delalloc
)
6450 struct btrfs_root
*root
= orig_root
->fs_info
->extent_root
;
6451 struct btrfs_free_cluster
*last_ptr
= NULL
;
6452 struct btrfs_block_group_cache
*block_group
= NULL
;
6453 u64 search_start
= 0;
6454 u64 max_extent_size
= 0;
6455 int empty_cluster
= 2 * 1024 * 1024;
6456 struct btrfs_space_info
*space_info
;
6458 int index
= __get_raid_index(flags
);
6459 int alloc_type
= (flags
& BTRFS_BLOCK_GROUP_DATA
) ?
6460 RESERVE_ALLOC_NO_ACCOUNT
: RESERVE_ALLOC
;
6461 bool failed_cluster_refill
= false;
6462 bool failed_alloc
= false;
6463 bool use_cluster
= true;
6464 bool have_caching_bg
= false;
6466 WARN_ON(num_bytes
< root
->sectorsize
);
6467 btrfs_set_key_type(ins
, BTRFS_EXTENT_ITEM_KEY
);
6471 trace_find_free_extent(orig_root
, num_bytes
, empty_size
, flags
);
6473 space_info
= __find_space_info(root
->fs_info
, flags
);
6475 btrfs_err(root
->fs_info
, "No space info for %llu", flags
);
6480 * If the space info is for both data and metadata it means we have a
6481 * small filesystem and we can't use the clustering stuff.
6483 if (btrfs_mixed_space_info(space_info
))
6484 use_cluster
= false;
6486 if (flags
& BTRFS_BLOCK_GROUP_METADATA
&& use_cluster
) {
6487 last_ptr
= &root
->fs_info
->meta_alloc_cluster
;
6488 if (!btrfs_test_opt(root
, SSD
))
6489 empty_cluster
= 64 * 1024;
6492 if ((flags
& BTRFS_BLOCK_GROUP_DATA
) && use_cluster
&&
6493 btrfs_test_opt(root
, SSD
)) {
6494 last_ptr
= &root
->fs_info
->data_alloc_cluster
;
6498 spin_lock(&last_ptr
->lock
);
6499 if (last_ptr
->block_group
)
6500 hint_byte
= last_ptr
->window_start
;
6501 spin_unlock(&last_ptr
->lock
);
6504 search_start
= max(search_start
, first_logical_byte(root
, 0));
6505 search_start
= max(search_start
, hint_byte
);
6510 if (search_start
== hint_byte
) {
6511 block_group
= btrfs_lookup_block_group(root
->fs_info
,
6514 * we don't want to use the block group if it doesn't match our
6515 * allocation bits, or if its not cached.
6517 * However if we are re-searching with an ideal block group
6518 * picked out then we don't care that the block group is cached.
6520 if (block_group
&& block_group_bits(block_group
, flags
) &&
6521 block_group
->cached
!= BTRFS_CACHE_NO
) {
6522 down_read(&space_info
->groups_sem
);
6523 if (list_empty(&block_group
->list
) ||
6526 * someone is removing this block group,
6527 * we can't jump into the have_block_group
6528 * target because our list pointers are not
6531 btrfs_put_block_group(block_group
);
6532 up_read(&space_info
->groups_sem
);
6534 index
= get_block_group_index(block_group
);
6535 btrfs_lock_block_group(block_group
, delalloc
);
6536 goto have_block_group
;
6538 } else if (block_group
) {
6539 btrfs_put_block_group(block_group
);
6543 have_caching_bg
= false;
6544 down_read(&space_info
->groups_sem
);
6545 list_for_each_entry(block_group
, &space_info
->block_groups
[index
],
6550 btrfs_grab_block_group(block_group
, delalloc
);
6551 search_start
= block_group
->key
.objectid
;
6554 * this can happen if we end up cycling through all the
6555 * raid types, but we want to make sure we only allocate
6556 * for the proper type.
6558 if (!block_group_bits(block_group
, flags
)) {
6559 u64 extra
= BTRFS_BLOCK_GROUP_DUP
|
6560 BTRFS_BLOCK_GROUP_RAID1
|
6561 BTRFS_BLOCK_GROUP_RAID5
|
6562 BTRFS_BLOCK_GROUP_RAID6
|
6563 BTRFS_BLOCK_GROUP_RAID10
;
6566 * if they asked for extra copies and this block group
6567 * doesn't provide them, bail. This does allow us to
6568 * fill raid0 from raid1.
6570 if ((flags
& extra
) && !(block_group
->flags
& extra
))
6575 cached
= block_group_cache_done(block_group
);
6576 if (unlikely(!cached
)) {
6577 ret
= cache_block_group(block_group
, 0);
6582 if (unlikely(block_group
->cached
== BTRFS_CACHE_ERROR
))
6584 if (unlikely(block_group
->ro
))
6588 * Ok we want to try and use the cluster allocator, so
6592 struct btrfs_block_group_cache
*used_block_group
;
6593 unsigned long aligned_cluster
;
6595 * the refill lock keeps out other
6596 * people trying to start a new cluster
6598 used_block_group
= btrfs_lock_cluster(block_group
,
6601 if (!used_block_group
)
6602 goto refill_cluster
;
6604 if (used_block_group
!= block_group
&&
6605 (used_block_group
->ro
||
6606 !block_group_bits(used_block_group
, flags
)))
6607 goto release_cluster
;
6609 offset
= btrfs_alloc_from_cluster(used_block_group
,
6612 used_block_group
->key
.objectid
,
6615 /* we have a block, we're done */
6616 spin_unlock(&last_ptr
->refill_lock
);
6617 trace_btrfs_reserve_extent_cluster(root
,
6619 search_start
, num_bytes
);
6620 if (used_block_group
!= block_group
) {
6621 btrfs_release_block_group(block_group
,
6623 block_group
= used_block_group
;
6628 WARN_ON(last_ptr
->block_group
!= used_block_group
);
6630 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6631 * set up a new clusters, so lets just skip it
6632 * and let the allocator find whatever block
6633 * it can find. If we reach this point, we
6634 * will have tried the cluster allocator
6635 * plenty of times and not have found
6636 * anything, so we are likely way too
6637 * fragmented for the clustering stuff to find
6640 * However, if the cluster is taken from the
6641 * current block group, release the cluster
6642 * first, so that we stand a better chance of
6643 * succeeding in the unclustered
6645 if (loop
>= LOOP_NO_EMPTY_SIZE
&&
6646 used_block_group
!= block_group
) {
6647 spin_unlock(&last_ptr
->refill_lock
);
6648 btrfs_release_block_group(used_block_group
,
6650 goto unclustered_alloc
;
6654 * this cluster didn't work out, free it and
6657 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
6659 if (used_block_group
!= block_group
)
6660 btrfs_release_block_group(used_block_group
,
6663 if (loop
>= LOOP_NO_EMPTY_SIZE
) {
6664 spin_unlock(&last_ptr
->refill_lock
);
6665 goto unclustered_alloc
;
6668 aligned_cluster
= max_t(unsigned long,
6669 empty_cluster
+ empty_size
,
6670 block_group
->full_stripe_len
);
6672 /* allocate a cluster in this block group */
6673 ret
= btrfs_find_space_cluster(root
, block_group
,
6674 last_ptr
, search_start
,
6679 * now pull our allocation out of this
6682 offset
= btrfs_alloc_from_cluster(block_group
,
6688 /* we found one, proceed */
6689 spin_unlock(&last_ptr
->refill_lock
);
6690 trace_btrfs_reserve_extent_cluster(root
,
6691 block_group
, search_start
,
6695 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
6696 && !failed_cluster_refill
) {
6697 spin_unlock(&last_ptr
->refill_lock
);
6699 failed_cluster_refill
= true;
6700 wait_block_group_cache_progress(block_group
,
6701 num_bytes
+ empty_cluster
+ empty_size
);
6702 goto have_block_group
;
6706 * at this point we either didn't find a cluster
6707 * or we weren't able to allocate a block from our
6708 * cluster. Free the cluster we've been trying
6709 * to use, and go to the next block group
6711 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
6712 spin_unlock(&last_ptr
->refill_lock
);
6717 spin_lock(&block_group
->free_space_ctl
->tree_lock
);
6719 block_group
->free_space_ctl
->free_space
<
6720 num_bytes
+ empty_cluster
+ empty_size
) {
6721 if (block_group
->free_space_ctl
->free_space
>
6724 block_group
->free_space_ctl
->free_space
;
6725 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
6728 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
6730 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
6731 num_bytes
, empty_size
,
6734 * If we didn't find a chunk, and we haven't failed on this
6735 * block group before, and this block group is in the middle of
6736 * caching and we are ok with waiting, then go ahead and wait
6737 * for progress to be made, and set failed_alloc to true.
6739 * If failed_alloc is true then we've already waited on this
6740 * block group once and should move on to the next block group.
6742 if (!offset
&& !failed_alloc
&& !cached
&&
6743 loop
> LOOP_CACHING_NOWAIT
) {
6744 wait_block_group_cache_progress(block_group
,
6745 num_bytes
+ empty_size
);
6746 failed_alloc
= true;
6747 goto have_block_group
;
6748 } else if (!offset
) {
6750 have_caching_bg
= true;
6754 search_start
= stripe_align(root
, block_group
,
6757 /* move on to the next group */
6758 if (search_start
+ num_bytes
>
6759 block_group
->key
.objectid
+ block_group
->key
.offset
) {
6760 btrfs_add_free_space(block_group
, offset
, num_bytes
);
6764 if (offset
< search_start
)
6765 btrfs_add_free_space(block_group
, offset
,
6766 search_start
- offset
);
6767 BUG_ON(offset
> search_start
);
6769 ret
= btrfs_update_reserved_bytes(block_group
, num_bytes
,
6770 alloc_type
, delalloc
);
6771 if (ret
== -EAGAIN
) {
6772 btrfs_add_free_space(block_group
, offset
, num_bytes
);
6776 /* we are all good, lets return */
6777 ins
->objectid
= search_start
;
6778 ins
->offset
= num_bytes
;
6780 trace_btrfs_reserve_extent(orig_root
, block_group
,
6781 search_start
, num_bytes
);
6782 btrfs_release_block_group(block_group
, delalloc
);
6785 failed_cluster_refill
= false;
6786 failed_alloc
= false;
6787 BUG_ON(index
!= get_block_group_index(block_group
));
6788 btrfs_release_block_group(block_group
, delalloc
);
6790 up_read(&space_info
->groups_sem
);
6792 if (!ins
->objectid
&& loop
>= LOOP_CACHING_WAIT
&& have_caching_bg
)
6795 if (!ins
->objectid
&& ++index
< BTRFS_NR_RAID_TYPES
)
6799 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6800 * caching kthreads as we move along
6801 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6802 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6803 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6806 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
) {
6809 if (loop
== LOOP_ALLOC_CHUNK
) {
6810 struct btrfs_trans_handle
*trans
;
6813 trans
= current
->journal_info
;
6817 trans
= btrfs_join_transaction(root
);
6819 if (IS_ERR(trans
)) {
6820 ret
= PTR_ERR(trans
);
6824 ret
= do_chunk_alloc(trans
, root
, flags
,
6827 * Do not bail out on ENOSPC since we
6828 * can do more things.
6830 if (ret
< 0 && ret
!= -ENOSPC
)
6831 btrfs_abort_transaction(trans
,
6836 btrfs_end_transaction(trans
, root
);
6841 if (loop
== LOOP_NO_EMPTY_SIZE
) {
6847 } else if (!ins
->objectid
) {
6849 } else if (ins
->objectid
) {
6854 ins
->offset
= max_extent_size
;
6858 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
6859 int dump_block_groups
)
6861 struct btrfs_block_group_cache
*cache
;
6864 spin_lock(&info
->lock
);
6865 printk(KERN_INFO
"BTRFS: space_info %llu has %llu free, is %sfull\n",
6867 info
->total_bytes
- info
->bytes_used
- info
->bytes_pinned
-
6868 info
->bytes_reserved
- info
->bytes_readonly
,
6869 (info
->full
) ? "" : "not ");
6870 printk(KERN_INFO
"BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
6871 "reserved=%llu, may_use=%llu, readonly=%llu\n",
6872 info
->total_bytes
, info
->bytes_used
, info
->bytes_pinned
,
6873 info
->bytes_reserved
, info
->bytes_may_use
,
6874 info
->bytes_readonly
);
6875 spin_unlock(&info
->lock
);
6877 if (!dump_block_groups
)
6880 down_read(&info
->groups_sem
);
6882 list_for_each_entry(cache
, &info
->block_groups
[index
], list
) {
6883 spin_lock(&cache
->lock
);
6884 printk(KERN_INFO
"BTRFS: "
6885 "block group %llu has %llu bytes, "
6886 "%llu used %llu pinned %llu reserved %s\n",
6887 cache
->key
.objectid
, cache
->key
.offset
,
6888 btrfs_block_group_used(&cache
->item
), cache
->pinned
,
6889 cache
->reserved
, cache
->ro
? "[readonly]" : "");
6890 btrfs_dump_free_space(cache
, bytes
);
6891 spin_unlock(&cache
->lock
);
6893 if (++index
< BTRFS_NR_RAID_TYPES
)
6895 up_read(&info
->groups_sem
);
6898 int btrfs_reserve_extent(struct btrfs_root
*root
,
6899 u64 num_bytes
, u64 min_alloc_size
,
6900 u64 empty_size
, u64 hint_byte
,
6901 struct btrfs_key
*ins
, int is_data
, int delalloc
)
6903 bool final_tried
= false;
6907 flags
= btrfs_get_alloc_profile(root
, is_data
);
6909 WARN_ON(num_bytes
< root
->sectorsize
);
6910 ret
= find_free_extent(root
, num_bytes
, empty_size
, hint_byte
, ins
,
6913 if (ret
== -ENOSPC
) {
6914 if (!final_tried
&& ins
->offset
) {
6915 num_bytes
= min(num_bytes
>> 1, ins
->offset
);
6916 num_bytes
= round_down(num_bytes
, root
->sectorsize
);
6917 num_bytes
= max(num_bytes
, min_alloc_size
);
6918 if (num_bytes
== min_alloc_size
)
6921 } else if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
6922 struct btrfs_space_info
*sinfo
;
6924 sinfo
= __find_space_info(root
->fs_info
, flags
);
6925 btrfs_err(root
->fs_info
, "allocation failed flags %llu, wanted %llu",
6928 dump_space_info(sinfo
, num_bytes
, 1);
6935 static int __btrfs_free_reserved_extent(struct btrfs_root
*root
,
6937 int pin
, int delalloc
)
6939 struct btrfs_block_group_cache
*cache
;
6942 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
6944 btrfs_err(root
->fs_info
, "Unable to find block group for %llu",
6949 if (btrfs_test_opt(root
, DISCARD
))
6950 ret
= btrfs_discard_extent(root
, start
, len
, NULL
);
6953 pin_down_extent(root
, cache
, start
, len
, 1);
6955 btrfs_add_free_space(cache
, start
, len
);
6956 btrfs_update_reserved_bytes(cache
, len
, RESERVE_FREE
, delalloc
);
6958 btrfs_put_block_group(cache
);
6960 trace_btrfs_reserved_extent_free(root
, start
, len
);
6965 int btrfs_free_reserved_extent(struct btrfs_root
*root
,
6966 u64 start
, u64 len
, int delalloc
)
6968 return __btrfs_free_reserved_extent(root
, start
, len
, 0, delalloc
);
6971 int btrfs_free_and_pin_reserved_extent(struct btrfs_root
*root
,
6974 return __btrfs_free_reserved_extent(root
, start
, len
, 1, 0);
6977 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
6978 struct btrfs_root
*root
,
6979 u64 parent
, u64 root_objectid
,
6980 u64 flags
, u64 owner
, u64 offset
,
6981 struct btrfs_key
*ins
, int ref_mod
)
6984 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6985 struct btrfs_extent_item
*extent_item
;
6986 struct btrfs_extent_inline_ref
*iref
;
6987 struct btrfs_path
*path
;
6988 struct extent_buffer
*leaf
;
6993 type
= BTRFS_SHARED_DATA_REF_KEY
;
6995 type
= BTRFS_EXTENT_DATA_REF_KEY
;
6997 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
6999 path
= btrfs_alloc_path();
7003 path
->leave_spinning
= 1;
7004 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
7007 btrfs_free_path(path
);
7011 leaf
= path
->nodes
[0];
7012 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
7013 struct btrfs_extent_item
);
7014 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
7015 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
7016 btrfs_set_extent_flags(leaf
, extent_item
,
7017 flags
| BTRFS_EXTENT_FLAG_DATA
);
7019 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
7020 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
7022 struct btrfs_shared_data_ref
*ref
;
7023 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
7024 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
7025 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
7027 struct btrfs_extent_data_ref
*ref
;
7028 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
7029 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
7030 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
7031 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
7032 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
7035 btrfs_mark_buffer_dirty(path
->nodes
[0]);
7036 btrfs_free_path(path
);
7038 /* Always set parent to 0 here since its exclusive anyway. */
7039 ret
= btrfs_qgroup_record_ref(trans
, fs_info
, root_objectid
,
7040 ins
->objectid
, ins
->offset
,
7041 BTRFS_QGROUP_OPER_ADD_EXCL
, 0);
7045 ret
= update_block_group(root
, ins
->objectid
, ins
->offset
, 1);
7046 if (ret
) { /* -ENOENT, logic error */
7047 btrfs_err(fs_info
, "update block group failed for %llu %llu",
7048 ins
->objectid
, ins
->offset
);
7051 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, ins
->offset
);
7055 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
7056 struct btrfs_root
*root
,
7057 u64 parent
, u64 root_objectid
,
7058 u64 flags
, struct btrfs_disk_key
*key
,
7059 int level
, struct btrfs_key
*ins
,
7063 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7064 struct btrfs_extent_item
*extent_item
;
7065 struct btrfs_tree_block_info
*block_info
;
7066 struct btrfs_extent_inline_ref
*iref
;
7067 struct btrfs_path
*path
;
7068 struct extent_buffer
*leaf
;
7069 u32 size
= sizeof(*extent_item
) + sizeof(*iref
);
7070 u64 num_bytes
= ins
->offset
;
7071 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
7074 if (!skinny_metadata
)
7075 size
+= sizeof(*block_info
);
7077 path
= btrfs_alloc_path();
7079 btrfs_free_and_pin_reserved_extent(root
, ins
->objectid
,
7084 path
->leave_spinning
= 1;
7085 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
7088 btrfs_free_and_pin_reserved_extent(root
, ins
->objectid
,
7090 btrfs_free_path(path
);
7094 leaf
= path
->nodes
[0];
7095 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
7096 struct btrfs_extent_item
);
7097 btrfs_set_extent_refs(leaf
, extent_item
, 1);
7098 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
7099 btrfs_set_extent_flags(leaf
, extent_item
,
7100 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
7102 if (skinny_metadata
) {
7103 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
7104 num_bytes
= root
->leafsize
;
7106 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
7107 btrfs_set_tree_block_key(leaf
, block_info
, key
);
7108 btrfs_set_tree_block_level(leaf
, block_info
, level
);
7109 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
7113 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
7114 btrfs_set_extent_inline_ref_type(leaf
, iref
,
7115 BTRFS_SHARED_BLOCK_REF_KEY
);
7116 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
7118 btrfs_set_extent_inline_ref_type(leaf
, iref
,
7119 BTRFS_TREE_BLOCK_REF_KEY
);
7120 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
7123 btrfs_mark_buffer_dirty(leaf
);
7124 btrfs_free_path(path
);
7127 ret
= btrfs_qgroup_record_ref(trans
, fs_info
, root_objectid
,
7128 ins
->objectid
, num_bytes
,
7129 BTRFS_QGROUP_OPER_ADD_EXCL
, 0);
7134 ret
= update_block_group(root
, ins
->objectid
, root
->leafsize
, 1);
7135 if (ret
) { /* -ENOENT, logic error */
7136 btrfs_err(fs_info
, "update block group failed for %llu %llu",
7137 ins
->objectid
, ins
->offset
);
7141 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, root
->leafsize
);
7145 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
7146 struct btrfs_root
*root
,
7147 u64 root_objectid
, u64 owner
,
7148 u64 offset
, struct btrfs_key
*ins
)
7152 BUG_ON(root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
7154 ret
= btrfs_add_delayed_data_ref(root
->fs_info
, trans
, ins
->objectid
,
7156 root_objectid
, owner
, offset
,
7157 BTRFS_ADD_DELAYED_EXTENT
, NULL
, 0);
7162 * this is used by the tree logging recovery code. It records that
7163 * an extent has been allocated and makes sure to clear the free
7164 * space cache bits as well
7166 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
7167 struct btrfs_root
*root
,
7168 u64 root_objectid
, u64 owner
, u64 offset
,
7169 struct btrfs_key
*ins
)
7172 struct btrfs_block_group_cache
*block_group
;
7175 * Mixed block groups will exclude before processing the log so we only
7176 * need to do the exlude dance if this fs isn't mixed.
7178 if (!btrfs_fs_incompat(root
->fs_info
, MIXED_GROUPS
)) {
7179 ret
= __exclude_logged_extent(root
, ins
->objectid
, ins
->offset
);
7184 block_group
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
7188 ret
= btrfs_update_reserved_bytes(block_group
, ins
->offset
,
7189 RESERVE_ALLOC_NO_ACCOUNT
, 0);
7190 BUG_ON(ret
); /* logic error */
7191 ret
= alloc_reserved_file_extent(trans
, root
, 0, root_objectid
,
7192 0, owner
, offset
, ins
, 1);
7193 btrfs_put_block_group(block_group
);
7197 static struct extent_buffer
*
7198 btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
7199 u64 bytenr
, u32 blocksize
, int level
)
7201 struct extent_buffer
*buf
;
7203 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
7205 return ERR_PTR(-ENOMEM
);
7206 btrfs_set_header_generation(buf
, trans
->transid
);
7207 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, buf
, level
);
7208 btrfs_tree_lock(buf
);
7209 clean_tree_block(trans
, root
, buf
);
7210 clear_bit(EXTENT_BUFFER_STALE
, &buf
->bflags
);
7212 btrfs_set_lock_blocking(buf
);
7213 btrfs_set_buffer_uptodate(buf
);
7215 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
7217 * we allow two log transactions at a time, use different
7218 * EXENT bit to differentiate dirty pages.
7220 if (root
->log_transid
% 2 == 0)
7221 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
7222 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
7224 set_extent_new(&root
->dirty_log_pages
, buf
->start
,
7225 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
7227 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
7228 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
7230 trans
->blocks_used
++;
7231 /* this returns a buffer locked for blocking */
7235 static struct btrfs_block_rsv
*
7236 use_block_rsv(struct btrfs_trans_handle
*trans
,
7237 struct btrfs_root
*root
, u32 blocksize
)
7239 struct btrfs_block_rsv
*block_rsv
;
7240 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
7242 bool global_updated
= false;
7244 block_rsv
= get_block_rsv(trans
, root
);
7246 if (unlikely(block_rsv
->size
== 0))
7249 ret
= block_rsv_use_bytes(block_rsv
, blocksize
);
7253 if (block_rsv
->failfast
)
7254 return ERR_PTR(ret
);
7256 if (block_rsv
->type
== BTRFS_BLOCK_RSV_GLOBAL
&& !global_updated
) {
7257 global_updated
= true;
7258 update_global_block_rsv(root
->fs_info
);
7262 if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
7263 static DEFINE_RATELIMIT_STATE(_rs
,
7264 DEFAULT_RATELIMIT_INTERVAL
* 10,
7265 /*DEFAULT_RATELIMIT_BURST*/ 1);
7266 if (__ratelimit(&_rs
))
7268 "BTRFS: block rsv returned %d\n", ret
);
7271 ret
= reserve_metadata_bytes(root
, block_rsv
, blocksize
,
7272 BTRFS_RESERVE_NO_FLUSH
);
7276 * If we couldn't reserve metadata bytes try and use some from
7277 * the global reserve if its space type is the same as the global
7280 if (block_rsv
->type
!= BTRFS_BLOCK_RSV_GLOBAL
&&
7281 block_rsv
->space_info
== global_rsv
->space_info
) {
7282 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
7286 return ERR_PTR(ret
);
7289 static void unuse_block_rsv(struct btrfs_fs_info
*fs_info
,
7290 struct btrfs_block_rsv
*block_rsv
, u32 blocksize
)
7292 block_rsv_add_bytes(block_rsv
, blocksize
, 0);
7293 block_rsv_release_bytes(fs_info
, block_rsv
, NULL
, 0);
7297 * finds a free extent and does all the dirty work required for allocation
7298 * returns the key for the extent through ins, and a tree buffer for
7299 * the first block of the extent through buf.
7301 * returns the tree buffer or NULL.
7303 struct extent_buffer
*btrfs_alloc_free_block(struct btrfs_trans_handle
*trans
,
7304 struct btrfs_root
*root
, u32 blocksize
,
7305 u64 parent
, u64 root_objectid
,
7306 struct btrfs_disk_key
*key
, int level
,
7307 u64 hint
, u64 empty_size
)
7309 struct btrfs_key ins
;
7310 struct btrfs_block_rsv
*block_rsv
;
7311 struct extent_buffer
*buf
;
7314 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
7317 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
7318 if (unlikely(test_bit(BTRFS_ROOT_DUMMY_ROOT
, &root
->state
))) {
7319 buf
= btrfs_init_new_buffer(trans
, root
, root
->alloc_bytenr
,
7322 root
->alloc_bytenr
+= blocksize
;
7326 block_rsv
= use_block_rsv(trans
, root
, blocksize
);
7327 if (IS_ERR(block_rsv
))
7328 return ERR_CAST(block_rsv
);
7330 ret
= btrfs_reserve_extent(root
, blocksize
, blocksize
,
7331 empty_size
, hint
, &ins
, 0, 0);
7333 unuse_block_rsv(root
->fs_info
, block_rsv
, blocksize
);
7334 return ERR_PTR(ret
);
7337 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
,
7339 BUG_ON(IS_ERR(buf
)); /* -ENOMEM */
7341 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
7343 parent
= ins
.objectid
;
7344 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
7348 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
7349 struct btrfs_delayed_extent_op
*extent_op
;
7350 extent_op
= btrfs_alloc_delayed_extent_op();
7351 BUG_ON(!extent_op
); /* -ENOMEM */
7353 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
7355 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
7356 extent_op
->flags_to_set
= flags
;
7357 if (skinny_metadata
)
7358 extent_op
->update_key
= 0;
7360 extent_op
->update_key
= 1;
7361 extent_op
->update_flags
= 1;
7362 extent_op
->is_data
= 0;
7363 extent_op
->level
= level
;
7365 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
7367 ins
.offset
, parent
, root_objectid
,
7368 level
, BTRFS_ADD_DELAYED_EXTENT
,
7370 BUG_ON(ret
); /* -ENOMEM */
7375 struct walk_control
{
7376 u64 refs
[BTRFS_MAX_LEVEL
];
7377 u64 flags
[BTRFS_MAX_LEVEL
];
7378 struct btrfs_key update_progress
;
7389 #define DROP_REFERENCE 1
7390 #define UPDATE_BACKREF 2
7392 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
7393 struct btrfs_root
*root
,
7394 struct walk_control
*wc
,
7395 struct btrfs_path
*path
)
7403 struct btrfs_key key
;
7404 struct extent_buffer
*eb
;
7409 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
7410 wc
->reada_count
= wc
->reada_count
* 2 / 3;
7411 wc
->reada_count
= max(wc
->reada_count
, 2);
7413 wc
->reada_count
= wc
->reada_count
* 3 / 2;
7414 wc
->reada_count
= min_t(int, wc
->reada_count
,
7415 BTRFS_NODEPTRS_PER_BLOCK(root
));
7418 eb
= path
->nodes
[wc
->level
];
7419 nritems
= btrfs_header_nritems(eb
);
7420 blocksize
= btrfs_level_size(root
, wc
->level
- 1);
7422 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
7423 if (nread
>= wc
->reada_count
)
7427 bytenr
= btrfs_node_blockptr(eb
, slot
);
7428 generation
= btrfs_node_ptr_generation(eb
, slot
);
7430 if (slot
== path
->slots
[wc
->level
])
7433 if (wc
->stage
== UPDATE_BACKREF
&&
7434 generation
<= root
->root_key
.offset
)
7437 /* We don't lock the tree block, it's OK to be racy here */
7438 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
,
7439 wc
->level
- 1, 1, &refs
,
7441 /* We don't care about errors in readahead. */
7446 if (wc
->stage
== DROP_REFERENCE
) {
7450 if (wc
->level
== 1 &&
7451 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
7453 if (!wc
->update_ref
||
7454 generation
<= root
->root_key
.offset
)
7456 btrfs_node_key_to_cpu(eb
, &key
, slot
);
7457 ret
= btrfs_comp_cpu_keys(&key
,
7458 &wc
->update_progress
);
7462 if (wc
->level
== 1 &&
7463 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
7467 ret
= readahead_tree_block(root
, bytenr
, blocksize
,
7473 wc
->reada_slot
= slot
;
7476 static int account_leaf_items(struct btrfs_trans_handle
*trans
,
7477 struct btrfs_root
*root
,
7478 struct extent_buffer
*eb
)
7480 int nr
= btrfs_header_nritems(eb
);
7481 int i
, extent_type
, ret
;
7482 struct btrfs_key key
;
7483 struct btrfs_file_extent_item
*fi
;
7484 u64 bytenr
, num_bytes
;
7486 for (i
= 0; i
< nr
; i
++) {
7487 btrfs_item_key_to_cpu(eb
, &key
, i
);
7489 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
7492 fi
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
7493 /* filter out non qgroup-accountable extents */
7494 extent_type
= btrfs_file_extent_type(eb
, fi
);
7496 if (extent_type
== BTRFS_FILE_EXTENT_INLINE
)
7499 bytenr
= btrfs_file_extent_disk_bytenr(eb
, fi
);
7503 num_bytes
= btrfs_file_extent_disk_num_bytes(eb
, fi
);
7505 ret
= btrfs_qgroup_record_ref(trans
, root
->fs_info
,
7508 BTRFS_QGROUP_OPER_SUB_SUBTREE
, 0);
7516 * Walk up the tree from the bottom, freeing leaves and any interior
7517 * nodes which have had all slots visited. If a node (leaf or
7518 * interior) is freed, the node above it will have it's slot
7519 * incremented. The root node will never be freed.
7521 * At the end of this function, we should have a path which has all
7522 * slots incremented to the next position for a search. If we need to
7523 * read a new node it will be NULL and the node above it will have the
7524 * correct slot selected for a later read.
7526 * If we increment the root nodes slot counter past the number of
7527 * elements, 1 is returned to signal completion of the search.
7529 static int adjust_slots_upwards(struct btrfs_root
*root
,
7530 struct btrfs_path
*path
, int root_level
)
7534 struct extent_buffer
*eb
;
7536 if (root_level
== 0)
7539 while (level
<= root_level
) {
7540 eb
= path
->nodes
[level
];
7541 nr
= btrfs_header_nritems(eb
);
7542 path
->slots
[level
]++;
7543 slot
= path
->slots
[level
];
7544 if (slot
>= nr
|| level
== 0) {
7546 * Don't free the root - we will detect this
7547 * condition after our loop and return a
7548 * positive value for caller to stop walking the tree.
7550 if (level
!= root_level
) {
7551 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7552 path
->locks
[level
] = 0;
7554 free_extent_buffer(eb
);
7555 path
->nodes
[level
] = NULL
;
7556 path
->slots
[level
] = 0;
7560 * We have a valid slot to walk back down
7561 * from. Stop here so caller can process these
7570 eb
= path
->nodes
[root_level
];
7571 if (path
->slots
[root_level
] >= btrfs_header_nritems(eb
))
7578 * root_eb is the subtree root and is locked before this function is called.
7580 static int account_shared_subtree(struct btrfs_trans_handle
*trans
,
7581 struct btrfs_root
*root
,
7582 struct extent_buffer
*root_eb
,
7588 struct extent_buffer
*eb
= root_eb
;
7589 struct btrfs_path
*path
= NULL
;
7591 BUG_ON(root_level
< 0 || root_level
> BTRFS_MAX_LEVEL
);
7592 BUG_ON(root_eb
== NULL
);
7594 if (!root
->fs_info
->quota_enabled
)
7597 if (!extent_buffer_uptodate(root_eb
)) {
7598 ret
= btrfs_read_buffer(root_eb
, root_gen
);
7603 if (root_level
== 0) {
7604 ret
= account_leaf_items(trans
, root
, root_eb
);
7608 path
= btrfs_alloc_path();
7613 * Walk down the tree. Missing extent blocks are filled in as
7614 * we go. Metadata is accounted every time we read a new
7617 * When we reach a leaf, we account for file extent items in it,
7618 * walk back up the tree (adjusting slot pointers as we go)
7619 * and restart the search process.
7621 extent_buffer_get(root_eb
); /* For path */
7622 path
->nodes
[root_level
] = root_eb
;
7623 path
->slots
[root_level
] = 0;
7624 path
->locks
[root_level
] = 0; /* so release_path doesn't try to unlock */
7627 while (level
>= 0) {
7628 if (path
->nodes
[level
] == NULL
) {
7629 int child_bsize
= root
->nodesize
;
7634 /* We need to get child blockptr/gen from
7635 * parent before we can read it. */
7636 eb
= path
->nodes
[level
+ 1];
7637 parent_slot
= path
->slots
[level
+ 1];
7638 child_bytenr
= btrfs_node_blockptr(eb
, parent_slot
);
7639 child_gen
= btrfs_node_ptr_generation(eb
, parent_slot
);
7641 eb
= read_tree_block(root
, child_bytenr
, child_bsize
,
7643 if (!eb
|| !extent_buffer_uptodate(eb
)) {
7648 path
->nodes
[level
] = eb
;
7649 path
->slots
[level
] = 0;
7651 btrfs_tree_read_lock(eb
);
7652 btrfs_set_lock_blocking_rw(eb
, BTRFS_READ_LOCK
);
7653 path
->locks
[level
] = BTRFS_READ_LOCK_BLOCKING
;
7655 ret
= btrfs_qgroup_record_ref(trans
, root
->fs_info
,
7659 BTRFS_QGROUP_OPER_SUB_SUBTREE
,
7667 ret
= account_leaf_items(trans
, root
, path
->nodes
[level
]);
7671 /* Nonzero return here means we completed our search */
7672 ret
= adjust_slots_upwards(root
, path
, root_level
);
7676 /* Restart search with new slots */
7685 btrfs_free_path(path
);
7691 * helper to process tree block while walking down the tree.
7693 * when wc->stage == UPDATE_BACKREF, this function updates
7694 * back refs for pointers in the block.
7696 * NOTE: return value 1 means we should stop walking down.
7698 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
7699 struct btrfs_root
*root
,
7700 struct btrfs_path
*path
,
7701 struct walk_control
*wc
, int lookup_info
)
7703 int level
= wc
->level
;
7704 struct extent_buffer
*eb
= path
->nodes
[level
];
7705 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
7708 if (wc
->stage
== UPDATE_BACKREF
&&
7709 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
7713 * when reference count of tree block is 1, it won't increase
7714 * again. once full backref flag is set, we never clear it.
7717 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
7718 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
7719 BUG_ON(!path
->locks
[level
]);
7720 ret
= btrfs_lookup_extent_info(trans
, root
,
7721 eb
->start
, level
, 1,
7724 BUG_ON(ret
== -ENOMEM
);
7727 BUG_ON(wc
->refs
[level
] == 0);
7730 if (wc
->stage
== DROP_REFERENCE
) {
7731 if (wc
->refs
[level
] > 1)
7734 if (path
->locks
[level
] && !wc
->keep_locks
) {
7735 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7736 path
->locks
[level
] = 0;
7741 /* wc->stage == UPDATE_BACKREF */
7742 if (!(wc
->flags
[level
] & flag
)) {
7743 BUG_ON(!path
->locks
[level
]);
7744 ret
= btrfs_inc_ref(trans
, root
, eb
, 1);
7745 BUG_ON(ret
); /* -ENOMEM */
7746 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
7747 BUG_ON(ret
); /* -ENOMEM */
7748 ret
= btrfs_set_disk_extent_flags(trans
, root
, eb
->start
,
7750 btrfs_header_level(eb
), 0);
7751 BUG_ON(ret
); /* -ENOMEM */
7752 wc
->flags
[level
] |= flag
;
7756 * the block is shared by multiple trees, so it's not good to
7757 * keep the tree lock
7759 if (path
->locks
[level
] && level
> 0) {
7760 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7761 path
->locks
[level
] = 0;
7767 * helper to process tree block pointer.
7769 * when wc->stage == DROP_REFERENCE, this function checks
7770 * reference count of the block pointed to. if the block
7771 * is shared and we need update back refs for the subtree
7772 * rooted at the block, this function changes wc->stage to
7773 * UPDATE_BACKREF. if the block is shared and there is no
7774 * need to update back, this function drops the reference
7777 * NOTE: return value 1 means we should stop walking down.
7779 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
7780 struct btrfs_root
*root
,
7781 struct btrfs_path
*path
,
7782 struct walk_control
*wc
, int *lookup_info
)
7788 struct btrfs_key key
;
7789 struct extent_buffer
*next
;
7790 int level
= wc
->level
;
7793 bool need_account
= false;
7795 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
7796 path
->slots
[level
]);
7798 * if the lower level block was created before the snapshot
7799 * was created, we know there is no need to update back refs
7802 if (wc
->stage
== UPDATE_BACKREF
&&
7803 generation
<= root
->root_key
.offset
) {
7808 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
7809 blocksize
= btrfs_level_size(root
, level
- 1);
7811 next
= btrfs_find_tree_block(root
, bytenr
, blocksize
);
7813 next
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
7816 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, next
,
7820 btrfs_tree_lock(next
);
7821 btrfs_set_lock_blocking(next
);
7823 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, level
- 1, 1,
7824 &wc
->refs
[level
- 1],
7825 &wc
->flags
[level
- 1]);
7827 btrfs_tree_unlock(next
);
7831 if (unlikely(wc
->refs
[level
- 1] == 0)) {
7832 btrfs_err(root
->fs_info
, "Missing references.");
7837 if (wc
->stage
== DROP_REFERENCE
) {
7838 if (wc
->refs
[level
- 1] > 1) {
7839 need_account
= true;
7841 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
7844 if (!wc
->update_ref
||
7845 generation
<= root
->root_key
.offset
)
7848 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
7849 path
->slots
[level
]);
7850 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
7854 wc
->stage
= UPDATE_BACKREF
;
7855 wc
->shared_level
= level
- 1;
7859 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
7863 if (!btrfs_buffer_uptodate(next
, generation
, 0)) {
7864 btrfs_tree_unlock(next
);
7865 free_extent_buffer(next
);
7871 if (reada
&& level
== 1)
7872 reada_walk_down(trans
, root
, wc
, path
);
7873 next
= read_tree_block(root
, bytenr
, blocksize
, generation
);
7874 if (!next
|| !extent_buffer_uptodate(next
)) {
7875 free_extent_buffer(next
);
7878 btrfs_tree_lock(next
);
7879 btrfs_set_lock_blocking(next
);
7883 BUG_ON(level
!= btrfs_header_level(next
));
7884 path
->nodes
[level
] = next
;
7885 path
->slots
[level
] = 0;
7886 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
7892 wc
->refs
[level
- 1] = 0;
7893 wc
->flags
[level
- 1] = 0;
7894 if (wc
->stage
== DROP_REFERENCE
) {
7895 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
7896 parent
= path
->nodes
[level
]->start
;
7898 BUG_ON(root
->root_key
.objectid
!=
7899 btrfs_header_owner(path
->nodes
[level
]));
7904 ret
= account_shared_subtree(trans
, root
, next
,
7905 generation
, level
- 1);
7907 printk_ratelimited(KERN_ERR
"BTRFS: %s Error "
7908 "%d accounting shared subtree. Quota "
7909 "is out of sync, rescan required.\n",
7910 root
->fs_info
->sb
->s_id
, ret
);
7913 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
, parent
,
7914 root
->root_key
.objectid
, level
- 1, 0, 0);
7915 BUG_ON(ret
); /* -ENOMEM */
7917 btrfs_tree_unlock(next
);
7918 free_extent_buffer(next
);
7924 * helper to process tree block while walking up the tree.
7926 * when wc->stage == DROP_REFERENCE, this function drops
7927 * reference count on the block.
7929 * when wc->stage == UPDATE_BACKREF, this function changes
7930 * wc->stage back to DROP_REFERENCE if we changed wc->stage
7931 * to UPDATE_BACKREF previously while processing the block.
7933 * NOTE: return value 1 means we should stop walking up.
7935 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
7936 struct btrfs_root
*root
,
7937 struct btrfs_path
*path
,
7938 struct walk_control
*wc
)
7941 int level
= wc
->level
;
7942 struct extent_buffer
*eb
= path
->nodes
[level
];
7945 if (wc
->stage
== UPDATE_BACKREF
) {
7946 BUG_ON(wc
->shared_level
< level
);
7947 if (level
< wc
->shared_level
)
7950 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
7954 wc
->stage
= DROP_REFERENCE
;
7955 wc
->shared_level
= -1;
7956 path
->slots
[level
] = 0;
7959 * check reference count again if the block isn't locked.
7960 * we should start walking down the tree again if reference
7963 if (!path
->locks
[level
]) {
7965 btrfs_tree_lock(eb
);
7966 btrfs_set_lock_blocking(eb
);
7967 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
7969 ret
= btrfs_lookup_extent_info(trans
, root
,
7970 eb
->start
, level
, 1,
7974 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7975 path
->locks
[level
] = 0;
7978 BUG_ON(wc
->refs
[level
] == 0);
7979 if (wc
->refs
[level
] == 1) {
7980 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7981 path
->locks
[level
] = 0;
7987 /* wc->stage == DROP_REFERENCE */
7988 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
7990 if (wc
->refs
[level
] == 1) {
7992 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
7993 ret
= btrfs_dec_ref(trans
, root
, eb
, 1);
7995 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
7996 BUG_ON(ret
); /* -ENOMEM */
7997 ret
= account_leaf_items(trans
, root
, eb
);
7999 printk_ratelimited(KERN_ERR
"BTRFS: %s Error "
8000 "%d accounting leaf items. Quota "
8001 "is out of sync, rescan required.\n",
8002 root
->fs_info
->sb
->s_id
, ret
);
8005 /* make block locked assertion in clean_tree_block happy */
8006 if (!path
->locks
[level
] &&
8007 btrfs_header_generation(eb
) == trans
->transid
) {
8008 btrfs_tree_lock(eb
);
8009 btrfs_set_lock_blocking(eb
);
8010 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8012 clean_tree_block(trans
, root
, eb
);
8015 if (eb
== root
->node
) {
8016 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8019 BUG_ON(root
->root_key
.objectid
!=
8020 btrfs_header_owner(eb
));
8022 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8023 parent
= path
->nodes
[level
+ 1]->start
;
8025 BUG_ON(root
->root_key
.objectid
!=
8026 btrfs_header_owner(path
->nodes
[level
+ 1]));
8029 btrfs_free_tree_block(trans
, root
, eb
, parent
, wc
->refs
[level
] == 1);
8031 wc
->refs
[level
] = 0;
8032 wc
->flags
[level
] = 0;
8036 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
8037 struct btrfs_root
*root
,
8038 struct btrfs_path
*path
,
8039 struct walk_control
*wc
)
8041 int level
= wc
->level
;
8042 int lookup_info
= 1;
8045 while (level
>= 0) {
8046 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
8053 if (path
->slots
[level
] >=
8054 btrfs_header_nritems(path
->nodes
[level
]))
8057 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
8059 path
->slots
[level
]++;
8068 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
8069 struct btrfs_root
*root
,
8070 struct btrfs_path
*path
,
8071 struct walk_control
*wc
, int max_level
)
8073 int level
= wc
->level
;
8076 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
8077 while (level
< max_level
&& path
->nodes
[level
]) {
8079 if (path
->slots
[level
] + 1 <
8080 btrfs_header_nritems(path
->nodes
[level
])) {
8081 path
->slots
[level
]++;
8084 ret
= walk_up_proc(trans
, root
, path
, wc
);
8088 if (path
->locks
[level
]) {
8089 btrfs_tree_unlock_rw(path
->nodes
[level
],
8090 path
->locks
[level
]);
8091 path
->locks
[level
] = 0;
8093 free_extent_buffer(path
->nodes
[level
]);
8094 path
->nodes
[level
] = NULL
;
8102 * drop a subvolume tree.
8104 * this function traverses the tree freeing any blocks that only
8105 * referenced by the tree.
8107 * when a shared tree block is found. this function decreases its
8108 * reference count by one. if update_ref is true, this function
8109 * also make sure backrefs for the shared block and all lower level
8110 * blocks are properly updated.
8112 * If called with for_reloc == 0, may exit early with -EAGAIN
8114 int btrfs_drop_snapshot(struct btrfs_root
*root
,
8115 struct btrfs_block_rsv
*block_rsv
, int update_ref
,
8118 struct btrfs_path
*path
;
8119 struct btrfs_trans_handle
*trans
;
8120 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
8121 struct btrfs_root_item
*root_item
= &root
->root_item
;
8122 struct walk_control
*wc
;
8123 struct btrfs_key key
;
8127 bool root_dropped
= false;
8129 btrfs_debug(root
->fs_info
, "Drop subvolume %llu", root
->objectid
);
8131 path
= btrfs_alloc_path();
8137 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
8139 btrfs_free_path(path
);
8144 trans
= btrfs_start_transaction(tree_root
, 0);
8145 if (IS_ERR(trans
)) {
8146 err
= PTR_ERR(trans
);
8151 trans
->block_rsv
= block_rsv
;
8153 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
8154 level
= btrfs_header_level(root
->node
);
8155 path
->nodes
[level
] = btrfs_lock_root_node(root
);
8156 btrfs_set_lock_blocking(path
->nodes
[level
]);
8157 path
->slots
[level
] = 0;
8158 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8159 memset(&wc
->update_progress
, 0,
8160 sizeof(wc
->update_progress
));
8162 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
8163 memcpy(&wc
->update_progress
, &key
,
8164 sizeof(wc
->update_progress
));
8166 level
= root_item
->drop_level
;
8168 path
->lowest_level
= level
;
8169 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
8170 path
->lowest_level
= 0;
8178 * unlock our path, this is safe because only this
8179 * function is allowed to delete this snapshot
8181 btrfs_unlock_up_safe(path
, 0);
8183 level
= btrfs_header_level(root
->node
);
8185 btrfs_tree_lock(path
->nodes
[level
]);
8186 btrfs_set_lock_blocking(path
->nodes
[level
]);
8187 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8189 ret
= btrfs_lookup_extent_info(trans
, root
,
8190 path
->nodes
[level
]->start
,
8191 level
, 1, &wc
->refs
[level
],
8197 BUG_ON(wc
->refs
[level
] == 0);
8199 if (level
== root_item
->drop_level
)
8202 btrfs_tree_unlock(path
->nodes
[level
]);
8203 path
->locks
[level
] = 0;
8204 WARN_ON(wc
->refs
[level
] != 1);
8210 wc
->shared_level
= -1;
8211 wc
->stage
= DROP_REFERENCE
;
8212 wc
->update_ref
= update_ref
;
8214 wc
->for_reloc
= for_reloc
;
8215 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
8219 ret
= walk_down_tree(trans
, root
, path
, wc
);
8225 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
8232 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
8236 if (wc
->stage
== DROP_REFERENCE
) {
8238 btrfs_node_key(path
->nodes
[level
],
8239 &root_item
->drop_progress
,
8240 path
->slots
[level
]);
8241 root_item
->drop_level
= level
;
8244 BUG_ON(wc
->level
== 0);
8245 if (btrfs_should_end_transaction(trans
, tree_root
) ||
8246 (!for_reloc
&& btrfs_need_cleaner_sleep(root
))) {
8247 ret
= btrfs_update_root(trans
, tree_root
,
8251 btrfs_abort_transaction(trans
, tree_root
, ret
);
8257 * Qgroup update accounting is run from
8258 * delayed ref handling. This usually works
8259 * out because delayed refs are normally the
8260 * only way qgroup updates are added. However,
8261 * we may have added updates during our tree
8262 * walk so run qgroups here to make sure we
8263 * don't lose any updates.
8265 ret
= btrfs_delayed_qgroup_accounting(trans
,
8268 printk_ratelimited(KERN_ERR
"BTRFS: Failure %d "
8269 "running qgroup updates "
8270 "during snapshot delete. "
8271 "Quota is out of sync, "
8272 "rescan required.\n", ret
);
8274 btrfs_end_transaction_throttle(trans
, tree_root
);
8275 if (!for_reloc
&& btrfs_need_cleaner_sleep(root
)) {
8276 pr_debug("BTRFS: drop snapshot early exit\n");
8281 trans
= btrfs_start_transaction(tree_root
, 0);
8282 if (IS_ERR(trans
)) {
8283 err
= PTR_ERR(trans
);
8287 trans
->block_rsv
= block_rsv
;
8290 btrfs_release_path(path
);
8294 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
8296 btrfs_abort_transaction(trans
, tree_root
, ret
);
8300 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
8301 ret
= btrfs_find_root(tree_root
, &root
->root_key
, path
,
8304 btrfs_abort_transaction(trans
, tree_root
, ret
);
8307 } else if (ret
> 0) {
8308 /* if we fail to delete the orphan item this time
8309 * around, it'll get picked up the next time.
8311 * The most common failure here is just -ENOENT.
8313 btrfs_del_orphan_item(trans
, tree_root
,
8314 root
->root_key
.objectid
);
8318 if (test_bit(BTRFS_ROOT_IN_RADIX
, &root
->state
)) {
8319 btrfs_drop_and_free_fs_root(tree_root
->fs_info
, root
);
8321 free_extent_buffer(root
->node
);
8322 free_extent_buffer(root
->commit_root
);
8323 btrfs_put_fs_root(root
);
8325 root_dropped
= true;
8327 ret
= btrfs_delayed_qgroup_accounting(trans
, tree_root
->fs_info
);
8329 printk_ratelimited(KERN_ERR
"BTRFS: Failure %d "
8330 "running qgroup updates "
8331 "during snapshot delete. "
8332 "Quota is out of sync, "
8333 "rescan required.\n", ret
);
8335 btrfs_end_transaction_throttle(trans
, tree_root
);
8338 btrfs_free_path(path
);
8341 * So if we need to stop dropping the snapshot for whatever reason we
8342 * need to make sure to add it back to the dead root list so that we
8343 * keep trying to do the work later. This also cleans up roots if we
8344 * don't have it in the radix (like when we recover after a power fail
8345 * or unmount) so we don't leak memory.
8347 if (!for_reloc
&& root_dropped
== false)
8348 btrfs_add_dead_root(root
);
8349 if (err
&& err
!= -EAGAIN
)
8350 btrfs_std_error(root
->fs_info
, err
);
8355 * drop subtree rooted at tree block 'node'.
8357 * NOTE: this function will unlock and release tree block 'node'
8358 * only used by relocation code
8360 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
8361 struct btrfs_root
*root
,
8362 struct extent_buffer
*node
,
8363 struct extent_buffer
*parent
)
8365 struct btrfs_path
*path
;
8366 struct walk_control
*wc
;
8372 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
8374 path
= btrfs_alloc_path();
8378 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
8380 btrfs_free_path(path
);
8384 btrfs_assert_tree_locked(parent
);
8385 parent_level
= btrfs_header_level(parent
);
8386 extent_buffer_get(parent
);
8387 path
->nodes
[parent_level
] = parent
;
8388 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
8390 btrfs_assert_tree_locked(node
);
8391 level
= btrfs_header_level(node
);
8392 path
->nodes
[level
] = node
;
8393 path
->slots
[level
] = 0;
8394 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8396 wc
->refs
[parent_level
] = 1;
8397 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
8399 wc
->shared_level
= -1;
8400 wc
->stage
= DROP_REFERENCE
;
8404 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
8407 wret
= walk_down_tree(trans
, root
, path
, wc
);
8413 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
8421 btrfs_free_path(path
);
8425 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
8431 * if restripe for this chunk_type is on pick target profile and
8432 * return, otherwise do the usual balance
8434 stripped
= get_restripe_target(root
->fs_info
, flags
);
8436 return extended_to_chunk(stripped
);
8438 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
8440 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
8441 BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
|
8442 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
8444 if (num_devices
== 1) {
8445 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
8446 stripped
= flags
& ~stripped
;
8448 /* turn raid0 into single device chunks */
8449 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
8452 /* turn mirroring into duplication */
8453 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
8454 BTRFS_BLOCK_GROUP_RAID10
))
8455 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
8457 /* they already had raid on here, just return */
8458 if (flags
& stripped
)
8461 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
8462 stripped
= flags
& ~stripped
;
8464 /* switch duplicated blocks with raid1 */
8465 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
8466 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
8468 /* this is drive concat, leave it alone */
8474 static int set_block_group_ro(struct btrfs_block_group_cache
*cache
, int force
)
8476 struct btrfs_space_info
*sinfo
= cache
->space_info
;
8478 u64 min_allocable_bytes
;
8483 * We need some metadata space and system metadata space for
8484 * allocating chunks in some corner cases until we force to set
8485 * it to be readonly.
8488 (BTRFS_BLOCK_GROUP_SYSTEM
| BTRFS_BLOCK_GROUP_METADATA
)) &&
8490 min_allocable_bytes
= 1 * 1024 * 1024;
8492 min_allocable_bytes
= 0;
8494 spin_lock(&sinfo
->lock
);
8495 spin_lock(&cache
->lock
);
8502 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
8503 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
8505 if (sinfo
->bytes_used
+ sinfo
->bytes_reserved
+ sinfo
->bytes_pinned
+
8506 sinfo
->bytes_may_use
+ sinfo
->bytes_readonly
+ num_bytes
+
8507 min_allocable_bytes
<= sinfo
->total_bytes
) {
8508 sinfo
->bytes_readonly
+= num_bytes
;
8513 spin_unlock(&cache
->lock
);
8514 spin_unlock(&sinfo
->lock
);
8518 int btrfs_set_block_group_ro(struct btrfs_root
*root
,
8519 struct btrfs_block_group_cache
*cache
)
8522 struct btrfs_trans_handle
*trans
;
8528 trans
= btrfs_join_transaction(root
);
8530 return PTR_ERR(trans
);
8532 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
8533 if (alloc_flags
!= cache
->flags
) {
8534 ret
= do_chunk_alloc(trans
, root
, alloc_flags
,
8540 ret
= set_block_group_ro(cache
, 0);
8543 alloc_flags
= get_alloc_profile(root
, cache
->space_info
->flags
);
8544 ret
= do_chunk_alloc(trans
, root
, alloc_flags
,
8548 ret
= set_block_group_ro(cache
, 0);
8550 btrfs_end_transaction(trans
, root
);
8554 int btrfs_force_chunk_alloc(struct btrfs_trans_handle
*trans
,
8555 struct btrfs_root
*root
, u64 type
)
8557 u64 alloc_flags
= get_alloc_profile(root
, type
);
8558 return do_chunk_alloc(trans
, root
, alloc_flags
,
8563 * helper to account the unused space of all the readonly block group in the
8564 * list. takes mirrors into account.
8566 static u64
__btrfs_get_ro_block_group_free_space(struct list_head
*groups_list
)
8568 struct btrfs_block_group_cache
*block_group
;
8572 list_for_each_entry(block_group
, groups_list
, list
) {
8573 spin_lock(&block_group
->lock
);
8575 if (!block_group
->ro
) {
8576 spin_unlock(&block_group
->lock
);
8580 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_RAID1
|
8581 BTRFS_BLOCK_GROUP_RAID10
|
8582 BTRFS_BLOCK_GROUP_DUP
))
8587 free_bytes
+= (block_group
->key
.offset
-
8588 btrfs_block_group_used(&block_group
->item
)) *
8591 spin_unlock(&block_group
->lock
);
8598 * helper to account the unused space of all the readonly block group in the
8599 * space_info. takes mirrors into account.
8601 u64
btrfs_account_ro_block_groups_free_space(struct btrfs_space_info
*sinfo
)
8606 spin_lock(&sinfo
->lock
);
8608 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
8609 if (!list_empty(&sinfo
->block_groups
[i
]))
8610 free_bytes
+= __btrfs_get_ro_block_group_free_space(
8611 &sinfo
->block_groups
[i
]);
8613 spin_unlock(&sinfo
->lock
);
8618 void btrfs_set_block_group_rw(struct btrfs_root
*root
,
8619 struct btrfs_block_group_cache
*cache
)
8621 struct btrfs_space_info
*sinfo
= cache
->space_info
;
8626 spin_lock(&sinfo
->lock
);
8627 spin_lock(&cache
->lock
);
8628 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
8629 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
8630 sinfo
->bytes_readonly
-= num_bytes
;
8632 spin_unlock(&cache
->lock
);
8633 spin_unlock(&sinfo
->lock
);
8637 * checks to see if its even possible to relocate this block group.
8639 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8640 * ok to go ahead and try.
8642 int btrfs_can_relocate(struct btrfs_root
*root
, u64 bytenr
)
8644 struct btrfs_block_group_cache
*block_group
;
8645 struct btrfs_space_info
*space_info
;
8646 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
8647 struct btrfs_device
*device
;
8648 struct btrfs_trans_handle
*trans
;
8657 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
8659 /* odd, couldn't find the block group, leave it alone */
8663 min_free
= btrfs_block_group_used(&block_group
->item
);
8665 /* no bytes used, we're good */
8669 space_info
= block_group
->space_info
;
8670 spin_lock(&space_info
->lock
);
8672 full
= space_info
->full
;
8675 * if this is the last block group we have in this space, we can't
8676 * relocate it unless we're able to allocate a new chunk below.
8678 * Otherwise, we need to make sure we have room in the space to handle
8679 * all of the extents from this block group. If we can, we're good
8681 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
8682 (space_info
->bytes_used
+ space_info
->bytes_reserved
+
8683 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
8684 min_free
< space_info
->total_bytes
)) {
8685 spin_unlock(&space_info
->lock
);
8688 spin_unlock(&space_info
->lock
);
8691 * ok we don't have enough space, but maybe we have free space on our
8692 * devices to allocate new chunks for relocation, so loop through our
8693 * alloc devices and guess if we have enough space. if this block
8694 * group is going to be restriped, run checks against the target
8695 * profile instead of the current one.
8707 target
= get_restripe_target(root
->fs_info
, block_group
->flags
);
8709 index
= __get_raid_index(extended_to_chunk(target
));
8712 * this is just a balance, so if we were marked as full
8713 * we know there is no space for a new chunk
8718 index
= get_block_group_index(block_group
);
8721 if (index
== BTRFS_RAID_RAID10
) {
8725 } else if (index
== BTRFS_RAID_RAID1
) {
8727 } else if (index
== BTRFS_RAID_DUP
) {
8730 } else if (index
== BTRFS_RAID_RAID0
) {
8731 dev_min
= fs_devices
->rw_devices
;
8732 do_div(min_free
, dev_min
);
8735 /* We need to do this so that we can look at pending chunks */
8736 trans
= btrfs_join_transaction(root
);
8737 if (IS_ERR(trans
)) {
8738 ret
= PTR_ERR(trans
);
8742 mutex_lock(&root
->fs_info
->chunk_mutex
);
8743 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
8747 * check to make sure we can actually find a chunk with enough
8748 * space to fit our block group in.
8750 if (device
->total_bytes
> device
->bytes_used
+ min_free
&&
8751 !device
->is_tgtdev_for_dev_replace
) {
8752 ret
= find_free_dev_extent(trans
, device
, min_free
,
8757 if (dev_nr
>= dev_min
)
8763 mutex_unlock(&root
->fs_info
->chunk_mutex
);
8764 btrfs_end_transaction(trans
, root
);
8766 btrfs_put_block_group(block_group
);
8770 static int find_first_block_group(struct btrfs_root
*root
,
8771 struct btrfs_path
*path
, struct btrfs_key
*key
)
8774 struct btrfs_key found_key
;
8775 struct extent_buffer
*leaf
;
8778 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
8783 slot
= path
->slots
[0];
8784 leaf
= path
->nodes
[0];
8785 if (slot
>= btrfs_header_nritems(leaf
)) {
8786 ret
= btrfs_next_leaf(root
, path
);
8793 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
8795 if (found_key
.objectid
>= key
->objectid
&&
8796 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
8806 void btrfs_put_block_group_cache(struct btrfs_fs_info
*info
)
8808 struct btrfs_block_group_cache
*block_group
;
8812 struct inode
*inode
;
8814 block_group
= btrfs_lookup_first_block_group(info
, last
);
8815 while (block_group
) {
8816 spin_lock(&block_group
->lock
);
8817 if (block_group
->iref
)
8819 spin_unlock(&block_group
->lock
);
8820 block_group
= next_block_group(info
->tree_root
,
8830 inode
= block_group
->inode
;
8831 block_group
->iref
= 0;
8832 block_group
->inode
= NULL
;
8833 spin_unlock(&block_group
->lock
);
8835 last
= block_group
->key
.objectid
+ block_group
->key
.offset
;
8836 btrfs_put_block_group(block_group
);
8840 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
8842 struct btrfs_block_group_cache
*block_group
;
8843 struct btrfs_space_info
*space_info
;
8844 struct btrfs_caching_control
*caching_ctl
;
8847 down_write(&info
->commit_root_sem
);
8848 while (!list_empty(&info
->caching_block_groups
)) {
8849 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
8850 struct btrfs_caching_control
, list
);
8851 list_del(&caching_ctl
->list
);
8852 put_caching_control(caching_ctl
);
8854 up_write(&info
->commit_root_sem
);
8856 spin_lock(&info
->block_group_cache_lock
);
8857 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
8858 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
8860 rb_erase(&block_group
->cache_node
,
8861 &info
->block_group_cache_tree
);
8862 spin_unlock(&info
->block_group_cache_lock
);
8864 down_write(&block_group
->space_info
->groups_sem
);
8865 list_del(&block_group
->list
);
8866 up_write(&block_group
->space_info
->groups_sem
);
8868 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
8869 wait_block_group_cache_done(block_group
);
8872 * We haven't cached this block group, which means we could
8873 * possibly have excluded extents on this block group.
8875 if (block_group
->cached
== BTRFS_CACHE_NO
||
8876 block_group
->cached
== BTRFS_CACHE_ERROR
)
8877 free_excluded_extents(info
->extent_root
, block_group
);
8879 btrfs_remove_free_space_cache(block_group
);
8880 btrfs_put_block_group(block_group
);
8882 spin_lock(&info
->block_group_cache_lock
);
8884 spin_unlock(&info
->block_group_cache_lock
);
8886 /* now that all the block groups are freed, go through and
8887 * free all the space_info structs. This is only called during
8888 * the final stages of unmount, and so we know nobody is
8889 * using them. We call synchronize_rcu() once before we start,
8890 * just to be on the safe side.
8894 release_global_block_rsv(info
);
8896 while (!list_empty(&info
->space_info
)) {
8899 space_info
= list_entry(info
->space_info
.next
,
8900 struct btrfs_space_info
,
8902 if (btrfs_test_opt(info
->tree_root
, ENOSPC_DEBUG
)) {
8903 if (WARN_ON(space_info
->bytes_pinned
> 0 ||
8904 space_info
->bytes_reserved
> 0 ||
8905 space_info
->bytes_may_use
> 0)) {
8906 dump_space_info(space_info
, 0, 0);
8909 list_del(&space_info
->list
);
8910 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++) {
8911 struct kobject
*kobj
;
8912 kobj
= space_info
->block_group_kobjs
[i
];
8913 space_info
->block_group_kobjs
[i
] = NULL
;
8919 kobject_del(&space_info
->kobj
);
8920 kobject_put(&space_info
->kobj
);
8925 static void __link_block_group(struct btrfs_space_info
*space_info
,
8926 struct btrfs_block_group_cache
*cache
)
8928 int index
= get_block_group_index(cache
);
8931 down_write(&space_info
->groups_sem
);
8932 if (list_empty(&space_info
->block_groups
[index
]))
8934 list_add_tail(&cache
->list
, &space_info
->block_groups
[index
]);
8935 up_write(&space_info
->groups_sem
);
8938 struct raid_kobject
*rkobj
;
8941 rkobj
= kzalloc(sizeof(*rkobj
), GFP_NOFS
);
8944 rkobj
->raid_type
= index
;
8945 kobject_init(&rkobj
->kobj
, &btrfs_raid_ktype
);
8946 ret
= kobject_add(&rkobj
->kobj
, &space_info
->kobj
,
8947 "%s", get_raid_name(index
));
8949 kobject_put(&rkobj
->kobj
);
8952 space_info
->block_group_kobjs
[index
] = &rkobj
->kobj
;
8957 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
8960 static struct btrfs_block_group_cache
*
8961 btrfs_create_block_group_cache(struct btrfs_root
*root
, u64 start
, u64 size
)
8963 struct btrfs_block_group_cache
*cache
;
8965 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
8969 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
8971 if (!cache
->free_space_ctl
) {
8976 cache
->key
.objectid
= start
;
8977 cache
->key
.offset
= size
;
8978 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
8980 cache
->sectorsize
= root
->sectorsize
;
8981 cache
->fs_info
= root
->fs_info
;
8982 cache
->full_stripe_len
= btrfs_full_stripe_len(root
,
8983 &root
->fs_info
->mapping_tree
,
8985 atomic_set(&cache
->count
, 1);
8986 spin_lock_init(&cache
->lock
);
8987 init_rwsem(&cache
->data_rwsem
);
8988 INIT_LIST_HEAD(&cache
->list
);
8989 INIT_LIST_HEAD(&cache
->cluster_list
);
8990 INIT_LIST_HEAD(&cache
->new_bg_list
);
8991 btrfs_init_free_space_ctl(cache
);
8996 int btrfs_read_block_groups(struct btrfs_root
*root
)
8998 struct btrfs_path
*path
;
9000 struct btrfs_block_group_cache
*cache
;
9001 struct btrfs_fs_info
*info
= root
->fs_info
;
9002 struct btrfs_space_info
*space_info
;
9003 struct btrfs_key key
;
9004 struct btrfs_key found_key
;
9005 struct extent_buffer
*leaf
;
9009 root
= info
->extent_root
;
9012 btrfs_set_key_type(&key
, BTRFS_BLOCK_GROUP_ITEM_KEY
);
9013 path
= btrfs_alloc_path();
9018 cache_gen
= btrfs_super_cache_generation(root
->fs_info
->super_copy
);
9019 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
9020 btrfs_super_generation(root
->fs_info
->super_copy
) != cache_gen
)
9022 if (btrfs_test_opt(root
, CLEAR_CACHE
))
9026 ret
= find_first_block_group(root
, path
, &key
);
9032 leaf
= path
->nodes
[0];
9033 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
9035 cache
= btrfs_create_block_group_cache(root
, found_key
.objectid
,
9044 * When we mount with old space cache, we need to
9045 * set BTRFS_DC_CLEAR and set dirty flag.
9047 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9048 * truncate the old free space cache inode and
9050 * b) Setting 'dirty flag' makes sure that we flush
9051 * the new space cache info onto disk.
9053 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
9054 if (btrfs_test_opt(root
, SPACE_CACHE
))
9058 read_extent_buffer(leaf
, &cache
->item
,
9059 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
9060 sizeof(cache
->item
));
9061 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
9063 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
9064 btrfs_release_path(path
);
9067 * We need to exclude the super stripes now so that the space
9068 * info has super bytes accounted for, otherwise we'll think
9069 * we have more space than we actually do.
9071 ret
= exclude_super_stripes(root
, cache
);
9074 * We may have excluded something, so call this just in
9077 free_excluded_extents(root
, cache
);
9078 btrfs_put_block_group(cache
);
9083 * check for two cases, either we are full, and therefore
9084 * don't need to bother with the caching work since we won't
9085 * find any space, or we are empty, and we can just add all
9086 * the space in and be done with it. This saves us _alot_ of
9087 * time, particularly in the full case.
9089 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
9090 cache
->last_byte_to_unpin
= (u64
)-1;
9091 cache
->cached
= BTRFS_CACHE_FINISHED
;
9092 free_excluded_extents(root
, cache
);
9093 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
9094 cache
->last_byte_to_unpin
= (u64
)-1;
9095 cache
->cached
= BTRFS_CACHE_FINISHED
;
9096 add_new_free_space(cache
, root
->fs_info
,
9098 found_key
.objectid
+
9100 free_excluded_extents(root
, cache
);
9103 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
9105 btrfs_remove_free_space_cache(cache
);
9106 btrfs_put_block_group(cache
);
9110 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
9111 btrfs_block_group_used(&cache
->item
),
9114 btrfs_remove_free_space_cache(cache
);
9115 spin_lock(&info
->block_group_cache_lock
);
9116 rb_erase(&cache
->cache_node
,
9117 &info
->block_group_cache_tree
);
9118 spin_unlock(&info
->block_group_cache_lock
);
9119 btrfs_put_block_group(cache
);
9123 cache
->space_info
= space_info
;
9124 spin_lock(&cache
->space_info
->lock
);
9125 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
9126 spin_unlock(&cache
->space_info
->lock
);
9128 __link_block_group(space_info
, cache
);
9130 set_avail_alloc_bits(root
->fs_info
, cache
->flags
);
9131 if (btrfs_chunk_readonly(root
, cache
->key
.objectid
))
9132 set_block_group_ro(cache
, 1);
9135 list_for_each_entry_rcu(space_info
, &root
->fs_info
->space_info
, list
) {
9136 if (!(get_alloc_profile(root
, space_info
->flags
) &
9137 (BTRFS_BLOCK_GROUP_RAID10
|
9138 BTRFS_BLOCK_GROUP_RAID1
|
9139 BTRFS_BLOCK_GROUP_RAID5
|
9140 BTRFS_BLOCK_GROUP_RAID6
|
9141 BTRFS_BLOCK_GROUP_DUP
)))
9144 * avoid allocating from un-mirrored block group if there are
9145 * mirrored block groups.
9147 list_for_each_entry(cache
,
9148 &space_info
->block_groups
[BTRFS_RAID_RAID0
],
9150 set_block_group_ro(cache
, 1);
9151 list_for_each_entry(cache
,
9152 &space_info
->block_groups
[BTRFS_RAID_SINGLE
],
9154 set_block_group_ro(cache
, 1);
9157 init_global_block_rsv(info
);
9160 btrfs_free_path(path
);
9164 void btrfs_create_pending_block_groups(struct btrfs_trans_handle
*trans
,
9165 struct btrfs_root
*root
)
9167 struct btrfs_block_group_cache
*block_group
, *tmp
;
9168 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
9169 struct btrfs_block_group_item item
;
9170 struct btrfs_key key
;
9173 list_for_each_entry_safe(block_group
, tmp
, &trans
->new_bgs
,
9175 list_del_init(&block_group
->new_bg_list
);
9180 spin_lock(&block_group
->lock
);
9181 memcpy(&item
, &block_group
->item
, sizeof(item
));
9182 memcpy(&key
, &block_group
->key
, sizeof(key
));
9183 spin_unlock(&block_group
->lock
);
9185 ret
= btrfs_insert_item(trans
, extent_root
, &key
, &item
,
9188 btrfs_abort_transaction(trans
, extent_root
, ret
);
9189 ret
= btrfs_finish_chunk_alloc(trans
, extent_root
,
9190 key
.objectid
, key
.offset
);
9192 btrfs_abort_transaction(trans
, extent_root
, ret
);
9196 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
9197 struct btrfs_root
*root
, u64 bytes_used
,
9198 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
9202 struct btrfs_root
*extent_root
;
9203 struct btrfs_block_group_cache
*cache
;
9205 extent_root
= root
->fs_info
->extent_root
;
9207 btrfs_set_log_full_commit(root
->fs_info
, trans
);
9209 cache
= btrfs_create_block_group_cache(root
, chunk_offset
, size
);
9213 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
9214 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
9215 btrfs_set_block_group_flags(&cache
->item
, type
);
9217 cache
->flags
= type
;
9218 cache
->last_byte_to_unpin
= (u64
)-1;
9219 cache
->cached
= BTRFS_CACHE_FINISHED
;
9220 ret
= exclude_super_stripes(root
, cache
);
9223 * We may have excluded something, so call this just in
9226 free_excluded_extents(root
, cache
);
9227 btrfs_put_block_group(cache
);
9231 add_new_free_space(cache
, root
->fs_info
, chunk_offset
,
9232 chunk_offset
+ size
);
9234 free_excluded_extents(root
, cache
);
9236 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
9238 btrfs_remove_free_space_cache(cache
);
9239 btrfs_put_block_group(cache
);
9243 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
9244 &cache
->space_info
);
9246 btrfs_remove_free_space_cache(cache
);
9247 spin_lock(&root
->fs_info
->block_group_cache_lock
);
9248 rb_erase(&cache
->cache_node
,
9249 &root
->fs_info
->block_group_cache_tree
);
9250 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
9251 btrfs_put_block_group(cache
);
9254 update_global_block_rsv(root
->fs_info
);
9256 spin_lock(&cache
->space_info
->lock
);
9257 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
9258 spin_unlock(&cache
->space_info
->lock
);
9260 __link_block_group(cache
->space_info
, cache
);
9262 list_add_tail(&cache
->new_bg_list
, &trans
->new_bgs
);
9264 set_avail_alloc_bits(extent_root
->fs_info
, type
);
9269 static void clear_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
9271 u64 extra_flags
= chunk_to_extended(flags
) &
9272 BTRFS_EXTENDED_PROFILE_MASK
;
9274 write_seqlock(&fs_info
->profiles_lock
);
9275 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
9276 fs_info
->avail_data_alloc_bits
&= ~extra_flags
;
9277 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
9278 fs_info
->avail_metadata_alloc_bits
&= ~extra_flags
;
9279 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
9280 fs_info
->avail_system_alloc_bits
&= ~extra_flags
;
9281 write_sequnlock(&fs_info
->profiles_lock
);
9284 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
9285 struct btrfs_root
*root
, u64 group_start
)
9287 struct btrfs_path
*path
;
9288 struct btrfs_block_group_cache
*block_group
;
9289 struct btrfs_free_cluster
*cluster
;
9290 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
9291 struct btrfs_key key
;
9292 struct inode
*inode
;
9293 struct kobject
*kobj
= NULL
;
9298 root
= root
->fs_info
->extent_root
;
9300 block_group
= btrfs_lookup_block_group(root
->fs_info
, group_start
);
9301 BUG_ON(!block_group
);
9302 BUG_ON(!block_group
->ro
);
9305 * Free the reserved super bytes from this block group before
9308 free_excluded_extents(root
, block_group
);
9310 memcpy(&key
, &block_group
->key
, sizeof(key
));
9311 index
= get_block_group_index(block_group
);
9312 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
9313 BTRFS_BLOCK_GROUP_RAID1
|
9314 BTRFS_BLOCK_GROUP_RAID10
))
9319 /* make sure this block group isn't part of an allocation cluster */
9320 cluster
= &root
->fs_info
->data_alloc_cluster
;
9321 spin_lock(&cluster
->refill_lock
);
9322 btrfs_return_cluster_to_free_space(block_group
, cluster
);
9323 spin_unlock(&cluster
->refill_lock
);
9326 * make sure this block group isn't part of a metadata
9327 * allocation cluster
9329 cluster
= &root
->fs_info
->meta_alloc_cluster
;
9330 spin_lock(&cluster
->refill_lock
);
9331 btrfs_return_cluster_to_free_space(block_group
, cluster
);
9332 spin_unlock(&cluster
->refill_lock
);
9334 path
= btrfs_alloc_path();
9340 inode
= lookup_free_space_inode(tree_root
, block_group
, path
);
9341 if (!IS_ERR(inode
)) {
9342 ret
= btrfs_orphan_add(trans
, inode
);
9344 btrfs_add_delayed_iput(inode
);
9348 /* One for the block groups ref */
9349 spin_lock(&block_group
->lock
);
9350 if (block_group
->iref
) {
9351 block_group
->iref
= 0;
9352 block_group
->inode
= NULL
;
9353 spin_unlock(&block_group
->lock
);
9356 spin_unlock(&block_group
->lock
);
9358 /* One for our lookup ref */
9359 btrfs_add_delayed_iput(inode
);
9362 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
9363 key
.offset
= block_group
->key
.objectid
;
9366 ret
= btrfs_search_slot(trans
, tree_root
, &key
, path
, -1, 1);
9370 btrfs_release_path(path
);
9372 ret
= btrfs_del_item(trans
, tree_root
, path
);
9375 btrfs_release_path(path
);
9378 spin_lock(&root
->fs_info
->block_group_cache_lock
);
9379 rb_erase(&block_group
->cache_node
,
9380 &root
->fs_info
->block_group_cache_tree
);
9382 if (root
->fs_info
->first_logical_byte
== block_group
->key
.objectid
)
9383 root
->fs_info
->first_logical_byte
= (u64
)-1;
9384 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
9386 down_write(&block_group
->space_info
->groups_sem
);
9388 * we must use list_del_init so people can check to see if they
9389 * are still on the list after taking the semaphore
9391 list_del_init(&block_group
->list
);
9392 if (list_empty(&block_group
->space_info
->block_groups
[index
])) {
9393 kobj
= block_group
->space_info
->block_group_kobjs
[index
];
9394 block_group
->space_info
->block_group_kobjs
[index
] = NULL
;
9395 clear_avail_alloc_bits(root
->fs_info
, block_group
->flags
);
9397 up_write(&block_group
->space_info
->groups_sem
);
9403 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
9404 wait_block_group_cache_done(block_group
);
9406 btrfs_remove_free_space_cache(block_group
);
9408 spin_lock(&block_group
->space_info
->lock
);
9409 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
9410 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
9411 block_group
->space_info
->disk_total
-= block_group
->key
.offset
* factor
;
9412 spin_unlock(&block_group
->space_info
->lock
);
9414 memcpy(&key
, &block_group
->key
, sizeof(key
));
9416 btrfs_clear_space_info_full(root
->fs_info
);
9418 btrfs_put_block_group(block_group
);
9419 btrfs_put_block_group(block_group
);
9421 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
9427 ret
= btrfs_del_item(trans
, root
, path
);
9429 btrfs_free_path(path
);
9433 int btrfs_init_space_info(struct btrfs_fs_info
*fs_info
)
9435 struct btrfs_space_info
*space_info
;
9436 struct btrfs_super_block
*disk_super
;
9442 disk_super
= fs_info
->super_copy
;
9443 if (!btrfs_super_root(disk_super
))
9446 features
= btrfs_super_incompat_flags(disk_super
);
9447 if (features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
9450 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
9451 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
9456 flags
= BTRFS_BLOCK_GROUP_METADATA
| BTRFS_BLOCK_GROUP_DATA
;
9457 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
9459 flags
= BTRFS_BLOCK_GROUP_METADATA
;
9460 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
9464 flags
= BTRFS_BLOCK_GROUP_DATA
;
9465 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
9471 int btrfs_error_unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
9473 return unpin_extent_range(root
, start
, end
);
9476 int btrfs_error_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
9477 u64 num_bytes
, u64
*actual_bytes
)
9479 return btrfs_discard_extent(root
, bytenr
, num_bytes
, actual_bytes
);
9482 int btrfs_trim_fs(struct btrfs_root
*root
, struct fstrim_range
*range
)
9484 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
9485 struct btrfs_block_group_cache
*cache
= NULL
;
9490 u64 total_bytes
= btrfs_super_total_bytes(fs_info
->super_copy
);
9494 * try to trim all FS space, our block group may start from non-zero.
9496 if (range
->len
== total_bytes
)
9497 cache
= btrfs_lookup_first_block_group(fs_info
, range
->start
);
9499 cache
= btrfs_lookup_block_group(fs_info
, range
->start
);
9502 if (cache
->key
.objectid
>= (range
->start
+ range
->len
)) {
9503 btrfs_put_block_group(cache
);
9507 start
= max(range
->start
, cache
->key
.objectid
);
9508 end
= min(range
->start
+ range
->len
,
9509 cache
->key
.objectid
+ cache
->key
.offset
);
9511 if (end
- start
>= range
->minlen
) {
9512 if (!block_group_cache_done(cache
)) {
9513 ret
= cache_block_group(cache
, 0);
9515 btrfs_put_block_group(cache
);
9518 ret
= wait_block_group_cache_done(cache
);
9520 btrfs_put_block_group(cache
);
9524 ret
= btrfs_trim_block_group(cache
,
9530 trimmed
+= group_trimmed
;
9532 btrfs_put_block_group(cache
);
9537 cache
= next_block_group(fs_info
->tree_root
, cache
);
9540 range
->len
= trimmed
;
9545 * btrfs_{start,end}_write() is similar to mnt_{want, drop}_write(),
9546 * they are used to prevent the some tasks writing data into the page cache
9547 * by nocow before the subvolume is snapshoted, but flush the data into
9548 * the disk after the snapshot creation.
9550 void btrfs_end_nocow_write(struct btrfs_root
*root
)
9552 percpu_counter_dec(&root
->subv_writers
->counter
);
9554 * Make sure counter is updated before we wake up
9558 if (waitqueue_active(&root
->subv_writers
->wait
))
9559 wake_up(&root
->subv_writers
->wait
);
9562 int btrfs_start_nocow_write(struct btrfs_root
*root
)
9564 if (unlikely(atomic_read(&root
->will_be_snapshoted
)))
9567 percpu_counter_inc(&root
->subv_writers
->counter
);
9569 * Make sure counter is updated before we check for snapshot creation.
9572 if (unlikely(atomic_read(&root
->will_be_snapshoted
))) {
9573 btrfs_end_nocow_write(root
);