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 locked_ref
->processing
= 0;
2651 btrfs_delayed_ref_unlock(locked_ref
);
2652 btrfs_put_delayed_ref(ref
);
2653 btrfs_debug(fs_info
, "run_one_delayed_ref returned %d", ret
);
2658 * If this node is a head, that means all the refs in this head
2659 * have been dealt with, and we will pick the next head to deal
2660 * with, so we must unlock the head and drop it from the cluster
2661 * list before we release it.
2663 if (btrfs_delayed_ref_is_head(ref
)) {
2664 if (locked_ref
->is_data
&&
2665 locked_ref
->total_ref_mod
< 0) {
2666 spin_lock(&delayed_refs
->lock
);
2667 delayed_refs
->pending_csums
-= ref
->num_bytes
;
2668 spin_unlock(&delayed_refs
->lock
);
2670 btrfs_delayed_ref_unlock(locked_ref
);
2673 btrfs_put_delayed_ref(ref
);
2679 * We don't want to include ref heads since we can have empty ref heads
2680 * and those will drastically skew our runtime down since we just do
2681 * accounting, no actual extent tree updates.
2683 if (actual_count
> 0) {
2684 u64 runtime
= ktime_to_ns(ktime_sub(ktime_get(), start
));
2688 * We weigh the current average higher than our current runtime
2689 * to avoid large swings in the average.
2691 spin_lock(&delayed_refs
->lock
);
2692 avg
= fs_info
->avg_delayed_ref_runtime
* 3 + runtime
;
2693 fs_info
->avg_delayed_ref_runtime
= avg
>> 2; /* div by 4 */
2694 spin_unlock(&delayed_refs
->lock
);
2699 #ifdef SCRAMBLE_DELAYED_REFS
2701 * Normally delayed refs get processed in ascending bytenr order. This
2702 * correlates in most cases to the order added. To expose dependencies on this
2703 * order, we start to process the tree in the middle instead of the beginning
2705 static u64
find_middle(struct rb_root
*root
)
2707 struct rb_node
*n
= root
->rb_node
;
2708 struct btrfs_delayed_ref_node
*entry
;
2711 u64 first
= 0, last
= 0;
2715 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2716 first
= entry
->bytenr
;
2720 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2721 last
= entry
->bytenr
;
2726 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2727 WARN_ON(!entry
->in_tree
);
2729 middle
= entry
->bytenr
;
2742 static inline u64
heads_to_leaves(struct btrfs_root
*root
, u64 heads
)
2746 num_bytes
= heads
* (sizeof(struct btrfs_extent_item
) +
2747 sizeof(struct btrfs_extent_inline_ref
));
2748 if (!btrfs_fs_incompat(root
->fs_info
, SKINNY_METADATA
))
2749 num_bytes
+= heads
* sizeof(struct btrfs_tree_block_info
);
2752 * We don't ever fill up leaves all the way so multiply by 2 just to be
2753 * closer to what we're really going to want to use.
2755 return div_u64(num_bytes
, BTRFS_LEAF_DATA_SIZE(root
));
2759 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2760 * would require to store the csums for that many bytes.
2762 u64
btrfs_csum_bytes_to_leaves(struct btrfs_root
*root
, u64 csum_bytes
)
2765 u64 num_csums_per_leaf
;
2768 csum_size
= BTRFS_MAX_ITEM_SIZE(root
);
2769 num_csums_per_leaf
= div64_u64(csum_size
,
2770 (u64
)btrfs_super_csum_size(root
->fs_info
->super_copy
));
2771 num_csums
= div64_u64(csum_bytes
, root
->sectorsize
);
2772 num_csums
+= num_csums_per_leaf
- 1;
2773 num_csums
= div64_u64(num_csums
, num_csums_per_leaf
);
2777 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle
*trans
,
2778 struct btrfs_root
*root
)
2780 struct btrfs_block_rsv
*global_rsv
;
2781 u64 num_heads
= trans
->transaction
->delayed_refs
.num_heads_ready
;
2782 u64 csum_bytes
= trans
->transaction
->delayed_refs
.pending_csums
;
2783 u64 num_dirty_bgs
= trans
->transaction
->num_dirty_bgs
;
2784 u64 num_bytes
, num_dirty_bgs_bytes
;
2787 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
2788 num_heads
= heads_to_leaves(root
, num_heads
);
2790 num_bytes
+= (num_heads
- 1) * root
->nodesize
;
2792 num_bytes
+= btrfs_csum_bytes_to_leaves(root
, csum_bytes
) * root
->nodesize
;
2793 num_dirty_bgs_bytes
= btrfs_calc_trans_metadata_size(root
,
2795 global_rsv
= &root
->fs_info
->global_block_rsv
;
2798 * If we can't allocate any more chunks lets make sure we have _lots_ of
2799 * wiggle room since running delayed refs can create more delayed refs.
2801 if (global_rsv
->space_info
->full
) {
2802 num_dirty_bgs_bytes
<<= 1;
2806 spin_lock(&global_rsv
->lock
);
2807 if (global_rsv
->reserved
<= num_bytes
+ num_dirty_bgs_bytes
)
2809 spin_unlock(&global_rsv
->lock
);
2813 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle
*trans
,
2814 struct btrfs_root
*root
)
2816 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2818 atomic_read(&trans
->transaction
->delayed_refs
.num_entries
);
2823 avg_runtime
= fs_info
->avg_delayed_ref_runtime
;
2824 val
= num_entries
* avg_runtime
;
2825 if (num_entries
* avg_runtime
>= NSEC_PER_SEC
)
2827 if (val
>= NSEC_PER_SEC
/ 2)
2830 return btrfs_check_space_for_delayed_refs(trans
, root
);
2833 struct async_delayed_refs
{
2834 struct btrfs_root
*root
;
2839 struct completion wait
;
2840 struct btrfs_work work
;
2843 static void delayed_ref_async_start(struct btrfs_work
*work
)
2845 struct async_delayed_refs
*async
;
2846 struct btrfs_trans_handle
*trans
;
2849 async
= container_of(work
, struct async_delayed_refs
, work
);
2851 /* if the commit is already started, we don't need to wait here */
2852 if (btrfs_transaction_blocked(async
->root
->fs_info
))
2855 trans
= btrfs_join_transaction(async
->root
);
2856 if (IS_ERR(trans
)) {
2857 async
->error
= PTR_ERR(trans
);
2862 * trans->sync means that when we call end_transaction, we won't
2863 * wait on delayed refs
2867 /* Don't bother flushing if we got into a different transaction */
2868 if (trans
->transid
> async
->transid
)
2871 ret
= btrfs_run_delayed_refs(trans
, async
->root
, async
->count
);
2875 ret
= btrfs_end_transaction(trans
, async
->root
);
2876 if (ret
&& !async
->error
)
2880 complete(&async
->wait
);
2885 int btrfs_async_run_delayed_refs(struct btrfs_root
*root
,
2886 unsigned long count
, u64 transid
, int wait
)
2888 struct async_delayed_refs
*async
;
2891 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
2895 async
->root
= root
->fs_info
->tree_root
;
2896 async
->count
= count
;
2898 async
->transid
= transid
;
2903 init_completion(&async
->wait
);
2905 btrfs_init_work(&async
->work
, btrfs_extent_refs_helper
,
2906 delayed_ref_async_start
, NULL
, NULL
);
2908 btrfs_queue_work(root
->fs_info
->extent_workers
, &async
->work
);
2911 wait_for_completion(&async
->wait
);
2920 * this starts processing the delayed reference count updates and
2921 * extent insertions we have queued up so far. count can be
2922 * 0, which means to process everything in the tree at the start
2923 * of the run (but not newly added entries), or it can be some target
2924 * number you'd like to process.
2926 * Returns 0 on success or if called with an aborted transaction
2927 * Returns <0 on error and aborts the transaction
2929 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2930 struct btrfs_root
*root
, unsigned long count
)
2932 struct rb_node
*node
;
2933 struct btrfs_delayed_ref_root
*delayed_refs
;
2934 struct btrfs_delayed_ref_head
*head
;
2936 int run_all
= count
== (unsigned long)-1;
2937 bool can_flush_pending_bgs
= trans
->can_flush_pending_bgs
;
2939 /* We'll clean this up in btrfs_cleanup_transaction */
2943 if (root
->fs_info
->creating_free_space_tree
)
2946 if (root
== root
->fs_info
->extent_root
)
2947 root
= root
->fs_info
->tree_root
;
2949 delayed_refs
= &trans
->transaction
->delayed_refs
;
2951 count
= atomic_read(&delayed_refs
->num_entries
) * 2;
2954 #ifdef SCRAMBLE_DELAYED_REFS
2955 delayed_refs
->run_delayed_start
= find_middle(&delayed_refs
->root
);
2957 trans
->can_flush_pending_bgs
= false;
2958 ret
= __btrfs_run_delayed_refs(trans
, root
, count
);
2960 btrfs_abort_transaction(trans
, ret
);
2965 if (!list_empty(&trans
->new_bgs
))
2966 btrfs_create_pending_block_groups(trans
, root
);
2968 spin_lock(&delayed_refs
->lock
);
2969 node
= rb_first(&delayed_refs
->href_root
);
2971 spin_unlock(&delayed_refs
->lock
);
2974 count
= (unsigned long)-1;
2977 head
= rb_entry(node
, struct btrfs_delayed_ref_head
,
2979 if (btrfs_delayed_ref_is_head(&head
->node
)) {
2980 struct btrfs_delayed_ref_node
*ref
;
2983 atomic_inc(&ref
->refs
);
2985 spin_unlock(&delayed_refs
->lock
);
2987 * Mutex was contended, block until it's
2988 * released and try again
2990 mutex_lock(&head
->mutex
);
2991 mutex_unlock(&head
->mutex
);
2993 btrfs_put_delayed_ref(ref
);
2999 node
= rb_next(node
);
3001 spin_unlock(&delayed_refs
->lock
);
3006 assert_qgroups_uptodate(trans
);
3007 trans
->can_flush_pending_bgs
= can_flush_pending_bgs
;
3011 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
3012 struct btrfs_root
*root
,
3013 u64 bytenr
, u64 num_bytes
, u64 flags
,
3014 int level
, int is_data
)
3016 struct btrfs_delayed_extent_op
*extent_op
;
3019 extent_op
= btrfs_alloc_delayed_extent_op();
3023 extent_op
->flags_to_set
= flags
;
3024 extent_op
->update_flags
= true;
3025 extent_op
->update_key
= false;
3026 extent_op
->is_data
= is_data
? true : false;
3027 extent_op
->level
= level
;
3029 ret
= btrfs_add_delayed_extent_op(root
->fs_info
, trans
, bytenr
,
3030 num_bytes
, extent_op
);
3032 btrfs_free_delayed_extent_op(extent_op
);
3036 static noinline
int check_delayed_ref(struct btrfs_trans_handle
*trans
,
3037 struct btrfs_root
*root
,
3038 struct btrfs_path
*path
,
3039 u64 objectid
, u64 offset
, u64 bytenr
)
3041 struct btrfs_delayed_ref_head
*head
;
3042 struct btrfs_delayed_ref_node
*ref
;
3043 struct btrfs_delayed_data_ref
*data_ref
;
3044 struct btrfs_delayed_ref_root
*delayed_refs
;
3047 delayed_refs
= &trans
->transaction
->delayed_refs
;
3048 spin_lock(&delayed_refs
->lock
);
3049 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
3051 spin_unlock(&delayed_refs
->lock
);
3055 if (!mutex_trylock(&head
->mutex
)) {
3056 atomic_inc(&head
->node
.refs
);
3057 spin_unlock(&delayed_refs
->lock
);
3059 btrfs_release_path(path
);
3062 * Mutex was contended, block until it's released and let
3065 mutex_lock(&head
->mutex
);
3066 mutex_unlock(&head
->mutex
);
3067 btrfs_put_delayed_ref(&head
->node
);
3070 spin_unlock(&delayed_refs
->lock
);
3072 spin_lock(&head
->lock
);
3073 list_for_each_entry(ref
, &head
->ref_list
, list
) {
3074 /* If it's a shared ref we know a cross reference exists */
3075 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
) {
3080 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
3083 * If our ref doesn't match the one we're currently looking at
3084 * then we have a cross reference.
3086 if (data_ref
->root
!= root
->root_key
.objectid
||
3087 data_ref
->objectid
!= objectid
||
3088 data_ref
->offset
!= offset
) {
3093 spin_unlock(&head
->lock
);
3094 mutex_unlock(&head
->mutex
);
3098 static noinline
int check_committed_ref(struct btrfs_trans_handle
*trans
,
3099 struct btrfs_root
*root
,
3100 struct btrfs_path
*path
,
3101 u64 objectid
, u64 offset
, u64 bytenr
)
3103 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
3104 struct extent_buffer
*leaf
;
3105 struct btrfs_extent_data_ref
*ref
;
3106 struct btrfs_extent_inline_ref
*iref
;
3107 struct btrfs_extent_item
*ei
;
3108 struct btrfs_key key
;
3112 key
.objectid
= bytenr
;
3113 key
.offset
= (u64
)-1;
3114 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
3116 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
3119 BUG_ON(ret
== 0); /* Corruption */
3122 if (path
->slots
[0] == 0)
3126 leaf
= path
->nodes
[0];
3127 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
3129 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
3133 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
3134 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3135 if (item_size
< sizeof(*ei
)) {
3136 WARN_ON(item_size
!= sizeof(struct btrfs_extent_item_v0
));
3140 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
3142 if (item_size
!= sizeof(*ei
) +
3143 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
3146 if (btrfs_extent_generation(leaf
, ei
) <=
3147 btrfs_root_last_snapshot(&root
->root_item
))
3150 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
3151 if (btrfs_extent_inline_ref_type(leaf
, iref
) !=
3152 BTRFS_EXTENT_DATA_REF_KEY
)
3155 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
3156 if (btrfs_extent_refs(leaf
, ei
) !=
3157 btrfs_extent_data_ref_count(leaf
, ref
) ||
3158 btrfs_extent_data_ref_root(leaf
, ref
) !=
3159 root
->root_key
.objectid
||
3160 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
3161 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
3169 int btrfs_cross_ref_exist(struct btrfs_trans_handle
*trans
,
3170 struct btrfs_root
*root
,
3171 u64 objectid
, u64 offset
, u64 bytenr
)
3173 struct btrfs_path
*path
;
3177 path
= btrfs_alloc_path();
3182 ret
= check_committed_ref(trans
, root
, path
, objectid
,
3184 if (ret
&& ret
!= -ENOENT
)
3187 ret2
= check_delayed_ref(trans
, root
, path
, objectid
,
3189 } while (ret2
== -EAGAIN
);
3191 if (ret2
&& ret2
!= -ENOENT
) {
3196 if (ret
!= -ENOENT
|| ret2
!= -ENOENT
)
3199 btrfs_free_path(path
);
3200 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
3205 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
3206 struct btrfs_root
*root
,
3207 struct extent_buffer
*buf
,
3208 int full_backref
, int inc
)
3215 struct btrfs_key key
;
3216 struct btrfs_file_extent_item
*fi
;
3220 int (*process_func
)(struct btrfs_trans_handle
*, struct btrfs_root
*,
3221 u64
, u64
, u64
, u64
, u64
, u64
);
3224 if (btrfs_is_testing(root
->fs_info
))
3227 ref_root
= btrfs_header_owner(buf
);
3228 nritems
= btrfs_header_nritems(buf
);
3229 level
= btrfs_header_level(buf
);
3231 if (!test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) && level
== 0)
3235 process_func
= btrfs_inc_extent_ref
;
3237 process_func
= btrfs_free_extent
;
3240 parent
= buf
->start
;
3244 for (i
= 0; i
< nritems
; i
++) {
3246 btrfs_item_key_to_cpu(buf
, &key
, i
);
3247 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
3249 fi
= btrfs_item_ptr(buf
, i
,
3250 struct btrfs_file_extent_item
);
3251 if (btrfs_file_extent_type(buf
, fi
) ==
3252 BTRFS_FILE_EXTENT_INLINE
)
3254 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
3258 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
3259 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
3260 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
3261 parent
, ref_root
, key
.objectid
,
3266 bytenr
= btrfs_node_blockptr(buf
, i
);
3267 num_bytes
= root
->nodesize
;
3268 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
3269 parent
, ref_root
, level
- 1, 0);
3279 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3280 struct extent_buffer
*buf
, int full_backref
)
3282 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1);
3285 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3286 struct extent_buffer
*buf
, int full_backref
)
3288 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0);
3291 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
3292 struct btrfs_root
*root
,
3293 struct btrfs_path
*path
,
3294 struct btrfs_block_group_cache
*cache
)
3297 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
3299 struct extent_buffer
*leaf
;
3301 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
3308 leaf
= path
->nodes
[0];
3309 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
3310 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
3311 btrfs_mark_buffer_dirty(leaf
);
3313 btrfs_release_path(path
);
3318 static struct btrfs_block_group_cache
*
3319 next_block_group(struct btrfs_root
*root
,
3320 struct btrfs_block_group_cache
*cache
)
3322 struct rb_node
*node
;
3324 spin_lock(&root
->fs_info
->block_group_cache_lock
);
3326 /* If our block group was removed, we need a full search. */
3327 if (RB_EMPTY_NODE(&cache
->cache_node
)) {
3328 const u64 next_bytenr
= cache
->key
.objectid
+ cache
->key
.offset
;
3330 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
3331 btrfs_put_block_group(cache
);
3332 cache
= btrfs_lookup_first_block_group(root
->fs_info
,
3336 node
= rb_next(&cache
->cache_node
);
3337 btrfs_put_block_group(cache
);
3339 cache
= rb_entry(node
, struct btrfs_block_group_cache
,
3341 btrfs_get_block_group(cache
);
3344 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
3348 static int cache_save_setup(struct btrfs_block_group_cache
*block_group
,
3349 struct btrfs_trans_handle
*trans
,
3350 struct btrfs_path
*path
)
3352 struct btrfs_root
*root
= block_group
->fs_info
->tree_root
;
3353 struct inode
*inode
= NULL
;
3355 int dcs
= BTRFS_DC_ERROR
;
3361 * If this block group is smaller than 100 megs don't bother caching the
3364 if (block_group
->key
.offset
< (100 * SZ_1M
)) {
3365 spin_lock(&block_group
->lock
);
3366 block_group
->disk_cache_state
= BTRFS_DC_WRITTEN
;
3367 spin_unlock(&block_group
->lock
);
3374 inode
= lookup_free_space_inode(root
, block_group
, path
);
3375 if (IS_ERR(inode
) && PTR_ERR(inode
) != -ENOENT
) {
3376 ret
= PTR_ERR(inode
);
3377 btrfs_release_path(path
);
3381 if (IS_ERR(inode
)) {
3385 if (block_group
->ro
)
3388 ret
= create_free_space_inode(root
, trans
, block_group
, path
);
3394 /* We've already setup this transaction, go ahead and exit */
3395 if (block_group
->cache_generation
== trans
->transid
&&
3396 i_size_read(inode
)) {
3397 dcs
= BTRFS_DC_SETUP
;
3402 * We want to set the generation to 0, that way if anything goes wrong
3403 * from here on out we know not to trust this cache when we load up next
3406 BTRFS_I(inode
)->generation
= 0;
3407 ret
= btrfs_update_inode(trans
, root
, inode
);
3410 * So theoretically we could recover from this, simply set the
3411 * super cache generation to 0 so we know to invalidate the
3412 * cache, but then we'd have to keep track of the block groups
3413 * that fail this way so we know we _have_ to reset this cache
3414 * before the next commit or risk reading stale cache. So to
3415 * limit our exposure to horrible edge cases lets just abort the
3416 * transaction, this only happens in really bad situations
3419 btrfs_abort_transaction(trans
, ret
);
3424 if (i_size_read(inode
) > 0) {
3425 ret
= btrfs_check_trunc_cache_free_space(root
,
3426 &root
->fs_info
->global_block_rsv
);
3430 ret
= btrfs_truncate_free_space_cache(root
, trans
, NULL
, inode
);
3435 spin_lock(&block_group
->lock
);
3436 if (block_group
->cached
!= BTRFS_CACHE_FINISHED
||
3437 !btrfs_test_opt(root
->fs_info
, SPACE_CACHE
)) {
3439 * don't bother trying to write stuff out _if_
3440 * a) we're not cached,
3441 * b) we're with nospace_cache mount option.
3443 dcs
= BTRFS_DC_WRITTEN
;
3444 spin_unlock(&block_group
->lock
);
3447 spin_unlock(&block_group
->lock
);
3450 * We hit an ENOSPC when setting up the cache in this transaction, just
3451 * skip doing the setup, we've already cleared the cache so we're safe.
3453 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC
, &trans
->transaction
->flags
)) {
3459 * Try to preallocate enough space based on how big the block group is.
3460 * Keep in mind this has to include any pinned space which could end up
3461 * taking up quite a bit since it's not folded into the other space
3464 num_pages
= div_u64(block_group
->key
.offset
, SZ_256M
);
3469 num_pages
*= PAGE_SIZE
;
3471 ret
= btrfs_check_data_free_space(inode
, 0, num_pages
);
3475 ret
= btrfs_prealloc_file_range_trans(inode
, trans
, 0, 0, num_pages
,
3476 num_pages
, num_pages
,
3479 * Our cache requires contiguous chunks so that we don't modify a bunch
3480 * of metadata or split extents when writing the cache out, which means
3481 * we can enospc if we are heavily fragmented in addition to just normal
3482 * out of space conditions. So if we hit this just skip setting up any
3483 * other block groups for this transaction, maybe we'll unpin enough
3484 * space the next time around.
3487 dcs
= BTRFS_DC_SETUP
;
3488 else if (ret
== -ENOSPC
)
3489 set_bit(BTRFS_TRANS_CACHE_ENOSPC
, &trans
->transaction
->flags
);
3494 btrfs_release_path(path
);
3496 spin_lock(&block_group
->lock
);
3497 if (!ret
&& dcs
== BTRFS_DC_SETUP
)
3498 block_group
->cache_generation
= trans
->transid
;
3499 block_group
->disk_cache_state
= dcs
;
3500 spin_unlock(&block_group
->lock
);
3505 int btrfs_setup_space_cache(struct btrfs_trans_handle
*trans
,
3506 struct btrfs_root
*root
)
3508 struct btrfs_block_group_cache
*cache
, *tmp
;
3509 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3510 struct btrfs_path
*path
;
3512 if (list_empty(&cur_trans
->dirty_bgs
) ||
3513 !btrfs_test_opt(root
->fs_info
, SPACE_CACHE
))
3516 path
= btrfs_alloc_path();
3520 /* Could add new block groups, use _safe just in case */
3521 list_for_each_entry_safe(cache
, tmp
, &cur_trans
->dirty_bgs
,
3523 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
)
3524 cache_save_setup(cache
, trans
, path
);
3527 btrfs_free_path(path
);
3532 * transaction commit does final block group cache writeback during a
3533 * critical section where nothing is allowed to change the FS. This is
3534 * required in order for the cache to actually match the block group,
3535 * but can introduce a lot of latency into the commit.
3537 * So, btrfs_start_dirty_block_groups is here to kick off block group
3538 * cache IO. There's a chance we'll have to redo some of it if the
3539 * block group changes again during the commit, but it greatly reduces
3540 * the commit latency by getting rid of the easy block groups while
3541 * we're still allowing others to join the commit.
3543 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle
*trans
,
3544 struct btrfs_root
*root
)
3546 struct btrfs_block_group_cache
*cache
;
3547 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3550 struct btrfs_path
*path
= NULL
;
3552 struct list_head
*io
= &cur_trans
->io_bgs
;
3553 int num_started
= 0;
3556 spin_lock(&cur_trans
->dirty_bgs_lock
);
3557 if (list_empty(&cur_trans
->dirty_bgs
)) {
3558 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3561 list_splice_init(&cur_trans
->dirty_bgs
, &dirty
);
3562 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3566 * make sure all the block groups on our dirty list actually
3569 btrfs_create_pending_block_groups(trans
, root
);
3572 path
= btrfs_alloc_path();
3578 * cache_write_mutex is here only to save us from balance or automatic
3579 * removal of empty block groups deleting this block group while we are
3580 * writing out the cache
3582 mutex_lock(&trans
->transaction
->cache_write_mutex
);
3583 while (!list_empty(&dirty
)) {
3584 cache
= list_first_entry(&dirty
,
3585 struct btrfs_block_group_cache
,
3588 * this can happen if something re-dirties a block
3589 * group that is already under IO. Just wait for it to
3590 * finish and then do it all again
3592 if (!list_empty(&cache
->io_list
)) {
3593 list_del_init(&cache
->io_list
);
3594 btrfs_wait_cache_io(root
, trans
, cache
,
3595 &cache
->io_ctl
, path
,
3596 cache
->key
.objectid
);
3597 btrfs_put_block_group(cache
);
3602 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3603 * if it should update the cache_state. Don't delete
3604 * until after we wait.
3606 * Since we're not running in the commit critical section
3607 * we need the dirty_bgs_lock to protect from update_block_group
3609 spin_lock(&cur_trans
->dirty_bgs_lock
);
3610 list_del_init(&cache
->dirty_list
);
3611 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3615 cache_save_setup(cache
, trans
, path
);
3617 if (cache
->disk_cache_state
== BTRFS_DC_SETUP
) {
3618 cache
->io_ctl
.inode
= NULL
;
3619 ret
= btrfs_write_out_cache(root
, trans
, cache
, path
);
3620 if (ret
== 0 && cache
->io_ctl
.inode
) {
3625 * the cache_write_mutex is protecting
3628 list_add_tail(&cache
->io_list
, io
);
3631 * if we failed to write the cache, the
3632 * generation will be bad and life goes on
3638 ret
= write_one_cache_group(trans
, root
, path
, cache
);
3640 * Our block group might still be attached to the list
3641 * of new block groups in the transaction handle of some
3642 * other task (struct btrfs_trans_handle->new_bgs). This
3643 * means its block group item isn't yet in the extent
3644 * tree. If this happens ignore the error, as we will
3645 * try again later in the critical section of the
3646 * transaction commit.
3648 if (ret
== -ENOENT
) {
3650 spin_lock(&cur_trans
->dirty_bgs_lock
);
3651 if (list_empty(&cache
->dirty_list
)) {
3652 list_add_tail(&cache
->dirty_list
,
3653 &cur_trans
->dirty_bgs
);
3654 btrfs_get_block_group(cache
);
3656 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3658 btrfs_abort_transaction(trans
, ret
);
3662 /* if its not on the io list, we need to put the block group */
3664 btrfs_put_block_group(cache
);
3670 * Avoid blocking other tasks for too long. It might even save
3671 * us from writing caches for block groups that are going to be
3674 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
3675 mutex_lock(&trans
->transaction
->cache_write_mutex
);
3677 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
3680 * go through delayed refs for all the stuff we've just kicked off
3681 * and then loop back (just once)
3683 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
3684 if (!ret
&& loops
== 0) {
3686 spin_lock(&cur_trans
->dirty_bgs_lock
);
3687 list_splice_init(&cur_trans
->dirty_bgs
, &dirty
);
3689 * dirty_bgs_lock protects us from concurrent block group
3690 * deletes too (not just cache_write_mutex).
3692 if (!list_empty(&dirty
)) {
3693 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3696 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3697 } else if (ret
< 0) {
3698 btrfs_cleanup_dirty_bgs(cur_trans
, root
);
3701 btrfs_free_path(path
);
3705 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
3706 struct btrfs_root
*root
)
3708 struct btrfs_block_group_cache
*cache
;
3709 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3712 struct btrfs_path
*path
;
3713 struct list_head
*io
= &cur_trans
->io_bgs
;
3714 int num_started
= 0;
3716 path
= btrfs_alloc_path();
3721 * Even though we are in the critical section of the transaction commit,
3722 * we can still have concurrent tasks adding elements to this
3723 * transaction's list of dirty block groups. These tasks correspond to
3724 * endio free space workers started when writeback finishes for a
3725 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3726 * allocate new block groups as a result of COWing nodes of the root
3727 * tree when updating the free space inode. The writeback for the space
3728 * caches is triggered by an earlier call to
3729 * btrfs_start_dirty_block_groups() and iterations of the following
3731 * Also we want to do the cache_save_setup first and then run the
3732 * delayed refs to make sure we have the best chance at doing this all
3735 spin_lock(&cur_trans
->dirty_bgs_lock
);
3736 while (!list_empty(&cur_trans
->dirty_bgs
)) {
3737 cache
= list_first_entry(&cur_trans
->dirty_bgs
,
3738 struct btrfs_block_group_cache
,
3742 * this can happen if cache_save_setup re-dirties a block
3743 * group that is already under IO. Just wait for it to
3744 * finish and then do it all again
3746 if (!list_empty(&cache
->io_list
)) {
3747 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3748 list_del_init(&cache
->io_list
);
3749 btrfs_wait_cache_io(root
, trans
, cache
,
3750 &cache
->io_ctl
, path
,
3751 cache
->key
.objectid
);
3752 btrfs_put_block_group(cache
);
3753 spin_lock(&cur_trans
->dirty_bgs_lock
);
3757 * don't remove from the dirty list until after we've waited
3760 list_del_init(&cache
->dirty_list
);
3761 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3764 cache_save_setup(cache
, trans
, path
);
3767 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long) -1);
3769 if (!ret
&& cache
->disk_cache_state
== BTRFS_DC_SETUP
) {
3770 cache
->io_ctl
.inode
= NULL
;
3771 ret
= btrfs_write_out_cache(root
, trans
, cache
, path
);
3772 if (ret
== 0 && cache
->io_ctl
.inode
) {
3775 list_add_tail(&cache
->io_list
, io
);
3778 * if we failed to write the cache, the
3779 * generation will be bad and life goes on
3785 ret
= write_one_cache_group(trans
, root
, path
, cache
);
3787 * One of the free space endio workers might have
3788 * created a new block group while updating a free space
3789 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3790 * and hasn't released its transaction handle yet, in
3791 * which case the new block group is still attached to
3792 * its transaction handle and its creation has not
3793 * finished yet (no block group item in the extent tree
3794 * yet, etc). If this is the case, wait for all free
3795 * space endio workers to finish and retry. This is a
3796 * a very rare case so no need for a more efficient and
3799 if (ret
== -ENOENT
) {
3800 wait_event(cur_trans
->writer_wait
,
3801 atomic_read(&cur_trans
->num_writers
) == 1);
3802 ret
= write_one_cache_group(trans
, root
, path
,
3806 btrfs_abort_transaction(trans
, ret
);
3809 /* if its not on the io list, we need to put the block group */
3811 btrfs_put_block_group(cache
);
3812 spin_lock(&cur_trans
->dirty_bgs_lock
);
3814 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3816 while (!list_empty(io
)) {
3817 cache
= list_first_entry(io
, struct btrfs_block_group_cache
,
3819 list_del_init(&cache
->io_list
);
3820 btrfs_wait_cache_io(root
, trans
, cache
,
3821 &cache
->io_ctl
, path
, cache
->key
.objectid
);
3822 btrfs_put_block_group(cache
);
3825 btrfs_free_path(path
);
3829 int btrfs_extent_readonly(struct btrfs_root
*root
, u64 bytenr
)
3831 struct btrfs_block_group_cache
*block_group
;
3834 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
3835 if (!block_group
|| block_group
->ro
)
3838 btrfs_put_block_group(block_group
);
3842 bool btrfs_inc_nocow_writers(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
3844 struct btrfs_block_group_cache
*bg
;
3847 bg
= btrfs_lookup_block_group(fs_info
, bytenr
);
3851 spin_lock(&bg
->lock
);
3855 atomic_inc(&bg
->nocow_writers
);
3856 spin_unlock(&bg
->lock
);
3858 /* no put on block group, done by btrfs_dec_nocow_writers */
3860 btrfs_put_block_group(bg
);
3866 void btrfs_dec_nocow_writers(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
3868 struct btrfs_block_group_cache
*bg
;
3870 bg
= btrfs_lookup_block_group(fs_info
, bytenr
);
3872 if (atomic_dec_and_test(&bg
->nocow_writers
))
3873 wake_up_atomic_t(&bg
->nocow_writers
);
3875 * Once for our lookup and once for the lookup done by a previous call
3876 * to btrfs_inc_nocow_writers()
3878 btrfs_put_block_group(bg
);
3879 btrfs_put_block_group(bg
);
3882 static int btrfs_wait_nocow_writers_atomic_t(atomic_t
*a
)
3888 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache
*bg
)
3890 wait_on_atomic_t(&bg
->nocow_writers
,
3891 btrfs_wait_nocow_writers_atomic_t
,
3892 TASK_UNINTERRUPTIBLE
);
3895 static const char *alloc_name(u64 flags
)
3898 case BTRFS_BLOCK_GROUP_METADATA
|BTRFS_BLOCK_GROUP_DATA
:
3900 case BTRFS_BLOCK_GROUP_METADATA
:
3902 case BTRFS_BLOCK_GROUP_DATA
:
3904 case BTRFS_BLOCK_GROUP_SYSTEM
:
3908 return "invalid-combination";
3912 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
3913 u64 total_bytes
, u64 bytes_used
,
3915 struct btrfs_space_info
**space_info
)
3917 struct btrfs_space_info
*found
;
3922 if (flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3923 BTRFS_BLOCK_GROUP_RAID10
))
3928 found
= __find_space_info(info
, flags
);
3930 spin_lock(&found
->lock
);
3931 found
->total_bytes
+= total_bytes
;
3932 found
->disk_total
+= total_bytes
* factor
;
3933 found
->bytes_used
+= bytes_used
;
3934 found
->disk_used
+= bytes_used
* factor
;
3935 found
->bytes_readonly
+= bytes_readonly
;
3936 if (total_bytes
> 0)
3938 space_info_add_new_bytes(info
, found
, total_bytes
-
3939 bytes_used
- bytes_readonly
);
3940 spin_unlock(&found
->lock
);
3941 *space_info
= found
;
3944 found
= kzalloc(sizeof(*found
), GFP_NOFS
);
3948 ret
= percpu_counter_init(&found
->total_bytes_pinned
, 0, GFP_KERNEL
);
3954 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
3955 INIT_LIST_HEAD(&found
->block_groups
[i
]);
3956 init_rwsem(&found
->groups_sem
);
3957 spin_lock_init(&found
->lock
);
3958 found
->flags
= flags
& BTRFS_BLOCK_GROUP_TYPE_MASK
;
3959 found
->total_bytes
= total_bytes
;
3960 found
->disk_total
= total_bytes
* factor
;
3961 found
->bytes_used
= bytes_used
;
3962 found
->disk_used
= bytes_used
* factor
;
3963 found
->bytes_pinned
= 0;
3964 found
->bytes_reserved
= 0;
3965 found
->bytes_readonly
= bytes_readonly
;
3966 found
->bytes_may_use
= 0;
3968 found
->max_extent_size
= 0;
3969 found
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3970 found
->chunk_alloc
= 0;
3972 init_waitqueue_head(&found
->wait
);
3973 INIT_LIST_HEAD(&found
->ro_bgs
);
3974 INIT_LIST_HEAD(&found
->tickets
);
3975 INIT_LIST_HEAD(&found
->priority_tickets
);
3977 ret
= kobject_init_and_add(&found
->kobj
, &space_info_ktype
,
3978 info
->space_info_kobj
, "%s",
3979 alloc_name(found
->flags
));
3985 *space_info
= found
;
3986 list_add_rcu(&found
->list
, &info
->space_info
);
3987 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3988 info
->data_sinfo
= found
;
3993 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
3995 u64 extra_flags
= chunk_to_extended(flags
) &
3996 BTRFS_EXTENDED_PROFILE_MASK
;
3998 write_seqlock(&fs_info
->profiles_lock
);
3999 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
4000 fs_info
->avail_data_alloc_bits
|= extra_flags
;
4001 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
4002 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
4003 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
4004 fs_info
->avail_system_alloc_bits
|= extra_flags
;
4005 write_sequnlock(&fs_info
->profiles_lock
);
4009 * returns target flags in extended format or 0 if restripe for this
4010 * chunk_type is not in progress
4012 * should be called with either volume_mutex or balance_lock held
4014 static u64
get_restripe_target(struct btrfs_fs_info
*fs_info
, u64 flags
)
4016 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
4022 if (flags
& BTRFS_BLOCK_GROUP_DATA
&&
4023 bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
4024 target
= BTRFS_BLOCK_GROUP_DATA
| bctl
->data
.target
;
4025 } else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
&&
4026 bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
4027 target
= BTRFS_BLOCK_GROUP_SYSTEM
| bctl
->sys
.target
;
4028 } else if (flags
& BTRFS_BLOCK_GROUP_METADATA
&&
4029 bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
4030 target
= BTRFS_BLOCK_GROUP_METADATA
| bctl
->meta
.target
;
4037 * @flags: available profiles in extended format (see ctree.h)
4039 * Returns reduced profile in chunk format. If profile changing is in
4040 * progress (either running or paused) picks the target profile (if it's
4041 * already available), otherwise falls back to plain reducing.
4043 static u64
btrfs_reduce_alloc_profile(struct btrfs_root
*root
, u64 flags
)
4045 u64 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
4051 * see if restripe for this chunk_type is in progress, if so
4052 * try to reduce to the target profile
4054 spin_lock(&root
->fs_info
->balance_lock
);
4055 target
= get_restripe_target(root
->fs_info
, flags
);
4057 /* pick target profile only if it's already available */
4058 if ((flags
& target
) & BTRFS_EXTENDED_PROFILE_MASK
) {
4059 spin_unlock(&root
->fs_info
->balance_lock
);
4060 return extended_to_chunk(target
);
4063 spin_unlock(&root
->fs_info
->balance_lock
);
4065 /* First, mask out the RAID levels which aren't possible */
4066 for (raid_type
= 0; raid_type
< BTRFS_NR_RAID_TYPES
; raid_type
++) {
4067 if (num_devices
>= btrfs_raid_array
[raid_type
].devs_min
)
4068 allowed
|= btrfs_raid_group
[raid_type
];
4072 if (allowed
& BTRFS_BLOCK_GROUP_RAID6
)
4073 allowed
= BTRFS_BLOCK_GROUP_RAID6
;
4074 else if (allowed
& BTRFS_BLOCK_GROUP_RAID5
)
4075 allowed
= BTRFS_BLOCK_GROUP_RAID5
;
4076 else if (allowed
& BTRFS_BLOCK_GROUP_RAID10
)
4077 allowed
= BTRFS_BLOCK_GROUP_RAID10
;
4078 else if (allowed
& BTRFS_BLOCK_GROUP_RAID1
)
4079 allowed
= BTRFS_BLOCK_GROUP_RAID1
;
4080 else if (allowed
& BTRFS_BLOCK_GROUP_RAID0
)
4081 allowed
= BTRFS_BLOCK_GROUP_RAID0
;
4083 flags
&= ~BTRFS_BLOCK_GROUP_PROFILE_MASK
;
4085 return extended_to_chunk(flags
| allowed
);
4088 static u64
get_alloc_profile(struct btrfs_root
*root
, u64 orig_flags
)
4095 seq
= read_seqbegin(&root
->fs_info
->profiles_lock
);
4097 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
4098 flags
|= root
->fs_info
->avail_data_alloc_bits
;
4099 else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
4100 flags
|= root
->fs_info
->avail_system_alloc_bits
;
4101 else if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
4102 flags
|= root
->fs_info
->avail_metadata_alloc_bits
;
4103 } while (read_seqretry(&root
->fs_info
->profiles_lock
, seq
));
4105 return btrfs_reduce_alloc_profile(root
, flags
);
4108 u64
btrfs_get_alloc_profile(struct btrfs_root
*root
, int data
)
4114 flags
= BTRFS_BLOCK_GROUP_DATA
;
4115 else if (root
== root
->fs_info
->chunk_root
)
4116 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
4118 flags
= BTRFS_BLOCK_GROUP_METADATA
;
4120 ret
= get_alloc_profile(root
, flags
);
4124 int btrfs_alloc_data_chunk_ondemand(struct inode
*inode
, u64 bytes
)
4126 struct btrfs_space_info
*data_sinfo
;
4127 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4128 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4131 int need_commit
= 2;
4132 int have_pinned_space
;
4134 /* make sure bytes are sectorsize aligned */
4135 bytes
= ALIGN(bytes
, root
->sectorsize
);
4137 if (btrfs_is_free_space_inode(inode
)) {
4139 ASSERT(current
->journal_info
);
4142 data_sinfo
= fs_info
->data_sinfo
;
4147 /* make sure we have enough space to handle the data first */
4148 spin_lock(&data_sinfo
->lock
);
4149 used
= data_sinfo
->bytes_used
+ data_sinfo
->bytes_reserved
+
4150 data_sinfo
->bytes_pinned
+ data_sinfo
->bytes_readonly
+
4151 data_sinfo
->bytes_may_use
;
4153 if (used
+ bytes
> data_sinfo
->total_bytes
) {
4154 struct btrfs_trans_handle
*trans
;
4157 * if we don't have enough free bytes in this space then we need
4158 * to alloc a new chunk.
4160 if (!data_sinfo
->full
) {
4163 data_sinfo
->force_alloc
= CHUNK_ALLOC_FORCE
;
4164 spin_unlock(&data_sinfo
->lock
);
4166 alloc_target
= btrfs_get_alloc_profile(root
, 1);
4168 * It is ugly that we don't call nolock join
4169 * transaction for the free space inode case here.
4170 * But it is safe because we only do the data space
4171 * reservation for the free space cache in the
4172 * transaction context, the common join transaction
4173 * just increase the counter of the current transaction
4174 * handler, doesn't try to acquire the trans_lock of
4177 trans
= btrfs_join_transaction(root
);
4179 return PTR_ERR(trans
);
4181 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
4183 CHUNK_ALLOC_NO_FORCE
);
4184 btrfs_end_transaction(trans
, root
);
4189 have_pinned_space
= 1;
4195 data_sinfo
= fs_info
->data_sinfo
;
4201 * If we don't have enough pinned space to deal with this
4202 * allocation, and no removed chunk in current transaction,
4203 * don't bother committing the transaction.
4205 have_pinned_space
= percpu_counter_compare(
4206 &data_sinfo
->total_bytes_pinned
,
4207 used
+ bytes
- data_sinfo
->total_bytes
);
4208 spin_unlock(&data_sinfo
->lock
);
4210 /* commit the current transaction and try again */
4213 !atomic_read(&root
->fs_info
->open_ioctl_trans
)) {
4216 if (need_commit
> 0) {
4217 btrfs_start_delalloc_roots(fs_info
, 0, -1);
4218 btrfs_wait_ordered_roots(fs_info
, -1, 0, (u64
)-1);
4221 trans
= btrfs_join_transaction(root
);
4223 return PTR_ERR(trans
);
4224 if (have_pinned_space
>= 0 ||
4225 test_bit(BTRFS_TRANS_HAVE_FREE_BGS
,
4226 &trans
->transaction
->flags
) ||
4228 ret
= btrfs_commit_transaction(trans
, root
);
4232 * The cleaner kthread might still be doing iput
4233 * operations. Wait for it to finish so that
4234 * more space is released.
4236 mutex_lock(&root
->fs_info
->cleaner_delayed_iput_mutex
);
4237 mutex_unlock(&root
->fs_info
->cleaner_delayed_iput_mutex
);
4240 btrfs_end_transaction(trans
, root
);
4244 trace_btrfs_space_reservation(root
->fs_info
,
4245 "space_info:enospc",
4246 data_sinfo
->flags
, bytes
, 1);
4249 data_sinfo
->bytes_may_use
+= bytes
;
4250 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
4251 data_sinfo
->flags
, bytes
, 1);
4252 spin_unlock(&data_sinfo
->lock
);
4258 * New check_data_free_space() with ability for precious data reservation
4259 * Will replace old btrfs_check_data_free_space(), but for patch split,
4260 * add a new function first and then replace it.
4262 int btrfs_check_data_free_space(struct inode
*inode
, u64 start
, u64 len
)
4264 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4267 /* align the range */
4268 len
= round_up(start
+ len
, root
->sectorsize
) -
4269 round_down(start
, root
->sectorsize
);
4270 start
= round_down(start
, root
->sectorsize
);
4272 ret
= btrfs_alloc_data_chunk_ondemand(inode
, len
);
4277 * Use new btrfs_qgroup_reserve_data to reserve precious data space
4279 * TODO: Find a good method to avoid reserve data space for NOCOW
4280 * range, but don't impact performance on quota disable case.
4282 ret
= btrfs_qgroup_reserve_data(inode
, start
, len
);
4287 * Called if we need to clear a data reservation for this inode
4288 * Normally in a error case.
4290 * This one will *NOT* use accurate qgroup reserved space API, just for case
4291 * which we can't sleep and is sure it won't affect qgroup reserved space.
4292 * Like clear_bit_hook().
4294 void btrfs_free_reserved_data_space_noquota(struct inode
*inode
, u64 start
,
4297 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4298 struct btrfs_space_info
*data_sinfo
;
4300 /* Make sure the range is aligned to sectorsize */
4301 len
= round_up(start
+ len
, root
->sectorsize
) -
4302 round_down(start
, root
->sectorsize
);
4303 start
= round_down(start
, root
->sectorsize
);
4305 data_sinfo
= root
->fs_info
->data_sinfo
;
4306 spin_lock(&data_sinfo
->lock
);
4307 if (WARN_ON(data_sinfo
->bytes_may_use
< len
))
4308 data_sinfo
->bytes_may_use
= 0;
4310 data_sinfo
->bytes_may_use
-= len
;
4311 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
4312 data_sinfo
->flags
, len
, 0);
4313 spin_unlock(&data_sinfo
->lock
);
4317 * Called if we need to clear a data reservation for this inode
4318 * Normally in a error case.
4320 * This one will handle the per-inode data rsv map for accurate reserved
4323 void btrfs_free_reserved_data_space(struct inode
*inode
, u64 start
, u64 len
)
4325 btrfs_free_reserved_data_space_noquota(inode
, start
, len
);
4326 btrfs_qgroup_free_data(inode
, start
, len
);
4329 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
4331 struct list_head
*head
= &info
->space_info
;
4332 struct btrfs_space_info
*found
;
4335 list_for_each_entry_rcu(found
, head
, list
) {
4336 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
4337 found
->force_alloc
= CHUNK_ALLOC_FORCE
;
4342 static inline u64
calc_global_rsv_need_space(struct btrfs_block_rsv
*global
)
4344 return (global
->size
<< 1);
4347 static int should_alloc_chunk(struct btrfs_root
*root
,
4348 struct btrfs_space_info
*sinfo
, int force
)
4350 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
4351 u64 num_bytes
= sinfo
->total_bytes
- sinfo
->bytes_readonly
;
4352 u64 num_allocated
= sinfo
->bytes_used
+ sinfo
->bytes_reserved
;
4355 if (force
== CHUNK_ALLOC_FORCE
)
4359 * We need to take into account the global rsv because for all intents
4360 * and purposes it's used space. Don't worry about locking the
4361 * global_rsv, it doesn't change except when the transaction commits.
4363 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
4364 num_allocated
+= calc_global_rsv_need_space(global_rsv
);
4367 * in limited mode, we want to have some free space up to
4368 * about 1% of the FS size.
4370 if (force
== CHUNK_ALLOC_LIMITED
) {
4371 thresh
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
4372 thresh
= max_t(u64
, SZ_64M
, div_factor_fine(thresh
, 1));
4374 if (num_bytes
- num_allocated
< thresh
)
4378 if (num_allocated
+ SZ_2M
< div_factor(num_bytes
, 8))
4383 static u64
get_profile_num_devs(struct btrfs_root
*root
, u64 type
)
4387 if (type
& (BTRFS_BLOCK_GROUP_RAID10
|
4388 BTRFS_BLOCK_GROUP_RAID0
|
4389 BTRFS_BLOCK_GROUP_RAID5
|
4390 BTRFS_BLOCK_GROUP_RAID6
))
4391 num_dev
= root
->fs_info
->fs_devices
->rw_devices
;
4392 else if (type
& BTRFS_BLOCK_GROUP_RAID1
)
4395 num_dev
= 1; /* DUP or single */
4401 * If @is_allocation is true, reserve space in the system space info necessary
4402 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4405 void check_system_chunk(struct btrfs_trans_handle
*trans
,
4406 struct btrfs_root
*root
,
4409 struct btrfs_space_info
*info
;
4416 * Needed because we can end up allocating a system chunk and for an
4417 * atomic and race free space reservation in the chunk block reserve.
4419 ASSERT(mutex_is_locked(&root
->fs_info
->chunk_mutex
));
4421 info
= __find_space_info(root
->fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
4422 spin_lock(&info
->lock
);
4423 left
= info
->total_bytes
- info
->bytes_used
- info
->bytes_pinned
-
4424 info
->bytes_reserved
- info
->bytes_readonly
-
4425 info
->bytes_may_use
;
4426 spin_unlock(&info
->lock
);
4428 num_devs
= get_profile_num_devs(root
, type
);
4430 /* num_devs device items to update and 1 chunk item to add or remove */
4431 thresh
= btrfs_calc_trunc_metadata_size(root
, num_devs
) +
4432 btrfs_calc_trans_metadata_size(root
, 1);
4434 if (left
< thresh
&& btrfs_test_opt(root
->fs_info
, ENOSPC_DEBUG
)) {
4435 btrfs_info(root
->fs_info
, "left=%llu, need=%llu, flags=%llu",
4436 left
, thresh
, type
);
4437 dump_space_info(info
, 0, 0);
4440 if (left
< thresh
) {
4443 flags
= btrfs_get_alloc_profile(root
->fs_info
->chunk_root
, 0);
4445 * Ignore failure to create system chunk. We might end up not
4446 * needing it, as we might not need to COW all nodes/leafs from
4447 * the paths we visit in the chunk tree (they were already COWed
4448 * or created in the current transaction for example).
4450 ret
= btrfs_alloc_chunk(trans
, root
, flags
);
4454 ret
= btrfs_block_rsv_add(root
->fs_info
->chunk_root
,
4455 &root
->fs_info
->chunk_block_rsv
,
4456 thresh
, BTRFS_RESERVE_NO_FLUSH
);
4458 trans
->chunk_bytes_reserved
+= thresh
;
4463 * If force is CHUNK_ALLOC_FORCE:
4464 * - return 1 if it successfully allocates a chunk,
4465 * - return errors including -ENOSPC otherwise.
4466 * If force is NOT CHUNK_ALLOC_FORCE:
4467 * - return 0 if it doesn't need to allocate a new chunk,
4468 * - return 1 if it successfully allocates a chunk,
4469 * - return errors including -ENOSPC otherwise.
4471 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
4472 struct btrfs_root
*extent_root
, u64 flags
, int force
)
4474 struct btrfs_space_info
*space_info
;
4475 struct btrfs_fs_info
*fs_info
= extent_root
->fs_info
;
4476 int wait_for_alloc
= 0;
4479 /* Don't re-enter if we're already allocating a chunk */
4480 if (trans
->allocating_chunk
)
4483 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
4485 ret
= update_space_info(extent_root
->fs_info
, flags
,
4486 0, 0, 0, &space_info
);
4487 BUG_ON(ret
); /* -ENOMEM */
4489 BUG_ON(!space_info
); /* Logic error */
4492 spin_lock(&space_info
->lock
);
4493 if (force
< space_info
->force_alloc
)
4494 force
= space_info
->force_alloc
;
4495 if (space_info
->full
) {
4496 if (should_alloc_chunk(extent_root
, space_info
, force
))
4500 spin_unlock(&space_info
->lock
);
4504 if (!should_alloc_chunk(extent_root
, space_info
, force
)) {
4505 spin_unlock(&space_info
->lock
);
4507 } else if (space_info
->chunk_alloc
) {
4510 space_info
->chunk_alloc
= 1;
4513 spin_unlock(&space_info
->lock
);
4515 mutex_lock(&fs_info
->chunk_mutex
);
4518 * The chunk_mutex is held throughout the entirety of a chunk
4519 * allocation, so once we've acquired the chunk_mutex we know that the
4520 * other guy is done and we need to recheck and see if we should
4523 if (wait_for_alloc
) {
4524 mutex_unlock(&fs_info
->chunk_mutex
);
4529 trans
->allocating_chunk
= true;
4532 * If we have mixed data/metadata chunks we want to make sure we keep
4533 * allocating mixed chunks instead of individual chunks.
4535 if (btrfs_mixed_space_info(space_info
))
4536 flags
|= (BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
);
4539 * if we're doing a data chunk, go ahead and make sure that
4540 * we keep a reasonable number of metadata chunks allocated in the
4543 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
4544 fs_info
->data_chunk_allocations
++;
4545 if (!(fs_info
->data_chunk_allocations
%
4546 fs_info
->metadata_ratio
))
4547 force_metadata_allocation(fs_info
);
4551 * Check if we have enough space in SYSTEM chunk because we may need
4552 * to update devices.
4554 check_system_chunk(trans
, extent_root
, flags
);
4556 ret
= btrfs_alloc_chunk(trans
, extent_root
, flags
);
4557 trans
->allocating_chunk
= false;
4559 spin_lock(&space_info
->lock
);
4560 if (ret
< 0 && ret
!= -ENOSPC
)
4563 space_info
->full
= 1;
4567 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
4569 space_info
->chunk_alloc
= 0;
4570 spin_unlock(&space_info
->lock
);
4571 mutex_unlock(&fs_info
->chunk_mutex
);
4573 * When we allocate a new chunk we reserve space in the chunk block
4574 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4575 * add new nodes/leafs to it if we end up needing to do it when
4576 * inserting the chunk item and updating device items as part of the
4577 * second phase of chunk allocation, performed by
4578 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4579 * large number of new block groups to create in our transaction
4580 * handle's new_bgs list to avoid exhausting the chunk block reserve
4581 * in extreme cases - like having a single transaction create many new
4582 * block groups when starting to write out the free space caches of all
4583 * the block groups that were made dirty during the lifetime of the
4586 if (trans
->can_flush_pending_bgs
&&
4587 trans
->chunk_bytes_reserved
>= (u64
)SZ_2M
) {
4588 btrfs_create_pending_block_groups(trans
, extent_root
);
4589 btrfs_trans_release_chunk_metadata(trans
);
4594 static int can_overcommit(struct btrfs_root
*root
,
4595 struct btrfs_space_info
*space_info
, u64 bytes
,
4596 enum btrfs_reserve_flush_enum flush
)
4598 struct btrfs_block_rsv
*global_rsv
;
4604 /* Don't overcommit when in mixed mode. */
4605 if (space_info
->flags
& BTRFS_BLOCK_GROUP_DATA
)
4608 BUG_ON(root
->fs_info
== NULL
);
4609 global_rsv
= &root
->fs_info
->global_block_rsv
;
4610 profile
= btrfs_get_alloc_profile(root
, 0);
4611 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4612 space_info
->bytes_pinned
+ space_info
->bytes_readonly
;
4615 * We only want to allow over committing if we have lots of actual space
4616 * free, but if we don't have enough space to handle the global reserve
4617 * space then we could end up having a real enospc problem when trying
4618 * to allocate a chunk or some other such important allocation.
4620 spin_lock(&global_rsv
->lock
);
4621 space_size
= calc_global_rsv_need_space(global_rsv
);
4622 spin_unlock(&global_rsv
->lock
);
4623 if (used
+ space_size
>= space_info
->total_bytes
)
4626 used
+= space_info
->bytes_may_use
;
4628 spin_lock(&root
->fs_info
->free_chunk_lock
);
4629 avail
= root
->fs_info
->free_chunk_space
;
4630 spin_unlock(&root
->fs_info
->free_chunk_lock
);
4633 * If we have dup, raid1 or raid10 then only half of the free
4634 * space is actually useable. For raid56, the space info used
4635 * doesn't include the parity drive, so we don't have to
4638 if (profile
& (BTRFS_BLOCK_GROUP_DUP
|
4639 BTRFS_BLOCK_GROUP_RAID1
|
4640 BTRFS_BLOCK_GROUP_RAID10
))
4644 * If we aren't flushing all things, let us overcommit up to
4645 * 1/2th of the space. If we can flush, don't let us overcommit
4646 * too much, let it overcommit up to 1/8 of the space.
4648 if (flush
== BTRFS_RESERVE_FLUSH_ALL
)
4653 if (used
+ bytes
< space_info
->total_bytes
+ avail
)
4658 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root
*root
,
4659 unsigned long nr_pages
, int nr_items
)
4661 struct super_block
*sb
= root
->fs_info
->sb
;
4663 if (down_read_trylock(&sb
->s_umount
)) {
4664 writeback_inodes_sb_nr(sb
, nr_pages
, WB_REASON_FS_FREE_SPACE
);
4665 up_read(&sb
->s_umount
);
4668 * We needn't worry the filesystem going from r/w to r/o though
4669 * we don't acquire ->s_umount mutex, because the filesystem
4670 * should guarantee the delalloc inodes list be empty after
4671 * the filesystem is readonly(all dirty pages are written to
4674 btrfs_start_delalloc_roots(root
->fs_info
, 0, nr_items
);
4675 if (!current
->journal_info
)
4676 btrfs_wait_ordered_roots(root
->fs_info
, nr_items
,
4681 static inline int calc_reclaim_items_nr(struct btrfs_root
*root
, u64 to_reclaim
)
4686 bytes
= btrfs_calc_trans_metadata_size(root
, 1);
4687 nr
= (int)div64_u64(to_reclaim
, bytes
);
4693 #define EXTENT_SIZE_PER_ITEM SZ_256K
4696 * shrink metadata reservation for delalloc
4698 static void shrink_delalloc(struct btrfs_root
*root
, u64 to_reclaim
, u64 orig
,
4701 struct btrfs_block_rsv
*block_rsv
;
4702 struct btrfs_space_info
*space_info
;
4703 struct btrfs_trans_handle
*trans
;
4707 unsigned long nr_pages
;
4710 enum btrfs_reserve_flush_enum flush
;
4712 /* Calc the number of the pages we need flush for space reservation */
4713 items
= calc_reclaim_items_nr(root
, to_reclaim
);
4714 to_reclaim
= (u64
)items
* EXTENT_SIZE_PER_ITEM
;
4716 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4717 block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
4718 space_info
= block_rsv
->space_info
;
4720 delalloc_bytes
= percpu_counter_sum_positive(
4721 &root
->fs_info
->delalloc_bytes
);
4722 if (delalloc_bytes
== 0) {
4726 btrfs_wait_ordered_roots(root
->fs_info
, items
,
4732 while (delalloc_bytes
&& loops
< 3) {
4733 max_reclaim
= min(delalloc_bytes
, to_reclaim
);
4734 nr_pages
= max_reclaim
>> PAGE_SHIFT
;
4735 btrfs_writeback_inodes_sb_nr(root
, nr_pages
, items
);
4737 * We need to wait for the async pages to actually start before
4740 max_reclaim
= atomic_read(&root
->fs_info
->async_delalloc_pages
);
4744 if (max_reclaim
<= nr_pages
)
4747 max_reclaim
-= nr_pages
;
4749 wait_event(root
->fs_info
->async_submit_wait
,
4750 atomic_read(&root
->fs_info
->async_delalloc_pages
) <=
4754 flush
= BTRFS_RESERVE_FLUSH_ALL
;
4756 flush
= BTRFS_RESERVE_NO_FLUSH
;
4757 spin_lock(&space_info
->lock
);
4758 if (can_overcommit(root
, space_info
, orig
, flush
)) {
4759 spin_unlock(&space_info
->lock
);
4762 if (list_empty(&space_info
->tickets
) &&
4763 list_empty(&space_info
->priority_tickets
)) {
4764 spin_unlock(&space_info
->lock
);
4767 spin_unlock(&space_info
->lock
);
4770 if (wait_ordered
&& !trans
) {
4771 btrfs_wait_ordered_roots(root
->fs_info
, items
,
4774 time_left
= schedule_timeout_killable(1);
4778 delalloc_bytes
= percpu_counter_sum_positive(
4779 &root
->fs_info
->delalloc_bytes
);
4784 * maybe_commit_transaction - possibly commit the transaction if its ok to
4785 * @root - the root we're allocating for
4786 * @bytes - the number of bytes we want to reserve
4787 * @force - force the commit
4789 * This will check to make sure that committing the transaction will actually
4790 * get us somewhere and then commit the transaction if it does. Otherwise it
4791 * will return -ENOSPC.
4793 static int may_commit_transaction(struct btrfs_root
*root
,
4794 struct btrfs_space_info
*space_info
,
4795 u64 bytes
, int force
)
4797 struct btrfs_block_rsv
*delayed_rsv
= &root
->fs_info
->delayed_block_rsv
;
4798 struct btrfs_trans_handle
*trans
;
4800 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4807 /* See if there is enough pinned space to make this reservation */
4808 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4813 * See if there is some space in the delayed insertion reservation for
4816 if (space_info
!= delayed_rsv
->space_info
)
4819 spin_lock(&delayed_rsv
->lock
);
4820 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4821 bytes
- delayed_rsv
->size
) >= 0) {
4822 spin_unlock(&delayed_rsv
->lock
);
4825 spin_unlock(&delayed_rsv
->lock
);
4828 trans
= btrfs_join_transaction(root
);
4832 return btrfs_commit_transaction(trans
, root
);
4835 struct reserve_ticket
{
4838 struct list_head list
;
4839 wait_queue_head_t wait
;
4842 static int flush_space(struct btrfs_root
*root
,
4843 struct btrfs_space_info
*space_info
, u64 num_bytes
,
4844 u64 orig_bytes
, int state
)
4846 struct btrfs_trans_handle
*trans
;
4851 case FLUSH_DELAYED_ITEMS_NR
:
4852 case FLUSH_DELAYED_ITEMS
:
4853 if (state
== FLUSH_DELAYED_ITEMS_NR
)
4854 nr
= calc_reclaim_items_nr(root
, num_bytes
) * 2;
4858 trans
= btrfs_join_transaction(root
);
4859 if (IS_ERR(trans
)) {
4860 ret
= PTR_ERR(trans
);
4863 ret
= btrfs_run_delayed_items_nr(trans
, root
, nr
);
4864 btrfs_end_transaction(trans
, root
);
4866 case FLUSH_DELALLOC
:
4867 case FLUSH_DELALLOC_WAIT
:
4868 shrink_delalloc(root
, num_bytes
* 2, orig_bytes
,
4869 state
== FLUSH_DELALLOC_WAIT
);
4872 trans
= btrfs_join_transaction(root
);
4873 if (IS_ERR(trans
)) {
4874 ret
= PTR_ERR(trans
);
4877 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
4878 btrfs_get_alloc_profile(root
, 0),
4879 CHUNK_ALLOC_NO_FORCE
);
4880 btrfs_end_transaction(trans
, root
);
4881 if (ret
> 0 || ret
== -ENOSPC
)
4885 ret
= may_commit_transaction(root
, space_info
, orig_bytes
, 0);
4892 trace_btrfs_flush_space(root
->fs_info
, space_info
->flags
, num_bytes
,
4893 orig_bytes
, state
, ret
);
4898 btrfs_calc_reclaim_metadata_size(struct btrfs_root
*root
,
4899 struct btrfs_space_info
*space_info
)
4901 struct reserve_ticket
*ticket
;
4906 list_for_each_entry(ticket
, &space_info
->tickets
, list
)
4907 to_reclaim
+= ticket
->bytes
;
4908 list_for_each_entry(ticket
, &space_info
->priority_tickets
, list
)
4909 to_reclaim
+= ticket
->bytes
;
4913 to_reclaim
= min_t(u64
, num_online_cpus() * SZ_1M
, SZ_16M
);
4914 if (can_overcommit(root
, space_info
, to_reclaim
,
4915 BTRFS_RESERVE_FLUSH_ALL
))
4918 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4919 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4920 space_info
->bytes_may_use
;
4921 if (can_overcommit(root
, space_info
, SZ_1M
, BTRFS_RESERVE_FLUSH_ALL
))
4922 expected
= div_factor_fine(space_info
->total_bytes
, 95);
4924 expected
= div_factor_fine(space_info
->total_bytes
, 90);
4926 if (used
> expected
)
4927 to_reclaim
= used
- expected
;
4930 to_reclaim
= min(to_reclaim
, space_info
->bytes_may_use
+
4931 space_info
->bytes_reserved
);
4935 static inline int need_do_async_reclaim(struct btrfs_space_info
*space_info
,
4936 struct btrfs_root
*root
, u64 used
)
4938 u64 thresh
= div_factor_fine(space_info
->total_bytes
, 98);
4940 /* If we're just plain full then async reclaim just slows us down. */
4941 if ((space_info
->bytes_used
+ space_info
->bytes_reserved
) >= thresh
)
4944 if (!btrfs_calc_reclaim_metadata_size(root
, space_info
))
4947 return (used
>= thresh
&& !btrfs_fs_closing(root
->fs_info
) &&
4948 !test_bit(BTRFS_FS_STATE_REMOUNTING
,
4949 &root
->fs_info
->fs_state
));
4952 static void wake_all_tickets(struct list_head
*head
)
4954 struct reserve_ticket
*ticket
;
4956 while (!list_empty(head
)) {
4957 ticket
= list_first_entry(head
, struct reserve_ticket
, list
);
4958 list_del_init(&ticket
->list
);
4959 ticket
->error
= -ENOSPC
;
4960 wake_up(&ticket
->wait
);
4965 * This is for normal flushers, we can wait all goddamned day if we want to. We
4966 * will loop and continuously try to flush as long as we are making progress.
4967 * We count progress as clearing off tickets each time we have to loop.
4969 static void btrfs_async_reclaim_metadata_space(struct work_struct
*work
)
4971 struct reserve_ticket
*last_ticket
= NULL
;
4972 struct btrfs_fs_info
*fs_info
;
4973 struct btrfs_space_info
*space_info
;
4976 int commit_cycles
= 0;
4978 fs_info
= container_of(work
, struct btrfs_fs_info
, async_reclaim_work
);
4979 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4981 spin_lock(&space_info
->lock
);
4982 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
->fs_root
,
4985 space_info
->flush
= 0;
4986 spin_unlock(&space_info
->lock
);
4989 last_ticket
= list_first_entry(&space_info
->tickets
,
4990 struct reserve_ticket
, list
);
4991 spin_unlock(&space_info
->lock
);
4993 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
4995 struct reserve_ticket
*ticket
;
4998 ret
= flush_space(fs_info
->fs_root
, space_info
, to_reclaim
,
4999 to_reclaim
, flush_state
);
5000 spin_lock(&space_info
->lock
);
5001 if (list_empty(&space_info
->tickets
)) {
5002 space_info
->flush
= 0;
5003 spin_unlock(&space_info
->lock
);
5006 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
->fs_root
,
5008 ticket
= list_first_entry(&space_info
->tickets
,
5009 struct reserve_ticket
, list
);
5010 if (last_ticket
== ticket
) {
5013 last_ticket
= ticket
;
5014 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
5019 if (flush_state
> COMMIT_TRANS
) {
5021 if (commit_cycles
> 2) {
5022 wake_all_tickets(&space_info
->tickets
);
5023 space_info
->flush
= 0;
5025 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
5028 spin_unlock(&space_info
->lock
);
5029 } while (flush_state
<= COMMIT_TRANS
);
5032 void btrfs_init_async_reclaim_work(struct work_struct
*work
)
5034 INIT_WORK(work
, btrfs_async_reclaim_metadata_space
);
5037 static void priority_reclaim_metadata_space(struct btrfs_fs_info
*fs_info
,
5038 struct btrfs_space_info
*space_info
,
5039 struct reserve_ticket
*ticket
)
5042 int flush_state
= FLUSH_DELAYED_ITEMS_NR
;
5044 spin_lock(&space_info
->lock
);
5045 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
->fs_root
,
5048 spin_unlock(&space_info
->lock
);
5051 spin_unlock(&space_info
->lock
);
5054 flush_space(fs_info
->fs_root
, space_info
, to_reclaim
,
5055 to_reclaim
, flush_state
);
5057 spin_lock(&space_info
->lock
);
5058 if (ticket
->bytes
== 0) {
5059 spin_unlock(&space_info
->lock
);
5062 spin_unlock(&space_info
->lock
);
5065 * Priority flushers can't wait on delalloc without
5068 if (flush_state
== FLUSH_DELALLOC
||
5069 flush_state
== FLUSH_DELALLOC_WAIT
)
5070 flush_state
= ALLOC_CHUNK
;
5071 } while (flush_state
< COMMIT_TRANS
);
5074 static int wait_reserve_ticket(struct btrfs_fs_info
*fs_info
,
5075 struct btrfs_space_info
*space_info
,
5076 struct reserve_ticket
*ticket
, u64 orig_bytes
)
5082 spin_lock(&space_info
->lock
);
5083 while (ticket
->bytes
> 0 && ticket
->error
== 0) {
5084 ret
= prepare_to_wait_event(&ticket
->wait
, &wait
, TASK_KILLABLE
);
5089 spin_unlock(&space_info
->lock
);
5093 finish_wait(&ticket
->wait
, &wait
);
5094 spin_lock(&space_info
->lock
);
5097 ret
= ticket
->error
;
5098 if (!list_empty(&ticket
->list
))
5099 list_del_init(&ticket
->list
);
5100 if (ticket
->bytes
&& ticket
->bytes
< orig_bytes
) {
5101 u64 num_bytes
= orig_bytes
- ticket
->bytes
;
5102 space_info
->bytes_may_use
-= num_bytes
;
5103 trace_btrfs_space_reservation(fs_info
, "space_info",
5104 space_info
->flags
, num_bytes
, 0);
5106 spin_unlock(&space_info
->lock
);
5112 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5113 * @root - the root we're allocating for
5114 * @space_info - the space info we want to allocate from
5115 * @orig_bytes - the number of bytes we want
5116 * @flush - whether or not we can flush to make our reservation
5118 * This will reserve orig_bytes number of bytes from the space info associated
5119 * with the block_rsv. If there is not enough space it will make an attempt to
5120 * flush out space to make room. It will do this by flushing delalloc if
5121 * possible or committing the transaction. If flush is 0 then no attempts to
5122 * regain reservations will be made and this will fail if there is not enough
5125 static int __reserve_metadata_bytes(struct btrfs_root
*root
,
5126 struct btrfs_space_info
*space_info
,
5128 enum btrfs_reserve_flush_enum flush
)
5130 struct reserve_ticket ticket
;
5135 ASSERT(!current
->journal_info
|| flush
!= BTRFS_RESERVE_FLUSH_ALL
);
5137 spin_lock(&space_info
->lock
);
5139 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
5140 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
5141 space_info
->bytes_may_use
;
5144 * If we have enough space then hooray, make our reservation and carry
5145 * on. If not see if we can overcommit, and if we can, hooray carry on.
5146 * If not things get more complicated.
5148 if (used
+ orig_bytes
<= space_info
->total_bytes
) {
5149 space_info
->bytes_may_use
+= orig_bytes
;
5150 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
5151 space_info
->flags
, orig_bytes
,
5154 } else if (can_overcommit(root
, space_info
, orig_bytes
, flush
)) {
5155 space_info
->bytes_may_use
+= orig_bytes
;
5156 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
5157 space_info
->flags
, orig_bytes
,
5163 * If we couldn't make a reservation then setup our reservation ticket
5164 * and kick the async worker if it's not already running.
5166 * If we are a priority flusher then we just need to add our ticket to
5167 * the list and we will do our own flushing further down.
5169 if (ret
&& flush
!= BTRFS_RESERVE_NO_FLUSH
) {
5170 ticket
.bytes
= orig_bytes
;
5172 init_waitqueue_head(&ticket
.wait
);
5173 if (flush
== BTRFS_RESERVE_FLUSH_ALL
) {
5174 list_add_tail(&ticket
.list
, &space_info
->tickets
);
5175 if (!space_info
->flush
) {
5176 space_info
->flush
= 1;
5177 trace_btrfs_trigger_flush(root
->fs_info
,
5181 queue_work(system_unbound_wq
,
5182 &root
->fs_info
->async_reclaim_work
);
5185 list_add_tail(&ticket
.list
,
5186 &space_info
->priority_tickets
);
5188 } else if (!ret
&& space_info
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
5191 * We will do the space reservation dance during log replay,
5192 * which means we won't have fs_info->fs_root set, so don't do
5193 * the async reclaim as we will panic.
5195 if (!root
->fs_info
->log_root_recovering
&&
5196 need_do_async_reclaim(space_info
, root
, used
) &&
5197 !work_busy(&root
->fs_info
->async_reclaim_work
)) {
5198 trace_btrfs_trigger_flush(root
->fs_info
,
5202 queue_work(system_unbound_wq
,
5203 &root
->fs_info
->async_reclaim_work
);
5206 spin_unlock(&space_info
->lock
);
5207 if (!ret
|| flush
== BTRFS_RESERVE_NO_FLUSH
)
5210 if (flush
== BTRFS_RESERVE_FLUSH_ALL
)
5211 return wait_reserve_ticket(root
->fs_info
, space_info
, &ticket
,
5215 priority_reclaim_metadata_space(root
->fs_info
, space_info
, &ticket
);
5216 spin_lock(&space_info
->lock
);
5218 if (ticket
.bytes
< orig_bytes
) {
5219 u64 num_bytes
= orig_bytes
- ticket
.bytes
;
5220 space_info
->bytes_may_use
-= num_bytes
;
5221 trace_btrfs_space_reservation(root
->fs_info
,
5222 "space_info", space_info
->flags
,
5226 list_del_init(&ticket
.list
);
5229 spin_unlock(&space_info
->lock
);
5230 ASSERT(list_empty(&ticket
.list
));
5235 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5236 * @root - the root we're allocating for
5237 * @block_rsv - the block_rsv we're allocating for
5238 * @orig_bytes - the number of bytes we want
5239 * @flush - whether or not we can flush to make our reservation
5241 * This will reserve orgi_bytes number of bytes from the space info associated
5242 * with the block_rsv. If there is not enough space it will make an attempt to
5243 * flush out space to make room. It will do this by flushing delalloc if
5244 * possible or committing the transaction. If flush is 0 then no attempts to
5245 * regain reservations will be made and this will fail if there is not enough
5248 static int reserve_metadata_bytes(struct btrfs_root
*root
,
5249 struct btrfs_block_rsv
*block_rsv
,
5251 enum btrfs_reserve_flush_enum flush
)
5255 ret
= __reserve_metadata_bytes(root
, block_rsv
->space_info
, orig_bytes
,
5257 if (ret
== -ENOSPC
&&
5258 unlikely(root
->orphan_cleanup_state
== ORPHAN_CLEANUP_STARTED
)) {
5259 struct btrfs_block_rsv
*global_rsv
=
5260 &root
->fs_info
->global_block_rsv
;
5262 if (block_rsv
!= global_rsv
&&
5263 !block_rsv_use_bytes(global_rsv
, orig_bytes
))
5267 trace_btrfs_space_reservation(root
->fs_info
,
5268 "space_info:enospc",
5269 block_rsv
->space_info
->flags
,
5274 static struct btrfs_block_rsv
*get_block_rsv(
5275 const struct btrfs_trans_handle
*trans
,
5276 const struct btrfs_root
*root
)
5278 struct btrfs_block_rsv
*block_rsv
= NULL
;
5280 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
5281 (root
== root
->fs_info
->csum_root
&& trans
->adding_csums
) ||
5282 (root
== root
->fs_info
->uuid_root
))
5283 block_rsv
= trans
->block_rsv
;
5286 block_rsv
= root
->block_rsv
;
5289 block_rsv
= &root
->fs_info
->empty_block_rsv
;
5294 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
5298 spin_lock(&block_rsv
->lock
);
5299 if (block_rsv
->reserved
>= num_bytes
) {
5300 block_rsv
->reserved
-= num_bytes
;
5301 if (block_rsv
->reserved
< block_rsv
->size
)
5302 block_rsv
->full
= 0;
5305 spin_unlock(&block_rsv
->lock
);
5309 static void block_rsv_add_bytes(struct btrfs_block_rsv
*block_rsv
,
5310 u64 num_bytes
, int update_size
)
5312 spin_lock(&block_rsv
->lock
);
5313 block_rsv
->reserved
+= num_bytes
;
5315 block_rsv
->size
+= num_bytes
;
5316 else if (block_rsv
->reserved
>= block_rsv
->size
)
5317 block_rsv
->full
= 1;
5318 spin_unlock(&block_rsv
->lock
);
5321 int btrfs_cond_migrate_bytes(struct btrfs_fs_info
*fs_info
,
5322 struct btrfs_block_rsv
*dest
, u64 num_bytes
,
5325 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5328 if (global_rsv
->space_info
!= dest
->space_info
)
5331 spin_lock(&global_rsv
->lock
);
5332 min_bytes
= div_factor(global_rsv
->size
, min_factor
);
5333 if (global_rsv
->reserved
< min_bytes
+ num_bytes
) {
5334 spin_unlock(&global_rsv
->lock
);
5337 global_rsv
->reserved
-= num_bytes
;
5338 if (global_rsv
->reserved
< global_rsv
->size
)
5339 global_rsv
->full
= 0;
5340 spin_unlock(&global_rsv
->lock
);
5342 block_rsv_add_bytes(dest
, num_bytes
, 1);
5347 * This is for space we already have accounted in space_info->bytes_may_use, so
5348 * basically when we're returning space from block_rsv's.
5350 static void space_info_add_old_bytes(struct btrfs_fs_info
*fs_info
,
5351 struct btrfs_space_info
*space_info
,
5354 struct reserve_ticket
*ticket
;
5355 struct list_head
*head
;
5357 enum btrfs_reserve_flush_enum flush
= BTRFS_RESERVE_NO_FLUSH
;
5358 bool check_overcommit
= false;
5360 spin_lock(&space_info
->lock
);
5361 head
= &space_info
->priority_tickets
;
5364 * If we are over our limit then we need to check and see if we can
5365 * overcommit, and if we can't then we just need to free up our space
5366 * and not satisfy any requests.
5368 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
5369 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
5370 space_info
->bytes_may_use
;
5371 if (used
- num_bytes
>= space_info
->total_bytes
)
5372 check_overcommit
= true;
5374 while (!list_empty(head
) && num_bytes
) {
5375 ticket
= list_first_entry(head
, struct reserve_ticket
,
5378 * We use 0 bytes because this space is already reserved, so
5379 * adding the ticket space would be a double count.
5381 if (check_overcommit
&&
5382 !can_overcommit(fs_info
->extent_root
, space_info
, 0,
5385 if (num_bytes
>= ticket
->bytes
) {
5386 list_del_init(&ticket
->list
);
5387 num_bytes
-= ticket
->bytes
;
5389 wake_up(&ticket
->wait
);
5391 ticket
->bytes
-= num_bytes
;
5396 if (num_bytes
&& head
== &space_info
->priority_tickets
) {
5397 head
= &space_info
->tickets
;
5398 flush
= BTRFS_RESERVE_FLUSH_ALL
;
5401 space_info
->bytes_may_use
-= num_bytes
;
5402 trace_btrfs_space_reservation(fs_info
, "space_info",
5403 space_info
->flags
, num_bytes
, 0);
5404 spin_unlock(&space_info
->lock
);
5408 * This is for newly allocated space that isn't accounted in
5409 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5410 * we use this helper.
5412 static void space_info_add_new_bytes(struct btrfs_fs_info
*fs_info
,
5413 struct btrfs_space_info
*space_info
,
5416 struct reserve_ticket
*ticket
;
5417 struct list_head
*head
= &space_info
->priority_tickets
;
5420 while (!list_empty(head
) && num_bytes
) {
5421 ticket
= list_first_entry(head
, struct reserve_ticket
,
5423 if (num_bytes
>= ticket
->bytes
) {
5424 trace_btrfs_space_reservation(fs_info
, "space_info",
5427 list_del_init(&ticket
->list
);
5428 num_bytes
-= ticket
->bytes
;
5429 space_info
->bytes_may_use
+= ticket
->bytes
;
5431 wake_up(&ticket
->wait
);
5433 trace_btrfs_space_reservation(fs_info
, "space_info",
5436 space_info
->bytes_may_use
+= num_bytes
;
5437 ticket
->bytes
-= num_bytes
;
5442 if (num_bytes
&& head
== &space_info
->priority_tickets
) {
5443 head
= &space_info
->tickets
;
5448 static void block_rsv_release_bytes(struct btrfs_fs_info
*fs_info
,
5449 struct btrfs_block_rsv
*block_rsv
,
5450 struct btrfs_block_rsv
*dest
, u64 num_bytes
)
5452 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
5454 spin_lock(&block_rsv
->lock
);
5455 if (num_bytes
== (u64
)-1)
5456 num_bytes
= block_rsv
->size
;
5457 block_rsv
->size
-= num_bytes
;
5458 if (block_rsv
->reserved
>= block_rsv
->size
) {
5459 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
5460 block_rsv
->reserved
= block_rsv
->size
;
5461 block_rsv
->full
= 1;
5465 spin_unlock(&block_rsv
->lock
);
5467 if (num_bytes
> 0) {
5469 spin_lock(&dest
->lock
);
5473 bytes_to_add
= dest
->size
- dest
->reserved
;
5474 bytes_to_add
= min(num_bytes
, bytes_to_add
);
5475 dest
->reserved
+= bytes_to_add
;
5476 if (dest
->reserved
>= dest
->size
)
5478 num_bytes
-= bytes_to_add
;
5480 spin_unlock(&dest
->lock
);
5483 space_info_add_old_bytes(fs_info
, space_info
,
5488 int btrfs_block_rsv_migrate(struct btrfs_block_rsv
*src
,
5489 struct btrfs_block_rsv
*dst
, u64 num_bytes
,
5494 ret
= block_rsv_use_bytes(src
, num_bytes
);
5498 block_rsv_add_bytes(dst
, num_bytes
, update_size
);
5502 void btrfs_init_block_rsv(struct btrfs_block_rsv
*rsv
, unsigned short type
)
5504 memset(rsv
, 0, sizeof(*rsv
));
5505 spin_lock_init(&rsv
->lock
);
5509 struct btrfs_block_rsv
*btrfs_alloc_block_rsv(struct btrfs_root
*root
,
5510 unsigned short type
)
5512 struct btrfs_block_rsv
*block_rsv
;
5513 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5515 block_rsv
= kmalloc(sizeof(*block_rsv
), GFP_NOFS
);
5519 btrfs_init_block_rsv(block_rsv
, type
);
5520 block_rsv
->space_info
= __find_space_info(fs_info
,
5521 BTRFS_BLOCK_GROUP_METADATA
);
5525 void btrfs_free_block_rsv(struct btrfs_root
*root
,
5526 struct btrfs_block_rsv
*rsv
)
5530 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
5534 void __btrfs_free_block_rsv(struct btrfs_block_rsv
*rsv
)
5539 int btrfs_block_rsv_add(struct btrfs_root
*root
,
5540 struct btrfs_block_rsv
*block_rsv
, u64 num_bytes
,
5541 enum btrfs_reserve_flush_enum flush
)
5548 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
5550 block_rsv_add_bytes(block_rsv
, num_bytes
, 1);
5557 int btrfs_block_rsv_check(struct btrfs_root
*root
,
5558 struct btrfs_block_rsv
*block_rsv
, int min_factor
)
5566 spin_lock(&block_rsv
->lock
);
5567 num_bytes
= div_factor(block_rsv
->size
, min_factor
);
5568 if (block_rsv
->reserved
>= num_bytes
)
5570 spin_unlock(&block_rsv
->lock
);
5575 int btrfs_block_rsv_refill(struct btrfs_root
*root
,
5576 struct btrfs_block_rsv
*block_rsv
, u64 min_reserved
,
5577 enum btrfs_reserve_flush_enum flush
)
5585 spin_lock(&block_rsv
->lock
);
5586 num_bytes
= min_reserved
;
5587 if (block_rsv
->reserved
>= num_bytes
)
5590 num_bytes
-= block_rsv
->reserved
;
5591 spin_unlock(&block_rsv
->lock
);
5596 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
5598 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
5605 void btrfs_block_rsv_release(struct btrfs_root
*root
,
5606 struct btrfs_block_rsv
*block_rsv
,
5609 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
5610 if (global_rsv
== block_rsv
||
5611 block_rsv
->space_info
!= global_rsv
->space_info
)
5613 block_rsv_release_bytes(root
->fs_info
, block_rsv
, global_rsv
,
5617 static void update_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5619 struct btrfs_block_rsv
*block_rsv
= &fs_info
->global_block_rsv
;
5620 struct btrfs_space_info
*sinfo
= block_rsv
->space_info
;
5624 * The global block rsv is based on the size of the extent tree, the
5625 * checksum tree and the root tree. If the fs is empty we want to set
5626 * it to a minimal amount for safety.
5628 num_bytes
= btrfs_root_used(&fs_info
->extent_root
->root_item
) +
5629 btrfs_root_used(&fs_info
->csum_root
->root_item
) +
5630 btrfs_root_used(&fs_info
->tree_root
->root_item
);
5631 num_bytes
= max_t(u64
, num_bytes
, SZ_16M
);
5633 spin_lock(&sinfo
->lock
);
5634 spin_lock(&block_rsv
->lock
);
5636 block_rsv
->size
= min_t(u64
, num_bytes
, SZ_512M
);
5638 if (block_rsv
->reserved
< block_rsv
->size
) {
5639 num_bytes
= sinfo
->bytes_used
+ sinfo
->bytes_pinned
+
5640 sinfo
->bytes_reserved
+ sinfo
->bytes_readonly
+
5641 sinfo
->bytes_may_use
;
5642 if (sinfo
->total_bytes
> num_bytes
) {
5643 num_bytes
= sinfo
->total_bytes
- num_bytes
;
5644 num_bytes
= min(num_bytes
,
5645 block_rsv
->size
- block_rsv
->reserved
);
5646 block_rsv
->reserved
+= num_bytes
;
5647 sinfo
->bytes_may_use
+= num_bytes
;
5648 trace_btrfs_space_reservation(fs_info
, "space_info",
5649 sinfo
->flags
, num_bytes
,
5652 } else if (block_rsv
->reserved
> block_rsv
->size
) {
5653 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
5654 sinfo
->bytes_may_use
-= num_bytes
;
5655 trace_btrfs_space_reservation(fs_info
, "space_info",
5656 sinfo
->flags
, num_bytes
, 0);
5657 block_rsv
->reserved
= block_rsv
->size
;
5660 if (block_rsv
->reserved
== block_rsv
->size
)
5661 block_rsv
->full
= 1;
5663 block_rsv
->full
= 0;
5665 spin_unlock(&block_rsv
->lock
);
5666 spin_unlock(&sinfo
->lock
);
5669 static void init_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5671 struct btrfs_space_info
*space_info
;
5673 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
5674 fs_info
->chunk_block_rsv
.space_info
= space_info
;
5676 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
5677 fs_info
->global_block_rsv
.space_info
= space_info
;
5678 fs_info
->delalloc_block_rsv
.space_info
= space_info
;
5679 fs_info
->trans_block_rsv
.space_info
= space_info
;
5680 fs_info
->empty_block_rsv
.space_info
= space_info
;
5681 fs_info
->delayed_block_rsv
.space_info
= space_info
;
5683 fs_info
->extent_root
->block_rsv
= &fs_info
->global_block_rsv
;
5684 fs_info
->csum_root
->block_rsv
= &fs_info
->global_block_rsv
;
5685 fs_info
->dev_root
->block_rsv
= &fs_info
->global_block_rsv
;
5686 fs_info
->tree_root
->block_rsv
= &fs_info
->global_block_rsv
;
5687 if (fs_info
->quota_root
)
5688 fs_info
->quota_root
->block_rsv
= &fs_info
->global_block_rsv
;
5689 fs_info
->chunk_root
->block_rsv
= &fs_info
->chunk_block_rsv
;
5691 update_global_block_rsv(fs_info
);
5694 static void release_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5696 block_rsv_release_bytes(fs_info
, &fs_info
->global_block_rsv
, NULL
,
5698 WARN_ON(fs_info
->delalloc_block_rsv
.size
> 0);
5699 WARN_ON(fs_info
->delalloc_block_rsv
.reserved
> 0);
5700 WARN_ON(fs_info
->trans_block_rsv
.size
> 0);
5701 WARN_ON(fs_info
->trans_block_rsv
.reserved
> 0);
5702 WARN_ON(fs_info
->chunk_block_rsv
.size
> 0);
5703 WARN_ON(fs_info
->chunk_block_rsv
.reserved
> 0);
5704 WARN_ON(fs_info
->delayed_block_rsv
.size
> 0);
5705 WARN_ON(fs_info
->delayed_block_rsv
.reserved
> 0);
5708 void btrfs_trans_release_metadata(struct btrfs_trans_handle
*trans
,
5709 struct btrfs_root
*root
)
5711 if (!trans
->block_rsv
)
5714 if (!trans
->bytes_reserved
)
5717 trace_btrfs_space_reservation(root
->fs_info
, "transaction",
5718 trans
->transid
, trans
->bytes_reserved
, 0);
5719 btrfs_block_rsv_release(root
, trans
->block_rsv
, trans
->bytes_reserved
);
5720 trans
->bytes_reserved
= 0;
5724 * To be called after all the new block groups attached to the transaction
5725 * handle have been created (btrfs_create_pending_block_groups()).
5727 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle
*trans
)
5729 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
5731 if (!trans
->chunk_bytes_reserved
)
5734 WARN_ON_ONCE(!list_empty(&trans
->new_bgs
));
5736 block_rsv_release_bytes(fs_info
, &fs_info
->chunk_block_rsv
, NULL
,
5737 trans
->chunk_bytes_reserved
);
5738 trans
->chunk_bytes_reserved
= 0;
5741 /* Can only return 0 or -ENOSPC */
5742 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle
*trans
,
5743 struct inode
*inode
)
5745 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5747 * We always use trans->block_rsv here as we will have reserved space
5748 * for our orphan when starting the transaction, using get_block_rsv()
5749 * here will sometimes make us choose the wrong block rsv as we could be
5750 * doing a reloc inode for a non refcounted root.
5752 struct btrfs_block_rsv
*src_rsv
= trans
->block_rsv
;
5753 struct btrfs_block_rsv
*dst_rsv
= root
->orphan_block_rsv
;
5756 * We need to hold space in order to delete our orphan item once we've
5757 * added it, so this takes the reservation so we can release it later
5758 * when we are truly done with the orphan item.
5760 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
5761 trace_btrfs_space_reservation(root
->fs_info
, "orphan",
5762 btrfs_ino(inode
), num_bytes
, 1);
5763 return btrfs_block_rsv_migrate(src_rsv
, dst_rsv
, num_bytes
, 1);
5766 void btrfs_orphan_release_metadata(struct inode
*inode
)
5768 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5769 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
5770 trace_btrfs_space_reservation(root
->fs_info
, "orphan",
5771 btrfs_ino(inode
), num_bytes
, 0);
5772 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
, num_bytes
);
5776 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5777 * root: the root of the parent directory
5778 * rsv: block reservation
5779 * items: the number of items that we need do reservation
5780 * qgroup_reserved: used to return the reserved size in qgroup
5782 * This function is used to reserve the space for snapshot/subvolume
5783 * creation and deletion. Those operations are different with the
5784 * common file/directory operations, they change two fs/file trees
5785 * and root tree, the number of items that the qgroup reserves is
5786 * different with the free space reservation. So we can not use
5787 * the space reservation mechanism in start_transaction().
5789 int btrfs_subvolume_reserve_metadata(struct btrfs_root
*root
,
5790 struct btrfs_block_rsv
*rsv
,
5792 u64
*qgroup_reserved
,
5793 bool use_global_rsv
)
5797 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
5799 if (root
->fs_info
->quota_enabled
) {
5800 /* One for parent inode, two for dir entries */
5801 num_bytes
= 3 * root
->nodesize
;
5802 ret
= btrfs_qgroup_reserve_meta(root
, num_bytes
);
5809 *qgroup_reserved
= num_bytes
;
5811 num_bytes
= btrfs_calc_trans_metadata_size(root
, items
);
5812 rsv
->space_info
= __find_space_info(root
->fs_info
,
5813 BTRFS_BLOCK_GROUP_METADATA
);
5814 ret
= btrfs_block_rsv_add(root
, rsv
, num_bytes
,
5815 BTRFS_RESERVE_FLUSH_ALL
);
5817 if (ret
== -ENOSPC
&& use_global_rsv
)
5818 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
, num_bytes
, 1);
5820 if (ret
&& *qgroup_reserved
)
5821 btrfs_qgroup_free_meta(root
, *qgroup_reserved
);
5826 void btrfs_subvolume_release_metadata(struct btrfs_root
*root
,
5827 struct btrfs_block_rsv
*rsv
,
5828 u64 qgroup_reserved
)
5830 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
5834 * drop_outstanding_extent - drop an outstanding extent
5835 * @inode: the inode we're dropping the extent for
5836 * @num_bytes: the number of bytes we're releasing.
5838 * This is called when we are freeing up an outstanding extent, either called
5839 * after an error or after an extent is written. This will return the number of
5840 * reserved extents that need to be freed. This must be called with
5841 * BTRFS_I(inode)->lock held.
5843 static unsigned drop_outstanding_extent(struct inode
*inode
, u64 num_bytes
)
5845 unsigned drop_inode_space
= 0;
5846 unsigned dropped_extents
= 0;
5847 unsigned num_extents
= 0;
5849 num_extents
= (unsigned)div64_u64(num_bytes
+
5850 BTRFS_MAX_EXTENT_SIZE
- 1,
5851 BTRFS_MAX_EXTENT_SIZE
);
5852 ASSERT(num_extents
);
5853 ASSERT(BTRFS_I(inode
)->outstanding_extents
>= num_extents
);
5854 BTRFS_I(inode
)->outstanding_extents
-= num_extents
;
5856 if (BTRFS_I(inode
)->outstanding_extents
== 0 &&
5857 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5858 &BTRFS_I(inode
)->runtime_flags
))
5859 drop_inode_space
= 1;
5862 * If we have more or the same amount of outstanding extents than we have
5863 * reserved then we need to leave the reserved extents count alone.
5865 if (BTRFS_I(inode
)->outstanding_extents
>=
5866 BTRFS_I(inode
)->reserved_extents
)
5867 return drop_inode_space
;
5869 dropped_extents
= BTRFS_I(inode
)->reserved_extents
-
5870 BTRFS_I(inode
)->outstanding_extents
;
5871 BTRFS_I(inode
)->reserved_extents
-= dropped_extents
;
5872 return dropped_extents
+ drop_inode_space
;
5876 * calc_csum_metadata_size - return the amount of metadata space that must be
5877 * reserved/freed for the given bytes.
5878 * @inode: the inode we're manipulating
5879 * @num_bytes: the number of bytes in question
5880 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5882 * This adjusts the number of csum_bytes in the inode and then returns the
5883 * correct amount of metadata that must either be reserved or freed. We
5884 * calculate how many checksums we can fit into one leaf and then divide the
5885 * number of bytes that will need to be checksumed by this value to figure out
5886 * how many checksums will be required. If we are adding bytes then the number
5887 * may go up and we will return the number of additional bytes that must be
5888 * reserved. If it is going down we will return the number of bytes that must
5891 * This must be called with BTRFS_I(inode)->lock held.
5893 static u64
calc_csum_metadata_size(struct inode
*inode
, u64 num_bytes
,
5896 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5897 u64 old_csums
, num_csums
;
5899 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
&&
5900 BTRFS_I(inode
)->csum_bytes
== 0)
5903 old_csums
= btrfs_csum_bytes_to_leaves(root
, BTRFS_I(inode
)->csum_bytes
);
5905 BTRFS_I(inode
)->csum_bytes
+= num_bytes
;
5907 BTRFS_I(inode
)->csum_bytes
-= num_bytes
;
5908 num_csums
= btrfs_csum_bytes_to_leaves(root
, BTRFS_I(inode
)->csum_bytes
);
5910 /* No change, no need to reserve more */
5911 if (old_csums
== num_csums
)
5915 return btrfs_calc_trans_metadata_size(root
,
5916 num_csums
- old_csums
);
5918 return btrfs_calc_trans_metadata_size(root
, old_csums
- num_csums
);
5921 int btrfs_delalloc_reserve_metadata(struct inode
*inode
, u64 num_bytes
)
5923 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5924 struct btrfs_block_rsv
*block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
5927 unsigned nr_extents
= 0;
5928 enum btrfs_reserve_flush_enum flush
= BTRFS_RESERVE_FLUSH_ALL
;
5930 bool delalloc_lock
= true;
5933 bool release_extra
= false;
5935 /* If we are a free space inode we need to not flush since we will be in
5936 * the middle of a transaction commit. We also don't need the delalloc
5937 * mutex since we won't race with anybody. We need this mostly to make
5938 * lockdep shut its filthy mouth.
5940 * If we have a transaction open (can happen if we call truncate_block
5941 * from truncate), then we need FLUSH_LIMIT so we don't deadlock.
5943 if (btrfs_is_free_space_inode(inode
)) {
5944 flush
= BTRFS_RESERVE_NO_FLUSH
;
5945 delalloc_lock
= false;
5946 } else if (current
->journal_info
) {
5947 flush
= BTRFS_RESERVE_FLUSH_LIMIT
;
5950 if (flush
!= BTRFS_RESERVE_NO_FLUSH
&&
5951 btrfs_transaction_in_commit(root
->fs_info
))
5952 schedule_timeout(1);
5955 mutex_lock(&BTRFS_I(inode
)->delalloc_mutex
);
5957 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
5959 spin_lock(&BTRFS_I(inode
)->lock
);
5960 nr_extents
= (unsigned)div64_u64(num_bytes
+
5961 BTRFS_MAX_EXTENT_SIZE
- 1,
5962 BTRFS_MAX_EXTENT_SIZE
);
5963 BTRFS_I(inode
)->outstanding_extents
+= nr_extents
;
5966 if (BTRFS_I(inode
)->outstanding_extents
>
5967 BTRFS_I(inode
)->reserved_extents
)
5968 nr_extents
+= BTRFS_I(inode
)->outstanding_extents
-
5969 BTRFS_I(inode
)->reserved_extents
;
5971 /* We always want to reserve a slot for updating the inode. */
5972 to_reserve
= btrfs_calc_trans_metadata_size(root
, nr_extents
+ 1);
5973 to_reserve
+= calc_csum_metadata_size(inode
, num_bytes
, 1);
5974 csum_bytes
= BTRFS_I(inode
)->csum_bytes
;
5975 spin_unlock(&BTRFS_I(inode
)->lock
);
5977 if (root
->fs_info
->quota_enabled
) {
5978 ret
= btrfs_qgroup_reserve_meta(root
,
5979 nr_extents
* root
->nodesize
);
5984 ret
= btrfs_block_rsv_add(root
, block_rsv
, to_reserve
, flush
);
5985 if (unlikely(ret
)) {
5986 btrfs_qgroup_free_meta(root
, nr_extents
* root
->nodesize
);
5990 spin_lock(&BTRFS_I(inode
)->lock
);
5991 if (test_and_set_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5992 &BTRFS_I(inode
)->runtime_flags
)) {
5993 to_reserve
-= btrfs_calc_trans_metadata_size(root
, 1);
5994 release_extra
= true;
5996 BTRFS_I(inode
)->reserved_extents
+= nr_extents
;
5997 spin_unlock(&BTRFS_I(inode
)->lock
);
6000 mutex_unlock(&BTRFS_I(inode
)->delalloc_mutex
);
6003 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
6004 btrfs_ino(inode
), to_reserve
, 1);
6006 btrfs_block_rsv_release(root
, block_rsv
,
6007 btrfs_calc_trans_metadata_size(root
,
6012 spin_lock(&BTRFS_I(inode
)->lock
);
6013 dropped
= drop_outstanding_extent(inode
, num_bytes
);
6015 * If the inodes csum_bytes is the same as the original
6016 * csum_bytes then we know we haven't raced with any free()ers
6017 * so we can just reduce our inodes csum bytes and carry on.
6019 if (BTRFS_I(inode
)->csum_bytes
== csum_bytes
) {
6020 calc_csum_metadata_size(inode
, num_bytes
, 0);
6022 u64 orig_csum_bytes
= BTRFS_I(inode
)->csum_bytes
;
6026 * This is tricky, but first we need to figure out how much we
6027 * freed from any free-ers that occurred during this
6028 * reservation, so we reset ->csum_bytes to the csum_bytes
6029 * before we dropped our lock, and then call the free for the
6030 * number of bytes that were freed while we were trying our
6033 bytes
= csum_bytes
- BTRFS_I(inode
)->csum_bytes
;
6034 BTRFS_I(inode
)->csum_bytes
= csum_bytes
;
6035 to_free
= calc_csum_metadata_size(inode
, bytes
, 0);
6039 * Now we need to see how much we would have freed had we not
6040 * been making this reservation and our ->csum_bytes were not
6041 * artificially inflated.
6043 BTRFS_I(inode
)->csum_bytes
= csum_bytes
- num_bytes
;
6044 bytes
= csum_bytes
- orig_csum_bytes
;
6045 bytes
= calc_csum_metadata_size(inode
, bytes
, 0);
6048 * Now reset ->csum_bytes to what it should be. If bytes is
6049 * more than to_free then we would have freed more space had we
6050 * not had an artificially high ->csum_bytes, so we need to free
6051 * the remainder. If bytes is the same or less then we don't
6052 * need to do anything, the other free-ers did the correct
6055 BTRFS_I(inode
)->csum_bytes
= orig_csum_bytes
- num_bytes
;
6056 if (bytes
> to_free
)
6057 to_free
= bytes
- to_free
;
6061 spin_unlock(&BTRFS_I(inode
)->lock
);
6063 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
6066 btrfs_block_rsv_release(root
, block_rsv
, to_free
);
6067 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
6068 btrfs_ino(inode
), to_free
, 0);
6071 mutex_unlock(&BTRFS_I(inode
)->delalloc_mutex
);
6076 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
6077 * @inode: the inode to release the reservation for
6078 * @num_bytes: the number of bytes we're releasing
6080 * This will release the metadata reservation for an inode. This can be called
6081 * once we complete IO for a given set of bytes to release their metadata
6084 void btrfs_delalloc_release_metadata(struct inode
*inode
, u64 num_bytes
)
6086 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6090 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
6091 spin_lock(&BTRFS_I(inode
)->lock
);
6092 dropped
= drop_outstanding_extent(inode
, num_bytes
);
6095 to_free
= calc_csum_metadata_size(inode
, num_bytes
, 0);
6096 spin_unlock(&BTRFS_I(inode
)->lock
);
6098 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
6100 if (btrfs_is_testing(root
->fs_info
))
6103 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
6104 btrfs_ino(inode
), to_free
, 0);
6106 btrfs_block_rsv_release(root
, &root
->fs_info
->delalloc_block_rsv
,
6111 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6113 * @inode: inode we're writing to
6114 * @start: start range we are writing to
6115 * @len: how long the range we are writing to
6117 * TODO: This function will finally replace old btrfs_delalloc_reserve_space()
6119 * This will do the following things
6121 * o reserve space in data space info for num bytes
6122 * and reserve precious corresponding qgroup space
6123 * (Done in check_data_free_space)
6125 * o reserve space for metadata space, based on the number of outstanding
6126 * extents and how much csums will be needed
6127 * also reserve metadata space in a per root over-reserve method.
6128 * o add to the inodes->delalloc_bytes
6129 * o add it to the fs_info's delalloc inodes list.
6130 * (Above 3 all done in delalloc_reserve_metadata)
6132 * Return 0 for success
6133 * Return <0 for error(-ENOSPC or -EQUOT)
6135 int btrfs_delalloc_reserve_space(struct inode
*inode
, u64 start
, u64 len
)
6139 ret
= btrfs_check_data_free_space(inode
, start
, len
);
6142 ret
= btrfs_delalloc_reserve_metadata(inode
, len
);
6144 btrfs_free_reserved_data_space(inode
, start
, len
);
6149 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6150 * @inode: inode we're releasing space for
6151 * @start: start position of the space already reserved
6152 * @len: the len of the space already reserved
6154 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
6155 * called in the case that we don't need the metadata AND data reservations
6156 * anymore. So if there is an error or we insert an inline extent.
6158 * This function will release the metadata space that was not used and will
6159 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
6160 * list if there are no delalloc bytes left.
6161 * Also it will handle the qgroup reserved space.
6163 void btrfs_delalloc_release_space(struct inode
*inode
, u64 start
, u64 len
)
6165 btrfs_delalloc_release_metadata(inode
, len
);
6166 btrfs_free_reserved_data_space(inode
, start
, len
);
6169 static int update_block_group(struct btrfs_trans_handle
*trans
,
6170 struct btrfs_root
*root
, u64 bytenr
,
6171 u64 num_bytes
, int alloc
)
6173 struct btrfs_block_group_cache
*cache
= NULL
;
6174 struct btrfs_fs_info
*info
= root
->fs_info
;
6175 u64 total
= num_bytes
;
6180 /* block accounting for super block */
6181 spin_lock(&info
->delalloc_root_lock
);
6182 old_val
= btrfs_super_bytes_used(info
->super_copy
);
6184 old_val
+= num_bytes
;
6186 old_val
-= num_bytes
;
6187 btrfs_set_super_bytes_used(info
->super_copy
, old_val
);
6188 spin_unlock(&info
->delalloc_root_lock
);
6191 cache
= btrfs_lookup_block_group(info
, bytenr
);
6194 if (cache
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
6195 BTRFS_BLOCK_GROUP_RAID1
|
6196 BTRFS_BLOCK_GROUP_RAID10
))
6201 * If this block group has free space cache written out, we
6202 * need to make sure to load it if we are removing space. This
6203 * is because we need the unpinning stage to actually add the
6204 * space back to the block group, otherwise we will leak space.
6206 if (!alloc
&& cache
->cached
== BTRFS_CACHE_NO
)
6207 cache_block_group(cache
, 1);
6209 byte_in_group
= bytenr
- cache
->key
.objectid
;
6210 WARN_ON(byte_in_group
> cache
->key
.offset
);
6212 spin_lock(&cache
->space_info
->lock
);
6213 spin_lock(&cache
->lock
);
6215 if (btrfs_test_opt(root
->fs_info
, SPACE_CACHE
) &&
6216 cache
->disk_cache_state
< BTRFS_DC_CLEAR
)
6217 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
6219 old_val
= btrfs_block_group_used(&cache
->item
);
6220 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
6222 old_val
+= num_bytes
;
6223 btrfs_set_block_group_used(&cache
->item
, old_val
);
6224 cache
->reserved
-= num_bytes
;
6225 cache
->space_info
->bytes_reserved
-= num_bytes
;
6226 cache
->space_info
->bytes_used
+= num_bytes
;
6227 cache
->space_info
->disk_used
+= num_bytes
* factor
;
6228 spin_unlock(&cache
->lock
);
6229 spin_unlock(&cache
->space_info
->lock
);
6231 old_val
-= num_bytes
;
6232 btrfs_set_block_group_used(&cache
->item
, old_val
);
6233 cache
->pinned
+= num_bytes
;
6234 cache
->space_info
->bytes_pinned
+= num_bytes
;
6235 cache
->space_info
->bytes_used
-= num_bytes
;
6236 cache
->space_info
->disk_used
-= num_bytes
* factor
;
6237 spin_unlock(&cache
->lock
);
6238 spin_unlock(&cache
->space_info
->lock
);
6240 trace_btrfs_space_reservation(root
->fs_info
, "pinned",
6241 cache
->space_info
->flags
,
6243 set_extent_dirty(info
->pinned_extents
,
6244 bytenr
, bytenr
+ num_bytes
- 1,
6245 GFP_NOFS
| __GFP_NOFAIL
);
6248 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
6249 if (list_empty(&cache
->dirty_list
)) {
6250 list_add_tail(&cache
->dirty_list
,
6251 &trans
->transaction
->dirty_bgs
);
6252 trans
->transaction
->num_dirty_bgs
++;
6253 btrfs_get_block_group(cache
);
6255 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
6258 * No longer have used bytes in this block group, queue it for
6259 * deletion. We do this after adding the block group to the
6260 * dirty list to avoid races between cleaner kthread and space
6263 if (!alloc
&& old_val
== 0) {
6264 spin_lock(&info
->unused_bgs_lock
);
6265 if (list_empty(&cache
->bg_list
)) {
6266 btrfs_get_block_group(cache
);
6267 list_add_tail(&cache
->bg_list
,
6270 spin_unlock(&info
->unused_bgs_lock
);
6273 btrfs_put_block_group(cache
);
6275 bytenr
+= num_bytes
;
6280 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
6282 struct btrfs_block_group_cache
*cache
;
6285 spin_lock(&root
->fs_info
->block_group_cache_lock
);
6286 bytenr
= root
->fs_info
->first_logical_byte
;
6287 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
6289 if (bytenr
< (u64
)-1)
6292 cache
= btrfs_lookup_first_block_group(root
->fs_info
, search_start
);
6296 bytenr
= cache
->key
.objectid
;
6297 btrfs_put_block_group(cache
);
6302 static int pin_down_extent(struct btrfs_root
*root
,
6303 struct btrfs_block_group_cache
*cache
,
6304 u64 bytenr
, u64 num_bytes
, int reserved
)
6306 spin_lock(&cache
->space_info
->lock
);
6307 spin_lock(&cache
->lock
);
6308 cache
->pinned
+= num_bytes
;
6309 cache
->space_info
->bytes_pinned
+= num_bytes
;
6311 cache
->reserved
-= num_bytes
;
6312 cache
->space_info
->bytes_reserved
-= num_bytes
;
6314 spin_unlock(&cache
->lock
);
6315 spin_unlock(&cache
->space_info
->lock
);
6317 trace_btrfs_space_reservation(root
->fs_info
, "pinned",
6318 cache
->space_info
->flags
, num_bytes
, 1);
6319 set_extent_dirty(root
->fs_info
->pinned_extents
, bytenr
,
6320 bytenr
+ num_bytes
- 1, GFP_NOFS
| __GFP_NOFAIL
);
6325 * this function must be called within transaction
6327 int btrfs_pin_extent(struct btrfs_root
*root
,
6328 u64 bytenr
, u64 num_bytes
, int reserved
)
6330 struct btrfs_block_group_cache
*cache
;
6332 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
6333 BUG_ON(!cache
); /* Logic error */
6335 pin_down_extent(root
, cache
, bytenr
, num_bytes
, reserved
);
6337 btrfs_put_block_group(cache
);
6342 * this function must be called within transaction
6344 int btrfs_pin_extent_for_log_replay(struct btrfs_root
*root
,
6345 u64 bytenr
, u64 num_bytes
)
6347 struct btrfs_block_group_cache
*cache
;
6350 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
6355 * pull in the free space cache (if any) so that our pin
6356 * removes the free space from the cache. We have load_only set
6357 * to one because the slow code to read in the free extents does check
6358 * the pinned extents.
6360 cache_block_group(cache
, 1);
6362 pin_down_extent(root
, cache
, bytenr
, num_bytes
, 0);
6364 /* remove us from the free space cache (if we're there at all) */
6365 ret
= btrfs_remove_free_space(cache
, bytenr
, num_bytes
);
6366 btrfs_put_block_group(cache
);
6370 static int __exclude_logged_extent(struct btrfs_root
*root
, u64 start
, u64 num_bytes
)
6373 struct btrfs_block_group_cache
*block_group
;
6374 struct btrfs_caching_control
*caching_ctl
;
6376 block_group
= btrfs_lookup_block_group(root
->fs_info
, start
);
6380 cache_block_group(block_group
, 0);
6381 caching_ctl
= get_caching_control(block_group
);
6385 BUG_ON(!block_group_cache_done(block_group
));
6386 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
6388 mutex_lock(&caching_ctl
->mutex
);
6390 if (start
>= caching_ctl
->progress
) {
6391 ret
= add_excluded_extent(root
, start
, num_bytes
);
6392 } else if (start
+ num_bytes
<= caching_ctl
->progress
) {
6393 ret
= btrfs_remove_free_space(block_group
,
6396 num_bytes
= caching_ctl
->progress
- start
;
6397 ret
= btrfs_remove_free_space(block_group
,
6402 num_bytes
= (start
+ num_bytes
) -
6403 caching_ctl
->progress
;
6404 start
= caching_ctl
->progress
;
6405 ret
= add_excluded_extent(root
, start
, num_bytes
);
6408 mutex_unlock(&caching_ctl
->mutex
);
6409 put_caching_control(caching_ctl
);
6411 btrfs_put_block_group(block_group
);
6415 int btrfs_exclude_logged_extents(struct btrfs_root
*log
,
6416 struct extent_buffer
*eb
)
6418 struct btrfs_file_extent_item
*item
;
6419 struct btrfs_key key
;
6423 if (!btrfs_fs_incompat(log
->fs_info
, MIXED_GROUPS
))
6426 for (i
= 0; i
< btrfs_header_nritems(eb
); i
++) {
6427 btrfs_item_key_to_cpu(eb
, &key
, i
);
6428 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
6430 item
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
6431 found_type
= btrfs_file_extent_type(eb
, item
);
6432 if (found_type
== BTRFS_FILE_EXTENT_INLINE
)
6434 if (btrfs_file_extent_disk_bytenr(eb
, item
) == 0)
6436 key
.objectid
= btrfs_file_extent_disk_bytenr(eb
, item
);
6437 key
.offset
= btrfs_file_extent_disk_num_bytes(eb
, item
);
6438 __exclude_logged_extent(log
, key
.objectid
, key
.offset
);
6445 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache
*bg
)
6447 atomic_inc(&bg
->reservations
);
6450 void btrfs_dec_block_group_reservations(struct btrfs_fs_info
*fs_info
,
6453 struct btrfs_block_group_cache
*bg
;
6455 bg
= btrfs_lookup_block_group(fs_info
, start
);
6457 if (atomic_dec_and_test(&bg
->reservations
))
6458 wake_up_atomic_t(&bg
->reservations
);
6459 btrfs_put_block_group(bg
);
6462 static int btrfs_wait_bg_reservations_atomic_t(atomic_t
*a
)
6468 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache
*bg
)
6470 struct btrfs_space_info
*space_info
= bg
->space_info
;
6474 if (!(bg
->flags
& BTRFS_BLOCK_GROUP_DATA
))
6478 * Our block group is read only but before we set it to read only,
6479 * some task might have had allocated an extent from it already, but it
6480 * has not yet created a respective ordered extent (and added it to a
6481 * root's list of ordered extents).
6482 * Therefore wait for any task currently allocating extents, since the
6483 * block group's reservations counter is incremented while a read lock
6484 * on the groups' semaphore is held and decremented after releasing
6485 * the read access on that semaphore and creating the ordered extent.
6487 down_write(&space_info
->groups_sem
);
6488 up_write(&space_info
->groups_sem
);
6490 wait_on_atomic_t(&bg
->reservations
,
6491 btrfs_wait_bg_reservations_atomic_t
,
6492 TASK_UNINTERRUPTIBLE
);
6496 * btrfs_add_reserved_bytes - update the block_group and space info counters
6497 * @cache: The cache we are manipulating
6498 * @ram_bytes: The number of bytes of file content, and will be same to
6499 * @num_bytes except for the compress path.
6500 * @num_bytes: The number of bytes in question
6501 * @delalloc: The blocks are allocated for the delalloc write
6503 * This is called by the allocator when it reserves space. Metadata
6504 * reservations should be called with RESERVE_ALLOC so we do the proper
6505 * ENOSPC accounting. For data we handle the reservation through clearing the
6506 * delalloc bits in the io_tree. We have to do this since we could end up
6507 * allocating less disk space for the amount of data we have reserved in the
6508 * case of compression.
6510 * If this is a reservation and the block group has become read only we cannot
6511 * make the reservation and return -EAGAIN, otherwise this function always
6514 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache
*cache
,
6515 u64 ram_bytes
, u64 num_bytes
, int delalloc
)
6517 struct btrfs_space_info
*space_info
= cache
->space_info
;
6520 spin_lock(&space_info
->lock
);
6521 spin_lock(&cache
->lock
);
6525 cache
->reserved
+= num_bytes
;
6526 space_info
->bytes_reserved
+= num_bytes
;
6528 trace_btrfs_space_reservation(cache
->fs_info
,
6529 "space_info", space_info
->flags
,
6531 space_info
->bytes_may_use
-= ram_bytes
;
6533 cache
->delalloc_bytes
+= num_bytes
;
6535 spin_unlock(&cache
->lock
);
6536 spin_unlock(&space_info
->lock
);
6541 * btrfs_free_reserved_bytes - update the block_group and space info counters
6542 * @cache: The cache we are manipulating
6543 * @num_bytes: The number of bytes in question
6544 * @delalloc: The blocks are allocated for the delalloc write
6546 * This is called by somebody who is freeing space that was never actually used
6547 * on disk. For example if you reserve some space for a new leaf in transaction
6548 * A and before transaction A commits you free that leaf, you call this with
6549 * reserve set to 0 in order to clear the reservation.
6552 static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache
*cache
,
6553 u64 num_bytes
, int delalloc
)
6555 struct btrfs_space_info
*space_info
= cache
->space_info
;
6558 spin_lock(&space_info
->lock
);
6559 spin_lock(&cache
->lock
);
6561 space_info
->bytes_readonly
+= num_bytes
;
6562 cache
->reserved
-= num_bytes
;
6563 space_info
->bytes_reserved
-= num_bytes
;
6566 cache
->delalloc_bytes
-= num_bytes
;
6567 spin_unlock(&cache
->lock
);
6568 spin_unlock(&space_info
->lock
);
6571 void btrfs_prepare_extent_commit(struct btrfs_trans_handle
*trans
,
6572 struct btrfs_root
*root
)
6574 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6575 struct btrfs_caching_control
*next
;
6576 struct btrfs_caching_control
*caching_ctl
;
6577 struct btrfs_block_group_cache
*cache
;
6579 down_write(&fs_info
->commit_root_sem
);
6581 list_for_each_entry_safe(caching_ctl
, next
,
6582 &fs_info
->caching_block_groups
, list
) {
6583 cache
= caching_ctl
->block_group
;
6584 if (block_group_cache_done(cache
)) {
6585 cache
->last_byte_to_unpin
= (u64
)-1;
6586 list_del_init(&caching_ctl
->list
);
6587 put_caching_control(caching_ctl
);
6589 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
6593 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
6594 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
6596 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
6598 up_write(&fs_info
->commit_root_sem
);
6600 update_global_block_rsv(fs_info
);
6604 * Returns the free cluster for the given space info and sets empty_cluster to
6605 * what it should be based on the mount options.
6607 static struct btrfs_free_cluster
*
6608 fetch_cluster_info(struct btrfs_root
*root
, struct btrfs_space_info
*space_info
,
6611 struct btrfs_free_cluster
*ret
= NULL
;
6612 bool ssd
= btrfs_test_opt(root
->fs_info
, SSD
);
6615 if (btrfs_mixed_space_info(space_info
))
6619 *empty_cluster
= SZ_2M
;
6620 if (space_info
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
6621 ret
= &root
->fs_info
->meta_alloc_cluster
;
6623 *empty_cluster
= SZ_64K
;
6624 } else if ((space_info
->flags
& BTRFS_BLOCK_GROUP_DATA
) && ssd
) {
6625 ret
= &root
->fs_info
->data_alloc_cluster
;
6631 static int unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
,
6632 const bool return_free_space
)
6634 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6635 struct btrfs_block_group_cache
*cache
= NULL
;
6636 struct btrfs_space_info
*space_info
;
6637 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
6638 struct btrfs_free_cluster
*cluster
= NULL
;
6640 u64 total_unpinned
= 0;
6641 u64 empty_cluster
= 0;
6644 while (start
<= end
) {
6647 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
6649 btrfs_put_block_group(cache
);
6651 cache
= btrfs_lookup_block_group(fs_info
, start
);
6652 BUG_ON(!cache
); /* Logic error */
6654 cluster
= fetch_cluster_info(root
,
6657 empty_cluster
<<= 1;
6660 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
6661 len
= min(len
, end
+ 1 - start
);
6663 if (start
< cache
->last_byte_to_unpin
) {
6664 len
= min(len
, cache
->last_byte_to_unpin
- start
);
6665 if (return_free_space
)
6666 btrfs_add_free_space(cache
, start
, len
);
6670 total_unpinned
+= len
;
6671 space_info
= cache
->space_info
;
6674 * If this space cluster has been marked as fragmented and we've
6675 * unpinned enough in this block group to potentially allow a
6676 * cluster to be created inside of it go ahead and clear the
6679 if (cluster
&& cluster
->fragmented
&&
6680 total_unpinned
> empty_cluster
) {
6681 spin_lock(&cluster
->lock
);
6682 cluster
->fragmented
= 0;
6683 spin_unlock(&cluster
->lock
);
6686 spin_lock(&space_info
->lock
);
6687 spin_lock(&cache
->lock
);
6688 cache
->pinned
-= len
;
6689 space_info
->bytes_pinned
-= len
;
6691 trace_btrfs_space_reservation(fs_info
, "pinned",
6692 space_info
->flags
, len
, 0);
6693 space_info
->max_extent_size
= 0;
6694 percpu_counter_add(&space_info
->total_bytes_pinned
, -len
);
6696 space_info
->bytes_readonly
+= len
;
6699 spin_unlock(&cache
->lock
);
6700 if (!readonly
&& return_free_space
&&
6701 global_rsv
->space_info
== space_info
) {
6703 WARN_ON(!return_free_space
);
6704 spin_lock(&global_rsv
->lock
);
6705 if (!global_rsv
->full
) {
6706 to_add
= min(len
, global_rsv
->size
-
6707 global_rsv
->reserved
);
6708 global_rsv
->reserved
+= to_add
;
6709 space_info
->bytes_may_use
+= to_add
;
6710 if (global_rsv
->reserved
>= global_rsv
->size
)
6711 global_rsv
->full
= 1;
6712 trace_btrfs_space_reservation(fs_info
,
6718 spin_unlock(&global_rsv
->lock
);
6719 /* Add to any tickets we may have */
6721 space_info_add_new_bytes(fs_info
, space_info
,
6724 spin_unlock(&space_info
->lock
);
6728 btrfs_put_block_group(cache
);
6732 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
6733 struct btrfs_root
*root
)
6735 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6736 struct btrfs_block_group_cache
*block_group
, *tmp
;
6737 struct list_head
*deleted_bgs
;
6738 struct extent_io_tree
*unpin
;
6743 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
6744 unpin
= &fs_info
->freed_extents
[1];
6746 unpin
= &fs_info
->freed_extents
[0];
6748 while (!trans
->aborted
) {
6749 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
6750 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
6751 EXTENT_DIRTY
, NULL
);
6753 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
6757 if (btrfs_test_opt(root
->fs_info
, DISCARD
))
6758 ret
= btrfs_discard_extent(root
, start
,
6759 end
+ 1 - start
, NULL
);
6761 clear_extent_dirty(unpin
, start
, end
);
6762 unpin_extent_range(root
, start
, end
, true);
6763 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
6768 * Transaction is finished. We don't need the lock anymore. We
6769 * do need to clean up the block groups in case of a transaction
6772 deleted_bgs
= &trans
->transaction
->deleted_bgs
;
6773 list_for_each_entry_safe(block_group
, tmp
, deleted_bgs
, bg_list
) {
6777 if (!trans
->aborted
)
6778 ret
= btrfs_discard_extent(root
,
6779 block_group
->key
.objectid
,
6780 block_group
->key
.offset
,
6783 list_del_init(&block_group
->bg_list
);
6784 btrfs_put_block_group_trimming(block_group
);
6785 btrfs_put_block_group(block_group
);
6788 const char *errstr
= btrfs_decode_error(ret
);
6790 "Discard failed while removing blockgroup: errno=%d %s\n",
6798 static void add_pinned_bytes(struct btrfs_fs_info
*fs_info
, u64 num_bytes
,
6799 u64 owner
, u64 root_objectid
)
6801 struct btrfs_space_info
*space_info
;
6804 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
6805 if (root_objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
6806 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
6808 flags
= BTRFS_BLOCK_GROUP_METADATA
;
6810 flags
= BTRFS_BLOCK_GROUP_DATA
;
6813 space_info
= __find_space_info(fs_info
, flags
);
6814 BUG_ON(!space_info
); /* Logic bug */
6815 percpu_counter_add(&space_info
->total_bytes_pinned
, num_bytes
);
6819 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
6820 struct btrfs_root
*root
,
6821 struct btrfs_delayed_ref_node
*node
, u64 parent
,
6822 u64 root_objectid
, u64 owner_objectid
,
6823 u64 owner_offset
, int refs_to_drop
,
6824 struct btrfs_delayed_extent_op
*extent_op
)
6826 struct btrfs_key key
;
6827 struct btrfs_path
*path
;
6828 struct btrfs_fs_info
*info
= root
->fs_info
;
6829 struct btrfs_root
*extent_root
= info
->extent_root
;
6830 struct extent_buffer
*leaf
;
6831 struct btrfs_extent_item
*ei
;
6832 struct btrfs_extent_inline_ref
*iref
;
6835 int extent_slot
= 0;
6836 int found_extent
= 0;
6840 u64 bytenr
= node
->bytenr
;
6841 u64 num_bytes
= node
->num_bytes
;
6843 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
6846 path
= btrfs_alloc_path();
6850 path
->reada
= READA_FORWARD
;
6851 path
->leave_spinning
= 1;
6853 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
6854 BUG_ON(!is_data
&& refs_to_drop
!= 1);
6857 skinny_metadata
= 0;
6859 ret
= lookup_extent_backref(trans
, extent_root
, path
, &iref
,
6860 bytenr
, num_bytes
, parent
,
6861 root_objectid
, owner_objectid
,
6864 extent_slot
= path
->slots
[0];
6865 while (extent_slot
>= 0) {
6866 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
6868 if (key
.objectid
!= bytenr
)
6870 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
6871 key
.offset
== num_bytes
) {
6875 if (key
.type
== BTRFS_METADATA_ITEM_KEY
&&
6876 key
.offset
== owner_objectid
) {
6880 if (path
->slots
[0] - extent_slot
> 5)
6884 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6885 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
6886 if (found_extent
&& item_size
< sizeof(*ei
))
6889 if (!found_extent
) {
6891 ret
= remove_extent_backref(trans
, extent_root
, path
,
6893 is_data
, &last_ref
);
6895 btrfs_abort_transaction(trans
, ret
);
6898 btrfs_release_path(path
);
6899 path
->leave_spinning
= 1;
6901 key
.objectid
= bytenr
;
6902 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6903 key
.offset
= num_bytes
;
6905 if (!is_data
&& skinny_metadata
) {
6906 key
.type
= BTRFS_METADATA_ITEM_KEY
;
6907 key
.offset
= owner_objectid
;
6910 ret
= btrfs_search_slot(trans
, extent_root
,
6912 if (ret
> 0 && skinny_metadata
&& path
->slots
[0]) {
6914 * Couldn't find our skinny metadata item,
6915 * see if we have ye olde extent item.
6918 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
6920 if (key
.objectid
== bytenr
&&
6921 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
6922 key
.offset
== num_bytes
)
6926 if (ret
> 0 && skinny_metadata
) {
6927 skinny_metadata
= false;
6928 key
.objectid
= bytenr
;
6929 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6930 key
.offset
= num_bytes
;
6931 btrfs_release_path(path
);
6932 ret
= btrfs_search_slot(trans
, extent_root
,
6937 btrfs_err(info
, "umm, got %d back from search, was looking for %llu",
6940 btrfs_print_leaf(extent_root
,
6944 btrfs_abort_transaction(trans
, ret
);
6947 extent_slot
= path
->slots
[0];
6949 } else if (WARN_ON(ret
== -ENOENT
)) {
6950 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
6952 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6953 bytenr
, parent
, root_objectid
, owner_objectid
,
6955 btrfs_abort_transaction(trans
, ret
);
6958 btrfs_abort_transaction(trans
, ret
);
6962 leaf
= path
->nodes
[0];
6963 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
6964 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6965 if (item_size
< sizeof(*ei
)) {
6966 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
6967 ret
= convert_extent_item_v0(trans
, extent_root
, path
,
6970 btrfs_abort_transaction(trans
, ret
);
6974 btrfs_release_path(path
);
6975 path
->leave_spinning
= 1;
6977 key
.objectid
= bytenr
;
6978 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6979 key
.offset
= num_bytes
;
6981 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
6984 btrfs_err(info
, "umm, got %d back from search, was looking for %llu",
6986 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
6989 btrfs_abort_transaction(trans
, ret
);
6993 extent_slot
= path
->slots
[0];
6994 leaf
= path
->nodes
[0];
6995 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
6998 BUG_ON(item_size
< sizeof(*ei
));
6999 ei
= btrfs_item_ptr(leaf
, extent_slot
,
7000 struct btrfs_extent_item
);
7001 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
7002 key
.type
== BTRFS_EXTENT_ITEM_KEY
) {
7003 struct btrfs_tree_block_info
*bi
;
7004 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
7005 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
7006 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
7009 refs
= btrfs_extent_refs(leaf
, ei
);
7010 if (refs
< refs_to_drop
) {
7011 btrfs_err(info
, "trying to drop %d refs but we only have %Lu "
7012 "for bytenr %Lu", refs_to_drop
, refs
, bytenr
);
7014 btrfs_abort_transaction(trans
, ret
);
7017 refs
-= refs_to_drop
;
7021 __run_delayed_extent_op(extent_op
, leaf
, ei
);
7023 * In the case of inline back ref, reference count will
7024 * be updated by remove_extent_backref
7027 BUG_ON(!found_extent
);
7029 btrfs_set_extent_refs(leaf
, ei
, refs
);
7030 btrfs_mark_buffer_dirty(leaf
);
7033 ret
= remove_extent_backref(trans
, extent_root
, path
,
7035 is_data
, &last_ref
);
7037 btrfs_abort_transaction(trans
, ret
);
7041 add_pinned_bytes(root
->fs_info
, -num_bytes
, owner_objectid
,
7045 BUG_ON(is_data
&& refs_to_drop
!=
7046 extent_data_ref_count(path
, iref
));
7048 BUG_ON(path
->slots
[0] != extent_slot
);
7050 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
7051 path
->slots
[0] = extent_slot
;
7057 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
7060 btrfs_abort_transaction(trans
, ret
);
7063 btrfs_release_path(path
);
7066 ret
= btrfs_del_csums(trans
, root
, bytenr
, num_bytes
);
7068 btrfs_abort_transaction(trans
, ret
);
7073 ret
= add_to_free_space_tree(trans
, root
->fs_info
, bytenr
,
7076 btrfs_abort_transaction(trans
, ret
);
7080 ret
= update_block_group(trans
, root
, bytenr
, num_bytes
, 0);
7082 btrfs_abort_transaction(trans
, ret
);
7086 btrfs_release_path(path
);
7089 btrfs_free_path(path
);
7094 * when we free an block, it is possible (and likely) that we free the last
7095 * delayed ref for that extent as well. This searches the delayed ref tree for
7096 * a given extent, and if there are no other delayed refs to be processed, it
7097 * removes it from the tree.
7099 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
7100 struct btrfs_root
*root
, u64 bytenr
)
7102 struct btrfs_delayed_ref_head
*head
;
7103 struct btrfs_delayed_ref_root
*delayed_refs
;
7106 delayed_refs
= &trans
->transaction
->delayed_refs
;
7107 spin_lock(&delayed_refs
->lock
);
7108 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
7110 goto out_delayed_unlock
;
7112 spin_lock(&head
->lock
);
7113 if (!list_empty(&head
->ref_list
))
7116 if (head
->extent_op
) {
7117 if (!head
->must_insert_reserved
)
7119 btrfs_free_delayed_extent_op(head
->extent_op
);
7120 head
->extent_op
= NULL
;
7124 * waiting for the lock here would deadlock. If someone else has it
7125 * locked they are already in the process of dropping it anyway
7127 if (!mutex_trylock(&head
->mutex
))
7131 * at this point we have a head with no other entries. Go
7132 * ahead and process it.
7134 head
->node
.in_tree
= 0;
7135 rb_erase(&head
->href_node
, &delayed_refs
->href_root
);
7137 atomic_dec(&delayed_refs
->num_entries
);
7140 * we don't take a ref on the node because we're removing it from the
7141 * tree, so we just steal the ref the tree was holding.
7143 delayed_refs
->num_heads
--;
7144 if (head
->processing
== 0)
7145 delayed_refs
->num_heads_ready
--;
7146 head
->processing
= 0;
7147 spin_unlock(&head
->lock
);
7148 spin_unlock(&delayed_refs
->lock
);
7150 BUG_ON(head
->extent_op
);
7151 if (head
->must_insert_reserved
)
7154 mutex_unlock(&head
->mutex
);
7155 btrfs_put_delayed_ref(&head
->node
);
7158 spin_unlock(&head
->lock
);
7161 spin_unlock(&delayed_refs
->lock
);
7165 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
7166 struct btrfs_root
*root
,
7167 struct extent_buffer
*buf
,
7168 u64 parent
, int last_ref
)
7173 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
7174 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
7175 buf
->start
, buf
->len
,
7176 parent
, root
->root_key
.objectid
,
7177 btrfs_header_level(buf
),
7178 BTRFS_DROP_DELAYED_REF
, NULL
);
7179 BUG_ON(ret
); /* -ENOMEM */
7185 if (btrfs_header_generation(buf
) == trans
->transid
) {
7186 struct btrfs_block_group_cache
*cache
;
7188 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
7189 ret
= check_ref_cleanup(trans
, root
, buf
->start
);
7194 cache
= btrfs_lookup_block_group(root
->fs_info
, buf
->start
);
7196 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
7197 pin_down_extent(root
, cache
, buf
->start
, buf
->len
, 1);
7198 btrfs_put_block_group(cache
);
7202 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
7204 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
7205 btrfs_free_reserved_bytes(cache
, buf
->len
, 0);
7206 btrfs_put_block_group(cache
);
7207 trace_btrfs_reserved_extent_free(root
, buf
->start
, buf
->len
);
7212 add_pinned_bytes(root
->fs_info
, buf
->len
,
7213 btrfs_header_level(buf
),
7214 root
->root_key
.objectid
);
7217 * Deleting the buffer, clear the corrupt flag since it doesn't matter
7220 clear_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
);
7223 /* Can return -ENOMEM */
7224 int btrfs_free_extent(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
7225 u64 bytenr
, u64 num_bytes
, u64 parent
, u64 root_objectid
,
7226 u64 owner
, u64 offset
)
7229 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7231 if (btrfs_is_testing(fs_info
))
7234 add_pinned_bytes(root
->fs_info
, num_bytes
, owner
, root_objectid
);
7237 * tree log blocks never actually go into the extent allocation
7238 * tree, just update pinning info and exit early.
7240 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
7241 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
7242 /* unlocks the pinned mutex */
7243 btrfs_pin_extent(root
, bytenr
, num_bytes
, 1);
7245 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
7246 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
7248 parent
, root_objectid
, (int)owner
,
7249 BTRFS_DROP_DELAYED_REF
, NULL
);
7251 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
7253 parent
, root_objectid
, owner
,
7255 BTRFS_DROP_DELAYED_REF
, NULL
);
7261 * when we wait for progress in the block group caching, its because
7262 * our allocation attempt failed at least once. So, we must sleep
7263 * and let some progress happen before we try again.
7265 * This function will sleep at least once waiting for new free space to
7266 * show up, and then it will check the block group free space numbers
7267 * for our min num_bytes. Another option is to have it go ahead
7268 * and look in the rbtree for a free extent of a given size, but this
7271 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7272 * any of the information in this block group.
7274 static noinline
void
7275 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
7278 struct btrfs_caching_control
*caching_ctl
;
7280 caching_ctl
= get_caching_control(cache
);
7284 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
7285 (cache
->free_space_ctl
->free_space
>= num_bytes
));
7287 put_caching_control(caching_ctl
);
7291 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
7293 struct btrfs_caching_control
*caching_ctl
;
7296 caching_ctl
= get_caching_control(cache
);
7298 return (cache
->cached
== BTRFS_CACHE_ERROR
) ? -EIO
: 0;
7300 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
7301 if (cache
->cached
== BTRFS_CACHE_ERROR
)
7303 put_caching_control(caching_ctl
);
7307 int __get_raid_index(u64 flags
)
7309 if (flags
& BTRFS_BLOCK_GROUP_RAID10
)
7310 return BTRFS_RAID_RAID10
;
7311 else if (flags
& BTRFS_BLOCK_GROUP_RAID1
)
7312 return BTRFS_RAID_RAID1
;
7313 else if (flags
& BTRFS_BLOCK_GROUP_DUP
)
7314 return BTRFS_RAID_DUP
;
7315 else if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
7316 return BTRFS_RAID_RAID0
;
7317 else if (flags
& BTRFS_BLOCK_GROUP_RAID5
)
7318 return BTRFS_RAID_RAID5
;
7319 else if (flags
& BTRFS_BLOCK_GROUP_RAID6
)
7320 return BTRFS_RAID_RAID6
;
7322 return BTRFS_RAID_SINGLE
; /* BTRFS_BLOCK_GROUP_SINGLE */
7325 int get_block_group_index(struct btrfs_block_group_cache
*cache
)
7327 return __get_raid_index(cache
->flags
);
7330 static const char *btrfs_raid_type_names
[BTRFS_NR_RAID_TYPES
] = {
7331 [BTRFS_RAID_RAID10
] = "raid10",
7332 [BTRFS_RAID_RAID1
] = "raid1",
7333 [BTRFS_RAID_DUP
] = "dup",
7334 [BTRFS_RAID_RAID0
] = "raid0",
7335 [BTRFS_RAID_SINGLE
] = "single",
7336 [BTRFS_RAID_RAID5
] = "raid5",
7337 [BTRFS_RAID_RAID6
] = "raid6",
7340 static const char *get_raid_name(enum btrfs_raid_types type
)
7342 if (type
>= BTRFS_NR_RAID_TYPES
)
7345 return btrfs_raid_type_names
[type
];
7348 enum btrfs_loop_type
{
7349 LOOP_CACHING_NOWAIT
= 0,
7350 LOOP_CACHING_WAIT
= 1,
7351 LOOP_ALLOC_CHUNK
= 2,
7352 LOOP_NO_EMPTY_SIZE
= 3,
7356 btrfs_lock_block_group(struct btrfs_block_group_cache
*cache
,
7360 down_read(&cache
->data_rwsem
);
7364 btrfs_grab_block_group(struct btrfs_block_group_cache
*cache
,
7367 btrfs_get_block_group(cache
);
7369 down_read(&cache
->data_rwsem
);
7372 static struct btrfs_block_group_cache
*
7373 btrfs_lock_cluster(struct btrfs_block_group_cache
*block_group
,
7374 struct btrfs_free_cluster
*cluster
,
7377 struct btrfs_block_group_cache
*used_bg
= NULL
;
7379 spin_lock(&cluster
->refill_lock
);
7381 used_bg
= cluster
->block_group
;
7385 if (used_bg
== block_group
)
7388 btrfs_get_block_group(used_bg
);
7393 if (down_read_trylock(&used_bg
->data_rwsem
))
7396 spin_unlock(&cluster
->refill_lock
);
7398 down_read(&used_bg
->data_rwsem
);
7400 spin_lock(&cluster
->refill_lock
);
7401 if (used_bg
== cluster
->block_group
)
7404 up_read(&used_bg
->data_rwsem
);
7405 btrfs_put_block_group(used_bg
);
7410 btrfs_release_block_group(struct btrfs_block_group_cache
*cache
,
7414 up_read(&cache
->data_rwsem
);
7415 btrfs_put_block_group(cache
);
7419 * walks the btree of allocated extents and find a hole of a given size.
7420 * The key ins is changed to record the hole:
7421 * ins->objectid == start position
7422 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7423 * ins->offset == the size of the hole.
7424 * Any available blocks before search_start are skipped.
7426 * If there is no suitable free space, we will record the max size of
7427 * the free space extent currently.
7429 static noinline
int find_free_extent(struct btrfs_root
*orig_root
,
7430 u64 ram_bytes
, u64 num_bytes
, u64 empty_size
,
7431 u64 hint_byte
, struct btrfs_key
*ins
,
7432 u64 flags
, int delalloc
)
7435 struct btrfs_root
*root
= orig_root
->fs_info
->extent_root
;
7436 struct btrfs_free_cluster
*last_ptr
= NULL
;
7437 struct btrfs_block_group_cache
*block_group
= NULL
;
7438 u64 search_start
= 0;
7439 u64 max_extent_size
= 0;
7440 u64 empty_cluster
= 0;
7441 struct btrfs_space_info
*space_info
;
7443 int index
= __get_raid_index(flags
);
7444 bool failed_cluster_refill
= false;
7445 bool failed_alloc
= false;
7446 bool use_cluster
= true;
7447 bool have_caching_bg
= false;
7448 bool orig_have_caching_bg
= false;
7449 bool full_search
= false;
7451 WARN_ON(num_bytes
< root
->sectorsize
);
7452 ins
->type
= BTRFS_EXTENT_ITEM_KEY
;
7456 trace_find_free_extent(orig_root
, num_bytes
, empty_size
, flags
);
7458 space_info
= __find_space_info(root
->fs_info
, flags
);
7460 btrfs_err(root
->fs_info
, "No space info for %llu", flags
);
7465 * If our free space is heavily fragmented we may not be able to make
7466 * big contiguous allocations, so instead of doing the expensive search
7467 * for free space, simply return ENOSPC with our max_extent_size so we
7468 * can go ahead and search for a more manageable chunk.
7470 * If our max_extent_size is large enough for our allocation simply
7471 * disable clustering since we will likely not be able to find enough
7472 * space to create a cluster and induce latency trying.
7474 if (unlikely(space_info
->max_extent_size
)) {
7475 spin_lock(&space_info
->lock
);
7476 if (space_info
->max_extent_size
&&
7477 num_bytes
> space_info
->max_extent_size
) {
7478 ins
->offset
= space_info
->max_extent_size
;
7479 spin_unlock(&space_info
->lock
);
7481 } else if (space_info
->max_extent_size
) {
7482 use_cluster
= false;
7484 spin_unlock(&space_info
->lock
);
7487 last_ptr
= fetch_cluster_info(orig_root
, space_info
, &empty_cluster
);
7489 spin_lock(&last_ptr
->lock
);
7490 if (last_ptr
->block_group
)
7491 hint_byte
= last_ptr
->window_start
;
7492 if (last_ptr
->fragmented
) {
7494 * We still set window_start so we can keep track of the
7495 * last place we found an allocation to try and save
7498 hint_byte
= last_ptr
->window_start
;
7499 use_cluster
= false;
7501 spin_unlock(&last_ptr
->lock
);
7504 search_start
= max(search_start
, first_logical_byte(root
, 0));
7505 search_start
= max(search_start
, hint_byte
);
7506 if (search_start
== hint_byte
) {
7507 block_group
= btrfs_lookup_block_group(root
->fs_info
,
7510 * we don't want to use the block group if it doesn't match our
7511 * allocation bits, or if its not cached.
7513 * However if we are re-searching with an ideal block group
7514 * picked out then we don't care that the block group is cached.
7516 if (block_group
&& block_group_bits(block_group
, flags
) &&
7517 block_group
->cached
!= BTRFS_CACHE_NO
) {
7518 down_read(&space_info
->groups_sem
);
7519 if (list_empty(&block_group
->list
) ||
7522 * someone is removing this block group,
7523 * we can't jump into the have_block_group
7524 * target because our list pointers are not
7527 btrfs_put_block_group(block_group
);
7528 up_read(&space_info
->groups_sem
);
7530 index
= get_block_group_index(block_group
);
7531 btrfs_lock_block_group(block_group
, delalloc
);
7532 goto have_block_group
;
7534 } else if (block_group
) {
7535 btrfs_put_block_group(block_group
);
7539 have_caching_bg
= false;
7540 if (index
== 0 || index
== __get_raid_index(flags
))
7542 down_read(&space_info
->groups_sem
);
7543 list_for_each_entry(block_group
, &space_info
->block_groups
[index
],
7548 btrfs_grab_block_group(block_group
, delalloc
);
7549 search_start
= block_group
->key
.objectid
;
7552 * this can happen if we end up cycling through all the
7553 * raid types, but we want to make sure we only allocate
7554 * for the proper type.
7556 if (!block_group_bits(block_group
, flags
)) {
7557 u64 extra
= BTRFS_BLOCK_GROUP_DUP
|
7558 BTRFS_BLOCK_GROUP_RAID1
|
7559 BTRFS_BLOCK_GROUP_RAID5
|
7560 BTRFS_BLOCK_GROUP_RAID6
|
7561 BTRFS_BLOCK_GROUP_RAID10
;
7564 * if they asked for extra copies and this block group
7565 * doesn't provide them, bail. This does allow us to
7566 * fill raid0 from raid1.
7568 if ((flags
& extra
) && !(block_group
->flags
& extra
))
7573 cached
= block_group_cache_done(block_group
);
7574 if (unlikely(!cached
)) {
7575 have_caching_bg
= true;
7576 ret
= cache_block_group(block_group
, 0);
7581 if (unlikely(block_group
->cached
== BTRFS_CACHE_ERROR
))
7583 if (unlikely(block_group
->ro
))
7587 * Ok we want to try and use the cluster allocator, so
7590 if (last_ptr
&& use_cluster
) {
7591 struct btrfs_block_group_cache
*used_block_group
;
7592 unsigned long aligned_cluster
;
7594 * the refill lock keeps out other
7595 * people trying to start a new cluster
7597 used_block_group
= btrfs_lock_cluster(block_group
,
7600 if (!used_block_group
)
7601 goto refill_cluster
;
7603 if (used_block_group
!= block_group
&&
7604 (used_block_group
->ro
||
7605 !block_group_bits(used_block_group
, flags
)))
7606 goto release_cluster
;
7608 offset
= btrfs_alloc_from_cluster(used_block_group
,
7611 used_block_group
->key
.objectid
,
7614 /* we have a block, we're done */
7615 spin_unlock(&last_ptr
->refill_lock
);
7616 trace_btrfs_reserve_extent_cluster(root
,
7618 search_start
, num_bytes
);
7619 if (used_block_group
!= block_group
) {
7620 btrfs_release_block_group(block_group
,
7622 block_group
= used_block_group
;
7627 WARN_ON(last_ptr
->block_group
!= used_block_group
);
7629 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7630 * set up a new clusters, so lets just skip it
7631 * and let the allocator find whatever block
7632 * it can find. If we reach this point, we
7633 * will have tried the cluster allocator
7634 * plenty of times and not have found
7635 * anything, so we are likely way too
7636 * fragmented for the clustering stuff to find
7639 * However, if the cluster is taken from the
7640 * current block group, release the cluster
7641 * first, so that we stand a better chance of
7642 * succeeding in the unclustered
7644 if (loop
>= LOOP_NO_EMPTY_SIZE
&&
7645 used_block_group
!= block_group
) {
7646 spin_unlock(&last_ptr
->refill_lock
);
7647 btrfs_release_block_group(used_block_group
,
7649 goto unclustered_alloc
;
7653 * this cluster didn't work out, free it and
7656 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
7658 if (used_block_group
!= block_group
)
7659 btrfs_release_block_group(used_block_group
,
7662 if (loop
>= LOOP_NO_EMPTY_SIZE
) {
7663 spin_unlock(&last_ptr
->refill_lock
);
7664 goto unclustered_alloc
;
7667 aligned_cluster
= max_t(unsigned long,
7668 empty_cluster
+ empty_size
,
7669 block_group
->full_stripe_len
);
7671 /* allocate a cluster in this block group */
7672 ret
= btrfs_find_space_cluster(root
, block_group
,
7673 last_ptr
, search_start
,
7678 * now pull our allocation out of this
7681 offset
= btrfs_alloc_from_cluster(block_group
,
7687 /* we found one, proceed */
7688 spin_unlock(&last_ptr
->refill_lock
);
7689 trace_btrfs_reserve_extent_cluster(root
,
7690 block_group
, search_start
,
7694 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
7695 && !failed_cluster_refill
) {
7696 spin_unlock(&last_ptr
->refill_lock
);
7698 failed_cluster_refill
= true;
7699 wait_block_group_cache_progress(block_group
,
7700 num_bytes
+ empty_cluster
+ empty_size
);
7701 goto have_block_group
;
7705 * at this point we either didn't find a cluster
7706 * or we weren't able to allocate a block from our
7707 * cluster. Free the cluster we've been trying
7708 * to use, and go to the next block group
7710 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
7711 spin_unlock(&last_ptr
->refill_lock
);
7717 * We are doing an unclustered alloc, set the fragmented flag so
7718 * we don't bother trying to setup a cluster again until we get
7721 if (unlikely(last_ptr
)) {
7722 spin_lock(&last_ptr
->lock
);
7723 last_ptr
->fragmented
= 1;
7724 spin_unlock(&last_ptr
->lock
);
7726 spin_lock(&block_group
->free_space_ctl
->tree_lock
);
7728 block_group
->free_space_ctl
->free_space
<
7729 num_bytes
+ empty_cluster
+ empty_size
) {
7730 if (block_group
->free_space_ctl
->free_space
>
7733 block_group
->free_space_ctl
->free_space
;
7734 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
7737 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
7739 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
7740 num_bytes
, empty_size
,
7743 * If we didn't find a chunk, and we haven't failed on this
7744 * block group before, and this block group is in the middle of
7745 * caching and we are ok with waiting, then go ahead and wait
7746 * for progress to be made, and set failed_alloc to true.
7748 * If failed_alloc is true then we've already waited on this
7749 * block group once and should move on to the next block group.
7751 if (!offset
&& !failed_alloc
&& !cached
&&
7752 loop
> LOOP_CACHING_NOWAIT
) {
7753 wait_block_group_cache_progress(block_group
,
7754 num_bytes
+ empty_size
);
7755 failed_alloc
= true;
7756 goto have_block_group
;
7757 } else if (!offset
) {
7761 search_start
= ALIGN(offset
, root
->stripesize
);
7763 /* move on to the next group */
7764 if (search_start
+ num_bytes
>
7765 block_group
->key
.objectid
+ block_group
->key
.offset
) {
7766 btrfs_add_free_space(block_group
, offset
, num_bytes
);
7770 if (offset
< search_start
)
7771 btrfs_add_free_space(block_group
, offset
,
7772 search_start
- offset
);
7773 BUG_ON(offset
> search_start
);
7775 ret
= btrfs_add_reserved_bytes(block_group
, ram_bytes
,
7776 num_bytes
, delalloc
);
7777 if (ret
== -EAGAIN
) {
7778 btrfs_add_free_space(block_group
, offset
, num_bytes
);
7781 btrfs_inc_block_group_reservations(block_group
);
7783 /* we are all good, lets return */
7784 ins
->objectid
= search_start
;
7785 ins
->offset
= num_bytes
;
7787 trace_btrfs_reserve_extent(orig_root
, block_group
,
7788 search_start
, num_bytes
);
7789 btrfs_release_block_group(block_group
, delalloc
);
7792 failed_cluster_refill
= false;
7793 failed_alloc
= false;
7794 BUG_ON(index
!= get_block_group_index(block_group
));
7795 btrfs_release_block_group(block_group
, delalloc
);
7797 up_read(&space_info
->groups_sem
);
7799 if ((loop
== LOOP_CACHING_NOWAIT
) && have_caching_bg
7800 && !orig_have_caching_bg
)
7801 orig_have_caching_bg
= true;
7803 if (!ins
->objectid
&& loop
>= LOOP_CACHING_WAIT
&& have_caching_bg
)
7806 if (!ins
->objectid
&& ++index
< BTRFS_NR_RAID_TYPES
)
7810 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7811 * caching kthreads as we move along
7812 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7813 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7814 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7817 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
) {
7819 if (loop
== LOOP_CACHING_NOWAIT
) {
7821 * We want to skip the LOOP_CACHING_WAIT step if we
7822 * don't have any uncached bgs and we've already done a
7823 * full search through.
7825 if (orig_have_caching_bg
|| !full_search
)
7826 loop
= LOOP_CACHING_WAIT
;
7828 loop
= LOOP_ALLOC_CHUNK
;
7833 if (loop
== LOOP_ALLOC_CHUNK
) {
7834 struct btrfs_trans_handle
*trans
;
7837 trans
= current
->journal_info
;
7841 trans
= btrfs_join_transaction(root
);
7843 if (IS_ERR(trans
)) {
7844 ret
= PTR_ERR(trans
);
7848 ret
= do_chunk_alloc(trans
, root
, flags
,
7852 * If we can't allocate a new chunk we've already looped
7853 * through at least once, move on to the NO_EMPTY_SIZE
7857 loop
= LOOP_NO_EMPTY_SIZE
;
7860 * Do not bail out on ENOSPC since we
7861 * can do more things.
7863 if (ret
< 0 && ret
!= -ENOSPC
)
7864 btrfs_abort_transaction(trans
, ret
);
7868 btrfs_end_transaction(trans
, root
);
7873 if (loop
== LOOP_NO_EMPTY_SIZE
) {
7875 * Don't loop again if we already have no empty_size and
7878 if (empty_size
== 0 &&
7879 empty_cluster
== 0) {
7888 } else if (!ins
->objectid
) {
7890 } else if (ins
->objectid
) {
7891 if (!use_cluster
&& last_ptr
) {
7892 spin_lock(&last_ptr
->lock
);
7893 last_ptr
->window_start
= ins
->objectid
;
7894 spin_unlock(&last_ptr
->lock
);
7899 if (ret
== -ENOSPC
) {
7900 spin_lock(&space_info
->lock
);
7901 space_info
->max_extent_size
= max_extent_size
;
7902 spin_unlock(&space_info
->lock
);
7903 ins
->offset
= max_extent_size
;
7908 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
7909 int dump_block_groups
)
7911 struct btrfs_block_group_cache
*cache
;
7914 spin_lock(&info
->lock
);
7915 printk(KERN_INFO
"BTRFS: space_info %llu has %llu free, is %sfull\n",
7917 info
->total_bytes
- info
->bytes_used
- info
->bytes_pinned
-
7918 info
->bytes_reserved
- info
->bytes_readonly
-
7919 info
->bytes_may_use
, (info
->full
) ? "" : "not ");
7920 printk(KERN_INFO
"BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7921 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7922 info
->total_bytes
, info
->bytes_used
, info
->bytes_pinned
,
7923 info
->bytes_reserved
, info
->bytes_may_use
,
7924 info
->bytes_readonly
);
7925 spin_unlock(&info
->lock
);
7927 if (!dump_block_groups
)
7930 down_read(&info
->groups_sem
);
7932 list_for_each_entry(cache
, &info
->block_groups
[index
], list
) {
7933 spin_lock(&cache
->lock
);
7934 printk(KERN_INFO
"BTRFS: "
7935 "block group %llu has %llu bytes, "
7936 "%llu used %llu pinned %llu reserved %s\n",
7937 cache
->key
.objectid
, cache
->key
.offset
,
7938 btrfs_block_group_used(&cache
->item
), cache
->pinned
,
7939 cache
->reserved
, cache
->ro
? "[readonly]" : "");
7940 btrfs_dump_free_space(cache
, bytes
);
7941 spin_unlock(&cache
->lock
);
7943 if (++index
< BTRFS_NR_RAID_TYPES
)
7945 up_read(&info
->groups_sem
);
7948 int btrfs_reserve_extent(struct btrfs_root
*root
, u64 ram_bytes
,
7949 u64 num_bytes
, u64 min_alloc_size
,
7950 u64 empty_size
, u64 hint_byte
,
7951 struct btrfs_key
*ins
, int is_data
, int delalloc
)
7953 bool final_tried
= num_bytes
== min_alloc_size
;
7957 flags
= btrfs_get_alloc_profile(root
, is_data
);
7959 WARN_ON(num_bytes
< root
->sectorsize
);
7960 ret
= find_free_extent(root
, ram_bytes
, num_bytes
, empty_size
,
7961 hint_byte
, ins
, flags
, delalloc
);
7962 if (!ret
&& !is_data
) {
7963 btrfs_dec_block_group_reservations(root
->fs_info
,
7965 } else if (ret
== -ENOSPC
) {
7966 if (!final_tried
&& ins
->offset
) {
7967 num_bytes
= min(num_bytes
>> 1, ins
->offset
);
7968 num_bytes
= round_down(num_bytes
, root
->sectorsize
);
7969 num_bytes
= max(num_bytes
, min_alloc_size
);
7970 ram_bytes
= num_bytes
;
7971 if (num_bytes
== min_alloc_size
)
7974 } else if (btrfs_test_opt(root
->fs_info
, ENOSPC_DEBUG
)) {
7975 struct btrfs_space_info
*sinfo
;
7977 sinfo
= __find_space_info(root
->fs_info
, flags
);
7978 btrfs_err(root
->fs_info
, "allocation failed flags %llu, wanted %llu",
7981 dump_space_info(sinfo
, num_bytes
, 1);
7988 static int __btrfs_free_reserved_extent(struct btrfs_root
*root
,
7990 int pin
, int delalloc
)
7992 struct btrfs_block_group_cache
*cache
;
7995 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
7997 btrfs_err(root
->fs_info
, "Unable to find block group for %llu",
8003 pin_down_extent(root
, cache
, start
, len
, 1);
8005 if (btrfs_test_opt(root
->fs_info
, DISCARD
))
8006 ret
= btrfs_discard_extent(root
, start
, len
, NULL
);
8007 btrfs_add_free_space(cache
, start
, len
);
8008 btrfs_free_reserved_bytes(cache
, len
, delalloc
);
8009 trace_btrfs_reserved_extent_free(root
, start
, len
);
8012 btrfs_put_block_group(cache
);
8016 int btrfs_free_reserved_extent(struct btrfs_root
*root
,
8017 u64 start
, u64 len
, int delalloc
)
8019 return __btrfs_free_reserved_extent(root
, start
, len
, 0, delalloc
);
8022 int btrfs_free_and_pin_reserved_extent(struct btrfs_root
*root
,
8025 return __btrfs_free_reserved_extent(root
, start
, len
, 1, 0);
8028 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
8029 struct btrfs_root
*root
,
8030 u64 parent
, u64 root_objectid
,
8031 u64 flags
, u64 owner
, u64 offset
,
8032 struct btrfs_key
*ins
, int ref_mod
)
8035 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8036 struct btrfs_extent_item
*extent_item
;
8037 struct btrfs_extent_inline_ref
*iref
;
8038 struct btrfs_path
*path
;
8039 struct extent_buffer
*leaf
;
8044 type
= BTRFS_SHARED_DATA_REF_KEY
;
8046 type
= BTRFS_EXTENT_DATA_REF_KEY
;
8048 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
8050 path
= btrfs_alloc_path();
8054 path
->leave_spinning
= 1;
8055 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
8058 btrfs_free_path(path
);
8062 leaf
= path
->nodes
[0];
8063 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
8064 struct btrfs_extent_item
);
8065 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
8066 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
8067 btrfs_set_extent_flags(leaf
, extent_item
,
8068 flags
| BTRFS_EXTENT_FLAG_DATA
);
8070 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
8071 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
8073 struct btrfs_shared_data_ref
*ref
;
8074 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
8075 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
8076 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
8078 struct btrfs_extent_data_ref
*ref
;
8079 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
8080 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
8081 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
8082 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
8083 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
8086 btrfs_mark_buffer_dirty(path
->nodes
[0]);
8087 btrfs_free_path(path
);
8089 ret
= remove_from_free_space_tree(trans
, fs_info
, ins
->objectid
,
8094 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1);
8095 if (ret
) { /* -ENOENT, logic error */
8096 btrfs_err(fs_info
, "update block group failed for %llu %llu",
8097 ins
->objectid
, ins
->offset
);
8100 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, ins
->offset
);
8104 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
8105 struct btrfs_root
*root
,
8106 u64 parent
, u64 root_objectid
,
8107 u64 flags
, struct btrfs_disk_key
*key
,
8108 int level
, struct btrfs_key
*ins
)
8111 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8112 struct btrfs_extent_item
*extent_item
;
8113 struct btrfs_tree_block_info
*block_info
;
8114 struct btrfs_extent_inline_ref
*iref
;
8115 struct btrfs_path
*path
;
8116 struct extent_buffer
*leaf
;
8117 u32 size
= sizeof(*extent_item
) + sizeof(*iref
);
8118 u64 num_bytes
= ins
->offset
;
8119 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
8122 if (!skinny_metadata
)
8123 size
+= sizeof(*block_info
);
8125 path
= btrfs_alloc_path();
8127 btrfs_free_and_pin_reserved_extent(root
, ins
->objectid
,
8132 path
->leave_spinning
= 1;
8133 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
8136 btrfs_free_path(path
);
8137 btrfs_free_and_pin_reserved_extent(root
, ins
->objectid
,
8142 leaf
= path
->nodes
[0];
8143 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
8144 struct btrfs_extent_item
);
8145 btrfs_set_extent_refs(leaf
, extent_item
, 1);
8146 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
8147 btrfs_set_extent_flags(leaf
, extent_item
,
8148 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
8150 if (skinny_metadata
) {
8151 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
8152 num_bytes
= root
->nodesize
;
8154 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
8155 btrfs_set_tree_block_key(leaf
, block_info
, key
);
8156 btrfs_set_tree_block_level(leaf
, block_info
, level
);
8157 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
8161 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
8162 btrfs_set_extent_inline_ref_type(leaf
, iref
,
8163 BTRFS_SHARED_BLOCK_REF_KEY
);
8164 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
8166 btrfs_set_extent_inline_ref_type(leaf
, iref
,
8167 BTRFS_TREE_BLOCK_REF_KEY
);
8168 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
8171 btrfs_mark_buffer_dirty(leaf
);
8172 btrfs_free_path(path
);
8174 ret
= remove_from_free_space_tree(trans
, fs_info
, ins
->objectid
,
8179 ret
= update_block_group(trans
, root
, ins
->objectid
, root
->nodesize
,
8181 if (ret
) { /* -ENOENT, logic error */
8182 btrfs_err(fs_info
, "update block group failed for %llu %llu",
8183 ins
->objectid
, ins
->offset
);
8187 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, root
->nodesize
);
8191 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
8192 struct btrfs_root
*root
,
8193 u64 root_objectid
, u64 owner
,
8194 u64 offset
, u64 ram_bytes
,
8195 struct btrfs_key
*ins
)
8199 BUG_ON(root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
8201 ret
= btrfs_add_delayed_data_ref(root
->fs_info
, trans
, ins
->objectid
,
8203 root_objectid
, owner
, offset
,
8204 ram_bytes
, BTRFS_ADD_DELAYED_EXTENT
,
8210 * this is used by the tree logging recovery code. It records that
8211 * an extent has been allocated and makes sure to clear the free
8212 * space cache bits as well
8214 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
8215 struct btrfs_root
*root
,
8216 u64 root_objectid
, u64 owner
, u64 offset
,
8217 struct btrfs_key
*ins
)
8220 struct btrfs_block_group_cache
*block_group
;
8223 * Mixed block groups will exclude before processing the log so we only
8224 * need to do the exclude dance if this fs isn't mixed.
8226 if (!btrfs_fs_incompat(root
->fs_info
, MIXED_GROUPS
)) {
8227 ret
= __exclude_logged_extent(root
, ins
->objectid
, ins
->offset
);
8232 block_group
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
8236 ret
= btrfs_add_reserved_bytes(block_group
, ins
->offset
,
8238 BUG_ON(ret
); /* logic error */
8239 ret
= alloc_reserved_file_extent(trans
, root
, 0, root_objectid
,
8240 0, owner
, offset
, ins
, 1);
8241 btrfs_put_block_group(block_group
);
8245 static struct extent_buffer
*
8246 btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
8247 u64 bytenr
, int level
)
8249 struct extent_buffer
*buf
;
8251 buf
= btrfs_find_create_tree_block(root
, bytenr
);
8255 btrfs_set_header_generation(buf
, trans
->transid
);
8256 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, buf
, level
);
8257 btrfs_tree_lock(buf
);
8258 clean_tree_block(trans
, root
->fs_info
, buf
);
8259 clear_bit(EXTENT_BUFFER_STALE
, &buf
->bflags
);
8261 btrfs_set_lock_blocking(buf
);
8262 set_extent_buffer_uptodate(buf
);
8264 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
8265 buf
->log_index
= root
->log_transid
% 2;
8267 * we allow two log transactions at a time, use different
8268 * EXENT bit to differentiate dirty pages.
8270 if (buf
->log_index
== 0)
8271 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
8272 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
8274 set_extent_new(&root
->dirty_log_pages
, buf
->start
,
8275 buf
->start
+ buf
->len
- 1);
8277 buf
->log_index
= -1;
8278 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
8279 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
8281 trans
->dirty
= true;
8282 /* this returns a buffer locked for blocking */
8286 static struct btrfs_block_rsv
*
8287 use_block_rsv(struct btrfs_trans_handle
*trans
,
8288 struct btrfs_root
*root
, u32 blocksize
)
8290 struct btrfs_block_rsv
*block_rsv
;
8291 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
8293 bool global_updated
= false;
8295 block_rsv
= get_block_rsv(trans
, root
);
8297 if (unlikely(block_rsv
->size
== 0))
8300 ret
= block_rsv_use_bytes(block_rsv
, blocksize
);
8304 if (block_rsv
->failfast
)
8305 return ERR_PTR(ret
);
8307 if (block_rsv
->type
== BTRFS_BLOCK_RSV_GLOBAL
&& !global_updated
) {
8308 global_updated
= true;
8309 update_global_block_rsv(root
->fs_info
);
8313 if (btrfs_test_opt(root
->fs_info
, ENOSPC_DEBUG
)) {
8314 static DEFINE_RATELIMIT_STATE(_rs
,
8315 DEFAULT_RATELIMIT_INTERVAL
* 10,
8316 /*DEFAULT_RATELIMIT_BURST*/ 1);
8317 if (__ratelimit(&_rs
))
8319 "BTRFS: block rsv returned %d\n", ret
);
8322 ret
= reserve_metadata_bytes(root
, block_rsv
, blocksize
,
8323 BTRFS_RESERVE_NO_FLUSH
);
8327 * If we couldn't reserve metadata bytes try and use some from
8328 * the global reserve if its space type is the same as the global
8331 if (block_rsv
->type
!= BTRFS_BLOCK_RSV_GLOBAL
&&
8332 block_rsv
->space_info
== global_rsv
->space_info
) {
8333 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
8337 return ERR_PTR(ret
);
8340 static void unuse_block_rsv(struct btrfs_fs_info
*fs_info
,
8341 struct btrfs_block_rsv
*block_rsv
, u32 blocksize
)
8343 block_rsv_add_bytes(block_rsv
, blocksize
, 0);
8344 block_rsv_release_bytes(fs_info
, block_rsv
, NULL
, 0);
8348 * finds a free extent and does all the dirty work required for allocation
8349 * returns the tree buffer or an ERR_PTR on error.
8351 struct extent_buffer
*btrfs_alloc_tree_block(struct btrfs_trans_handle
*trans
,
8352 struct btrfs_root
*root
,
8353 u64 parent
, u64 root_objectid
,
8354 struct btrfs_disk_key
*key
, int level
,
8355 u64 hint
, u64 empty_size
)
8357 struct btrfs_key ins
;
8358 struct btrfs_block_rsv
*block_rsv
;
8359 struct extent_buffer
*buf
;
8360 struct btrfs_delayed_extent_op
*extent_op
;
8363 u32 blocksize
= root
->nodesize
;
8364 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
8367 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8368 if (btrfs_is_testing(root
->fs_info
)) {
8369 buf
= btrfs_init_new_buffer(trans
, root
, root
->alloc_bytenr
,
8372 root
->alloc_bytenr
+= blocksize
;
8377 block_rsv
= use_block_rsv(trans
, root
, blocksize
);
8378 if (IS_ERR(block_rsv
))
8379 return ERR_CAST(block_rsv
);
8381 ret
= btrfs_reserve_extent(root
, blocksize
, blocksize
, blocksize
,
8382 empty_size
, hint
, &ins
, 0, 0);
8386 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
, level
);
8389 goto out_free_reserved
;
8392 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
8394 parent
= ins
.objectid
;
8395 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
8399 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
8400 extent_op
= btrfs_alloc_delayed_extent_op();
8406 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
8408 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
8409 extent_op
->flags_to_set
= flags
;
8410 extent_op
->update_key
= skinny_metadata
? false : true;
8411 extent_op
->update_flags
= true;
8412 extent_op
->is_data
= false;
8413 extent_op
->level
= level
;
8415 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
8416 ins
.objectid
, ins
.offset
,
8417 parent
, root_objectid
, level
,
8418 BTRFS_ADD_DELAYED_EXTENT
,
8421 goto out_free_delayed
;
8426 btrfs_free_delayed_extent_op(extent_op
);
8428 free_extent_buffer(buf
);
8430 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
, 0);
8432 unuse_block_rsv(root
->fs_info
, block_rsv
, blocksize
);
8433 return ERR_PTR(ret
);
8436 struct walk_control
{
8437 u64 refs
[BTRFS_MAX_LEVEL
];
8438 u64 flags
[BTRFS_MAX_LEVEL
];
8439 struct btrfs_key update_progress
;
8450 #define DROP_REFERENCE 1
8451 #define UPDATE_BACKREF 2
8453 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
8454 struct btrfs_root
*root
,
8455 struct walk_control
*wc
,
8456 struct btrfs_path
*path
)
8464 struct btrfs_key key
;
8465 struct extent_buffer
*eb
;
8470 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
8471 wc
->reada_count
= wc
->reada_count
* 2 / 3;
8472 wc
->reada_count
= max(wc
->reada_count
, 2);
8474 wc
->reada_count
= wc
->reada_count
* 3 / 2;
8475 wc
->reada_count
= min_t(int, wc
->reada_count
,
8476 BTRFS_NODEPTRS_PER_BLOCK(root
));
8479 eb
= path
->nodes
[wc
->level
];
8480 nritems
= btrfs_header_nritems(eb
);
8481 blocksize
= root
->nodesize
;
8483 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
8484 if (nread
>= wc
->reada_count
)
8488 bytenr
= btrfs_node_blockptr(eb
, slot
);
8489 generation
= btrfs_node_ptr_generation(eb
, slot
);
8491 if (slot
== path
->slots
[wc
->level
])
8494 if (wc
->stage
== UPDATE_BACKREF
&&
8495 generation
<= root
->root_key
.offset
)
8498 /* We don't lock the tree block, it's OK to be racy here */
8499 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
,
8500 wc
->level
- 1, 1, &refs
,
8502 /* We don't care about errors in readahead. */
8507 if (wc
->stage
== DROP_REFERENCE
) {
8511 if (wc
->level
== 1 &&
8512 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8514 if (!wc
->update_ref
||
8515 generation
<= root
->root_key
.offset
)
8517 btrfs_node_key_to_cpu(eb
, &key
, slot
);
8518 ret
= btrfs_comp_cpu_keys(&key
,
8519 &wc
->update_progress
);
8523 if (wc
->level
== 1 &&
8524 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8528 readahead_tree_block(root
, bytenr
);
8531 wc
->reada_slot
= slot
;
8534 static int account_leaf_items(struct btrfs_trans_handle
*trans
,
8535 struct btrfs_root
*root
,
8536 struct extent_buffer
*eb
)
8538 int nr
= btrfs_header_nritems(eb
);
8539 int i
, extent_type
, ret
;
8540 struct btrfs_key key
;
8541 struct btrfs_file_extent_item
*fi
;
8542 u64 bytenr
, num_bytes
;
8544 /* We can be called directly from walk_up_proc() */
8545 if (!root
->fs_info
->quota_enabled
)
8548 for (i
= 0; i
< nr
; i
++) {
8549 btrfs_item_key_to_cpu(eb
, &key
, i
);
8551 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
8554 fi
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
8555 /* filter out non qgroup-accountable extents */
8556 extent_type
= btrfs_file_extent_type(eb
, fi
);
8558 if (extent_type
== BTRFS_FILE_EXTENT_INLINE
)
8561 bytenr
= btrfs_file_extent_disk_bytenr(eb
, fi
);
8565 num_bytes
= btrfs_file_extent_disk_num_bytes(eb
, fi
);
8567 ret
= btrfs_qgroup_insert_dirty_extent(trans
, root
->fs_info
,
8568 bytenr
, num_bytes
, GFP_NOFS
);
8576 * Walk up the tree from the bottom, freeing leaves and any interior
8577 * nodes which have had all slots visited. If a node (leaf or
8578 * interior) is freed, the node above it will have it's slot
8579 * incremented. The root node will never be freed.
8581 * At the end of this function, we should have a path which has all
8582 * slots incremented to the next position for a search. If we need to
8583 * read a new node it will be NULL and the node above it will have the
8584 * correct slot selected for a later read.
8586 * If we increment the root nodes slot counter past the number of
8587 * elements, 1 is returned to signal completion of the search.
8589 static int adjust_slots_upwards(struct btrfs_root
*root
,
8590 struct btrfs_path
*path
, int root_level
)
8594 struct extent_buffer
*eb
;
8596 if (root_level
== 0)
8599 while (level
<= root_level
) {
8600 eb
= path
->nodes
[level
];
8601 nr
= btrfs_header_nritems(eb
);
8602 path
->slots
[level
]++;
8603 slot
= path
->slots
[level
];
8604 if (slot
>= nr
|| level
== 0) {
8606 * Don't free the root - we will detect this
8607 * condition after our loop and return a
8608 * positive value for caller to stop walking the tree.
8610 if (level
!= root_level
) {
8611 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8612 path
->locks
[level
] = 0;
8614 free_extent_buffer(eb
);
8615 path
->nodes
[level
] = NULL
;
8616 path
->slots
[level
] = 0;
8620 * We have a valid slot to walk back down
8621 * from. Stop here so caller can process these
8630 eb
= path
->nodes
[root_level
];
8631 if (path
->slots
[root_level
] >= btrfs_header_nritems(eb
))
8638 * root_eb is the subtree root and is locked before this function is called.
8640 static int account_shared_subtree(struct btrfs_trans_handle
*trans
,
8641 struct btrfs_root
*root
,
8642 struct extent_buffer
*root_eb
,
8648 struct extent_buffer
*eb
= root_eb
;
8649 struct btrfs_path
*path
= NULL
;
8651 BUG_ON(root_level
< 0 || root_level
> BTRFS_MAX_LEVEL
);
8652 BUG_ON(root_eb
== NULL
);
8654 if (!root
->fs_info
->quota_enabled
)
8657 if (!extent_buffer_uptodate(root_eb
)) {
8658 ret
= btrfs_read_buffer(root_eb
, root_gen
);
8663 if (root_level
== 0) {
8664 ret
= account_leaf_items(trans
, root
, root_eb
);
8668 path
= btrfs_alloc_path();
8673 * Walk down the tree. Missing extent blocks are filled in as
8674 * we go. Metadata is accounted every time we read a new
8677 * When we reach a leaf, we account for file extent items in it,
8678 * walk back up the tree (adjusting slot pointers as we go)
8679 * and restart the search process.
8681 extent_buffer_get(root_eb
); /* For path */
8682 path
->nodes
[root_level
] = root_eb
;
8683 path
->slots
[root_level
] = 0;
8684 path
->locks
[root_level
] = 0; /* so release_path doesn't try to unlock */
8687 while (level
>= 0) {
8688 if (path
->nodes
[level
] == NULL
) {
8693 /* We need to get child blockptr/gen from
8694 * parent before we can read it. */
8695 eb
= path
->nodes
[level
+ 1];
8696 parent_slot
= path
->slots
[level
+ 1];
8697 child_bytenr
= btrfs_node_blockptr(eb
, parent_slot
);
8698 child_gen
= btrfs_node_ptr_generation(eb
, parent_slot
);
8700 eb
= read_tree_block(root
, child_bytenr
, child_gen
);
8704 } else if (!extent_buffer_uptodate(eb
)) {
8705 free_extent_buffer(eb
);
8710 path
->nodes
[level
] = eb
;
8711 path
->slots
[level
] = 0;
8713 btrfs_tree_read_lock(eb
);
8714 btrfs_set_lock_blocking_rw(eb
, BTRFS_READ_LOCK
);
8715 path
->locks
[level
] = BTRFS_READ_LOCK_BLOCKING
;
8717 ret
= btrfs_qgroup_insert_dirty_extent(trans
,
8718 root
->fs_info
, child_bytenr
,
8719 root
->nodesize
, GFP_NOFS
);
8725 ret
= account_leaf_items(trans
, root
, path
->nodes
[level
]);
8729 /* Nonzero return here means we completed our search */
8730 ret
= adjust_slots_upwards(root
, path
, root_level
);
8734 /* Restart search with new slots */
8743 btrfs_free_path(path
);
8749 * helper to process tree block while walking down the tree.
8751 * when wc->stage == UPDATE_BACKREF, this function updates
8752 * back refs for pointers in the block.
8754 * NOTE: return value 1 means we should stop walking down.
8756 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
8757 struct btrfs_root
*root
,
8758 struct btrfs_path
*path
,
8759 struct walk_control
*wc
, int lookup_info
)
8761 int level
= wc
->level
;
8762 struct extent_buffer
*eb
= path
->nodes
[level
];
8763 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
8766 if (wc
->stage
== UPDATE_BACKREF
&&
8767 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
8771 * when reference count of tree block is 1, it won't increase
8772 * again. once full backref flag is set, we never clear it.
8775 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
8776 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
8777 BUG_ON(!path
->locks
[level
]);
8778 ret
= btrfs_lookup_extent_info(trans
, root
,
8779 eb
->start
, level
, 1,
8782 BUG_ON(ret
== -ENOMEM
);
8785 BUG_ON(wc
->refs
[level
] == 0);
8788 if (wc
->stage
== DROP_REFERENCE
) {
8789 if (wc
->refs
[level
] > 1)
8792 if (path
->locks
[level
] && !wc
->keep_locks
) {
8793 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8794 path
->locks
[level
] = 0;
8799 /* wc->stage == UPDATE_BACKREF */
8800 if (!(wc
->flags
[level
] & flag
)) {
8801 BUG_ON(!path
->locks
[level
]);
8802 ret
= btrfs_inc_ref(trans
, root
, eb
, 1);
8803 BUG_ON(ret
); /* -ENOMEM */
8804 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
8805 BUG_ON(ret
); /* -ENOMEM */
8806 ret
= btrfs_set_disk_extent_flags(trans
, root
, eb
->start
,
8808 btrfs_header_level(eb
), 0);
8809 BUG_ON(ret
); /* -ENOMEM */
8810 wc
->flags
[level
] |= flag
;
8814 * the block is shared by multiple trees, so it's not good to
8815 * keep the tree lock
8817 if (path
->locks
[level
] && level
> 0) {
8818 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8819 path
->locks
[level
] = 0;
8825 * helper to process tree block pointer.
8827 * when wc->stage == DROP_REFERENCE, this function checks
8828 * reference count of the block pointed to. if the block
8829 * is shared and we need update back refs for the subtree
8830 * rooted at the block, this function changes wc->stage to
8831 * UPDATE_BACKREF. if the block is shared and there is no
8832 * need to update back, this function drops the reference
8835 * NOTE: return value 1 means we should stop walking down.
8837 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
8838 struct btrfs_root
*root
,
8839 struct btrfs_path
*path
,
8840 struct walk_control
*wc
, int *lookup_info
)
8846 struct btrfs_key key
;
8847 struct extent_buffer
*next
;
8848 int level
= wc
->level
;
8851 bool need_account
= false;
8853 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
8854 path
->slots
[level
]);
8856 * if the lower level block was created before the snapshot
8857 * was created, we know there is no need to update back refs
8860 if (wc
->stage
== UPDATE_BACKREF
&&
8861 generation
<= root
->root_key
.offset
) {
8866 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
8867 blocksize
= root
->nodesize
;
8869 next
= btrfs_find_tree_block(root
->fs_info
, bytenr
);
8871 next
= btrfs_find_create_tree_block(root
, bytenr
);
8873 return PTR_ERR(next
);
8875 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, next
,
8879 btrfs_tree_lock(next
);
8880 btrfs_set_lock_blocking(next
);
8882 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, level
- 1, 1,
8883 &wc
->refs
[level
- 1],
8884 &wc
->flags
[level
- 1]);
8886 btrfs_tree_unlock(next
);
8890 if (unlikely(wc
->refs
[level
- 1] == 0)) {
8891 btrfs_err(root
->fs_info
, "Missing references.");
8896 if (wc
->stage
== DROP_REFERENCE
) {
8897 if (wc
->refs
[level
- 1] > 1) {
8898 need_account
= true;
8900 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8903 if (!wc
->update_ref
||
8904 generation
<= root
->root_key
.offset
)
8907 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
8908 path
->slots
[level
]);
8909 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
8913 wc
->stage
= UPDATE_BACKREF
;
8914 wc
->shared_level
= level
- 1;
8918 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8922 if (!btrfs_buffer_uptodate(next
, generation
, 0)) {
8923 btrfs_tree_unlock(next
);
8924 free_extent_buffer(next
);
8930 if (reada
&& level
== 1)
8931 reada_walk_down(trans
, root
, wc
, path
);
8932 next
= read_tree_block(root
, bytenr
, generation
);
8934 return PTR_ERR(next
);
8935 } else if (!extent_buffer_uptodate(next
)) {
8936 free_extent_buffer(next
);
8939 btrfs_tree_lock(next
);
8940 btrfs_set_lock_blocking(next
);
8944 BUG_ON(level
!= btrfs_header_level(next
));
8945 path
->nodes
[level
] = next
;
8946 path
->slots
[level
] = 0;
8947 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8953 wc
->refs
[level
- 1] = 0;
8954 wc
->flags
[level
- 1] = 0;
8955 if (wc
->stage
== DROP_REFERENCE
) {
8956 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
8957 parent
= path
->nodes
[level
]->start
;
8959 BUG_ON(root
->root_key
.objectid
!=
8960 btrfs_header_owner(path
->nodes
[level
]));
8965 ret
= account_shared_subtree(trans
, root
, next
,
8966 generation
, level
- 1);
8968 btrfs_err_rl(root
->fs_info
,
8970 "%d accounting shared subtree. Quota "
8971 "is out of sync, rescan required.",
8975 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
, parent
,
8976 root
->root_key
.objectid
, level
- 1, 0);
8977 BUG_ON(ret
); /* -ENOMEM */
8979 btrfs_tree_unlock(next
);
8980 free_extent_buffer(next
);
8986 * helper to process tree block while walking up the tree.
8988 * when wc->stage == DROP_REFERENCE, this function drops
8989 * reference count on the block.
8991 * when wc->stage == UPDATE_BACKREF, this function changes
8992 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8993 * to UPDATE_BACKREF previously while processing the block.
8995 * NOTE: return value 1 means we should stop walking up.
8997 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
8998 struct btrfs_root
*root
,
8999 struct btrfs_path
*path
,
9000 struct walk_control
*wc
)
9003 int level
= wc
->level
;
9004 struct extent_buffer
*eb
= path
->nodes
[level
];
9007 if (wc
->stage
== UPDATE_BACKREF
) {
9008 BUG_ON(wc
->shared_level
< level
);
9009 if (level
< wc
->shared_level
)
9012 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
9016 wc
->stage
= DROP_REFERENCE
;
9017 wc
->shared_level
= -1;
9018 path
->slots
[level
] = 0;
9021 * check reference count again if the block isn't locked.
9022 * we should start walking down the tree again if reference
9025 if (!path
->locks
[level
]) {
9027 btrfs_tree_lock(eb
);
9028 btrfs_set_lock_blocking(eb
);
9029 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
9031 ret
= btrfs_lookup_extent_info(trans
, root
,
9032 eb
->start
, level
, 1,
9036 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
9037 path
->locks
[level
] = 0;
9040 BUG_ON(wc
->refs
[level
] == 0);
9041 if (wc
->refs
[level
] == 1) {
9042 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
9043 path
->locks
[level
] = 0;
9049 /* wc->stage == DROP_REFERENCE */
9050 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
9052 if (wc
->refs
[level
] == 1) {
9054 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
9055 ret
= btrfs_dec_ref(trans
, root
, eb
, 1);
9057 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
9058 BUG_ON(ret
); /* -ENOMEM */
9059 ret
= account_leaf_items(trans
, root
, eb
);
9061 btrfs_err_rl(root
->fs_info
,
9063 "%d accounting leaf items. Quota "
9064 "is out of sync, rescan required.",
9068 /* make block locked assertion in clean_tree_block happy */
9069 if (!path
->locks
[level
] &&
9070 btrfs_header_generation(eb
) == trans
->transid
) {
9071 btrfs_tree_lock(eb
);
9072 btrfs_set_lock_blocking(eb
);
9073 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
9075 clean_tree_block(trans
, root
->fs_info
, eb
);
9078 if (eb
== root
->node
) {
9079 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
9082 BUG_ON(root
->root_key
.objectid
!=
9083 btrfs_header_owner(eb
));
9085 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
9086 parent
= path
->nodes
[level
+ 1]->start
;
9088 BUG_ON(root
->root_key
.objectid
!=
9089 btrfs_header_owner(path
->nodes
[level
+ 1]));
9092 btrfs_free_tree_block(trans
, root
, eb
, parent
, wc
->refs
[level
] == 1);
9094 wc
->refs
[level
] = 0;
9095 wc
->flags
[level
] = 0;
9099 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
9100 struct btrfs_root
*root
,
9101 struct btrfs_path
*path
,
9102 struct walk_control
*wc
)
9104 int level
= wc
->level
;
9105 int lookup_info
= 1;
9108 while (level
>= 0) {
9109 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
9116 if (path
->slots
[level
] >=
9117 btrfs_header_nritems(path
->nodes
[level
]))
9120 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
9122 path
->slots
[level
]++;
9131 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
9132 struct btrfs_root
*root
,
9133 struct btrfs_path
*path
,
9134 struct walk_control
*wc
, int max_level
)
9136 int level
= wc
->level
;
9139 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
9140 while (level
< max_level
&& path
->nodes
[level
]) {
9142 if (path
->slots
[level
] + 1 <
9143 btrfs_header_nritems(path
->nodes
[level
])) {
9144 path
->slots
[level
]++;
9147 ret
= walk_up_proc(trans
, root
, path
, wc
);
9151 if (path
->locks
[level
]) {
9152 btrfs_tree_unlock_rw(path
->nodes
[level
],
9153 path
->locks
[level
]);
9154 path
->locks
[level
] = 0;
9156 free_extent_buffer(path
->nodes
[level
]);
9157 path
->nodes
[level
] = NULL
;
9165 * drop a subvolume tree.
9167 * this function traverses the tree freeing any blocks that only
9168 * referenced by the tree.
9170 * when a shared tree block is found. this function decreases its
9171 * reference count by one. if update_ref is true, this function
9172 * also make sure backrefs for the shared block and all lower level
9173 * blocks are properly updated.
9175 * If called with for_reloc == 0, may exit early with -EAGAIN
9177 int btrfs_drop_snapshot(struct btrfs_root
*root
,
9178 struct btrfs_block_rsv
*block_rsv
, int update_ref
,
9181 struct btrfs_path
*path
;
9182 struct btrfs_trans_handle
*trans
;
9183 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
9184 struct btrfs_root_item
*root_item
= &root
->root_item
;
9185 struct walk_control
*wc
;
9186 struct btrfs_key key
;
9190 bool root_dropped
= false;
9192 btrfs_debug(root
->fs_info
, "Drop subvolume %llu", root
->objectid
);
9194 path
= btrfs_alloc_path();
9200 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
9202 btrfs_free_path(path
);
9207 trans
= btrfs_start_transaction(tree_root
, 0);
9208 if (IS_ERR(trans
)) {
9209 err
= PTR_ERR(trans
);
9214 trans
->block_rsv
= block_rsv
;
9216 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
9217 level
= btrfs_header_level(root
->node
);
9218 path
->nodes
[level
] = btrfs_lock_root_node(root
);
9219 btrfs_set_lock_blocking(path
->nodes
[level
]);
9220 path
->slots
[level
] = 0;
9221 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
9222 memset(&wc
->update_progress
, 0,
9223 sizeof(wc
->update_progress
));
9225 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
9226 memcpy(&wc
->update_progress
, &key
,
9227 sizeof(wc
->update_progress
));
9229 level
= root_item
->drop_level
;
9231 path
->lowest_level
= level
;
9232 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
9233 path
->lowest_level
= 0;
9241 * unlock our path, this is safe because only this
9242 * function is allowed to delete this snapshot
9244 btrfs_unlock_up_safe(path
, 0);
9246 level
= btrfs_header_level(root
->node
);
9248 btrfs_tree_lock(path
->nodes
[level
]);
9249 btrfs_set_lock_blocking(path
->nodes
[level
]);
9250 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
9252 ret
= btrfs_lookup_extent_info(trans
, root
,
9253 path
->nodes
[level
]->start
,
9254 level
, 1, &wc
->refs
[level
],
9260 BUG_ON(wc
->refs
[level
] == 0);
9262 if (level
== root_item
->drop_level
)
9265 btrfs_tree_unlock(path
->nodes
[level
]);
9266 path
->locks
[level
] = 0;
9267 WARN_ON(wc
->refs
[level
] != 1);
9273 wc
->shared_level
= -1;
9274 wc
->stage
= DROP_REFERENCE
;
9275 wc
->update_ref
= update_ref
;
9277 wc
->for_reloc
= for_reloc
;
9278 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
9282 ret
= walk_down_tree(trans
, root
, path
, wc
);
9288 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
9295 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
9299 if (wc
->stage
== DROP_REFERENCE
) {
9301 btrfs_node_key(path
->nodes
[level
],
9302 &root_item
->drop_progress
,
9303 path
->slots
[level
]);
9304 root_item
->drop_level
= level
;
9307 BUG_ON(wc
->level
== 0);
9308 if (btrfs_should_end_transaction(trans
, tree_root
) ||
9309 (!for_reloc
&& btrfs_need_cleaner_sleep(root
))) {
9310 ret
= btrfs_update_root(trans
, tree_root
,
9314 btrfs_abort_transaction(trans
, ret
);
9319 btrfs_end_transaction_throttle(trans
, tree_root
);
9320 if (!for_reloc
&& btrfs_need_cleaner_sleep(root
)) {
9321 pr_debug("BTRFS: drop snapshot early exit\n");
9326 trans
= btrfs_start_transaction(tree_root
, 0);
9327 if (IS_ERR(trans
)) {
9328 err
= PTR_ERR(trans
);
9332 trans
->block_rsv
= block_rsv
;
9335 btrfs_release_path(path
);
9339 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
9341 btrfs_abort_transaction(trans
, ret
);
9345 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
9346 ret
= btrfs_find_root(tree_root
, &root
->root_key
, path
,
9349 btrfs_abort_transaction(trans
, ret
);
9352 } else if (ret
> 0) {
9353 /* if we fail to delete the orphan item this time
9354 * around, it'll get picked up the next time.
9356 * The most common failure here is just -ENOENT.
9358 btrfs_del_orphan_item(trans
, tree_root
,
9359 root
->root_key
.objectid
);
9363 if (test_bit(BTRFS_ROOT_IN_RADIX
, &root
->state
)) {
9364 btrfs_add_dropped_root(trans
, root
);
9366 free_extent_buffer(root
->node
);
9367 free_extent_buffer(root
->commit_root
);
9368 btrfs_put_fs_root(root
);
9370 root_dropped
= true;
9372 btrfs_end_transaction_throttle(trans
, tree_root
);
9375 btrfs_free_path(path
);
9378 * So if we need to stop dropping the snapshot for whatever reason we
9379 * need to make sure to add it back to the dead root list so that we
9380 * keep trying to do the work later. This also cleans up roots if we
9381 * don't have it in the radix (like when we recover after a power fail
9382 * or unmount) so we don't leak memory.
9384 if (!for_reloc
&& root_dropped
== false)
9385 btrfs_add_dead_root(root
);
9386 if (err
&& err
!= -EAGAIN
)
9387 btrfs_handle_fs_error(root
->fs_info
, err
, NULL
);
9392 * drop subtree rooted at tree block 'node'.
9394 * NOTE: this function will unlock and release tree block 'node'
9395 * only used by relocation code
9397 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
9398 struct btrfs_root
*root
,
9399 struct extent_buffer
*node
,
9400 struct extent_buffer
*parent
)
9402 struct btrfs_path
*path
;
9403 struct walk_control
*wc
;
9409 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
9411 path
= btrfs_alloc_path();
9415 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
9417 btrfs_free_path(path
);
9421 btrfs_assert_tree_locked(parent
);
9422 parent_level
= btrfs_header_level(parent
);
9423 extent_buffer_get(parent
);
9424 path
->nodes
[parent_level
] = parent
;
9425 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
9427 btrfs_assert_tree_locked(node
);
9428 level
= btrfs_header_level(node
);
9429 path
->nodes
[level
] = node
;
9430 path
->slots
[level
] = 0;
9431 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
9433 wc
->refs
[parent_level
] = 1;
9434 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
9436 wc
->shared_level
= -1;
9437 wc
->stage
= DROP_REFERENCE
;
9441 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
9444 wret
= walk_down_tree(trans
, root
, path
, wc
);
9450 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
9458 btrfs_free_path(path
);
9462 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
9468 * if restripe for this chunk_type is on pick target profile and
9469 * return, otherwise do the usual balance
9471 stripped
= get_restripe_target(root
->fs_info
, flags
);
9473 return extended_to_chunk(stripped
);
9475 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
9477 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
9478 BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
|
9479 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
9481 if (num_devices
== 1) {
9482 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
9483 stripped
= flags
& ~stripped
;
9485 /* turn raid0 into single device chunks */
9486 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
9489 /* turn mirroring into duplication */
9490 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
9491 BTRFS_BLOCK_GROUP_RAID10
))
9492 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
9494 /* they already had raid on here, just return */
9495 if (flags
& stripped
)
9498 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
9499 stripped
= flags
& ~stripped
;
9501 /* switch duplicated blocks with raid1 */
9502 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
9503 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
9505 /* this is drive concat, leave it alone */
9511 static int inc_block_group_ro(struct btrfs_block_group_cache
*cache
, int force
)
9513 struct btrfs_space_info
*sinfo
= cache
->space_info
;
9515 u64 min_allocable_bytes
;
9519 * We need some metadata space and system metadata space for
9520 * allocating chunks in some corner cases until we force to set
9521 * it to be readonly.
9524 (BTRFS_BLOCK_GROUP_SYSTEM
| BTRFS_BLOCK_GROUP_METADATA
)) &&
9526 min_allocable_bytes
= SZ_1M
;
9528 min_allocable_bytes
= 0;
9530 spin_lock(&sinfo
->lock
);
9531 spin_lock(&cache
->lock
);
9539 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
9540 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
9542 if (sinfo
->bytes_used
+ sinfo
->bytes_reserved
+ sinfo
->bytes_pinned
+
9543 sinfo
->bytes_may_use
+ sinfo
->bytes_readonly
+ num_bytes
+
9544 min_allocable_bytes
<= sinfo
->total_bytes
) {
9545 sinfo
->bytes_readonly
+= num_bytes
;
9547 list_add_tail(&cache
->ro_list
, &sinfo
->ro_bgs
);
9551 spin_unlock(&cache
->lock
);
9552 spin_unlock(&sinfo
->lock
);
9556 int btrfs_inc_block_group_ro(struct btrfs_root
*root
,
9557 struct btrfs_block_group_cache
*cache
)
9560 struct btrfs_trans_handle
*trans
;
9565 trans
= btrfs_join_transaction(root
);
9567 return PTR_ERR(trans
);
9570 * we're not allowed to set block groups readonly after the dirty
9571 * block groups cache has started writing. If it already started,
9572 * back off and let this transaction commit
9574 mutex_lock(&root
->fs_info
->ro_block_group_mutex
);
9575 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN
, &trans
->transaction
->flags
)) {
9576 u64 transid
= trans
->transid
;
9578 mutex_unlock(&root
->fs_info
->ro_block_group_mutex
);
9579 btrfs_end_transaction(trans
, root
);
9581 ret
= btrfs_wait_for_commit(root
, transid
);
9588 * if we are changing raid levels, try to allocate a corresponding
9589 * block group with the new raid level.
9591 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
9592 if (alloc_flags
!= cache
->flags
) {
9593 ret
= do_chunk_alloc(trans
, root
, alloc_flags
,
9596 * ENOSPC is allowed here, we may have enough space
9597 * already allocated at the new raid level to
9606 ret
= inc_block_group_ro(cache
, 0);
9609 alloc_flags
= get_alloc_profile(root
, cache
->space_info
->flags
);
9610 ret
= do_chunk_alloc(trans
, root
, alloc_flags
,
9614 ret
= inc_block_group_ro(cache
, 0);
9616 if (cache
->flags
& BTRFS_BLOCK_GROUP_SYSTEM
) {
9617 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
9618 lock_chunks(root
->fs_info
->chunk_root
);
9619 check_system_chunk(trans
, root
, alloc_flags
);
9620 unlock_chunks(root
->fs_info
->chunk_root
);
9622 mutex_unlock(&root
->fs_info
->ro_block_group_mutex
);
9624 btrfs_end_transaction(trans
, root
);
9628 int btrfs_force_chunk_alloc(struct btrfs_trans_handle
*trans
,
9629 struct btrfs_root
*root
, u64 type
)
9631 u64 alloc_flags
= get_alloc_profile(root
, type
);
9632 return do_chunk_alloc(trans
, root
, alloc_flags
,
9637 * helper to account the unused space of all the readonly block group in the
9638 * space_info. takes mirrors into account.
9640 u64
btrfs_account_ro_block_groups_free_space(struct btrfs_space_info
*sinfo
)
9642 struct btrfs_block_group_cache
*block_group
;
9646 /* It's df, we don't care if it's racy */
9647 if (list_empty(&sinfo
->ro_bgs
))
9650 spin_lock(&sinfo
->lock
);
9651 list_for_each_entry(block_group
, &sinfo
->ro_bgs
, ro_list
) {
9652 spin_lock(&block_group
->lock
);
9654 if (!block_group
->ro
) {
9655 spin_unlock(&block_group
->lock
);
9659 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_RAID1
|
9660 BTRFS_BLOCK_GROUP_RAID10
|
9661 BTRFS_BLOCK_GROUP_DUP
))
9666 free_bytes
+= (block_group
->key
.offset
-
9667 btrfs_block_group_used(&block_group
->item
)) *
9670 spin_unlock(&block_group
->lock
);
9672 spin_unlock(&sinfo
->lock
);
9677 void btrfs_dec_block_group_ro(struct btrfs_root
*root
,
9678 struct btrfs_block_group_cache
*cache
)
9680 struct btrfs_space_info
*sinfo
= cache
->space_info
;
9685 spin_lock(&sinfo
->lock
);
9686 spin_lock(&cache
->lock
);
9688 num_bytes
= cache
->key
.offset
- cache
->reserved
-
9689 cache
->pinned
- cache
->bytes_super
-
9690 btrfs_block_group_used(&cache
->item
);
9691 sinfo
->bytes_readonly
-= num_bytes
;
9692 list_del_init(&cache
->ro_list
);
9694 spin_unlock(&cache
->lock
);
9695 spin_unlock(&sinfo
->lock
);
9699 * checks to see if its even possible to relocate this block group.
9701 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9702 * ok to go ahead and try.
9704 int btrfs_can_relocate(struct btrfs_root
*root
, u64 bytenr
)
9706 struct btrfs_block_group_cache
*block_group
;
9707 struct btrfs_space_info
*space_info
;
9708 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
9709 struct btrfs_device
*device
;
9710 struct btrfs_trans_handle
*trans
;
9720 debug
= btrfs_test_opt(root
->fs_info
, ENOSPC_DEBUG
);
9722 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
9724 /* odd, couldn't find the block group, leave it alone */
9727 btrfs_warn(root
->fs_info
,
9728 "can't find block group for bytenr %llu",
9733 min_free
= btrfs_block_group_used(&block_group
->item
);
9735 /* no bytes used, we're good */
9739 space_info
= block_group
->space_info
;
9740 spin_lock(&space_info
->lock
);
9742 full
= space_info
->full
;
9745 * if this is the last block group we have in this space, we can't
9746 * relocate it unless we're able to allocate a new chunk below.
9748 * Otherwise, we need to make sure we have room in the space to handle
9749 * all of the extents from this block group. If we can, we're good
9751 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
9752 (space_info
->bytes_used
+ space_info
->bytes_reserved
+
9753 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
9754 min_free
< space_info
->total_bytes
)) {
9755 spin_unlock(&space_info
->lock
);
9758 spin_unlock(&space_info
->lock
);
9761 * ok we don't have enough space, but maybe we have free space on our
9762 * devices to allocate new chunks for relocation, so loop through our
9763 * alloc devices and guess if we have enough space. if this block
9764 * group is going to be restriped, run checks against the target
9765 * profile instead of the current one.
9777 target
= get_restripe_target(root
->fs_info
, block_group
->flags
);
9779 index
= __get_raid_index(extended_to_chunk(target
));
9782 * this is just a balance, so if we were marked as full
9783 * we know there is no space for a new chunk
9787 btrfs_warn(root
->fs_info
,
9788 "no space to alloc new chunk for block group %llu",
9789 block_group
->key
.objectid
);
9793 index
= get_block_group_index(block_group
);
9796 if (index
== BTRFS_RAID_RAID10
) {
9800 } else if (index
== BTRFS_RAID_RAID1
) {
9802 } else if (index
== BTRFS_RAID_DUP
) {
9805 } else if (index
== BTRFS_RAID_RAID0
) {
9806 dev_min
= fs_devices
->rw_devices
;
9807 min_free
= div64_u64(min_free
, dev_min
);
9810 /* We need to do this so that we can look at pending chunks */
9811 trans
= btrfs_join_transaction(root
);
9812 if (IS_ERR(trans
)) {
9813 ret
= PTR_ERR(trans
);
9817 mutex_lock(&root
->fs_info
->chunk_mutex
);
9818 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
9822 * check to make sure we can actually find a chunk with enough
9823 * space to fit our block group in.
9825 if (device
->total_bytes
> device
->bytes_used
+ min_free
&&
9826 !device
->is_tgtdev_for_dev_replace
) {
9827 ret
= find_free_dev_extent(trans
, device
, min_free
,
9832 if (dev_nr
>= dev_min
)
9838 if (debug
&& ret
== -1)
9839 btrfs_warn(root
->fs_info
,
9840 "no space to allocate a new chunk for block group %llu",
9841 block_group
->key
.objectid
);
9842 mutex_unlock(&root
->fs_info
->chunk_mutex
);
9843 btrfs_end_transaction(trans
, root
);
9845 btrfs_put_block_group(block_group
);
9849 static int find_first_block_group(struct btrfs_root
*root
,
9850 struct btrfs_path
*path
, struct btrfs_key
*key
)
9853 struct btrfs_key found_key
;
9854 struct extent_buffer
*leaf
;
9857 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
9862 slot
= path
->slots
[0];
9863 leaf
= path
->nodes
[0];
9864 if (slot
>= btrfs_header_nritems(leaf
)) {
9865 ret
= btrfs_next_leaf(root
, path
);
9872 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
9874 if (found_key
.objectid
>= key
->objectid
&&
9875 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
9876 struct extent_map_tree
*em_tree
;
9877 struct extent_map
*em
;
9879 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
9880 read_lock(&em_tree
->lock
);
9881 em
= lookup_extent_mapping(em_tree
, found_key
.objectid
,
9883 read_unlock(&em_tree
->lock
);
9885 btrfs_err(root
->fs_info
,
9886 "logical %llu len %llu found bg but no related chunk",
9887 found_key
.objectid
, found_key
.offset
);
9892 free_extent_map(em
);
9901 void btrfs_put_block_group_cache(struct btrfs_fs_info
*info
)
9903 struct btrfs_block_group_cache
*block_group
;
9907 struct inode
*inode
;
9909 block_group
= btrfs_lookup_first_block_group(info
, last
);
9910 while (block_group
) {
9911 spin_lock(&block_group
->lock
);
9912 if (block_group
->iref
)
9914 spin_unlock(&block_group
->lock
);
9915 block_group
= next_block_group(info
->tree_root
,
9925 inode
= block_group
->inode
;
9926 block_group
->iref
= 0;
9927 block_group
->inode
= NULL
;
9928 spin_unlock(&block_group
->lock
);
9929 ASSERT(block_group
->io_ctl
.inode
== NULL
);
9931 last
= block_group
->key
.objectid
+ block_group
->key
.offset
;
9932 btrfs_put_block_group(block_group
);
9936 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
9938 struct btrfs_block_group_cache
*block_group
;
9939 struct btrfs_space_info
*space_info
;
9940 struct btrfs_caching_control
*caching_ctl
;
9943 down_write(&info
->commit_root_sem
);
9944 while (!list_empty(&info
->caching_block_groups
)) {
9945 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
9946 struct btrfs_caching_control
, list
);
9947 list_del(&caching_ctl
->list
);
9948 put_caching_control(caching_ctl
);
9950 up_write(&info
->commit_root_sem
);
9952 spin_lock(&info
->unused_bgs_lock
);
9953 while (!list_empty(&info
->unused_bgs
)) {
9954 block_group
= list_first_entry(&info
->unused_bgs
,
9955 struct btrfs_block_group_cache
,
9957 list_del_init(&block_group
->bg_list
);
9958 btrfs_put_block_group(block_group
);
9960 spin_unlock(&info
->unused_bgs_lock
);
9962 spin_lock(&info
->block_group_cache_lock
);
9963 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
9964 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
9966 rb_erase(&block_group
->cache_node
,
9967 &info
->block_group_cache_tree
);
9968 RB_CLEAR_NODE(&block_group
->cache_node
);
9969 spin_unlock(&info
->block_group_cache_lock
);
9971 down_write(&block_group
->space_info
->groups_sem
);
9972 list_del(&block_group
->list
);
9973 up_write(&block_group
->space_info
->groups_sem
);
9975 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
9976 wait_block_group_cache_done(block_group
);
9979 * We haven't cached this block group, which means we could
9980 * possibly have excluded extents on this block group.
9982 if (block_group
->cached
== BTRFS_CACHE_NO
||
9983 block_group
->cached
== BTRFS_CACHE_ERROR
)
9984 free_excluded_extents(info
->extent_root
, block_group
);
9986 btrfs_remove_free_space_cache(block_group
);
9987 ASSERT(list_empty(&block_group
->dirty_list
));
9988 ASSERT(list_empty(&block_group
->io_list
));
9989 ASSERT(list_empty(&block_group
->bg_list
));
9990 ASSERT(atomic_read(&block_group
->count
) == 1);
9991 btrfs_put_block_group(block_group
);
9993 spin_lock(&info
->block_group_cache_lock
);
9995 spin_unlock(&info
->block_group_cache_lock
);
9997 /* now that all the block groups are freed, go through and
9998 * free all the space_info structs. This is only called during
9999 * the final stages of unmount, and so we know nobody is
10000 * using them. We call synchronize_rcu() once before we start,
10001 * just to be on the safe side.
10005 release_global_block_rsv(info
);
10007 while (!list_empty(&info
->space_info
)) {
10010 space_info
= list_entry(info
->space_info
.next
,
10011 struct btrfs_space_info
,
10015 * Do not hide this behind enospc_debug, this is actually
10016 * important and indicates a real bug if this happens.
10018 if (WARN_ON(space_info
->bytes_pinned
> 0 ||
10019 space_info
->bytes_reserved
> 0 ||
10020 space_info
->bytes_may_use
> 0))
10021 dump_space_info(space_info
, 0, 0);
10022 list_del(&space_info
->list
);
10023 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++) {
10024 struct kobject
*kobj
;
10025 kobj
= space_info
->block_group_kobjs
[i
];
10026 space_info
->block_group_kobjs
[i
] = NULL
;
10032 kobject_del(&space_info
->kobj
);
10033 kobject_put(&space_info
->kobj
);
10038 static void __link_block_group(struct btrfs_space_info
*space_info
,
10039 struct btrfs_block_group_cache
*cache
)
10041 int index
= get_block_group_index(cache
);
10042 bool first
= false;
10044 down_write(&space_info
->groups_sem
);
10045 if (list_empty(&space_info
->block_groups
[index
]))
10047 list_add_tail(&cache
->list
, &space_info
->block_groups
[index
]);
10048 up_write(&space_info
->groups_sem
);
10051 struct raid_kobject
*rkobj
;
10054 rkobj
= kzalloc(sizeof(*rkobj
), GFP_NOFS
);
10057 rkobj
->raid_type
= index
;
10058 kobject_init(&rkobj
->kobj
, &btrfs_raid_ktype
);
10059 ret
= kobject_add(&rkobj
->kobj
, &space_info
->kobj
,
10060 "%s", get_raid_name(index
));
10062 kobject_put(&rkobj
->kobj
);
10065 space_info
->block_group_kobjs
[index
] = &rkobj
->kobj
;
10070 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
10073 static struct btrfs_block_group_cache
*
10074 btrfs_create_block_group_cache(struct btrfs_root
*root
, u64 start
, u64 size
)
10076 struct btrfs_block_group_cache
*cache
;
10078 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
10082 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
10084 if (!cache
->free_space_ctl
) {
10089 cache
->key
.objectid
= start
;
10090 cache
->key
.offset
= size
;
10091 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
10093 cache
->sectorsize
= root
->sectorsize
;
10094 cache
->fs_info
= root
->fs_info
;
10095 cache
->full_stripe_len
= btrfs_full_stripe_len(root
,
10096 &root
->fs_info
->mapping_tree
,
10098 set_free_space_tree_thresholds(cache
);
10100 atomic_set(&cache
->count
, 1);
10101 spin_lock_init(&cache
->lock
);
10102 init_rwsem(&cache
->data_rwsem
);
10103 INIT_LIST_HEAD(&cache
->list
);
10104 INIT_LIST_HEAD(&cache
->cluster_list
);
10105 INIT_LIST_HEAD(&cache
->bg_list
);
10106 INIT_LIST_HEAD(&cache
->ro_list
);
10107 INIT_LIST_HEAD(&cache
->dirty_list
);
10108 INIT_LIST_HEAD(&cache
->io_list
);
10109 btrfs_init_free_space_ctl(cache
);
10110 atomic_set(&cache
->trimming
, 0);
10111 mutex_init(&cache
->free_space_lock
);
10116 int btrfs_read_block_groups(struct btrfs_root
*root
)
10118 struct btrfs_path
*path
;
10120 struct btrfs_block_group_cache
*cache
;
10121 struct btrfs_fs_info
*info
= root
->fs_info
;
10122 struct btrfs_space_info
*space_info
;
10123 struct btrfs_key key
;
10124 struct btrfs_key found_key
;
10125 struct extent_buffer
*leaf
;
10126 int need_clear
= 0;
10129 root
= info
->extent_root
;
10132 key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
10133 path
= btrfs_alloc_path();
10136 path
->reada
= READA_FORWARD
;
10138 cache_gen
= btrfs_super_cache_generation(root
->fs_info
->super_copy
);
10139 if (btrfs_test_opt(root
->fs_info
, SPACE_CACHE
) &&
10140 btrfs_super_generation(root
->fs_info
->super_copy
) != cache_gen
)
10142 if (btrfs_test_opt(root
->fs_info
, CLEAR_CACHE
))
10146 ret
= find_first_block_group(root
, path
, &key
);
10152 leaf
= path
->nodes
[0];
10153 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
10155 cache
= btrfs_create_block_group_cache(root
, found_key
.objectid
,
10164 * When we mount with old space cache, we need to
10165 * set BTRFS_DC_CLEAR and set dirty flag.
10167 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
10168 * truncate the old free space cache inode and
10170 * b) Setting 'dirty flag' makes sure that we flush
10171 * the new space cache info onto disk.
10173 if (btrfs_test_opt(root
->fs_info
, SPACE_CACHE
))
10174 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
10177 read_extent_buffer(leaf
, &cache
->item
,
10178 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
10179 sizeof(cache
->item
));
10180 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
10182 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
10183 btrfs_release_path(path
);
10186 * We need to exclude the super stripes now so that the space
10187 * info has super bytes accounted for, otherwise we'll think
10188 * we have more space than we actually do.
10190 ret
= exclude_super_stripes(root
, cache
);
10193 * We may have excluded something, so call this just in
10196 free_excluded_extents(root
, cache
);
10197 btrfs_put_block_group(cache
);
10202 * check for two cases, either we are full, and therefore
10203 * don't need to bother with the caching work since we won't
10204 * find any space, or we are empty, and we can just add all
10205 * the space in and be done with it. This saves us _alot_ of
10206 * time, particularly in the full case.
10208 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
10209 cache
->last_byte_to_unpin
= (u64
)-1;
10210 cache
->cached
= BTRFS_CACHE_FINISHED
;
10211 free_excluded_extents(root
, cache
);
10212 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
10213 cache
->last_byte_to_unpin
= (u64
)-1;
10214 cache
->cached
= BTRFS_CACHE_FINISHED
;
10215 add_new_free_space(cache
, root
->fs_info
,
10216 found_key
.objectid
,
10217 found_key
.objectid
+
10219 free_excluded_extents(root
, cache
);
10222 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
10224 btrfs_remove_free_space_cache(cache
);
10225 btrfs_put_block_group(cache
);
10229 trace_btrfs_add_block_group(root
->fs_info
, cache
, 0);
10230 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
10231 btrfs_block_group_used(&cache
->item
),
10232 cache
->bytes_super
, &space_info
);
10234 btrfs_remove_free_space_cache(cache
);
10235 spin_lock(&info
->block_group_cache_lock
);
10236 rb_erase(&cache
->cache_node
,
10237 &info
->block_group_cache_tree
);
10238 RB_CLEAR_NODE(&cache
->cache_node
);
10239 spin_unlock(&info
->block_group_cache_lock
);
10240 btrfs_put_block_group(cache
);
10244 cache
->space_info
= space_info
;
10246 __link_block_group(space_info
, cache
);
10248 set_avail_alloc_bits(root
->fs_info
, cache
->flags
);
10249 if (btrfs_chunk_readonly(root
, cache
->key
.objectid
)) {
10250 inc_block_group_ro(cache
, 1);
10251 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
10252 spin_lock(&info
->unused_bgs_lock
);
10253 /* Should always be true but just in case. */
10254 if (list_empty(&cache
->bg_list
)) {
10255 btrfs_get_block_group(cache
);
10256 list_add_tail(&cache
->bg_list
,
10257 &info
->unused_bgs
);
10259 spin_unlock(&info
->unused_bgs_lock
);
10263 list_for_each_entry_rcu(space_info
, &root
->fs_info
->space_info
, list
) {
10264 if (!(get_alloc_profile(root
, space_info
->flags
) &
10265 (BTRFS_BLOCK_GROUP_RAID10
|
10266 BTRFS_BLOCK_GROUP_RAID1
|
10267 BTRFS_BLOCK_GROUP_RAID5
|
10268 BTRFS_BLOCK_GROUP_RAID6
|
10269 BTRFS_BLOCK_GROUP_DUP
)))
10272 * avoid allocating from un-mirrored block group if there are
10273 * mirrored block groups.
10275 list_for_each_entry(cache
,
10276 &space_info
->block_groups
[BTRFS_RAID_RAID0
],
10278 inc_block_group_ro(cache
, 1);
10279 list_for_each_entry(cache
,
10280 &space_info
->block_groups
[BTRFS_RAID_SINGLE
],
10282 inc_block_group_ro(cache
, 1);
10285 init_global_block_rsv(info
);
10288 btrfs_free_path(path
);
10292 void btrfs_create_pending_block_groups(struct btrfs_trans_handle
*trans
,
10293 struct btrfs_root
*root
)
10295 struct btrfs_block_group_cache
*block_group
, *tmp
;
10296 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
10297 struct btrfs_block_group_item item
;
10298 struct btrfs_key key
;
10300 bool can_flush_pending_bgs
= trans
->can_flush_pending_bgs
;
10302 trans
->can_flush_pending_bgs
= false;
10303 list_for_each_entry_safe(block_group
, tmp
, &trans
->new_bgs
, bg_list
) {
10307 spin_lock(&block_group
->lock
);
10308 memcpy(&item
, &block_group
->item
, sizeof(item
));
10309 memcpy(&key
, &block_group
->key
, sizeof(key
));
10310 spin_unlock(&block_group
->lock
);
10312 ret
= btrfs_insert_item(trans
, extent_root
, &key
, &item
,
10315 btrfs_abort_transaction(trans
, ret
);
10316 ret
= btrfs_finish_chunk_alloc(trans
, extent_root
,
10317 key
.objectid
, key
.offset
);
10319 btrfs_abort_transaction(trans
, ret
);
10320 add_block_group_free_space(trans
, root
->fs_info
, block_group
);
10321 /* already aborted the transaction if it failed. */
10323 list_del_init(&block_group
->bg_list
);
10325 trans
->can_flush_pending_bgs
= can_flush_pending_bgs
;
10328 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
10329 struct btrfs_root
*root
, u64 bytes_used
,
10330 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
10334 struct btrfs_root
*extent_root
;
10335 struct btrfs_block_group_cache
*cache
;
10336 extent_root
= root
->fs_info
->extent_root
;
10338 btrfs_set_log_full_commit(root
->fs_info
, trans
);
10340 cache
= btrfs_create_block_group_cache(root
, chunk_offset
, size
);
10344 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
10345 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
10346 btrfs_set_block_group_flags(&cache
->item
, type
);
10348 cache
->flags
= type
;
10349 cache
->last_byte_to_unpin
= (u64
)-1;
10350 cache
->cached
= BTRFS_CACHE_FINISHED
;
10351 cache
->needs_free_space
= 1;
10352 ret
= exclude_super_stripes(root
, cache
);
10355 * We may have excluded something, so call this just in
10358 free_excluded_extents(root
, cache
);
10359 btrfs_put_block_group(cache
);
10363 add_new_free_space(cache
, root
->fs_info
, chunk_offset
,
10364 chunk_offset
+ size
);
10366 free_excluded_extents(root
, cache
);
10368 #ifdef CONFIG_BTRFS_DEBUG
10369 if (btrfs_should_fragment_free_space(root
, cache
)) {
10370 u64 new_bytes_used
= size
- bytes_used
;
10372 bytes_used
+= new_bytes_used
>> 1;
10373 fragment_free_space(root
, cache
);
10377 * Call to ensure the corresponding space_info object is created and
10378 * assigned to our block group, but don't update its counters just yet.
10379 * We want our bg to be added to the rbtree with its ->space_info set.
10381 ret
= update_space_info(root
->fs_info
, cache
->flags
, 0, 0, 0,
10382 &cache
->space_info
);
10384 btrfs_remove_free_space_cache(cache
);
10385 btrfs_put_block_group(cache
);
10389 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
10391 btrfs_remove_free_space_cache(cache
);
10392 btrfs_put_block_group(cache
);
10397 * Now that our block group has its ->space_info set and is inserted in
10398 * the rbtree, update the space info's counters.
10400 trace_btrfs_add_block_group(root
->fs_info
, cache
, 1);
10401 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
10402 cache
->bytes_super
, &cache
->space_info
);
10404 btrfs_remove_free_space_cache(cache
);
10405 spin_lock(&root
->fs_info
->block_group_cache_lock
);
10406 rb_erase(&cache
->cache_node
,
10407 &root
->fs_info
->block_group_cache_tree
);
10408 RB_CLEAR_NODE(&cache
->cache_node
);
10409 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
10410 btrfs_put_block_group(cache
);
10413 update_global_block_rsv(root
->fs_info
);
10415 __link_block_group(cache
->space_info
, cache
);
10417 list_add_tail(&cache
->bg_list
, &trans
->new_bgs
);
10419 set_avail_alloc_bits(extent_root
->fs_info
, type
);
10423 static void clear_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
10425 u64 extra_flags
= chunk_to_extended(flags
) &
10426 BTRFS_EXTENDED_PROFILE_MASK
;
10428 write_seqlock(&fs_info
->profiles_lock
);
10429 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
10430 fs_info
->avail_data_alloc_bits
&= ~extra_flags
;
10431 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
10432 fs_info
->avail_metadata_alloc_bits
&= ~extra_flags
;
10433 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
10434 fs_info
->avail_system_alloc_bits
&= ~extra_flags
;
10435 write_sequnlock(&fs_info
->profiles_lock
);
10438 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
10439 struct btrfs_root
*root
, u64 group_start
,
10440 struct extent_map
*em
)
10442 struct btrfs_path
*path
;
10443 struct btrfs_block_group_cache
*block_group
;
10444 struct btrfs_free_cluster
*cluster
;
10445 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
10446 struct btrfs_key key
;
10447 struct inode
*inode
;
10448 struct kobject
*kobj
= NULL
;
10452 struct btrfs_caching_control
*caching_ctl
= NULL
;
10455 root
= root
->fs_info
->extent_root
;
10457 block_group
= btrfs_lookup_block_group(root
->fs_info
, group_start
);
10458 BUG_ON(!block_group
);
10459 BUG_ON(!block_group
->ro
);
10462 * Free the reserved super bytes from this block group before
10465 free_excluded_extents(root
, block_group
);
10467 memcpy(&key
, &block_group
->key
, sizeof(key
));
10468 index
= get_block_group_index(block_group
);
10469 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
10470 BTRFS_BLOCK_GROUP_RAID1
|
10471 BTRFS_BLOCK_GROUP_RAID10
))
10476 /* make sure this block group isn't part of an allocation cluster */
10477 cluster
= &root
->fs_info
->data_alloc_cluster
;
10478 spin_lock(&cluster
->refill_lock
);
10479 btrfs_return_cluster_to_free_space(block_group
, cluster
);
10480 spin_unlock(&cluster
->refill_lock
);
10483 * make sure this block group isn't part of a metadata
10484 * allocation cluster
10486 cluster
= &root
->fs_info
->meta_alloc_cluster
;
10487 spin_lock(&cluster
->refill_lock
);
10488 btrfs_return_cluster_to_free_space(block_group
, cluster
);
10489 spin_unlock(&cluster
->refill_lock
);
10491 path
= btrfs_alloc_path();
10498 * get the inode first so any iput calls done for the io_list
10499 * aren't the final iput (no unlinks allowed now)
10501 inode
= lookup_free_space_inode(tree_root
, block_group
, path
);
10503 mutex_lock(&trans
->transaction
->cache_write_mutex
);
10505 * make sure our free spache cache IO is done before remove the
10508 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
10509 if (!list_empty(&block_group
->io_list
)) {
10510 list_del_init(&block_group
->io_list
);
10512 WARN_ON(!IS_ERR(inode
) && inode
!= block_group
->io_ctl
.inode
);
10514 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
10515 btrfs_wait_cache_io(root
, trans
, block_group
,
10516 &block_group
->io_ctl
, path
,
10517 block_group
->key
.objectid
);
10518 btrfs_put_block_group(block_group
);
10519 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
10522 if (!list_empty(&block_group
->dirty_list
)) {
10523 list_del_init(&block_group
->dirty_list
);
10524 btrfs_put_block_group(block_group
);
10526 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
10527 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
10529 if (!IS_ERR(inode
)) {
10530 ret
= btrfs_orphan_add(trans
, inode
);
10532 btrfs_add_delayed_iput(inode
);
10535 clear_nlink(inode
);
10536 /* One for the block groups ref */
10537 spin_lock(&block_group
->lock
);
10538 if (block_group
->iref
) {
10539 block_group
->iref
= 0;
10540 block_group
->inode
= NULL
;
10541 spin_unlock(&block_group
->lock
);
10544 spin_unlock(&block_group
->lock
);
10546 /* One for our lookup ref */
10547 btrfs_add_delayed_iput(inode
);
10550 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
10551 key
.offset
= block_group
->key
.objectid
;
10554 ret
= btrfs_search_slot(trans
, tree_root
, &key
, path
, -1, 1);
10558 btrfs_release_path(path
);
10560 ret
= btrfs_del_item(trans
, tree_root
, path
);
10563 btrfs_release_path(path
);
10566 spin_lock(&root
->fs_info
->block_group_cache_lock
);
10567 rb_erase(&block_group
->cache_node
,
10568 &root
->fs_info
->block_group_cache_tree
);
10569 RB_CLEAR_NODE(&block_group
->cache_node
);
10571 if (root
->fs_info
->first_logical_byte
== block_group
->key
.objectid
)
10572 root
->fs_info
->first_logical_byte
= (u64
)-1;
10573 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
10575 down_write(&block_group
->space_info
->groups_sem
);
10577 * we must use list_del_init so people can check to see if they
10578 * are still on the list after taking the semaphore
10580 list_del_init(&block_group
->list
);
10581 if (list_empty(&block_group
->space_info
->block_groups
[index
])) {
10582 kobj
= block_group
->space_info
->block_group_kobjs
[index
];
10583 block_group
->space_info
->block_group_kobjs
[index
] = NULL
;
10584 clear_avail_alloc_bits(root
->fs_info
, block_group
->flags
);
10586 up_write(&block_group
->space_info
->groups_sem
);
10592 if (block_group
->has_caching_ctl
)
10593 caching_ctl
= get_caching_control(block_group
);
10594 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
10595 wait_block_group_cache_done(block_group
);
10596 if (block_group
->has_caching_ctl
) {
10597 down_write(&root
->fs_info
->commit_root_sem
);
10598 if (!caching_ctl
) {
10599 struct btrfs_caching_control
*ctl
;
10601 list_for_each_entry(ctl
,
10602 &root
->fs_info
->caching_block_groups
, list
)
10603 if (ctl
->block_group
== block_group
) {
10605 atomic_inc(&caching_ctl
->count
);
10610 list_del_init(&caching_ctl
->list
);
10611 up_write(&root
->fs_info
->commit_root_sem
);
10613 /* Once for the caching bgs list and once for us. */
10614 put_caching_control(caching_ctl
);
10615 put_caching_control(caching_ctl
);
10619 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
10620 if (!list_empty(&block_group
->dirty_list
)) {
10623 if (!list_empty(&block_group
->io_list
)) {
10626 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
10627 btrfs_remove_free_space_cache(block_group
);
10629 spin_lock(&block_group
->space_info
->lock
);
10630 list_del_init(&block_group
->ro_list
);
10632 if (btrfs_test_opt(root
->fs_info
, ENOSPC_DEBUG
)) {
10633 WARN_ON(block_group
->space_info
->total_bytes
10634 < block_group
->key
.offset
);
10635 WARN_ON(block_group
->space_info
->bytes_readonly
10636 < block_group
->key
.offset
);
10637 WARN_ON(block_group
->space_info
->disk_total
10638 < block_group
->key
.offset
* factor
);
10640 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
10641 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
10642 block_group
->space_info
->disk_total
-= block_group
->key
.offset
* factor
;
10644 spin_unlock(&block_group
->space_info
->lock
);
10646 memcpy(&key
, &block_group
->key
, sizeof(key
));
10649 if (!list_empty(&em
->list
)) {
10650 /* We're in the transaction->pending_chunks list. */
10651 free_extent_map(em
);
10653 spin_lock(&block_group
->lock
);
10654 block_group
->removed
= 1;
10656 * At this point trimming can't start on this block group, because we
10657 * removed the block group from the tree fs_info->block_group_cache_tree
10658 * so no one can't find it anymore and even if someone already got this
10659 * block group before we removed it from the rbtree, they have already
10660 * incremented block_group->trimming - if they didn't, they won't find
10661 * any free space entries because we already removed them all when we
10662 * called btrfs_remove_free_space_cache().
10664 * And we must not remove the extent map from the fs_info->mapping_tree
10665 * to prevent the same logical address range and physical device space
10666 * ranges from being reused for a new block group. This is because our
10667 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10668 * completely transactionless, so while it is trimming a range the
10669 * currently running transaction might finish and a new one start,
10670 * allowing for new block groups to be created that can reuse the same
10671 * physical device locations unless we take this special care.
10673 * There may also be an implicit trim operation if the file system
10674 * is mounted with -odiscard. The same protections must remain
10675 * in place until the extents have been discarded completely when
10676 * the transaction commit has completed.
10678 remove_em
= (atomic_read(&block_group
->trimming
) == 0);
10680 * Make sure a trimmer task always sees the em in the pinned_chunks list
10681 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10682 * before checking block_group->removed).
10686 * Our em might be in trans->transaction->pending_chunks which
10687 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10688 * and so is the fs_info->pinned_chunks list.
10690 * So at this point we must be holding the chunk_mutex to avoid
10691 * any races with chunk allocation (more specifically at
10692 * volumes.c:contains_pending_extent()), to ensure it always
10693 * sees the em, either in the pending_chunks list or in the
10694 * pinned_chunks list.
10696 list_move_tail(&em
->list
, &root
->fs_info
->pinned_chunks
);
10698 spin_unlock(&block_group
->lock
);
10701 struct extent_map_tree
*em_tree
;
10703 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
10704 write_lock(&em_tree
->lock
);
10706 * The em might be in the pending_chunks list, so make sure the
10707 * chunk mutex is locked, since remove_extent_mapping() will
10708 * delete us from that list.
10710 remove_extent_mapping(em_tree
, em
);
10711 write_unlock(&em_tree
->lock
);
10712 /* once for the tree */
10713 free_extent_map(em
);
10716 unlock_chunks(root
);
10718 ret
= remove_block_group_free_space(trans
, root
->fs_info
, block_group
);
10722 btrfs_put_block_group(block_group
);
10723 btrfs_put_block_group(block_group
);
10725 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
10731 ret
= btrfs_del_item(trans
, root
, path
);
10733 btrfs_free_path(path
);
10737 struct btrfs_trans_handle
*
10738 btrfs_start_trans_remove_block_group(struct btrfs_fs_info
*fs_info
,
10739 const u64 chunk_offset
)
10741 struct extent_map_tree
*em_tree
= &fs_info
->mapping_tree
.map_tree
;
10742 struct extent_map
*em
;
10743 struct map_lookup
*map
;
10744 unsigned int num_items
;
10746 read_lock(&em_tree
->lock
);
10747 em
= lookup_extent_mapping(em_tree
, chunk_offset
, 1);
10748 read_unlock(&em_tree
->lock
);
10749 ASSERT(em
&& em
->start
== chunk_offset
);
10752 * We need to reserve 3 + N units from the metadata space info in order
10753 * to remove a block group (done at btrfs_remove_chunk() and at
10754 * btrfs_remove_block_group()), which are used for:
10756 * 1 unit for adding the free space inode's orphan (located in the tree
10758 * 1 unit for deleting the block group item (located in the extent
10760 * 1 unit for deleting the free space item (located in tree of tree
10762 * N units for deleting N device extent items corresponding to each
10763 * stripe (located in the device tree).
10765 * In order to remove a block group we also need to reserve units in the
10766 * system space info in order to update the chunk tree (update one or
10767 * more device items and remove one chunk item), but this is done at
10768 * btrfs_remove_chunk() through a call to check_system_chunk().
10770 map
= em
->map_lookup
;
10771 num_items
= 3 + map
->num_stripes
;
10772 free_extent_map(em
);
10774 return btrfs_start_transaction_fallback_global_rsv(fs_info
->extent_root
,
10779 * Process the unused_bgs list and remove any that don't have any allocated
10780 * space inside of them.
10782 void btrfs_delete_unused_bgs(struct btrfs_fs_info
*fs_info
)
10784 struct btrfs_block_group_cache
*block_group
;
10785 struct btrfs_space_info
*space_info
;
10786 struct btrfs_root
*root
= fs_info
->extent_root
;
10787 struct btrfs_trans_handle
*trans
;
10790 if (!fs_info
->open
)
10793 spin_lock(&fs_info
->unused_bgs_lock
);
10794 while (!list_empty(&fs_info
->unused_bgs
)) {
10798 block_group
= list_first_entry(&fs_info
->unused_bgs
,
10799 struct btrfs_block_group_cache
,
10801 list_del_init(&block_group
->bg_list
);
10803 space_info
= block_group
->space_info
;
10805 if (ret
|| btrfs_mixed_space_info(space_info
)) {
10806 btrfs_put_block_group(block_group
);
10809 spin_unlock(&fs_info
->unused_bgs_lock
);
10811 mutex_lock(&fs_info
->delete_unused_bgs_mutex
);
10813 /* Don't want to race with allocators so take the groups_sem */
10814 down_write(&space_info
->groups_sem
);
10815 spin_lock(&block_group
->lock
);
10816 if (block_group
->reserved
||
10817 btrfs_block_group_used(&block_group
->item
) ||
10819 list_is_singular(&block_group
->list
)) {
10821 * We want to bail if we made new allocations or have
10822 * outstanding allocations in this block group. We do
10823 * the ro check in case balance is currently acting on
10824 * this block group.
10826 spin_unlock(&block_group
->lock
);
10827 up_write(&space_info
->groups_sem
);
10830 spin_unlock(&block_group
->lock
);
10832 /* We don't want to force the issue, only flip if it's ok. */
10833 ret
= inc_block_group_ro(block_group
, 0);
10834 up_write(&space_info
->groups_sem
);
10841 * Want to do this before we do anything else so we can recover
10842 * properly if we fail to join the transaction.
10844 trans
= btrfs_start_trans_remove_block_group(fs_info
,
10845 block_group
->key
.objectid
);
10846 if (IS_ERR(trans
)) {
10847 btrfs_dec_block_group_ro(root
, block_group
);
10848 ret
= PTR_ERR(trans
);
10853 * We could have pending pinned extents for this block group,
10854 * just delete them, we don't care about them anymore.
10856 start
= block_group
->key
.objectid
;
10857 end
= start
+ block_group
->key
.offset
- 1;
10859 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10860 * btrfs_finish_extent_commit(). If we are at transaction N,
10861 * another task might be running finish_extent_commit() for the
10862 * previous transaction N - 1, and have seen a range belonging
10863 * to the block group in freed_extents[] before we were able to
10864 * clear the whole block group range from freed_extents[]. This
10865 * means that task can lookup for the block group after we
10866 * unpinned it from freed_extents[] and removed it, leading to
10867 * a BUG_ON() at btrfs_unpin_extent_range().
10869 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
10870 ret
= clear_extent_bits(&fs_info
->freed_extents
[0], start
, end
,
10873 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10874 btrfs_dec_block_group_ro(root
, block_group
);
10877 ret
= clear_extent_bits(&fs_info
->freed_extents
[1], start
, end
,
10880 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10881 btrfs_dec_block_group_ro(root
, block_group
);
10884 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10886 /* Reset pinned so btrfs_put_block_group doesn't complain */
10887 spin_lock(&space_info
->lock
);
10888 spin_lock(&block_group
->lock
);
10890 space_info
->bytes_pinned
-= block_group
->pinned
;
10891 space_info
->bytes_readonly
+= block_group
->pinned
;
10892 percpu_counter_add(&space_info
->total_bytes_pinned
,
10893 -block_group
->pinned
);
10894 block_group
->pinned
= 0;
10896 spin_unlock(&block_group
->lock
);
10897 spin_unlock(&space_info
->lock
);
10899 /* DISCARD can flip during remount */
10900 trimming
= btrfs_test_opt(root
->fs_info
, DISCARD
);
10902 /* Implicit trim during transaction commit. */
10904 btrfs_get_block_group_trimming(block_group
);
10907 * Btrfs_remove_chunk will abort the transaction if things go
10910 ret
= btrfs_remove_chunk(trans
, root
,
10911 block_group
->key
.objectid
);
10915 btrfs_put_block_group_trimming(block_group
);
10920 * If we're not mounted with -odiscard, we can just forget
10921 * about this block group. Otherwise we'll need to wait
10922 * until transaction commit to do the actual discard.
10925 spin_lock(&fs_info
->unused_bgs_lock
);
10927 * A concurrent scrub might have added us to the list
10928 * fs_info->unused_bgs, so use a list_move operation
10929 * to add the block group to the deleted_bgs list.
10931 list_move(&block_group
->bg_list
,
10932 &trans
->transaction
->deleted_bgs
);
10933 spin_unlock(&fs_info
->unused_bgs_lock
);
10934 btrfs_get_block_group(block_group
);
10937 btrfs_end_transaction(trans
, root
);
10939 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
10940 btrfs_put_block_group(block_group
);
10941 spin_lock(&fs_info
->unused_bgs_lock
);
10943 spin_unlock(&fs_info
->unused_bgs_lock
);
10946 int btrfs_init_space_info(struct btrfs_fs_info
*fs_info
)
10948 struct btrfs_space_info
*space_info
;
10949 struct btrfs_super_block
*disk_super
;
10955 disk_super
= fs_info
->super_copy
;
10956 if (!btrfs_super_root(disk_super
))
10959 features
= btrfs_super_incompat_flags(disk_super
);
10960 if (features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
10963 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
10964 ret
= update_space_info(fs_info
, flags
, 0, 0, 0, &space_info
);
10969 flags
= BTRFS_BLOCK_GROUP_METADATA
| BTRFS_BLOCK_GROUP_DATA
;
10970 ret
= update_space_info(fs_info
, flags
, 0, 0, 0, &space_info
);
10972 flags
= BTRFS_BLOCK_GROUP_METADATA
;
10973 ret
= update_space_info(fs_info
, flags
, 0, 0, 0, &space_info
);
10977 flags
= BTRFS_BLOCK_GROUP_DATA
;
10978 ret
= update_space_info(fs_info
, flags
, 0, 0, 0, &space_info
);
10984 int btrfs_error_unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
10986 return unpin_extent_range(root
, start
, end
, false);
10990 * It used to be that old block groups would be left around forever.
10991 * Iterating over them would be enough to trim unused space. Since we
10992 * now automatically remove them, we also need to iterate over unallocated
10995 * We don't want a transaction for this since the discard may take a
10996 * substantial amount of time. We don't require that a transaction be
10997 * running, but we do need to take a running transaction into account
10998 * to ensure that we're not discarding chunks that were released in
10999 * the current transaction.
11001 * Holding the chunks lock will prevent other threads from allocating
11002 * or releasing chunks, but it won't prevent a running transaction
11003 * from committing and releasing the memory that the pending chunks
11004 * list head uses. For that, we need to take a reference to the
11007 static int btrfs_trim_free_extents(struct btrfs_device
*device
,
11008 u64 minlen
, u64
*trimmed
)
11010 u64 start
= 0, len
= 0;
11015 /* Not writeable = nothing to do. */
11016 if (!device
->writeable
)
11019 /* No free space = nothing to do. */
11020 if (device
->total_bytes
<= device
->bytes_used
)
11026 struct btrfs_fs_info
*fs_info
= device
->dev_root
->fs_info
;
11027 struct btrfs_transaction
*trans
;
11030 ret
= mutex_lock_interruptible(&fs_info
->chunk_mutex
);
11034 down_read(&fs_info
->commit_root_sem
);
11036 spin_lock(&fs_info
->trans_lock
);
11037 trans
= fs_info
->running_transaction
;
11039 atomic_inc(&trans
->use_count
);
11040 spin_unlock(&fs_info
->trans_lock
);
11042 ret
= find_free_dev_extent_start(trans
, device
, minlen
, start
,
11045 btrfs_put_transaction(trans
);
11048 up_read(&fs_info
->commit_root_sem
);
11049 mutex_unlock(&fs_info
->chunk_mutex
);
11050 if (ret
== -ENOSPC
)
11055 ret
= btrfs_issue_discard(device
->bdev
, start
, len
, &bytes
);
11056 up_read(&fs_info
->commit_root_sem
);
11057 mutex_unlock(&fs_info
->chunk_mutex
);
11065 if (fatal_signal_pending(current
)) {
11066 ret
= -ERESTARTSYS
;
11076 int btrfs_trim_fs(struct btrfs_root
*root
, struct fstrim_range
*range
)
11078 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
11079 struct btrfs_block_group_cache
*cache
= NULL
;
11080 struct btrfs_device
*device
;
11081 struct list_head
*devices
;
11086 u64 total_bytes
= btrfs_super_total_bytes(fs_info
->super_copy
);
11090 * try to trim all FS space, our block group may start from non-zero.
11092 if (range
->len
== total_bytes
)
11093 cache
= btrfs_lookup_first_block_group(fs_info
, range
->start
);
11095 cache
= btrfs_lookup_block_group(fs_info
, range
->start
);
11098 if (cache
->key
.objectid
>= (range
->start
+ range
->len
)) {
11099 btrfs_put_block_group(cache
);
11103 start
= max(range
->start
, cache
->key
.objectid
);
11104 end
= min(range
->start
+ range
->len
,
11105 cache
->key
.objectid
+ cache
->key
.offset
);
11107 if (end
- start
>= range
->minlen
) {
11108 if (!block_group_cache_done(cache
)) {
11109 ret
= cache_block_group(cache
, 0);
11111 btrfs_put_block_group(cache
);
11114 ret
= wait_block_group_cache_done(cache
);
11116 btrfs_put_block_group(cache
);
11120 ret
= btrfs_trim_block_group(cache
,
11126 trimmed
+= group_trimmed
;
11128 btrfs_put_block_group(cache
);
11133 cache
= next_block_group(fs_info
->tree_root
, cache
);
11136 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
11137 devices
= &root
->fs_info
->fs_devices
->alloc_list
;
11138 list_for_each_entry(device
, devices
, dev_alloc_list
) {
11139 ret
= btrfs_trim_free_extents(device
, range
->minlen
,
11144 trimmed
+= group_trimmed
;
11146 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
11148 range
->len
= trimmed
;
11153 * btrfs_{start,end}_write_no_snapshoting() are similar to
11154 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
11155 * data into the page cache through nocow before the subvolume is snapshoted,
11156 * but flush the data into disk after the snapshot creation, or to prevent
11157 * operations while snapshoting is ongoing and that cause the snapshot to be
11158 * inconsistent (writes followed by expanding truncates for example).
11160 void btrfs_end_write_no_snapshoting(struct btrfs_root
*root
)
11162 percpu_counter_dec(&root
->subv_writers
->counter
);
11164 * Make sure counter is updated before we wake up waiters.
11167 if (waitqueue_active(&root
->subv_writers
->wait
))
11168 wake_up(&root
->subv_writers
->wait
);
11171 int btrfs_start_write_no_snapshoting(struct btrfs_root
*root
)
11173 if (atomic_read(&root
->will_be_snapshoted
))
11176 percpu_counter_inc(&root
->subv_writers
->counter
);
11178 * Make sure counter is updated before we check for snapshot creation.
11181 if (atomic_read(&root
->will_be_snapshoted
)) {
11182 btrfs_end_write_no_snapshoting(root
);
11188 static int wait_snapshoting_atomic_t(atomic_t
*a
)
11194 void btrfs_wait_for_snapshot_creation(struct btrfs_root
*root
)
11199 ret
= btrfs_start_write_no_snapshoting(root
);
11202 wait_on_atomic_t(&root
->will_be_snapshoted
,
11203 wait_snapshoting_atomic_t
,
11204 TASK_UNINTERRUPTIBLE
);