2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include <linux/slab.h>
26 #include <linux/ratelimit.h>
27 #include <linux/percpu_counter.h>
31 #include "print-tree.h"
35 #include "free-space-cache.h"
36 #include "free-space-tree.h"
41 #undef SCRAMBLE_DELAYED_REFS
44 * control flags for do_chunk_alloc's force field
45 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
46 * if we really need one.
48 * CHUNK_ALLOC_LIMITED means to only try and allocate one
49 * if we have very few chunks already allocated. This is
50 * used as part of the clustering code to help make sure
51 * we have a good pool of storage to cluster in, without
52 * filling the FS with empty chunks
54 * CHUNK_ALLOC_FORCE means it must try to allocate one
58 CHUNK_ALLOC_NO_FORCE
= 0,
59 CHUNK_ALLOC_LIMITED
= 1,
60 CHUNK_ALLOC_FORCE
= 2,
63 static int update_block_group(struct btrfs_trans_handle
*trans
,
64 struct btrfs_root
*root
, u64 bytenr
,
65 u64 num_bytes
, int alloc
);
66 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
67 struct btrfs_root
*root
,
68 struct btrfs_delayed_ref_node
*node
, u64 parent
,
69 u64 root_objectid
, u64 owner_objectid
,
70 u64 owner_offset
, int refs_to_drop
,
71 struct btrfs_delayed_extent_op
*extra_op
);
72 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
73 struct extent_buffer
*leaf
,
74 struct btrfs_extent_item
*ei
);
75 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
76 struct btrfs_root
*root
,
77 u64 parent
, u64 root_objectid
,
78 u64 flags
, u64 owner
, u64 offset
,
79 struct btrfs_key
*ins
, int ref_mod
);
80 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
81 struct btrfs_root
*root
,
82 u64 parent
, u64 root_objectid
,
83 u64 flags
, struct btrfs_disk_key
*key
,
84 int level
, struct btrfs_key
*ins
);
85 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
86 struct btrfs_root
*extent_root
, u64 flags
,
88 static int find_next_key(struct btrfs_path
*path
, int level
,
89 struct btrfs_key
*key
);
90 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
91 int dump_block_groups
);
92 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache
*cache
,
93 u64 ram_bytes
, u64 num_bytes
, int delalloc
);
94 static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache
*cache
,
95 u64 num_bytes
, int delalloc
);
96 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
98 int btrfs_pin_extent(struct btrfs_root
*root
,
99 u64 bytenr
, u64 num_bytes
, int reserved
);
100 static int __reserve_metadata_bytes(struct btrfs_root
*root
,
101 struct btrfs_space_info
*space_info
,
103 enum btrfs_reserve_flush_enum flush
);
104 static void space_info_add_new_bytes(struct btrfs_fs_info
*fs_info
,
105 struct btrfs_space_info
*space_info
,
107 static void space_info_add_old_bytes(struct btrfs_fs_info
*fs_info
,
108 struct btrfs_space_info
*space_info
,
112 block_group_cache_done(struct btrfs_block_group_cache
*cache
)
115 return cache
->cached
== BTRFS_CACHE_FINISHED
||
116 cache
->cached
== BTRFS_CACHE_ERROR
;
119 static int block_group_bits(struct btrfs_block_group_cache
*cache
, u64 bits
)
121 return (cache
->flags
& bits
) == bits
;
124 void btrfs_get_block_group(struct btrfs_block_group_cache
*cache
)
126 atomic_inc(&cache
->count
);
129 void btrfs_put_block_group(struct btrfs_block_group_cache
*cache
)
131 if (atomic_dec_and_test(&cache
->count
)) {
132 WARN_ON(cache
->pinned
> 0);
133 WARN_ON(cache
->reserved
> 0);
134 kfree(cache
->free_space_ctl
);
140 * this adds the block group to the fs_info rb tree for the block group
143 static int btrfs_add_block_group_cache(struct btrfs_fs_info
*info
,
144 struct btrfs_block_group_cache
*block_group
)
147 struct rb_node
*parent
= NULL
;
148 struct btrfs_block_group_cache
*cache
;
150 spin_lock(&info
->block_group_cache_lock
);
151 p
= &info
->block_group_cache_tree
.rb_node
;
155 cache
= rb_entry(parent
, struct btrfs_block_group_cache
,
157 if (block_group
->key
.objectid
< cache
->key
.objectid
) {
159 } else if (block_group
->key
.objectid
> cache
->key
.objectid
) {
162 spin_unlock(&info
->block_group_cache_lock
);
167 rb_link_node(&block_group
->cache_node
, parent
, p
);
168 rb_insert_color(&block_group
->cache_node
,
169 &info
->block_group_cache_tree
);
171 if (info
->first_logical_byte
> block_group
->key
.objectid
)
172 info
->first_logical_byte
= block_group
->key
.objectid
;
174 spin_unlock(&info
->block_group_cache_lock
);
180 * This will return the block group at or after bytenr if contains is 0, else
181 * it will return the block group that contains the bytenr
183 static struct btrfs_block_group_cache
*
184 block_group_cache_tree_search(struct btrfs_fs_info
*info
, u64 bytenr
,
187 struct btrfs_block_group_cache
*cache
, *ret
= NULL
;
191 spin_lock(&info
->block_group_cache_lock
);
192 n
= info
->block_group_cache_tree
.rb_node
;
195 cache
= rb_entry(n
, struct btrfs_block_group_cache
,
197 end
= cache
->key
.objectid
+ cache
->key
.offset
- 1;
198 start
= cache
->key
.objectid
;
200 if (bytenr
< start
) {
201 if (!contains
&& (!ret
|| start
< ret
->key
.objectid
))
204 } else if (bytenr
> start
) {
205 if (contains
&& bytenr
<= end
) {
216 btrfs_get_block_group(ret
);
217 if (bytenr
== 0 && info
->first_logical_byte
> ret
->key
.objectid
)
218 info
->first_logical_byte
= ret
->key
.objectid
;
220 spin_unlock(&info
->block_group_cache_lock
);
225 static int add_excluded_extent(struct btrfs_root
*root
,
226 u64 start
, u64 num_bytes
)
228 u64 end
= start
+ num_bytes
- 1;
229 set_extent_bits(&root
->fs_info
->freed_extents
[0],
230 start
, end
, EXTENT_UPTODATE
);
231 set_extent_bits(&root
->fs_info
->freed_extents
[1],
232 start
, end
, EXTENT_UPTODATE
);
236 static void free_excluded_extents(struct btrfs_root
*root
,
237 struct btrfs_block_group_cache
*cache
)
241 start
= cache
->key
.objectid
;
242 end
= start
+ cache
->key
.offset
- 1;
244 clear_extent_bits(&root
->fs_info
->freed_extents
[0],
245 start
, end
, EXTENT_UPTODATE
);
246 clear_extent_bits(&root
->fs_info
->freed_extents
[1],
247 start
, end
, EXTENT_UPTODATE
);
250 static int exclude_super_stripes(struct btrfs_root
*root
,
251 struct btrfs_block_group_cache
*cache
)
258 if (cache
->key
.objectid
< BTRFS_SUPER_INFO_OFFSET
) {
259 stripe_len
= BTRFS_SUPER_INFO_OFFSET
- cache
->key
.objectid
;
260 cache
->bytes_super
+= stripe_len
;
261 ret
= add_excluded_extent(root
, cache
->key
.objectid
,
267 for (i
= 0; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
268 bytenr
= btrfs_sb_offset(i
);
269 ret
= btrfs_rmap_block(&root
->fs_info
->mapping_tree
,
270 cache
->key
.objectid
, bytenr
,
271 0, &logical
, &nr
, &stripe_len
);
278 if (logical
[nr
] > cache
->key
.objectid
+
282 if (logical
[nr
] + stripe_len
<= cache
->key
.objectid
)
286 if (start
< cache
->key
.objectid
) {
287 start
= cache
->key
.objectid
;
288 len
= (logical
[nr
] + stripe_len
) - start
;
290 len
= min_t(u64
, stripe_len
,
291 cache
->key
.objectid
+
292 cache
->key
.offset
- start
);
295 cache
->bytes_super
+= len
;
296 ret
= add_excluded_extent(root
, start
, len
);
308 static struct btrfs_caching_control
*
309 get_caching_control(struct btrfs_block_group_cache
*cache
)
311 struct btrfs_caching_control
*ctl
;
313 spin_lock(&cache
->lock
);
314 if (!cache
->caching_ctl
) {
315 spin_unlock(&cache
->lock
);
319 ctl
= cache
->caching_ctl
;
320 atomic_inc(&ctl
->count
);
321 spin_unlock(&cache
->lock
);
325 static void put_caching_control(struct btrfs_caching_control
*ctl
)
327 if (atomic_dec_and_test(&ctl
->count
))
331 #ifdef CONFIG_BTRFS_DEBUG
332 static void fragment_free_space(struct btrfs_root
*root
,
333 struct btrfs_block_group_cache
*block_group
)
335 u64 start
= block_group
->key
.objectid
;
336 u64 len
= block_group
->key
.offset
;
337 u64 chunk
= block_group
->flags
& BTRFS_BLOCK_GROUP_METADATA
?
338 root
->nodesize
: root
->sectorsize
;
339 u64 step
= chunk
<< 1;
341 while (len
> chunk
) {
342 btrfs_remove_free_space(block_group
, start
, chunk
);
353 * this is only called by cache_block_group, since we could have freed extents
354 * we need to check the pinned_extents for any extents that can't be used yet
355 * since their free space will be released as soon as the transaction commits.
357 u64
add_new_free_space(struct btrfs_block_group_cache
*block_group
,
358 struct btrfs_fs_info
*info
, u64 start
, u64 end
)
360 u64 extent_start
, extent_end
, size
, total_added
= 0;
363 while (start
< end
) {
364 ret
= find_first_extent_bit(info
->pinned_extents
, start
,
365 &extent_start
, &extent_end
,
366 EXTENT_DIRTY
| EXTENT_UPTODATE
,
371 if (extent_start
<= start
) {
372 start
= extent_end
+ 1;
373 } else if (extent_start
> start
&& extent_start
< end
) {
374 size
= extent_start
- start
;
376 ret
= btrfs_add_free_space(block_group
, start
,
378 BUG_ON(ret
); /* -ENOMEM or logic error */
379 start
= extent_end
+ 1;
388 ret
= btrfs_add_free_space(block_group
, start
, size
);
389 BUG_ON(ret
); /* -ENOMEM or logic error */
395 static int load_extent_tree_free(struct btrfs_caching_control
*caching_ctl
)
397 struct btrfs_block_group_cache
*block_group
;
398 struct btrfs_fs_info
*fs_info
;
399 struct btrfs_root
*extent_root
;
400 struct btrfs_path
*path
;
401 struct extent_buffer
*leaf
;
402 struct btrfs_key key
;
409 block_group
= caching_ctl
->block_group
;
410 fs_info
= block_group
->fs_info
;
411 extent_root
= fs_info
->extent_root
;
413 path
= btrfs_alloc_path();
417 last
= max_t(u64
, block_group
->key
.objectid
, BTRFS_SUPER_INFO_OFFSET
);
419 #ifdef CONFIG_BTRFS_DEBUG
421 * If we're fragmenting we don't want to make anybody think we can
422 * allocate from this block group until we've had a chance to fragment
425 if (btrfs_should_fragment_free_space(extent_root
, block_group
))
429 * We don't want to deadlock with somebody trying to allocate a new
430 * extent for the extent root while also trying to search the extent
431 * root to add free space. So we skip locking and search the commit
432 * root, since its read-only
434 path
->skip_locking
= 1;
435 path
->search_commit_root
= 1;
436 path
->reada
= READA_FORWARD
;
440 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
443 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
447 leaf
= path
->nodes
[0];
448 nritems
= btrfs_header_nritems(leaf
);
451 if (btrfs_fs_closing(fs_info
) > 1) {
456 if (path
->slots
[0] < nritems
) {
457 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
459 ret
= find_next_key(path
, 0, &key
);
463 if (need_resched() ||
464 rwsem_is_contended(&fs_info
->commit_root_sem
)) {
466 caching_ctl
->progress
= last
;
467 btrfs_release_path(path
);
468 up_read(&fs_info
->commit_root_sem
);
469 mutex_unlock(&caching_ctl
->mutex
);
471 mutex_lock(&caching_ctl
->mutex
);
472 down_read(&fs_info
->commit_root_sem
);
476 ret
= btrfs_next_leaf(extent_root
, path
);
481 leaf
= path
->nodes
[0];
482 nritems
= btrfs_header_nritems(leaf
);
486 if (key
.objectid
< last
) {
489 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
492 caching_ctl
->progress
= last
;
493 btrfs_release_path(path
);
497 if (key
.objectid
< block_group
->key
.objectid
) {
502 if (key
.objectid
>= block_group
->key
.objectid
+
503 block_group
->key
.offset
)
506 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
||
507 key
.type
== BTRFS_METADATA_ITEM_KEY
) {
508 total_found
+= add_new_free_space(block_group
,
511 if (key
.type
== BTRFS_METADATA_ITEM_KEY
)
512 last
= key
.objectid
+
513 fs_info
->tree_root
->nodesize
;
515 last
= key
.objectid
+ key
.offset
;
517 if (total_found
> CACHING_CTL_WAKE_UP
) {
520 wake_up(&caching_ctl
->wait
);
527 total_found
+= add_new_free_space(block_group
, fs_info
, last
,
528 block_group
->key
.objectid
+
529 block_group
->key
.offset
);
530 caching_ctl
->progress
= (u64
)-1;
533 btrfs_free_path(path
);
537 static noinline
void caching_thread(struct btrfs_work
*work
)
539 struct btrfs_block_group_cache
*block_group
;
540 struct btrfs_fs_info
*fs_info
;
541 struct btrfs_caching_control
*caching_ctl
;
542 struct btrfs_root
*extent_root
;
545 caching_ctl
= container_of(work
, struct btrfs_caching_control
, work
);
546 block_group
= caching_ctl
->block_group
;
547 fs_info
= block_group
->fs_info
;
548 extent_root
= fs_info
->extent_root
;
550 mutex_lock(&caching_ctl
->mutex
);
551 down_read(&fs_info
->commit_root_sem
);
553 if (btrfs_fs_compat_ro(fs_info
, FREE_SPACE_TREE
))
554 ret
= load_free_space_tree(caching_ctl
);
556 ret
= load_extent_tree_free(caching_ctl
);
558 spin_lock(&block_group
->lock
);
559 block_group
->caching_ctl
= NULL
;
560 block_group
->cached
= ret
? BTRFS_CACHE_ERROR
: BTRFS_CACHE_FINISHED
;
561 spin_unlock(&block_group
->lock
);
563 #ifdef CONFIG_BTRFS_DEBUG
564 if (btrfs_should_fragment_free_space(extent_root
, block_group
)) {
567 spin_lock(&block_group
->space_info
->lock
);
568 spin_lock(&block_group
->lock
);
569 bytes_used
= block_group
->key
.offset
-
570 btrfs_block_group_used(&block_group
->item
);
571 block_group
->space_info
->bytes_used
+= bytes_used
>> 1;
572 spin_unlock(&block_group
->lock
);
573 spin_unlock(&block_group
->space_info
->lock
);
574 fragment_free_space(extent_root
, block_group
);
578 caching_ctl
->progress
= (u64
)-1;
580 up_read(&fs_info
->commit_root_sem
);
581 free_excluded_extents(fs_info
->extent_root
, block_group
);
582 mutex_unlock(&caching_ctl
->mutex
);
584 wake_up(&caching_ctl
->wait
);
586 put_caching_control(caching_ctl
);
587 btrfs_put_block_group(block_group
);
590 static int cache_block_group(struct btrfs_block_group_cache
*cache
,
594 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
595 struct btrfs_caching_control
*caching_ctl
;
598 caching_ctl
= kzalloc(sizeof(*caching_ctl
), GFP_NOFS
);
602 INIT_LIST_HEAD(&caching_ctl
->list
);
603 mutex_init(&caching_ctl
->mutex
);
604 init_waitqueue_head(&caching_ctl
->wait
);
605 caching_ctl
->block_group
= cache
;
606 caching_ctl
->progress
= cache
->key
.objectid
;
607 atomic_set(&caching_ctl
->count
, 1);
608 btrfs_init_work(&caching_ctl
->work
, btrfs_cache_helper
,
609 caching_thread
, NULL
, NULL
);
611 spin_lock(&cache
->lock
);
613 * This should be a rare occasion, but this could happen I think in the
614 * case where one thread starts to load the space cache info, and then
615 * some other thread starts a transaction commit which tries to do an
616 * allocation while the other thread is still loading the space cache
617 * info. The previous loop should have kept us from choosing this block
618 * group, but if we've moved to the state where we will wait on caching
619 * block groups we need to first check if we're doing a fast load here,
620 * so we can wait for it to finish, otherwise we could end up allocating
621 * from a block group who's cache gets evicted for one reason or
624 while (cache
->cached
== BTRFS_CACHE_FAST
) {
625 struct btrfs_caching_control
*ctl
;
627 ctl
= cache
->caching_ctl
;
628 atomic_inc(&ctl
->count
);
629 prepare_to_wait(&ctl
->wait
, &wait
, TASK_UNINTERRUPTIBLE
);
630 spin_unlock(&cache
->lock
);
634 finish_wait(&ctl
->wait
, &wait
);
635 put_caching_control(ctl
);
636 spin_lock(&cache
->lock
);
639 if (cache
->cached
!= BTRFS_CACHE_NO
) {
640 spin_unlock(&cache
->lock
);
644 WARN_ON(cache
->caching_ctl
);
645 cache
->caching_ctl
= caching_ctl
;
646 cache
->cached
= BTRFS_CACHE_FAST
;
647 spin_unlock(&cache
->lock
);
649 if (fs_info
->mount_opt
& BTRFS_MOUNT_SPACE_CACHE
) {
650 mutex_lock(&caching_ctl
->mutex
);
651 ret
= load_free_space_cache(fs_info
, cache
);
653 spin_lock(&cache
->lock
);
655 cache
->caching_ctl
= NULL
;
656 cache
->cached
= BTRFS_CACHE_FINISHED
;
657 cache
->last_byte_to_unpin
= (u64
)-1;
658 caching_ctl
->progress
= (u64
)-1;
660 if (load_cache_only
) {
661 cache
->caching_ctl
= NULL
;
662 cache
->cached
= BTRFS_CACHE_NO
;
664 cache
->cached
= BTRFS_CACHE_STARTED
;
665 cache
->has_caching_ctl
= 1;
668 spin_unlock(&cache
->lock
);
669 #ifdef CONFIG_BTRFS_DEBUG
671 btrfs_should_fragment_free_space(fs_info
->extent_root
,
675 spin_lock(&cache
->space_info
->lock
);
676 spin_lock(&cache
->lock
);
677 bytes_used
= cache
->key
.offset
-
678 btrfs_block_group_used(&cache
->item
);
679 cache
->space_info
->bytes_used
+= bytes_used
>> 1;
680 spin_unlock(&cache
->lock
);
681 spin_unlock(&cache
->space_info
->lock
);
682 fragment_free_space(fs_info
->extent_root
, cache
);
685 mutex_unlock(&caching_ctl
->mutex
);
687 wake_up(&caching_ctl
->wait
);
689 put_caching_control(caching_ctl
);
690 free_excluded_extents(fs_info
->extent_root
, cache
);
695 * We're either using the free space tree or no caching at all.
696 * Set cached to the appropriate value and wakeup any waiters.
698 spin_lock(&cache
->lock
);
699 if (load_cache_only
) {
700 cache
->caching_ctl
= NULL
;
701 cache
->cached
= BTRFS_CACHE_NO
;
703 cache
->cached
= BTRFS_CACHE_STARTED
;
704 cache
->has_caching_ctl
= 1;
706 spin_unlock(&cache
->lock
);
707 wake_up(&caching_ctl
->wait
);
710 if (load_cache_only
) {
711 put_caching_control(caching_ctl
);
715 down_write(&fs_info
->commit_root_sem
);
716 atomic_inc(&caching_ctl
->count
);
717 list_add_tail(&caching_ctl
->list
, &fs_info
->caching_block_groups
);
718 up_write(&fs_info
->commit_root_sem
);
720 btrfs_get_block_group(cache
);
722 btrfs_queue_work(fs_info
->caching_workers
, &caching_ctl
->work
);
728 * return the block group that starts at or after bytenr
730 static struct btrfs_block_group_cache
*
731 btrfs_lookup_first_block_group(struct btrfs_fs_info
*info
, u64 bytenr
)
733 struct btrfs_block_group_cache
*cache
;
735 cache
= block_group_cache_tree_search(info
, bytenr
, 0);
741 * return the block group that contains the given bytenr
743 struct btrfs_block_group_cache
*btrfs_lookup_block_group(
744 struct btrfs_fs_info
*info
,
747 struct btrfs_block_group_cache
*cache
;
749 cache
= block_group_cache_tree_search(info
, bytenr
, 1);
754 static struct btrfs_space_info
*__find_space_info(struct btrfs_fs_info
*info
,
757 struct list_head
*head
= &info
->space_info
;
758 struct btrfs_space_info
*found
;
760 flags
&= BTRFS_BLOCK_GROUP_TYPE_MASK
;
763 list_for_each_entry_rcu(found
, head
, list
) {
764 if (found
->flags
& flags
) {
774 * after adding space to the filesystem, we need to clear the full flags
775 * on all the space infos.
777 void btrfs_clear_space_info_full(struct btrfs_fs_info
*info
)
779 struct list_head
*head
= &info
->space_info
;
780 struct btrfs_space_info
*found
;
783 list_for_each_entry_rcu(found
, head
, list
)
788 /* simple helper to search for an existing data extent at a given offset */
789 int btrfs_lookup_data_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
792 struct btrfs_key key
;
793 struct btrfs_path
*path
;
795 path
= btrfs_alloc_path();
799 key
.objectid
= start
;
801 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
802 ret
= btrfs_search_slot(NULL
, root
->fs_info
->extent_root
, &key
, path
,
804 btrfs_free_path(path
);
809 * helper function to lookup reference count and flags of a tree block.
811 * the head node for delayed ref is used to store the sum of all the
812 * reference count modifications queued up in the rbtree. the head
813 * node may also store the extent flags to set. This way you can check
814 * to see what the reference count and extent flags would be if all of
815 * the delayed refs are not processed.
817 int btrfs_lookup_extent_info(struct btrfs_trans_handle
*trans
,
818 struct btrfs_root
*root
, u64 bytenr
,
819 u64 offset
, int metadata
, u64
*refs
, u64
*flags
)
821 struct btrfs_delayed_ref_head
*head
;
822 struct btrfs_delayed_ref_root
*delayed_refs
;
823 struct btrfs_path
*path
;
824 struct btrfs_extent_item
*ei
;
825 struct extent_buffer
*leaf
;
826 struct btrfs_key key
;
833 * If we don't have skinny metadata, don't bother doing anything
836 if (metadata
&& !btrfs_fs_incompat(root
->fs_info
, SKINNY_METADATA
)) {
837 offset
= root
->nodesize
;
841 path
= btrfs_alloc_path();
846 path
->skip_locking
= 1;
847 path
->search_commit_root
= 1;
851 key
.objectid
= bytenr
;
854 key
.type
= BTRFS_METADATA_ITEM_KEY
;
856 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
858 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
,
863 if (ret
> 0 && metadata
&& key
.type
== BTRFS_METADATA_ITEM_KEY
) {
864 if (path
->slots
[0]) {
866 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
868 if (key
.objectid
== bytenr
&&
869 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
870 key
.offset
== root
->nodesize
)
876 leaf
= path
->nodes
[0];
877 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
878 if (item_size
>= sizeof(*ei
)) {
879 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
880 struct btrfs_extent_item
);
881 num_refs
= btrfs_extent_refs(leaf
, ei
);
882 extent_flags
= btrfs_extent_flags(leaf
, ei
);
884 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
885 struct btrfs_extent_item_v0
*ei0
;
886 BUG_ON(item_size
!= sizeof(*ei0
));
887 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
888 struct btrfs_extent_item_v0
);
889 num_refs
= btrfs_extent_refs_v0(leaf
, ei0
);
890 /* FIXME: this isn't correct for data */
891 extent_flags
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
896 BUG_ON(num_refs
== 0);
906 delayed_refs
= &trans
->transaction
->delayed_refs
;
907 spin_lock(&delayed_refs
->lock
);
908 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
910 if (!mutex_trylock(&head
->mutex
)) {
911 atomic_inc(&head
->node
.refs
);
912 spin_unlock(&delayed_refs
->lock
);
914 btrfs_release_path(path
);
917 * Mutex was contended, block until it's released and try
920 mutex_lock(&head
->mutex
);
921 mutex_unlock(&head
->mutex
);
922 btrfs_put_delayed_ref(&head
->node
);
925 spin_lock(&head
->lock
);
926 if (head
->extent_op
&& head
->extent_op
->update_flags
)
927 extent_flags
|= head
->extent_op
->flags_to_set
;
929 BUG_ON(num_refs
== 0);
931 num_refs
+= head
->node
.ref_mod
;
932 spin_unlock(&head
->lock
);
933 mutex_unlock(&head
->mutex
);
935 spin_unlock(&delayed_refs
->lock
);
937 WARN_ON(num_refs
== 0);
941 *flags
= extent_flags
;
943 btrfs_free_path(path
);
948 * Back reference rules. Back refs have three main goals:
950 * 1) differentiate between all holders of references to an extent so that
951 * when a reference is dropped we can make sure it was a valid reference
952 * before freeing the extent.
954 * 2) Provide enough information to quickly find the holders of an extent
955 * if we notice a given block is corrupted or bad.
957 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
958 * maintenance. This is actually the same as #2, but with a slightly
959 * different use case.
961 * There are two kinds of back refs. The implicit back refs is optimized
962 * for pointers in non-shared tree blocks. For a given pointer in a block,
963 * back refs of this kind provide information about the block's owner tree
964 * and the pointer's key. These information allow us to find the block by
965 * b-tree searching. The full back refs is for pointers in tree blocks not
966 * referenced by their owner trees. The location of tree block is recorded
967 * in the back refs. Actually the full back refs is generic, and can be
968 * used in all cases the implicit back refs is used. The major shortcoming
969 * of the full back refs is its overhead. Every time a tree block gets
970 * COWed, we have to update back refs entry for all pointers in it.
972 * For a newly allocated tree block, we use implicit back refs for
973 * pointers in it. This means most tree related operations only involve
974 * implicit back refs. For a tree block created in old transaction, the
975 * only way to drop a reference to it is COW it. So we can detect the
976 * event that tree block loses its owner tree's reference and do the
977 * back refs conversion.
979 * When a tree block is COWed through a tree, there are four cases:
981 * The reference count of the block is one and the tree is the block's
982 * owner tree. Nothing to do in this case.
984 * The reference count of the block is one and the tree is not the
985 * block's owner tree. In this case, full back refs is used for pointers
986 * in the block. Remove these full back refs, add implicit back refs for
987 * every pointers in the new block.
989 * The reference count of the block is greater than one and the tree is
990 * the block's owner tree. In this case, implicit back refs is used for
991 * pointers in the block. Add full back refs for every pointers in the
992 * block, increase lower level extents' reference counts. The original
993 * implicit back refs are entailed to the new block.
995 * The reference count of the block is greater than one and the tree is
996 * not the block's owner tree. Add implicit back refs for every pointer in
997 * the new block, increase lower level extents' reference count.
999 * Back Reference Key composing:
1001 * The key objectid corresponds to the first byte in the extent,
1002 * The key type is used to differentiate between types of back refs.
1003 * There are different meanings of the key offset for different types
1006 * File extents can be referenced by:
1008 * - multiple snapshots, subvolumes, or different generations in one subvol
1009 * - different files inside a single subvolume
1010 * - different offsets inside a file (bookend extents in file.c)
1012 * The extent ref structure for the implicit back refs has fields for:
1014 * - Objectid of the subvolume root
1015 * - objectid of the file holding the reference
1016 * - original offset in the file
1017 * - how many bookend extents
1019 * The key offset for the implicit back refs is hash of the first
1022 * The extent ref structure for the full back refs has field for:
1024 * - number of pointers in the tree leaf
1026 * The key offset for the implicit back refs is the first byte of
1029 * When a file extent is allocated, The implicit back refs is used.
1030 * the fields are filled in:
1032 * (root_key.objectid, inode objectid, offset in file, 1)
1034 * When a file extent is removed file truncation, we find the
1035 * corresponding implicit back refs and check the following fields:
1037 * (btrfs_header_owner(leaf), inode objectid, offset in file)
1039 * Btree extents can be referenced by:
1041 * - Different subvolumes
1043 * Both the implicit back refs and the full back refs for tree blocks
1044 * only consist of key. The key offset for the implicit back refs is
1045 * objectid of block's owner tree. The key offset for the full back refs
1046 * is the first byte of parent block.
1048 * When implicit back refs is used, information about the lowest key and
1049 * level of the tree block are required. These information are stored in
1050 * tree block info structure.
1053 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1054 static int convert_extent_item_v0(struct btrfs_trans_handle
*trans
,
1055 struct btrfs_root
*root
,
1056 struct btrfs_path
*path
,
1057 u64 owner
, u32 extra_size
)
1059 struct btrfs_extent_item
*item
;
1060 struct btrfs_extent_item_v0
*ei0
;
1061 struct btrfs_extent_ref_v0
*ref0
;
1062 struct btrfs_tree_block_info
*bi
;
1063 struct extent_buffer
*leaf
;
1064 struct btrfs_key key
;
1065 struct btrfs_key found_key
;
1066 u32 new_size
= sizeof(*item
);
1070 leaf
= path
->nodes
[0];
1071 BUG_ON(btrfs_item_size_nr(leaf
, path
->slots
[0]) != sizeof(*ei0
));
1073 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1074 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1075 struct btrfs_extent_item_v0
);
1076 refs
= btrfs_extent_refs_v0(leaf
, ei0
);
1078 if (owner
== (u64
)-1) {
1080 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1081 ret
= btrfs_next_leaf(root
, path
);
1084 BUG_ON(ret
> 0); /* Corruption */
1085 leaf
= path
->nodes
[0];
1087 btrfs_item_key_to_cpu(leaf
, &found_key
,
1089 BUG_ON(key
.objectid
!= found_key
.objectid
);
1090 if (found_key
.type
!= BTRFS_EXTENT_REF_V0_KEY
) {
1094 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1095 struct btrfs_extent_ref_v0
);
1096 owner
= btrfs_ref_objectid_v0(leaf
, ref0
);
1100 btrfs_release_path(path
);
1102 if (owner
< BTRFS_FIRST_FREE_OBJECTID
)
1103 new_size
+= sizeof(*bi
);
1105 new_size
-= sizeof(*ei0
);
1106 ret
= btrfs_search_slot(trans
, root
, &key
, path
,
1107 new_size
+ extra_size
, 1);
1110 BUG_ON(ret
); /* Corruption */
1112 btrfs_extend_item(root
, path
, new_size
);
1114 leaf
= path
->nodes
[0];
1115 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1116 btrfs_set_extent_refs(leaf
, item
, refs
);
1117 /* FIXME: get real generation */
1118 btrfs_set_extent_generation(leaf
, item
, 0);
1119 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1120 btrfs_set_extent_flags(leaf
, item
,
1121 BTRFS_EXTENT_FLAG_TREE_BLOCK
|
1122 BTRFS_BLOCK_FLAG_FULL_BACKREF
);
1123 bi
= (struct btrfs_tree_block_info
*)(item
+ 1);
1124 /* FIXME: get first key of the block */
1125 memset_extent_buffer(leaf
, 0, (unsigned long)bi
, sizeof(*bi
));
1126 btrfs_set_tree_block_level(leaf
, bi
, (int)owner
);
1128 btrfs_set_extent_flags(leaf
, item
, BTRFS_EXTENT_FLAG_DATA
);
1130 btrfs_mark_buffer_dirty(leaf
);
1135 static u64
hash_extent_data_ref(u64 root_objectid
, u64 owner
, u64 offset
)
1137 u32 high_crc
= ~(u32
)0;
1138 u32 low_crc
= ~(u32
)0;
1141 lenum
= cpu_to_le64(root_objectid
);
1142 high_crc
= btrfs_crc32c(high_crc
, &lenum
, sizeof(lenum
));
1143 lenum
= cpu_to_le64(owner
);
1144 low_crc
= btrfs_crc32c(low_crc
, &lenum
, sizeof(lenum
));
1145 lenum
= cpu_to_le64(offset
);
1146 low_crc
= btrfs_crc32c(low_crc
, &lenum
, sizeof(lenum
));
1148 return ((u64
)high_crc
<< 31) ^ (u64
)low_crc
;
1151 static u64
hash_extent_data_ref_item(struct extent_buffer
*leaf
,
1152 struct btrfs_extent_data_ref
*ref
)
1154 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf
, ref
),
1155 btrfs_extent_data_ref_objectid(leaf
, ref
),
1156 btrfs_extent_data_ref_offset(leaf
, ref
));
1159 static int match_extent_data_ref(struct extent_buffer
*leaf
,
1160 struct btrfs_extent_data_ref
*ref
,
1161 u64 root_objectid
, u64 owner
, u64 offset
)
1163 if (btrfs_extent_data_ref_root(leaf
, ref
) != root_objectid
||
1164 btrfs_extent_data_ref_objectid(leaf
, ref
) != owner
||
1165 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
1170 static noinline
int lookup_extent_data_ref(struct btrfs_trans_handle
*trans
,
1171 struct btrfs_root
*root
,
1172 struct btrfs_path
*path
,
1173 u64 bytenr
, u64 parent
,
1175 u64 owner
, u64 offset
)
1177 struct btrfs_key key
;
1178 struct btrfs_extent_data_ref
*ref
;
1179 struct extent_buffer
*leaf
;
1185 key
.objectid
= bytenr
;
1187 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1188 key
.offset
= parent
;
1190 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1191 key
.offset
= hash_extent_data_ref(root_objectid
,
1196 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1205 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1206 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1207 btrfs_release_path(path
);
1208 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1219 leaf
= path
->nodes
[0];
1220 nritems
= btrfs_header_nritems(leaf
);
1222 if (path
->slots
[0] >= nritems
) {
1223 ret
= btrfs_next_leaf(root
, path
);
1229 leaf
= path
->nodes
[0];
1230 nritems
= btrfs_header_nritems(leaf
);
1234 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1235 if (key
.objectid
!= bytenr
||
1236 key
.type
!= BTRFS_EXTENT_DATA_REF_KEY
)
1239 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1240 struct btrfs_extent_data_ref
);
1242 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1245 btrfs_release_path(path
);
1257 static noinline
int insert_extent_data_ref(struct btrfs_trans_handle
*trans
,
1258 struct btrfs_root
*root
,
1259 struct btrfs_path
*path
,
1260 u64 bytenr
, u64 parent
,
1261 u64 root_objectid
, u64 owner
,
1262 u64 offset
, int refs_to_add
)
1264 struct btrfs_key key
;
1265 struct extent_buffer
*leaf
;
1270 key
.objectid
= bytenr
;
1272 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1273 key
.offset
= parent
;
1274 size
= sizeof(struct btrfs_shared_data_ref
);
1276 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1277 key
.offset
= hash_extent_data_ref(root_objectid
,
1279 size
= sizeof(struct btrfs_extent_data_ref
);
1282 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, size
);
1283 if (ret
&& ret
!= -EEXIST
)
1286 leaf
= path
->nodes
[0];
1288 struct btrfs_shared_data_ref
*ref
;
1289 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1290 struct btrfs_shared_data_ref
);
1292 btrfs_set_shared_data_ref_count(leaf
, ref
, refs_to_add
);
1294 num_refs
= btrfs_shared_data_ref_count(leaf
, ref
);
1295 num_refs
+= refs_to_add
;
1296 btrfs_set_shared_data_ref_count(leaf
, ref
, num_refs
);
1299 struct btrfs_extent_data_ref
*ref
;
1300 while (ret
== -EEXIST
) {
1301 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1302 struct btrfs_extent_data_ref
);
1303 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1306 btrfs_release_path(path
);
1308 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1310 if (ret
&& ret
!= -EEXIST
)
1313 leaf
= path
->nodes
[0];
1315 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1316 struct btrfs_extent_data_ref
);
1318 btrfs_set_extent_data_ref_root(leaf
, ref
,
1320 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
1321 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
1322 btrfs_set_extent_data_ref_count(leaf
, ref
, refs_to_add
);
1324 num_refs
= btrfs_extent_data_ref_count(leaf
, ref
);
1325 num_refs
+= refs_to_add
;
1326 btrfs_set_extent_data_ref_count(leaf
, ref
, num_refs
);
1329 btrfs_mark_buffer_dirty(leaf
);
1332 btrfs_release_path(path
);
1336 static noinline
int remove_extent_data_ref(struct btrfs_trans_handle
*trans
,
1337 struct btrfs_root
*root
,
1338 struct btrfs_path
*path
,
1339 int refs_to_drop
, int *last_ref
)
1341 struct btrfs_key key
;
1342 struct btrfs_extent_data_ref
*ref1
= NULL
;
1343 struct btrfs_shared_data_ref
*ref2
= NULL
;
1344 struct extent_buffer
*leaf
;
1348 leaf
= path
->nodes
[0];
1349 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1351 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1352 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1353 struct btrfs_extent_data_ref
);
1354 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1355 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1356 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1357 struct btrfs_shared_data_ref
);
1358 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1359 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1360 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1361 struct btrfs_extent_ref_v0
*ref0
;
1362 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1363 struct btrfs_extent_ref_v0
);
1364 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1370 BUG_ON(num_refs
< refs_to_drop
);
1371 num_refs
-= refs_to_drop
;
1373 if (num_refs
== 0) {
1374 ret
= btrfs_del_item(trans
, root
, path
);
1377 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
)
1378 btrfs_set_extent_data_ref_count(leaf
, ref1
, num_refs
);
1379 else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
)
1380 btrfs_set_shared_data_ref_count(leaf
, ref2
, num_refs
);
1381 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1383 struct btrfs_extent_ref_v0
*ref0
;
1384 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1385 struct btrfs_extent_ref_v0
);
1386 btrfs_set_ref_count_v0(leaf
, ref0
, num_refs
);
1389 btrfs_mark_buffer_dirty(leaf
);
1394 static noinline u32
extent_data_ref_count(struct btrfs_path
*path
,
1395 struct btrfs_extent_inline_ref
*iref
)
1397 struct btrfs_key key
;
1398 struct extent_buffer
*leaf
;
1399 struct btrfs_extent_data_ref
*ref1
;
1400 struct btrfs_shared_data_ref
*ref2
;
1403 leaf
= path
->nodes
[0];
1404 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1406 if (btrfs_extent_inline_ref_type(leaf
, iref
) ==
1407 BTRFS_EXTENT_DATA_REF_KEY
) {
1408 ref1
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1409 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1411 ref2
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1412 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1414 } else if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1415 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1416 struct btrfs_extent_data_ref
);
1417 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1418 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1419 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1420 struct btrfs_shared_data_ref
);
1421 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1422 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1423 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1424 struct btrfs_extent_ref_v0
*ref0
;
1425 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1426 struct btrfs_extent_ref_v0
);
1427 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1435 static noinline
int lookup_tree_block_ref(struct btrfs_trans_handle
*trans
,
1436 struct btrfs_root
*root
,
1437 struct btrfs_path
*path
,
1438 u64 bytenr
, u64 parent
,
1441 struct btrfs_key key
;
1444 key
.objectid
= bytenr
;
1446 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1447 key
.offset
= parent
;
1449 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1450 key
.offset
= root_objectid
;
1453 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1456 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1457 if (ret
== -ENOENT
&& parent
) {
1458 btrfs_release_path(path
);
1459 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1460 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1468 static noinline
int insert_tree_block_ref(struct btrfs_trans_handle
*trans
,
1469 struct btrfs_root
*root
,
1470 struct btrfs_path
*path
,
1471 u64 bytenr
, u64 parent
,
1474 struct btrfs_key key
;
1477 key
.objectid
= bytenr
;
1479 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1480 key
.offset
= parent
;
1482 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1483 key
.offset
= root_objectid
;
1486 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
1487 btrfs_release_path(path
);
1491 static inline int extent_ref_type(u64 parent
, u64 owner
)
1494 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1496 type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1498 type
= BTRFS_TREE_BLOCK_REF_KEY
;
1501 type
= BTRFS_SHARED_DATA_REF_KEY
;
1503 type
= BTRFS_EXTENT_DATA_REF_KEY
;
1508 static int find_next_key(struct btrfs_path
*path
, int level
,
1509 struct btrfs_key
*key
)
1512 for (; level
< BTRFS_MAX_LEVEL
; level
++) {
1513 if (!path
->nodes
[level
])
1515 if (path
->slots
[level
] + 1 >=
1516 btrfs_header_nritems(path
->nodes
[level
]))
1519 btrfs_item_key_to_cpu(path
->nodes
[level
], key
,
1520 path
->slots
[level
] + 1);
1522 btrfs_node_key_to_cpu(path
->nodes
[level
], key
,
1523 path
->slots
[level
] + 1);
1530 * look for inline back ref. if back ref is found, *ref_ret is set
1531 * to the address of inline back ref, and 0 is returned.
1533 * if back ref isn't found, *ref_ret is set to the address where it
1534 * should be inserted, and -ENOENT is returned.
1536 * if insert is true and there are too many inline back refs, the path
1537 * points to the extent item, and -EAGAIN is returned.
1539 * NOTE: inline back refs are ordered in the same way that back ref
1540 * items in the tree are ordered.
1542 static noinline_for_stack
1543 int lookup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1544 struct btrfs_root
*root
,
1545 struct btrfs_path
*path
,
1546 struct btrfs_extent_inline_ref
**ref_ret
,
1547 u64 bytenr
, u64 num_bytes
,
1548 u64 parent
, u64 root_objectid
,
1549 u64 owner
, u64 offset
, int insert
)
1551 struct btrfs_key key
;
1552 struct extent_buffer
*leaf
;
1553 struct btrfs_extent_item
*ei
;
1554 struct btrfs_extent_inline_ref
*iref
;
1564 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
1567 key
.objectid
= bytenr
;
1568 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1569 key
.offset
= num_bytes
;
1571 want
= extent_ref_type(parent
, owner
);
1573 extra_size
= btrfs_extent_inline_ref_size(want
);
1574 path
->keep_locks
= 1;
1579 * Owner is our parent level, so we can just add one to get the level
1580 * for the block we are interested in.
1582 if (skinny_metadata
&& owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1583 key
.type
= BTRFS_METADATA_ITEM_KEY
;
1588 ret
= btrfs_search_slot(trans
, root
, &key
, path
, extra_size
, 1);
1595 * We may be a newly converted file system which still has the old fat
1596 * extent entries for metadata, so try and see if we have one of those.
1598 if (ret
> 0 && skinny_metadata
) {
1599 skinny_metadata
= false;
1600 if (path
->slots
[0]) {
1602 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
1604 if (key
.objectid
== bytenr
&&
1605 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
1606 key
.offset
== num_bytes
)
1610 key
.objectid
= bytenr
;
1611 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1612 key
.offset
= num_bytes
;
1613 btrfs_release_path(path
);
1618 if (ret
&& !insert
) {
1621 } else if (WARN_ON(ret
)) {
1626 leaf
= path
->nodes
[0];
1627 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1628 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1629 if (item_size
< sizeof(*ei
)) {
1634 ret
= convert_extent_item_v0(trans
, root
, path
, owner
,
1640 leaf
= path
->nodes
[0];
1641 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1644 BUG_ON(item_size
< sizeof(*ei
));
1646 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1647 flags
= btrfs_extent_flags(leaf
, ei
);
1649 ptr
= (unsigned long)(ei
+ 1);
1650 end
= (unsigned long)ei
+ item_size
;
1652 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
&& !skinny_metadata
) {
1653 ptr
+= sizeof(struct btrfs_tree_block_info
);
1663 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1664 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1668 ptr
+= btrfs_extent_inline_ref_size(type
);
1672 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1673 struct btrfs_extent_data_ref
*dref
;
1674 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1675 if (match_extent_data_ref(leaf
, dref
, root_objectid
,
1680 if (hash_extent_data_ref_item(leaf
, dref
) <
1681 hash_extent_data_ref(root_objectid
, owner
, offset
))
1685 ref_offset
= btrfs_extent_inline_ref_offset(leaf
, iref
);
1687 if (parent
== ref_offset
) {
1691 if (ref_offset
< parent
)
1694 if (root_objectid
== ref_offset
) {
1698 if (ref_offset
< root_objectid
)
1702 ptr
+= btrfs_extent_inline_ref_size(type
);
1704 if (err
== -ENOENT
&& insert
) {
1705 if (item_size
+ extra_size
>=
1706 BTRFS_MAX_EXTENT_ITEM_SIZE(root
)) {
1711 * To add new inline back ref, we have to make sure
1712 * there is no corresponding back ref item.
1713 * For simplicity, we just do not add new inline back
1714 * ref if there is any kind of item for this block
1716 if (find_next_key(path
, 0, &key
) == 0 &&
1717 key
.objectid
== bytenr
&&
1718 key
.type
< BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1723 *ref_ret
= (struct btrfs_extent_inline_ref
*)ptr
;
1726 path
->keep_locks
= 0;
1727 btrfs_unlock_up_safe(path
, 1);
1733 * helper to add new inline back ref
1735 static noinline_for_stack
1736 void setup_inline_extent_backref(struct btrfs_root
*root
,
1737 struct btrfs_path
*path
,
1738 struct btrfs_extent_inline_ref
*iref
,
1739 u64 parent
, u64 root_objectid
,
1740 u64 owner
, u64 offset
, int refs_to_add
,
1741 struct btrfs_delayed_extent_op
*extent_op
)
1743 struct extent_buffer
*leaf
;
1744 struct btrfs_extent_item
*ei
;
1747 unsigned long item_offset
;
1752 leaf
= path
->nodes
[0];
1753 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1754 item_offset
= (unsigned long)iref
- (unsigned long)ei
;
1756 type
= extent_ref_type(parent
, owner
);
1757 size
= btrfs_extent_inline_ref_size(type
);
1759 btrfs_extend_item(root
, path
, size
);
1761 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1762 refs
= btrfs_extent_refs(leaf
, ei
);
1763 refs
+= refs_to_add
;
1764 btrfs_set_extent_refs(leaf
, ei
, refs
);
1766 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1768 ptr
= (unsigned long)ei
+ item_offset
;
1769 end
= (unsigned long)ei
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
1770 if (ptr
< end
- size
)
1771 memmove_extent_buffer(leaf
, ptr
+ size
, ptr
,
1774 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1775 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
1776 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1777 struct btrfs_extent_data_ref
*dref
;
1778 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1779 btrfs_set_extent_data_ref_root(leaf
, dref
, root_objectid
);
1780 btrfs_set_extent_data_ref_objectid(leaf
, dref
, owner
);
1781 btrfs_set_extent_data_ref_offset(leaf
, dref
, offset
);
1782 btrfs_set_extent_data_ref_count(leaf
, dref
, refs_to_add
);
1783 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1784 struct btrfs_shared_data_ref
*sref
;
1785 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1786 btrfs_set_shared_data_ref_count(leaf
, sref
, refs_to_add
);
1787 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1788 } else if (type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
1789 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1791 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
1793 btrfs_mark_buffer_dirty(leaf
);
1796 static int lookup_extent_backref(struct btrfs_trans_handle
*trans
,
1797 struct btrfs_root
*root
,
1798 struct btrfs_path
*path
,
1799 struct btrfs_extent_inline_ref
**ref_ret
,
1800 u64 bytenr
, u64 num_bytes
, u64 parent
,
1801 u64 root_objectid
, u64 owner
, u64 offset
)
1805 ret
= lookup_inline_extent_backref(trans
, root
, path
, ref_ret
,
1806 bytenr
, num_bytes
, parent
,
1807 root_objectid
, owner
, offset
, 0);
1811 btrfs_release_path(path
);
1814 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1815 ret
= lookup_tree_block_ref(trans
, root
, path
, bytenr
, parent
,
1818 ret
= lookup_extent_data_ref(trans
, root
, path
, bytenr
, parent
,
1819 root_objectid
, owner
, offset
);
1825 * helper to update/remove inline back ref
1827 static noinline_for_stack
1828 void update_inline_extent_backref(struct btrfs_root
*root
,
1829 struct btrfs_path
*path
,
1830 struct btrfs_extent_inline_ref
*iref
,
1832 struct btrfs_delayed_extent_op
*extent_op
,
1835 struct extent_buffer
*leaf
;
1836 struct btrfs_extent_item
*ei
;
1837 struct btrfs_extent_data_ref
*dref
= NULL
;
1838 struct btrfs_shared_data_ref
*sref
= NULL
;
1846 leaf
= path
->nodes
[0];
1847 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1848 refs
= btrfs_extent_refs(leaf
, ei
);
1849 WARN_ON(refs_to_mod
< 0 && refs
+ refs_to_mod
<= 0);
1850 refs
+= refs_to_mod
;
1851 btrfs_set_extent_refs(leaf
, ei
, refs
);
1853 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1855 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1857 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1858 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1859 refs
= btrfs_extent_data_ref_count(leaf
, dref
);
1860 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1861 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1862 refs
= btrfs_shared_data_ref_count(leaf
, sref
);
1865 BUG_ON(refs_to_mod
!= -1);
1868 BUG_ON(refs_to_mod
< 0 && refs
< -refs_to_mod
);
1869 refs
+= refs_to_mod
;
1872 if (type
== BTRFS_EXTENT_DATA_REF_KEY
)
1873 btrfs_set_extent_data_ref_count(leaf
, dref
, refs
);
1875 btrfs_set_shared_data_ref_count(leaf
, sref
, refs
);
1878 size
= btrfs_extent_inline_ref_size(type
);
1879 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1880 ptr
= (unsigned long)iref
;
1881 end
= (unsigned long)ei
+ item_size
;
1882 if (ptr
+ size
< end
)
1883 memmove_extent_buffer(leaf
, ptr
, ptr
+ size
,
1886 btrfs_truncate_item(root
, path
, item_size
, 1);
1888 btrfs_mark_buffer_dirty(leaf
);
1891 static noinline_for_stack
1892 int insert_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1893 struct btrfs_root
*root
,
1894 struct btrfs_path
*path
,
1895 u64 bytenr
, u64 num_bytes
, u64 parent
,
1896 u64 root_objectid
, u64 owner
,
1897 u64 offset
, int refs_to_add
,
1898 struct btrfs_delayed_extent_op
*extent_op
)
1900 struct btrfs_extent_inline_ref
*iref
;
1903 ret
= lookup_inline_extent_backref(trans
, root
, path
, &iref
,
1904 bytenr
, num_bytes
, parent
,
1905 root_objectid
, owner
, offset
, 1);
1907 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
);
1908 update_inline_extent_backref(root
, path
, iref
,
1909 refs_to_add
, extent_op
, NULL
);
1910 } else if (ret
== -ENOENT
) {
1911 setup_inline_extent_backref(root
, path
, iref
, parent
,
1912 root_objectid
, owner
, offset
,
1913 refs_to_add
, extent_op
);
1919 static int insert_extent_backref(struct btrfs_trans_handle
*trans
,
1920 struct btrfs_root
*root
,
1921 struct btrfs_path
*path
,
1922 u64 bytenr
, u64 parent
, u64 root_objectid
,
1923 u64 owner
, u64 offset
, int refs_to_add
)
1926 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1927 BUG_ON(refs_to_add
!= 1);
1928 ret
= insert_tree_block_ref(trans
, root
, path
, bytenr
,
1929 parent
, root_objectid
);
1931 ret
= insert_extent_data_ref(trans
, root
, path
, bytenr
,
1932 parent
, root_objectid
,
1933 owner
, offset
, refs_to_add
);
1938 static int remove_extent_backref(struct btrfs_trans_handle
*trans
,
1939 struct btrfs_root
*root
,
1940 struct btrfs_path
*path
,
1941 struct btrfs_extent_inline_ref
*iref
,
1942 int refs_to_drop
, int is_data
, int *last_ref
)
1946 BUG_ON(!is_data
&& refs_to_drop
!= 1);
1948 update_inline_extent_backref(root
, path
, iref
,
1949 -refs_to_drop
, NULL
, last_ref
);
1950 } else if (is_data
) {
1951 ret
= remove_extent_data_ref(trans
, root
, path
, refs_to_drop
,
1955 ret
= btrfs_del_item(trans
, root
, path
);
1960 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
1961 static int btrfs_issue_discard(struct block_device
*bdev
, u64 start
, u64 len
,
1962 u64
*discarded_bytes
)
1965 u64 bytes_left
, end
;
1966 u64 aligned_start
= ALIGN(start
, 1 << 9);
1968 if (WARN_ON(start
!= aligned_start
)) {
1969 len
-= aligned_start
- start
;
1970 len
= round_down(len
, 1 << 9);
1971 start
= aligned_start
;
1974 *discarded_bytes
= 0;
1982 /* Skip any superblocks on this device. */
1983 for (j
= 0; j
< BTRFS_SUPER_MIRROR_MAX
; j
++) {
1984 u64 sb_start
= btrfs_sb_offset(j
);
1985 u64 sb_end
= sb_start
+ BTRFS_SUPER_INFO_SIZE
;
1986 u64 size
= sb_start
- start
;
1988 if (!in_range(sb_start
, start
, bytes_left
) &&
1989 !in_range(sb_end
, start
, bytes_left
) &&
1990 !in_range(start
, sb_start
, BTRFS_SUPER_INFO_SIZE
))
1994 * Superblock spans beginning of range. Adjust start and
1997 if (sb_start
<= start
) {
1998 start
+= sb_end
- start
;
2003 bytes_left
= end
- start
;
2008 ret
= blkdev_issue_discard(bdev
, start
>> 9, size
>> 9,
2011 *discarded_bytes
+= size
;
2012 else if (ret
!= -EOPNOTSUPP
)
2021 bytes_left
= end
- start
;
2025 ret
= blkdev_issue_discard(bdev
, start
>> 9, bytes_left
>> 9,
2028 *discarded_bytes
+= bytes_left
;
2033 int btrfs_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
2034 u64 num_bytes
, u64
*actual_bytes
)
2037 u64 discarded_bytes
= 0;
2038 struct btrfs_bio
*bbio
= NULL
;
2042 * Avoid races with device replace and make sure our bbio has devices
2043 * associated to its stripes that don't go away while we are discarding.
2045 btrfs_bio_counter_inc_blocked(root
->fs_info
);
2046 /* Tell the block device(s) that the sectors can be discarded */
2047 ret
= btrfs_map_block(root
->fs_info
, REQ_OP_DISCARD
,
2048 bytenr
, &num_bytes
, &bbio
, 0);
2049 /* Error condition is -ENOMEM */
2051 struct btrfs_bio_stripe
*stripe
= bbio
->stripes
;
2055 for (i
= 0; i
< bbio
->num_stripes
; i
++, stripe
++) {
2057 if (!stripe
->dev
->can_discard
)
2060 ret
= btrfs_issue_discard(stripe
->dev
->bdev
,
2065 discarded_bytes
+= bytes
;
2066 else if (ret
!= -EOPNOTSUPP
)
2067 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2070 * Just in case we get back EOPNOTSUPP for some reason,
2071 * just ignore the return value so we don't screw up
2072 * people calling discard_extent.
2076 btrfs_put_bbio(bbio
);
2078 btrfs_bio_counter_dec(root
->fs_info
);
2081 *actual_bytes
= discarded_bytes
;
2084 if (ret
== -EOPNOTSUPP
)
2089 /* Can return -ENOMEM */
2090 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
2091 struct btrfs_root
*root
,
2092 u64 bytenr
, u64 num_bytes
, u64 parent
,
2093 u64 root_objectid
, u64 owner
, u64 offset
)
2096 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2098 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
&&
2099 root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
2101 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
2102 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
2104 parent
, root_objectid
, (int)owner
,
2105 BTRFS_ADD_DELAYED_REF
, NULL
);
2107 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
2108 num_bytes
, parent
, root_objectid
,
2110 BTRFS_ADD_DELAYED_REF
, NULL
);
2115 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
2116 struct btrfs_root
*root
,
2117 struct btrfs_delayed_ref_node
*node
,
2118 u64 parent
, u64 root_objectid
,
2119 u64 owner
, u64 offset
, int refs_to_add
,
2120 struct btrfs_delayed_extent_op
*extent_op
)
2122 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2123 struct btrfs_path
*path
;
2124 struct extent_buffer
*leaf
;
2125 struct btrfs_extent_item
*item
;
2126 struct btrfs_key key
;
2127 u64 bytenr
= node
->bytenr
;
2128 u64 num_bytes
= node
->num_bytes
;
2132 path
= btrfs_alloc_path();
2136 path
->reada
= READA_FORWARD
;
2137 path
->leave_spinning
= 1;
2138 /* this will setup the path even if it fails to insert the back ref */
2139 ret
= insert_inline_extent_backref(trans
, fs_info
->extent_root
, path
,
2140 bytenr
, num_bytes
, parent
,
2141 root_objectid
, owner
, offset
,
2142 refs_to_add
, extent_op
);
2143 if ((ret
< 0 && ret
!= -EAGAIN
) || !ret
)
2147 * Ok we had -EAGAIN which means we didn't have space to insert and
2148 * inline extent ref, so just update the reference count and add a
2151 leaf
= path
->nodes
[0];
2152 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2153 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2154 refs
= btrfs_extent_refs(leaf
, item
);
2155 btrfs_set_extent_refs(leaf
, item
, refs
+ refs_to_add
);
2157 __run_delayed_extent_op(extent_op
, leaf
, item
);
2159 btrfs_mark_buffer_dirty(leaf
);
2160 btrfs_release_path(path
);
2162 path
->reada
= READA_FORWARD
;
2163 path
->leave_spinning
= 1;
2164 /* now insert the actual backref */
2165 ret
= insert_extent_backref(trans
, root
->fs_info
->extent_root
,
2166 path
, bytenr
, parent
, root_objectid
,
2167 owner
, offset
, refs_to_add
);
2169 btrfs_abort_transaction(trans
, ret
);
2171 btrfs_free_path(path
);
2175 static int run_delayed_data_ref(struct btrfs_trans_handle
*trans
,
2176 struct btrfs_root
*root
,
2177 struct btrfs_delayed_ref_node
*node
,
2178 struct btrfs_delayed_extent_op
*extent_op
,
2179 int insert_reserved
)
2182 struct btrfs_delayed_data_ref
*ref
;
2183 struct btrfs_key ins
;
2188 ins
.objectid
= node
->bytenr
;
2189 ins
.offset
= node
->num_bytes
;
2190 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2192 ref
= btrfs_delayed_node_to_data_ref(node
);
2193 trace_run_delayed_data_ref(root
->fs_info
, node
, ref
, node
->action
);
2195 if (node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2196 parent
= ref
->parent
;
2197 ref_root
= ref
->root
;
2199 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2201 flags
|= extent_op
->flags_to_set
;
2202 ret
= alloc_reserved_file_extent(trans
, root
,
2203 parent
, ref_root
, flags
,
2204 ref
->objectid
, ref
->offset
,
2205 &ins
, node
->ref_mod
);
2206 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2207 ret
= __btrfs_inc_extent_ref(trans
, root
, node
, parent
,
2208 ref_root
, ref
->objectid
,
2209 ref
->offset
, node
->ref_mod
,
2211 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2212 ret
= __btrfs_free_extent(trans
, root
, node
, parent
,
2213 ref_root
, ref
->objectid
,
2214 ref
->offset
, node
->ref_mod
,
2222 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
2223 struct extent_buffer
*leaf
,
2224 struct btrfs_extent_item
*ei
)
2226 u64 flags
= btrfs_extent_flags(leaf
, ei
);
2227 if (extent_op
->update_flags
) {
2228 flags
|= extent_op
->flags_to_set
;
2229 btrfs_set_extent_flags(leaf
, ei
, flags
);
2232 if (extent_op
->update_key
) {
2233 struct btrfs_tree_block_info
*bi
;
2234 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
));
2235 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
2236 btrfs_set_tree_block_key(leaf
, bi
, &extent_op
->key
);
2240 static int run_delayed_extent_op(struct btrfs_trans_handle
*trans
,
2241 struct btrfs_root
*root
,
2242 struct btrfs_delayed_ref_node
*node
,
2243 struct btrfs_delayed_extent_op
*extent_op
)
2245 struct btrfs_key key
;
2246 struct btrfs_path
*path
;
2247 struct btrfs_extent_item
*ei
;
2248 struct extent_buffer
*leaf
;
2252 int metadata
= !extent_op
->is_data
;
2257 if (metadata
&& !btrfs_fs_incompat(root
->fs_info
, SKINNY_METADATA
))
2260 path
= btrfs_alloc_path();
2264 key
.objectid
= node
->bytenr
;
2267 key
.type
= BTRFS_METADATA_ITEM_KEY
;
2268 key
.offset
= extent_op
->level
;
2270 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2271 key
.offset
= node
->num_bytes
;
2275 path
->reada
= READA_FORWARD
;
2276 path
->leave_spinning
= 1;
2277 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
,
2285 if (path
->slots
[0] > 0) {
2287 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
2289 if (key
.objectid
== node
->bytenr
&&
2290 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
2291 key
.offset
== node
->num_bytes
)
2295 btrfs_release_path(path
);
2298 key
.objectid
= node
->bytenr
;
2299 key
.offset
= node
->num_bytes
;
2300 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2309 leaf
= path
->nodes
[0];
2310 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2311 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2312 if (item_size
< sizeof(*ei
)) {
2313 ret
= convert_extent_item_v0(trans
, root
->fs_info
->extent_root
,
2319 leaf
= path
->nodes
[0];
2320 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2323 BUG_ON(item_size
< sizeof(*ei
));
2324 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2325 __run_delayed_extent_op(extent_op
, leaf
, ei
);
2327 btrfs_mark_buffer_dirty(leaf
);
2329 btrfs_free_path(path
);
2333 static int run_delayed_tree_ref(struct btrfs_trans_handle
*trans
,
2334 struct btrfs_root
*root
,
2335 struct btrfs_delayed_ref_node
*node
,
2336 struct btrfs_delayed_extent_op
*extent_op
,
2337 int insert_reserved
)
2340 struct btrfs_delayed_tree_ref
*ref
;
2341 struct btrfs_key ins
;
2344 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
2347 ref
= btrfs_delayed_node_to_tree_ref(node
);
2348 trace_run_delayed_tree_ref(root
->fs_info
, node
, ref
, node
->action
);
2350 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2351 parent
= ref
->parent
;
2352 ref_root
= ref
->root
;
2354 ins
.objectid
= node
->bytenr
;
2355 if (skinny_metadata
) {
2356 ins
.offset
= ref
->level
;
2357 ins
.type
= BTRFS_METADATA_ITEM_KEY
;
2359 ins
.offset
= node
->num_bytes
;
2360 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2363 BUG_ON(node
->ref_mod
!= 1);
2364 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2365 BUG_ON(!extent_op
|| !extent_op
->update_flags
);
2366 ret
= alloc_reserved_tree_block(trans
, root
,
2368 extent_op
->flags_to_set
,
2371 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2372 ret
= __btrfs_inc_extent_ref(trans
, root
, node
,
2376 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2377 ret
= __btrfs_free_extent(trans
, root
, node
,
2379 ref
->level
, 0, 1, extent_op
);
2386 /* helper function to actually process a single delayed ref entry */
2387 static int run_one_delayed_ref(struct btrfs_trans_handle
*trans
,
2388 struct btrfs_root
*root
,
2389 struct btrfs_delayed_ref_node
*node
,
2390 struct btrfs_delayed_extent_op
*extent_op
,
2391 int insert_reserved
)
2395 if (trans
->aborted
) {
2396 if (insert_reserved
)
2397 btrfs_pin_extent(root
, node
->bytenr
,
2398 node
->num_bytes
, 1);
2402 if (btrfs_delayed_ref_is_head(node
)) {
2403 struct btrfs_delayed_ref_head
*head
;
2405 * we've hit the end of the chain and we were supposed
2406 * to insert this extent into the tree. But, it got
2407 * deleted before we ever needed to insert it, so all
2408 * we have to do is clean up the accounting
2411 head
= btrfs_delayed_node_to_head(node
);
2412 trace_run_delayed_ref_head(root
->fs_info
, node
, head
,
2415 if (insert_reserved
) {
2416 btrfs_pin_extent(root
, node
->bytenr
,
2417 node
->num_bytes
, 1);
2418 if (head
->is_data
) {
2419 ret
= btrfs_del_csums(trans
, root
,
2425 /* Also free its reserved qgroup space */
2426 btrfs_qgroup_free_delayed_ref(root
->fs_info
,
2427 head
->qgroup_ref_root
,
2428 head
->qgroup_reserved
);
2432 if (node
->type
== BTRFS_TREE_BLOCK_REF_KEY
||
2433 node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2434 ret
= run_delayed_tree_ref(trans
, root
, node
, extent_op
,
2436 else if (node
->type
== BTRFS_EXTENT_DATA_REF_KEY
||
2437 node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2438 ret
= run_delayed_data_ref(trans
, root
, node
, extent_op
,
2445 static inline struct btrfs_delayed_ref_node
*
2446 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
2448 struct btrfs_delayed_ref_node
*ref
;
2450 if (list_empty(&head
->ref_list
))
2454 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2455 * This is to prevent a ref count from going down to zero, which deletes
2456 * the extent item from the extent tree, when there still are references
2457 * to add, which would fail because they would not find the extent item.
2459 list_for_each_entry(ref
, &head
->ref_list
, list
) {
2460 if (ref
->action
== BTRFS_ADD_DELAYED_REF
)
2464 return list_entry(head
->ref_list
.next
, struct btrfs_delayed_ref_node
,
2469 * Returns 0 on success or if called with an already aborted transaction.
2470 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2472 static noinline
int __btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2473 struct btrfs_root
*root
,
2476 struct btrfs_delayed_ref_root
*delayed_refs
;
2477 struct btrfs_delayed_ref_node
*ref
;
2478 struct btrfs_delayed_ref_head
*locked_ref
= NULL
;
2479 struct btrfs_delayed_extent_op
*extent_op
;
2480 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2481 ktime_t start
= ktime_get();
2483 unsigned long count
= 0;
2484 unsigned long actual_count
= 0;
2485 int must_insert_reserved
= 0;
2487 delayed_refs
= &trans
->transaction
->delayed_refs
;
2493 spin_lock(&delayed_refs
->lock
);
2494 locked_ref
= btrfs_select_ref_head(trans
);
2496 spin_unlock(&delayed_refs
->lock
);
2500 /* grab the lock that says we are going to process
2501 * all the refs for this head */
2502 ret
= btrfs_delayed_ref_lock(trans
, locked_ref
);
2503 spin_unlock(&delayed_refs
->lock
);
2505 * we may have dropped the spin lock to get the head
2506 * mutex lock, and that might have given someone else
2507 * time to free the head. If that's true, it has been
2508 * removed from our list and we can move on.
2510 if (ret
== -EAGAIN
) {
2518 * We need to try and merge add/drops of the same ref since we
2519 * can run into issues with relocate dropping the implicit ref
2520 * and then it being added back again before the drop can
2521 * finish. If we merged anything we need to re-loop so we can
2523 * Or we can get node references of the same type that weren't
2524 * merged when created due to bumps in the tree mod seq, and
2525 * we need to merge them to prevent adding an inline extent
2526 * backref before dropping it (triggering a BUG_ON at
2527 * insert_inline_extent_backref()).
2529 spin_lock(&locked_ref
->lock
);
2530 btrfs_merge_delayed_refs(trans
, fs_info
, delayed_refs
,
2534 * locked_ref is the head node, so we have to go one
2535 * node back for any delayed ref updates
2537 ref
= select_delayed_ref(locked_ref
);
2539 if (ref
&& ref
->seq
&&
2540 btrfs_check_delayed_seq(fs_info
, delayed_refs
, ref
->seq
)) {
2541 spin_unlock(&locked_ref
->lock
);
2542 btrfs_delayed_ref_unlock(locked_ref
);
2543 spin_lock(&delayed_refs
->lock
);
2544 locked_ref
->processing
= 0;
2545 delayed_refs
->num_heads_ready
++;
2546 spin_unlock(&delayed_refs
->lock
);
2554 * record the must insert reserved flag before we
2555 * drop the spin lock.
2557 must_insert_reserved
= locked_ref
->must_insert_reserved
;
2558 locked_ref
->must_insert_reserved
= 0;
2560 extent_op
= locked_ref
->extent_op
;
2561 locked_ref
->extent_op
= NULL
;
2566 /* All delayed refs have been processed, Go ahead
2567 * and send the head node to run_one_delayed_ref,
2568 * so that any accounting fixes can happen
2570 ref
= &locked_ref
->node
;
2572 if (extent_op
&& must_insert_reserved
) {
2573 btrfs_free_delayed_extent_op(extent_op
);
2578 spin_unlock(&locked_ref
->lock
);
2579 ret
= run_delayed_extent_op(trans
, root
,
2581 btrfs_free_delayed_extent_op(extent_op
);
2585 * Need to reset must_insert_reserved if
2586 * there was an error so the abort stuff
2587 * can cleanup the reserved space
2590 if (must_insert_reserved
)
2591 locked_ref
->must_insert_reserved
= 1;
2592 locked_ref
->processing
= 0;
2593 btrfs_debug(fs_info
, "run_delayed_extent_op returned %d", ret
);
2594 btrfs_delayed_ref_unlock(locked_ref
);
2601 * Need to drop our head ref lock and re-acquire the
2602 * delayed ref lock and then re-check to make sure
2605 spin_unlock(&locked_ref
->lock
);
2606 spin_lock(&delayed_refs
->lock
);
2607 spin_lock(&locked_ref
->lock
);
2608 if (!list_empty(&locked_ref
->ref_list
) ||
2609 locked_ref
->extent_op
) {
2610 spin_unlock(&locked_ref
->lock
);
2611 spin_unlock(&delayed_refs
->lock
);
2615 delayed_refs
->num_heads
--;
2616 rb_erase(&locked_ref
->href_node
,
2617 &delayed_refs
->href_root
);
2618 spin_unlock(&delayed_refs
->lock
);
2622 list_del(&ref
->list
);
2624 atomic_dec(&delayed_refs
->num_entries
);
2626 if (!btrfs_delayed_ref_is_head(ref
)) {
2628 * when we play the delayed ref, also correct the
2631 switch (ref
->action
) {
2632 case BTRFS_ADD_DELAYED_REF
:
2633 case BTRFS_ADD_DELAYED_EXTENT
:
2634 locked_ref
->node
.ref_mod
-= ref
->ref_mod
;
2636 case BTRFS_DROP_DELAYED_REF
:
2637 locked_ref
->node
.ref_mod
+= ref
->ref_mod
;
2643 spin_unlock(&locked_ref
->lock
);
2645 ret
= run_one_delayed_ref(trans
, root
, ref
, extent_op
,
2646 must_insert_reserved
);
2648 btrfs_free_delayed_extent_op(extent_op
);
2650 spin_lock(&delayed_refs
->lock
);
2651 locked_ref
->processing
= 0;
2652 delayed_refs
->num_heads_ready
++;
2653 spin_unlock(&delayed_refs
->lock
);
2654 btrfs_delayed_ref_unlock(locked_ref
);
2655 btrfs_put_delayed_ref(ref
);
2656 btrfs_debug(fs_info
, "run_one_delayed_ref returned %d", ret
);
2661 * If this node is a head, that means all the refs in this head
2662 * have been dealt with, and we will pick the next head to deal
2663 * with, so we must unlock the head and drop it from the cluster
2664 * list before we release it.
2666 if (btrfs_delayed_ref_is_head(ref
)) {
2667 if (locked_ref
->is_data
&&
2668 locked_ref
->total_ref_mod
< 0) {
2669 spin_lock(&delayed_refs
->lock
);
2670 delayed_refs
->pending_csums
-= ref
->num_bytes
;
2671 spin_unlock(&delayed_refs
->lock
);
2673 btrfs_delayed_ref_unlock(locked_ref
);
2676 btrfs_put_delayed_ref(ref
);
2682 * We don't want to include ref heads since we can have empty ref heads
2683 * and those will drastically skew our runtime down since we just do
2684 * accounting, no actual extent tree updates.
2686 if (actual_count
> 0) {
2687 u64 runtime
= ktime_to_ns(ktime_sub(ktime_get(), start
));
2691 * We weigh the current average higher than our current runtime
2692 * to avoid large swings in the average.
2694 spin_lock(&delayed_refs
->lock
);
2695 avg
= fs_info
->avg_delayed_ref_runtime
* 3 + runtime
;
2696 fs_info
->avg_delayed_ref_runtime
= avg
>> 2; /* div by 4 */
2697 spin_unlock(&delayed_refs
->lock
);
2702 #ifdef SCRAMBLE_DELAYED_REFS
2704 * Normally delayed refs get processed in ascending bytenr order. This
2705 * correlates in most cases to the order added. To expose dependencies on this
2706 * order, we start to process the tree in the middle instead of the beginning
2708 static u64
find_middle(struct rb_root
*root
)
2710 struct rb_node
*n
= root
->rb_node
;
2711 struct btrfs_delayed_ref_node
*entry
;
2714 u64 first
= 0, last
= 0;
2718 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2719 first
= entry
->bytenr
;
2723 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2724 last
= entry
->bytenr
;
2729 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2730 WARN_ON(!entry
->in_tree
);
2732 middle
= entry
->bytenr
;
2745 static inline u64
heads_to_leaves(struct btrfs_root
*root
, u64 heads
)
2749 num_bytes
= heads
* (sizeof(struct btrfs_extent_item
) +
2750 sizeof(struct btrfs_extent_inline_ref
));
2751 if (!btrfs_fs_incompat(root
->fs_info
, SKINNY_METADATA
))
2752 num_bytes
+= heads
* sizeof(struct btrfs_tree_block_info
);
2755 * We don't ever fill up leaves all the way so multiply by 2 just to be
2756 * closer to what we're really going to want to use.
2758 return div_u64(num_bytes
, BTRFS_LEAF_DATA_SIZE(root
));
2762 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2763 * would require to store the csums for that many bytes.
2765 u64
btrfs_csum_bytes_to_leaves(struct btrfs_root
*root
, u64 csum_bytes
)
2768 u64 num_csums_per_leaf
;
2771 csum_size
= BTRFS_MAX_ITEM_SIZE(root
);
2772 num_csums_per_leaf
= div64_u64(csum_size
,
2773 (u64
)btrfs_super_csum_size(root
->fs_info
->super_copy
));
2774 num_csums
= div64_u64(csum_bytes
, root
->sectorsize
);
2775 num_csums
+= num_csums_per_leaf
- 1;
2776 num_csums
= div64_u64(num_csums
, num_csums_per_leaf
);
2780 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle
*trans
,
2781 struct btrfs_root
*root
)
2783 struct btrfs_block_rsv
*global_rsv
;
2784 u64 num_heads
= trans
->transaction
->delayed_refs
.num_heads_ready
;
2785 u64 csum_bytes
= trans
->transaction
->delayed_refs
.pending_csums
;
2786 u64 num_dirty_bgs
= trans
->transaction
->num_dirty_bgs
;
2787 u64 num_bytes
, num_dirty_bgs_bytes
;
2790 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
2791 num_heads
= heads_to_leaves(root
, num_heads
);
2793 num_bytes
+= (num_heads
- 1) * root
->nodesize
;
2795 num_bytes
+= btrfs_csum_bytes_to_leaves(root
, csum_bytes
) * root
->nodesize
;
2796 num_dirty_bgs_bytes
= btrfs_calc_trans_metadata_size(root
,
2798 global_rsv
= &root
->fs_info
->global_block_rsv
;
2801 * If we can't allocate any more chunks lets make sure we have _lots_ of
2802 * wiggle room since running delayed refs can create more delayed refs.
2804 if (global_rsv
->space_info
->full
) {
2805 num_dirty_bgs_bytes
<<= 1;
2809 spin_lock(&global_rsv
->lock
);
2810 if (global_rsv
->reserved
<= num_bytes
+ num_dirty_bgs_bytes
)
2812 spin_unlock(&global_rsv
->lock
);
2816 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle
*trans
,
2817 struct btrfs_root
*root
)
2819 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2821 atomic_read(&trans
->transaction
->delayed_refs
.num_entries
);
2826 avg_runtime
= fs_info
->avg_delayed_ref_runtime
;
2827 val
= num_entries
* avg_runtime
;
2828 if (num_entries
* avg_runtime
>= NSEC_PER_SEC
)
2830 if (val
>= NSEC_PER_SEC
/ 2)
2833 return btrfs_check_space_for_delayed_refs(trans
, root
);
2836 struct async_delayed_refs
{
2837 struct btrfs_root
*root
;
2842 struct completion wait
;
2843 struct btrfs_work work
;
2846 static void delayed_ref_async_start(struct btrfs_work
*work
)
2848 struct async_delayed_refs
*async
;
2849 struct btrfs_trans_handle
*trans
;
2852 async
= container_of(work
, struct async_delayed_refs
, work
);
2854 /* if the commit is already started, we don't need to wait here */
2855 if (btrfs_transaction_blocked(async
->root
->fs_info
))
2858 trans
= btrfs_join_transaction(async
->root
);
2859 if (IS_ERR(trans
)) {
2860 async
->error
= PTR_ERR(trans
);
2865 * trans->sync means that when we call end_transaction, we won't
2866 * wait on delayed refs
2870 /* Don't bother flushing if we got into a different transaction */
2871 if (trans
->transid
> async
->transid
)
2874 ret
= btrfs_run_delayed_refs(trans
, async
->root
, async
->count
);
2878 ret
= btrfs_end_transaction(trans
, async
->root
);
2879 if (ret
&& !async
->error
)
2883 complete(&async
->wait
);
2888 int btrfs_async_run_delayed_refs(struct btrfs_root
*root
,
2889 unsigned long count
, u64 transid
, int wait
)
2891 struct async_delayed_refs
*async
;
2894 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
2898 async
->root
= root
->fs_info
->tree_root
;
2899 async
->count
= count
;
2901 async
->transid
= transid
;
2906 init_completion(&async
->wait
);
2908 btrfs_init_work(&async
->work
, btrfs_extent_refs_helper
,
2909 delayed_ref_async_start
, NULL
, NULL
);
2911 btrfs_queue_work(root
->fs_info
->extent_workers
, &async
->work
);
2914 wait_for_completion(&async
->wait
);
2923 * this starts processing the delayed reference count updates and
2924 * extent insertions we have queued up so far. count can be
2925 * 0, which means to process everything in the tree at the start
2926 * of the run (but not newly added entries), or it can be some target
2927 * number you'd like to process.
2929 * Returns 0 on success or if called with an aborted transaction
2930 * Returns <0 on error and aborts the transaction
2932 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2933 struct btrfs_root
*root
, unsigned long count
)
2935 struct rb_node
*node
;
2936 struct btrfs_delayed_ref_root
*delayed_refs
;
2937 struct btrfs_delayed_ref_head
*head
;
2939 int run_all
= count
== (unsigned long)-1;
2940 bool can_flush_pending_bgs
= trans
->can_flush_pending_bgs
;
2942 /* We'll clean this up in btrfs_cleanup_transaction */
2946 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE
, &root
->fs_info
->flags
))
2949 if (root
== root
->fs_info
->extent_root
)
2950 root
= root
->fs_info
->tree_root
;
2952 delayed_refs
= &trans
->transaction
->delayed_refs
;
2954 count
= atomic_read(&delayed_refs
->num_entries
) * 2;
2957 #ifdef SCRAMBLE_DELAYED_REFS
2958 delayed_refs
->run_delayed_start
= find_middle(&delayed_refs
->root
);
2960 trans
->can_flush_pending_bgs
= false;
2961 ret
= __btrfs_run_delayed_refs(trans
, root
, count
);
2963 btrfs_abort_transaction(trans
, ret
);
2968 if (!list_empty(&trans
->new_bgs
))
2969 btrfs_create_pending_block_groups(trans
, root
);
2971 spin_lock(&delayed_refs
->lock
);
2972 node
= rb_first(&delayed_refs
->href_root
);
2974 spin_unlock(&delayed_refs
->lock
);
2977 count
= (unsigned long)-1;
2980 head
= rb_entry(node
, struct btrfs_delayed_ref_head
,
2982 if (btrfs_delayed_ref_is_head(&head
->node
)) {
2983 struct btrfs_delayed_ref_node
*ref
;
2986 atomic_inc(&ref
->refs
);
2988 spin_unlock(&delayed_refs
->lock
);
2990 * Mutex was contended, block until it's
2991 * released and try again
2993 mutex_lock(&head
->mutex
);
2994 mutex_unlock(&head
->mutex
);
2996 btrfs_put_delayed_ref(ref
);
3002 node
= rb_next(node
);
3004 spin_unlock(&delayed_refs
->lock
);
3009 assert_qgroups_uptodate(trans
);
3010 trans
->can_flush_pending_bgs
= can_flush_pending_bgs
;
3014 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
3015 struct btrfs_root
*root
,
3016 u64 bytenr
, u64 num_bytes
, u64 flags
,
3017 int level
, int is_data
)
3019 struct btrfs_delayed_extent_op
*extent_op
;
3022 extent_op
= btrfs_alloc_delayed_extent_op();
3026 extent_op
->flags_to_set
= flags
;
3027 extent_op
->update_flags
= true;
3028 extent_op
->update_key
= false;
3029 extent_op
->is_data
= is_data
? true : false;
3030 extent_op
->level
= level
;
3032 ret
= btrfs_add_delayed_extent_op(root
->fs_info
, trans
, bytenr
,
3033 num_bytes
, extent_op
);
3035 btrfs_free_delayed_extent_op(extent_op
);
3039 static noinline
int check_delayed_ref(struct btrfs_trans_handle
*trans
,
3040 struct btrfs_root
*root
,
3041 struct btrfs_path
*path
,
3042 u64 objectid
, u64 offset
, u64 bytenr
)
3044 struct btrfs_delayed_ref_head
*head
;
3045 struct btrfs_delayed_ref_node
*ref
;
3046 struct btrfs_delayed_data_ref
*data_ref
;
3047 struct btrfs_delayed_ref_root
*delayed_refs
;
3050 delayed_refs
= &trans
->transaction
->delayed_refs
;
3051 spin_lock(&delayed_refs
->lock
);
3052 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
3054 spin_unlock(&delayed_refs
->lock
);
3058 if (!mutex_trylock(&head
->mutex
)) {
3059 atomic_inc(&head
->node
.refs
);
3060 spin_unlock(&delayed_refs
->lock
);
3062 btrfs_release_path(path
);
3065 * Mutex was contended, block until it's released and let
3068 mutex_lock(&head
->mutex
);
3069 mutex_unlock(&head
->mutex
);
3070 btrfs_put_delayed_ref(&head
->node
);
3073 spin_unlock(&delayed_refs
->lock
);
3075 spin_lock(&head
->lock
);
3076 list_for_each_entry(ref
, &head
->ref_list
, list
) {
3077 /* If it's a shared ref we know a cross reference exists */
3078 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
) {
3083 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
3086 * If our ref doesn't match the one we're currently looking at
3087 * then we have a cross reference.
3089 if (data_ref
->root
!= root
->root_key
.objectid
||
3090 data_ref
->objectid
!= objectid
||
3091 data_ref
->offset
!= offset
) {
3096 spin_unlock(&head
->lock
);
3097 mutex_unlock(&head
->mutex
);
3101 static noinline
int check_committed_ref(struct btrfs_trans_handle
*trans
,
3102 struct btrfs_root
*root
,
3103 struct btrfs_path
*path
,
3104 u64 objectid
, u64 offset
, u64 bytenr
)
3106 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
3107 struct extent_buffer
*leaf
;
3108 struct btrfs_extent_data_ref
*ref
;
3109 struct btrfs_extent_inline_ref
*iref
;
3110 struct btrfs_extent_item
*ei
;
3111 struct btrfs_key key
;
3115 key
.objectid
= bytenr
;
3116 key
.offset
= (u64
)-1;
3117 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
3119 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
3122 BUG_ON(ret
== 0); /* Corruption */
3125 if (path
->slots
[0] == 0)
3129 leaf
= path
->nodes
[0];
3130 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
3132 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
3136 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
3137 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3138 if (item_size
< sizeof(*ei
)) {
3139 WARN_ON(item_size
!= sizeof(struct btrfs_extent_item_v0
));
3143 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
3145 if (item_size
!= sizeof(*ei
) +
3146 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
3149 if (btrfs_extent_generation(leaf
, ei
) <=
3150 btrfs_root_last_snapshot(&root
->root_item
))
3153 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
3154 if (btrfs_extent_inline_ref_type(leaf
, iref
) !=
3155 BTRFS_EXTENT_DATA_REF_KEY
)
3158 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
3159 if (btrfs_extent_refs(leaf
, ei
) !=
3160 btrfs_extent_data_ref_count(leaf
, ref
) ||
3161 btrfs_extent_data_ref_root(leaf
, ref
) !=
3162 root
->root_key
.objectid
||
3163 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
3164 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
3172 int btrfs_cross_ref_exist(struct btrfs_trans_handle
*trans
,
3173 struct btrfs_root
*root
,
3174 u64 objectid
, u64 offset
, u64 bytenr
)
3176 struct btrfs_path
*path
;
3180 path
= btrfs_alloc_path();
3185 ret
= check_committed_ref(trans
, root
, path
, objectid
,
3187 if (ret
&& ret
!= -ENOENT
)
3190 ret2
= check_delayed_ref(trans
, root
, path
, objectid
,
3192 } while (ret2
== -EAGAIN
);
3194 if (ret2
&& ret2
!= -ENOENT
) {
3199 if (ret
!= -ENOENT
|| ret2
!= -ENOENT
)
3202 btrfs_free_path(path
);
3203 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
3208 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
3209 struct btrfs_root
*root
,
3210 struct extent_buffer
*buf
,
3211 int full_backref
, int inc
)
3218 struct btrfs_key key
;
3219 struct btrfs_file_extent_item
*fi
;
3223 int (*process_func
)(struct btrfs_trans_handle
*, struct btrfs_root
*,
3224 u64
, u64
, u64
, u64
, u64
, u64
);
3227 if (btrfs_is_testing(root
->fs_info
))
3230 ref_root
= btrfs_header_owner(buf
);
3231 nritems
= btrfs_header_nritems(buf
);
3232 level
= btrfs_header_level(buf
);
3234 if (!test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) && level
== 0)
3238 process_func
= btrfs_inc_extent_ref
;
3240 process_func
= btrfs_free_extent
;
3243 parent
= buf
->start
;
3247 for (i
= 0; i
< nritems
; i
++) {
3249 btrfs_item_key_to_cpu(buf
, &key
, i
);
3250 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
3252 fi
= btrfs_item_ptr(buf
, i
,
3253 struct btrfs_file_extent_item
);
3254 if (btrfs_file_extent_type(buf
, fi
) ==
3255 BTRFS_FILE_EXTENT_INLINE
)
3257 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
3261 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
3262 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
3263 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
3264 parent
, ref_root
, key
.objectid
,
3269 bytenr
= btrfs_node_blockptr(buf
, i
);
3270 num_bytes
= root
->nodesize
;
3271 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
3272 parent
, ref_root
, level
- 1, 0);
3282 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3283 struct extent_buffer
*buf
, int full_backref
)
3285 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1);
3288 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3289 struct extent_buffer
*buf
, int full_backref
)
3291 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0);
3294 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
3295 struct btrfs_root
*root
,
3296 struct btrfs_path
*path
,
3297 struct btrfs_block_group_cache
*cache
)
3300 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
3302 struct extent_buffer
*leaf
;
3304 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
3311 leaf
= path
->nodes
[0];
3312 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
3313 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
3314 btrfs_mark_buffer_dirty(leaf
);
3316 btrfs_release_path(path
);
3321 static struct btrfs_block_group_cache
*
3322 next_block_group(struct btrfs_root
*root
,
3323 struct btrfs_block_group_cache
*cache
)
3325 struct rb_node
*node
;
3327 spin_lock(&root
->fs_info
->block_group_cache_lock
);
3329 /* If our block group was removed, we need a full search. */
3330 if (RB_EMPTY_NODE(&cache
->cache_node
)) {
3331 const u64 next_bytenr
= cache
->key
.objectid
+ cache
->key
.offset
;
3333 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
3334 btrfs_put_block_group(cache
);
3335 cache
= btrfs_lookup_first_block_group(root
->fs_info
,
3339 node
= rb_next(&cache
->cache_node
);
3340 btrfs_put_block_group(cache
);
3342 cache
= rb_entry(node
, struct btrfs_block_group_cache
,
3344 btrfs_get_block_group(cache
);
3347 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
3351 static int cache_save_setup(struct btrfs_block_group_cache
*block_group
,
3352 struct btrfs_trans_handle
*trans
,
3353 struct btrfs_path
*path
)
3355 struct btrfs_root
*root
= block_group
->fs_info
->tree_root
;
3356 struct inode
*inode
= NULL
;
3358 int dcs
= BTRFS_DC_ERROR
;
3364 * If this block group is smaller than 100 megs don't bother caching the
3367 if (block_group
->key
.offset
< (100 * SZ_1M
)) {
3368 spin_lock(&block_group
->lock
);
3369 block_group
->disk_cache_state
= BTRFS_DC_WRITTEN
;
3370 spin_unlock(&block_group
->lock
);
3377 inode
= lookup_free_space_inode(root
, block_group
, path
);
3378 if (IS_ERR(inode
) && PTR_ERR(inode
) != -ENOENT
) {
3379 ret
= PTR_ERR(inode
);
3380 btrfs_release_path(path
);
3384 if (IS_ERR(inode
)) {
3388 if (block_group
->ro
)
3391 ret
= create_free_space_inode(root
, trans
, block_group
, path
);
3397 /* We've already setup this transaction, go ahead and exit */
3398 if (block_group
->cache_generation
== trans
->transid
&&
3399 i_size_read(inode
)) {
3400 dcs
= BTRFS_DC_SETUP
;
3405 * We want to set the generation to 0, that way if anything goes wrong
3406 * from here on out we know not to trust this cache when we load up next
3409 BTRFS_I(inode
)->generation
= 0;
3410 ret
= btrfs_update_inode(trans
, root
, inode
);
3413 * So theoretically we could recover from this, simply set the
3414 * super cache generation to 0 so we know to invalidate the
3415 * cache, but then we'd have to keep track of the block groups
3416 * that fail this way so we know we _have_ to reset this cache
3417 * before the next commit or risk reading stale cache. So to
3418 * limit our exposure to horrible edge cases lets just abort the
3419 * transaction, this only happens in really bad situations
3422 btrfs_abort_transaction(trans
, ret
);
3427 if (i_size_read(inode
) > 0) {
3428 ret
= btrfs_check_trunc_cache_free_space(root
,
3429 &root
->fs_info
->global_block_rsv
);
3433 ret
= btrfs_truncate_free_space_cache(root
, trans
, NULL
, inode
);
3438 spin_lock(&block_group
->lock
);
3439 if (block_group
->cached
!= BTRFS_CACHE_FINISHED
||
3440 !btrfs_test_opt(root
->fs_info
, SPACE_CACHE
)) {
3442 * don't bother trying to write stuff out _if_
3443 * a) we're not cached,
3444 * b) we're with nospace_cache mount option.
3446 dcs
= BTRFS_DC_WRITTEN
;
3447 spin_unlock(&block_group
->lock
);
3450 spin_unlock(&block_group
->lock
);
3453 * We hit an ENOSPC when setting up the cache in this transaction, just
3454 * skip doing the setup, we've already cleared the cache so we're safe.
3456 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC
, &trans
->transaction
->flags
)) {
3462 * Try to preallocate enough space based on how big the block group is.
3463 * Keep in mind this has to include any pinned space which could end up
3464 * taking up quite a bit since it's not folded into the other space
3467 num_pages
= div_u64(block_group
->key
.offset
, SZ_256M
);
3472 num_pages
*= PAGE_SIZE
;
3474 ret
= btrfs_check_data_free_space(inode
, 0, num_pages
);
3478 ret
= btrfs_prealloc_file_range_trans(inode
, trans
, 0, 0, num_pages
,
3479 num_pages
, num_pages
,
3482 * Our cache requires contiguous chunks so that we don't modify a bunch
3483 * of metadata or split extents when writing the cache out, which means
3484 * we can enospc if we are heavily fragmented in addition to just normal
3485 * out of space conditions. So if we hit this just skip setting up any
3486 * other block groups for this transaction, maybe we'll unpin enough
3487 * space the next time around.
3490 dcs
= BTRFS_DC_SETUP
;
3491 else if (ret
== -ENOSPC
)
3492 set_bit(BTRFS_TRANS_CACHE_ENOSPC
, &trans
->transaction
->flags
);
3497 btrfs_release_path(path
);
3499 spin_lock(&block_group
->lock
);
3500 if (!ret
&& dcs
== BTRFS_DC_SETUP
)
3501 block_group
->cache_generation
= trans
->transid
;
3502 block_group
->disk_cache_state
= dcs
;
3503 spin_unlock(&block_group
->lock
);
3508 int btrfs_setup_space_cache(struct btrfs_trans_handle
*trans
,
3509 struct btrfs_root
*root
)
3511 struct btrfs_block_group_cache
*cache
, *tmp
;
3512 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3513 struct btrfs_path
*path
;
3515 if (list_empty(&cur_trans
->dirty_bgs
) ||
3516 !btrfs_test_opt(root
->fs_info
, SPACE_CACHE
))
3519 path
= btrfs_alloc_path();
3523 /* Could add new block groups, use _safe just in case */
3524 list_for_each_entry_safe(cache
, tmp
, &cur_trans
->dirty_bgs
,
3526 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
)
3527 cache_save_setup(cache
, trans
, path
);
3530 btrfs_free_path(path
);
3535 * transaction commit does final block group cache writeback during a
3536 * critical section where nothing is allowed to change the FS. This is
3537 * required in order for the cache to actually match the block group,
3538 * but can introduce a lot of latency into the commit.
3540 * So, btrfs_start_dirty_block_groups is here to kick off block group
3541 * cache IO. There's a chance we'll have to redo some of it if the
3542 * block group changes again during the commit, but it greatly reduces
3543 * the commit latency by getting rid of the easy block groups while
3544 * we're still allowing others to join the commit.
3546 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle
*trans
,
3547 struct btrfs_root
*root
)
3549 struct btrfs_block_group_cache
*cache
;
3550 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3553 struct btrfs_path
*path
= NULL
;
3555 struct list_head
*io
= &cur_trans
->io_bgs
;
3556 int num_started
= 0;
3559 spin_lock(&cur_trans
->dirty_bgs_lock
);
3560 if (list_empty(&cur_trans
->dirty_bgs
)) {
3561 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3564 list_splice_init(&cur_trans
->dirty_bgs
, &dirty
);
3565 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3569 * make sure all the block groups on our dirty list actually
3572 btrfs_create_pending_block_groups(trans
, root
);
3575 path
= btrfs_alloc_path();
3581 * cache_write_mutex is here only to save us from balance or automatic
3582 * removal of empty block groups deleting this block group while we are
3583 * writing out the cache
3585 mutex_lock(&trans
->transaction
->cache_write_mutex
);
3586 while (!list_empty(&dirty
)) {
3587 cache
= list_first_entry(&dirty
,
3588 struct btrfs_block_group_cache
,
3591 * this can happen if something re-dirties a block
3592 * group that is already under IO. Just wait for it to
3593 * finish and then do it all again
3595 if (!list_empty(&cache
->io_list
)) {
3596 list_del_init(&cache
->io_list
);
3597 btrfs_wait_cache_io(root
, trans
, cache
,
3598 &cache
->io_ctl
, path
,
3599 cache
->key
.objectid
);
3600 btrfs_put_block_group(cache
);
3605 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3606 * if it should update the cache_state. Don't delete
3607 * until after we wait.
3609 * Since we're not running in the commit critical section
3610 * we need the dirty_bgs_lock to protect from update_block_group
3612 spin_lock(&cur_trans
->dirty_bgs_lock
);
3613 list_del_init(&cache
->dirty_list
);
3614 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3618 cache_save_setup(cache
, trans
, path
);
3620 if (cache
->disk_cache_state
== BTRFS_DC_SETUP
) {
3621 cache
->io_ctl
.inode
= NULL
;
3622 ret
= btrfs_write_out_cache(root
, trans
, cache
, path
);
3623 if (ret
== 0 && cache
->io_ctl
.inode
) {
3628 * the cache_write_mutex is protecting
3631 list_add_tail(&cache
->io_list
, io
);
3634 * if we failed to write the cache, the
3635 * generation will be bad and life goes on
3641 ret
= write_one_cache_group(trans
, root
, path
, cache
);
3643 * Our block group might still be attached to the list
3644 * of new block groups in the transaction handle of some
3645 * other task (struct btrfs_trans_handle->new_bgs). This
3646 * means its block group item isn't yet in the extent
3647 * tree. If this happens ignore the error, as we will
3648 * try again later in the critical section of the
3649 * transaction commit.
3651 if (ret
== -ENOENT
) {
3653 spin_lock(&cur_trans
->dirty_bgs_lock
);
3654 if (list_empty(&cache
->dirty_list
)) {
3655 list_add_tail(&cache
->dirty_list
,
3656 &cur_trans
->dirty_bgs
);
3657 btrfs_get_block_group(cache
);
3659 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3661 btrfs_abort_transaction(trans
, ret
);
3665 /* if its not on the io list, we need to put the block group */
3667 btrfs_put_block_group(cache
);
3673 * Avoid blocking other tasks for too long. It might even save
3674 * us from writing caches for block groups that are going to be
3677 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
3678 mutex_lock(&trans
->transaction
->cache_write_mutex
);
3680 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
3683 * go through delayed refs for all the stuff we've just kicked off
3684 * and then loop back (just once)
3686 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
3687 if (!ret
&& loops
== 0) {
3689 spin_lock(&cur_trans
->dirty_bgs_lock
);
3690 list_splice_init(&cur_trans
->dirty_bgs
, &dirty
);
3692 * dirty_bgs_lock protects us from concurrent block group
3693 * deletes too (not just cache_write_mutex).
3695 if (!list_empty(&dirty
)) {
3696 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3699 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3700 } else if (ret
< 0) {
3701 btrfs_cleanup_dirty_bgs(cur_trans
, root
);
3704 btrfs_free_path(path
);
3708 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
3709 struct btrfs_root
*root
)
3711 struct btrfs_block_group_cache
*cache
;
3712 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3715 struct btrfs_path
*path
;
3716 struct list_head
*io
= &cur_trans
->io_bgs
;
3717 int num_started
= 0;
3719 path
= btrfs_alloc_path();
3724 * Even though we are in the critical section of the transaction commit,
3725 * we can still have concurrent tasks adding elements to this
3726 * transaction's list of dirty block groups. These tasks correspond to
3727 * endio free space workers started when writeback finishes for a
3728 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3729 * allocate new block groups as a result of COWing nodes of the root
3730 * tree when updating the free space inode. The writeback for the space
3731 * caches is triggered by an earlier call to
3732 * btrfs_start_dirty_block_groups() and iterations of the following
3734 * Also we want to do the cache_save_setup first and then run the
3735 * delayed refs to make sure we have the best chance at doing this all
3738 spin_lock(&cur_trans
->dirty_bgs_lock
);
3739 while (!list_empty(&cur_trans
->dirty_bgs
)) {
3740 cache
= list_first_entry(&cur_trans
->dirty_bgs
,
3741 struct btrfs_block_group_cache
,
3745 * this can happen if cache_save_setup re-dirties a block
3746 * group that is already under IO. Just wait for it to
3747 * finish and then do it all again
3749 if (!list_empty(&cache
->io_list
)) {
3750 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3751 list_del_init(&cache
->io_list
);
3752 btrfs_wait_cache_io(root
, trans
, cache
,
3753 &cache
->io_ctl
, path
,
3754 cache
->key
.objectid
);
3755 btrfs_put_block_group(cache
);
3756 spin_lock(&cur_trans
->dirty_bgs_lock
);
3760 * don't remove from the dirty list until after we've waited
3763 list_del_init(&cache
->dirty_list
);
3764 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3767 cache_save_setup(cache
, trans
, path
);
3770 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long) -1);
3772 if (!ret
&& cache
->disk_cache_state
== BTRFS_DC_SETUP
) {
3773 cache
->io_ctl
.inode
= NULL
;
3774 ret
= btrfs_write_out_cache(root
, trans
, cache
, path
);
3775 if (ret
== 0 && cache
->io_ctl
.inode
) {
3778 list_add_tail(&cache
->io_list
, io
);
3781 * if we failed to write the cache, the
3782 * generation will be bad and life goes on
3788 ret
= write_one_cache_group(trans
, root
, path
, cache
);
3790 * One of the free space endio workers might have
3791 * created a new block group while updating a free space
3792 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3793 * and hasn't released its transaction handle yet, in
3794 * which case the new block group is still attached to
3795 * its transaction handle and its creation has not
3796 * finished yet (no block group item in the extent tree
3797 * yet, etc). If this is the case, wait for all free
3798 * space endio workers to finish and retry. This is a
3799 * a very rare case so no need for a more efficient and
3802 if (ret
== -ENOENT
) {
3803 wait_event(cur_trans
->writer_wait
,
3804 atomic_read(&cur_trans
->num_writers
) == 1);
3805 ret
= write_one_cache_group(trans
, root
, path
,
3809 btrfs_abort_transaction(trans
, ret
);
3812 /* if its not on the io list, we need to put the block group */
3814 btrfs_put_block_group(cache
);
3815 spin_lock(&cur_trans
->dirty_bgs_lock
);
3817 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3819 while (!list_empty(io
)) {
3820 cache
= list_first_entry(io
, struct btrfs_block_group_cache
,
3822 list_del_init(&cache
->io_list
);
3823 btrfs_wait_cache_io(root
, trans
, cache
,
3824 &cache
->io_ctl
, path
, cache
->key
.objectid
);
3825 btrfs_put_block_group(cache
);
3828 btrfs_free_path(path
);
3832 int btrfs_extent_readonly(struct btrfs_root
*root
, u64 bytenr
)
3834 struct btrfs_block_group_cache
*block_group
;
3837 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
3838 if (!block_group
|| block_group
->ro
)
3841 btrfs_put_block_group(block_group
);
3845 bool btrfs_inc_nocow_writers(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
3847 struct btrfs_block_group_cache
*bg
;
3850 bg
= btrfs_lookup_block_group(fs_info
, bytenr
);
3854 spin_lock(&bg
->lock
);
3858 atomic_inc(&bg
->nocow_writers
);
3859 spin_unlock(&bg
->lock
);
3861 /* no put on block group, done by btrfs_dec_nocow_writers */
3863 btrfs_put_block_group(bg
);
3869 void btrfs_dec_nocow_writers(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
3871 struct btrfs_block_group_cache
*bg
;
3873 bg
= btrfs_lookup_block_group(fs_info
, bytenr
);
3875 if (atomic_dec_and_test(&bg
->nocow_writers
))
3876 wake_up_atomic_t(&bg
->nocow_writers
);
3878 * Once for our lookup and once for the lookup done by a previous call
3879 * to btrfs_inc_nocow_writers()
3881 btrfs_put_block_group(bg
);
3882 btrfs_put_block_group(bg
);
3885 static int btrfs_wait_nocow_writers_atomic_t(atomic_t
*a
)
3891 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache
*bg
)
3893 wait_on_atomic_t(&bg
->nocow_writers
,
3894 btrfs_wait_nocow_writers_atomic_t
,
3895 TASK_UNINTERRUPTIBLE
);
3898 static const char *alloc_name(u64 flags
)
3901 case BTRFS_BLOCK_GROUP_METADATA
|BTRFS_BLOCK_GROUP_DATA
:
3903 case BTRFS_BLOCK_GROUP_METADATA
:
3905 case BTRFS_BLOCK_GROUP_DATA
:
3907 case BTRFS_BLOCK_GROUP_SYSTEM
:
3911 return "invalid-combination";
3915 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
3916 u64 total_bytes
, u64 bytes_used
,
3918 struct btrfs_space_info
**space_info
)
3920 struct btrfs_space_info
*found
;
3925 if (flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3926 BTRFS_BLOCK_GROUP_RAID10
))
3931 found
= __find_space_info(info
, flags
);
3933 spin_lock(&found
->lock
);
3934 found
->total_bytes
+= total_bytes
;
3935 found
->disk_total
+= total_bytes
* factor
;
3936 found
->bytes_used
+= bytes_used
;
3937 found
->disk_used
+= bytes_used
* factor
;
3938 found
->bytes_readonly
+= bytes_readonly
;
3939 if (total_bytes
> 0)
3941 space_info_add_new_bytes(info
, found
, total_bytes
-
3942 bytes_used
- bytes_readonly
);
3943 spin_unlock(&found
->lock
);
3944 *space_info
= found
;
3947 found
= kzalloc(sizeof(*found
), GFP_NOFS
);
3951 ret
= percpu_counter_init(&found
->total_bytes_pinned
, 0, GFP_KERNEL
);
3957 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
3958 INIT_LIST_HEAD(&found
->block_groups
[i
]);
3959 init_rwsem(&found
->groups_sem
);
3960 spin_lock_init(&found
->lock
);
3961 found
->flags
= flags
& BTRFS_BLOCK_GROUP_TYPE_MASK
;
3962 found
->total_bytes
= total_bytes
;
3963 found
->disk_total
= total_bytes
* factor
;
3964 found
->bytes_used
= bytes_used
;
3965 found
->disk_used
= bytes_used
* factor
;
3966 found
->bytes_pinned
= 0;
3967 found
->bytes_reserved
= 0;
3968 found
->bytes_readonly
= bytes_readonly
;
3969 found
->bytes_may_use
= 0;
3971 found
->max_extent_size
= 0;
3972 found
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3973 found
->chunk_alloc
= 0;
3975 init_waitqueue_head(&found
->wait
);
3976 INIT_LIST_HEAD(&found
->ro_bgs
);
3977 INIT_LIST_HEAD(&found
->tickets
);
3978 INIT_LIST_HEAD(&found
->priority_tickets
);
3980 ret
= kobject_init_and_add(&found
->kobj
, &space_info_ktype
,
3981 info
->space_info_kobj
, "%s",
3982 alloc_name(found
->flags
));
3988 *space_info
= found
;
3989 list_add_rcu(&found
->list
, &info
->space_info
);
3990 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3991 info
->data_sinfo
= found
;
3996 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
3998 u64 extra_flags
= chunk_to_extended(flags
) &
3999 BTRFS_EXTENDED_PROFILE_MASK
;
4001 write_seqlock(&fs_info
->profiles_lock
);
4002 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
4003 fs_info
->avail_data_alloc_bits
|= extra_flags
;
4004 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
4005 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
4006 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
4007 fs_info
->avail_system_alloc_bits
|= extra_flags
;
4008 write_sequnlock(&fs_info
->profiles_lock
);
4012 * returns target flags in extended format or 0 if restripe for this
4013 * chunk_type is not in progress
4015 * should be called with either volume_mutex or balance_lock held
4017 static u64
get_restripe_target(struct btrfs_fs_info
*fs_info
, u64 flags
)
4019 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
4025 if (flags
& BTRFS_BLOCK_GROUP_DATA
&&
4026 bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
4027 target
= BTRFS_BLOCK_GROUP_DATA
| bctl
->data
.target
;
4028 } else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
&&
4029 bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
4030 target
= BTRFS_BLOCK_GROUP_SYSTEM
| bctl
->sys
.target
;
4031 } else if (flags
& BTRFS_BLOCK_GROUP_METADATA
&&
4032 bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
4033 target
= BTRFS_BLOCK_GROUP_METADATA
| bctl
->meta
.target
;
4040 * @flags: available profiles in extended format (see ctree.h)
4042 * Returns reduced profile in chunk format. If profile changing is in
4043 * progress (either running or paused) picks the target profile (if it's
4044 * already available), otherwise falls back to plain reducing.
4046 static u64
btrfs_reduce_alloc_profile(struct btrfs_root
*root
, u64 flags
)
4048 u64 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
4054 * see if restripe for this chunk_type is in progress, if so
4055 * try to reduce to the target profile
4057 spin_lock(&root
->fs_info
->balance_lock
);
4058 target
= get_restripe_target(root
->fs_info
, flags
);
4060 /* pick target profile only if it's already available */
4061 if ((flags
& target
) & BTRFS_EXTENDED_PROFILE_MASK
) {
4062 spin_unlock(&root
->fs_info
->balance_lock
);
4063 return extended_to_chunk(target
);
4066 spin_unlock(&root
->fs_info
->balance_lock
);
4068 /* First, mask out the RAID levels which aren't possible */
4069 for (raid_type
= 0; raid_type
< BTRFS_NR_RAID_TYPES
; raid_type
++) {
4070 if (num_devices
>= btrfs_raid_array
[raid_type
].devs_min
)
4071 allowed
|= btrfs_raid_group
[raid_type
];
4075 if (allowed
& BTRFS_BLOCK_GROUP_RAID6
)
4076 allowed
= BTRFS_BLOCK_GROUP_RAID6
;
4077 else if (allowed
& BTRFS_BLOCK_GROUP_RAID5
)
4078 allowed
= BTRFS_BLOCK_GROUP_RAID5
;
4079 else if (allowed
& BTRFS_BLOCK_GROUP_RAID10
)
4080 allowed
= BTRFS_BLOCK_GROUP_RAID10
;
4081 else if (allowed
& BTRFS_BLOCK_GROUP_RAID1
)
4082 allowed
= BTRFS_BLOCK_GROUP_RAID1
;
4083 else if (allowed
& BTRFS_BLOCK_GROUP_RAID0
)
4084 allowed
= BTRFS_BLOCK_GROUP_RAID0
;
4086 flags
&= ~BTRFS_BLOCK_GROUP_PROFILE_MASK
;
4088 return extended_to_chunk(flags
| allowed
);
4091 static u64
get_alloc_profile(struct btrfs_root
*root
, u64 orig_flags
)
4098 seq
= read_seqbegin(&root
->fs_info
->profiles_lock
);
4100 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
4101 flags
|= root
->fs_info
->avail_data_alloc_bits
;
4102 else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
4103 flags
|= root
->fs_info
->avail_system_alloc_bits
;
4104 else if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
4105 flags
|= root
->fs_info
->avail_metadata_alloc_bits
;
4106 } while (read_seqretry(&root
->fs_info
->profiles_lock
, seq
));
4108 return btrfs_reduce_alloc_profile(root
, flags
);
4111 u64
btrfs_get_alloc_profile(struct btrfs_root
*root
, int data
)
4117 flags
= BTRFS_BLOCK_GROUP_DATA
;
4118 else if (root
== root
->fs_info
->chunk_root
)
4119 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
4121 flags
= BTRFS_BLOCK_GROUP_METADATA
;
4123 ret
= get_alloc_profile(root
, flags
);
4127 int btrfs_alloc_data_chunk_ondemand(struct inode
*inode
, u64 bytes
)
4129 struct btrfs_space_info
*data_sinfo
;
4130 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4131 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4134 int need_commit
= 2;
4135 int have_pinned_space
;
4137 /* make sure bytes are sectorsize aligned */
4138 bytes
= ALIGN(bytes
, root
->sectorsize
);
4140 if (btrfs_is_free_space_inode(inode
)) {
4142 ASSERT(current
->journal_info
);
4145 data_sinfo
= fs_info
->data_sinfo
;
4150 /* make sure we have enough space to handle the data first */
4151 spin_lock(&data_sinfo
->lock
);
4152 used
= data_sinfo
->bytes_used
+ data_sinfo
->bytes_reserved
+
4153 data_sinfo
->bytes_pinned
+ data_sinfo
->bytes_readonly
+
4154 data_sinfo
->bytes_may_use
;
4156 if (used
+ bytes
> data_sinfo
->total_bytes
) {
4157 struct btrfs_trans_handle
*trans
;
4160 * if we don't have enough free bytes in this space then we need
4161 * to alloc a new chunk.
4163 if (!data_sinfo
->full
) {
4166 data_sinfo
->force_alloc
= CHUNK_ALLOC_FORCE
;
4167 spin_unlock(&data_sinfo
->lock
);
4169 alloc_target
= btrfs_get_alloc_profile(root
, 1);
4171 * It is ugly that we don't call nolock join
4172 * transaction for the free space inode case here.
4173 * But it is safe because we only do the data space
4174 * reservation for the free space cache in the
4175 * transaction context, the common join transaction
4176 * just increase the counter of the current transaction
4177 * handler, doesn't try to acquire the trans_lock of
4180 trans
= btrfs_join_transaction(root
);
4182 return PTR_ERR(trans
);
4184 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
4186 CHUNK_ALLOC_NO_FORCE
);
4187 btrfs_end_transaction(trans
, root
);
4192 have_pinned_space
= 1;
4198 data_sinfo
= fs_info
->data_sinfo
;
4204 * If we don't have enough pinned space to deal with this
4205 * allocation, and no removed chunk in current transaction,
4206 * don't bother committing the transaction.
4208 have_pinned_space
= percpu_counter_compare(
4209 &data_sinfo
->total_bytes_pinned
,
4210 used
+ bytes
- data_sinfo
->total_bytes
);
4211 spin_unlock(&data_sinfo
->lock
);
4213 /* commit the current transaction and try again */
4216 !atomic_read(&root
->fs_info
->open_ioctl_trans
)) {
4219 if (need_commit
> 0) {
4220 btrfs_start_delalloc_roots(fs_info
, 0, -1);
4221 btrfs_wait_ordered_roots(fs_info
, -1, 0, (u64
)-1);
4224 trans
= btrfs_join_transaction(root
);
4226 return PTR_ERR(trans
);
4227 if (have_pinned_space
>= 0 ||
4228 test_bit(BTRFS_TRANS_HAVE_FREE_BGS
,
4229 &trans
->transaction
->flags
) ||
4231 ret
= btrfs_commit_transaction(trans
, root
);
4235 * The cleaner kthread might still be doing iput
4236 * operations. Wait for it to finish so that
4237 * more space is released.
4239 mutex_lock(&root
->fs_info
->cleaner_delayed_iput_mutex
);
4240 mutex_unlock(&root
->fs_info
->cleaner_delayed_iput_mutex
);
4243 btrfs_end_transaction(trans
, root
);
4247 trace_btrfs_space_reservation(root
->fs_info
,
4248 "space_info:enospc",
4249 data_sinfo
->flags
, bytes
, 1);
4252 data_sinfo
->bytes_may_use
+= bytes
;
4253 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
4254 data_sinfo
->flags
, bytes
, 1);
4255 spin_unlock(&data_sinfo
->lock
);
4261 * New check_data_free_space() with ability for precious data reservation
4262 * Will replace old btrfs_check_data_free_space(), but for patch split,
4263 * add a new function first and then replace it.
4265 int btrfs_check_data_free_space(struct inode
*inode
, u64 start
, u64 len
)
4267 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4270 /* align the range */
4271 len
= round_up(start
+ len
, root
->sectorsize
) -
4272 round_down(start
, root
->sectorsize
);
4273 start
= round_down(start
, root
->sectorsize
);
4275 ret
= btrfs_alloc_data_chunk_ondemand(inode
, len
);
4280 * Use new btrfs_qgroup_reserve_data to reserve precious data space
4282 * TODO: Find a good method to avoid reserve data space for NOCOW
4283 * range, but don't impact performance on quota disable case.
4285 ret
= btrfs_qgroup_reserve_data(inode
, start
, len
);
4290 * Called if we need to clear a data reservation for this inode
4291 * Normally in a error case.
4293 * This one will *NOT* use accurate qgroup reserved space API, just for case
4294 * which we can't sleep and is sure it won't affect qgroup reserved space.
4295 * Like clear_bit_hook().
4297 void btrfs_free_reserved_data_space_noquota(struct inode
*inode
, u64 start
,
4300 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4301 struct btrfs_space_info
*data_sinfo
;
4303 /* Make sure the range is aligned to sectorsize */
4304 len
= round_up(start
+ len
, root
->sectorsize
) -
4305 round_down(start
, root
->sectorsize
);
4306 start
= round_down(start
, root
->sectorsize
);
4308 data_sinfo
= root
->fs_info
->data_sinfo
;
4309 spin_lock(&data_sinfo
->lock
);
4310 if (WARN_ON(data_sinfo
->bytes_may_use
< len
))
4311 data_sinfo
->bytes_may_use
= 0;
4313 data_sinfo
->bytes_may_use
-= len
;
4314 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
4315 data_sinfo
->flags
, len
, 0);
4316 spin_unlock(&data_sinfo
->lock
);
4320 * Called if we need to clear a data reservation for this inode
4321 * Normally in a error case.
4323 * This one will handle the per-inode data rsv map for accurate reserved
4326 void btrfs_free_reserved_data_space(struct inode
*inode
, u64 start
, u64 len
)
4328 btrfs_free_reserved_data_space_noquota(inode
, start
, len
);
4329 btrfs_qgroup_free_data(inode
, start
, len
);
4332 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
4334 struct list_head
*head
= &info
->space_info
;
4335 struct btrfs_space_info
*found
;
4338 list_for_each_entry_rcu(found
, head
, list
) {
4339 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
4340 found
->force_alloc
= CHUNK_ALLOC_FORCE
;
4345 static inline u64
calc_global_rsv_need_space(struct btrfs_block_rsv
*global
)
4347 return (global
->size
<< 1);
4350 static int should_alloc_chunk(struct btrfs_root
*root
,
4351 struct btrfs_space_info
*sinfo
, int force
)
4353 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
4354 u64 num_bytes
= sinfo
->total_bytes
- sinfo
->bytes_readonly
;
4355 u64 num_allocated
= sinfo
->bytes_used
+ sinfo
->bytes_reserved
;
4358 if (force
== CHUNK_ALLOC_FORCE
)
4362 * We need to take into account the global rsv because for all intents
4363 * and purposes it's used space. Don't worry about locking the
4364 * global_rsv, it doesn't change except when the transaction commits.
4366 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
4367 num_allocated
+= calc_global_rsv_need_space(global_rsv
);
4370 * in limited mode, we want to have some free space up to
4371 * about 1% of the FS size.
4373 if (force
== CHUNK_ALLOC_LIMITED
) {
4374 thresh
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
4375 thresh
= max_t(u64
, SZ_64M
, div_factor_fine(thresh
, 1));
4377 if (num_bytes
- num_allocated
< thresh
)
4381 if (num_allocated
+ SZ_2M
< div_factor(num_bytes
, 8))
4386 static u64
get_profile_num_devs(struct btrfs_root
*root
, u64 type
)
4390 if (type
& (BTRFS_BLOCK_GROUP_RAID10
|
4391 BTRFS_BLOCK_GROUP_RAID0
|
4392 BTRFS_BLOCK_GROUP_RAID5
|
4393 BTRFS_BLOCK_GROUP_RAID6
))
4394 num_dev
= root
->fs_info
->fs_devices
->rw_devices
;
4395 else if (type
& BTRFS_BLOCK_GROUP_RAID1
)
4398 num_dev
= 1; /* DUP or single */
4404 * If @is_allocation is true, reserve space in the system space info necessary
4405 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4408 void check_system_chunk(struct btrfs_trans_handle
*trans
,
4409 struct btrfs_root
*root
,
4412 struct btrfs_space_info
*info
;
4419 * Needed because we can end up allocating a system chunk and for an
4420 * atomic and race free space reservation in the chunk block reserve.
4422 ASSERT(mutex_is_locked(&root
->fs_info
->chunk_mutex
));
4424 info
= __find_space_info(root
->fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
4425 spin_lock(&info
->lock
);
4426 left
= info
->total_bytes
- info
->bytes_used
- info
->bytes_pinned
-
4427 info
->bytes_reserved
- info
->bytes_readonly
-
4428 info
->bytes_may_use
;
4429 spin_unlock(&info
->lock
);
4431 num_devs
= get_profile_num_devs(root
, type
);
4433 /* num_devs device items to update and 1 chunk item to add or remove */
4434 thresh
= btrfs_calc_trunc_metadata_size(root
, num_devs
) +
4435 btrfs_calc_trans_metadata_size(root
, 1);
4437 if (left
< thresh
&& btrfs_test_opt(root
->fs_info
, ENOSPC_DEBUG
)) {
4438 btrfs_info(root
->fs_info
, "left=%llu, need=%llu, flags=%llu",
4439 left
, thresh
, type
);
4440 dump_space_info(info
, 0, 0);
4443 if (left
< thresh
) {
4446 flags
= btrfs_get_alloc_profile(root
->fs_info
->chunk_root
, 0);
4448 * Ignore failure to create system chunk. We might end up not
4449 * needing it, as we might not need to COW all nodes/leafs from
4450 * the paths we visit in the chunk tree (they were already COWed
4451 * or created in the current transaction for example).
4453 ret
= btrfs_alloc_chunk(trans
, root
, flags
);
4457 ret
= btrfs_block_rsv_add(root
->fs_info
->chunk_root
,
4458 &root
->fs_info
->chunk_block_rsv
,
4459 thresh
, BTRFS_RESERVE_NO_FLUSH
);
4461 trans
->chunk_bytes_reserved
+= thresh
;
4466 * If force is CHUNK_ALLOC_FORCE:
4467 * - return 1 if it successfully allocates a chunk,
4468 * - return errors including -ENOSPC otherwise.
4469 * If force is NOT CHUNK_ALLOC_FORCE:
4470 * - return 0 if it doesn't need to allocate a new chunk,
4471 * - return 1 if it successfully allocates a chunk,
4472 * - return errors including -ENOSPC otherwise.
4474 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
4475 struct btrfs_root
*extent_root
, u64 flags
, int force
)
4477 struct btrfs_space_info
*space_info
;
4478 struct btrfs_fs_info
*fs_info
= extent_root
->fs_info
;
4479 int wait_for_alloc
= 0;
4482 /* Don't re-enter if we're already allocating a chunk */
4483 if (trans
->allocating_chunk
)
4486 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
4488 ret
= update_space_info(extent_root
->fs_info
, flags
,
4489 0, 0, 0, &space_info
);
4490 BUG_ON(ret
); /* -ENOMEM */
4492 BUG_ON(!space_info
); /* Logic error */
4495 spin_lock(&space_info
->lock
);
4496 if (force
< space_info
->force_alloc
)
4497 force
= space_info
->force_alloc
;
4498 if (space_info
->full
) {
4499 if (should_alloc_chunk(extent_root
, space_info
, force
))
4503 spin_unlock(&space_info
->lock
);
4507 if (!should_alloc_chunk(extent_root
, space_info
, force
)) {
4508 spin_unlock(&space_info
->lock
);
4510 } else if (space_info
->chunk_alloc
) {
4513 space_info
->chunk_alloc
= 1;
4516 spin_unlock(&space_info
->lock
);
4518 mutex_lock(&fs_info
->chunk_mutex
);
4521 * The chunk_mutex is held throughout the entirety of a chunk
4522 * allocation, so once we've acquired the chunk_mutex we know that the
4523 * other guy is done and we need to recheck and see if we should
4526 if (wait_for_alloc
) {
4527 mutex_unlock(&fs_info
->chunk_mutex
);
4532 trans
->allocating_chunk
= true;
4535 * If we have mixed data/metadata chunks we want to make sure we keep
4536 * allocating mixed chunks instead of individual chunks.
4538 if (btrfs_mixed_space_info(space_info
))
4539 flags
|= (BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
);
4542 * if we're doing a data chunk, go ahead and make sure that
4543 * we keep a reasonable number of metadata chunks allocated in the
4546 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
4547 fs_info
->data_chunk_allocations
++;
4548 if (!(fs_info
->data_chunk_allocations
%
4549 fs_info
->metadata_ratio
))
4550 force_metadata_allocation(fs_info
);
4554 * Check if we have enough space in SYSTEM chunk because we may need
4555 * to update devices.
4557 check_system_chunk(trans
, extent_root
, flags
);
4559 ret
= btrfs_alloc_chunk(trans
, extent_root
, flags
);
4560 trans
->allocating_chunk
= false;
4562 spin_lock(&space_info
->lock
);
4563 if (ret
< 0 && ret
!= -ENOSPC
)
4566 space_info
->full
= 1;
4570 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
4572 space_info
->chunk_alloc
= 0;
4573 spin_unlock(&space_info
->lock
);
4574 mutex_unlock(&fs_info
->chunk_mutex
);
4576 * When we allocate a new chunk we reserve space in the chunk block
4577 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4578 * add new nodes/leafs to it if we end up needing to do it when
4579 * inserting the chunk item and updating device items as part of the
4580 * second phase of chunk allocation, performed by
4581 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4582 * large number of new block groups to create in our transaction
4583 * handle's new_bgs list to avoid exhausting the chunk block reserve
4584 * in extreme cases - like having a single transaction create many new
4585 * block groups when starting to write out the free space caches of all
4586 * the block groups that were made dirty during the lifetime of the
4589 if (trans
->can_flush_pending_bgs
&&
4590 trans
->chunk_bytes_reserved
>= (u64
)SZ_2M
) {
4591 btrfs_create_pending_block_groups(trans
, extent_root
);
4592 btrfs_trans_release_chunk_metadata(trans
);
4597 static int can_overcommit(struct btrfs_root
*root
,
4598 struct btrfs_space_info
*space_info
, u64 bytes
,
4599 enum btrfs_reserve_flush_enum flush
)
4601 struct btrfs_block_rsv
*global_rsv
;
4607 /* Don't overcommit when in mixed mode. */
4608 if (space_info
->flags
& BTRFS_BLOCK_GROUP_DATA
)
4611 BUG_ON(root
->fs_info
== NULL
);
4612 global_rsv
= &root
->fs_info
->global_block_rsv
;
4613 profile
= btrfs_get_alloc_profile(root
, 0);
4614 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4615 space_info
->bytes_pinned
+ space_info
->bytes_readonly
;
4618 * We only want to allow over committing if we have lots of actual space
4619 * free, but if we don't have enough space to handle the global reserve
4620 * space then we could end up having a real enospc problem when trying
4621 * to allocate a chunk or some other such important allocation.
4623 spin_lock(&global_rsv
->lock
);
4624 space_size
= calc_global_rsv_need_space(global_rsv
);
4625 spin_unlock(&global_rsv
->lock
);
4626 if (used
+ space_size
>= space_info
->total_bytes
)
4629 used
+= space_info
->bytes_may_use
;
4631 spin_lock(&root
->fs_info
->free_chunk_lock
);
4632 avail
= root
->fs_info
->free_chunk_space
;
4633 spin_unlock(&root
->fs_info
->free_chunk_lock
);
4636 * If we have dup, raid1 or raid10 then only half of the free
4637 * space is actually useable. For raid56, the space info used
4638 * doesn't include the parity drive, so we don't have to
4641 if (profile
& (BTRFS_BLOCK_GROUP_DUP
|
4642 BTRFS_BLOCK_GROUP_RAID1
|
4643 BTRFS_BLOCK_GROUP_RAID10
))
4647 * If we aren't flushing all things, let us overcommit up to
4648 * 1/2th of the space. If we can flush, don't let us overcommit
4649 * too much, let it overcommit up to 1/8 of the space.
4651 if (flush
== BTRFS_RESERVE_FLUSH_ALL
)
4656 if (used
+ bytes
< space_info
->total_bytes
+ avail
)
4661 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root
*root
,
4662 unsigned long nr_pages
, int nr_items
)
4664 struct super_block
*sb
= root
->fs_info
->sb
;
4666 if (down_read_trylock(&sb
->s_umount
)) {
4667 writeback_inodes_sb_nr(sb
, nr_pages
, WB_REASON_FS_FREE_SPACE
);
4668 up_read(&sb
->s_umount
);
4671 * We needn't worry the filesystem going from r/w to r/o though
4672 * we don't acquire ->s_umount mutex, because the filesystem
4673 * should guarantee the delalloc inodes list be empty after
4674 * the filesystem is readonly(all dirty pages are written to
4677 btrfs_start_delalloc_roots(root
->fs_info
, 0, nr_items
);
4678 if (!current
->journal_info
)
4679 btrfs_wait_ordered_roots(root
->fs_info
, nr_items
,
4684 static inline int calc_reclaim_items_nr(struct btrfs_root
*root
, u64 to_reclaim
)
4689 bytes
= btrfs_calc_trans_metadata_size(root
, 1);
4690 nr
= (int)div64_u64(to_reclaim
, bytes
);
4696 #define EXTENT_SIZE_PER_ITEM SZ_256K
4699 * shrink metadata reservation for delalloc
4701 static void shrink_delalloc(struct btrfs_root
*root
, u64 to_reclaim
, u64 orig
,
4704 struct btrfs_block_rsv
*block_rsv
;
4705 struct btrfs_space_info
*space_info
;
4706 struct btrfs_trans_handle
*trans
;
4710 unsigned long nr_pages
;
4713 enum btrfs_reserve_flush_enum flush
;
4715 /* Calc the number of the pages we need flush for space reservation */
4716 items
= calc_reclaim_items_nr(root
, to_reclaim
);
4717 to_reclaim
= (u64
)items
* EXTENT_SIZE_PER_ITEM
;
4719 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4720 block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
4721 space_info
= block_rsv
->space_info
;
4723 delalloc_bytes
= percpu_counter_sum_positive(
4724 &root
->fs_info
->delalloc_bytes
);
4725 if (delalloc_bytes
== 0) {
4729 btrfs_wait_ordered_roots(root
->fs_info
, items
,
4735 while (delalloc_bytes
&& loops
< 3) {
4736 max_reclaim
= min(delalloc_bytes
, to_reclaim
);
4737 nr_pages
= max_reclaim
>> PAGE_SHIFT
;
4738 btrfs_writeback_inodes_sb_nr(root
, nr_pages
, items
);
4740 * We need to wait for the async pages to actually start before
4743 max_reclaim
= atomic_read(&root
->fs_info
->async_delalloc_pages
);
4747 if (max_reclaim
<= nr_pages
)
4750 max_reclaim
-= nr_pages
;
4752 wait_event(root
->fs_info
->async_submit_wait
,
4753 atomic_read(&root
->fs_info
->async_delalloc_pages
) <=
4757 flush
= BTRFS_RESERVE_FLUSH_ALL
;
4759 flush
= BTRFS_RESERVE_NO_FLUSH
;
4760 spin_lock(&space_info
->lock
);
4761 if (can_overcommit(root
, space_info
, orig
, flush
)) {
4762 spin_unlock(&space_info
->lock
);
4765 if (list_empty(&space_info
->tickets
) &&
4766 list_empty(&space_info
->priority_tickets
)) {
4767 spin_unlock(&space_info
->lock
);
4770 spin_unlock(&space_info
->lock
);
4773 if (wait_ordered
&& !trans
) {
4774 btrfs_wait_ordered_roots(root
->fs_info
, items
,
4777 time_left
= schedule_timeout_killable(1);
4781 delalloc_bytes
= percpu_counter_sum_positive(
4782 &root
->fs_info
->delalloc_bytes
);
4787 * maybe_commit_transaction - possibly commit the transaction if its ok to
4788 * @root - the root we're allocating for
4789 * @bytes - the number of bytes we want to reserve
4790 * @force - force the commit
4792 * This will check to make sure that committing the transaction will actually
4793 * get us somewhere and then commit the transaction if it does. Otherwise it
4794 * will return -ENOSPC.
4796 static int may_commit_transaction(struct btrfs_root
*root
,
4797 struct btrfs_space_info
*space_info
,
4798 u64 bytes
, int force
)
4800 struct btrfs_block_rsv
*delayed_rsv
= &root
->fs_info
->delayed_block_rsv
;
4801 struct btrfs_trans_handle
*trans
;
4803 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4810 /* See if there is enough pinned space to make this reservation */
4811 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4816 * See if there is some space in the delayed insertion reservation for
4819 if (space_info
!= delayed_rsv
->space_info
)
4822 spin_lock(&delayed_rsv
->lock
);
4823 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4824 bytes
- delayed_rsv
->size
) >= 0) {
4825 spin_unlock(&delayed_rsv
->lock
);
4828 spin_unlock(&delayed_rsv
->lock
);
4831 trans
= btrfs_join_transaction(root
);
4835 return btrfs_commit_transaction(trans
, root
);
4838 struct reserve_ticket
{
4841 struct list_head list
;
4842 wait_queue_head_t wait
;
4845 static int flush_space(struct btrfs_root
*root
,
4846 struct btrfs_space_info
*space_info
, u64 num_bytes
,
4847 u64 orig_bytes
, int state
)
4849 struct btrfs_trans_handle
*trans
;
4854 case FLUSH_DELAYED_ITEMS_NR
:
4855 case FLUSH_DELAYED_ITEMS
:
4856 if (state
== FLUSH_DELAYED_ITEMS_NR
)
4857 nr
= calc_reclaim_items_nr(root
, num_bytes
) * 2;
4861 trans
= btrfs_join_transaction(root
);
4862 if (IS_ERR(trans
)) {
4863 ret
= PTR_ERR(trans
);
4866 ret
= btrfs_run_delayed_items_nr(trans
, root
, nr
);
4867 btrfs_end_transaction(trans
, root
);
4869 case FLUSH_DELALLOC
:
4870 case FLUSH_DELALLOC_WAIT
:
4871 shrink_delalloc(root
, num_bytes
* 2, orig_bytes
,
4872 state
== FLUSH_DELALLOC_WAIT
);
4875 trans
= btrfs_join_transaction(root
);
4876 if (IS_ERR(trans
)) {
4877 ret
= PTR_ERR(trans
);
4880 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
4881 btrfs_get_alloc_profile(root
, 0),
4882 CHUNK_ALLOC_NO_FORCE
);
4883 btrfs_end_transaction(trans
, root
);
4884 if (ret
> 0 || ret
== -ENOSPC
)
4888 ret
= may_commit_transaction(root
, space_info
, orig_bytes
, 0);
4895 trace_btrfs_flush_space(root
->fs_info
, space_info
->flags
, num_bytes
,
4896 orig_bytes
, state
, ret
);
4901 btrfs_calc_reclaim_metadata_size(struct btrfs_root
*root
,
4902 struct btrfs_space_info
*space_info
)
4904 struct reserve_ticket
*ticket
;
4909 list_for_each_entry(ticket
, &space_info
->tickets
, list
)
4910 to_reclaim
+= ticket
->bytes
;
4911 list_for_each_entry(ticket
, &space_info
->priority_tickets
, list
)
4912 to_reclaim
+= ticket
->bytes
;
4916 to_reclaim
= min_t(u64
, num_online_cpus() * SZ_1M
, SZ_16M
);
4917 if (can_overcommit(root
, space_info
, to_reclaim
,
4918 BTRFS_RESERVE_FLUSH_ALL
))
4921 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4922 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4923 space_info
->bytes_may_use
;
4924 if (can_overcommit(root
, space_info
, SZ_1M
, BTRFS_RESERVE_FLUSH_ALL
))
4925 expected
= div_factor_fine(space_info
->total_bytes
, 95);
4927 expected
= div_factor_fine(space_info
->total_bytes
, 90);
4929 if (used
> expected
)
4930 to_reclaim
= used
- expected
;
4933 to_reclaim
= min(to_reclaim
, space_info
->bytes_may_use
+
4934 space_info
->bytes_reserved
);
4938 static inline int need_do_async_reclaim(struct btrfs_space_info
*space_info
,
4939 struct btrfs_root
*root
, u64 used
)
4941 u64 thresh
= div_factor_fine(space_info
->total_bytes
, 98);
4943 /* If we're just plain full then async reclaim just slows us down. */
4944 if ((space_info
->bytes_used
+ space_info
->bytes_reserved
) >= thresh
)
4947 if (!btrfs_calc_reclaim_metadata_size(root
, space_info
))
4950 return (used
>= thresh
&& !btrfs_fs_closing(root
->fs_info
) &&
4951 !test_bit(BTRFS_FS_STATE_REMOUNTING
,
4952 &root
->fs_info
->fs_state
));
4955 static void wake_all_tickets(struct list_head
*head
)
4957 struct reserve_ticket
*ticket
;
4959 while (!list_empty(head
)) {
4960 ticket
= list_first_entry(head
, struct reserve_ticket
, list
);
4961 list_del_init(&ticket
->list
);
4962 ticket
->error
= -ENOSPC
;
4963 wake_up(&ticket
->wait
);
4968 * This is for normal flushers, we can wait all goddamned day if we want to. We
4969 * will loop and continuously try to flush as long as we are making progress.
4970 * We count progress as clearing off tickets each time we have to loop.
4972 static void btrfs_async_reclaim_metadata_space(struct work_struct
*work
)
4974 struct reserve_ticket
*last_ticket
= NULL
;
4975 struct btrfs_fs_info
*fs_info
;
4976 struct btrfs_space_info
*space_info
;
4979 int commit_cycles
= 0;
4981 fs_info
= container_of(work
, struct btrfs_fs_info
, async_reclaim_work
);
4982 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4984 spin_lock(&space_info
->lock
);
4985 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
->fs_root
,
4988 space_info
->flush
= 0;
4989 spin_unlock(&space_info
->lock
);
4992 last_ticket
= list_first_entry(&space_info
->tickets
,
4993 struct reserve_ticket
, list
);
4994 spin_unlock(&space_info
->lock
);
4996 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
4998 struct reserve_ticket
*ticket
;
5001 ret
= flush_space(fs_info
->fs_root
, space_info
, to_reclaim
,
5002 to_reclaim
, flush_state
);
5003 spin_lock(&space_info
->lock
);
5004 if (list_empty(&space_info
->tickets
)) {
5005 space_info
->flush
= 0;
5006 spin_unlock(&space_info
->lock
);
5009 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
->fs_root
,
5011 ticket
= list_first_entry(&space_info
->tickets
,
5012 struct reserve_ticket
, list
);
5013 if (last_ticket
== ticket
) {
5016 last_ticket
= ticket
;
5017 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
5022 if (flush_state
> COMMIT_TRANS
) {
5024 if (commit_cycles
> 2) {
5025 wake_all_tickets(&space_info
->tickets
);
5026 space_info
->flush
= 0;
5028 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
5031 spin_unlock(&space_info
->lock
);
5032 } while (flush_state
<= COMMIT_TRANS
);
5035 void btrfs_init_async_reclaim_work(struct work_struct
*work
)
5037 INIT_WORK(work
, btrfs_async_reclaim_metadata_space
);
5040 static void priority_reclaim_metadata_space(struct btrfs_fs_info
*fs_info
,
5041 struct btrfs_space_info
*space_info
,
5042 struct reserve_ticket
*ticket
)
5045 int flush_state
= FLUSH_DELAYED_ITEMS_NR
;
5047 spin_lock(&space_info
->lock
);
5048 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
->fs_root
,
5051 spin_unlock(&space_info
->lock
);
5054 spin_unlock(&space_info
->lock
);
5057 flush_space(fs_info
->fs_root
, space_info
, to_reclaim
,
5058 to_reclaim
, flush_state
);
5060 spin_lock(&space_info
->lock
);
5061 if (ticket
->bytes
== 0) {
5062 spin_unlock(&space_info
->lock
);
5065 spin_unlock(&space_info
->lock
);
5068 * Priority flushers can't wait on delalloc without
5071 if (flush_state
== FLUSH_DELALLOC
||
5072 flush_state
== FLUSH_DELALLOC_WAIT
)
5073 flush_state
= ALLOC_CHUNK
;
5074 } while (flush_state
< COMMIT_TRANS
);
5077 static int wait_reserve_ticket(struct btrfs_fs_info
*fs_info
,
5078 struct btrfs_space_info
*space_info
,
5079 struct reserve_ticket
*ticket
, u64 orig_bytes
)
5085 spin_lock(&space_info
->lock
);
5086 while (ticket
->bytes
> 0 && ticket
->error
== 0) {
5087 ret
= prepare_to_wait_event(&ticket
->wait
, &wait
, TASK_KILLABLE
);
5092 spin_unlock(&space_info
->lock
);
5096 finish_wait(&ticket
->wait
, &wait
);
5097 spin_lock(&space_info
->lock
);
5100 ret
= ticket
->error
;
5101 if (!list_empty(&ticket
->list
))
5102 list_del_init(&ticket
->list
);
5103 if (ticket
->bytes
&& ticket
->bytes
< orig_bytes
) {
5104 u64 num_bytes
= orig_bytes
- ticket
->bytes
;
5105 space_info
->bytes_may_use
-= num_bytes
;
5106 trace_btrfs_space_reservation(fs_info
, "space_info",
5107 space_info
->flags
, num_bytes
, 0);
5109 spin_unlock(&space_info
->lock
);
5115 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5116 * @root - the root we're allocating for
5117 * @space_info - the space info we want to allocate from
5118 * @orig_bytes - the number of bytes we want
5119 * @flush - whether or not we can flush to make our reservation
5121 * This will reserve orig_bytes number of bytes from the space info associated
5122 * with the block_rsv. If there is not enough space it will make an attempt to
5123 * flush out space to make room. It will do this by flushing delalloc if
5124 * possible or committing the transaction. If flush is 0 then no attempts to
5125 * regain reservations will be made and this will fail if there is not enough
5128 static int __reserve_metadata_bytes(struct btrfs_root
*root
,
5129 struct btrfs_space_info
*space_info
,
5131 enum btrfs_reserve_flush_enum flush
)
5133 struct reserve_ticket ticket
;
5138 ASSERT(!current
->journal_info
|| flush
!= BTRFS_RESERVE_FLUSH_ALL
);
5140 spin_lock(&space_info
->lock
);
5142 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
5143 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
5144 space_info
->bytes_may_use
;
5147 * If we have enough space then hooray, make our reservation and carry
5148 * on. If not see if we can overcommit, and if we can, hooray carry on.
5149 * If not things get more complicated.
5151 if (used
+ orig_bytes
<= space_info
->total_bytes
) {
5152 space_info
->bytes_may_use
+= orig_bytes
;
5153 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
5154 space_info
->flags
, orig_bytes
,
5157 } else if (can_overcommit(root
, space_info
, orig_bytes
, flush
)) {
5158 space_info
->bytes_may_use
+= orig_bytes
;
5159 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
5160 space_info
->flags
, orig_bytes
,
5166 * If we couldn't make a reservation then setup our reservation ticket
5167 * and kick the async worker if it's not already running.
5169 * If we are a priority flusher then we just need to add our ticket to
5170 * the list and we will do our own flushing further down.
5172 if (ret
&& flush
!= BTRFS_RESERVE_NO_FLUSH
) {
5173 ticket
.bytes
= orig_bytes
;
5175 init_waitqueue_head(&ticket
.wait
);
5176 if (flush
== BTRFS_RESERVE_FLUSH_ALL
) {
5177 list_add_tail(&ticket
.list
, &space_info
->tickets
);
5178 if (!space_info
->flush
) {
5179 space_info
->flush
= 1;
5180 trace_btrfs_trigger_flush(root
->fs_info
,
5184 queue_work(system_unbound_wq
,
5185 &root
->fs_info
->async_reclaim_work
);
5188 list_add_tail(&ticket
.list
,
5189 &space_info
->priority_tickets
);
5191 } else if (!ret
&& space_info
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
5194 * We will do the space reservation dance during log replay,
5195 * which means we won't have fs_info->fs_root set, so don't do
5196 * the async reclaim as we will panic.
5198 if (!test_bit(BTRFS_FS_LOG_RECOVERING
, &root
->fs_info
->flags
) &&
5199 need_do_async_reclaim(space_info
, root
, used
) &&
5200 !work_busy(&root
->fs_info
->async_reclaim_work
)) {
5201 trace_btrfs_trigger_flush(root
->fs_info
,
5205 queue_work(system_unbound_wq
,
5206 &root
->fs_info
->async_reclaim_work
);
5209 spin_unlock(&space_info
->lock
);
5210 if (!ret
|| flush
== BTRFS_RESERVE_NO_FLUSH
)
5213 if (flush
== BTRFS_RESERVE_FLUSH_ALL
)
5214 return wait_reserve_ticket(root
->fs_info
, space_info
, &ticket
,
5218 priority_reclaim_metadata_space(root
->fs_info
, space_info
, &ticket
);
5219 spin_lock(&space_info
->lock
);
5221 if (ticket
.bytes
< orig_bytes
) {
5222 u64 num_bytes
= orig_bytes
- ticket
.bytes
;
5223 space_info
->bytes_may_use
-= num_bytes
;
5224 trace_btrfs_space_reservation(root
->fs_info
,
5225 "space_info", space_info
->flags
,
5229 list_del_init(&ticket
.list
);
5232 spin_unlock(&space_info
->lock
);
5233 ASSERT(list_empty(&ticket
.list
));
5238 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5239 * @root - the root we're allocating for
5240 * @block_rsv - the block_rsv we're allocating for
5241 * @orig_bytes - the number of bytes we want
5242 * @flush - whether or not we can flush to make our reservation
5244 * This will reserve orgi_bytes number of bytes from the space info associated
5245 * with the block_rsv. If there is not enough space it will make an attempt to
5246 * flush out space to make room. It will do this by flushing delalloc if
5247 * possible or committing the transaction. If flush is 0 then no attempts to
5248 * regain reservations will be made and this will fail if there is not enough
5251 static int reserve_metadata_bytes(struct btrfs_root
*root
,
5252 struct btrfs_block_rsv
*block_rsv
,
5254 enum btrfs_reserve_flush_enum flush
)
5258 ret
= __reserve_metadata_bytes(root
, block_rsv
->space_info
, orig_bytes
,
5260 if (ret
== -ENOSPC
&&
5261 unlikely(root
->orphan_cleanup_state
== ORPHAN_CLEANUP_STARTED
)) {
5262 struct btrfs_block_rsv
*global_rsv
=
5263 &root
->fs_info
->global_block_rsv
;
5265 if (block_rsv
!= global_rsv
&&
5266 !block_rsv_use_bytes(global_rsv
, orig_bytes
))
5270 trace_btrfs_space_reservation(root
->fs_info
,
5271 "space_info:enospc",
5272 block_rsv
->space_info
->flags
,
5277 static struct btrfs_block_rsv
*get_block_rsv(
5278 const struct btrfs_trans_handle
*trans
,
5279 const struct btrfs_root
*root
)
5281 struct btrfs_block_rsv
*block_rsv
= NULL
;
5283 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
5284 (root
== root
->fs_info
->csum_root
&& trans
->adding_csums
) ||
5285 (root
== root
->fs_info
->uuid_root
))
5286 block_rsv
= trans
->block_rsv
;
5289 block_rsv
= root
->block_rsv
;
5292 block_rsv
= &root
->fs_info
->empty_block_rsv
;
5297 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
5301 spin_lock(&block_rsv
->lock
);
5302 if (block_rsv
->reserved
>= num_bytes
) {
5303 block_rsv
->reserved
-= num_bytes
;
5304 if (block_rsv
->reserved
< block_rsv
->size
)
5305 block_rsv
->full
= 0;
5308 spin_unlock(&block_rsv
->lock
);
5312 static void block_rsv_add_bytes(struct btrfs_block_rsv
*block_rsv
,
5313 u64 num_bytes
, int update_size
)
5315 spin_lock(&block_rsv
->lock
);
5316 block_rsv
->reserved
+= num_bytes
;
5318 block_rsv
->size
+= num_bytes
;
5319 else if (block_rsv
->reserved
>= block_rsv
->size
)
5320 block_rsv
->full
= 1;
5321 spin_unlock(&block_rsv
->lock
);
5324 int btrfs_cond_migrate_bytes(struct btrfs_fs_info
*fs_info
,
5325 struct btrfs_block_rsv
*dest
, u64 num_bytes
,
5328 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5331 if (global_rsv
->space_info
!= dest
->space_info
)
5334 spin_lock(&global_rsv
->lock
);
5335 min_bytes
= div_factor(global_rsv
->size
, min_factor
);
5336 if (global_rsv
->reserved
< min_bytes
+ num_bytes
) {
5337 spin_unlock(&global_rsv
->lock
);
5340 global_rsv
->reserved
-= num_bytes
;
5341 if (global_rsv
->reserved
< global_rsv
->size
)
5342 global_rsv
->full
= 0;
5343 spin_unlock(&global_rsv
->lock
);
5345 block_rsv_add_bytes(dest
, num_bytes
, 1);
5350 * This is for space we already have accounted in space_info->bytes_may_use, so
5351 * basically when we're returning space from block_rsv's.
5353 static void space_info_add_old_bytes(struct btrfs_fs_info
*fs_info
,
5354 struct btrfs_space_info
*space_info
,
5357 struct reserve_ticket
*ticket
;
5358 struct list_head
*head
;
5360 enum btrfs_reserve_flush_enum flush
= BTRFS_RESERVE_NO_FLUSH
;
5361 bool check_overcommit
= false;
5363 spin_lock(&space_info
->lock
);
5364 head
= &space_info
->priority_tickets
;
5367 * If we are over our limit then we need to check and see if we can
5368 * overcommit, and if we can't then we just need to free up our space
5369 * and not satisfy any requests.
5371 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
5372 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
5373 space_info
->bytes_may_use
;
5374 if (used
- num_bytes
>= space_info
->total_bytes
)
5375 check_overcommit
= true;
5377 while (!list_empty(head
) && num_bytes
) {
5378 ticket
= list_first_entry(head
, struct reserve_ticket
,
5381 * We use 0 bytes because this space is already reserved, so
5382 * adding the ticket space would be a double count.
5384 if (check_overcommit
&&
5385 !can_overcommit(fs_info
->extent_root
, space_info
, 0,
5388 if (num_bytes
>= ticket
->bytes
) {
5389 list_del_init(&ticket
->list
);
5390 num_bytes
-= ticket
->bytes
;
5392 wake_up(&ticket
->wait
);
5394 ticket
->bytes
-= num_bytes
;
5399 if (num_bytes
&& head
== &space_info
->priority_tickets
) {
5400 head
= &space_info
->tickets
;
5401 flush
= BTRFS_RESERVE_FLUSH_ALL
;
5404 space_info
->bytes_may_use
-= num_bytes
;
5405 trace_btrfs_space_reservation(fs_info
, "space_info",
5406 space_info
->flags
, num_bytes
, 0);
5407 spin_unlock(&space_info
->lock
);
5411 * This is for newly allocated space that isn't accounted in
5412 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5413 * we use this helper.
5415 static void space_info_add_new_bytes(struct btrfs_fs_info
*fs_info
,
5416 struct btrfs_space_info
*space_info
,
5419 struct reserve_ticket
*ticket
;
5420 struct list_head
*head
= &space_info
->priority_tickets
;
5423 while (!list_empty(head
) && num_bytes
) {
5424 ticket
= list_first_entry(head
, struct reserve_ticket
,
5426 if (num_bytes
>= ticket
->bytes
) {
5427 trace_btrfs_space_reservation(fs_info
, "space_info",
5430 list_del_init(&ticket
->list
);
5431 num_bytes
-= ticket
->bytes
;
5432 space_info
->bytes_may_use
+= ticket
->bytes
;
5434 wake_up(&ticket
->wait
);
5436 trace_btrfs_space_reservation(fs_info
, "space_info",
5439 space_info
->bytes_may_use
+= num_bytes
;
5440 ticket
->bytes
-= num_bytes
;
5445 if (num_bytes
&& head
== &space_info
->priority_tickets
) {
5446 head
= &space_info
->tickets
;
5451 static void block_rsv_release_bytes(struct btrfs_fs_info
*fs_info
,
5452 struct btrfs_block_rsv
*block_rsv
,
5453 struct btrfs_block_rsv
*dest
, u64 num_bytes
)
5455 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
5457 spin_lock(&block_rsv
->lock
);
5458 if (num_bytes
== (u64
)-1)
5459 num_bytes
= block_rsv
->size
;
5460 block_rsv
->size
-= num_bytes
;
5461 if (block_rsv
->reserved
>= block_rsv
->size
) {
5462 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
5463 block_rsv
->reserved
= block_rsv
->size
;
5464 block_rsv
->full
= 1;
5468 spin_unlock(&block_rsv
->lock
);
5470 if (num_bytes
> 0) {
5472 spin_lock(&dest
->lock
);
5476 bytes_to_add
= dest
->size
- dest
->reserved
;
5477 bytes_to_add
= min(num_bytes
, bytes_to_add
);
5478 dest
->reserved
+= bytes_to_add
;
5479 if (dest
->reserved
>= dest
->size
)
5481 num_bytes
-= bytes_to_add
;
5483 spin_unlock(&dest
->lock
);
5486 space_info_add_old_bytes(fs_info
, space_info
,
5491 int btrfs_block_rsv_migrate(struct btrfs_block_rsv
*src
,
5492 struct btrfs_block_rsv
*dst
, u64 num_bytes
,
5497 ret
= block_rsv_use_bytes(src
, num_bytes
);
5501 block_rsv_add_bytes(dst
, num_bytes
, update_size
);
5505 void btrfs_init_block_rsv(struct btrfs_block_rsv
*rsv
, unsigned short type
)
5507 memset(rsv
, 0, sizeof(*rsv
));
5508 spin_lock_init(&rsv
->lock
);
5512 struct btrfs_block_rsv
*btrfs_alloc_block_rsv(struct btrfs_root
*root
,
5513 unsigned short type
)
5515 struct btrfs_block_rsv
*block_rsv
;
5516 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5518 block_rsv
= kmalloc(sizeof(*block_rsv
), GFP_NOFS
);
5522 btrfs_init_block_rsv(block_rsv
, type
);
5523 block_rsv
->space_info
= __find_space_info(fs_info
,
5524 BTRFS_BLOCK_GROUP_METADATA
);
5528 void btrfs_free_block_rsv(struct btrfs_root
*root
,
5529 struct btrfs_block_rsv
*rsv
)
5533 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
5537 void __btrfs_free_block_rsv(struct btrfs_block_rsv
*rsv
)
5542 int btrfs_block_rsv_add(struct btrfs_root
*root
,
5543 struct btrfs_block_rsv
*block_rsv
, u64 num_bytes
,
5544 enum btrfs_reserve_flush_enum flush
)
5551 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
5553 block_rsv_add_bytes(block_rsv
, num_bytes
, 1);
5560 int btrfs_block_rsv_check(struct btrfs_root
*root
,
5561 struct btrfs_block_rsv
*block_rsv
, int min_factor
)
5569 spin_lock(&block_rsv
->lock
);
5570 num_bytes
= div_factor(block_rsv
->size
, min_factor
);
5571 if (block_rsv
->reserved
>= num_bytes
)
5573 spin_unlock(&block_rsv
->lock
);
5578 int btrfs_block_rsv_refill(struct btrfs_root
*root
,
5579 struct btrfs_block_rsv
*block_rsv
, u64 min_reserved
,
5580 enum btrfs_reserve_flush_enum flush
)
5588 spin_lock(&block_rsv
->lock
);
5589 num_bytes
= min_reserved
;
5590 if (block_rsv
->reserved
>= num_bytes
)
5593 num_bytes
-= block_rsv
->reserved
;
5594 spin_unlock(&block_rsv
->lock
);
5599 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
5601 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
5608 void btrfs_block_rsv_release(struct btrfs_root
*root
,
5609 struct btrfs_block_rsv
*block_rsv
,
5612 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
5613 if (global_rsv
== block_rsv
||
5614 block_rsv
->space_info
!= global_rsv
->space_info
)
5616 block_rsv_release_bytes(root
->fs_info
, block_rsv
, global_rsv
,
5620 static void update_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5622 struct btrfs_block_rsv
*block_rsv
= &fs_info
->global_block_rsv
;
5623 struct btrfs_space_info
*sinfo
= block_rsv
->space_info
;
5627 * The global block rsv is based on the size of the extent tree, the
5628 * checksum tree and the root tree. If the fs is empty we want to set
5629 * it to a minimal amount for safety.
5631 num_bytes
= btrfs_root_used(&fs_info
->extent_root
->root_item
) +
5632 btrfs_root_used(&fs_info
->csum_root
->root_item
) +
5633 btrfs_root_used(&fs_info
->tree_root
->root_item
);
5634 num_bytes
= max_t(u64
, num_bytes
, SZ_16M
);
5636 spin_lock(&sinfo
->lock
);
5637 spin_lock(&block_rsv
->lock
);
5639 block_rsv
->size
= min_t(u64
, num_bytes
, SZ_512M
);
5641 if (block_rsv
->reserved
< block_rsv
->size
) {
5642 num_bytes
= sinfo
->bytes_used
+ sinfo
->bytes_pinned
+
5643 sinfo
->bytes_reserved
+ sinfo
->bytes_readonly
+
5644 sinfo
->bytes_may_use
;
5645 if (sinfo
->total_bytes
> num_bytes
) {
5646 num_bytes
= sinfo
->total_bytes
- num_bytes
;
5647 num_bytes
= min(num_bytes
,
5648 block_rsv
->size
- block_rsv
->reserved
);
5649 block_rsv
->reserved
+= num_bytes
;
5650 sinfo
->bytes_may_use
+= num_bytes
;
5651 trace_btrfs_space_reservation(fs_info
, "space_info",
5652 sinfo
->flags
, num_bytes
,
5655 } else if (block_rsv
->reserved
> block_rsv
->size
) {
5656 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
5657 sinfo
->bytes_may_use
-= num_bytes
;
5658 trace_btrfs_space_reservation(fs_info
, "space_info",
5659 sinfo
->flags
, num_bytes
, 0);
5660 block_rsv
->reserved
= block_rsv
->size
;
5663 if (block_rsv
->reserved
== block_rsv
->size
)
5664 block_rsv
->full
= 1;
5666 block_rsv
->full
= 0;
5668 spin_unlock(&block_rsv
->lock
);
5669 spin_unlock(&sinfo
->lock
);
5672 static void init_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5674 struct btrfs_space_info
*space_info
;
5676 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
5677 fs_info
->chunk_block_rsv
.space_info
= space_info
;
5679 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
5680 fs_info
->global_block_rsv
.space_info
= space_info
;
5681 fs_info
->delalloc_block_rsv
.space_info
= space_info
;
5682 fs_info
->trans_block_rsv
.space_info
= space_info
;
5683 fs_info
->empty_block_rsv
.space_info
= space_info
;
5684 fs_info
->delayed_block_rsv
.space_info
= space_info
;
5686 fs_info
->extent_root
->block_rsv
= &fs_info
->global_block_rsv
;
5687 fs_info
->csum_root
->block_rsv
= &fs_info
->global_block_rsv
;
5688 fs_info
->dev_root
->block_rsv
= &fs_info
->global_block_rsv
;
5689 fs_info
->tree_root
->block_rsv
= &fs_info
->global_block_rsv
;
5690 if (fs_info
->quota_root
)
5691 fs_info
->quota_root
->block_rsv
= &fs_info
->global_block_rsv
;
5692 fs_info
->chunk_root
->block_rsv
= &fs_info
->chunk_block_rsv
;
5694 update_global_block_rsv(fs_info
);
5697 static void release_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5699 block_rsv_release_bytes(fs_info
, &fs_info
->global_block_rsv
, NULL
,
5701 WARN_ON(fs_info
->delalloc_block_rsv
.size
> 0);
5702 WARN_ON(fs_info
->delalloc_block_rsv
.reserved
> 0);
5703 WARN_ON(fs_info
->trans_block_rsv
.size
> 0);
5704 WARN_ON(fs_info
->trans_block_rsv
.reserved
> 0);
5705 WARN_ON(fs_info
->chunk_block_rsv
.size
> 0);
5706 WARN_ON(fs_info
->chunk_block_rsv
.reserved
> 0);
5707 WARN_ON(fs_info
->delayed_block_rsv
.size
> 0);
5708 WARN_ON(fs_info
->delayed_block_rsv
.reserved
> 0);
5711 void btrfs_trans_release_metadata(struct btrfs_trans_handle
*trans
,
5712 struct btrfs_root
*root
)
5714 if (!trans
->block_rsv
)
5717 if (!trans
->bytes_reserved
)
5720 trace_btrfs_space_reservation(root
->fs_info
, "transaction",
5721 trans
->transid
, trans
->bytes_reserved
, 0);
5722 btrfs_block_rsv_release(root
, trans
->block_rsv
, trans
->bytes_reserved
);
5723 trans
->bytes_reserved
= 0;
5727 * To be called after all the new block groups attached to the transaction
5728 * handle have been created (btrfs_create_pending_block_groups()).
5730 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle
*trans
)
5732 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
5734 if (!trans
->chunk_bytes_reserved
)
5737 WARN_ON_ONCE(!list_empty(&trans
->new_bgs
));
5739 block_rsv_release_bytes(fs_info
, &fs_info
->chunk_block_rsv
, NULL
,
5740 trans
->chunk_bytes_reserved
);
5741 trans
->chunk_bytes_reserved
= 0;
5744 /* Can only return 0 or -ENOSPC */
5745 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle
*trans
,
5746 struct inode
*inode
)
5748 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5750 * We always use trans->block_rsv here as we will have reserved space
5751 * for our orphan when starting the transaction, using get_block_rsv()
5752 * here will sometimes make us choose the wrong block rsv as we could be
5753 * doing a reloc inode for a non refcounted root.
5755 struct btrfs_block_rsv
*src_rsv
= trans
->block_rsv
;
5756 struct btrfs_block_rsv
*dst_rsv
= root
->orphan_block_rsv
;
5759 * We need to hold space in order to delete our orphan item once we've
5760 * added it, so this takes the reservation so we can release it later
5761 * when we are truly done with the orphan item.
5763 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
5764 trace_btrfs_space_reservation(root
->fs_info
, "orphan",
5765 btrfs_ino(inode
), num_bytes
, 1);
5766 return btrfs_block_rsv_migrate(src_rsv
, dst_rsv
, num_bytes
, 1);
5769 void btrfs_orphan_release_metadata(struct inode
*inode
)
5771 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5772 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
5773 trace_btrfs_space_reservation(root
->fs_info
, "orphan",
5774 btrfs_ino(inode
), num_bytes
, 0);
5775 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
, num_bytes
);
5779 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5780 * root: the root of the parent directory
5781 * rsv: block reservation
5782 * items: the number of items that we need do reservation
5783 * qgroup_reserved: used to return the reserved size in qgroup
5785 * This function is used to reserve the space for snapshot/subvolume
5786 * creation and deletion. Those operations are different with the
5787 * common file/directory operations, they change two fs/file trees
5788 * and root tree, the number of items that the qgroup reserves is
5789 * different with the free space reservation. So we can not use
5790 * the space reservation mechanism in start_transaction().
5792 int btrfs_subvolume_reserve_metadata(struct btrfs_root
*root
,
5793 struct btrfs_block_rsv
*rsv
,
5795 u64
*qgroup_reserved
,
5796 bool use_global_rsv
)
5800 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
5802 if (test_bit(BTRFS_FS_QUOTA_ENABLED
, &root
->fs_info
->flags
)) {
5803 /* One for parent inode, two for dir entries */
5804 num_bytes
= 3 * root
->nodesize
;
5805 ret
= btrfs_qgroup_reserve_meta(root
, num_bytes
);
5812 *qgroup_reserved
= num_bytes
;
5814 num_bytes
= btrfs_calc_trans_metadata_size(root
, items
);
5815 rsv
->space_info
= __find_space_info(root
->fs_info
,
5816 BTRFS_BLOCK_GROUP_METADATA
);
5817 ret
= btrfs_block_rsv_add(root
, rsv
, num_bytes
,
5818 BTRFS_RESERVE_FLUSH_ALL
);
5820 if (ret
== -ENOSPC
&& use_global_rsv
)
5821 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
, num_bytes
, 1);
5823 if (ret
&& *qgroup_reserved
)
5824 btrfs_qgroup_free_meta(root
, *qgroup_reserved
);
5829 void btrfs_subvolume_release_metadata(struct btrfs_root
*root
,
5830 struct btrfs_block_rsv
*rsv
,
5831 u64 qgroup_reserved
)
5833 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
5837 * drop_outstanding_extent - drop an outstanding extent
5838 * @inode: the inode we're dropping the extent for
5839 * @num_bytes: the number of bytes we're releasing.
5841 * This is called when we are freeing up an outstanding extent, either called
5842 * after an error or after an extent is written. This will return the number of
5843 * reserved extents that need to be freed. This must be called with
5844 * BTRFS_I(inode)->lock held.
5846 static unsigned drop_outstanding_extent(struct inode
*inode
, u64 num_bytes
)
5848 unsigned drop_inode_space
= 0;
5849 unsigned dropped_extents
= 0;
5850 unsigned num_extents
= 0;
5852 num_extents
= (unsigned)div64_u64(num_bytes
+
5853 BTRFS_MAX_EXTENT_SIZE
- 1,
5854 BTRFS_MAX_EXTENT_SIZE
);
5855 ASSERT(num_extents
);
5856 ASSERT(BTRFS_I(inode
)->outstanding_extents
>= num_extents
);
5857 BTRFS_I(inode
)->outstanding_extents
-= num_extents
;
5859 if (BTRFS_I(inode
)->outstanding_extents
== 0 &&
5860 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5861 &BTRFS_I(inode
)->runtime_flags
))
5862 drop_inode_space
= 1;
5865 * If we have more or the same amount of outstanding extents than we have
5866 * reserved then we need to leave the reserved extents count alone.
5868 if (BTRFS_I(inode
)->outstanding_extents
>=
5869 BTRFS_I(inode
)->reserved_extents
)
5870 return drop_inode_space
;
5872 dropped_extents
= BTRFS_I(inode
)->reserved_extents
-
5873 BTRFS_I(inode
)->outstanding_extents
;
5874 BTRFS_I(inode
)->reserved_extents
-= dropped_extents
;
5875 return dropped_extents
+ drop_inode_space
;
5879 * calc_csum_metadata_size - return the amount of metadata space that must be
5880 * reserved/freed for the given bytes.
5881 * @inode: the inode we're manipulating
5882 * @num_bytes: the number of bytes in question
5883 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5885 * This adjusts the number of csum_bytes in the inode and then returns the
5886 * correct amount of metadata that must either be reserved or freed. We
5887 * calculate how many checksums we can fit into one leaf and then divide the
5888 * number of bytes that will need to be checksumed by this value to figure out
5889 * how many checksums will be required. If we are adding bytes then the number
5890 * may go up and we will return the number of additional bytes that must be
5891 * reserved. If it is going down we will return the number of bytes that must
5894 * This must be called with BTRFS_I(inode)->lock held.
5896 static u64
calc_csum_metadata_size(struct inode
*inode
, u64 num_bytes
,
5899 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5900 u64 old_csums
, num_csums
;
5902 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
&&
5903 BTRFS_I(inode
)->csum_bytes
== 0)
5906 old_csums
= btrfs_csum_bytes_to_leaves(root
, BTRFS_I(inode
)->csum_bytes
);
5908 BTRFS_I(inode
)->csum_bytes
+= num_bytes
;
5910 BTRFS_I(inode
)->csum_bytes
-= num_bytes
;
5911 num_csums
= btrfs_csum_bytes_to_leaves(root
, BTRFS_I(inode
)->csum_bytes
);
5913 /* No change, no need to reserve more */
5914 if (old_csums
== num_csums
)
5918 return btrfs_calc_trans_metadata_size(root
,
5919 num_csums
- old_csums
);
5921 return btrfs_calc_trans_metadata_size(root
, old_csums
- num_csums
);
5924 int btrfs_delalloc_reserve_metadata(struct inode
*inode
, u64 num_bytes
)
5926 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5927 struct btrfs_block_rsv
*block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
5930 unsigned nr_extents
= 0;
5931 enum btrfs_reserve_flush_enum flush
= BTRFS_RESERVE_FLUSH_ALL
;
5933 bool delalloc_lock
= true;
5936 bool release_extra
= false;
5938 /* If we are a free space inode we need to not flush since we will be in
5939 * the middle of a transaction commit. We also don't need the delalloc
5940 * mutex since we won't race with anybody. We need this mostly to make
5941 * lockdep shut its filthy mouth.
5943 * If we have a transaction open (can happen if we call truncate_block
5944 * from truncate), then we need FLUSH_LIMIT so we don't deadlock.
5946 if (btrfs_is_free_space_inode(inode
)) {
5947 flush
= BTRFS_RESERVE_NO_FLUSH
;
5948 delalloc_lock
= false;
5949 } else if (current
->journal_info
) {
5950 flush
= BTRFS_RESERVE_FLUSH_LIMIT
;
5953 if (flush
!= BTRFS_RESERVE_NO_FLUSH
&&
5954 btrfs_transaction_in_commit(root
->fs_info
))
5955 schedule_timeout(1);
5958 mutex_lock(&BTRFS_I(inode
)->delalloc_mutex
);
5960 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
5962 spin_lock(&BTRFS_I(inode
)->lock
);
5963 nr_extents
= (unsigned)div64_u64(num_bytes
+
5964 BTRFS_MAX_EXTENT_SIZE
- 1,
5965 BTRFS_MAX_EXTENT_SIZE
);
5966 BTRFS_I(inode
)->outstanding_extents
+= nr_extents
;
5969 if (BTRFS_I(inode
)->outstanding_extents
>
5970 BTRFS_I(inode
)->reserved_extents
)
5971 nr_extents
+= BTRFS_I(inode
)->outstanding_extents
-
5972 BTRFS_I(inode
)->reserved_extents
;
5974 /* We always want to reserve a slot for updating the inode. */
5975 to_reserve
= btrfs_calc_trans_metadata_size(root
, nr_extents
+ 1);
5976 to_reserve
+= calc_csum_metadata_size(inode
, num_bytes
, 1);
5977 csum_bytes
= BTRFS_I(inode
)->csum_bytes
;
5978 spin_unlock(&BTRFS_I(inode
)->lock
);
5980 if (test_bit(BTRFS_FS_QUOTA_ENABLED
, &root
->fs_info
->flags
)) {
5981 ret
= btrfs_qgroup_reserve_meta(root
,
5982 nr_extents
* root
->nodesize
);
5987 ret
= btrfs_block_rsv_add(root
, block_rsv
, to_reserve
, flush
);
5988 if (unlikely(ret
)) {
5989 btrfs_qgroup_free_meta(root
, nr_extents
* root
->nodesize
);
5993 spin_lock(&BTRFS_I(inode
)->lock
);
5994 if (test_and_set_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5995 &BTRFS_I(inode
)->runtime_flags
)) {
5996 to_reserve
-= btrfs_calc_trans_metadata_size(root
, 1);
5997 release_extra
= true;
5999 BTRFS_I(inode
)->reserved_extents
+= nr_extents
;
6000 spin_unlock(&BTRFS_I(inode
)->lock
);
6003 mutex_unlock(&BTRFS_I(inode
)->delalloc_mutex
);
6006 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
6007 btrfs_ino(inode
), to_reserve
, 1);
6009 btrfs_block_rsv_release(root
, block_rsv
,
6010 btrfs_calc_trans_metadata_size(root
,
6015 spin_lock(&BTRFS_I(inode
)->lock
);
6016 dropped
= drop_outstanding_extent(inode
, num_bytes
);
6018 * If the inodes csum_bytes is the same as the original
6019 * csum_bytes then we know we haven't raced with any free()ers
6020 * so we can just reduce our inodes csum bytes and carry on.
6022 if (BTRFS_I(inode
)->csum_bytes
== csum_bytes
) {
6023 calc_csum_metadata_size(inode
, num_bytes
, 0);
6025 u64 orig_csum_bytes
= BTRFS_I(inode
)->csum_bytes
;
6029 * This is tricky, but first we need to figure out how much we
6030 * freed from any free-ers that occurred during this
6031 * reservation, so we reset ->csum_bytes to the csum_bytes
6032 * before we dropped our lock, and then call the free for the
6033 * number of bytes that were freed while we were trying our
6036 bytes
= csum_bytes
- BTRFS_I(inode
)->csum_bytes
;
6037 BTRFS_I(inode
)->csum_bytes
= csum_bytes
;
6038 to_free
= calc_csum_metadata_size(inode
, bytes
, 0);
6042 * Now we need to see how much we would have freed had we not
6043 * been making this reservation and our ->csum_bytes were not
6044 * artificially inflated.
6046 BTRFS_I(inode
)->csum_bytes
= csum_bytes
- num_bytes
;
6047 bytes
= csum_bytes
- orig_csum_bytes
;
6048 bytes
= calc_csum_metadata_size(inode
, bytes
, 0);
6051 * Now reset ->csum_bytes to what it should be. If bytes is
6052 * more than to_free then we would have freed more space had we
6053 * not had an artificially high ->csum_bytes, so we need to free
6054 * the remainder. If bytes is the same or less then we don't
6055 * need to do anything, the other free-ers did the correct
6058 BTRFS_I(inode
)->csum_bytes
= orig_csum_bytes
- num_bytes
;
6059 if (bytes
> to_free
)
6060 to_free
= bytes
- to_free
;
6064 spin_unlock(&BTRFS_I(inode
)->lock
);
6066 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
6069 btrfs_block_rsv_release(root
, block_rsv
, to_free
);
6070 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
6071 btrfs_ino(inode
), to_free
, 0);
6074 mutex_unlock(&BTRFS_I(inode
)->delalloc_mutex
);
6079 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
6080 * @inode: the inode to release the reservation for
6081 * @num_bytes: the number of bytes we're releasing
6083 * This will release the metadata reservation for an inode. This can be called
6084 * once we complete IO for a given set of bytes to release their metadata
6087 void btrfs_delalloc_release_metadata(struct inode
*inode
, u64 num_bytes
)
6089 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6093 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
6094 spin_lock(&BTRFS_I(inode
)->lock
);
6095 dropped
= drop_outstanding_extent(inode
, num_bytes
);
6098 to_free
= calc_csum_metadata_size(inode
, num_bytes
, 0);
6099 spin_unlock(&BTRFS_I(inode
)->lock
);
6101 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
6103 if (btrfs_is_testing(root
->fs_info
))
6106 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
6107 btrfs_ino(inode
), to_free
, 0);
6109 btrfs_block_rsv_release(root
, &root
->fs_info
->delalloc_block_rsv
,
6114 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6116 * @inode: inode we're writing to
6117 * @start: start range we are writing to
6118 * @len: how long the range we are writing to
6120 * TODO: This function will finally replace old btrfs_delalloc_reserve_space()
6122 * This will do the following things
6124 * o reserve space in data space info for num bytes
6125 * and reserve precious corresponding qgroup space
6126 * (Done in check_data_free_space)
6128 * o reserve space for metadata space, based on the number of outstanding
6129 * extents and how much csums will be needed
6130 * also reserve metadata space in a per root over-reserve method.
6131 * o add to the inodes->delalloc_bytes
6132 * o add it to the fs_info's delalloc inodes list.
6133 * (Above 3 all done in delalloc_reserve_metadata)
6135 * Return 0 for success
6136 * Return <0 for error(-ENOSPC or -EQUOT)
6138 int btrfs_delalloc_reserve_space(struct inode
*inode
, u64 start
, u64 len
)
6142 ret
= btrfs_check_data_free_space(inode
, start
, len
);
6145 ret
= btrfs_delalloc_reserve_metadata(inode
, len
);
6147 btrfs_free_reserved_data_space(inode
, start
, len
);
6152 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6153 * @inode: inode we're releasing space for
6154 * @start: start position of the space already reserved
6155 * @len: the len of the space already reserved
6157 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
6158 * called in the case that we don't need the metadata AND data reservations
6159 * anymore. So if there is an error or we insert an inline extent.
6161 * This function will release the metadata space that was not used and will
6162 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
6163 * list if there are no delalloc bytes left.
6164 * Also it will handle the qgroup reserved space.
6166 void btrfs_delalloc_release_space(struct inode
*inode
, u64 start
, u64 len
)
6168 btrfs_delalloc_release_metadata(inode
, len
);
6169 btrfs_free_reserved_data_space(inode
, start
, len
);
6172 static int update_block_group(struct btrfs_trans_handle
*trans
,
6173 struct btrfs_root
*root
, u64 bytenr
,
6174 u64 num_bytes
, int alloc
)
6176 struct btrfs_block_group_cache
*cache
= NULL
;
6177 struct btrfs_fs_info
*info
= root
->fs_info
;
6178 u64 total
= num_bytes
;
6183 /* block accounting for super block */
6184 spin_lock(&info
->delalloc_root_lock
);
6185 old_val
= btrfs_super_bytes_used(info
->super_copy
);
6187 old_val
+= num_bytes
;
6189 old_val
-= num_bytes
;
6190 btrfs_set_super_bytes_used(info
->super_copy
, old_val
);
6191 spin_unlock(&info
->delalloc_root_lock
);
6194 cache
= btrfs_lookup_block_group(info
, bytenr
);
6197 if (cache
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
6198 BTRFS_BLOCK_GROUP_RAID1
|
6199 BTRFS_BLOCK_GROUP_RAID10
))
6204 * If this block group has free space cache written out, we
6205 * need to make sure to load it if we are removing space. This
6206 * is because we need the unpinning stage to actually add the
6207 * space back to the block group, otherwise we will leak space.
6209 if (!alloc
&& cache
->cached
== BTRFS_CACHE_NO
)
6210 cache_block_group(cache
, 1);
6212 byte_in_group
= bytenr
- cache
->key
.objectid
;
6213 WARN_ON(byte_in_group
> cache
->key
.offset
);
6215 spin_lock(&cache
->space_info
->lock
);
6216 spin_lock(&cache
->lock
);
6218 if (btrfs_test_opt(root
->fs_info
, SPACE_CACHE
) &&
6219 cache
->disk_cache_state
< BTRFS_DC_CLEAR
)
6220 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
6222 old_val
= btrfs_block_group_used(&cache
->item
);
6223 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
6225 old_val
+= num_bytes
;
6226 btrfs_set_block_group_used(&cache
->item
, old_val
);
6227 cache
->reserved
-= num_bytes
;
6228 cache
->space_info
->bytes_reserved
-= num_bytes
;
6229 cache
->space_info
->bytes_used
+= num_bytes
;
6230 cache
->space_info
->disk_used
+= num_bytes
* factor
;
6231 spin_unlock(&cache
->lock
);
6232 spin_unlock(&cache
->space_info
->lock
);
6234 old_val
-= num_bytes
;
6235 btrfs_set_block_group_used(&cache
->item
, old_val
);
6236 cache
->pinned
+= num_bytes
;
6237 cache
->space_info
->bytes_pinned
+= num_bytes
;
6238 cache
->space_info
->bytes_used
-= num_bytes
;
6239 cache
->space_info
->disk_used
-= num_bytes
* factor
;
6240 spin_unlock(&cache
->lock
);
6241 spin_unlock(&cache
->space_info
->lock
);
6243 trace_btrfs_space_reservation(root
->fs_info
, "pinned",
6244 cache
->space_info
->flags
,
6246 set_extent_dirty(info
->pinned_extents
,
6247 bytenr
, bytenr
+ num_bytes
- 1,
6248 GFP_NOFS
| __GFP_NOFAIL
);
6251 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
6252 if (list_empty(&cache
->dirty_list
)) {
6253 list_add_tail(&cache
->dirty_list
,
6254 &trans
->transaction
->dirty_bgs
);
6255 trans
->transaction
->num_dirty_bgs
++;
6256 btrfs_get_block_group(cache
);
6258 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
6261 * No longer have used bytes in this block group, queue it for
6262 * deletion. We do this after adding the block group to the
6263 * dirty list to avoid races between cleaner kthread and space
6266 if (!alloc
&& old_val
== 0) {
6267 spin_lock(&info
->unused_bgs_lock
);
6268 if (list_empty(&cache
->bg_list
)) {
6269 btrfs_get_block_group(cache
);
6270 list_add_tail(&cache
->bg_list
,
6273 spin_unlock(&info
->unused_bgs_lock
);
6276 btrfs_put_block_group(cache
);
6278 bytenr
+= num_bytes
;
6283 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
6285 struct btrfs_block_group_cache
*cache
;
6288 spin_lock(&root
->fs_info
->block_group_cache_lock
);
6289 bytenr
= root
->fs_info
->first_logical_byte
;
6290 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
6292 if (bytenr
< (u64
)-1)
6295 cache
= btrfs_lookup_first_block_group(root
->fs_info
, search_start
);
6299 bytenr
= cache
->key
.objectid
;
6300 btrfs_put_block_group(cache
);
6305 static int pin_down_extent(struct btrfs_root
*root
,
6306 struct btrfs_block_group_cache
*cache
,
6307 u64 bytenr
, u64 num_bytes
, int reserved
)
6309 spin_lock(&cache
->space_info
->lock
);
6310 spin_lock(&cache
->lock
);
6311 cache
->pinned
+= num_bytes
;
6312 cache
->space_info
->bytes_pinned
+= num_bytes
;
6314 cache
->reserved
-= num_bytes
;
6315 cache
->space_info
->bytes_reserved
-= num_bytes
;
6317 spin_unlock(&cache
->lock
);
6318 spin_unlock(&cache
->space_info
->lock
);
6320 trace_btrfs_space_reservation(root
->fs_info
, "pinned",
6321 cache
->space_info
->flags
, num_bytes
, 1);
6322 set_extent_dirty(root
->fs_info
->pinned_extents
, bytenr
,
6323 bytenr
+ num_bytes
- 1, GFP_NOFS
| __GFP_NOFAIL
);
6328 * this function must be called within transaction
6330 int btrfs_pin_extent(struct btrfs_root
*root
,
6331 u64 bytenr
, u64 num_bytes
, int reserved
)
6333 struct btrfs_block_group_cache
*cache
;
6335 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
6336 BUG_ON(!cache
); /* Logic error */
6338 pin_down_extent(root
, cache
, bytenr
, num_bytes
, reserved
);
6340 btrfs_put_block_group(cache
);
6345 * this function must be called within transaction
6347 int btrfs_pin_extent_for_log_replay(struct btrfs_root
*root
,
6348 u64 bytenr
, u64 num_bytes
)
6350 struct btrfs_block_group_cache
*cache
;
6353 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
6358 * pull in the free space cache (if any) so that our pin
6359 * removes the free space from the cache. We have load_only set
6360 * to one because the slow code to read in the free extents does check
6361 * the pinned extents.
6363 cache_block_group(cache
, 1);
6365 pin_down_extent(root
, cache
, bytenr
, num_bytes
, 0);
6367 /* remove us from the free space cache (if we're there at all) */
6368 ret
= btrfs_remove_free_space(cache
, bytenr
, num_bytes
);
6369 btrfs_put_block_group(cache
);
6373 static int __exclude_logged_extent(struct btrfs_root
*root
, u64 start
, u64 num_bytes
)
6376 struct btrfs_block_group_cache
*block_group
;
6377 struct btrfs_caching_control
*caching_ctl
;
6379 block_group
= btrfs_lookup_block_group(root
->fs_info
, start
);
6383 cache_block_group(block_group
, 0);
6384 caching_ctl
= get_caching_control(block_group
);
6388 BUG_ON(!block_group_cache_done(block_group
));
6389 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
6391 mutex_lock(&caching_ctl
->mutex
);
6393 if (start
>= caching_ctl
->progress
) {
6394 ret
= add_excluded_extent(root
, start
, num_bytes
);
6395 } else if (start
+ num_bytes
<= caching_ctl
->progress
) {
6396 ret
= btrfs_remove_free_space(block_group
,
6399 num_bytes
= caching_ctl
->progress
- start
;
6400 ret
= btrfs_remove_free_space(block_group
,
6405 num_bytes
= (start
+ num_bytes
) -
6406 caching_ctl
->progress
;
6407 start
= caching_ctl
->progress
;
6408 ret
= add_excluded_extent(root
, start
, num_bytes
);
6411 mutex_unlock(&caching_ctl
->mutex
);
6412 put_caching_control(caching_ctl
);
6414 btrfs_put_block_group(block_group
);
6418 int btrfs_exclude_logged_extents(struct btrfs_root
*log
,
6419 struct extent_buffer
*eb
)
6421 struct btrfs_file_extent_item
*item
;
6422 struct btrfs_key key
;
6426 if (!btrfs_fs_incompat(log
->fs_info
, MIXED_GROUPS
))
6429 for (i
= 0; i
< btrfs_header_nritems(eb
); i
++) {
6430 btrfs_item_key_to_cpu(eb
, &key
, i
);
6431 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
6433 item
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
6434 found_type
= btrfs_file_extent_type(eb
, item
);
6435 if (found_type
== BTRFS_FILE_EXTENT_INLINE
)
6437 if (btrfs_file_extent_disk_bytenr(eb
, item
) == 0)
6439 key
.objectid
= btrfs_file_extent_disk_bytenr(eb
, item
);
6440 key
.offset
= btrfs_file_extent_disk_num_bytes(eb
, item
);
6441 __exclude_logged_extent(log
, key
.objectid
, key
.offset
);
6448 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache
*bg
)
6450 atomic_inc(&bg
->reservations
);
6453 void btrfs_dec_block_group_reservations(struct btrfs_fs_info
*fs_info
,
6456 struct btrfs_block_group_cache
*bg
;
6458 bg
= btrfs_lookup_block_group(fs_info
, start
);
6460 if (atomic_dec_and_test(&bg
->reservations
))
6461 wake_up_atomic_t(&bg
->reservations
);
6462 btrfs_put_block_group(bg
);
6465 static int btrfs_wait_bg_reservations_atomic_t(atomic_t
*a
)
6471 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache
*bg
)
6473 struct btrfs_space_info
*space_info
= bg
->space_info
;
6477 if (!(bg
->flags
& BTRFS_BLOCK_GROUP_DATA
))
6481 * Our block group is read only but before we set it to read only,
6482 * some task might have had allocated an extent from it already, but it
6483 * has not yet created a respective ordered extent (and added it to a
6484 * root's list of ordered extents).
6485 * Therefore wait for any task currently allocating extents, since the
6486 * block group's reservations counter is incremented while a read lock
6487 * on the groups' semaphore is held and decremented after releasing
6488 * the read access on that semaphore and creating the ordered extent.
6490 down_write(&space_info
->groups_sem
);
6491 up_write(&space_info
->groups_sem
);
6493 wait_on_atomic_t(&bg
->reservations
,
6494 btrfs_wait_bg_reservations_atomic_t
,
6495 TASK_UNINTERRUPTIBLE
);
6499 * btrfs_add_reserved_bytes - update the block_group and space info counters
6500 * @cache: The cache we are manipulating
6501 * @ram_bytes: The number of bytes of file content, and will be same to
6502 * @num_bytes except for the compress path.
6503 * @num_bytes: The number of bytes in question
6504 * @delalloc: The blocks are allocated for the delalloc write
6506 * This is called by the allocator when it reserves space. Metadata
6507 * reservations should be called with RESERVE_ALLOC so we do the proper
6508 * ENOSPC accounting. For data we handle the reservation through clearing the
6509 * delalloc bits in the io_tree. We have to do this since we could end up
6510 * allocating less disk space for the amount of data we have reserved in the
6511 * case of compression.
6513 * If this is a reservation and the block group has become read only we cannot
6514 * make the reservation and return -EAGAIN, otherwise this function always
6517 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache
*cache
,
6518 u64 ram_bytes
, u64 num_bytes
, int delalloc
)
6520 struct btrfs_space_info
*space_info
= cache
->space_info
;
6523 spin_lock(&space_info
->lock
);
6524 spin_lock(&cache
->lock
);
6528 cache
->reserved
+= num_bytes
;
6529 space_info
->bytes_reserved
+= num_bytes
;
6531 trace_btrfs_space_reservation(cache
->fs_info
,
6532 "space_info", space_info
->flags
,
6534 space_info
->bytes_may_use
-= ram_bytes
;
6536 cache
->delalloc_bytes
+= num_bytes
;
6538 spin_unlock(&cache
->lock
);
6539 spin_unlock(&space_info
->lock
);
6544 * btrfs_free_reserved_bytes - update the block_group and space info counters
6545 * @cache: The cache we are manipulating
6546 * @num_bytes: The number of bytes in question
6547 * @delalloc: The blocks are allocated for the delalloc write
6549 * This is called by somebody who is freeing space that was never actually used
6550 * on disk. For example if you reserve some space for a new leaf in transaction
6551 * A and before transaction A commits you free that leaf, you call this with
6552 * reserve set to 0 in order to clear the reservation.
6555 static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache
*cache
,
6556 u64 num_bytes
, int delalloc
)
6558 struct btrfs_space_info
*space_info
= cache
->space_info
;
6561 spin_lock(&space_info
->lock
);
6562 spin_lock(&cache
->lock
);
6564 space_info
->bytes_readonly
+= num_bytes
;
6565 cache
->reserved
-= num_bytes
;
6566 space_info
->bytes_reserved
-= num_bytes
;
6569 cache
->delalloc_bytes
-= num_bytes
;
6570 spin_unlock(&cache
->lock
);
6571 spin_unlock(&space_info
->lock
);
6574 void btrfs_prepare_extent_commit(struct btrfs_trans_handle
*trans
,
6575 struct btrfs_root
*root
)
6577 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6578 struct btrfs_caching_control
*next
;
6579 struct btrfs_caching_control
*caching_ctl
;
6580 struct btrfs_block_group_cache
*cache
;
6582 down_write(&fs_info
->commit_root_sem
);
6584 list_for_each_entry_safe(caching_ctl
, next
,
6585 &fs_info
->caching_block_groups
, list
) {
6586 cache
= caching_ctl
->block_group
;
6587 if (block_group_cache_done(cache
)) {
6588 cache
->last_byte_to_unpin
= (u64
)-1;
6589 list_del_init(&caching_ctl
->list
);
6590 put_caching_control(caching_ctl
);
6592 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
6596 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
6597 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
6599 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
6601 up_write(&fs_info
->commit_root_sem
);
6603 update_global_block_rsv(fs_info
);
6607 * Returns the free cluster for the given space info and sets empty_cluster to
6608 * what it should be based on the mount options.
6610 static struct btrfs_free_cluster
*
6611 fetch_cluster_info(struct btrfs_root
*root
, struct btrfs_space_info
*space_info
,
6614 struct btrfs_free_cluster
*ret
= NULL
;
6615 bool ssd
= btrfs_test_opt(root
->fs_info
, SSD
);
6618 if (btrfs_mixed_space_info(space_info
))
6622 *empty_cluster
= SZ_2M
;
6623 if (space_info
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
6624 ret
= &root
->fs_info
->meta_alloc_cluster
;
6626 *empty_cluster
= SZ_64K
;
6627 } else if ((space_info
->flags
& BTRFS_BLOCK_GROUP_DATA
) && ssd
) {
6628 ret
= &root
->fs_info
->data_alloc_cluster
;
6634 static int unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
,
6635 const bool return_free_space
)
6637 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6638 struct btrfs_block_group_cache
*cache
= NULL
;
6639 struct btrfs_space_info
*space_info
;
6640 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
6641 struct btrfs_free_cluster
*cluster
= NULL
;
6643 u64 total_unpinned
= 0;
6644 u64 empty_cluster
= 0;
6647 while (start
<= end
) {
6650 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
6652 btrfs_put_block_group(cache
);
6654 cache
= btrfs_lookup_block_group(fs_info
, start
);
6655 BUG_ON(!cache
); /* Logic error */
6657 cluster
= fetch_cluster_info(root
,
6660 empty_cluster
<<= 1;
6663 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
6664 len
= min(len
, end
+ 1 - start
);
6666 if (start
< cache
->last_byte_to_unpin
) {
6667 len
= min(len
, cache
->last_byte_to_unpin
- start
);
6668 if (return_free_space
)
6669 btrfs_add_free_space(cache
, start
, len
);
6673 total_unpinned
+= len
;
6674 space_info
= cache
->space_info
;
6677 * If this space cluster has been marked as fragmented and we've
6678 * unpinned enough in this block group to potentially allow a
6679 * cluster to be created inside of it go ahead and clear the
6682 if (cluster
&& cluster
->fragmented
&&
6683 total_unpinned
> empty_cluster
) {
6684 spin_lock(&cluster
->lock
);
6685 cluster
->fragmented
= 0;
6686 spin_unlock(&cluster
->lock
);
6689 spin_lock(&space_info
->lock
);
6690 spin_lock(&cache
->lock
);
6691 cache
->pinned
-= len
;
6692 space_info
->bytes_pinned
-= len
;
6694 trace_btrfs_space_reservation(fs_info
, "pinned",
6695 space_info
->flags
, len
, 0);
6696 space_info
->max_extent_size
= 0;
6697 percpu_counter_add(&space_info
->total_bytes_pinned
, -len
);
6699 space_info
->bytes_readonly
+= len
;
6702 spin_unlock(&cache
->lock
);
6703 if (!readonly
&& return_free_space
&&
6704 global_rsv
->space_info
== space_info
) {
6706 WARN_ON(!return_free_space
);
6707 spin_lock(&global_rsv
->lock
);
6708 if (!global_rsv
->full
) {
6709 to_add
= min(len
, global_rsv
->size
-
6710 global_rsv
->reserved
);
6711 global_rsv
->reserved
+= to_add
;
6712 space_info
->bytes_may_use
+= to_add
;
6713 if (global_rsv
->reserved
>= global_rsv
->size
)
6714 global_rsv
->full
= 1;
6715 trace_btrfs_space_reservation(fs_info
,
6721 spin_unlock(&global_rsv
->lock
);
6722 /* Add to any tickets we may have */
6724 space_info_add_new_bytes(fs_info
, space_info
,
6727 spin_unlock(&space_info
->lock
);
6731 btrfs_put_block_group(cache
);
6735 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
6736 struct btrfs_root
*root
)
6738 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6739 struct btrfs_block_group_cache
*block_group
, *tmp
;
6740 struct list_head
*deleted_bgs
;
6741 struct extent_io_tree
*unpin
;
6746 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
6747 unpin
= &fs_info
->freed_extents
[1];
6749 unpin
= &fs_info
->freed_extents
[0];
6751 while (!trans
->aborted
) {
6752 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
6753 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
6754 EXTENT_DIRTY
, NULL
);
6756 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
6760 if (btrfs_test_opt(root
->fs_info
, DISCARD
))
6761 ret
= btrfs_discard_extent(root
, start
,
6762 end
+ 1 - start
, NULL
);
6764 clear_extent_dirty(unpin
, start
, end
);
6765 unpin_extent_range(root
, start
, end
, true);
6766 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
6771 * Transaction is finished. We don't need the lock anymore. We
6772 * do need to clean up the block groups in case of a transaction
6775 deleted_bgs
= &trans
->transaction
->deleted_bgs
;
6776 list_for_each_entry_safe(block_group
, tmp
, deleted_bgs
, bg_list
) {
6780 if (!trans
->aborted
)
6781 ret
= btrfs_discard_extent(root
,
6782 block_group
->key
.objectid
,
6783 block_group
->key
.offset
,
6786 list_del_init(&block_group
->bg_list
);
6787 btrfs_put_block_group_trimming(block_group
);
6788 btrfs_put_block_group(block_group
);
6791 const char *errstr
= btrfs_decode_error(ret
);
6793 "Discard failed while removing blockgroup: errno=%d %s\n",
6801 static void add_pinned_bytes(struct btrfs_fs_info
*fs_info
, u64 num_bytes
,
6802 u64 owner
, u64 root_objectid
)
6804 struct btrfs_space_info
*space_info
;
6807 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
6808 if (root_objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
6809 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
6811 flags
= BTRFS_BLOCK_GROUP_METADATA
;
6813 flags
= BTRFS_BLOCK_GROUP_DATA
;
6816 space_info
= __find_space_info(fs_info
, flags
);
6817 BUG_ON(!space_info
); /* Logic bug */
6818 percpu_counter_add(&space_info
->total_bytes_pinned
, num_bytes
);
6822 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
6823 struct btrfs_root
*root
,
6824 struct btrfs_delayed_ref_node
*node
, u64 parent
,
6825 u64 root_objectid
, u64 owner_objectid
,
6826 u64 owner_offset
, int refs_to_drop
,
6827 struct btrfs_delayed_extent_op
*extent_op
)
6829 struct btrfs_key key
;
6830 struct btrfs_path
*path
;
6831 struct btrfs_fs_info
*info
= root
->fs_info
;
6832 struct btrfs_root
*extent_root
= info
->extent_root
;
6833 struct extent_buffer
*leaf
;
6834 struct btrfs_extent_item
*ei
;
6835 struct btrfs_extent_inline_ref
*iref
;
6838 int extent_slot
= 0;
6839 int found_extent
= 0;
6843 u64 bytenr
= node
->bytenr
;
6844 u64 num_bytes
= node
->num_bytes
;
6846 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
6849 path
= btrfs_alloc_path();
6853 path
->reada
= READA_FORWARD
;
6854 path
->leave_spinning
= 1;
6856 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
6857 BUG_ON(!is_data
&& refs_to_drop
!= 1);
6860 skinny_metadata
= 0;
6862 ret
= lookup_extent_backref(trans
, extent_root
, path
, &iref
,
6863 bytenr
, num_bytes
, parent
,
6864 root_objectid
, owner_objectid
,
6867 extent_slot
= path
->slots
[0];
6868 while (extent_slot
>= 0) {
6869 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
6871 if (key
.objectid
!= bytenr
)
6873 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
6874 key
.offset
== num_bytes
) {
6878 if (key
.type
== BTRFS_METADATA_ITEM_KEY
&&
6879 key
.offset
== owner_objectid
) {
6883 if (path
->slots
[0] - extent_slot
> 5)
6887 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6888 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
6889 if (found_extent
&& item_size
< sizeof(*ei
))
6892 if (!found_extent
) {
6894 ret
= remove_extent_backref(trans
, extent_root
, path
,
6896 is_data
, &last_ref
);
6898 btrfs_abort_transaction(trans
, ret
);
6901 btrfs_release_path(path
);
6902 path
->leave_spinning
= 1;
6904 key
.objectid
= bytenr
;
6905 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6906 key
.offset
= num_bytes
;
6908 if (!is_data
&& skinny_metadata
) {
6909 key
.type
= BTRFS_METADATA_ITEM_KEY
;
6910 key
.offset
= owner_objectid
;
6913 ret
= btrfs_search_slot(trans
, extent_root
,
6915 if (ret
> 0 && skinny_metadata
&& path
->slots
[0]) {
6917 * Couldn't find our skinny metadata item,
6918 * see if we have ye olde extent item.
6921 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
6923 if (key
.objectid
== bytenr
&&
6924 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
6925 key
.offset
== num_bytes
)
6929 if (ret
> 0 && skinny_metadata
) {
6930 skinny_metadata
= false;
6931 key
.objectid
= bytenr
;
6932 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6933 key
.offset
= num_bytes
;
6934 btrfs_release_path(path
);
6935 ret
= btrfs_search_slot(trans
, extent_root
,
6940 btrfs_err(info
, "umm, got %d back from search, was looking for %llu",
6943 btrfs_print_leaf(extent_root
,
6947 btrfs_abort_transaction(trans
, ret
);
6950 extent_slot
= path
->slots
[0];
6952 } else if (WARN_ON(ret
== -ENOENT
)) {
6953 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
6955 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6956 bytenr
, parent
, root_objectid
, owner_objectid
,
6958 btrfs_abort_transaction(trans
, ret
);
6961 btrfs_abort_transaction(trans
, ret
);
6965 leaf
= path
->nodes
[0];
6966 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
6967 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6968 if (item_size
< sizeof(*ei
)) {
6969 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
6970 ret
= convert_extent_item_v0(trans
, extent_root
, path
,
6973 btrfs_abort_transaction(trans
, ret
);
6977 btrfs_release_path(path
);
6978 path
->leave_spinning
= 1;
6980 key
.objectid
= bytenr
;
6981 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6982 key
.offset
= num_bytes
;
6984 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
6987 btrfs_err(info
, "umm, got %d back from search, was looking for %llu",
6989 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
6992 btrfs_abort_transaction(trans
, ret
);
6996 extent_slot
= path
->slots
[0];
6997 leaf
= path
->nodes
[0];
6998 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
7001 BUG_ON(item_size
< sizeof(*ei
));
7002 ei
= btrfs_item_ptr(leaf
, extent_slot
,
7003 struct btrfs_extent_item
);
7004 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
7005 key
.type
== BTRFS_EXTENT_ITEM_KEY
) {
7006 struct btrfs_tree_block_info
*bi
;
7007 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
7008 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
7009 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
7012 refs
= btrfs_extent_refs(leaf
, ei
);
7013 if (refs
< refs_to_drop
) {
7014 btrfs_err(info
, "trying to drop %d refs but we only have %Lu "
7015 "for bytenr %Lu", refs_to_drop
, refs
, bytenr
);
7017 btrfs_abort_transaction(trans
, ret
);
7020 refs
-= refs_to_drop
;
7024 __run_delayed_extent_op(extent_op
, leaf
, ei
);
7026 * In the case of inline back ref, reference count will
7027 * be updated by remove_extent_backref
7030 BUG_ON(!found_extent
);
7032 btrfs_set_extent_refs(leaf
, ei
, refs
);
7033 btrfs_mark_buffer_dirty(leaf
);
7036 ret
= remove_extent_backref(trans
, extent_root
, path
,
7038 is_data
, &last_ref
);
7040 btrfs_abort_transaction(trans
, ret
);
7044 add_pinned_bytes(root
->fs_info
, -num_bytes
, owner_objectid
,
7048 BUG_ON(is_data
&& refs_to_drop
!=
7049 extent_data_ref_count(path
, iref
));
7051 BUG_ON(path
->slots
[0] != extent_slot
);
7053 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
7054 path
->slots
[0] = extent_slot
;
7060 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
7063 btrfs_abort_transaction(trans
, ret
);
7066 btrfs_release_path(path
);
7069 ret
= btrfs_del_csums(trans
, root
, bytenr
, num_bytes
);
7071 btrfs_abort_transaction(trans
, ret
);
7076 ret
= add_to_free_space_tree(trans
, root
->fs_info
, bytenr
,
7079 btrfs_abort_transaction(trans
, ret
);
7083 ret
= update_block_group(trans
, root
, bytenr
, num_bytes
, 0);
7085 btrfs_abort_transaction(trans
, ret
);
7089 btrfs_release_path(path
);
7092 btrfs_free_path(path
);
7097 * when we free an block, it is possible (and likely) that we free the last
7098 * delayed ref for that extent as well. This searches the delayed ref tree for
7099 * a given extent, and if there are no other delayed refs to be processed, it
7100 * removes it from the tree.
7102 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
7103 struct btrfs_root
*root
, u64 bytenr
)
7105 struct btrfs_delayed_ref_head
*head
;
7106 struct btrfs_delayed_ref_root
*delayed_refs
;
7109 delayed_refs
= &trans
->transaction
->delayed_refs
;
7110 spin_lock(&delayed_refs
->lock
);
7111 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
7113 goto out_delayed_unlock
;
7115 spin_lock(&head
->lock
);
7116 if (!list_empty(&head
->ref_list
))
7119 if (head
->extent_op
) {
7120 if (!head
->must_insert_reserved
)
7122 btrfs_free_delayed_extent_op(head
->extent_op
);
7123 head
->extent_op
= NULL
;
7127 * waiting for the lock here would deadlock. If someone else has it
7128 * locked they are already in the process of dropping it anyway
7130 if (!mutex_trylock(&head
->mutex
))
7134 * at this point we have a head with no other entries. Go
7135 * ahead and process it.
7137 head
->node
.in_tree
= 0;
7138 rb_erase(&head
->href_node
, &delayed_refs
->href_root
);
7140 atomic_dec(&delayed_refs
->num_entries
);
7143 * we don't take a ref on the node because we're removing it from the
7144 * tree, so we just steal the ref the tree was holding.
7146 delayed_refs
->num_heads
--;
7147 if (head
->processing
== 0)
7148 delayed_refs
->num_heads_ready
--;
7149 head
->processing
= 0;
7150 spin_unlock(&head
->lock
);
7151 spin_unlock(&delayed_refs
->lock
);
7153 BUG_ON(head
->extent_op
);
7154 if (head
->must_insert_reserved
)
7157 mutex_unlock(&head
->mutex
);
7158 btrfs_put_delayed_ref(&head
->node
);
7161 spin_unlock(&head
->lock
);
7164 spin_unlock(&delayed_refs
->lock
);
7168 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
7169 struct btrfs_root
*root
,
7170 struct extent_buffer
*buf
,
7171 u64 parent
, int last_ref
)
7176 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
7177 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
7178 buf
->start
, buf
->len
,
7179 parent
, root
->root_key
.objectid
,
7180 btrfs_header_level(buf
),
7181 BTRFS_DROP_DELAYED_REF
, NULL
);
7182 BUG_ON(ret
); /* -ENOMEM */
7188 if (btrfs_header_generation(buf
) == trans
->transid
) {
7189 struct btrfs_block_group_cache
*cache
;
7191 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
7192 ret
= check_ref_cleanup(trans
, root
, buf
->start
);
7197 cache
= btrfs_lookup_block_group(root
->fs_info
, buf
->start
);
7199 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
7200 pin_down_extent(root
, cache
, buf
->start
, buf
->len
, 1);
7201 btrfs_put_block_group(cache
);
7205 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
7207 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
7208 btrfs_free_reserved_bytes(cache
, buf
->len
, 0);
7209 btrfs_put_block_group(cache
);
7210 trace_btrfs_reserved_extent_free(root
, buf
->start
, buf
->len
);
7215 add_pinned_bytes(root
->fs_info
, buf
->len
,
7216 btrfs_header_level(buf
),
7217 root
->root_key
.objectid
);
7220 * Deleting the buffer, clear the corrupt flag since it doesn't matter
7223 clear_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
);
7226 /* Can return -ENOMEM */
7227 int btrfs_free_extent(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
7228 u64 bytenr
, u64 num_bytes
, u64 parent
, u64 root_objectid
,
7229 u64 owner
, u64 offset
)
7232 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7234 if (btrfs_is_testing(fs_info
))
7237 add_pinned_bytes(root
->fs_info
, num_bytes
, owner
, root_objectid
);
7240 * tree log blocks never actually go into the extent allocation
7241 * tree, just update pinning info and exit early.
7243 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
7244 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
7245 /* unlocks the pinned mutex */
7246 btrfs_pin_extent(root
, bytenr
, num_bytes
, 1);
7248 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
7249 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
7251 parent
, root_objectid
, (int)owner
,
7252 BTRFS_DROP_DELAYED_REF
, NULL
);
7254 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
7256 parent
, root_objectid
, owner
,
7258 BTRFS_DROP_DELAYED_REF
, NULL
);
7264 * when we wait for progress in the block group caching, its because
7265 * our allocation attempt failed at least once. So, we must sleep
7266 * and let some progress happen before we try again.
7268 * This function will sleep at least once waiting for new free space to
7269 * show up, and then it will check the block group free space numbers
7270 * for our min num_bytes. Another option is to have it go ahead
7271 * and look in the rbtree for a free extent of a given size, but this
7274 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7275 * any of the information in this block group.
7277 static noinline
void
7278 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
7281 struct btrfs_caching_control
*caching_ctl
;
7283 caching_ctl
= get_caching_control(cache
);
7287 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
7288 (cache
->free_space_ctl
->free_space
>= num_bytes
));
7290 put_caching_control(caching_ctl
);
7294 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
7296 struct btrfs_caching_control
*caching_ctl
;
7299 caching_ctl
= get_caching_control(cache
);
7301 return (cache
->cached
== BTRFS_CACHE_ERROR
) ? -EIO
: 0;
7303 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
7304 if (cache
->cached
== BTRFS_CACHE_ERROR
)
7306 put_caching_control(caching_ctl
);
7310 int __get_raid_index(u64 flags
)
7312 if (flags
& BTRFS_BLOCK_GROUP_RAID10
)
7313 return BTRFS_RAID_RAID10
;
7314 else if (flags
& BTRFS_BLOCK_GROUP_RAID1
)
7315 return BTRFS_RAID_RAID1
;
7316 else if (flags
& BTRFS_BLOCK_GROUP_DUP
)
7317 return BTRFS_RAID_DUP
;
7318 else if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
7319 return BTRFS_RAID_RAID0
;
7320 else if (flags
& BTRFS_BLOCK_GROUP_RAID5
)
7321 return BTRFS_RAID_RAID5
;
7322 else if (flags
& BTRFS_BLOCK_GROUP_RAID6
)
7323 return BTRFS_RAID_RAID6
;
7325 return BTRFS_RAID_SINGLE
; /* BTRFS_BLOCK_GROUP_SINGLE */
7328 int get_block_group_index(struct btrfs_block_group_cache
*cache
)
7330 return __get_raid_index(cache
->flags
);
7333 static const char *btrfs_raid_type_names
[BTRFS_NR_RAID_TYPES
] = {
7334 [BTRFS_RAID_RAID10
] = "raid10",
7335 [BTRFS_RAID_RAID1
] = "raid1",
7336 [BTRFS_RAID_DUP
] = "dup",
7337 [BTRFS_RAID_RAID0
] = "raid0",
7338 [BTRFS_RAID_SINGLE
] = "single",
7339 [BTRFS_RAID_RAID5
] = "raid5",
7340 [BTRFS_RAID_RAID6
] = "raid6",
7343 static const char *get_raid_name(enum btrfs_raid_types type
)
7345 if (type
>= BTRFS_NR_RAID_TYPES
)
7348 return btrfs_raid_type_names
[type
];
7351 enum btrfs_loop_type
{
7352 LOOP_CACHING_NOWAIT
= 0,
7353 LOOP_CACHING_WAIT
= 1,
7354 LOOP_ALLOC_CHUNK
= 2,
7355 LOOP_NO_EMPTY_SIZE
= 3,
7359 btrfs_lock_block_group(struct btrfs_block_group_cache
*cache
,
7363 down_read(&cache
->data_rwsem
);
7367 btrfs_grab_block_group(struct btrfs_block_group_cache
*cache
,
7370 btrfs_get_block_group(cache
);
7372 down_read(&cache
->data_rwsem
);
7375 static struct btrfs_block_group_cache
*
7376 btrfs_lock_cluster(struct btrfs_block_group_cache
*block_group
,
7377 struct btrfs_free_cluster
*cluster
,
7380 struct btrfs_block_group_cache
*used_bg
= NULL
;
7382 spin_lock(&cluster
->refill_lock
);
7384 used_bg
= cluster
->block_group
;
7388 if (used_bg
== block_group
)
7391 btrfs_get_block_group(used_bg
);
7396 if (down_read_trylock(&used_bg
->data_rwsem
))
7399 spin_unlock(&cluster
->refill_lock
);
7401 down_read(&used_bg
->data_rwsem
);
7403 spin_lock(&cluster
->refill_lock
);
7404 if (used_bg
== cluster
->block_group
)
7407 up_read(&used_bg
->data_rwsem
);
7408 btrfs_put_block_group(used_bg
);
7413 btrfs_release_block_group(struct btrfs_block_group_cache
*cache
,
7417 up_read(&cache
->data_rwsem
);
7418 btrfs_put_block_group(cache
);
7422 * walks the btree of allocated extents and find a hole of a given size.
7423 * The key ins is changed to record the hole:
7424 * ins->objectid == start position
7425 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7426 * ins->offset == the size of the hole.
7427 * Any available blocks before search_start are skipped.
7429 * If there is no suitable free space, we will record the max size of
7430 * the free space extent currently.
7432 static noinline
int find_free_extent(struct btrfs_root
*orig_root
,
7433 u64 ram_bytes
, u64 num_bytes
, u64 empty_size
,
7434 u64 hint_byte
, struct btrfs_key
*ins
,
7435 u64 flags
, int delalloc
)
7438 struct btrfs_root
*root
= orig_root
->fs_info
->extent_root
;
7439 struct btrfs_free_cluster
*last_ptr
= NULL
;
7440 struct btrfs_block_group_cache
*block_group
= NULL
;
7441 u64 search_start
= 0;
7442 u64 max_extent_size
= 0;
7443 u64 empty_cluster
= 0;
7444 struct btrfs_space_info
*space_info
;
7446 int index
= __get_raid_index(flags
);
7447 bool failed_cluster_refill
= false;
7448 bool failed_alloc
= false;
7449 bool use_cluster
= true;
7450 bool have_caching_bg
= false;
7451 bool orig_have_caching_bg
= false;
7452 bool full_search
= false;
7454 WARN_ON(num_bytes
< root
->sectorsize
);
7455 ins
->type
= BTRFS_EXTENT_ITEM_KEY
;
7459 trace_find_free_extent(orig_root
, num_bytes
, empty_size
, flags
);
7461 space_info
= __find_space_info(root
->fs_info
, flags
);
7463 btrfs_err(root
->fs_info
, "No space info for %llu", flags
);
7468 * If our free space is heavily fragmented we may not be able to make
7469 * big contiguous allocations, so instead of doing the expensive search
7470 * for free space, simply return ENOSPC with our max_extent_size so we
7471 * can go ahead and search for a more manageable chunk.
7473 * If our max_extent_size is large enough for our allocation simply
7474 * disable clustering since we will likely not be able to find enough
7475 * space to create a cluster and induce latency trying.
7477 if (unlikely(space_info
->max_extent_size
)) {
7478 spin_lock(&space_info
->lock
);
7479 if (space_info
->max_extent_size
&&
7480 num_bytes
> space_info
->max_extent_size
) {
7481 ins
->offset
= space_info
->max_extent_size
;
7482 spin_unlock(&space_info
->lock
);
7484 } else if (space_info
->max_extent_size
) {
7485 use_cluster
= false;
7487 spin_unlock(&space_info
->lock
);
7490 last_ptr
= fetch_cluster_info(orig_root
, space_info
, &empty_cluster
);
7492 spin_lock(&last_ptr
->lock
);
7493 if (last_ptr
->block_group
)
7494 hint_byte
= last_ptr
->window_start
;
7495 if (last_ptr
->fragmented
) {
7497 * We still set window_start so we can keep track of the
7498 * last place we found an allocation to try and save
7501 hint_byte
= last_ptr
->window_start
;
7502 use_cluster
= false;
7504 spin_unlock(&last_ptr
->lock
);
7507 search_start
= max(search_start
, first_logical_byte(root
, 0));
7508 search_start
= max(search_start
, hint_byte
);
7509 if (search_start
== hint_byte
) {
7510 block_group
= btrfs_lookup_block_group(root
->fs_info
,
7513 * we don't want to use the block group if it doesn't match our
7514 * allocation bits, or if its not cached.
7516 * However if we are re-searching with an ideal block group
7517 * picked out then we don't care that the block group is cached.
7519 if (block_group
&& block_group_bits(block_group
, flags
) &&
7520 block_group
->cached
!= BTRFS_CACHE_NO
) {
7521 down_read(&space_info
->groups_sem
);
7522 if (list_empty(&block_group
->list
) ||
7525 * someone is removing this block group,
7526 * we can't jump into the have_block_group
7527 * target because our list pointers are not
7530 btrfs_put_block_group(block_group
);
7531 up_read(&space_info
->groups_sem
);
7533 index
= get_block_group_index(block_group
);
7534 btrfs_lock_block_group(block_group
, delalloc
);
7535 goto have_block_group
;
7537 } else if (block_group
) {
7538 btrfs_put_block_group(block_group
);
7542 have_caching_bg
= false;
7543 if (index
== 0 || index
== __get_raid_index(flags
))
7545 down_read(&space_info
->groups_sem
);
7546 list_for_each_entry(block_group
, &space_info
->block_groups
[index
],
7551 btrfs_grab_block_group(block_group
, delalloc
);
7552 search_start
= block_group
->key
.objectid
;
7555 * this can happen if we end up cycling through all the
7556 * raid types, but we want to make sure we only allocate
7557 * for the proper type.
7559 if (!block_group_bits(block_group
, flags
)) {
7560 u64 extra
= BTRFS_BLOCK_GROUP_DUP
|
7561 BTRFS_BLOCK_GROUP_RAID1
|
7562 BTRFS_BLOCK_GROUP_RAID5
|
7563 BTRFS_BLOCK_GROUP_RAID6
|
7564 BTRFS_BLOCK_GROUP_RAID10
;
7567 * if they asked for extra copies and this block group
7568 * doesn't provide them, bail. This does allow us to
7569 * fill raid0 from raid1.
7571 if ((flags
& extra
) && !(block_group
->flags
& extra
))
7576 cached
= block_group_cache_done(block_group
);
7577 if (unlikely(!cached
)) {
7578 have_caching_bg
= true;
7579 ret
= cache_block_group(block_group
, 0);
7584 if (unlikely(block_group
->cached
== BTRFS_CACHE_ERROR
))
7586 if (unlikely(block_group
->ro
))
7590 * Ok we want to try and use the cluster allocator, so
7593 if (last_ptr
&& use_cluster
) {
7594 struct btrfs_block_group_cache
*used_block_group
;
7595 unsigned long aligned_cluster
;
7597 * the refill lock keeps out other
7598 * people trying to start a new cluster
7600 used_block_group
= btrfs_lock_cluster(block_group
,
7603 if (!used_block_group
)
7604 goto refill_cluster
;
7606 if (used_block_group
!= block_group
&&
7607 (used_block_group
->ro
||
7608 !block_group_bits(used_block_group
, flags
)))
7609 goto release_cluster
;
7611 offset
= btrfs_alloc_from_cluster(used_block_group
,
7614 used_block_group
->key
.objectid
,
7617 /* we have a block, we're done */
7618 spin_unlock(&last_ptr
->refill_lock
);
7619 trace_btrfs_reserve_extent_cluster(root
,
7621 search_start
, num_bytes
);
7622 if (used_block_group
!= block_group
) {
7623 btrfs_release_block_group(block_group
,
7625 block_group
= used_block_group
;
7630 WARN_ON(last_ptr
->block_group
!= used_block_group
);
7632 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7633 * set up a new clusters, so lets just skip it
7634 * and let the allocator find whatever block
7635 * it can find. If we reach this point, we
7636 * will have tried the cluster allocator
7637 * plenty of times and not have found
7638 * anything, so we are likely way too
7639 * fragmented for the clustering stuff to find
7642 * However, if the cluster is taken from the
7643 * current block group, release the cluster
7644 * first, so that we stand a better chance of
7645 * succeeding in the unclustered
7647 if (loop
>= LOOP_NO_EMPTY_SIZE
&&
7648 used_block_group
!= block_group
) {
7649 spin_unlock(&last_ptr
->refill_lock
);
7650 btrfs_release_block_group(used_block_group
,
7652 goto unclustered_alloc
;
7656 * this cluster didn't work out, free it and
7659 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
7661 if (used_block_group
!= block_group
)
7662 btrfs_release_block_group(used_block_group
,
7665 if (loop
>= LOOP_NO_EMPTY_SIZE
) {
7666 spin_unlock(&last_ptr
->refill_lock
);
7667 goto unclustered_alloc
;
7670 aligned_cluster
= max_t(unsigned long,
7671 empty_cluster
+ empty_size
,
7672 block_group
->full_stripe_len
);
7674 /* allocate a cluster in this block group */
7675 ret
= btrfs_find_space_cluster(root
, block_group
,
7676 last_ptr
, search_start
,
7681 * now pull our allocation out of this
7684 offset
= btrfs_alloc_from_cluster(block_group
,
7690 /* we found one, proceed */
7691 spin_unlock(&last_ptr
->refill_lock
);
7692 trace_btrfs_reserve_extent_cluster(root
,
7693 block_group
, search_start
,
7697 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
7698 && !failed_cluster_refill
) {
7699 spin_unlock(&last_ptr
->refill_lock
);
7701 failed_cluster_refill
= true;
7702 wait_block_group_cache_progress(block_group
,
7703 num_bytes
+ empty_cluster
+ empty_size
);
7704 goto have_block_group
;
7708 * at this point we either didn't find a cluster
7709 * or we weren't able to allocate a block from our
7710 * cluster. Free the cluster we've been trying
7711 * to use, and go to the next block group
7713 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
7714 spin_unlock(&last_ptr
->refill_lock
);
7720 * We are doing an unclustered alloc, set the fragmented flag so
7721 * we don't bother trying to setup a cluster again until we get
7724 if (unlikely(last_ptr
)) {
7725 spin_lock(&last_ptr
->lock
);
7726 last_ptr
->fragmented
= 1;
7727 spin_unlock(&last_ptr
->lock
);
7729 spin_lock(&block_group
->free_space_ctl
->tree_lock
);
7731 block_group
->free_space_ctl
->free_space
<
7732 num_bytes
+ empty_cluster
+ empty_size
) {
7733 if (block_group
->free_space_ctl
->free_space
>
7736 block_group
->free_space_ctl
->free_space
;
7737 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
7740 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
7742 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
7743 num_bytes
, empty_size
,
7746 * If we didn't find a chunk, and we haven't failed on this
7747 * block group before, and this block group is in the middle of
7748 * caching and we are ok with waiting, then go ahead and wait
7749 * for progress to be made, and set failed_alloc to true.
7751 * If failed_alloc is true then we've already waited on this
7752 * block group once and should move on to the next block group.
7754 if (!offset
&& !failed_alloc
&& !cached
&&
7755 loop
> LOOP_CACHING_NOWAIT
) {
7756 wait_block_group_cache_progress(block_group
,
7757 num_bytes
+ empty_size
);
7758 failed_alloc
= true;
7759 goto have_block_group
;
7760 } else if (!offset
) {
7764 search_start
= ALIGN(offset
, root
->stripesize
);
7766 /* move on to the next group */
7767 if (search_start
+ num_bytes
>
7768 block_group
->key
.objectid
+ block_group
->key
.offset
) {
7769 btrfs_add_free_space(block_group
, offset
, num_bytes
);
7773 if (offset
< search_start
)
7774 btrfs_add_free_space(block_group
, offset
,
7775 search_start
- offset
);
7776 BUG_ON(offset
> search_start
);
7778 ret
= btrfs_add_reserved_bytes(block_group
, ram_bytes
,
7779 num_bytes
, delalloc
);
7780 if (ret
== -EAGAIN
) {
7781 btrfs_add_free_space(block_group
, offset
, num_bytes
);
7784 btrfs_inc_block_group_reservations(block_group
);
7786 /* we are all good, lets return */
7787 ins
->objectid
= search_start
;
7788 ins
->offset
= num_bytes
;
7790 trace_btrfs_reserve_extent(orig_root
, block_group
,
7791 search_start
, num_bytes
);
7792 btrfs_release_block_group(block_group
, delalloc
);
7795 failed_cluster_refill
= false;
7796 failed_alloc
= false;
7797 BUG_ON(index
!= get_block_group_index(block_group
));
7798 btrfs_release_block_group(block_group
, delalloc
);
7800 up_read(&space_info
->groups_sem
);
7802 if ((loop
== LOOP_CACHING_NOWAIT
) && have_caching_bg
7803 && !orig_have_caching_bg
)
7804 orig_have_caching_bg
= true;
7806 if (!ins
->objectid
&& loop
>= LOOP_CACHING_WAIT
&& have_caching_bg
)
7809 if (!ins
->objectid
&& ++index
< BTRFS_NR_RAID_TYPES
)
7813 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7814 * caching kthreads as we move along
7815 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7816 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7817 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7820 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
) {
7822 if (loop
== LOOP_CACHING_NOWAIT
) {
7824 * We want to skip the LOOP_CACHING_WAIT step if we
7825 * don't have any uncached bgs and we've already done a
7826 * full search through.
7828 if (orig_have_caching_bg
|| !full_search
)
7829 loop
= LOOP_CACHING_WAIT
;
7831 loop
= LOOP_ALLOC_CHUNK
;
7836 if (loop
== LOOP_ALLOC_CHUNK
) {
7837 struct btrfs_trans_handle
*trans
;
7840 trans
= current
->journal_info
;
7844 trans
= btrfs_join_transaction(root
);
7846 if (IS_ERR(trans
)) {
7847 ret
= PTR_ERR(trans
);
7851 ret
= do_chunk_alloc(trans
, root
, flags
,
7855 * If we can't allocate a new chunk we've already looped
7856 * through at least once, move on to the NO_EMPTY_SIZE
7860 loop
= LOOP_NO_EMPTY_SIZE
;
7863 * Do not bail out on ENOSPC since we
7864 * can do more things.
7866 if (ret
< 0 && ret
!= -ENOSPC
)
7867 btrfs_abort_transaction(trans
, ret
);
7871 btrfs_end_transaction(trans
, root
);
7876 if (loop
== LOOP_NO_EMPTY_SIZE
) {
7878 * Don't loop again if we already have no empty_size and
7881 if (empty_size
== 0 &&
7882 empty_cluster
== 0) {
7891 } else if (!ins
->objectid
) {
7893 } else if (ins
->objectid
) {
7894 if (!use_cluster
&& last_ptr
) {
7895 spin_lock(&last_ptr
->lock
);
7896 last_ptr
->window_start
= ins
->objectid
;
7897 spin_unlock(&last_ptr
->lock
);
7902 if (ret
== -ENOSPC
) {
7903 spin_lock(&space_info
->lock
);
7904 space_info
->max_extent_size
= max_extent_size
;
7905 spin_unlock(&space_info
->lock
);
7906 ins
->offset
= max_extent_size
;
7911 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
7912 int dump_block_groups
)
7914 struct btrfs_block_group_cache
*cache
;
7917 spin_lock(&info
->lock
);
7918 printk(KERN_INFO
"BTRFS: space_info %llu has %llu free, is %sfull\n",
7920 info
->total_bytes
- info
->bytes_used
- info
->bytes_pinned
-
7921 info
->bytes_reserved
- info
->bytes_readonly
-
7922 info
->bytes_may_use
, (info
->full
) ? "" : "not ");
7923 printk(KERN_INFO
"BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7924 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7925 info
->total_bytes
, info
->bytes_used
, info
->bytes_pinned
,
7926 info
->bytes_reserved
, info
->bytes_may_use
,
7927 info
->bytes_readonly
);
7928 spin_unlock(&info
->lock
);
7930 if (!dump_block_groups
)
7933 down_read(&info
->groups_sem
);
7935 list_for_each_entry(cache
, &info
->block_groups
[index
], list
) {
7936 spin_lock(&cache
->lock
);
7937 printk(KERN_INFO
"BTRFS: "
7938 "block group %llu has %llu bytes, "
7939 "%llu used %llu pinned %llu reserved %s\n",
7940 cache
->key
.objectid
, cache
->key
.offset
,
7941 btrfs_block_group_used(&cache
->item
), cache
->pinned
,
7942 cache
->reserved
, cache
->ro
? "[readonly]" : "");
7943 btrfs_dump_free_space(cache
, bytes
);
7944 spin_unlock(&cache
->lock
);
7946 if (++index
< BTRFS_NR_RAID_TYPES
)
7948 up_read(&info
->groups_sem
);
7951 int btrfs_reserve_extent(struct btrfs_root
*root
, u64 ram_bytes
,
7952 u64 num_bytes
, u64 min_alloc_size
,
7953 u64 empty_size
, u64 hint_byte
,
7954 struct btrfs_key
*ins
, int is_data
, int delalloc
)
7956 bool final_tried
= num_bytes
== min_alloc_size
;
7960 flags
= btrfs_get_alloc_profile(root
, is_data
);
7962 WARN_ON(num_bytes
< root
->sectorsize
);
7963 ret
= find_free_extent(root
, ram_bytes
, num_bytes
, empty_size
,
7964 hint_byte
, ins
, flags
, delalloc
);
7965 if (!ret
&& !is_data
) {
7966 btrfs_dec_block_group_reservations(root
->fs_info
,
7968 } else if (ret
== -ENOSPC
) {
7969 if (!final_tried
&& ins
->offset
) {
7970 num_bytes
= min(num_bytes
>> 1, ins
->offset
);
7971 num_bytes
= round_down(num_bytes
, root
->sectorsize
);
7972 num_bytes
= max(num_bytes
, min_alloc_size
);
7973 ram_bytes
= num_bytes
;
7974 if (num_bytes
== min_alloc_size
)
7977 } else if (btrfs_test_opt(root
->fs_info
, ENOSPC_DEBUG
)) {
7978 struct btrfs_space_info
*sinfo
;
7980 sinfo
= __find_space_info(root
->fs_info
, flags
);
7981 btrfs_err(root
->fs_info
, "allocation failed flags %llu, wanted %llu",
7984 dump_space_info(sinfo
, num_bytes
, 1);
7991 static int __btrfs_free_reserved_extent(struct btrfs_root
*root
,
7993 int pin
, int delalloc
)
7995 struct btrfs_block_group_cache
*cache
;
7998 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
8000 btrfs_err(root
->fs_info
, "Unable to find block group for %llu",
8006 pin_down_extent(root
, cache
, start
, len
, 1);
8008 if (btrfs_test_opt(root
->fs_info
, DISCARD
))
8009 ret
= btrfs_discard_extent(root
, start
, len
, NULL
);
8010 btrfs_add_free_space(cache
, start
, len
);
8011 btrfs_free_reserved_bytes(cache
, len
, delalloc
);
8012 trace_btrfs_reserved_extent_free(root
, start
, len
);
8015 btrfs_put_block_group(cache
);
8019 int btrfs_free_reserved_extent(struct btrfs_root
*root
,
8020 u64 start
, u64 len
, int delalloc
)
8022 return __btrfs_free_reserved_extent(root
, start
, len
, 0, delalloc
);
8025 int btrfs_free_and_pin_reserved_extent(struct btrfs_root
*root
,
8028 return __btrfs_free_reserved_extent(root
, start
, len
, 1, 0);
8031 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
8032 struct btrfs_root
*root
,
8033 u64 parent
, u64 root_objectid
,
8034 u64 flags
, u64 owner
, u64 offset
,
8035 struct btrfs_key
*ins
, int ref_mod
)
8038 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8039 struct btrfs_extent_item
*extent_item
;
8040 struct btrfs_extent_inline_ref
*iref
;
8041 struct btrfs_path
*path
;
8042 struct extent_buffer
*leaf
;
8047 type
= BTRFS_SHARED_DATA_REF_KEY
;
8049 type
= BTRFS_EXTENT_DATA_REF_KEY
;
8051 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
8053 path
= btrfs_alloc_path();
8057 path
->leave_spinning
= 1;
8058 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
8061 btrfs_free_path(path
);
8065 leaf
= path
->nodes
[0];
8066 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
8067 struct btrfs_extent_item
);
8068 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
8069 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
8070 btrfs_set_extent_flags(leaf
, extent_item
,
8071 flags
| BTRFS_EXTENT_FLAG_DATA
);
8073 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
8074 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
8076 struct btrfs_shared_data_ref
*ref
;
8077 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
8078 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
8079 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
8081 struct btrfs_extent_data_ref
*ref
;
8082 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
8083 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
8084 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
8085 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
8086 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
8089 btrfs_mark_buffer_dirty(path
->nodes
[0]);
8090 btrfs_free_path(path
);
8092 ret
= remove_from_free_space_tree(trans
, fs_info
, ins
->objectid
,
8097 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1);
8098 if (ret
) { /* -ENOENT, logic error */
8099 btrfs_err(fs_info
, "update block group failed for %llu %llu",
8100 ins
->objectid
, ins
->offset
);
8103 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, ins
->offset
);
8107 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
8108 struct btrfs_root
*root
,
8109 u64 parent
, u64 root_objectid
,
8110 u64 flags
, struct btrfs_disk_key
*key
,
8111 int level
, struct btrfs_key
*ins
)
8114 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8115 struct btrfs_extent_item
*extent_item
;
8116 struct btrfs_tree_block_info
*block_info
;
8117 struct btrfs_extent_inline_ref
*iref
;
8118 struct btrfs_path
*path
;
8119 struct extent_buffer
*leaf
;
8120 u32 size
= sizeof(*extent_item
) + sizeof(*iref
);
8121 u64 num_bytes
= ins
->offset
;
8122 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
8125 if (!skinny_metadata
)
8126 size
+= sizeof(*block_info
);
8128 path
= btrfs_alloc_path();
8130 btrfs_free_and_pin_reserved_extent(root
, ins
->objectid
,
8135 path
->leave_spinning
= 1;
8136 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
8139 btrfs_free_path(path
);
8140 btrfs_free_and_pin_reserved_extent(root
, ins
->objectid
,
8145 leaf
= path
->nodes
[0];
8146 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
8147 struct btrfs_extent_item
);
8148 btrfs_set_extent_refs(leaf
, extent_item
, 1);
8149 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
8150 btrfs_set_extent_flags(leaf
, extent_item
,
8151 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
8153 if (skinny_metadata
) {
8154 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
8155 num_bytes
= root
->nodesize
;
8157 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
8158 btrfs_set_tree_block_key(leaf
, block_info
, key
);
8159 btrfs_set_tree_block_level(leaf
, block_info
, level
);
8160 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
8164 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
8165 btrfs_set_extent_inline_ref_type(leaf
, iref
,
8166 BTRFS_SHARED_BLOCK_REF_KEY
);
8167 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
8169 btrfs_set_extent_inline_ref_type(leaf
, iref
,
8170 BTRFS_TREE_BLOCK_REF_KEY
);
8171 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
8174 btrfs_mark_buffer_dirty(leaf
);
8175 btrfs_free_path(path
);
8177 ret
= remove_from_free_space_tree(trans
, fs_info
, ins
->objectid
,
8182 ret
= update_block_group(trans
, root
, ins
->objectid
, root
->nodesize
,
8184 if (ret
) { /* -ENOENT, logic error */
8185 btrfs_err(fs_info
, "update block group failed for %llu %llu",
8186 ins
->objectid
, ins
->offset
);
8190 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, root
->nodesize
);
8194 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
8195 struct btrfs_root
*root
,
8196 u64 root_objectid
, u64 owner
,
8197 u64 offset
, u64 ram_bytes
,
8198 struct btrfs_key
*ins
)
8202 BUG_ON(root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
8204 ret
= btrfs_add_delayed_data_ref(root
->fs_info
, trans
, ins
->objectid
,
8206 root_objectid
, owner
, offset
,
8207 ram_bytes
, BTRFS_ADD_DELAYED_EXTENT
,
8213 * this is used by the tree logging recovery code. It records that
8214 * an extent has been allocated and makes sure to clear the free
8215 * space cache bits as well
8217 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
8218 struct btrfs_root
*root
,
8219 u64 root_objectid
, u64 owner
, u64 offset
,
8220 struct btrfs_key
*ins
)
8223 struct btrfs_block_group_cache
*block_group
;
8226 * Mixed block groups will exclude before processing the log so we only
8227 * need to do the exclude dance if this fs isn't mixed.
8229 if (!btrfs_fs_incompat(root
->fs_info
, MIXED_GROUPS
)) {
8230 ret
= __exclude_logged_extent(root
, ins
->objectid
, ins
->offset
);
8235 block_group
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
8239 ret
= btrfs_add_reserved_bytes(block_group
, ins
->offset
,
8241 BUG_ON(ret
); /* logic error */
8242 ret
= alloc_reserved_file_extent(trans
, root
, 0, root_objectid
,
8243 0, owner
, offset
, ins
, 1);
8244 btrfs_put_block_group(block_group
);
8248 static struct extent_buffer
*
8249 btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
8250 u64 bytenr
, int level
)
8252 struct extent_buffer
*buf
;
8254 buf
= btrfs_find_create_tree_block(root
, bytenr
);
8258 btrfs_set_header_generation(buf
, trans
->transid
);
8259 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, buf
, level
);
8260 btrfs_tree_lock(buf
);
8261 clean_tree_block(trans
, root
->fs_info
, buf
);
8262 clear_bit(EXTENT_BUFFER_STALE
, &buf
->bflags
);
8264 btrfs_set_lock_blocking(buf
);
8265 set_extent_buffer_uptodate(buf
);
8267 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
8268 buf
->log_index
= root
->log_transid
% 2;
8270 * we allow two log transactions at a time, use different
8271 * EXENT bit to differentiate dirty pages.
8273 if (buf
->log_index
== 0)
8274 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
8275 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
8277 set_extent_new(&root
->dirty_log_pages
, buf
->start
,
8278 buf
->start
+ buf
->len
- 1);
8280 buf
->log_index
= -1;
8281 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
8282 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
8284 trans
->dirty
= true;
8285 /* this returns a buffer locked for blocking */
8289 static struct btrfs_block_rsv
*
8290 use_block_rsv(struct btrfs_trans_handle
*trans
,
8291 struct btrfs_root
*root
, u32 blocksize
)
8293 struct btrfs_block_rsv
*block_rsv
;
8294 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
8296 bool global_updated
= false;
8298 block_rsv
= get_block_rsv(trans
, root
);
8300 if (unlikely(block_rsv
->size
== 0))
8303 ret
= block_rsv_use_bytes(block_rsv
, blocksize
);
8307 if (block_rsv
->failfast
)
8308 return ERR_PTR(ret
);
8310 if (block_rsv
->type
== BTRFS_BLOCK_RSV_GLOBAL
&& !global_updated
) {
8311 global_updated
= true;
8312 update_global_block_rsv(root
->fs_info
);
8316 if (btrfs_test_opt(root
->fs_info
, ENOSPC_DEBUG
)) {
8317 static DEFINE_RATELIMIT_STATE(_rs
,
8318 DEFAULT_RATELIMIT_INTERVAL
* 10,
8319 /*DEFAULT_RATELIMIT_BURST*/ 1);
8320 if (__ratelimit(&_rs
))
8322 "BTRFS: block rsv returned %d\n", ret
);
8325 ret
= reserve_metadata_bytes(root
, block_rsv
, blocksize
,
8326 BTRFS_RESERVE_NO_FLUSH
);
8330 * If we couldn't reserve metadata bytes try and use some from
8331 * the global reserve if its space type is the same as the global
8334 if (block_rsv
->type
!= BTRFS_BLOCK_RSV_GLOBAL
&&
8335 block_rsv
->space_info
== global_rsv
->space_info
) {
8336 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
8340 return ERR_PTR(ret
);
8343 static void unuse_block_rsv(struct btrfs_fs_info
*fs_info
,
8344 struct btrfs_block_rsv
*block_rsv
, u32 blocksize
)
8346 block_rsv_add_bytes(block_rsv
, blocksize
, 0);
8347 block_rsv_release_bytes(fs_info
, block_rsv
, NULL
, 0);
8351 * finds a free extent and does all the dirty work required for allocation
8352 * returns the tree buffer or an ERR_PTR on error.
8354 struct extent_buffer
*btrfs_alloc_tree_block(struct btrfs_trans_handle
*trans
,
8355 struct btrfs_root
*root
,
8356 u64 parent
, u64 root_objectid
,
8357 struct btrfs_disk_key
*key
, int level
,
8358 u64 hint
, u64 empty_size
)
8360 struct btrfs_key ins
;
8361 struct btrfs_block_rsv
*block_rsv
;
8362 struct extent_buffer
*buf
;
8363 struct btrfs_delayed_extent_op
*extent_op
;
8366 u32 blocksize
= root
->nodesize
;
8367 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
8370 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8371 if (btrfs_is_testing(root
->fs_info
)) {
8372 buf
= btrfs_init_new_buffer(trans
, root
, root
->alloc_bytenr
,
8375 root
->alloc_bytenr
+= blocksize
;
8380 block_rsv
= use_block_rsv(trans
, root
, blocksize
);
8381 if (IS_ERR(block_rsv
))
8382 return ERR_CAST(block_rsv
);
8384 ret
= btrfs_reserve_extent(root
, blocksize
, blocksize
, blocksize
,
8385 empty_size
, hint
, &ins
, 0, 0);
8389 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
, level
);
8392 goto out_free_reserved
;
8395 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
8397 parent
= ins
.objectid
;
8398 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
8402 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
8403 extent_op
= btrfs_alloc_delayed_extent_op();
8409 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
8411 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
8412 extent_op
->flags_to_set
= flags
;
8413 extent_op
->update_key
= skinny_metadata
? false : true;
8414 extent_op
->update_flags
= true;
8415 extent_op
->is_data
= false;
8416 extent_op
->level
= level
;
8418 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
8419 ins
.objectid
, ins
.offset
,
8420 parent
, root_objectid
, level
,
8421 BTRFS_ADD_DELAYED_EXTENT
,
8424 goto out_free_delayed
;
8429 btrfs_free_delayed_extent_op(extent_op
);
8431 free_extent_buffer(buf
);
8433 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
, 0);
8435 unuse_block_rsv(root
->fs_info
, block_rsv
, blocksize
);
8436 return ERR_PTR(ret
);
8439 struct walk_control
{
8440 u64 refs
[BTRFS_MAX_LEVEL
];
8441 u64 flags
[BTRFS_MAX_LEVEL
];
8442 struct btrfs_key update_progress
;
8453 #define DROP_REFERENCE 1
8454 #define UPDATE_BACKREF 2
8456 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
8457 struct btrfs_root
*root
,
8458 struct walk_control
*wc
,
8459 struct btrfs_path
*path
)
8466 struct btrfs_key key
;
8467 struct extent_buffer
*eb
;
8472 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
8473 wc
->reada_count
= wc
->reada_count
* 2 / 3;
8474 wc
->reada_count
= max(wc
->reada_count
, 2);
8476 wc
->reada_count
= wc
->reada_count
* 3 / 2;
8477 wc
->reada_count
= min_t(int, wc
->reada_count
,
8478 BTRFS_NODEPTRS_PER_BLOCK(root
));
8481 eb
= path
->nodes
[wc
->level
];
8482 nritems
= btrfs_header_nritems(eb
);
8484 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
8485 if (nread
>= wc
->reada_count
)
8489 bytenr
= btrfs_node_blockptr(eb
, slot
);
8490 generation
= btrfs_node_ptr_generation(eb
, slot
);
8492 if (slot
== path
->slots
[wc
->level
])
8495 if (wc
->stage
== UPDATE_BACKREF
&&
8496 generation
<= root
->root_key
.offset
)
8499 /* We don't lock the tree block, it's OK to be racy here */
8500 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
,
8501 wc
->level
- 1, 1, &refs
,
8503 /* We don't care about errors in readahead. */
8508 if (wc
->stage
== DROP_REFERENCE
) {
8512 if (wc
->level
== 1 &&
8513 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8515 if (!wc
->update_ref
||
8516 generation
<= root
->root_key
.offset
)
8518 btrfs_node_key_to_cpu(eb
, &key
, slot
);
8519 ret
= btrfs_comp_cpu_keys(&key
,
8520 &wc
->update_progress
);
8524 if (wc
->level
== 1 &&
8525 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8529 readahead_tree_block(root
, bytenr
);
8532 wc
->reada_slot
= slot
;
8535 static int account_leaf_items(struct btrfs_trans_handle
*trans
,
8536 struct btrfs_root
*root
,
8537 struct extent_buffer
*eb
)
8539 int nr
= btrfs_header_nritems(eb
);
8540 int i
, extent_type
, ret
;
8541 struct btrfs_key key
;
8542 struct btrfs_file_extent_item
*fi
;
8543 u64 bytenr
, num_bytes
;
8545 /* We can be called directly from walk_up_proc() */
8546 if (!test_bit(BTRFS_FS_QUOTA_ENABLED
, &root
->fs_info
->flags
))
8549 for (i
= 0; i
< nr
; i
++) {
8550 btrfs_item_key_to_cpu(eb
, &key
, i
);
8552 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
8555 fi
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
8556 /* filter out non qgroup-accountable extents */
8557 extent_type
= btrfs_file_extent_type(eb
, fi
);
8559 if (extent_type
== BTRFS_FILE_EXTENT_INLINE
)
8562 bytenr
= btrfs_file_extent_disk_bytenr(eb
, fi
);
8566 num_bytes
= btrfs_file_extent_disk_num_bytes(eb
, fi
);
8568 ret
= btrfs_qgroup_insert_dirty_extent(trans
, root
->fs_info
,
8569 bytenr
, num_bytes
, GFP_NOFS
);
8577 * Walk up the tree from the bottom, freeing leaves and any interior
8578 * nodes which have had all slots visited. If a node (leaf or
8579 * interior) is freed, the node above it will have it's slot
8580 * incremented. The root node will never be freed.
8582 * At the end of this function, we should have a path which has all
8583 * slots incremented to the next position for a search. If we need to
8584 * read a new node it will be NULL and the node above it will have the
8585 * correct slot selected for a later read.
8587 * If we increment the root nodes slot counter past the number of
8588 * elements, 1 is returned to signal completion of the search.
8590 static int adjust_slots_upwards(struct btrfs_root
*root
,
8591 struct btrfs_path
*path
, int root_level
)
8595 struct extent_buffer
*eb
;
8597 if (root_level
== 0)
8600 while (level
<= root_level
) {
8601 eb
= path
->nodes
[level
];
8602 nr
= btrfs_header_nritems(eb
);
8603 path
->slots
[level
]++;
8604 slot
= path
->slots
[level
];
8605 if (slot
>= nr
|| level
== 0) {
8607 * Don't free the root - we will detect this
8608 * condition after our loop and return a
8609 * positive value for caller to stop walking the tree.
8611 if (level
!= root_level
) {
8612 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8613 path
->locks
[level
] = 0;
8615 free_extent_buffer(eb
);
8616 path
->nodes
[level
] = NULL
;
8617 path
->slots
[level
] = 0;
8621 * We have a valid slot to walk back down
8622 * from. Stop here so caller can process these
8631 eb
= path
->nodes
[root_level
];
8632 if (path
->slots
[root_level
] >= btrfs_header_nritems(eb
))
8639 * root_eb is the subtree root and is locked before this function is called.
8641 static int account_shared_subtree(struct btrfs_trans_handle
*trans
,
8642 struct btrfs_root
*root
,
8643 struct extent_buffer
*root_eb
,
8649 struct extent_buffer
*eb
= root_eb
;
8650 struct btrfs_path
*path
= NULL
;
8652 BUG_ON(root_level
< 0 || root_level
> BTRFS_MAX_LEVEL
);
8653 BUG_ON(root_eb
== NULL
);
8655 if (!test_bit(BTRFS_FS_QUOTA_ENABLED
, &root
->fs_info
->flags
))
8658 if (!extent_buffer_uptodate(root_eb
)) {
8659 ret
= btrfs_read_buffer(root_eb
, root_gen
);
8664 if (root_level
== 0) {
8665 ret
= account_leaf_items(trans
, root
, root_eb
);
8669 path
= btrfs_alloc_path();
8674 * Walk down the tree. Missing extent blocks are filled in as
8675 * we go. Metadata is accounted every time we read a new
8678 * When we reach a leaf, we account for file extent items in it,
8679 * walk back up the tree (adjusting slot pointers as we go)
8680 * and restart the search process.
8682 extent_buffer_get(root_eb
); /* For path */
8683 path
->nodes
[root_level
] = root_eb
;
8684 path
->slots
[root_level
] = 0;
8685 path
->locks
[root_level
] = 0; /* so release_path doesn't try to unlock */
8688 while (level
>= 0) {
8689 if (path
->nodes
[level
] == NULL
) {
8694 /* We need to get child blockptr/gen from
8695 * parent before we can read it. */
8696 eb
= path
->nodes
[level
+ 1];
8697 parent_slot
= path
->slots
[level
+ 1];
8698 child_bytenr
= btrfs_node_blockptr(eb
, parent_slot
);
8699 child_gen
= btrfs_node_ptr_generation(eb
, parent_slot
);
8701 eb
= read_tree_block(root
, child_bytenr
, child_gen
);
8705 } else if (!extent_buffer_uptodate(eb
)) {
8706 free_extent_buffer(eb
);
8711 path
->nodes
[level
] = eb
;
8712 path
->slots
[level
] = 0;
8714 btrfs_tree_read_lock(eb
);
8715 btrfs_set_lock_blocking_rw(eb
, BTRFS_READ_LOCK
);
8716 path
->locks
[level
] = BTRFS_READ_LOCK_BLOCKING
;
8718 ret
= btrfs_qgroup_insert_dirty_extent(trans
,
8719 root
->fs_info
, child_bytenr
,
8720 root
->nodesize
, GFP_NOFS
);
8726 ret
= account_leaf_items(trans
, root
, path
->nodes
[level
]);
8730 /* Nonzero return here means we completed our search */
8731 ret
= adjust_slots_upwards(root
, path
, root_level
);
8735 /* Restart search with new slots */
8744 btrfs_free_path(path
);
8750 * helper to process tree block while walking down the tree.
8752 * when wc->stage == UPDATE_BACKREF, this function updates
8753 * back refs for pointers in the block.
8755 * NOTE: return value 1 means we should stop walking down.
8757 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
8758 struct btrfs_root
*root
,
8759 struct btrfs_path
*path
,
8760 struct walk_control
*wc
, int lookup_info
)
8762 int level
= wc
->level
;
8763 struct extent_buffer
*eb
= path
->nodes
[level
];
8764 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
8767 if (wc
->stage
== UPDATE_BACKREF
&&
8768 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
8772 * when reference count of tree block is 1, it won't increase
8773 * again. once full backref flag is set, we never clear it.
8776 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
8777 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
8778 BUG_ON(!path
->locks
[level
]);
8779 ret
= btrfs_lookup_extent_info(trans
, root
,
8780 eb
->start
, level
, 1,
8783 BUG_ON(ret
== -ENOMEM
);
8786 BUG_ON(wc
->refs
[level
] == 0);
8789 if (wc
->stage
== DROP_REFERENCE
) {
8790 if (wc
->refs
[level
] > 1)
8793 if (path
->locks
[level
] && !wc
->keep_locks
) {
8794 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8795 path
->locks
[level
] = 0;
8800 /* wc->stage == UPDATE_BACKREF */
8801 if (!(wc
->flags
[level
] & flag
)) {
8802 BUG_ON(!path
->locks
[level
]);
8803 ret
= btrfs_inc_ref(trans
, root
, eb
, 1);
8804 BUG_ON(ret
); /* -ENOMEM */
8805 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
8806 BUG_ON(ret
); /* -ENOMEM */
8807 ret
= btrfs_set_disk_extent_flags(trans
, root
, eb
->start
,
8809 btrfs_header_level(eb
), 0);
8810 BUG_ON(ret
); /* -ENOMEM */
8811 wc
->flags
[level
] |= flag
;
8815 * the block is shared by multiple trees, so it's not good to
8816 * keep the tree lock
8818 if (path
->locks
[level
] && level
> 0) {
8819 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8820 path
->locks
[level
] = 0;
8826 * helper to process tree block pointer.
8828 * when wc->stage == DROP_REFERENCE, this function checks
8829 * reference count of the block pointed to. if the block
8830 * is shared and we need update back refs for the subtree
8831 * rooted at the block, this function changes wc->stage to
8832 * UPDATE_BACKREF. if the block is shared and there is no
8833 * need to update back, this function drops the reference
8836 * NOTE: return value 1 means we should stop walking down.
8838 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
8839 struct btrfs_root
*root
,
8840 struct btrfs_path
*path
,
8841 struct walk_control
*wc
, int *lookup_info
)
8847 struct btrfs_key key
;
8848 struct extent_buffer
*next
;
8849 int level
= wc
->level
;
8852 bool need_account
= false;
8854 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
8855 path
->slots
[level
]);
8857 * if the lower level block was created before the snapshot
8858 * was created, we know there is no need to update back refs
8861 if (wc
->stage
== UPDATE_BACKREF
&&
8862 generation
<= root
->root_key
.offset
) {
8867 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
8868 blocksize
= root
->nodesize
;
8870 next
= btrfs_find_tree_block(root
->fs_info
, bytenr
);
8872 next
= btrfs_find_create_tree_block(root
, bytenr
);
8874 return PTR_ERR(next
);
8876 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, next
,
8880 btrfs_tree_lock(next
);
8881 btrfs_set_lock_blocking(next
);
8883 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, level
- 1, 1,
8884 &wc
->refs
[level
- 1],
8885 &wc
->flags
[level
- 1]);
8887 btrfs_tree_unlock(next
);
8891 if (unlikely(wc
->refs
[level
- 1] == 0)) {
8892 btrfs_err(root
->fs_info
, "Missing references.");
8897 if (wc
->stage
== DROP_REFERENCE
) {
8898 if (wc
->refs
[level
- 1] > 1) {
8899 need_account
= true;
8901 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8904 if (!wc
->update_ref
||
8905 generation
<= root
->root_key
.offset
)
8908 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
8909 path
->slots
[level
]);
8910 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
8914 wc
->stage
= UPDATE_BACKREF
;
8915 wc
->shared_level
= level
- 1;
8919 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8923 if (!btrfs_buffer_uptodate(next
, generation
, 0)) {
8924 btrfs_tree_unlock(next
);
8925 free_extent_buffer(next
);
8931 if (reada
&& level
== 1)
8932 reada_walk_down(trans
, root
, wc
, path
);
8933 next
= read_tree_block(root
, bytenr
, generation
);
8935 return PTR_ERR(next
);
8936 } else if (!extent_buffer_uptodate(next
)) {
8937 free_extent_buffer(next
);
8940 btrfs_tree_lock(next
);
8941 btrfs_set_lock_blocking(next
);
8945 BUG_ON(level
!= btrfs_header_level(next
));
8946 path
->nodes
[level
] = next
;
8947 path
->slots
[level
] = 0;
8948 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8954 wc
->refs
[level
- 1] = 0;
8955 wc
->flags
[level
- 1] = 0;
8956 if (wc
->stage
== DROP_REFERENCE
) {
8957 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
8958 parent
= path
->nodes
[level
]->start
;
8960 BUG_ON(root
->root_key
.objectid
!=
8961 btrfs_header_owner(path
->nodes
[level
]));
8966 ret
= account_shared_subtree(trans
, root
, next
,
8967 generation
, level
- 1);
8969 btrfs_err_rl(root
->fs_info
,
8971 "%d accounting shared subtree. Quota "
8972 "is out of sync, rescan required.",
8976 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
, parent
,
8977 root
->root_key
.objectid
, level
- 1, 0);
8978 BUG_ON(ret
); /* -ENOMEM */
8980 btrfs_tree_unlock(next
);
8981 free_extent_buffer(next
);
8987 * helper to process tree block while walking up the tree.
8989 * when wc->stage == DROP_REFERENCE, this function drops
8990 * reference count on the block.
8992 * when wc->stage == UPDATE_BACKREF, this function changes
8993 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8994 * to UPDATE_BACKREF previously while processing the block.
8996 * NOTE: return value 1 means we should stop walking up.
8998 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
8999 struct btrfs_root
*root
,
9000 struct btrfs_path
*path
,
9001 struct walk_control
*wc
)
9004 int level
= wc
->level
;
9005 struct extent_buffer
*eb
= path
->nodes
[level
];
9008 if (wc
->stage
== UPDATE_BACKREF
) {
9009 BUG_ON(wc
->shared_level
< level
);
9010 if (level
< wc
->shared_level
)
9013 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
9017 wc
->stage
= DROP_REFERENCE
;
9018 wc
->shared_level
= -1;
9019 path
->slots
[level
] = 0;
9022 * check reference count again if the block isn't locked.
9023 * we should start walking down the tree again if reference
9026 if (!path
->locks
[level
]) {
9028 btrfs_tree_lock(eb
);
9029 btrfs_set_lock_blocking(eb
);
9030 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
9032 ret
= btrfs_lookup_extent_info(trans
, root
,
9033 eb
->start
, level
, 1,
9037 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
9038 path
->locks
[level
] = 0;
9041 BUG_ON(wc
->refs
[level
] == 0);
9042 if (wc
->refs
[level
] == 1) {
9043 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
9044 path
->locks
[level
] = 0;
9050 /* wc->stage == DROP_REFERENCE */
9051 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
9053 if (wc
->refs
[level
] == 1) {
9055 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
9056 ret
= btrfs_dec_ref(trans
, root
, eb
, 1);
9058 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
9059 BUG_ON(ret
); /* -ENOMEM */
9060 ret
= account_leaf_items(trans
, root
, eb
);
9062 btrfs_err_rl(root
->fs_info
,
9064 "%d accounting leaf items. Quota "
9065 "is out of sync, rescan required.",
9069 /* make block locked assertion in clean_tree_block happy */
9070 if (!path
->locks
[level
] &&
9071 btrfs_header_generation(eb
) == trans
->transid
) {
9072 btrfs_tree_lock(eb
);
9073 btrfs_set_lock_blocking(eb
);
9074 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
9076 clean_tree_block(trans
, root
->fs_info
, eb
);
9079 if (eb
== root
->node
) {
9080 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
9083 BUG_ON(root
->root_key
.objectid
!=
9084 btrfs_header_owner(eb
));
9086 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
9087 parent
= path
->nodes
[level
+ 1]->start
;
9089 BUG_ON(root
->root_key
.objectid
!=
9090 btrfs_header_owner(path
->nodes
[level
+ 1]));
9093 btrfs_free_tree_block(trans
, root
, eb
, parent
, wc
->refs
[level
] == 1);
9095 wc
->refs
[level
] = 0;
9096 wc
->flags
[level
] = 0;
9100 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
9101 struct btrfs_root
*root
,
9102 struct btrfs_path
*path
,
9103 struct walk_control
*wc
)
9105 int level
= wc
->level
;
9106 int lookup_info
= 1;
9109 while (level
>= 0) {
9110 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
9117 if (path
->slots
[level
] >=
9118 btrfs_header_nritems(path
->nodes
[level
]))
9121 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
9123 path
->slots
[level
]++;
9132 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
9133 struct btrfs_root
*root
,
9134 struct btrfs_path
*path
,
9135 struct walk_control
*wc
, int max_level
)
9137 int level
= wc
->level
;
9140 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
9141 while (level
< max_level
&& path
->nodes
[level
]) {
9143 if (path
->slots
[level
] + 1 <
9144 btrfs_header_nritems(path
->nodes
[level
])) {
9145 path
->slots
[level
]++;
9148 ret
= walk_up_proc(trans
, root
, path
, wc
);
9152 if (path
->locks
[level
]) {
9153 btrfs_tree_unlock_rw(path
->nodes
[level
],
9154 path
->locks
[level
]);
9155 path
->locks
[level
] = 0;
9157 free_extent_buffer(path
->nodes
[level
]);
9158 path
->nodes
[level
] = NULL
;
9166 * drop a subvolume tree.
9168 * this function traverses the tree freeing any blocks that only
9169 * referenced by the tree.
9171 * when a shared tree block is found. this function decreases its
9172 * reference count by one. if update_ref is true, this function
9173 * also make sure backrefs for the shared block and all lower level
9174 * blocks are properly updated.
9176 * If called with for_reloc == 0, may exit early with -EAGAIN
9178 int btrfs_drop_snapshot(struct btrfs_root
*root
,
9179 struct btrfs_block_rsv
*block_rsv
, int update_ref
,
9182 struct btrfs_path
*path
;
9183 struct btrfs_trans_handle
*trans
;
9184 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
9185 struct btrfs_root_item
*root_item
= &root
->root_item
;
9186 struct walk_control
*wc
;
9187 struct btrfs_key key
;
9191 bool root_dropped
= false;
9193 btrfs_debug(root
->fs_info
, "Drop subvolume %llu", root
->objectid
);
9195 path
= btrfs_alloc_path();
9201 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
9203 btrfs_free_path(path
);
9208 trans
= btrfs_start_transaction(tree_root
, 0);
9209 if (IS_ERR(trans
)) {
9210 err
= PTR_ERR(trans
);
9215 trans
->block_rsv
= block_rsv
;
9217 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
9218 level
= btrfs_header_level(root
->node
);
9219 path
->nodes
[level
] = btrfs_lock_root_node(root
);
9220 btrfs_set_lock_blocking(path
->nodes
[level
]);
9221 path
->slots
[level
] = 0;
9222 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
9223 memset(&wc
->update_progress
, 0,
9224 sizeof(wc
->update_progress
));
9226 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
9227 memcpy(&wc
->update_progress
, &key
,
9228 sizeof(wc
->update_progress
));
9230 level
= root_item
->drop_level
;
9232 path
->lowest_level
= level
;
9233 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
9234 path
->lowest_level
= 0;
9242 * unlock our path, this is safe because only this
9243 * function is allowed to delete this snapshot
9245 btrfs_unlock_up_safe(path
, 0);
9247 level
= btrfs_header_level(root
->node
);
9249 btrfs_tree_lock(path
->nodes
[level
]);
9250 btrfs_set_lock_blocking(path
->nodes
[level
]);
9251 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
9253 ret
= btrfs_lookup_extent_info(trans
, root
,
9254 path
->nodes
[level
]->start
,
9255 level
, 1, &wc
->refs
[level
],
9261 BUG_ON(wc
->refs
[level
] == 0);
9263 if (level
== root_item
->drop_level
)
9266 btrfs_tree_unlock(path
->nodes
[level
]);
9267 path
->locks
[level
] = 0;
9268 WARN_ON(wc
->refs
[level
] != 1);
9274 wc
->shared_level
= -1;
9275 wc
->stage
= DROP_REFERENCE
;
9276 wc
->update_ref
= update_ref
;
9278 wc
->for_reloc
= for_reloc
;
9279 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
9283 ret
= walk_down_tree(trans
, root
, path
, wc
);
9289 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
9296 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
9300 if (wc
->stage
== DROP_REFERENCE
) {
9302 btrfs_node_key(path
->nodes
[level
],
9303 &root_item
->drop_progress
,
9304 path
->slots
[level
]);
9305 root_item
->drop_level
= level
;
9308 BUG_ON(wc
->level
== 0);
9309 if (btrfs_should_end_transaction(trans
, tree_root
) ||
9310 (!for_reloc
&& btrfs_need_cleaner_sleep(root
))) {
9311 ret
= btrfs_update_root(trans
, tree_root
,
9315 btrfs_abort_transaction(trans
, ret
);
9320 btrfs_end_transaction_throttle(trans
, tree_root
);
9321 if (!for_reloc
&& btrfs_need_cleaner_sleep(root
)) {
9322 pr_debug("BTRFS: drop snapshot early exit\n");
9327 trans
= btrfs_start_transaction(tree_root
, 0);
9328 if (IS_ERR(trans
)) {
9329 err
= PTR_ERR(trans
);
9333 trans
->block_rsv
= block_rsv
;
9336 btrfs_release_path(path
);
9340 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
9342 btrfs_abort_transaction(trans
, ret
);
9346 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
9347 ret
= btrfs_find_root(tree_root
, &root
->root_key
, path
,
9350 btrfs_abort_transaction(trans
, ret
);
9353 } else if (ret
> 0) {
9354 /* if we fail to delete the orphan item this time
9355 * around, it'll get picked up the next time.
9357 * The most common failure here is just -ENOENT.
9359 btrfs_del_orphan_item(trans
, tree_root
,
9360 root
->root_key
.objectid
);
9364 if (test_bit(BTRFS_ROOT_IN_RADIX
, &root
->state
)) {
9365 btrfs_add_dropped_root(trans
, root
);
9367 free_extent_buffer(root
->node
);
9368 free_extent_buffer(root
->commit_root
);
9369 btrfs_put_fs_root(root
);
9371 root_dropped
= true;
9373 btrfs_end_transaction_throttle(trans
, tree_root
);
9376 btrfs_free_path(path
);
9379 * So if we need to stop dropping the snapshot for whatever reason we
9380 * need to make sure to add it back to the dead root list so that we
9381 * keep trying to do the work later. This also cleans up roots if we
9382 * don't have it in the radix (like when we recover after a power fail
9383 * or unmount) so we don't leak memory.
9385 if (!for_reloc
&& root_dropped
== false)
9386 btrfs_add_dead_root(root
);
9387 if (err
&& err
!= -EAGAIN
)
9388 btrfs_handle_fs_error(root
->fs_info
, err
, NULL
);
9393 * drop subtree rooted at tree block 'node'.
9395 * NOTE: this function will unlock and release tree block 'node'
9396 * only used by relocation code
9398 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
9399 struct btrfs_root
*root
,
9400 struct extent_buffer
*node
,
9401 struct extent_buffer
*parent
)
9403 struct btrfs_path
*path
;
9404 struct walk_control
*wc
;
9410 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
9412 path
= btrfs_alloc_path();
9416 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
9418 btrfs_free_path(path
);
9422 btrfs_assert_tree_locked(parent
);
9423 parent_level
= btrfs_header_level(parent
);
9424 extent_buffer_get(parent
);
9425 path
->nodes
[parent_level
] = parent
;
9426 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
9428 btrfs_assert_tree_locked(node
);
9429 level
= btrfs_header_level(node
);
9430 path
->nodes
[level
] = node
;
9431 path
->slots
[level
] = 0;
9432 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
9434 wc
->refs
[parent_level
] = 1;
9435 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
9437 wc
->shared_level
= -1;
9438 wc
->stage
= DROP_REFERENCE
;
9442 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
9445 wret
= walk_down_tree(trans
, root
, path
, wc
);
9451 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
9459 btrfs_free_path(path
);
9463 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
9469 * if restripe for this chunk_type is on pick target profile and
9470 * return, otherwise do the usual balance
9472 stripped
= get_restripe_target(root
->fs_info
, flags
);
9474 return extended_to_chunk(stripped
);
9476 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
9478 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
9479 BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
|
9480 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
9482 if (num_devices
== 1) {
9483 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
9484 stripped
= flags
& ~stripped
;
9486 /* turn raid0 into single device chunks */
9487 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
9490 /* turn mirroring into duplication */
9491 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
9492 BTRFS_BLOCK_GROUP_RAID10
))
9493 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
9495 /* they already had raid on here, just return */
9496 if (flags
& stripped
)
9499 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
9500 stripped
= flags
& ~stripped
;
9502 /* switch duplicated blocks with raid1 */
9503 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
9504 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
9506 /* this is drive concat, leave it alone */
9512 static int inc_block_group_ro(struct btrfs_block_group_cache
*cache
, int force
)
9514 struct btrfs_space_info
*sinfo
= cache
->space_info
;
9516 u64 min_allocable_bytes
;
9520 * We need some metadata space and system metadata space for
9521 * allocating chunks in some corner cases until we force to set
9522 * it to be readonly.
9525 (BTRFS_BLOCK_GROUP_SYSTEM
| BTRFS_BLOCK_GROUP_METADATA
)) &&
9527 min_allocable_bytes
= SZ_1M
;
9529 min_allocable_bytes
= 0;
9531 spin_lock(&sinfo
->lock
);
9532 spin_lock(&cache
->lock
);
9540 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
9541 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
9543 if (sinfo
->bytes_used
+ sinfo
->bytes_reserved
+ sinfo
->bytes_pinned
+
9544 sinfo
->bytes_may_use
+ sinfo
->bytes_readonly
+ num_bytes
+
9545 min_allocable_bytes
<= sinfo
->total_bytes
) {
9546 sinfo
->bytes_readonly
+= num_bytes
;
9548 list_add_tail(&cache
->ro_list
, &sinfo
->ro_bgs
);
9552 spin_unlock(&cache
->lock
);
9553 spin_unlock(&sinfo
->lock
);
9557 int btrfs_inc_block_group_ro(struct btrfs_root
*root
,
9558 struct btrfs_block_group_cache
*cache
)
9561 struct btrfs_trans_handle
*trans
;
9566 trans
= btrfs_join_transaction(root
);
9568 return PTR_ERR(trans
);
9571 * we're not allowed to set block groups readonly after the dirty
9572 * block groups cache has started writing. If it already started,
9573 * back off and let this transaction commit
9575 mutex_lock(&root
->fs_info
->ro_block_group_mutex
);
9576 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN
, &trans
->transaction
->flags
)) {
9577 u64 transid
= trans
->transid
;
9579 mutex_unlock(&root
->fs_info
->ro_block_group_mutex
);
9580 btrfs_end_transaction(trans
, root
);
9582 ret
= btrfs_wait_for_commit(root
, transid
);
9589 * if we are changing raid levels, try to allocate a corresponding
9590 * block group with the new raid level.
9592 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
9593 if (alloc_flags
!= cache
->flags
) {
9594 ret
= do_chunk_alloc(trans
, root
, alloc_flags
,
9597 * ENOSPC is allowed here, we may have enough space
9598 * already allocated at the new raid level to
9607 ret
= inc_block_group_ro(cache
, 0);
9610 alloc_flags
= get_alloc_profile(root
, cache
->space_info
->flags
);
9611 ret
= do_chunk_alloc(trans
, root
, alloc_flags
,
9615 ret
= inc_block_group_ro(cache
, 0);
9617 if (cache
->flags
& BTRFS_BLOCK_GROUP_SYSTEM
) {
9618 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
9619 lock_chunks(root
->fs_info
->chunk_root
);
9620 check_system_chunk(trans
, root
, alloc_flags
);
9621 unlock_chunks(root
->fs_info
->chunk_root
);
9623 mutex_unlock(&root
->fs_info
->ro_block_group_mutex
);
9625 btrfs_end_transaction(trans
, root
);
9629 int btrfs_force_chunk_alloc(struct btrfs_trans_handle
*trans
,
9630 struct btrfs_root
*root
, u64 type
)
9632 u64 alloc_flags
= get_alloc_profile(root
, type
);
9633 return do_chunk_alloc(trans
, root
, alloc_flags
,
9638 * helper to account the unused space of all the readonly block group in the
9639 * space_info. takes mirrors into account.
9641 u64
btrfs_account_ro_block_groups_free_space(struct btrfs_space_info
*sinfo
)
9643 struct btrfs_block_group_cache
*block_group
;
9647 /* It's df, we don't care if it's racy */
9648 if (list_empty(&sinfo
->ro_bgs
))
9651 spin_lock(&sinfo
->lock
);
9652 list_for_each_entry(block_group
, &sinfo
->ro_bgs
, ro_list
) {
9653 spin_lock(&block_group
->lock
);
9655 if (!block_group
->ro
) {
9656 spin_unlock(&block_group
->lock
);
9660 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_RAID1
|
9661 BTRFS_BLOCK_GROUP_RAID10
|
9662 BTRFS_BLOCK_GROUP_DUP
))
9667 free_bytes
+= (block_group
->key
.offset
-
9668 btrfs_block_group_used(&block_group
->item
)) *
9671 spin_unlock(&block_group
->lock
);
9673 spin_unlock(&sinfo
->lock
);
9678 void btrfs_dec_block_group_ro(struct btrfs_root
*root
,
9679 struct btrfs_block_group_cache
*cache
)
9681 struct btrfs_space_info
*sinfo
= cache
->space_info
;
9686 spin_lock(&sinfo
->lock
);
9687 spin_lock(&cache
->lock
);
9689 num_bytes
= cache
->key
.offset
- cache
->reserved
-
9690 cache
->pinned
- cache
->bytes_super
-
9691 btrfs_block_group_used(&cache
->item
);
9692 sinfo
->bytes_readonly
-= num_bytes
;
9693 list_del_init(&cache
->ro_list
);
9695 spin_unlock(&cache
->lock
);
9696 spin_unlock(&sinfo
->lock
);
9700 * checks to see if its even possible to relocate this block group.
9702 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9703 * ok to go ahead and try.
9705 int btrfs_can_relocate(struct btrfs_root
*root
, u64 bytenr
)
9707 struct btrfs_block_group_cache
*block_group
;
9708 struct btrfs_space_info
*space_info
;
9709 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
9710 struct btrfs_device
*device
;
9711 struct btrfs_trans_handle
*trans
;
9721 debug
= btrfs_test_opt(root
->fs_info
, ENOSPC_DEBUG
);
9723 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
9725 /* odd, couldn't find the block group, leave it alone */
9728 btrfs_warn(root
->fs_info
,
9729 "can't find block group for bytenr %llu",
9734 min_free
= btrfs_block_group_used(&block_group
->item
);
9736 /* no bytes used, we're good */
9740 space_info
= block_group
->space_info
;
9741 spin_lock(&space_info
->lock
);
9743 full
= space_info
->full
;
9746 * if this is the last block group we have in this space, we can't
9747 * relocate it unless we're able to allocate a new chunk below.
9749 * Otherwise, we need to make sure we have room in the space to handle
9750 * all of the extents from this block group. If we can, we're good
9752 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
9753 (space_info
->bytes_used
+ space_info
->bytes_reserved
+
9754 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
9755 min_free
< space_info
->total_bytes
)) {
9756 spin_unlock(&space_info
->lock
);
9759 spin_unlock(&space_info
->lock
);
9762 * ok we don't have enough space, but maybe we have free space on our
9763 * devices to allocate new chunks for relocation, so loop through our
9764 * alloc devices and guess if we have enough space. if this block
9765 * group is going to be restriped, run checks against the target
9766 * profile instead of the current one.
9778 target
= get_restripe_target(root
->fs_info
, block_group
->flags
);
9780 index
= __get_raid_index(extended_to_chunk(target
));
9783 * this is just a balance, so if we were marked as full
9784 * we know there is no space for a new chunk
9788 btrfs_warn(root
->fs_info
,
9789 "no space to alloc new chunk for block group %llu",
9790 block_group
->key
.objectid
);
9794 index
= get_block_group_index(block_group
);
9797 if (index
== BTRFS_RAID_RAID10
) {
9801 } else if (index
== BTRFS_RAID_RAID1
) {
9803 } else if (index
== BTRFS_RAID_DUP
) {
9806 } else if (index
== BTRFS_RAID_RAID0
) {
9807 dev_min
= fs_devices
->rw_devices
;
9808 min_free
= div64_u64(min_free
, dev_min
);
9811 /* We need to do this so that we can look at pending chunks */
9812 trans
= btrfs_join_transaction(root
);
9813 if (IS_ERR(trans
)) {
9814 ret
= PTR_ERR(trans
);
9818 mutex_lock(&root
->fs_info
->chunk_mutex
);
9819 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
9823 * check to make sure we can actually find a chunk with enough
9824 * space to fit our block group in.
9826 if (device
->total_bytes
> device
->bytes_used
+ min_free
&&
9827 !device
->is_tgtdev_for_dev_replace
) {
9828 ret
= find_free_dev_extent(trans
, device
, min_free
,
9833 if (dev_nr
>= dev_min
)
9839 if (debug
&& ret
== -1)
9840 btrfs_warn(root
->fs_info
,
9841 "no space to allocate a new chunk for block group %llu",
9842 block_group
->key
.objectid
);
9843 mutex_unlock(&root
->fs_info
->chunk_mutex
);
9844 btrfs_end_transaction(trans
, root
);
9846 btrfs_put_block_group(block_group
);
9850 static int find_first_block_group(struct btrfs_root
*root
,
9851 struct btrfs_path
*path
, struct btrfs_key
*key
)
9854 struct btrfs_key found_key
;
9855 struct extent_buffer
*leaf
;
9858 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
9863 slot
= path
->slots
[0];
9864 leaf
= path
->nodes
[0];
9865 if (slot
>= btrfs_header_nritems(leaf
)) {
9866 ret
= btrfs_next_leaf(root
, path
);
9873 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
9875 if (found_key
.objectid
>= key
->objectid
&&
9876 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
9877 struct extent_map_tree
*em_tree
;
9878 struct extent_map
*em
;
9880 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
9881 read_lock(&em_tree
->lock
);
9882 em
= lookup_extent_mapping(em_tree
, found_key
.objectid
,
9884 read_unlock(&em_tree
->lock
);
9886 btrfs_err(root
->fs_info
,
9887 "logical %llu len %llu found bg but no related chunk",
9888 found_key
.objectid
, found_key
.offset
);
9893 free_extent_map(em
);
9902 void btrfs_put_block_group_cache(struct btrfs_fs_info
*info
)
9904 struct btrfs_block_group_cache
*block_group
;
9908 struct inode
*inode
;
9910 block_group
= btrfs_lookup_first_block_group(info
, last
);
9911 while (block_group
) {
9912 spin_lock(&block_group
->lock
);
9913 if (block_group
->iref
)
9915 spin_unlock(&block_group
->lock
);
9916 block_group
= next_block_group(info
->tree_root
,
9926 inode
= block_group
->inode
;
9927 block_group
->iref
= 0;
9928 block_group
->inode
= NULL
;
9929 spin_unlock(&block_group
->lock
);
9930 ASSERT(block_group
->io_ctl
.inode
== NULL
);
9932 last
= block_group
->key
.objectid
+ block_group
->key
.offset
;
9933 btrfs_put_block_group(block_group
);
9937 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
9939 struct btrfs_block_group_cache
*block_group
;
9940 struct btrfs_space_info
*space_info
;
9941 struct btrfs_caching_control
*caching_ctl
;
9944 down_write(&info
->commit_root_sem
);
9945 while (!list_empty(&info
->caching_block_groups
)) {
9946 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
9947 struct btrfs_caching_control
, list
);
9948 list_del(&caching_ctl
->list
);
9949 put_caching_control(caching_ctl
);
9951 up_write(&info
->commit_root_sem
);
9953 spin_lock(&info
->unused_bgs_lock
);
9954 while (!list_empty(&info
->unused_bgs
)) {
9955 block_group
= list_first_entry(&info
->unused_bgs
,
9956 struct btrfs_block_group_cache
,
9958 list_del_init(&block_group
->bg_list
);
9959 btrfs_put_block_group(block_group
);
9961 spin_unlock(&info
->unused_bgs_lock
);
9963 spin_lock(&info
->block_group_cache_lock
);
9964 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
9965 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
9967 rb_erase(&block_group
->cache_node
,
9968 &info
->block_group_cache_tree
);
9969 RB_CLEAR_NODE(&block_group
->cache_node
);
9970 spin_unlock(&info
->block_group_cache_lock
);
9972 down_write(&block_group
->space_info
->groups_sem
);
9973 list_del(&block_group
->list
);
9974 up_write(&block_group
->space_info
->groups_sem
);
9976 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
9977 wait_block_group_cache_done(block_group
);
9980 * We haven't cached this block group, which means we could
9981 * possibly have excluded extents on this block group.
9983 if (block_group
->cached
== BTRFS_CACHE_NO
||
9984 block_group
->cached
== BTRFS_CACHE_ERROR
)
9985 free_excluded_extents(info
->extent_root
, block_group
);
9987 btrfs_remove_free_space_cache(block_group
);
9988 ASSERT(list_empty(&block_group
->dirty_list
));
9989 ASSERT(list_empty(&block_group
->io_list
));
9990 ASSERT(list_empty(&block_group
->bg_list
));
9991 ASSERT(atomic_read(&block_group
->count
) == 1);
9992 btrfs_put_block_group(block_group
);
9994 spin_lock(&info
->block_group_cache_lock
);
9996 spin_unlock(&info
->block_group_cache_lock
);
9998 /* now that all the block groups are freed, go through and
9999 * free all the space_info structs. This is only called during
10000 * the final stages of unmount, and so we know nobody is
10001 * using them. We call synchronize_rcu() once before we start,
10002 * just to be on the safe side.
10006 release_global_block_rsv(info
);
10008 while (!list_empty(&info
->space_info
)) {
10011 space_info
= list_entry(info
->space_info
.next
,
10012 struct btrfs_space_info
,
10016 * Do not hide this behind enospc_debug, this is actually
10017 * important and indicates a real bug if this happens.
10019 if (WARN_ON(space_info
->bytes_pinned
> 0 ||
10020 space_info
->bytes_reserved
> 0 ||
10021 space_info
->bytes_may_use
> 0))
10022 dump_space_info(space_info
, 0, 0);
10023 list_del(&space_info
->list
);
10024 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++) {
10025 struct kobject
*kobj
;
10026 kobj
= space_info
->block_group_kobjs
[i
];
10027 space_info
->block_group_kobjs
[i
] = NULL
;
10033 kobject_del(&space_info
->kobj
);
10034 kobject_put(&space_info
->kobj
);
10039 static void __link_block_group(struct btrfs_space_info
*space_info
,
10040 struct btrfs_block_group_cache
*cache
)
10042 int index
= get_block_group_index(cache
);
10043 bool first
= false;
10045 down_write(&space_info
->groups_sem
);
10046 if (list_empty(&space_info
->block_groups
[index
]))
10048 list_add_tail(&cache
->list
, &space_info
->block_groups
[index
]);
10049 up_write(&space_info
->groups_sem
);
10052 struct raid_kobject
*rkobj
;
10055 rkobj
= kzalloc(sizeof(*rkobj
), GFP_NOFS
);
10058 rkobj
->raid_type
= index
;
10059 kobject_init(&rkobj
->kobj
, &btrfs_raid_ktype
);
10060 ret
= kobject_add(&rkobj
->kobj
, &space_info
->kobj
,
10061 "%s", get_raid_name(index
));
10063 kobject_put(&rkobj
->kobj
);
10066 space_info
->block_group_kobjs
[index
] = &rkobj
->kobj
;
10071 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
10074 static struct btrfs_block_group_cache
*
10075 btrfs_create_block_group_cache(struct btrfs_root
*root
, u64 start
, u64 size
)
10077 struct btrfs_block_group_cache
*cache
;
10079 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
10083 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
10085 if (!cache
->free_space_ctl
) {
10090 cache
->key
.objectid
= start
;
10091 cache
->key
.offset
= size
;
10092 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
10094 cache
->sectorsize
= root
->sectorsize
;
10095 cache
->fs_info
= root
->fs_info
;
10096 cache
->full_stripe_len
= btrfs_full_stripe_len(root
,
10097 &root
->fs_info
->mapping_tree
,
10099 set_free_space_tree_thresholds(cache
);
10101 atomic_set(&cache
->count
, 1);
10102 spin_lock_init(&cache
->lock
);
10103 init_rwsem(&cache
->data_rwsem
);
10104 INIT_LIST_HEAD(&cache
->list
);
10105 INIT_LIST_HEAD(&cache
->cluster_list
);
10106 INIT_LIST_HEAD(&cache
->bg_list
);
10107 INIT_LIST_HEAD(&cache
->ro_list
);
10108 INIT_LIST_HEAD(&cache
->dirty_list
);
10109 INIT_LIST_HEAD(&cache
->io_list
);
10110 btrfs_init_free_space_ctl(cache
);
10111 atomic_set(&cache
->trimming
, 0);
10112 mutex_init(&cache
->free_space_lock
);
10117 int btrfs_read_block_groups(struct btrfs_root
*root
)
10119 struct btrfs_path
*path
;
10121 struct btrfs_block_group_cache
*cache
;
10122 struct btrfs_fs_info
*info
= root
->fs_info
;
10123 struct btrfs_space_info
*space_info
;
10124 struct btrfs_key key
;
10125 struct btrfs_key found_key
;
10126 struct extent_buffer
*leaf
;
10127 int need_clear
= 0;
10132 feature
= btrfs_super_incompat_flags(info
->super_copy
);
10133 mixed
= !!(feature
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
);
10135 root
= info
->extent_root
;
10138 key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
10139 path
= btrfs_alloc_path();
10142 path
->reada
= READA_FORWARD
;
10144 cache_gen
= btrfs_super_cache_generation(root
->fs_info
->super_copy
);
10145 if (btrfs_test_opt(root
->fs_info
, SPACE_CACHE
) &&
10146 btrfs_super_generation(root
->fs_info
->super_copy
) != cache_gen
)
10148 if (btrfs_test_opt(root
->fs_info
, CLEAR_CACHE
))
10152 ret
= find_first_block_group(root
, path
, &key
);
10158 leaf
= path
->nodes
[0];
10159 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
10161 cache
= btrfs_create_block_group_cache(root
, found_key
.objectid
,
10170 * When we mount with old space cache, we need to
10171 * set BTRFS_DC_CLEAR and set dirty flag.
10173 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
10174 * truncate the old free space cache inode and
10176 * b) Setting 'dirty flag' makes sure that we flush
10177 * the new space cache info onto disk.
10179 if (btrfs_test_opt(root
->fs_info
, SPACE_CACHE
))
10180 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
10183 read_extent_buffer(leaf
, &cache
->item
,
10184 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
10185 sizeof(cache
->item
));
10186 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
10188 ((cache
->flags
& BTRFS_BLOCK_GROUP_METADATA
) &&
10189 (cache
->flags
& BTRFS_BLOCK_GROUP_DATA
))) {
10191 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
10192 cache
->key
.objectid
);
10197 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
10198 btrfs_release_path(path
);
10201 * We need to exclude the super stripes now so that the space
10202 * info has super bytes accounted for, otherwise we'll think
10203 * we have more space than we actually do.
10205 ret
= exclude_super_stripes(root
, cache
);
10208 * We may have excluded something, so call this just in
10211 free_excluded_extents(root
, cache
);
10212 btrfs_put_block_group(cache
);
10217 * check for two cases, either we are full, and therefore
10218 * don't need to bother with the caching work since we won't
10219 * find any space, or we are empty, and we can just add all
10220 * the space in and be done with it. This saves us _alot_ of
10221 * time, particularly in the full case.
10223 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
10224 cache
->last_byte_to_unpin
= (u64
)-1;
10225 cache
->cached
= BTRFS_CACHE_FINISHED
;
10226 free_excluded_extents(root
, cache
);
10227 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
10228 cache
->last_byte_to_unpin
= (u64
)-1;
10229 cache
->cached
= BTRFS_CACHE_FINISHED
;
10230 add_new_free_space(cache
, root
->fs_info
,
10231 found_key
.objectid
,
10232 found_key
.objectid
+
10234 free_excluded_extents(root
, cache
);
10237 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
10239 btrfs_remove_free_space_cache(cache
);
10240 btrfs_put_block_group(cache
);
10244 trace_btrfs_add_block_group(root
->fs_info
, cache
, 0);
10245 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
10246 btrfs_block_group_used(&cache
->item
),
10247 cache
->bytes_super
, &space_info
);
10249 btrfs_remove_free_space_cache(cache
);
10250 spin_lock(&info
->block_group_cache_lock
);
10251 rb_erase(&cache
->cache_node
,
10252 &info
->block_group_cache_tree
);
10253 RB_CLEAR_NODE(&cache
->cache_node
);
10254 spin_unlock(&info
->block_group_cache_lock
);
10255 btrfs_put_block_group(cache
);
10259 cache
->space_info
= space_info
;
10261 __link_block_group(space_info
, cache
);
10263 set_avail_alloc_bits(root
->fs_info
, cache
->flags
);
10264 if (btrfs_chunk_readonly(root
, cache
->key
.objectid
)) {
10265 inc_block_group_ro(cache
, 1);
10266 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
10267 spin_lock(&info
->unused_bgs_lock
);
10268 /* Should always be true but just in case. */
10269 if (list_empty(&cache
->bg_list
)) {
10270 btrfs_get_block_group(cache
);
10271 list_add_tail(&cache
->bg_list
,
10272 &info
->unused_bgs
);
10274 spin_unlock(&info
->unused_bgs_lock
);
10278 list_for_each_entry_rcu(space_info
, &root
->fs_info
->space_info
, list
) {
10279 if (!(get_alloc_profile(root
, space_info
->flags
) &
10280 (BTRFS_BLOCK_GROUP_RAID10
|
10281 BTRFS_BLOCK_GROUP_RAID1
|
10282 BTRFS_BLOCK_GROUP_RAID5
|
10283 BTRFS_BLOCK_GROUP_RAID6
|
10284 BTRFS_BLOCK_GROUP_DUP
)))
10287 * avoid allocating from un-mirrored block group if there are
10288 * mirrored block groups.
10290 list_for_each_entry(cache
,
10291 &space_info
->block_groups
[BTRFS_RAID_RAID0
],
10293 inc_block_group_ro(cache
, 1);
10294 list_for_each_entry(cache
,
10295 &space_info
->block_groups
[BTRFS_RAID_SINGLE
],
10297 inc_block_group_ro(cache
, 1);
10300 init_global_block_rsv(info
);
10303 btrfs_free_path(path
);
10307 void btrfs_create_pending_block_groups(struct btrfs_trans_handle
*trans
,
10308 struct btrfs_root
*root
)
10310 struct btrfs_block_group_cache
*block_group
, *tmp
;
10311 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
10312 struct btrfs_block_group_item item
;
10313 struct btrfs_key key
;
10315 bool can_flush_pending_bgs
= trans
->can_flush_pending_bgs
;
10317 trans
->can_flush_pending_bgs
= false;
10318 list_for_each_entry_safe(block_group
, tmp
, &trans
->new_bgs
, bg_list
) {
10322 spin_lock(&block_group
->lock
);
10323 memcpy(&item
, &block_group
->item
, sizeof(item
));
10324 memcpy(&key
, &block_group
->key
, sizeof(key
));
10325 spin_unlock(&block_group
->lock
);
10327 ret
= btrfs_insert_item(trans
, extent_root
, &key
, &item
,
10330 btrfs_abort_transaction(trans
, ret
);
10331 ret
= btrfs_finish_chunk_alloc(trans
, extent_root
,
10332 key
.objectid
, key
.offset
);
10334 btrfs_abort_transaction(trans
, ret
);
10335 add_block_group_free_space(trans
, root
->fs_info
, block_group
);
10336 /* already aborted the transaction if it failed. */
10338 list_del_init(&block_group
->bg_list
);
10340 trans
->can_flush_pending_bgs
= can_flush_pending_bgs
;
10343 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
10344 struct btrfs_root
*root
, u64 bytes_used
,
10345 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
10349 struct btrfs_root
*extent_root
;
10350 struct btrfs_block_group_cache
*cache
;
10351 extent_root
= root
->fs_info
->extent_root
;
10353 btrfs_set_log_full_commit(root
->fs_info
, trans
);
10355 cache
= btrfs_create_block_group_cache(root
, chunk_offset
, size
);
10359 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
10360 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
10361 btrfs_set_block_group_flags(&cache
->item
, type
);
10363 cache
->flags
= type
;
10364 cache
->last_byte_to_unpin
= (u64
)-1;
10365 cache
->cached
= BTRFS_CACHE_FINISHED
;
10366 cache
->needs_free_space
= 1;
10367 ret
= exclude_super_stripes(root
, cache
);
10370 * We may have excluded something, so call this just in
10373 free_excluded_extents(root
, cache
);
10374 btrfs_put_block_group(cache
);
10378 add_new_free_space(cache
, root
->fs_info
, chunk_offset
,
10379 chunk_offset
+ size
);
10381 free_excluded_extents(root
, cache
);
10383 #ifdef CONFIG_BTRFS_DEBUG
10384 if (btrfs_should_fragment_free_space(root
, cache
)) {
10385 u64 new_bytes_used
= size
- bytes_used
;
10387 bytes_used
+= new_bytes_used
>> 1;
10388 fragment_free_space(root
, cache
);
10392 * Call to ensure the corresponding space_info object is created and
10393 * assigned to our block group, but don't update its counters just yet.
10394 * We want our bg to be added to the rbtree with its ->space_info set.
10396 ret
= update_space_info(root
->fs_info
, cache
->flags
, 0, 0, 0,
10397 &cache
->space_info
);
10399 btrfs_remove_free_space_cache(cache
);
10400 btrfs_put_block_group(cache
);
10404 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
10406 btrfs_remove_free_space_cache(cache
);
10407 btrfs_put_block_group(cache
);
10412 * Now that our block group has its ->space_info set and is inserted in
10413 * the rbtree, update the space info's counters.
10415 trace_btrfs_add_block_group(root
->fs_info
, cache
, 1);
10416 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
10417 cache
->bytes_super
, &cache
->space_info
);
10419 btrfs_remove_free_space_cache(cache
);
10420 spin_lock(&root
->fs_info
->block_group_cache_lock
);
10421 rb_erase(&cache
->cache_node
,
10422 &root
->fs_info
->block_group_cache_tree
);
10423 RB_CLEAR_NODE(&cache
->cache_node
);
10424 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
10425 btrfs_put_block_group(cache
);
10428 update_global_block_rsv(root
->fs_info
);
10430 __link_block_group(cache
->space_info
, cache
);
10432 list_add_tail(&cache
->bg_list
, &trans
->new_bgs
);
10434 set_avail_alloc_bits(extent_root
->fs_info
, type
);
10438 static void clear_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
10440 u64 extra_flags
= chunk_to_extended(flags
) &
10441 BTRFS_EXTENDED_PROFILE_MASK
;
10443 write_seqlock(&fs_info
->profiles_lock
);
10444 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
10445 fs_info
->avail_data_alloc_bits
&= ~extra_flags
;
10446 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
10447 fs_info
->avail_metadata_alloc_bits
&= ~extra_flags
;
10448 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
10449 fs_info
->avail_system_alloc_bits
&= ~extra_flags
;
10450 write_sequnlock(&fs_info
->profiles_lock
);
10453 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
10454 struct btrfs_root
*root
, u64 group_start
,
10455 struct extent_map
*em
)
10457 struct btrfs_path
*path
;
10458 struct btrfs_block_group_cache
*block_group
;
10459 struct btrfs_free_cluster
*cluster
;
10460 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
10461 struct btrfs_key key
;
10462 struct inode
*inode
;
10463 struct kobject
*kobj
= NULL
;
10467 struct btrfs_caching_control
*caching_ctl
= NULL
;
10470 root
= root
->fs_info
->extent_root
;
10472 block_group
= btrfs_lookup_block_group(root
->fs_info
, group_start
);
10473 BUG_ON(!block_group
);
10474 BUG_ON(!block_group
->ro
);
10477 * Free the reserved super bytes from this block group before
10480 free_excluded_extents(root
, block_group
);
10482 memcpy(&key
, &block_group
->key
, sizeof(key
));
10483 index
= get_block_group_index(block_group
);
10484 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
10485 BTRFS_BLOCK_GROUP_RAID1
|
10486 BTRFS_BLOCK_GROUP_RAID10
))
10491 /* make sure this block group isn't part of an allocation cluster */
10492 cluster
= &root
->fs_info
->data_alloc_cluster
;
10493 spin_lock(&cluster
->refill_lock
);
10494 btrfs_return_cluster_to_free_space(block_group
, cluster
);
10495 spin_unlock(&cluster
->refill_lock
);
10498 * make sure this block group isn't part of a metadata
10499 * allocation cluster
10501 cluster
= &root
->fs_info
->meta_alloc_cluster
;
10502 spin_lock(&cluster
->refill_lock
);
10503 btrfs_return_cluster_to_free_space(block_group
, cluster
);
10504 spin_unlock(&cluster
->refill_lock
);
10506 path
= btrfs_alloc_path();
10513 * get the inode first so any iput calls done for the io_list
10514 * aren't the final iput (no unlinks allowed now)
10516 inode
= lookup_free_space_inode(tree_root
, block_group
, path
);
10518 mutex_lock(&trans
->transaction
->cache_write_mutex
);
10520 * make sure our free spache cache IO is done before remove the
10523 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
10524 if (!list_empty(&block_group
->io_list
)) {
10525 list_del_init(&block_group
->io_list
);
10527 WARN_ON(!IS_ERR(inode
) && inode
!= block_group
->io_ctl
.inode
);
10529 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
10530 btrfs_wait_cache_io(root
, trans
, block_group
,
10531 &block_group
->io_ctl
, path
,
10532 block_group
->key
.objectid
);
10533 btrfs_put_block_group(block_group
);
10534 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
10537 if (!list_empty(&block_group
->dirty_list
)) {
10538 list_del_init(&block_group
->dirty_list
);
10539 btrfs_put_block_group(block_group
);
10541 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
10542 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
10544 if (!IS_ERR(inode
)) {
10545 ret
= btrfs_orphan_add(trans
, inode
);
10547 btrfs_add_delayed_iput(inode
);
10550 clear_nlink(inode
);
10551 /* One for the block groups ref */
10552 spin_lock(&block_group
->lock
);
10553 if (block_group
->iref
) {
10554 block_group
->iref
= 0;
10555 block_group
->inode
= NULL
;
10556 spin_unlock(&block_group
->lock
);
10559 spin_unlock(&block_group
->lock
);
10561 /* One for our lookup ref */
10562 btrfs_add_delayed_iput(inode
);
10565 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
10566 key
.offset
= block_group
->key
.objectid
;
10569 ret
= btrfs_search_slot(trans
, tree_root
, &key
, path
, -1, 1);
10573 btrfs_release_path(path
);
10575 ret
= btrfs_del_item(trans
, tree_root
, path
);
10578 btrfs_release_path(path
);
10581 spin_lock(&root
->fs_info
->block_group_cache_lock
);
10582 rb_erase(&block_group
->cache_node
,
10583 &root
->fs_info
->block_group_cache_tree
);
10584 RB_CLEAR_NODE(&block_group
->cache_node
);
10586 if (root
->fs_info
->first_logical_byte
== block_group
->key
.objectid
)
10587 root
->fs_info
->first_logical_byte
= (u64
)-1;
10588 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
10590 down_write(&block_group
->space_info
->groups_sem
);
10592 * we must use list_del_init so people can check to see if they
10593 * are still on the list after taking the semaphore
10595 list_del_init(&block_group
->list
);
10596 if (list_empty(&block_group
->space_info
->block_groups
[index
])) {
10597 kobj
= block_group
->space_info
->block_group_kobjs
[index
];
10598 block_group
->space_info
->block_group_kobjs
[index
] = NULL
;
10599 clear_avail_alloc_bits(root
->fs_info
, block_group
->flags
);
10601 up_write(&block_group
->space_info
->groups_sem
);
10607 if (block_group
->has_caching_ctl
)
10608 caching_ctl
= get_caching_control(block_group
);
10609 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
10610 wait_block_group_cache_done(block_group
);
10611 if (block_group
->has_caching_ctl
) {
10612 down_write(&root
->fs_info
->commit_root_sem
);
10613 if (!caching_ctl
) {
10614 struct btrfs_caching_control
*ctl
;
10616 list_for_each_entry(ctl
,
10617 &root
->fs_info
->caching_block_groups
, list
)
10618 if (ctl
->block_group
== block_group
) {
10620 atomic_inc(&caching_ctl
->count
);
10625 list_del_init(&caching_ctl
->list
);
10626 up_write(&root
->fs_info
->commit_root_sem
);
10628 /* Once for the caching bgs list and once for us. */
10629 put_caching_control(caching_ctl
);
10630 put_caching_control(caching_ctl
);
10634 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
10635 if (!list_empty(&block_group
->dirty_list
)) {
10638 if (!list_empty(&block_group
->io_list
)) {
10641 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
10642 btrfs_remove_free_space_cache(block_group
);
10644 spin_lock(&block_group
->space_info
->lock
);
10645 list_del_init(&block_group
->ro_list
);
10647 if (btrfs_test_opt(root
->fs_info
, ENOSPC_DEBUG
)) {
10648 WARN_ON(block_group
->space_info
->total_bytes
10649 < block_group
->key
.offset
);
10650 WARN_ON(block_group
->space_info
->bytes_readonly
10651 < block_group
->key
.offset
);
10652 WARN_ON(block_group
->space_info
->disk_total
10653 < block_group
->key
.offset
* factor
);
10655 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
10656 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
10657 block_group
->space_info
->disk_total
-= block_group
->key
.offset
* factor
;
10659 spin_unlock(&block_group
->space_info
->lock
);
10661 memcpy(&key
, &block_group
->key
, sizeof(key
));
10664 if (!list_empty(&em
->list
)) {
10665 /* We're in the transaction->pending_chunks list. */
10666 free_extent_map(em
);
10668 spin_lock(&block_group
->lock
);
10669 block_group
->removed
= 1;
10671 * At this point trimming can't start on this block group, because we
10672 * removed the block group from the tree fs_info->block_group_cache_tree
10673 * so no one can't find it anymore and even if someone already got this
10674 * block group before we removed it from the rbtree, they have already
10675 * incremented block_group->trimming - if they didn't, they won't find
10676 * any free space entries because we already removed them all when we
10677 * called btrfs_remove_free_space_cache().
10679 * And we must not remove the extent map from the fs_info->mapping_tree
10680 * to prevent the same logical address range and physical device space
10681 * ranges from being reused for a new block group. This is because our
10682 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10683 * completely transactionless, so while it is trimming a range the
10684 * currently running transaction might finish and a new one start,
10685 * allowing for new block groups to be created that can reuse the same
10686 * physical device locations unless we take this special care.
10688 * There may also be an implicit trim operation if the file system
10689 * is mounted with -odiscard. The same protections must remain
10690 * in place until the extents have been discarded completely when
10691 * the transaction commit has completed.
10693 remove_em
= (atomic_read(&block_group
->trimming
) == 0);
10695 * Make sure a trimmer task always sees the em in the pinned_chunks list
10696 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10697 * before checking block_group->removed).
10701 * Our em might be in trans->transaction->pending_chunks which
10702 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10703 * and so is the fs_info->pinned_chunks list.
10705 * So at this point we must be holding the chunk_mutex to avoid
10706 * any races with chunk allocation (more specifically at
10707 * volumes.c:contains_pending_extent()), to ensure it always
10708 * sees the em, either in the pending_chunks list or in the
10709 * pinned_chunks list.
10711 list_move_tail(&em
->list
, &root
->fs_info
->pinned_chunks
);
10713 spin_unlock(&block_group
->lock
);
10716 struct extent_map_tree
*em_tree
;
10718 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
10719 write_lock(&em_tree
->lock
);
10721 * The em might be in the pending_chunks list, so make sure the
10722 * chunk mutex is locked, since remove_extent_mapping() will
10723 * delete us from that list.
10725 remove_extent_mapping(em_tree
, em
);
10726 write_unlock(&em_tree
->lock
);
10727 /* once for the tree */
10728 free_extent_map(em
);
10731 unlock_chunks(root
);
10733 ret
= remove_block_group_free_space(trans
, root
->fs_info
, block_group
);
10737 btrfs_put_block_group(block_group
);
10738 btrfs_put_block_group(block_group
);
10740 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
10746 ret
= btrfs_del_item(trans
, root
, path
);
10748 btrfs_free_path(path
);
10752 struct btrfs_trans_handle
*
10753 btrfs_start_trans_remove_block_group(struct btrfs_fs_info
*fs_info
,
10754 const u64 chunk_offset
)
10756 struct extent_map_tree
*em_tree
= &fs_info
->mapping_tree
.map_tree
;
10757 struct extent_map
*em
;
10758 struct map_lookup
*map
;
10759 unsigned int num_items
;
10761 read_lock(&em_tree
->lock
);
10762 em
= lookup_extent_mapping(em_tree
, chunk_offset
, 1);
10763 read_unlock(&em_tree
->lock
);
10764 ASSERT(em
&& em
->start
== chunk_offset
);
10767 * We need to reserve 3 + N units from the metadata space info in order
10768 * to remove a block group (done at btrfs_remove_chunk() and at
10769 * btrfs_remove_block_group()), which are used for:
10771 * 1 unit for adding the free space inode's orphan (located in the tree
10773 * 1 unit for deleting the block group item (located in the extent
10775 * 1 unit for deleting the free space item (located in tree of tree
10777 * N units for deleting N device extent items corresponding to each
10778 * stripe (located in the device tree).
10780 * In order to remove a block group we also need to reserve units in the
10781 * system space info in order to update the chunk tree (update one or
10782 * more device items and remove one chunk item), but this is done at
10783 * btrfs_remove_chunk() through a call to check_system_chunk().
10785 map
= em
->map_lookup
;
10786 num_items
= 3 + map
->num_stripes
;
10787 free_extent_map(em
);
10789 return btrfs_start_transaction_fallback_global_rsv(fs_info
->extent_root
,
10794 * Process the unused_bgs list and remove any that don't have any allocated
10795 * space inside of them.
10797 void btrfs_delete_unused_bgs(struct btrfs_fs_info
*fs_info
)
10799 struct btrfs_block_group_cache
*block_group
;
10800 struct btrfs_space_info
*space_info
;
10801 struct btrfs_root
*root
= fs_info
->extent_root
;
10802 struct btrfs_trans_handle
*trans
;
10805 if (!test_bit(BTRFS_FS_OPEN
, &fs_info
->flags
))
10808 spin_lock(&fs_info
->unused_bgs_lock
);
10809 while (!list_empty(&fs_info
->unused_bgs
)) {
10813 block_group
= list_first_entry(&fs_info
->unused_bgs
,
10814 struct btrfs_block_group_cache
,
10816 list_del_init(&block_group
->bg_list
);
10818 space_info
= block_group
->space_info
;
10820 if (ret
|| btrfs_mixed_space_info(space_info
)) {
10821 btrfs_put_block_group(block_group
);
10824 spin_unlock(&fs_info
->unused_bgs_lock
);
10826 mutex_lock(&fs_info
->delete_unused_bgs_mutex
);
10828 /* Don't want to race with allocators so take the groups_sem */
10829 down_write(&space_info
->groups_sem
);
10830 spin_lock(&block_group
->lock
);
10831 if (block_group
->reserved
||
10832 btrfs_block_group_used(&block_group
->item
) ||
10833 (block_group
->ro
&& !block_group
->removed
) ||
10834 list_is_singular(&block_group
->list
)) {
10836 * We want to bail if we made new allocations or have
10837 * outstanding allocations in this block group. We do
10838 * the ro check in case balance is currently acting on
10839 * this block group.
10841 spin_unlock(&block_group
->lock
);
10842 up_write(&space_info
->groups_sem
);
10845 spin_unlock(&block_group
->lock
);
10847 /* We don't want to force the issue, only flip if it's ok. */
10848 ret
= inc_block_group_ro(block_group
, 0);
10849 up_write(&space_info
->groups_sem
);
10856 * Want to do this before we do anything else so we can recover
10857 * properly if we fail to join the transaction.
10859 trans
= btrfs_start_trans_remove_block_group(fs_info
,
10860 block_group
->key
.objectid
);
10861 if (IS_ERR(trans
)) {
10862 btrfs_dec_block_group_ro(root
, block_group
);
10863 ret
= PTR_ERR(trans
);
10868 * We could have pending pinned extents for this block group,
10869 * just delete them, we don't care about them anymore.
10871 start
= block_group
->key
.objectid
;
10872 end
= start
+ block_group
->key
.offset
- 1;
10874 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10875 * btrfs_finish_extent_commit(). If we are at transaction N,
10876 * another task might be running finish_extent_commit() for the
10877 * previous transaction N - 1, and have seen a range belonging
10878 * to the block group in freed_extents[] before we were able to
10879 * clear the whole block group range from freed_extents[]. This
10880 * means that task can lookup for the block group after we
10881 * unpinned it from freed_extents[] and removed it, leading to
10882 * a BUG_ON() at btrfs_unpin_extent_range().
10884 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
10885 ret
= clear_extent_bits(&fs_info
->freed_extents
[0], start
, end
,
10888 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10889 btrfs_dec_block_group_ro(root
, block_group
);
10892 ret
= clear_extent_bits(&fs_info
->freed_extents
[1], start
, end
,
10895 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10896 btrfs_dec_block_group_ro(root
, block_group
);
10899 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10901 /* Reset pinned so btrfs_put_block_group doesn't complain */
10902 spin_lock(&space_info
->lock
);
10903 spin_lock(&block_group
->lock
);
10905 space_info
->bytes_pinned
-= block_group
->pinned
;
10906 space_info
->bytes_readonly
+= block_group
->pinned
;
10907 percpu_counter_add(&space_info
->total_bytes_pinned
,
10908 -block_group
->pinned
);
10909 block_group
->pinned
= 0;
10911 spin_unlock(&block_group
->lock
);
10912 spin_unlock(&space_info
->lock
);
10914 /* DISCARD can flip during remount */
10915 trimming
= btrfs_test_opt(root
->fs_info
, DISCARD
);
10917 /* Implicit trim during transaction commit. */
10919 btrfs_get_block_group_trimming(block_group
);
10922 * Btrfs_remove_chunk will abort the transaction if things go
10925 ret
= btrfs_remove_chunk(trans
, root
,
10926 block_group
->key
.objectid
);
10930 btrfs_put_block_group_trimming(block_group
);
10935 * If we're not mounted with -odiscard, we can just forget
10936 * about this block group. Otherwise we'll need to wait
10937 * until transaction commit to do the actual discard.
10940 spin_lock(&fs_info
->unused_bgs_lock
);
10942 * A concurrent scrub might have added us to the list
10943 * fs_info->unused_bgs, so use a list_move operation
10944 * to add the block group to the deleted_bgs list.
10946 list_move(&block_group
->bg_list
,
10947 &trans
->transaction
->deleted_bgs
);
10948 spin_unlock(&fs_info
->unused_bgs_lock
);
10949 btrfs_get_block_group(block_group
);
10952 btrfs_end_transaction(trans
, root
);
10954 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
10955 btrfs_put_block_group(block_group
);
10956 spin_lock(&fs_info
->unused_bgs_lock
);
10958 spin_unlock(&fs_info
->unused_bgs_lock
);
10961 int btrfs_init_space_info(struct btrfs_fs_info
*fs_info
)
10963 struct btrfs_space_info
*space_info
;
10964 struct btrfs_super_block
*disk_super
;
10970 disk_super
= fs_info
->super_copy
;
10971 if (!btrfs_super_root(disk_super
))
10974 features
= btrfs_super_incompat_flags(disk_super
);
10975 if (features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
10978 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
10979 ret
= update_space_info(fs_info
, flags
, 0, 0, 0, &space_info
);
10984 flags
= BTRFS_BLOCK_GROUP_METADATA
| BTRFS_BLOCK_GROUP_DATA
;
10985 ret
= update_space_info(fs_info
, flags
, 0, 0, 0, &space_info
);
10987 flags
= BTRFS_BLOCK_GROUP_METADATA
;
10988 ret
= update_space_info(fs_info
, flags
, 0, 0, 0, &space_info
);
10992 flags
= BTRFS_BLOCK_GROUP_DATA
;
10993 ret
= update_space_info(fs_info
, flags
, 0, 0, 0, &space_info
);
10999 int btrfs_error_unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
11001 return unpin_extent_range(root
, start
, end
, false);
11005 * It used to be that old block groups would be left around forever.
11006 * Iterating over them would be enough to trim unused space. Since we
11007 * now automatically remove them, we also need to iterate over unallocated
11010 * We don't want a transaction for this since the discard may take a
11011 * substantial amount of time. We don't require that a transaction be
11012 * running, but we do need to take a running transaction into account
11013 * to ensure that we're not discarding chunks that were released in
11014 * the current transaction.
11016 * Holding the chunks lock will prevent other threads from allocating
11017 * or releasing chunks, but it won't prevent a running transaction
11018 * from committing and releasing the memory that the pending chunks
11019 * list head uses. For that, we need to take a reference to the
11022 static int btrfs_trim_free_extents(struct btrfs_device
*device
,
11023 u64 minlen
, u64
*trimmed
)
11025 u64 start
= 0, len
= 0;
11030 /* Not writeable = nothing to do. */
11031 if (!device
->writeable
)
11034 /* No free space = nothing to do. */
11035 if (device
->total_bytes
<= device
->bytes_used
)
11041 struct btrfs_fs_info
*fs_info
= device
->dev_root
->fs_info
;
11042 struct btrfs_transaction
*trans
;
11045 ret
= mutex_lock_interruptible(&fs_info
->chunk_mutex
);
11049 down_read(&fs_info
->commit_root_sem
);
11051 spin_lock(&fs_info
->trans_lock
);
11052 trans
= fs_info
->running_transaction
;
11054 atomic_inc(&trans
->use_count
);
11055 spin_unlock(&fs_info
->trans_lock
);
11057 ret
= find_free_dev_extent_start(trans
, device
, minlen
, start
,
11060 btrfs_put_transaction(trans
);
11063 up_read(&fs_info
->commit_root_sem
);
11064 mutex_unlock(&fs_info
->chunk_mutex
);
11065 if (ret
== -ENOSPC
)
11070 ret
= btrfs_issue_discard(device
->bdev
, start
, len
, &bytes
);
11071 up_read(&fs_info
->commit_root_sem
);
11072 mutex_unlock(&fs_info
->chunk_mutex
);
11080 if (fatal_signal_pending(current
)) {
11081 ret
= -ERESTARTSYS
;
11091 int btrfs_trim_fs(struct btrfs_root
*root
, struct fstrim_range
*range
)
11093 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
11094 struct btrfs_block_group_cache
*cache
= NULL
;
11095 struct btrfs_device
*device
;
11096 struct list_head
*devices
;
11101 u64 total_bytes
= btrfs_super_total_bytes(fs_info
->super_copy
);
11105 * try to trim all FS space, our block group may start from non-zero.
11107 if (range
->len
== total_bytes
)
11108 cache
= btrfs_lookup_first_block_group(fs_info
, range
->start
);
11110 cache
= btrfs_lookup_block_group(fs_info
, range
->start
);
11113 if (cache
->key
.objectid
>= (range
->start
+ range
->len
)) {
11114 btrfs_put_block_group(cache
);
11118 start
= max(range
->start
, cache
->key
.objectid
);
11119 end
= min(range
->start
+ range
->len
,
11120 cache
->key
.objectid
+ cache
->key
.offset
);
11122 if (end
- start
>= range
->minlen
) {
11123 if (!block_group_cache_done(cache
)) {
11124 ret
= cache_block_group(cache
, 0);
11126 btrfs_put_block_group(cache
);
11129 ret
= wait_block_group_cache_done(cache
);
11131 btrfs_put_block_group(cache
);
11135 ret
= btrfs_trim_block_group(cache
,
11141 trimmed
+= group_trimmed
;
11143 btrfs_put_block_group(cache
);
11148 cache
= next_block_group(fs_info
->tree_root
, cache
);
11151 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
11152 devices
= &root
->fs_info
->fs_devices
->alloc_list
;
11153 list_for_each_entry(device
, devices
, dev_alloc_list
) {
11154 ret
= btrfs_trim_free_extents(device
, range
->minlen
,
11159 trimmed
+= group_trimmed
;
11161 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
11163 range
->len
= trimmed
;
11168 * btrfs_{start,end}_write_no_snapshoting() are similar to
11169 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
11170 * data into the page cache through nocow before the subvolume is snapshoted,
11171 * but flush the data into disk after the snapshot creation, or to prevent
11172 * operations while snapshoting is ongoing and that cause the snapshot to be
11173 * inconsistent (writes followed by expanding truncates for example).
11175 void btrfs_end_write_no_snapshoting(struct btrfs_root
*root
)
11177 percpu_counter_dec(&root
->subv_writers
->counter
);
11179 * Make sure counter is updated before we wake up waiters.
11182 if (waitqueue_active(&root
->subv_writers
->wait
))
11183 wake_up(&root
->subv_writers
->wait
);
11186 int btrfs_start_write_no_snapshoting(struct btrfs_root
*root
)
11188 if (atomic_read(&root
->will_be_snapshoted
))
11191 percpu_counter_inc(&root
->subv_writers
->counter
);
11193 * Make sure counter is updated before we check for snapshot creation.
11196 if (atomic_read(&root
->will_be_snapshoted
)) {
11197 btrfs_end_write_no_snapshoting(root
);
11203 static int wait_snapshoting_atomic_t(atomic_t
*a
)
11209 void btrfs_wait_for_snapshot_creation(struct btrfs_root
*root
)
11214 ret
= btrfs_start_write_no_snapshoting(root
);
11217 wait_on_atomic_t(&root
->will_be_snapshoted
,
11218 wait_snapshoting_atomic_t
,
11219 TASK_UNINTERRUPTIBLE
);