2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include <linux/slab.h>
26 #include <linux/ratelimit.h>
27 #include <linux/percpu_counter.h>
31 #include "print-tree.h"
35 #include "free-space-cache.h"
40 #undef SCRAMBLE_DELAYED_REFS
43 * control flags for do_chunk_alloc's force field
44 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
45 * if we really need one.
47 * CHUNK_ALLOC_LIMITED means to only try and allocate one
48 * if we have very few chunks already allocated. This is
49 * used as part of the clustering code to help make sure
50 * we have a good pool of storage to cluster in, without
51 * filling the FS with empty chunks
53 * CHUNK_ALLOC_FORCE means it must try to allocate one
57 CHUNK_ALLOC_NO_FORCE
= 0,
58 CHUNK_ALLOC_LIMITED
= 1,
59 CHUNK_ALLOC_FORCE
= 2,
63 * Control how reservations are dealt with.
65 * RESERVE_FREE - freeing a reservation.
66 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
68 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
69 * bytes_may_use as the ENOSPC accounting is done elsewhere
74 RESERVE_ALLOC_NO_ACCOUNT
= 2,
77 static int update_block_group(struct btrfs_trans_handle
*trans
,
78 struct btrfs_root
*root
, u64 bytenr
,
79 u64 num_bytes
, int alloc
);
80 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
81 struct btrfs_root
*root
,
82 struct btrfs_delayed_ref_node
*node
, u64 parent
,
83 u64 root_objectid
, u64 owner_objectid
,
84 u64 owner_offset
, int refs_to_drop
,
85 struct btrfs_delayed_extent_op
*extra_op
);
86 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
87 struct extent_buffer
*leaf
,
88 struct btrfs_extent_item
*ei
);
89 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
90 struct btrfs_root
*root
,
91 u64 parent
, u64 root_objectid
,
92 u64 flags
, u64 owner
, u64 offset
,
93 struct btrfs_key
*ins
, int ref_mod
);
94 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
95 struct btrfs_root
*root
,
96 u64 parent
, u64 root_objectid
,
97 u64 flags
, struct btrfs_disk_key
*key
,
98 int level
, struct btrfs_key
*ins
,
100 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
101 struct btrfs_root
*extent_root
, u64 flags
,
103 static int find_next_key(struct btrfs_path
*path
, int level
,
104 struct btrfs_key
*key
);
105 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
106 int dump_block_groups
);
107 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
108 u64 num_bytes
, int reserve
,
110 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
112 int btrfs_pin_extent(struct btrfs_root
*root
,
113 u64 bytenr
, u64 num_bytes
, int reserved
);
116 block_group_cache_done(struct btrfs_block_group_cache
*cache
)
119 return cache
->cached
== BTRFS_CACHE_FINISHED
||
120 cache
->cached
== BTRFS_CACHE_ERROR
;
123 static int block_group_bits(struct btrfs_block_group_cache
*cache
, u64 bits
)
125 return (cache
->flags
& bits
) == bits
;
128 static void btrfs_get_block_group(struct btrfs_block_group_cache
*cache
)
130 atomic_inc(&cache
->count
);
133 void btrfs_put_block_group(struct btrfs_block_group_cache
*cache
)
135 if (atomic_dec_and_test(&cache
->count
)) {
136 WARN_ON(cache
->pinned
> 0);
137 WARN_ON(cache
->reserved
> 0);
138 kfree(cache
->free_space_ctl
);
144 * this adds the block group to the fs_info rb tree for the block group
147 static int btrfs_add_block_group_cache(struct btrfs_fs_info
*info
,
148 struct btrfs_block_group_cache
*block_group
)
151 struct rb_node
*parent
= NULL
;
152 struct btrfs_block_group_cache
*cache
;
154 spin_lock(&info
->block_group_cache_lock
);
155 p
= &info
->block_group_cache_tree
.rb_node
;
159 cache
= rb_entry(parent
, struct btrfs_block_group_cache
,
161 if (block_group
->key
.objectid
< cache
->key
.objectid
) {
163 } else if (block_group
->key
.objectid
> cache
->key
.objectid
) {
166 spin_unlock(&info
->block_group_cache_lock
);
171 rb_link_node(&block_group
->cache_node
, parent
, p
);
172 rb_insert_color(&block_group
->cache_node
,
173 &info
->block_group_cache_tree
);
175 if (info
->first_logical_byte
> block_group
->key
.objectid
)
176 info
->first_logical_byte
= block_group
->key
.objectid
;
178 spin_unlock(&info
->block_group_cache_lock
);
184 * This will return the block group at or after bytenr if contains is 0, else
185 * it will return the block group that contains the bytenr
187 static struct btrfs_block_group_cache
*
188 block_group_cache_tree_search(struct btrfs_fs_info
*info
, u64 bytenr
,
191 struct btrfs_block_group_cache
*cache
, *ret
= NULL
;
195 spin_lock(&info
->block_group_cache_lock
);
196 n
= info
->block_group_cache_tree
.rb_node
;
199 cache
= rb_entry(n
, struct btrfs_block_group_cache
,
201 end
= cache
->key
.objectid
+ cache
->key
.offset
- 1;
202 start
= cache
->key
.objectid
;
204 if (bytenr
< start
) {
205 if (!contains
&& (!ret
|| start
< ret
->key
.objectid
))
208 } else if (bytenr
> start
) {
209 if (contains
&& bytenr
<= end
) {
220 btrfs_get_block_group(ret
);
221 if (bytenr
== 0 && info
->first_logical_byte
> ret
->key
.objectid
)
222 info
->first_logical_byte
= ret
->key
.objectid
;
224 spin_unlock(&info
->block_group_cache_lock
);
229 static int add_excluded_extent(struct btrfs_root
*root
,
230 u64 start
, u64 num_bytes
)
232 u64 end
= start
+ num_bytes
- 1;
233 set_extent_bits(&root
->fs_info
->freed_extents
[0],
234 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
235 set_extent_bits(&root
->fs_info
->freed_extents
[1],
236 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
240 static void free_excluded_extents(struct btrfs_root
*root
,
241 struct btrfs_block_group_cache
*cache
)
245 start
= cache
->key
.objectid
;
246 end
= start
+ cache
->key
.offset
- 1;
248 clear_extent_bits(&root
->fs_info
->freed_extents
[0],
249 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
250 clear_extent_bits(&root
->fs_info
->freed_extents
[1],
251 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
254 static int exclude_super_stripes(struct btrfs_root
*root
,
255 struct btrfs_block_group_cache
*cache
)
262 if (cache
->key
.objectid
< BTRFS_SUPER_INFO_OFFSET
) {
263 stripe_len
= BTRFS_SUPER_INFO_OFFSET
- cache
->key
.objectid
;
264 cache
->bytes_super
+= stripe_len
;
265 ret
= add_excluded_extent(root
, cache
->key
.objectid
,
271 for (i
= 0; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
272 bytenr
= btrfs_sb_offset(i
);
273 ret
= btrfs_rmap_block(&root
->fs_info
->mapping_tree
,
274 cache
->key
.objectid
, bytenr
,
275 0, &logical
, &nr
, &stripe_len
);
282 if (logical
[nr
] > cache
->key
.objectid
+
286 if (logical
[nr
] + stripe_len
<= cache
->key
.objectid
)
290 if (start
< cache
->key
.objectid
) {
291 start
= cache
->key
.objectid
;
292 len
= (logical
[nr
] + stripe_len
) - start
;
294 len
= min_t(u64
, stripe_len
,
295 cache
->key
.objectid
+
296 cache
->key
.offset
- start
);
299 cache
->bytes_super
+= len
;
300 ret
= add_excluded_extent(root
, start
, len
);
312 static struct btrfs_caching_control
*
313 get_caching_control(struct btrfs_block_group_cache
*cache
)
315 struct btrfs_caching_control
*ctl
;
317 spin_lock(&cache
->lock
);
318 if (!cache
->caching_ctl
) {
319 spin_unlock(&cache
->lock
);
323 ctl
= cache
->caching_ctl
;
324 atomic_inc(&ctl
->count
);
325 spin_unlock(&cache
->lock
);
329 static void put_caching_control(struct btrfs_caching_control
*ctl
)
331 if (atomic_dec_and_test(&ctl
->count
))
336 * this is only called by cache_block_group, since we could have freed extents
337 * we need to check the pinned_extents for any extents that can't be used yet
338 * since their free space will be released as soon as the transaction commits.
340 static u64
add_new_free_space(struct btrfs_block_group_cache
*block_group
,
341 struct btrfs_fs_info
*info
, u64 start
, u64 end
)
343 u64 extent_start
, extent_end
, size
, total_added
= 0;
346 while (start
< end
) {
347 ret
= find_first_extent_bit(info
->pinned_extents
, start
,
348 &extent_start
, &extent_end
,
349 EXTENT_DIRTY
| EXTENT_UPTODATE
,
354 if (extent_start
<= start
) {
355 start
= extent_end
+ 1;
356 } else if (extent_start
> start
&& extent_start
< end
) {
357 size
= extent_start
- start
;
359 ret
= btrfs_add_free_space(block_group
, start
,
361 BUG_ON(ret
); /* -ENOMEM or logic error */
362 start
= extent_end
+ 1;
371 ret
= btrfs_add_free_space(block_group
, start
, size
);
372 BUG_ON(ret
); /* -ENOMEM or logic error */
378 static noinline
void caching_thread(struct btrfs_work
*work
)
380 struct btrfs_block_group_cache
*block_group
;
381 struct btrfs_fs_info
*fs_info
;
382 struct btrfs_caching_control
*caching_ctl
;
383 struct btrfs_root
*extent_root
;
384 struct btrfs_path
*path
;
385 struct extent_buffer
*leaf
;
386 struct btrfs_key key
;
392 caching_ctl
= container_of(work
, struct btrfs_caching_control
, work
);
393 block_group
= caching_ctl
->block_group
;
394 fs_info
= block_group
->fs_info
;
395 extent_root
= fs_info
->extent_root
;
397 path
= btrfs_alloc_path();
401 last
= max_t(u64
, block_group
->key
.objectid
, BTRFS_SUPER_INFO_OFFSET
);
404 * We don't want to deadlock with somebody trying to allocate a new
405 * extent for the extent root while also trying to search the extent
406 * root to add free space. So we skip locking and search the commit
407 * root, since its read-only
409 path
->skip_locking
= 1;
410 path
->search_commit_root
= 1;
415 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
417 mutex_lock(&caching_ctl
->mutex
);
418 /* need to make sure the commit_root doesn't disappear */
419 down_read(&fs_info
->commit_root_sem
);
422 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
426 leaf
= path
->nodes
[0];
427 nritems
= btrfs_header_nritems(leaf
);
430 if (btrfs_fs_closing(fs_info
) > 1) {
435 if (path
->slots
[0] < nritems
) {
436 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
438 ret
= find_next_key(path
, 0, &key
);
442 if (need_resched() ||
443 rwsem_is_contended(&fs_info
->commit_root_sem
)) {
444 caching_ctl
->progress
= last
;
445 btrfs_release_path(path
);
446 up_read(&fs_info
->commit_root_sem
);
447 mutex_unlock(&caching_ctl
->mutex
);
452 ret
= btrfs_next_leaf(extent_root
, path
);
457 leaf
= path
->nodes
[0];
458 nritems
= btrfs_header_nritems(leaf
);
462 if (key
.objectid
< last
) {
465 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
467 caching_ctl
->progress
= last
;
468 btrfs_release_path(path
);
472 if (key
.objectid
< block_group
->key
.objectid
) {
477 if (key
.objectid
>= block_group
->key
.objectid
+
478 block_group
->key
.offset
)
481 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
||
482 key
.type
== BTRFS_METADATA_ITEM_KEY
) {
483 total_found
+= add_new_free_space(block_group
,
486 if (key
.type
== BTRFS_METADATA_ITEM_KEY
)
487 last
= key
.objectid
+
488 fs_info
->tree_root
->nodesize
;
490 last
= key
.objectid
+ key
.offset
;
492 if (total_found
> (1024 * 1024 * 2)) {
494 wake_up(&caching_ctl
->wait
);
501 total_found
+= add_new_free_space(block_group
, fs_info
, last
,
502 block_group
->key
.objectid
+
503 block_group
->key
.offset
);
504 caching_ctl
->progress
= (u64
)-1;
506 spin_lock(&block_group
->lock
);
507 block_group
->caching_ctl
= NULL
;
508 block_group
->cached
= BTRFS_CACHE_FINISHED
;
509 spin_unlock(&block_group
->lock
);
512 btrfs_free_path(path
);
513 up_read(&fs_info
->commit_root_sem
);
515 free_excluded_extents(extent_root
, block_group
);
517 mutex_unlock(&caching_ctl
->mutex
);
520 spin_lock(&block_group
->lock
);
521 block_group
->caching_ctl
= NULL
;
522 block_group
->cached
= BTRFS_CACHE_ERROR
;
523 spin_unlock(&block_group
->lock
);
525 wake_up(&caching_ctl
->wait
);
527 put_caching_control(caching_ctl
);
528 btrfs_put_block_group(block_group
);
531 static int cache_block_group(struct btrfs_block_group_cache
*cache
,
535 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
536 struct btrfs_caching_control
*caching_ctl
;
539 caching_ctl
= kzalloc(sizeof(*caching_ctl
), GFP_NOFS
);
543 INIT_LIST_HEAD(&caching_ctl
->list
);
544 mutex_init(&caching_ctl
->mutex
);
545 init_waitqueue_head(&caching_ctl
->wait
);
546 caching_ctl
->block_group
= cache
;
547 caching_ctl
->progress
= cache
->key
.objectid
;
548 atomic_set(&caching_ctl
->count
, 1);
549 btrfs_init_work(&caching_ctl
->work
, btrfs_cache_helper
,
550 caching_thread
, NULL
, NULL
);
552 spin_lock(&cache
->lock
);
554 * This should be a rare occasion, but this could happen I think in the
555 * case where one thread starts to load the space cache info, and then
556 * some other thread starts a transaction commit which tries to do an
557 * allocation while the other thread is still loading the space cache
558 * info. The previous loop should have kept us from choosing this block
559 * group, but if we've moved to the state where we will wait on caching
560 * block groups we need to first check if we're doing a fast load here,
561 * so we can wait for it to finish, otherwise we could end up allocating
562 * from a block group who's cache gets evicted for one reason or
565 while (cache
->cached
== BTRFS_CACHE_FAST
) {
566 struct btrfs_caching_control
*ctl
;
568 ctl
= cache
->caching_ctl
;
569 atomic_inc(&ctl
->count
);
570 prepare_to_wait(&ctl
->wait
, &wait
, TASK_UNINTERRUPTIBLE
);
571 spin_unlock(&cache
->lock
);
575 finish_wait(&ctl
->wait
, &wait
);
576 put_caching_control(ctl
);
577 spin_lock(&cache
->lock
);
580 if (cache
->cached
!= BTRFS_CACHE_NO
) {
581 spin_unlock(&cache
->lock
);
585 WARN_ON(cache
->caching_ctl
);
586 cache
->caching_ctl
= caching_ctl
;
587 cache
->cached
= BTRFS_CACHE_FAST
;
588 spin_unlock(&cache
->lock
);
590 if (fs_info
->mount_opt
& BTRFS_MOUNT_SPACE_CACHE
) {
591 mutex_lock(&caching_ctl
->mutex
);
592 ret
= load_free_space_cache(fs_info
, cache
);
594 spin_lock(&cache
->lock
);
596 cache
->caching_ctl
= NULL
;
597 cache
->cached
= BTRFS_CACHE_FINISHED
;
598 cache
->last_byte_to_unpin
= (u64
)-1;
599 caching_ctl
->progress
= (u64
)-1;
601 if (load_cache_only
) {
602 cache
->caching_ctl
= NULL
;
603 cache
->cached
= BTRFS_CACHE_NO
;
605 cache
->cached
= BTRFS_CACHE_STARTED
;
606 cache
->has_caching_ctl
= 1;
609 spin_unlock(&cache
->lock
);
610 mutex_unlock(&caching_ctl
->mutex
);
612 wake_up(&caching_ctl
->wait
);
614 put_caching_control(caching_ctl
);
615 free_excluded_extents(fs_info
->extent_root
, cache
);
620 * We are not going to do the fast caching, set cached to the
621 * appropriate value and wakeup any waiters.
623 spin_lock(&cache
->lock
);
624 if (load_cache_only
) {
625 cache
->caching_ctl
= NULL
;
626 cache
->cached
= BTRFS_CACHE_NO
;
628 cache
->cached
= BTRFS_CACHE_STARTED
;
629 cache
->has_caching_ctl
= 1;
631 spin_unlock(&cache
->lock
);
632 wake_up(&caching_ctl
->wait
);
635 if (load_cache_only
) {
636 put_caching_control(caching_ctl
);
640 down_write(&fs_info
->commit_root_sem
);
641 atomic_inc(&caching_ctl
->count
);
642 list_add_tail(&caching_ctl
->list
, &fs_info
->caching_block_groups
);
643 up_write(&fs_info
->commit_root_sem
);
645 btrfs_get_block_group(cache
);
647 btrfs_queue_work(fs_info
->caching_workers
, &caching_ctl
->work
);
653 * return the block group that starts at or after bytenr
655 static struct btrfs_block_group_cache
*
656 btrfs_lookup_first_block_group(struct btrfs_fs_info
*info
, u64 bytenr
)
658 struct btrfs_block_group_cache
*cache
;
660 cache
= block_group_cache_tree_search(info
, bytenr
, 0);
666 * return the block group that contains the given bytenr
668 struct btrfs_block_group_cache
*btrfs_lookup_block_group(
669 struct btrfs_fs_info
*info
,
672 struct btrfs_block_group_cache
*cache
;
674 cache
= block_group_cache_tree_search(info
, bytenr
, 1);
679 static struct btrfs_space_info
*__find_space_info(struct btrfs_fs_info
*info
,
682 struct list_head
*head
= &info
->space_info
;
683 struct btrfs_space_info
*found
;
685 flags
&= BTRFS_BLOCK_GROUP_TYPE_MASK
;
688 list_for_each_entry_rcu(found
, head
, list
) {
689 if (found
->flags
& flags
) {
699 * after adding space to the filesystem, we need to clear the full flags
700 * on all the space infos.
702 void btrfs_clear_space_info_full(struct btrfs_fs_info
*info
)
704 struct list_head
*head
= &info
->space_info
;
705 struct btrfs_space_info
*found
;
708 list_for_each_entry_rcu(found
, head
, list
)
713 /* simple helper to search for an existing data extent at a given offset */
714 int btrfs_lookup_data_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
717 struct btrfs_key key
;
718 struct btrfs_path
*path
;
720 path
= btrfs_alloc_path();
724 key
.objectid
= start
;
726 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
727 ret
= btrfs_search_slot(NULL
, root
->fs_info
->extent_root
, &key
, path
,
729 btrfs_free_path(path
);
734 * helper function to lookup reference count and flags of a tree block.
736 * the head node for delayed ref is used to store the sum of all the
737 * reference count modifications queued up in the rbtree. the head
738 * node may also store the extent flags to set. This way you can check
739 * to see what the reference count and extent flags would be if all of
740 * the delayed refs are not processed.
742 int btrfs_lookup_extent_info(struct btrfs_trans_handle
*trans
,
743 struct btrfs_root
*root
, u64 bytenr
,
744 u64 offset
, int metadata
, u64
*refs
, u64
*flags
)
746 struct btrfs_delayed_ref_head
*head
;
747 struct btrfs_delayed_ref_root
*delayed_refs
;
748 struct btrfs_path
*path
;
749 struct btrfs_extent_item
*ei
;
750 struct extent_buffer
*leaf
;
751 struct btrfs_key key
;
758 * If we don't have skinny metadata, don't bother doing anything
761 if (metadata
&& !btrfs_fs_incompat(root
->fs_info
, SKINNY_METADATA
)) {
762 offset
= root
->nodesize
;
766 path
= btrfs_alloc_path();
771 path
->skip_locking
= 1;
772 path
->search_commit_root
= 1;
776 key
.objectid
= bytenr
;
779 key
.type
= BTRFS_METADATA_ITEM_KEY
;
781 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
783 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
,
788 if (ret
> 0 && metadata
&& key
.type
== BTRFS_METADATA_ITEM_KEY
) {
789 if (path
->slots
[0]) {
791 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
793 if (key
.objectid
== bytenr
&&
794 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
795 key
.offset
== root
->nodesize
)
801 leaf
= path
->nodes
[0];
802 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
803 if (item_size
>= sizeof(*ei
)) {
804 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
805 struct btrfs_extent_item
);
806 num_refs
= btrfs_extent_refs(leaf
, ei
);
807 extent_flags
= btrfs_extent_flags(leaf
, ei
);
809 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
810 struct btrfs_extent_item_v0
*ei0
;
811 BUG_ON(item_size
!= sizeof(*ei0
));
812 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
813 struct btrfs_extent_item_v0
);
814 num_refs
= btrfs_extent_refs_v0(leaf
, ei0
);
815 /* FIXME: this isn't correct for data */
816 extent_flags
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
821 BUG_ON(num_refs
== 0);
831 delayed_refs
= &trans
->transaction
->delayed_refs
;
832 spin_lock(&delayed_refs
->lock
);
833 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
835 if (!mutex_trylock(&head
->mutex
)) {
836 atomic_inc(&head
->node
.refs
);
837 spin_unlock(&delayed_refs
->lock
);
839 btrfs_release_path(path
);
842 * Mutex was contended, block until it's released and try
845 mutex_lock(&head
->mutex
);
846 mutex_unlock(&head
->mutex
);
847 btrfs_put_delayed_ref(&head
->node
);
850 spin_lock(&head
->lock
);
851 if (head
->extent_op
&& head
->extent_op
->update_flags
)
852 extent_flags
|= head
->extent_op
->flags_to_set
;
854 BUG_ON(num_refs
== 0);
856 num_refs
+= head
->node
.ref_mod
;
857 spin_unlock(&head
->lock
);
858 mutex_unlock(&head
->mutex
);
860 spin_unlock(&delayed_refs
->lock
);
862 WARN_ON(num_refs
== 0);
866 *flags
= extent_flags
;
868 btrfs_free_path(path
);
873 * Back reference rules. Back refs have three main goals:
875 * 1) differentiate between all holders of references to an extent so that
876 * when a reference is dropped we can make sure it was a valid reference
877 * before freeing the extent.
879 * 2) Provide enough information to quickly find the holders of an extent
880 * if we notice a given block is corrupted or bad.
882 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
883 * maintenance. This is actually the same as #2, but with a slightly
884 * different use case.
886 * There are two kinds of back refs. The implicit back refs is optimized
887 * for pointers in non-shared tree blocks. For a given pointer in a block,
888 * back refs of this kind provide information about the block's owner tree
889 * and the pointer's key. These information allow us to find the block by
890 * b-tree searching. The full back refs is for pointers in tree blocks not
891 * referenced by their owner trees. The location of tree block is recorded
892 * in the back refs. Actually the full back refs is generic, and can be
893 * used in all cases the implicit back refs is used. The major shortcoming
894 * of the full back refs is its overhead. Every time a tree block gets
895 * COWed, we have to update back refs entry for all pointers in it.
897 * For a newly allocated tree block, we use implicit back refs for
898 * pointers in it. This means most tree related operations only involve
899 * implicit back refs. For a tree block created in old transaction, the
900 * only way to drop a reference to it is COW it. So we can detect the
901 * event that tree block loses its owner tree's reference and do the
902 * back refs conversion.
904 * When a tree block is COW'd through a tree, there are four cases:
906 * The reference count of the block is one and the tree is the block's
907 * owner tree. Nothing to do in this case.
909 * The reference count of the block is one and the tree is not the
910 * block's owner tree. In this case, full back refs is used for pointers
911 * in the block. Remove these full back refs, add implicit back refs for
912 * every pointers in the new block.
914 * The reference count of the block is greater than one and the tree is
915 * the block's owner tree. In this case, implicit back refs is used for
916 * pointers in the block. Add full back refs for every pointers in the
917 * block, increase lower level extents' reference counts. The original
918 * implicit back refs are entailed to the new block.
920 * The reference count of the block is greater than one and the tree is
921 * not the block's owner tree. Add implicit back refs for every pointer in
922 * the new block, increase lower level extents' reference count.
924 * Back Reference Key composing:
926 * The key objectid corresponds to the first byte in the extent,
927 * The key type is used to differentiate between types of back refs.
928 * There are different meanings of the key offset for different types
931 * File extents can be referenced by:
933 * - multiple snapshots, subvolumes, or different generations in one subvol
934 * - different files inside a single subvolume
935 * - different offsets inside a file (bookend extents in file.c)
937 * The extent ref structure for the implicit back refs has fields for:
939 * - Objectid of the subvolume root
940 * - objectid of the file holding the reference
941 * - original offset in the file
942 * - how many bookend extents
944 * The key offset for the implicit back refs is hash of the first
947 * The extent ref structure for the full back refs has field for:
949 * - number of pointers in the tree leaf
951 * The key offset for the implicit back refs is the first byte of
954 * When a file extent is allocated, The implicit back refs is used.
955 * the fields are filled in:
957 * (root_key.objectid, inode objectid, offset in file, 1)
959 * When a file extent is removed file truncation, we find the
960 * corresponding implicit back refs and check the following fields:
962 * (btrfs_header_owner(leaf), inode objectid, offset in file)
964 * Btree extents can be referenced by:
966 * - Different subvolumes
968 * Both the implicit back refs and the full back refs for tree blocks
969 * only consist of key. The key offset for the implicit back refs is
970 * objectid of block's owner tree. The key offset for the full back refs
971 * is the first byte of parent block.
973 * When implicit back refs is used, information about the lowest key and
974 * level of the tree block are required. These information are stored in
975 * tree block info structure.
978 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
979 static int convert_extent_item_v0(struct btrfs_trans_handle
*trans
,
980 struct btrfs_root
*root
,
981 struct btrfs_path
*path
,
982 u64 owner
, u32 extra_size
)
984 struct btrfs_extent_item
*item
;
985 struct btrfs_extent_item_v0
*ei0
;
986 struct btrfs_extent_ref_v0
*ref0
;
987 struct btrfs_tree_block_info
*bi
;
988 struct extent_buffer
*leaf
;
989 struct btrfs_key key
;
990 struct btrfs_key found_key
;
991 u32 new_size
= sizeof(*item
);
995 leaf
= path
->nodes
[0];
996 BUG_ON(btrfs_item_size_nr(leaf
, path
->slots
[0]) != sizeof(*ei0
));
998 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
999 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1000 struct btrfs_extent_item_v0
);
1001 refs
= btrfs_extent_refs_v0(leaf
, ei0
);
1003 if (owner
== (u64
)-1) {
1005 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1006 ret
= btrfs_next_leaf(root
, path
);
1009 BUG_ON(ret
> 0); /* Corruption */
1010 leaf
= path
->nodes
[0];
1012 btrfs_item_key_to_cpu(leaf
, &found_key
,
1014 BUG_ON(key
.objectid
!= found_key
.objectid
);
1015 if (found_key
.type
!= BTRFS_EXTENT_REF_V0_KEY
) {
1019 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1020 struct btrfs_extent_ref_v0
);
1021 owner
= btrfs_ref_objectid_v0(leaf
, ref0
);
1025 btrfs_release_path(path
);
1027 if (owner
< BTRFS_FIRST_FREE_OBJECTID
)
1028 new_size
+= sizeof(*bi
);
1030 new_size
-= sizeof(*ei0
);
1031 ret
= btrfs_search_slot(trans
, root
, &key
, path
,
1032 new_size
+ extra_size
, 1);
1035 BUG_ON(ret
); /* Corruption */
1037 btrfs_extend_item(root
, path
, new_size
);
1039 leaf
= path
->nodes
[0];
1040 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1041 btrfs_set_extent_refs(leaf
, item
, refs
);
1042 /* FIXME: get real generation */
1043 btrfs_set_extent_generation(leaf
, item
, 0);
1044 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1045 btrfs_set_extent_flags(leaf
, item
,
1046 BTRFS_EXTENT_FLAG_TREE_BLOCK
|
1047 BTRFS_BLOCK_FLAG_FULL_BACKREF
);
1048 bi
= (struct btrfs_tree_block_info
*)(item
+ 1);
1049 /* FIXME: get first key of the block */
1050 memset_extent_buffer(leaf
, 0, (unsigned long)bi
, sizeof(*bi
));
1051 btrfs_set_tree_block_level(leaf
, bi
, (int)owner
);
1053 btrfs_set_extent_flags(leaf
, item
, BTRFS_EXTENT_FLAG_DATA
);
1055 btrfs_mark_buffer_dirty(leaf
);
1060 static u64
hash_extent_data_ref(u64 root_objectid
, u64 owner
, u64 offset
)
1062 u32 high_crc
= ~(u32
)0;
1063 u32 low_crc
= ~(u32
)0;
1066 lenum
= cpu_to_le64(root_objectid
);
1067 high_crc
= btrfs_crc32c(high_crc
, &lenum
, sizeof(lenum
));
1068 lenum
= cpu_to_le64(owner
);
1069 low_crc
= btrfs_crc32c(low_crc
, &lenum
, sizeof(lenum
));
1070 lenum
= cpu_to_le64(offset
);
1071 low_crc
= btrfs_crc32c(low_crc
, &lenum
, sizeof(lenum
));
1073 return ((u64
)high_crc
<< 31) ^ (u64
)low_crc
;
1076 static u64
hash_extent_data_ref_item(struct extent_buffer
*leaf
,
1077 struct btrfs_extent_data_ref
*ref
)
1079 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf
, ref
),
1080 btrfs_extent_data_ref_objectid(leaf
, ref
),
1081 btrfs_extent_data_ref_offset(leaf
, ref
));
1084 static int match_extent_data_ref(struct extent_buffer
*leaf
,
1085 struct btrfs_extent_data_ref
*ref
,
1086 u64 root_objectid
, u64 owner
, u64 offset
)
1088 if (btrfs_extent_data_ref_root(leaf
, ref
) != root_objectid
||
1089 btrfs_extent_data_ref_objectid(leaf
, ref
) != owner
||
1090 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
1095 static noinline
int lookup_extent_data_ref(struct btrfs_trans_handle
*trans
,
1096 struct btrfs_root
*root
,
1097 struct btrfs_path
*path
,
1098 u64 bytenr
, u64 parent
,
1100 u64 owner
, u64 offset
)
1102 struct btrfs_key key
;
1103 struct btrfs_extent_data_ref
*ref
;
1104 struct extent_buffer
*leaf
;
1110 key
.objectid
= bytenr
;
1112 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1113 key
.offset
= parent
;
1115 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1116 key
.offset
= hash_extent_data_ref(root_objectid
,
1121 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1130 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1131 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1132 btrfs_release_path(path
);
1133 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1144 leaf
= path
->nodes
[0];
1145 nritems
= btrfs_header_nritems(leaf
);
1147 if (path
->slots
[0] >= nritems
) {
1148 ret
= btrfs_next_leaf(root
, path
);
1154 leaf
= path
->nodes
[0];
1155 nritems
= btrfs_header_nritems(leaf
);
1159 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1160 if (key
.objectid
!= bytenr
||
1161 key
.type
!= BTRFS_EXTENT_DATA_REF_KEY
)
1164 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1165 struct btrfs_extent_data_ref
);
1167 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1170 btrfs_release_path(path
);
1182 static noinline
int insert_extent_data_ref(struct btrfs_trans_handle
*trans
,
1183 struct btrfs_root
*root
,
1184 struct btrfs_path
*path
,
1185 u64 bytenr
, u64 parent
,
1186 u64 root_objectid
, u64 owner
,
1187 u64 offset
, int refs_to_add
)
1189 struct btrfs_key key
;
1190 struct extent_buffer
*leaf
;
1195 key
.objectid
= bytenr
;
1197 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1198 key
.offset
= parent
;
1199 size
= sizeof(struct btrfs_shared_data_ref
);
1201 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1202 key
.offset
= hash_extent_data_ref(root_objectid
,
1204 size
= sizeof(struct btrfs_extent_data_ref
);
1207 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, size
);
1208 if (ret
&& ret
!= -EEXIST
)
1211 leaf
= path
->nodes
[0];
1213 struct btrfs_shared_data_ref
*ref
;
1214 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1215 struct btrfs_shared_data_ref
);
1217 btrfs_set_shared_data_ref_count(leaf
, ref
, refs_to_add
);
1219 num_refs
= btrfs_shared_data_ref_count(leaf
, ref
);
1220 num_refs
+= refs_to_add
;
1221 btrfs_set_shared_data_ref_count(leaf
, ref
, num_refs
);
1224 struct btrfs_extent_data_ref
*ref
;
1225 while (ret
== -EEXIST
) {
1226 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1227 struct btrfs_extent_data_ref
);
1228 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1231 btrfs_release_path(path
);
1233 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1235 if (ret
&& ret
!= -EEXIST
)
1238 leaf
= path
->nodes
[0];
1240 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1241 struct btrfs_extent_data_ref
);
1243 btrfs_set_extent_data_ref_root(leaf
, ref
,
1245 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
1246 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
1247 btrfs_set_extent_data_ref_count(leaf
, ref
, refs_to_add
);
1249 num_refs
= btrfs_extent_data_ref_count(leaf
, ref
);
1250 num_refs
+= refs_to_add
;
1251 btrfs_set_extent_data_ref_count(leaf
, ref
, num_refs
);
1254 btrfs_mark_buffer_dirty(leaf
);
1257 btrfs_release_path(path
);
1261 static noinline
int remove_extent_data_ref(struct btrfs_trans_handle
*trans
,
1262 struct btrfs_root
*root
,
1263 struct btrfs_path
*path
,
1264 int refs_to_drop
, int *last_ref
)
1266 struct btrfs_key key
;
1267 struct btrfs_extent_data_ref
*ref1
= NULL
;
1268 struct btrfs_shared_data_ref
*ref2
= NULL
;
1269 struct extent_buffer
*leaf
;
1273 leaf
= path
->nodes
[0];
1274 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1276 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1277 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1278 struct btrfs_extent_data_ref
);
1279 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1280 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1281 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1282 struct btrfs_shared_data_ref
);
1283 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1284 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1285 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1286 struct btrfs_extent_ref_v0
*ref0
;
1287 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1288 struct btrfs_extent_ref_v0
);
1289 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1295 BUG_ON(num_refs
< refs_to_drop
);
1296 num_refs
-= refs_to_drop
;
1298 if (num_refs
== 0) {
1299 ret
= btrfs_del_item(trans
, root
, path
);
1302 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
)
1303 btrfs_set_extent_data_ref_count(leaf
, ref1
, num_refs
);
1304 else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
)
1305 btrfs_set_shared_data_ref_count(leaf
, ref2
, num_refs
);
1306 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1308 struct btrfs_extent_ref_v0
*ref0
;
1309 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1310 struct btrfs_extent_ref_v0
);
1311 btrfs_set_ref_count_v0(leaf
, ref0
, num_refs
);
1314 btrfs_mark_buffer_dirty(leaf
);
1319 static noinline u32
extent_data_ref_count(struct btrfs_root
*root
,
1320 struct btrfs_path
*path
,
1321 struct btrfs_extent_inline_ref
*iref
)
1323 struct btrfs_key key
;
1324 struct extent_buffer
*leaf
;
1325 struct btrfs_extent_data_ref
*ref1
;
1326 struct btrfs_shared_data_ref
*ref2
;
1329 leaf
= path
->nodes
[0];
1330 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1332 if (btrfs_extent_inline_ref_type(leaf
, iref
) ==
1333 BTRFS_EXTENT_DATA_REF_KEY
) {
1334 ref1
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1335 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1337 ref2
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1338 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1340 } else if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1341 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1342 struct btrfs_extent_data_ref
);
1343 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1344 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1345 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1346 struct btrfs_shared_data_ref
);
1347 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1348 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1349 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1350 struct btrfs_extent_ref_v0
*ref0
;
1351 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1352 struct btrfs_extent_ref_v0
);
1353 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1361 static noinline
int lookup_tree_block_ref(struct btrfs_trans_handle
*trans
,
1362 struct btrfs_root
*root
,
1363 struct btrfs_path
*path
,
1364 u64 bytenr
, u64 parent
,
1367 struct btrfs_key key
;
1370 key
.objectid
= bytenr
;
1372 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1373 key
.offset
= parent
;
1375 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1376 key
.offset
= root_objectid
;
1379 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1382 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1383 if (ret
== -ENOENT
&& parent
) {
1384 btrfs_release_path(path
);
1385 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1386 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1394 static noinline
int insert_tree_block_ref(struct btrfs_trans_handle
*trans
,
1395 struct btrfs_root
*root
,
1396 struct btrfs_path
*path
,
1397 u64 bytenr
, u64 parent
,
1400 struct btrfs_key key
;
1403 key
.objectid
= bytenr
;
1405 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1406 key
.offset
= parent
;
1408 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1409 key
.offset
= root_objectid
;
1412 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
1413 btrfs_release_path(path
);
1417 static inline int extent_ref_type(u64 parent
, u64 owner
)
1420 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1422 type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1424 type
= BTRFS_TREE_BLOCK_REF_KEY
;
1427 type
= BTRFS_SHARED_DATA_REF_KEY
;
1429 type
= BTRFS_EXTENT_DATA_REF_KEY
;
1434 static int find_next_key(struct btrfs_path
*path
, int level
,
1435 struct btrfs_key
*key
)
1438 for (; level
< BTRFS_MAX_LEVEL
; level
++) {
1439 if (!path
->nodes
[level
])
1441 if (path
->slots
[level
] + 1 >=
1442 btrfs_header_nritems(path
->nodes
[level
]))
1445 btrfs_item_key_to_cpu(path
->nodes
[level
], key
,
1446 path
->slots
[level
] + 1);
1448 btrfs_node_key_to_cpu(path
->nodes
[level
], key
,
1449 path
->slots
[level
] + 1);
1456 * look for inline back ref. if back ref is found, *ref_ret is set
1457 * to the address of inline back ref, and 0 is returned.
1459 * if back ref isn't found, *ref_ret is set to the address where it
1460 * should be inserted, and -ENOENT is returned.
1462 * if insert is true and there are too many inline back refs, the path
1463 * points to the extent item, and -EAGAIN is returned.
1465 * NOTE: inline back refs are ordered in the same way that back ref
1466 * items in the tree are ordered.
1468 static noinline_for_stack
1469 int lookup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1470 struct btrfs_root
*root
,
1471 struct btrfs_path
*path
,
1472 struct btrfs_extent_inline_ref
**ref_ret
,
1473 u64 bytenr
, u64 num_bytes
,
1474 u64 parent
, u64 root_objectid
,
1475 u64 owner
, u64 offset
, int insert
)
1477 struct btrfs_key key
;
1478 struct extent_buffer
*leaf
;
1479 struct btrfs_extent_item
*ei
;
1480 struct btrfs_extent_inline_ref
*iref
;
1490 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
1493 key
.objectid
= bytenr
;
1494 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1495 key
.offset
= num_bytes
;
1497 want
= extent_ref_type(parent
, owner
);
1499 extra_size
= btrfs_extent_inline_ref_size(want
);
1500 path
->keep_locks
= 1;
1505 * Owner is our parent level, so we can just add one to get the level
1506 * for the block we are interested in.
1508 if (skinny_metadata
&& owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1509 key
.type
= BTRFS_METADATA_ITEM_KEY
;
1514 ret
= btrfs_search_slot(trans
, root
, &key
, path
, extra_size
, 1);
1521 * We may be a newly converted file system which still has the old fat
1522 * extent entries for metadata, so try and see if we have one of those.
1524 if (ret
> 0 && skinny_metadata
) {
1525 skinny_metadata
= false;
1526 if (path
->slots
[0]) {
1528 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
1530 if (key
.objectid
== bytenr
&&
1531 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
1532 key
.offset
== num_bytes
)
1536 key
.objectid
= bytenr
;
1537 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1538 key
.offset
= num_bytes
;
1539 btrfs_release_path(path
);
1544 if (ret
&& !insert
) {
1547 } else if (WARN_ON(ret
)) {
1552 leaf
= path
->nodes
[0];
1553 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1554 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1555 if (item_size
< sizeof(*ei
)) {
1560 ret
= convert_extent_item_v0(trans
, root
, path
, owner
,
1566 leaf
= path
->nodes
[0];
1567 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1570 BUG_ON(item_size
< sizeof(*ei
));
1572 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1573 flags
= btrfs_extent_flags(leaf
, ei
);
1575 ptr
= (unsigned long)(ei
+ 1);
1576 end
= (unsigned long)ei
+ item_size
;
1578 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
&& !skinny_metadata
) {
1579 ptr
+= sizeof(struct btrfs_tree_block_info
);
1589 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1590 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1594 ptr
+= btrfs_extent_inline_ref_size(type
);
1598 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1599 struct btrfs_extent_data_ref
*dref
;
1600 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1601 if (match_extent_data_ref(leaf
, dref
, root_objectid
,
1606 if (hash_extent_data_ref_item(leaf
, dref
) <
1607 hash_extent_data_ref(root_objectid
, owner
, offset
))
1611 ref_offset
= btrfs_extent_inline_ref_offset(leaf
, iref
);
1613 if (parent
== ref_offset
) {
1617 if (ref_offset
< parent
)
1620 if (root_objectid
== ref_offset
) {
1624 if (ref_offset
< root_objectid
)
1628 ptr
+= btrfs_extent_inline_ref_size(type
);
1630 if (err
== -ENOENT
&& insert
) {
1631 if (item_size
+ extra_size
>=
1632 BTRFS_MAX_EXTENT_ITEM_SIZE(root
)) {
1637 * To add new inline back ref, we have to make sure
1638 * there is no corresponding back ref item.
1639 * For simplicity, we just do not add new inline back
1640 * ref if there is any kind of item for this block
1642 if (find_next_key(path
, 0, &key
) == 0 &&
1643 key
.objectid
== bytenr
&&
1644 key
.type
< BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1649 *ref_ret
= (struct btrfs_extent_inline_ref
*)ptr
;
1652 path
->keep_locks
= 0;
1653 btrfs_unlock_up_safe(path
, 1);
1659 * helper to add new inline back ref
1661 static noinline_for_stack
1662 void setup_inline_extent_backref(struct btrfs_root
*root
,
1663 struct btrfs_path
*path
,
1664 struct btrfs_extent_inline_ref
*iref
,
1665 u64 parent
, u64 root_objectid
,
1666 u64 owner
, u64 offset
, int refs_to_add
,
1667 struct btrfs_delayed_extent_op
*extent_op
)
1669 struct extent_buffer
*leaf
;
1670 struct btrfs_extent_item
*ei
;
1673 unsigned long item_offset
;
1678 leaf
= path
->nodes
[0];
1679 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1680 item_offset
= (unsigned long)iref
- (unsigned long)ei
;
1682 type
= extent_ref_type(parent
, owner
);
1683 size
= btrfs_extent_inline_ref_size(type
);
1685 btrfs_extend_item(root
, path
, size
);
1687 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1688 refs
= btrfs_extent_refs(leaf
, ei
);
1689 refs
+= refs_to_add
;
1690 btrfs_set_extent_refs(leaf
, ei
, refs
);
1692 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1694 ptr
= (unsigned long)ei
+ item_offset
;
1695 end
= (unsigned long)ei
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
1696 if (ptr
< end
- size
)
1697 memmove_extent_buffer(leaf
, ptr
+ size
, ptr
,
1700 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1701 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
1702 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1703 struct btrfs_extent_data_ref
*dref
;
1704 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1705 btrfs_set_extent_data_ref_root(leaf
, dref
, root_objectid
);
1706 btrfs_set_extent_data_ref_objectid(leaf
, dref
, owner
);
1707 btrfs_set_extent_data_ref_offset(leaf
, dref
, offset
);
1708 btrfs_set_extent_data_ref_count(leaf
, dref
, refs_to_add
);
1709 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1710 struct btrfs_shared_data_ref
*sref
;
1711 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1712 btrfs_set_shared_data_ref_count(leaf
, sref
, refs_to_add
);
1713 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1714 } else if (type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
1715 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1717 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
1719 btrfs_mark_buffer_dirty(leaf
);
1722 static int lookup_extent_backref(struct btrfs_trans_handle
*trans
,
1723 struct btrfs_root
*root
,
1724 struct btrfs_path
*path
,
1725 struct btrfs_extent_inline_ref
**ref_ret
,
1726 u64 bytenr
, u64 num_bytes
, u64 parent
,
1727 u64 root_objectid
, u64 owner
, u64 offset
)
1731 ret
= lookup_inline_extent_backref(trans
, root
, path
, ref_ret
,
1732 bytenr
, num_bytes
, parent
,
1733 root_objectid
, owner
, offset
, 0);
1737 btrfs_release_path(path
);
1740 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1741 ret
= lookup_tree_block_ref(trans
, root
, path
, bytenr
, parent
,
1744 ret
= lookup_extent_data_ref(trans
, root
, path
, bytenr
, parent
,
1745 root_objectid
, owner
, offset
);
1751 * helper to update/remove inline back ref
1753 static noinline_for_stack
1754 void update_inline_extent_backref(struct btrfs_root
*root
,
1755 struct btrfs_path
*path
,
1756 struct btrfs_extent_inline_ref
*iref
,
1758 struct btrfs_delayed_extent_op
*extent_op
,
1761 struct extent_buffer
*leaf
;
1762 struct btrfs_extent_item
*ei
;
1763 struct btrfs_extent_data_ref
*dref
= NULL
;
1764 struct btrfs_shared_data_ref
*sref
= NULL
;
1772 leaf
= path
->nodes
[0];
1773 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1774 refs
= btrfs_extent_refs(leaf
, ei
);
1775 WARN_ON(refs_to_mod
< 0 && refs
+ refs_to_mod
<= 0);
1776 refs
+= refs_to_mod
;
1777 btrfs_set_extent_refs(leaf
, ei
, refs
);
1779 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1781 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1783 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1784 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1785 refs
= btrfs_extent_data_ref_count(leaf
, dref
);
1786 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1787 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1788 refs
= btrfs_shared_data_ref_count(leaf
, sref
);
1791 BUG_ON(refs_to_mod
!= -1);
1794 BUG_ON(refs_to_mod
< 0 && refs
< -refs_to_mod
);
1795 refs
+= refs_to_mod
;
1798 if (type
== BTRFS_EXTENT_DATA_REF_KEY
)
1799 btrfs_set_extent_data_ref_count(leaf
, dref
, refs
);
1801 btrfs_set_shared_data_ref_count(leaf
, sref
, refs
);
1804 size
= btrfs_extent_inline_ref_size(type
);
1805 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1806 ptr
= (unsigned long)iref
;
1807 end
= (unsigned long)ei
+ item_size
;
1808 if (ptr
+ size
< end
)
1809 memmove_extent_buffer(leaf
, ptr
, ptr
+ size
,
1812 btrfs_truncate_item(root
, path
, item_size
, 1);
1814 btrfs_mark_buffer_dirty(leaf
);
1817 static noinline_for_stack
1818 int insert_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1819 struct btrfs_root
*root
,
1820 struct btrfs_path
*path
,
1821 u64 bytenr
, u64 num_bytes
, u64 parent
,
1822 u64 root_objectid
, u64 owner
,
1823 u64 offset
, int refs_to_add
,
1824 struct btrfs_delayed_extent_op
*extent_op
)
1826 struct btrfs_extent_inline_ref
*iref
;
1829 ret
= lookup_inline_extent_backref(trans
, root
, path
, &iref
,
1830 bytenr
, num_bytes
, parent
,
1831 root_objectid
, owner
, offset
, 1);
1833 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
);
1834 update_inline_extent_backref(root
, path
, iref
,
1835 refs_to_add
, extent_op
, NULL
);
1836 } else if (ret
== -ENOENT
) {
1837 setup_inline_extent_backref(root
, path
, iref
, parent
,
1838 root_objectid
, owner
, offset
,
1839 refs_to_add
, extent_op
);
1845 static int insert_extent_backref(struct btrfs_trans_handle
*trans
,
1846 struct btrfs_root
*root
,
1847 struct btrfs_path
*path
,
1848 u64 bytenr
, u64 parent
, u64 root_objectid
,
1849 u64 owner
, u64 offset
, int refs_to_add
)
1852 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1853 BUG_ON(refs_to_add
!= 1);
1854 ret
= insert_tree_block_ref(trans
, root
, path
, bytenr
,
1855 parent
, root_objectid
);
1857 ret
= insert_extent_data_ref(trans
, root
, path
, bytenr
,
1858 parent
, root_objectid
,
1859 owner
, offset
, refs_to_add
);
1864 static int remove_extent_backref(struct btrfs_trans_handle
*trans
,
1865 struct btrfs_root
*root
,
1866 struct btrfs_path
*path
,
1867 struct btrfs_extent_inline_ref
*iref
,
1868 int refs_to_drop
, int is_data
, int *last_ref
)
1872 BUG_ON(!is_data
&& refs_to_drop
!= 1);
1874 update_inline_extent_backref(root
, path
, iref
,
1875 -refs_to_drop
, NULL
, last_ref
);
1876 } else if (is_data
) {
1877 ret
= remove_extent_data_ref(trans
, root
, path
, refs_to_drop
,
1881 ret
= btrfs_del_item(trans
, root
, path
);
1886 static int btrfs_issue_discard(struct block_device
*bdev
,
1889 return blkdev_issue_discard(bdev
, start
>> 9, len
>> 9, GFP_NOFS
, 0);
1892 int btrfs_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
1893 u64 num_bytes
, u64
*actual_bytes
)
1896 u64 discarded_bytes
= 0;
1897 struct btrfs_bio
*bbio
= NULL
;
1900 /* Tell the block device(s) that the sectors can be discarded */
1901 ret
= btrfs_map_block(root
->fs_info
, REQ_DISCARD
,
1902 bytenr
, &num_bytes
, &bbio
, 0);
1903 /* Error condition is -ENOMEM */
1905 struct btrfs_bio_stripe
*stripe
= bbio
->stripes
;
1909 for (i
= 0; i
< bbio
->num_stripes
; i
++, stripe
++) {
1910 if (!stripe
->dev
->can_discard
)
1913 ret
= btrfs_issue_discard(stripe
->dev
->bdev
,
1917 discarded_bytes
+= stripe
->length
;
1918 else if (ret
!= -EOPNOTSUPP
)
1919 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1922 * Just in case we get back EOPNOTSUPP for some reason,
1923 * just ignore the return value so we don't screw up
1924 * people calling discard_extent.
1928 btrfs_put_bbio(bbio
);
1932 *actual_bytes
= discarded_bytes
;
1935 if (ret
== -EOPNOTSUPP
)
1940 /* Can return -ENOMEM */
1941 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1942 struct btrfs_root
*root
,
1943 u64 bytenr
, u64 num_bytes
, u64 parent
,
1944 u64 root_objectid
, u64 owner
, u64 offset
,
1948 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1950 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
&&
1951 root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
1953 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1954 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
1956 parent
, root_objectid
, (int)owner
,
1957 BTRFS_ADD_DELAYED_REF
, NULL
, no_quota
);
1959 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
1961 parent
, root_objectid
, owner
, offset
,
1962 BTRFS_ADD_DELAYED_REF
, NULL
, no_quota
);
1967 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1968 struct btrfs_root
*root
,
1969 struct btrfs_delayed_ref_node
*node
,
1970 u64 parent
, u64 root_objectid
,
1971 u64 owner
, u64 offset
, int refs_to_add
,
1972 struct btrfs_delayed_extent_op
*extent_op
)
1974 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1975 struct btrfs_path
*path
;
1976 struct extent_buffer
*leaf
;
1977 struct btrfs_extent_item
*item
;
1978 struct btrfs_key key
;
1979 u64 bytenr
= node
->bytenr
;
1980 u64 num_bytes
= node
->num_bytes
;
1983 int no_quota
= node
->no_quota
;
1985 path
= btrfs_alloc_path();
1989 if (!is_fstree(root_objectid
) || !root
->fs_info
->quota_enabled
)
1993 path
->leave_spinning
= 1;
1994 /* this will setup the path even if it fails to insert the back ref */
1995 ret
= insert_inline_extent_backref(trans
, fs_info
->extent_root
, path
,
1996 bytenr
, num_bytes
, parent
,
1997 root_objectid
, owner
, offset
,
1998 refs_to_add
, extent_op
);
1999 if ((ret
< 0 && ret
!= -EAGAIN
) || !ret
)
2003 * Ok we had -EAGAIN which means we didn't have space to insert and
2004 * inline extent ref, so just update the reference count and add a
2007 leaf
= path
->nodes
[0];
2008 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2009 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2010 refs
= btrfs_extent_refs(leaf
, item
);
2011 btrfs_set_extent_refs(leaf
, item
, refs
+ refs_to_add
);
2013 __run_delayed_extent_op(extent_op
, leaf
, item
);
2015 btrfs_mark_buffer_dirty(leaf
);
2016 btrfs_release_path(path
);
2019 path
->leave_spinning
= 1;
2020 /* now insert the actual backref */
2021 ret
= insert_extent_backref(trans
, root
->fs_info
->extent_root
,
2022 path
, bytenr
, parent
, root_objectid
,
2023 owner
, offset
, refs_to_add
);
2025 btrfs_abort_transaction(trans
, root
, ret
);
2027 btrfs_free_path(path
);
2031 static int run_delayed_data_ref(struct btrfs_trans_handle
*trans
,
2032 struct btrfs_root
*root
,
2033 struct btrfs_delayed_ref_node
*node
,
2034 struct btrfs_delayed_extent_op
*extent_op
,
2035 int insert_reserved
)
2038 struct btrfs_delayed_data_ref
*ref
;
2039 struct btrfs_key ins
;
2044 ins
.objectid
= node
->bytenr
;
2045 ins
.offset
= node
->num_bytes
;
2046 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2048 ref
= btrfs_delayed_node_to_data_ref(node
);
2049 trace_run_delayed_data_ref(node
, ref
, node
->action
);
2051 if (node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2052 parent
= ref
->parent
;
2053 ref_root
= ref
->root
;
2055 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2057 flags
|= extent_op
->flags_to_set
;
2058 ret
= alloc_reserved_file_extent(trans
, root
,
2059 parent
, ref_root
, flags
,
2060 ref
->objectid
, ref
->offset
,
2061 &ins
, node
->ref_mod
);
2062 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2063 ret
= __btrfs_inc_extent_ref(trans
, root
, node
, parent
,
2064 ref_root
, ref
->objectid
,
2065 ref
->offset
, node
->ref_mod
,
2067 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2068 ret
= __btrfs_free_extent(trans
, root
, node
, parent
,
2069 ref_root
, ref
->objectid
,
2070 ref
->offset
, node
->ref_mod
,
2078 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
2079 struct extent_buffer
*leaf
,
2080 struct btrfs_extent_item
*ei
)
2082 u64 flags
= btrfs_extent_flags(leaf
, ei
);
2083 if (extent_op
->update_flags
) {
2084 flags
|= extent_op
->flags_to_set
;
2085 btrfs_set_extent_flags(leaf
, ei
, flags
);
2088 if (extent_op
->update_key
) {
2089 struct btrfs_tree_block_info
*bi
;
2090 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
));
2091 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
2092 btrfs_set_tree_block_key(leaf
, bi
, &extent_op
->key
);
2096 static int run_delayed_extent_op(struct btrfs_trans_handle
*trans
,
2097 struct btrfs_root
*root
,
2098 struct btrfs_delayed_ref_node
*node
,
2099 struct btrfs_delayed_extent_op
*extent_op
)
2101 struct btrfs_key key
;
2102 struct btrfs_path
*path
;
2103 struct btrfs_extent_item
*ei
;
2104 struct extent_buffer
*leaf
;
2108 int metadata
= !extent_op
->is_data
;
2113 if (metadata
&& !btrfs_fs_incompat(root
->fs_info
, SKINNY_METADATA
))
2116 path
= btrfs_alloc_path();
2120 key
.objectid
= node
->bytenr
;
2123 key
.type
= BTRFS_METADATA_ITEM_KEY
;
2124 key
.offset
= extent_op
->level
;
2126 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2127 key
.offset
= node
->num_bytes
;
2132 path
->leave_spinning
= 1;
2133 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
,
2141 if (path
->slots
[0] > 0) {
2143 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
2145 if (key
.objectid
== node
->bytenr
&&
2146 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
2147 key
.offset
== node
->num_bytes
)
2151 btrfs_release_path(path
);
2154 key
.objectid
= node
->bytenr
;
2155 key
.offset
= node
->num_bytes
;
2156 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2165 leaf
= path
->nodes
[0];
2166 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2167 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2168 if (item_size
< sizeof(*ei
)) {
2169 ret
= convert_extent_item_v0(trans
, root
->fs_info
->extent_root
,
2175 leaf
= path
->nodes
[0];
2176 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2179 BUG_ON(item_size
< sizeof(*ei
));
2180 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2181 __run_delayed_extent_op(extent_op
, leaf
, ei
);
2183 btrfs_mark_buffer_dirty(leaf
);
2185 btrfs_free_path(path
);
2189 static int run_delayed_tree_ref(struct btrfs_trans_handle
*trans
,
2190 struct btrfs_root
*root
,
2191 struct btrfs_delayed_ref_node
*node
,
2192 struct btrfs_delayed_extent_op
*extent_op
,
2193 int insert_reserved
)
2196 struct btrfs_delayed_tree_ref
*ref
;
2197 struct btrfs_key ins
;
2200 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
2203 ref
= btrfs_delayed_node_to_tree_ref(node
);
2204 trace_run_delayed_tree_ref(node
, ref
, node
->action
);
2206 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2207 parent
= ref
->parent
;
2208 ref_root
= ref
->root
;
2210 ins
.objectid
= node
->bytenr
;
2211 if (skinny_metadata
) {
2212 ins
.offset
= ref
->level
;
2213 ins
.type
= BTRFS_METADATA_ITEM_KEY
;
2215 ins
.offset
= node
->num_bytes
;
2216 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2219 BUG_ON(node
->ref_mod
!= 1);
2220 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2221 BUG_ON(!extent_op
|| !extent_op
->update_flags
);
2222 ret
= alloc_reserved_tree_block(trans
, root
,
2224 extent_op
->flags_to_set
,
2228 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2229 ret
= __btrfs_inc_extent_ref(trans
, root
, node
,
2233 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2234 ret
= __btrfs_free_extent(trans
, root
, node
,
2236 ref
->level
, 0, 1, extent_op
);
2243 /* helper function to actually process a single delayed ref entry */
2244 static int run_one_delayed_ref(struct btrfs_trans_handle
*trans
,
2245 struct btrfs_root
*root
,
2246 struct btrfs_delayed_ref_node
*node
,
2247 struct btrfs_delayed_extent_op
*extent_op
,
2248 int insert_reserved
)
2252 if (trans
->aborted
) {
2253 if (insert_reserved
)
2254 btrfs_pin_extent(root
, node
->bytenr
,
2255 node
->num_bytes
, 1);
2259 if (btrfs_delayed_ref_is_head(node
)) {
2260 struct btrfs_delayed_ref_head
*head
;
2262 * we've hit the end of the chain and we were supposed
2263 * to insert this extent into the tree. But, it got
2264 * deleted before we ever needed to insert it, so all
2265 * we have to do is clean up the accounting
2268 head
= btrfs_delayed_node_to_head(node
);
2269 trace_run_delayed_ref_head(node
, head
, node
->action
);
2271 if (insert_reserved
) {
2272 btrfs_pin_extent(root
, node
->bytenr
,
2273 node
->num_bytes
, 1);
2274 if (head
->is_data
) {
2275 ret
= btrfs_del_csums(trans
, root
,
2283 if (node
->type
== BTRFS_TREE_BLOCK_REF_KEY
||
2284 node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2285 ret
= run_delayed_tree_ref(trans
, root
, node
, extent_op
,
2287 else if (node
->type
== BTRFS_EXTENT_DATA_REF_KEY
||
2288 node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2289 ret
= run_delayed_data_ref(trans
, root
, node
, extent_op
,
2296 static inline struct btrfs_delayed_ref_node
*
2297 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
2299 if (list_empty(&head
->ref_list
))
2302 return list_entry(head
->ref_list
.next
, struct btrfs_delayed_ref_node
,
2307 * Returns 0 on success or if called with an already aborted transaction.
2308 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2310 static noinline
int __btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2311 struct btrfs_root
*root
,
2314 struct btrfs_delayed_ref_root
*delayed_refs
;
2315 struct btrfs_delayed_ref_node
*ref
;
2316 struct btrfs_delayed_ref_head
*locked_ref
= NULL
;
2317 struct btrfs_delayed_extent_op
*extent_op
;
2318 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2319 ktime_t start
= ktime_get();
2321 unsigned long count
= 0;
2322 unsigned long actual_count
= 0;
2323 int must_insert_reserved
= 0;
2325 delayed_refs
= &trans
->transaction
->delayed_refs
;
2331 spin_lock(&delayed_refs
->lock
);
2332 locked_ref
= btrfs_select_ref_head(trans
);
2334 spin_unlock(&delayed_refs
->lock
);
2338 /* grab the lock that says we are going to process
2339 * all the refs for this head */
2340 ret
= btrfs_delayed_ref_lock(trans
, locked_ref
);
2341 spin_unlock(&delayed_refs
->lock
);
2343 * we may have dropped the spin lock to get the head
2344 * mutex lock, and that might have given someone else
2345 * time to free the head. If that's true, it has been
2346 * removed from our list and we can move on.
2348 if (ret
== -EAGAIN
) {
2355 spin_lock(&locked_ref
->lock
);
2358 * locked_ref is the head node, so we have to go one
2359 * node back for any delayed ref updates
2361 ref
= select_delayed_ref(locked_ref
);
2363 if (ref
&& ref
->seq
&&
2364 btrfs_check_delayed_seq(fs_info
, delayed_refs
, ref
->seq
)) {
2365 spin_unlock(&locked_ref
->lock
);
2366 btrfs_delayed_ref_unlock(locked_ref
);
2367 spin_lock(&delayed_refs
->lock
);
2368 locked_ref
->processing
= 0;
2369 delayed_refs
->num_heads_ready
++;
2370 spin_unlock(&delayed_refs
->lock
);
2378 * record the must insert reserved flag before we
2379 * drop the spin lock.
2381 must_insert_reserved
= locked_ref
->must_insert_reserved
;
2382 locked_ref
->must_insert_reserved
= 0;
2384 extent_op
= locked_ref
->extent_op
;
2385 locked_ref
->extent_op
= NULL
;
2390 /* All delayed refs have been processed, Go ahead
2391 * and send the head node to run_one_delayed_ref,
2392 * so that any accounting fixes can happen
2394 ref
= &locked_ref
->node
;
2396 if (extent_op
&& must_insert_reserved
) {
2397 btrfs_free_delayed_extent_op(extent_op
);
2402 spin_unlock(&locked_ref
->lock
);
2403 ret
= run_delayed_extent_op(trans
, root
,
2405 btrfs_free_delayed_extent_op(extent_op
);
2409 * Need to reset must_insert_reserved if
2410 * there was an error so the abort stuff
2411 * can cleanup the reserved space
2414 if (must_insert_reserved
)
2415 locked_ref
->must_insert_reserved
= 1;
2416 locked_ref
->processing
= 0;
2417 btrfs_debug(fs_info
, "run_delayed_extent_op returned %d", ret
);
2418 btrfs_delayed_ref_unlock(locked_ref
);
2425 * Need to drop our head ref lock and re-aqcuire the
2426 * delayed ref lock and then re-check to make sure
2429 spin_unlock(&locked_ref
->lock
);
2430 spin_lock(&delayed_refs
->lock
);
2431 spin_lock(&locked_ref
->lock
);
2432 if (!list_empty(&locked_ref
->ref_list
) ||
2433 locked_ref
->extent_op
) {
2434 spin_unlock(&locked_ref
->lock
);
2435 spin_unlock(&delayed_refs
->lock
);
2439 delayed_refs
->num_heads
--;
2440 rb_erase(&locked_ref
->href_node
,
2441 &delayed_refs
->href_root
);
2442 spin_unlock(&delayed_refs
->lock
);
2446 list_del(&ref
->list
);
2448 atomic_dec(&delayed_refs
->num_entries
);
2450 if (!btrfs_delayed_ref_is_head(ref
)) {
2452 * when we play the delayed ref, also correct the
2455 switch (ref
->action
) {
2456 case BTRFS_ADD_DELAYED_REF
:
2457 case BTRFS_ADD_DELAYED_EXTENT
:
2458 locked_ref
->node
.ref_mod
-= ref
->ref_mod
;
2460 case BTRFS_DROP_DELAYED_REF
:
2461 locked_ref
->node
.ref_mod
+= ref
->ref_mod
;
2467 spin_unlock(&locked_ref
->lock
);
2469 ret
= run_one_delayed_ref(trans
, root
, ref
, extent_op
,
2470 must_insert_reserved
);
2472 btrfs_free_delayed_extent_op(extent_op
);
2474 locked_ref
->processing
= 0;
2475 btrfs_delayed_ref_unlock(locked_ref
);
2476 btrfs_put_delayed_ref(ref
);
2477 btrfs_debug(fs_info
, "run_one_delayed_ref returned %d", ret
);
2482 * If this node is a head, that means all the refs in this head
2483 * have been dealt with, and we will pick the next head to deal
2484 * with, so we must unlock the head and drop it from the cluster
2485 * list before we release it.
2487 if (btrfs_delayed_ref_is_head(ref
)) {
2488 if (locked_ref
->is_data
&&
2489 locked_ref
->total_ref_mod
< 0) {
2490 spin_lock(&delayed_refs
->lock
);
2491 delayed_refs
->pending_csums
-= ref
->num_bytes
;
2492 spin_unlock(&delayed_refs
->lock
);
2494 btrfs_delayed_ref_unlock(locked_ref
);
2497 btrfs_put_delayed_ref(ref
);
2503 * We don't want to include ref heads since we can have empty ref heads
2504 * and those will drastically skew our runtime down since we just do
2505 * accounting, no actual extent tree updates.
2507 if (actual_count
> 0) {
2508 u64 runtime
= ktime_to_ns(ktime_sub(ktime_get(), start
));
2512 * We weigh the current average higher than our current runtime
2513 * to avoid large swings in the average.
2515 spin_lock(&delayed_refs
->lock
);
2516 avg
= fs_info
->avg_delayed_ref_runtime
* 3 + runtime
;
2517 fs_info
->avg_delayed_ref_runtime
= avg
>> 2; /* div by 4 */
2518 spin_unlock(&delayed_refs
->lock
);
2523 #ifdef SCRAMBLE_DELAYED_REFS
2525 * Normally delayed refs get processed in ascending bytenr order. This
2526 * correlates in most cases to the order added. To expose dependencies on this
2527 * order, we start to process the tree in the middle instead of the beginning
2529 static u64
find_middle(struct rb_root
*root
)
2531 struct rb_node
*n
= root
->rb_node
;
2532 struct btrfs_delayed_ref_node
*entry
;
2535 u64 first
= 0, last
= 0;
2539 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2540 first
= entry
->bytenr
;
2544 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2545 last
= entry
->bytenr
;
2550 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2551 WARN_ON(!entry
->in_tree
);
2553 middle
= entry
->bytenr
;
2566 static inline u64
heads_to_leaves(struct btrfs_root
*root
, u64 heads
)
2570 num_bytes
= heads
* (sizeof(struct btrfs_extent_item
) +
2571 sizeof(struct btrfs_extent_inline_ref
));
2572 if (!btrfs_fs_incompat(root
->fs_info
, SKINNY_METADATA
))
2573 num_bytes
+= heads
* sizeof(struct btrfs_tree_block_info
);
2576 * We don't ever fill up leaves all the way so multiply by 2 just to be
2577 * closer to what we're really going to want to ouse.
2579 return div_u64(num_bytes
, BTRFS_LEAF_DATA_SIZE(root
));
2583 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2584 * would require to store the csums for that many bytes.
2586 u64
btrfs_csum_bytes_to_leaves(struct btrfs_root
*root
, u64 csum_bytes
)
2589 u64 num_csums_per_leaf
;
2592 csum_size
= BTRFS_LEAF_DATA_SIZE(root
) - sizeof(struct btrfs_item
);
2593 num_csums_per_leaf
= div64_u64(csum_size
,
2594 (u64
)btrfs_super_csum_size(root
->fs_info
->super_copy
));
2595 num_csums
= div64_u64(csum_bytes
, root
->sectorsize
);
2596 num_csums
+= num_csums_per_leaf
- 1;
2597 num_csums
= div64_u64(num_csums
, num_csums_per_leaf
);
2601 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle
*trans
,
2602 struct btrfs_root
*root
)
2604 struct btrfs_block_rsv
*global_rsv
;
2605 u64 num_heads
= trans
->transaction
->delayed_refs
.num_heads_ready
;
2606 u64 csum_bytes
= trans
->transaction
->delayed_refs
.pending_csums
;
2607 u64 num_dirty_bgs
= trans
->transaction
->num_dirty_bgs
;
2608 u64 num_bytes
, num_dirty_bgs_bytes
;
2611 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
2612 num_heads
= heads_to_leaves(root
, num_heads
);
2614 num_bytes
+= (num_heads
- 1) * root
->nodesize
;
2616 num_bytes
+= btrfs_csum_bytes_to_leaves(root
, csum_bytes
) * root
->nodesize
;
2617 num_dirty_bgs_bytes
= btrfs_calc_trans_metadata_size(root
,
2619 global_rsv
= &root
->fs_info
->global_block_rsv
;
2622 * If we can't allocate any more chunks lets make sure we have _lots_ of
2623 * wiggle room since running delayed refs can create more delayed refs.
2625 if (global_rsv
->space_info
->full
) {
2626 num_dirty_bgs_bytes
<<= 1;
2630 spin_lock(&global_rsv
->lock
);
2631 if (global_rsv
->reserved
<= num_bytes
+ num_dirty_bgs_bytes
)
2633 spin_unlock(&global_rsv
->lock
);
2637 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle
*trans
,
2638 struct btrfs_root
*root
)
2640 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2642 atomic_read(&trans
->transaction
->delayed_refs
.num_entries
);
2647 avg_runtime
= fs_info
->avg_delayed_ref_runtime
;
2648 val
= num_entries
* avg_runtime
;
2649 if (num_entries
* avg_runtime
>= NSEC_PER_SEC
)
2651 if (val
>= NSEC_PER_SEC
/ 2)
2654 return btrfs_check_space_for_delayed_refs(trans
, root
);
2657 struct async_delayed_refs
{
2658 struct btrfs_root
*root
;
2662 struct completion wait
;
2663 struct btrfs_work work
;
2666 static void delayed_ref_async_start(struct btrfs_work
*work
)
2668 struct async_delayed_refs
*async
;
2669 struct btrfs_trans_handle
*trans
;
2672 async
= container_of(work
, struct async_delayed_refs
, work
);
2674 trans
= btrfs_join_transaction(async
->root
);
2675 if (IS_ERR(trans
)) {
2676 async
->error
= PTR_ERR(trans
);
2681 * trans->sync means that when we call end_transaciton, we won't
2682 * wait on delayed refs
2685 ret
= btrfs_run_delayed_refs(trans
, async
->root
, async
->count
);
2689 ret
= btrfs_end_transaction(trans
, async
->root
);
2690 if (ret
&& !async
->error
)
2694 complete(&async
->wait
);
2699 int btrfs_async_run_delayed_refs(struct btrfs_root
*root
,
2700 unsigned long count
, int wait
)
2702 struct async_delayed_refs
*async
;
2705 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
2709 async
->root
= root
->fs_info
->tree_root
;
2710 async
->count
= count
;
2716 init_completion(&async
->wait
);
2718 btrfs_init_work(&async
->work
, btrfs_extent_refs_helper
,
2719 delayed_ref_async_start
, NULL
, NULL
);
2721 btrfs_queue_work(root
->fs_info
->extent_workers
, &async
->work
);
2724 wait_for_completion(&async
->wait
);
2733 * this starts processing the delayed reference count updates and
2734 * extent insertions we have queued up so far. count can be
2735 * 0, which means to process everything in the tree at the start
2736 * of the run (but not newly added entries), or it can be some target
2737 * number you'd like to process.
2739 * Returns 0 on success or if called with an aborted transaction
2740 * Returns <0 on error and aborts the transaction
2742 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2743 struct btrfs_root
*root
, unsigned long count
)
2745 struct rb_node
*node
;
2746 struct btrfs_delayed_ref_root
*delayed_refs
;
2747 struct btrfs_delayed_ref_head
*head
;
2749 int run_all
= count
== (unsigned long)-1;
2751 /* We'll clean this up in btrfs_cleanup_transaction */
2755 if (root
== root
->fs_info
->extent_root
)
2756 root
= root
->fs_info
->tree_root
;
2758 delayed_refs
= &trans
->transaction
->delayed_refs
;
2760 count
= atomic_read(&delayed_refs
->num_entries
) * 2;
2763 #ifdef SCRAMBLE_DELAYED_REFS
2764 delayed_refs
->run_delayed_start
= find_middle(&delayed_refs
->root
);
2766 ret
= __btrfs_run_delayed_refs(trans
, root
, count
);
2768 btrfs_abort_transaction(trans
, root
, ret
);
2773 if (!list_empty(&trans
->new_bgs
))
2774 btrfs_create_pending_block_groups(trans
, root
);
2776 spin_lock(&delayed_refs
->lock
);
2777 node
= rb_first(&delayed_refs
->href_root
);
2779 spin_unlock(&delayed_refs
->lock
);
2782 count
= (unsigned long)-1;
2785 head
= rb_entry(node
, struct btrfs_delayed_ref_head
,
2787 if (btrfs_delayed_ref_is_head(&head
->node
)) {
2788 struct btrfs_delayed_ref_node
*ref
;
2791 atomic_inc(&ref
->refs
);
2793 spin_unlock(&delayed_refs
->lock
);
2795 * Mutex was contended, block until it's
2796 * released and try again
2798 mutex_lock(&head
->mutex
);
2799 mutex_unlock(&head
->mutex
);
2801 btrfs_put_delayed_ref(ref
);
2807 node
= rb_next(node
);
2809 spin_unlock(&delayed_refs
->lock
);
2814 assert_qgroups_uptodate(trans
);
2818 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
2819 struct btrfs_root
*root
,
2820 u64 bytenr
, u64 num_bytes
, u64 flags
,
2821 int level
, int is_data
)
2823 struct btrfs_delayed_extent_op
*extent_op
;
2826 extent_op
= btrfs_alloc_delayed_extent_op();
2830 extent_op
->flags_to_set
= flags
;
2831 extent_op
->update_flags
= 1;
2832 extent_op
->update_key
= 0;
2833 extent_op
->is_data
= is_data
? 1 : 0;
2834 extent_op
->level
= level
;
2836 ret
= btrfs_add_delayed_extent_op(root
->fs_info
, trans
, bytenr
,
2837 num_bytes
, extent_op
);
2839 btrfs_free_delayed_extent_op(extent_op
);
2843 static noinline
int check_delayed_ref(struct btrfs_trans_handle
*trans
,
2844 struct btrfs_root
*root
,
2845 struct btrfs_path
*path
,
2846 u64 objectid
, u64 offset
, u64 bytenr
)
2848 struct btrfs_delayed_ref_head
*head
;
2849 struct btrfs_delayed_ref_node
*ref
;
2850 struct btrfs_delayed_data_ref
*data_ref
;
2851 struct btrfs_delayed_ref_root
*delayed_refs
;
2854 delayed_refs
= &trans
->transaction
->delayed_refs
;
2855 spin_lock(&delayed_refs
->lock
);
2856 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
2858 spin_unlock(&delayed_refs
->lock
);
2862 if (!mutex_trylock(&head
->mutex
)) {
2863 atomic_inc(&head
->node
.refs
);
2864 spin_unlock(&delayed_refs
->lock
);
2866 btrfs_release_path(path
);
2869 * Mutex was contended, block until it's released and let
2872 mutex_lock(&head
->mutex
);
2873 mutex_unlock(&head
->mutex
);
2874 btrfs_put_delayed_ref(&head
->node
);
2877 spin_unlock(&delayed_refs
->lock
);
2879 spin_lock(&head
->lock
);
2880 list_for_each_entry(ref
, &head
->ref_list
, list
) {
2881 /* If it's a shared ref we know a cross reference exists */
2882 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
) {
2887 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
2890 * If our ref doesn't match the one we're currently looking at
2891 * then we have a cross reference.
2893 if (data_ref
->root
!= root
->root_key
.objectid
||
2894 data_ref
->objectid
!= objectid
||
2895 data_ref
->offset
!= offset
) {
2900 spin_unlock(&head
->lock
);
2901 mutex_unlock(&head
->mutex
);
2905 static noinline
int check_committed_ref(struct btrfs_trans_handle
*trans
,
2906 struct btrfs_root
*root
,
2907 struct btrfs_path
*path
,
2908 u64 objectid
, u64 offset
, u64 bytenr
)
2910 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2911 struct extent_buffer
*leaf
;
2912 struct btrfs_extent_data_ref
*ref
;
2913 struct btrfs_extent_inline_ref
*iref
;
2914 struct btrfs_extent_item
*ei
;
2915 struct btrfs_key key
;
2919 key
.objectid
= bytenr
;
2920 key
.offset
= (u64
)-1;
2921 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2923 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
2926 BUG_ON(ret
== 0); /* Corruption */
2929 if (path
->slots
[0] == 0)
2933 leaf
= path
->nodes
[0];
2934 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2936 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
2940 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2941 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2942 if (item_size
< sizeof(*ei
)) {
2943 WARN_ON(item_size
!= sizeof(struct btrfs_extent_item_v0
));
2947 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2949 if (item_size
!= sizeof(*ei
) +
2950 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
2953 if (btrfs_extent_generation(leaf
, ei
) <=
2954 btrfs_root_last_snapshot(&root
->root_item
))
2957 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
2958 if (btrfs_extent_inline_ref_type(leaf
, iref
) !=
2959 BTRFS_EXTENT_DATA_REF_KEY
)
2962 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
2963 if (btrfs_extent_refs(leaf
, ei
) !=
2964 btrfs_extent_data_ref_count(leaf
, ref
) ||
2965 btrfs_extent_data_ref_root(leaf
, ref
) !=
2966 root
->root_key
.objectid
||
2967 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
2968 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
2976 int btrfs_cross_ref_exist(struct btrfs_trans_handle
*trans
,
2977 struct btrfs_root
*root
,
2978 u64 objectid
, u64 offset
, u64 bytenr
)
2980 struct btrfs_path
*path
;
2984 path
= btrfs_alloc_path();
2989 ret
= check_committed_ref(trans
, root
, path
, objectid
,
2991 if (ret
&& ret
!= -ENOENT
)
2994 ret2
= check_delayed_ref(trans
, root
, path
, objectid
,
2996 } while (ret2
== -EAGAIN
);
2998 if (ret2
&& ret2
!= -ENOENT
) {
3003 if (ret
!= -ENOENT
|| ret2
!= -ENOENT
)
3006 btrfs_free_path(path
);
3007 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
3012 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
3013 struct btrfs_root
*root
,
3014 struct extent_buffer
*buf
,
3015 int full_backref
, int inc
)
3022 struct btrfs_key key
;
3023 struct btrfs_file_extent_item
*fi
;
3027 int (*process_func
)(struct btrfs_trans_handle
*, struct btrfs_root
*,
3028 u64
, u64
, u64
, u64
, u64
, u64
, int);
3031 if (btrfs_test_is_dummy_root(root
))
3034 ref_root
= btrfs_header_owner(buf
);
3035 nritems
= btrfs_header_nritems(buf
);
3036 level
= btrfs_header_level(buf
);
3038 if (!test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) && level
== 0)
3042 process_func
= btrfs_inc_extent_ref
;
3044 process_func
= btrfs_free_extent
;
3047 parent
= buf
->start
;
3051 for (i
= 0; i
< nritems
; i
++) {
3053 btrfs_item_key_to_cpu(buf
, &key
, i
);
3054 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
3056 fi
= btrfs_item_ptr(buf
, i
,
3057 struct btrfs_file_extent_item
);
3058 if (btrfs_file_extent_type(buf
, fi
) ==
3059 BTRFS_FILE_EXTENT_INLINE
)
3061 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
3065 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
3066 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
3067 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
3068 parent
, ref_root
, key
.objectid
,
3073 bytenr
= btrfs_node_blockptr(buf
, i
);
3074 num_bytes
= root
->nodesize
;
3075 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
3076 parent
, ref_root
, level
- 1, 0,
3087 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3088 struct extent_buffer
*buf
, int full_backref
)
3090 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1);
3093 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3094 struct extent_buffer
*buf
, int full_backref
)
3096 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0);
3099 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
3100 struct btrfs_root
*root
,
3101 struct btrfs_path
*path
,
3102 struct btrfs_block_group_cache
*cache
)
3105 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
3107 struct extent_buffer
*leaf
;
3109 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
3116 leaf
= path
->nodes
[0];
3117 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
3118 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
3119 btrfs_mark_buffer_dirty(leaf
);
3121 btrfs_release_path(path
);
3126 static struct btrfs_block_group_cache
*
3127 next_block_group(struct btrfs_root
*root
,
3128 struct btrfs_block_group_cache
*cache
)
3130 struct rb_node
*node
;
3132 spin_lock(&root
->fs_info
->block_group_cache_lock
);
3134 /* If our block group was removed, we need a full search. */
3135 if (RB_EMPTY_NODE(&cache
->cache_node
)) {
3136 const u64 next_bytenr
= cache
->key
.objectid
+ cache
->key
.offset
;
3138 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
3139 btrfs_put_block_group(cache
);
3140 cache
= btrfs_lookup_first_block_group(root
->fs_info
,
3144 node
= rb_next(&cache
->cache_node
);
3145 btrfs_put_block_group(cache
);
3147 cache
= rb_entry(node
, struct btrfs_block_group_cache
,
3149 btrfs_get_block_group(cache
);
3152 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
3156 static int cache_save_setup(struct btrfs_block_group_cache
*block_group
,
3157 struct btrfs_trans_handle
*trans
,
3158 struct btrfs_path
*path
)
3160 struct btrfs_root
*root
= block_group
->fs_info
->tree_root
;
3161 struct inode
*inode
= NULL
;
3163 int dcs
= BTRFS_DC_ERROR
;
3169 * If this block group is smaller than 100 megs don't bother caching the
3172 if (block_group
->key
.offset
< (100 * 1024 * 1024)) {
3173 spin_lock(&block_group
->lock
);
3174 block_group
->disk_cache_state
= BTRFS_DC_WRITTEN
;
3175 spin_unlock(&block_group
->lock
);
3182 inode
= lookup_free_space_inode(root
, block_group
, path
);
3183 if (IS_ERR(inode
) && PTR_ERR(inode
) != -ENOENT
) {
3184 ret
= PTR_ERR(inode
);
3185 btrfs_release_path(path
);
3189 if (IS_ERR(inode
)) {
3193 if (block_group
->ro
)
3196 ret
= create_free_space_inode(root
, trans
, block_group
, path
);
3202 /* We've already setup this transaction, go ahead and exit */
3203 if (block_group
->cache_generation
== trans
->transid
&&
3204 i_size_read(inode
)) {
3205 dcs
= BTRFS_DC_SETUP
;
3210 * We want to set the generation to 0, that way if anything goes wrong
3211 * from here on out we know not to trust this cache when we load up next
3214 BTRFS_I(inode
)->generation
= 0;
3215 ret
= btrfs_update_inode(trans
, root
, inode
);
3218 * So theoretically we could recover from this, simply set the
3219 * super cache generation to 0 so we know to invalidate the
3220 * cache, but then we'd have to keep track of the block groups
3221 * that fail this way so we know we _have_ to reset this cache
3222 * before the next commit or risk reading stale cache. So to
3223 * limit our exposure to horrible edge cases lets just abort the
3224 * transaction, this only happens in really bad situations
3227 btrfs_abort_transaction(trans
, root
, ret
);
3232 if (i_size_read(inode
) > 0) {
3233 ret
= btrfs_check_trunc_cache_free_space(root
,
3234 &root
->fs_info
->global_block_rsv
);
3238 ret
= btrfs_truncate_free_space_cache(root
, trans
, NULL
, inode
);
3243 spin_lock(&block_group
->lock
);
3244 if (block_group
->cached
!= BTRFS_CACHE_FINISHED
||
3245 !btrfs_test_opt(root
, SPACE_CACHE
)) {
3247 * don't bother trying to write stuff out _if_
3248 * a) we're not cached,
3249 * b) we're with nospace_cache mount option.
3251 dcs
= BTRFS_DC_WRITTEN
;
3252 spin_unlock(&block_group
->lock
);
3255 spin_unlock(&block_group
->lock
);
3258 * Try to preallocate enough space based on how big the block group is.
3259 * Keep in mind this has to include any pinned space which could end up
3260 * taking up quite a bit since it's not folded into the other space
3263 num_pages
= div_u64(block_group
->key
.offset
, 256 * 1024 * 1024);
3268 num_pages
*= PAGE_CACHE_SIZE
;
3270 ret
= btrfs_check_data_free_space(inode
, num_pages
, num_pages
);
3274 ret
= btrfs_prealloc_file_range_trans(inode
, trans
, 0, 0, num_pages
,
3275 num_pages
, num_pages
,
3278 dcs
= BTRFS_DC_SETUP
;
3279 btrfs_free_reserved_data_space(inode
, num_pages
);
3284 btrfs_release_path(path
);
3286 spin_lock(&block_group
->lock
);
3287 if (!ret
&& dcs
== BTRFS_DC_SETUP
)
3288 block_group
->cache_generation
= trans
->transid
;
3289 block_group
->disk_cache_state
= dcs
;
3290 spin_unlock(&block_group
->lock
);
3295 int btrfs_setup_space_cache(struct btrfs_trans_handle
*trans
,
3296 struct btrfs_root
*root
)
3298 struct btrfs_block_group_cache
*cache
, *tmp
;
3299 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3300 struct btrfs_path
*path
;
3302 if (list_empty(&cur_trans
->dirty_bgs
) ||
3303 !btrfs_test_opt(root
, SPACE_CACHE
))
3306 path
= btrfs_alloc_path();
3310 /* Could add new block groups, use _safe just in case */
3311 list_for_each_entry_safe(cache
, tmp
, &cur_trans
->dirty_bgs
,
3313 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
)
3314 cache_save_setup(cache
, trans
, path
);
3317 btrfs_free_path(path
);
3322 * transaction commit does final block group cache writeback during a
3323 * critical section where nothing is allowed to change the FS. This is
3324 * required in order for the cache to actually match the block group,
3325 * but can introduce a lot of latency into the commit.
3327 * So, btrfs_start_dirty_block_groups is here to kick off block group
3328 * cache IO. There's a chance we'll have to redo some of it if the
3329 * block group changes again during the commit, but it greatly reduces
3330 * the commit latency by getting rid of the easy block groups while
3331 * we're still allowing others to join the commit.
3333 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle
*trans
,
3334 struct btrfs_root
*root
)
3336 struct btrfs_block_group_cache
*cache
;
3337 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3340 struct btrfs_path
*path
= NULL
;
3342 struct list_head
*io
= &cur_trans
->io_bgs
;
3343 int num_started
= 0;
3346 spin_lock(&cur_trans
->dirty_bgs_lock
);
3347 if (list_empty(&cur_trans
->dirty_bgs
)) {
3348 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3351 list_splice_init(&cur_trans
->dirty_bgs
, &dirty
);
3352 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3356 * make sure all the block groups on our dirty list actually
3359 btrfs_create_pending_block_groups(trans
, root
);
3362 path
= btrfs_alloc_path();
3368 * cache_write_mutex is here only to save us from balance or automatic
3369 * removal of empty block groups deleting this block group while we are
3370 * writing out the cache
3372 mutex_lock(&trans
->transaction
->cache_write_mutex
);
3373 while (!list_empty(&dirty
)) {
3374 cache
= list_first_entry(&dirty
,
3375 struct btrfs_block_group_cache
,
3378 * this can happen if something re-dirties a block
3379 * group that is already under IO. Just wait for it to
3380 * finish and then do it all again
3382 if (!list_empty(&cache
->io_list
)) {
3383 list_del_init(&cache
->io_list
);
3384 btrfs_wait_cache_io(root
, trans
, cache
,
3385 &cache
->io_ctl
, path
,
3386 cache
->key
.objectid
);
3387 btrfs_put_block_group(cache
);
3392 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3393 * if it should update the cache_state. Don't delete
3394 * until after we wait.
3396 * Since we're not running in the commit critical section
3397 * we need the dirty_bgs_lock to protect from update_block_group
3399 spin_lock(&cur_trans
->dirty_bgs_lock
);
3400 list_del_init(&cache
->dirty_list
);
3401 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3405 cache_save_setup(cache
, trans
, path
);
3407 if (cache
->disk_cache_state
== BTRFS_DC_SETUP
) {
3408 cache
->io_ctl
.inode
= NULL
;
3409 ret
= btrfs_write_out_cache(root
, trans
, cache
, path
);
3410 if (ret
== 0 && cache
->io_ctl
.inode
) {
3415 * the cache_write_mutex is protecting
3418 list_add_tail(&cache
->io_list
, io
);
3421 * if we failed to write the cache, the
3422 * generation will be bad and life goes on
3428 ret
= write_one_cache_group(trans
, root
, path
, cache
);
3430 * Our block group might still be attached to the list
3431 * of new block groups in the transaction handle of some
3432 * other task (struct btrfs_trans_handle->new_bgs). This
3433 * means its block group item isn't yet in the extent
3434 * tree. If this happens ignore the error, as we will
3435 * try again later in the critical section of the
3436 * transaction commit.
3438 if (ret
== -ENOENT
) {
3440 spin_lock(&cur_trans
->dirty_bgs_lock
);
3441 if (list_empty(&cache
->dirty_list
)) {
3442 list_add_tail(&cache
->dirty_list
,
3443 &cur_trans
->dirty_bgs
);
3444 btrfs_get_block_group(cache
);
3446 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3448 btrfs_abort_transaction(trans
, root
, ret
);
3452 /* if its not on the io list, we need to put the block group */
3454 btrfs_put_block_group(cache
);
3460 * Avoid blocking other tasks for too long. It might even save
3461 * us from writing caches for block groups that are going to be
3464 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
3465 mutex_lock(&trans
->transaction
->cache_write_mutex
);
3467 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
3470 * go through delayed refs for all the stuff we've just kicked off
3471 * and then loop back (just once)
3473 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
3474 if (!ret
&& loops
== 0) {
3476 spin_lock(&cur_trans
->dirty_bgs_lock
);
3477 list_splice_init(&cur_trans
->dirty_bgs
, &dirty
);
3479 * dirty_bgs_lock protects us from concurrent block group
3480 * deletes too (not just cache_write_mutex).
3482 if (!list_empty(&dirty
)) {
3483 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3486 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3489 btrfs_free_path(path
);
3493 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
3494 struct btrfs_root
*root
)
3496 struct btrfs_block_group_cache
*cache
;
3497 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3500 struct btrfs_path
*path
;
3501 struct list_head
*io
= &cur_trans
->io_bgs
;
3502 int num_started
= 0;
3504 path
= btrfs_alloc_path();
3509 * We don't need the lock here since we are protected by the transaction
3510 * commit. We want to do the cache_save_setup first and then run the
3511 * delayed refs to make sure we have the best chance at doing this all
3514 while (!list_empty(&cur_trans
->dirty_bgs
)) {
3515 cache
= list_first_entry(&cur_trans
->dirty_bgs
,
3516 struct btrfs_block_group_cache
,
3520 * this can happen if cache_save_setup re-dirties a block
3521 * group that is already under IO. Just wait for it to
3522 * finish and then do it all again
3524 if (!list_empty(&cache
->io_list
)) {
3525 list_del_init(&cache
->io_list
);
3526 btrfs_wait_cache_io(root
, trans
, cache
,
3527 &cache
->io_ctl
, path
,
3528 cache
->key
.objectid
);
3529 btrfs_put_block_group(cache
);
3533 * don't remove from the dirty list until after we've waited
3536 list_del_init(&cache
->dirty_list
);
3539 cache_save_setup(cache
, trans
, path
);
3542 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long) -1);
3544 if (!ret
&& cache
->disk_cache_state
== BTRFS_DC_SETUP
) {
3545 cache
->io_ctl
.inode
= NULL
;
3546 ret
= btrfs_write_out_cache(root
, trans
, cache
, path
);
3547 if (ret
== 0 && cache
->io_ctl
.inode
) {
3550 list_add_tail(&cache
->io_list
, io
);
3553 * if we failed to write the cache, the
3554 * generation will be bad and life goes on
3560 ret
= write_one_cache_group(trans
, root
, path
, cache
);
3562 btrfs_abort_transaction(trans
, root
, ret
);
3565 /* if its not on the io list, we need to put the block group */
3567 btrfs_put_block_group(cache
);
3570 while (!list_empty(io
)) {
3571 cache
= list_first_entry(io
, struct btrfs_block_group_cache
,
3573 list_del_init(&cache
->io_list
);
3574 btrfs_wait_cache_io(root
, trans
, cache
,
3575 &cache
->io_ctl
, path
, cache
->key
.objectid
);
3576 btrfs_put_block_group(cache
);
3579 btrfs_free_path(path
);
3583 int btrfs_extent_readonly(struct btrfs_root
*root
, u64 bytenr
)
3585 struct btrfs_block_group_cache
*block_group
;
3588 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
3589 if (!block_group
|| block_group
->ro
)
3592 btrfs_put_block_group(block_group
);
3596 static const char *alloc_name(u64 flags
)
3599 case BTRFS_BLOCK_GROUP_METADATA
|BTRFS_BLOCK_GROUP_DATA
:
3601 case BTRFS_BLOCK_GROUP_METADATA
:
3603 case BTRFS_BLOCK_GROUP_DATA
:
3605 case BTRFS_BLOCK_GROUP_SYSTEM
:
3609 return "invalid-combination";
3613 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
3614 u64 total_bytes
, u64 bytes_used
,
3615 struct btrfs_space_info
**space_info
)
3617 struct btrfs_space_info
*found
;
3622 if (flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3623 BTRFS_BLOCK_GROUP_RAID10
))
3628 found
= __find_space_info(info
, flags
);
3630 spin_lock(&found
->lock
);
3631 found
->total_bytes
+= total_bytes
;
3632 found
->disk_total
+= total_bytes
* factor
;
3633 found
->bytes_used
+= bytes_used
;
3634 found
->disk_used
+= bytes_used
* factor
;
3635 if (total_bytes
> 0)
3637 spin_unlock(&found
->lock
);
3638 *space_info
= found
;
3641 found
= kzalloc(sizeof(*found
), GFP_NOFS
);
3645 ret
= percpu_counter_init(&found
->total_bytes_pinned
, 0, GFP_KERNEL
);
3651 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
3652 INIT_LIST_HEAD(&found
->block_groups
[i
]);
3653 init_rwsem(&found
->groups_sem
);
3654 spin_lock_init(&found
->lock
);
3655 found
->flags
= flags
& BTRFS_BLOCK_GROUP_TYPE_MASK
;
3656 found
->total_bytes
= total_bytes
;
3657 found
->disk_total
= total_bytes
* factor
;
3658 found
->bytes_used
= bytes_used
;
3659 found
->disk_used
= bytes_used
* factor
;
3660 found
->bytes_pinned
= 0;
3661 found
->bytes_reserved
= 0;
3662 found
->bytes_readonly
= 0;
3663 found
->bytes_may_use
= 0;
3664 if (total_bytes
> 0)
3668 found
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3669 found
->chunk_alloc
= 0;
3671 init_waitqueue_head(&found
->wait
);
3672 INIT_LIST_HEAD(&found
->ro_bgs
);
3674 ret
= kobject_init_and_add(&found
->kobj
, &space_info_ktype
,
3675 info
->space_info_kobj
, "%s",
3676 alloc_name(found
->flags
));
3682 *space_info
= found
;
3683 list_add_rcu(&found
->list
, &info
->space_info
);
3684 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3685 info
->data_sinfo
= found
;
3690 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
3692 u64 extra_flags
= chunk_to_extended(flags
) &
3693 BTRFS_EXTENDED_PROFILE_MASK
;
3695 write_seqlock(&fs_info
->profiles_lock
);
3696 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3697 fs_info
->avail_data_alloc_bits
|= extra_flags
;
3698 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3699 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
3700 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3701 fs_info
->avail_system_alloc_bits
|= extra_flags
;
3702 write_sequnlock(&fs_info
->profiles_lock
);
3706 * returns target flags in extended format or 0 if restripe for this
3707 * chunk_type is not in progress
3709 * should be called with either volume_mutex or balance_lock held
3711 static u64
get_restripe_target(struct btrfs_fs_info
*fs_info
, u64 flags
)
3713 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3719 if (flags
& BTRFS_BLOCK_GROUP_DATA
&&
3720 bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3721 target
= BTRFS_BLOCK_GROUP_DATA
| bctl
->data
.target
;
3722 } else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
&&
3723 bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3724 target
= BTRFS_BLOCK_GROUP_SYSTEM
| bctl
->sys
.target
;
3725 } else if (flags
& BTRFS_BLOCK_GROUP_METADATA
&&
3726 bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3727 target
= BTRFS_BLOCK_GROUP_METADATA
| bctl
->meta
.target
;
3734 * @flags: available profiles in extended format (see ctree.h)
3736 * Returns reduced profile in chunk format. If profile changing is in
3737 * progress (either running or paused) picks the target profile (if it's
3738 * already available), otherwise falls back to plain reducing.
3740 static u64
btrfs_reduce_alloc_profile(struct btrfs_root
*root
, u64 flags
)
3742 u64 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
3747 * see if restripe for this chunk_type is in progress, if so
3748 * try to reduce to the target profile
3750 spin_lock(&root
->fs_info
->balance_lock
);
3751 target
= get_restripe_target(root
->fs_info
, flags
);
3753 /* pick target profile only if it's already available */
3754 if ((flags
& target
) & BTRFS_EXTENDED_PROFILE_MASK
) {
3755 spin_unlock(&root
->fs_info
->balance_lock
);
3756 return extended_to_chunk(target
);
3759 spin_unlock(&root
->fs_info
->balance_lock
);
3761 /* First, mask out the RAID levels which aren't possible */
3762 if (num_devices
== 1)
3763 flags
&= ~(BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID0
|
3764 BTRFS_BLOCK_GROUP_RAID5
);
3765 if (num_devices
< 3)
3766 flags
&= ~BTRFS_BLOCK_GROUP_RAID6
;
3767 if (num_devices
< 4)
3768 flags
&= ~BTRFS_BLOCK_GROUP_RAID10
;
3770 tmp
= flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID0
|
3771 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID5
|
3772 BTRFS_BLOCK_GROUP_RAID6
| BTRFS_BLOCK_GROUP_RAID10
);
3775 if (tmp
& BTRFS_BLOCK_GROUP_RAID6
)
3776 tmp
= BTRFS_BLOCK_GROUP_RAID6
;
3777 else if (tmp
& BTRFS_BLOCK_GROUP_RAID5
)
3778 tmp
= BTRFS_BLOCK_GROUP_RAID5
;
3779 else if (tmp
& BTRFS_BLOCK_GROUP_RAID10
)
3780 tmp
= BTRFS_BLOCK_GROUP_RAID10
;
3781 else if (tmp
& BTRFS_BLOCK_GROUP_RAID1
)
3782 tmp
= BTRFS_BLOCK_GROUP_RAID1
;
3783 else if (tmp
& BTRFS_BLOCK_GROUP_RAID0
)
3784 tmp
= BTRFS_BLOCK_GROUP_RAID0
;
3786 return extended_to_chunk(flags
| tmp
);
3789 static u64
get_alloc_profile(struct btrfs_root
*root
, u64 orig_flags
)
3796 seq
= read_seqbegin(&root
->fs_info
->profiles_lock
);
3798 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3799 flags
|= root
->fs_info
->avail_data_alloc_bits
;
3800 else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3801 flags
|= root
->fs_info
->avail_system_alloc_bits
;
3802 else if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3803 flags
|= root
->fs_info
->avail_metadata_alloc_bits
;
3804 } while (read_seqretry(&root
->fs_info
->profiles_lock
, seq
));
3806 return btrfs_reduce_alloc_profile(root
, flags
);
3809 u64
btrfs_get_alloc_profile(struct btrfs_root
*root
, int data
)
3815 flags
= BTRFS_BLOCK_GROUP_DATA
;
3816 else if (root
== root
->fs_info
->chunk_root
)
3817 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
3819 flags
= BTRFS_BLOCK_GROUP_METADATA
;
3821 ret
= get_alloc_profile(root
, flags
);
3826 * This will check the space that the inode allocates from to make sure we have
3827 * enough space for bytes.
3829 int btrfs_check_data_free_space(struct inode
*inode
, u64 bytes
, u64 write_bytes
)
3831 struct btrfs_space_info
*data_sinfo
;
3832 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3833 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3836 int need_commit
= 2;
3837 int have_pinned_space
;
3839 /* make sure bytes are sectorsize aligned */
3840 bytes
= ALIGN(bytes
, root
->sectorsize
);
3842 if (btrfs_is_free_space_inode(inode
)) {
3844 ASSERT(current
->journal_info
);
3847 data_sinfo
= fs_info
->data_sinfo
;
3852 /* make sure we have enough space to handle the data first */
3853 spin_lock(&data_sinfo
->lock
);
3854 used
= data_sinfo
->bytes_used
+ data_sinfo
->bytes_reserved
+
3855 data_sinfo
->bytes_pinned
+ data_sinfo
->bytes_readonly
+
3856 data_sinfo
->bytes_may_use
;
3858 if (used
+ bytes
> data_sinfo
->total_bytes
) {
3859 struct btrfs_trans_handle
*trans
;
3862 * if we don't have enough free bytes in this space then we need
3863 * to alloc a new chunk.
3865 if (!data_sinfo
->full
) {
3868 data_sinfo
->force_alloc
= CHUNK_ALLOC_FORCE
;
3869 spin_unlock(&data_sinfo
->lock
);
3871 alloc_target
= btrfs_get_alloc_profile(root
, 1);
3873 * It is ugly that we don't call nolock join
3874 * transaction for the free space inode case here.
3875 * But it is safe because we only do the data space
3876 * reservation for the free space cache in the
3877 * transaction context, the common join transaction
3878 * just increase the counter of the current transaction
3879 * handler, doesn't try to acquire the trans_lock of
3882 trans
= btrfs_join_transaction(root
);
3884 return PTR_ERR(trans
);
3886 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
3888 CHUNK_ALLOC_NO_FORCE
);
3889 btrfs_end_transaction(trans
, root
);
3894 have_pinned_space
= 1;
3900 data_sinfo
= fs_info
->data_sinfo
;
3906 * If we don't have enough pinned space to deal with this
3907 * allocation, and no removed chunk in current transaction,
3908 * don't bother committing the transaction.
3910 have_pinned_space
= percpu_counter_compare(
3911 &data_sinfo
->total_bytes_pinned
,
3912 used
+ bytes
- data_sinfo
->total_bytes
);
3913 spin_unlock(&data_sinfo
->lock
);
3915 /* commit the current transaction and try again */
3918 !atomic_read(&root
->fs_info
->open_ioctl_trans
)) {
3921 if (need_commit
> 0)
3922 btrfs_wait_ordered_roots(fs_info
, -1);
3924 trans
= btrfs_join_transaction(root
);
3926 return PTR_ERR(trans
);
3927 if (have_pinned_space
>= 0 ||
3928 trans
->transaction
->have_free_bgs
||
3930 ret
= btrfs_commit_transaction(trans
, root
);
3934 * make sure that all running delayed iput are
3937 down_write(&root
->fs_info
->delayed_iput_sem
);
3938 up_write(&root
->fs_info
->delayed_iput_sem
);
3941 btrfs_end_transaction(trans
, root
);
3945 trace_btrfs_space_reservation(root
->fs_info
,
3946 "space_info:enospc",
3947 data_sinfo
->flags
, bytes
, 1);
3950 ret
= btrfs_qgroup_reserve(root
, write_bytes
);
3953 data_sinfo
->bytes_may_use
+= bytes
;
3954 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
3955 data_sinfo
->flags
, bytes
, 1);
3957 spin_unlock(&data_sinfo
->lock
);
3963 * Called if we need to clear a data reservation for this inode.
3965 void btrfs_free_reserved_data_space(struct inode
*inode
, u64 bytes
)
3967 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3968 struct btrfs_space_info
*data_sinfo
;
3970 /* make sure bytes are sectorsize aligned */
3971 bytes
= ALIGN(bytes
, root
->sectorsize
);
3973 data_sinfo
= root
->fs_info
->data_sinfo
;
3974 spin_lock(&data_sinfo
->lock
);
3975 WARN_ON(data_sinfo
->bytes_may_use
< bytes
);
3976 data_sinfo
->bytes_may_use
-= bytes
;
3977 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
3978 data_sinfo
->flags
, bytes
, 0);
3979 spin_unlock(&data_sinfo
->lock
);
3982 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
3984 struct list_head
*head
= &info
->space_info
;
3985 struct btrfs_space_info
*found
;
3988 list_for_each_entry_rcu(found
, head
, list
) {
3989 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
3990 found
->force_alloc
= CHUNK_ALLOC_FORCE
;
3995 static inline u64
calc_global_rsv_need_space(struct btrfs_block_rsv
*global
)
3997 return (global
->size
<< 1);
4000 static int should_alloc_chunk(struct btrfs_root
*root
,
4001 struct btrfs_space_info
*sinfo
, int force
)
4003 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
4004 u64 num_bytes
= sinfo
->total_bytes
- sinfo
->bytes_readonly
;
4005 u64 num_allocated
= sinfo
->bytes_used
+ sinfo
->bytes_reserved
;
4008 if (force
== CHUNK_ALLOC_FORCE
)
4012 * We need to take into account the global rsv because for all intents
4013 * and purposes it's used space. Don't worry about locking the
4014 * global_rsv, it doesn't change except when the transaction commits.
4016 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
4017 num_allocated
+= calc_global_rsv_need_space(global_rsv
);
4020 * in limited mode, we want to have some free space up to
4021 * about 1% of the FS size.
4023 if (force
== CHUNK_ALLOC_LIMITED
) {
4024 thresh
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
4025 thresh
= max_t(u64
, 64 * 1024 * 1024,
4026 div_factor_fine(thresh
, 1));
4028 if (num_bytes
- num_allocated
< thresh
)
4032 if (num_allocated
+ 2 * 1024 * 1024 < div_factor(num_bytes
, 8))
4037 static u64
get_profile_num_devs(struct btrfs_root
*root
, u64 type
)
4041 if (type
& (BTRFS_BLOCK_GROUP_RAID10
|
4042 BTRFS_BLOCK_GROUP_RAID0
|
4043 BTRFS_BLOCK_GROUP_RAID5
|
4044 BTRFS_BLOCK_GROUP_RAID6
))
4045 num_dev
= root
->fs_info
->fs_devices
->rw_devices
;
4046 else if (type
& BTRFS_BLOCK_GROUP_RAID1
)
4049 num_dev
= 1; /* DUP or single */
4055 * If @is_allocation is true, reserve space in the system space info necessary
4056 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4059 void check_system_chunk(struct btrfs_trans_handle
*trans
,
4060 struct btrfs_root
*root
,
4063 struct btrfs_space_info
*info
;
4070 * Needed because we can end up allocating a system chunk and for an
4071 * atomic and race free space reservation in the chunk block reserve.
4073 ASSERT(mutex_is_locked(&root
->fs_info
->chunk_mutex
));
4075 info
= __find_space_info(root
->fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
4076 spin_lock(&info
->lock
);
4077 left
= info
->total_bytes
- info
->bytes_used
- info
->bytes_pinned
-
4078 info
->bytes_reserved
- info
->bytes_readonly
-
4079 info
->bytes_may_use
;
4080 spin_unlock(&info
->lock
);
4082 num_devs
= get_profile_num_devs(root
, type
);
4084 /* num_devs device items to update and 1 chunk item to add or remove */
4085 thresh
= btrfs_calc_trunc_metadata_size(root
, num_devs
) +
4086 btrfs_calc_trans_metadata_size(root
, 1);
4088 if (left
< thresh
&& btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
4089 btrfs_info(root
->fs_info
, "left=%llu, need=%llu, flags=%llu",
4090 left
, thresh
, type
);
4091 dump_space_info(info
, 0, 0);
4094 if (left
< thresh
) {
4097 flags
= btrfs_get_alloc_profile(root
->fs_info
->chunk_root
, 0);
4099 * Ignore failure to create system chunk. We might end up not
4100 * needing it, as we might not need to COW all nodes/leafs from
4101 * the paths we visit in the chunk tree (they were already COWed
4102 * or created in the current transaction for example).
4104 ret
= btrfs_alloc_chunk(trans
, root
, flags
);
4108 ret
= btrfs_block_rsv_add(root
->fs_info
->chunk_root
,
4109 &root
->fs_info
->chunk_block_rsv
,
4110 thresh
, BTRFS_RESERVE_NO_FLUSH
);
4112 trans
->chunk_bytes_reserved
+= thresh
;
4116 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
4117 struct btrfs_root
*extent_root
, u64 flags
, int force
)
4119 struct btrfs_space_info
*space_info
;
4120 struct btrfs_fs_info
*fs_info
= extent_root
->fs_info
;
4121 int wait_for_alloc
= 0;
4124 /* Don't re-enter if we're already allocating a chunk */
4125 if (trans
->allocating_chunk
)
4128 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
4130 ret
= update_space_info(extent_root
->fs_info
, flags
,
4132 BUG_ON(ret
); /* -ENOMEM */
4134 BUG_ON(!space_info
); /* Logic error */
4137 spin_lock(&space_info
->lock
);
4138 if (force
< space_info
->force_alloc
)
4139 force
= space_info
->force_alloc
;
4140 if (space_info
->full
) {
4141 if (should_alloc_chunk(extent_root
, space_info
, force
))
4145 spin_unlock(&space_info
->lock
);
4149 if (!should_alloc_chunk(extent_root
, space_info
, force
)) {
4150 spin_unlock(&space_info
->lock
);
4152 } else if (space_info
->chunk_alloc
) {
4155 space_info
->chunk_alloc
= 1;
4158 spin_unlock(&space_info
->lock
);
4160 mutex_lock(&fs_info
->chunk_mutex
);
4163 * The chunk_mutex is held throughout the entirety of a chunk
4164 * allocation, so once we've acquired the chunk_mutex we know that the
4165 * other guy is done and we need to recheck and see if we should
4168 if (wait_for_alloc
) {
4169 mutex_unlock(&fs_info
->chunk_mutex
);
4174 trans
->allocating_chunk
= true;
4177 * If we have mixed data/metadata chunks we want to make sure we keep
4178 * allocating mixed chunks instead of individual chunks.
4180 if (btrfs_mixed_space_info(space_info
))
4181 flags
|= (BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
);
4184 * if we're doing a data chunk, go ahead and make sure that
4185 * we keep a reasonable number of metadata chunks allocated in the
4188 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
4189 fs_info
->data_chunk_allocations
++;
4190 if (!(fs_info
->data_chunk_allocations
%
4191 fs_info
->metadata_ratio
))
4192 force_metadata_allocation(fs_info
);
4196 * Check if we have enough space in SYSTEM chunk because we may need
4197 * to update devices.
4199 check_system_chunk(trans
, extent_root
, flags
);
4201 ret
= btrfs_alloc_chunk(trans
, extent_root
, flags
);
4202 trans
->allocating_chunk
= false;
4204 spin_lock(&space_info
->lock
);
4205 if (ret
< 0 && ret
!= -ENOSPC
)
4208 space_info
->full
= 1;
4212 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
4214 space_info
->chunk_alloc
= 0;
4215 spin_unlock(&space_info
->lock
);
4216 mutex_unlock(&fs_info
->chunk_mutex
);
4220 static int can_overcommit(struct btrfs_root
*root
,
4221 struct btrfs_space_info
*space_info
, u64 bytes
,
4222 enum btrfs_reserve_flush_enum flush
)
4224 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
4225 u64 profile
= btrfs_get_alloc_profile(root
, 0);
4230 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4231 space_info
->bytes_pinned
+ space_info
->bytes_readonly
;
4234 * We only want to allow over committing if we have lots of actual space
4235 * free, but if we don't have enough space to handle the global reserve
4236 * space then we could end up having a real enospc problem when trying
4237 * to allocate a chunk or some other such important allocation.
4239 spin_lock(&global_rsv
->lock
);
4240 space_size
= calc_global_rsv_need_space(global_rsv
);
4241 spin_unlock(&global_rsv
->lock
);
4242 if (used
+ space_size
>= space_info
->total_bytes
)
4245 used
+= space_info
->bytes_may_use
;
4247 spin_lock(&root
->fs_info
->free_chunk_lock
);
4248 avail
= root
->fs_info
->free_chunk_space
;
4249 spin_unlock(&root
->fs_info
->free_chunk_lock
);
4252 * If we have dup, raid1 or raid10 then only half of the free
4253 * space is actually useable. For raid56, the space info used
4254 * doesn't include the parity drive, so we don't have to
4257 if (profile
& (BTRFS_BLOCK_GROUP_DUP
|
4258 BTRFS_BLOCK_GROUP_RAID1
|
4259 BTRFS_BLOCK_GROUP_RAID10
))
4263 * If we aren't flushing all things, let us overcommit up to
4264 * 1/2th of the space. If we can flush, don't let us overcommit
4265 * too much, let it overcommit up to 1/8 of the space.
4267 if (flush
== BTRFS_RESERVE_FLUSH_ALL
)
4272 if (used
+ bytes
< space_info
->total_bytes
+ avail
)
4277 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root
*root
,
4278 unsigned long nr_pages
, int nr_items
)
4280 struct super_block
*sb
= root
->fs_info
->sb
;
4282 if (down_read_trylock(&sb
->s_umount
)) {
4283 writeback_inodes_sb_nr(sb
, nr_pages
, WB_REASON_FS_FREE_SPACE
);
4284 up_read(&sb
->s_umount
);
4287 * We needn't worry the filesystem going from r/w to r/o though
4288 * we don't acquire ->s_umount mutex, because the filesystem
4289 * should guarantee the delalloc inodes list be empty after
4290 * the filesystem is readonly(all dirty pages are written to
4293 btrfs_start_delalloc_roots(root
->fs_info
, 0, nr_items
);
4294 if (!current
->journal_info
)
4295 btrfs_wait_ordered_roots(root
->fs_info
, nr_items
);
4299 static inline int calc_reclaim_items_nr(struct btrfs_root
*root
, u64 to_reclaim
)
4304 bytes
= btrfs_calc_trans_metadata_size(root
, 1);
4305 nr
= (int)div64_u64(to_reclaim
, bytes
);
4311 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4314 * shrink metadata reservation for delalloc
4316 static void shrink_delalloc(struct btrfs_root
*root
, u64 to_reclaim
, u64 orig
,
4319 struct btrfs_block_rsv
*block_rsv
;
4320 struct btrfs_space_info
*space_info
;
4321 struct btrfs_trans_handle
*trans
;
4325 unsigned long nr_pages
;
4328 enum btrfs_reserve_flush_enum flush
;
4330 /* Calc the number of the pages we need flush for space reservation */
4331 items
= calc_reclaim_items_nr(root
, to_reclaim
);
4332 to_reclaim
= items
* EXTENT_SIZE_PER_ITEM
;
4334 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4335 block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
4336 space_info
= block_rsv
->space_info
;
4338 delalloc_bytes
= percpu_counter_sum_positive(
4339 &root
->fs_info
->delalloc_bytes
);
4340 if (delalloc_bytes
== 0) {
4344 btrfs_wait_ordered_roots(root
->fs_info
, items
);
4349 while (delalloc_bytes
&& loops
< 3) {
4350 max_reclaim
= min(delalloc_bytes
, to_reclaim
);
4351 nr_pages
= max_reclaim
>> PAGE_CACHE_SHIFT
;
4352 btrfs_writeback_inodes_sb_nr(root
, nr_pages
, items
);
4354 * We need to wait for the async pages to actually start before
4357 max_reclaim
= atomic_read(&root
->fs_info
->async_delalloc_pages
);
4361 if (max_reclaim
<= nr_pages
)
4364 max_reclaim
-= nr_pages
;
4366 wait_event(root
->fs_info
->async_submit_wait
,
4367 atomic_read(&root
->fs_info
->async_delalloc_pages
) <=
4371 flush
= BTRFS_RESERVE_FLUSH_ALL
;
4373 flush
= BTRFS_RESERVE_NO_FLUSH
;
4374 spin_lock(&space_info
->lock
);
4375 if (can_overcommit(root
, space_info
, orig
, flush
)) {
4376 spin_unlock(&space_info
->lock
);
4379 spin_unlock(&space_info
->lock
);
4382 if (wait_ordered
&& !trans
) {
4383 btrfs_wait_ordered_roots(root
->fs_info
, items
);
4385 time_left
= schedule_timeout_killable(1);
4389 delalloc_bytes
= percpu_counter_sum_positive(
4390 &root
->fs_info
->delalloc_bytes
);
4395 * maybe_commit_transaction - possibly commit the transaction if its ok to
4396 * @root - the root we're allocating for
4397 * @bytes - the number of bytes we want to reserve
4398 * @force - force the commit
4400 * This will check to make sure that committing the transaction will actually
4401 * get us somewhere and then commit the transaction if it does. Otherwise it
4402 * will return -ENOSPC.
4404 static int may_commit_transaction(struct btrfs_root
*root
,
4405 struct btrfs_space_info
*space_info
,
4406 u64 bytes
, int force
)
4408 struct btrfs_block_rsv
*delayed_rsv
= &root
->fs_info
->delayed_block_rsv
;
4409 struct btrfs_trans_handle
*trans
;
4411 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4418 /* See if there is enough pinned space to make this reservation */
4419 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4424 * See if there is some space in the delayed insertion reservation for
4427 if (space_info
!= delayed_rsv
->space_info
)
4430 spin_lock(&delayed_rsv
->lock
);
4431 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4432 bytes
- delayed_rsv
->size
) >= 0) {
4433 spin_unlock(&delayed_rsv
->lock
);
4436 spin_unlock(&delayed_rsv
->lock
);
4439 trans
= btrfs_join_transaction(root
);
4443 return btrfs_commit_transaction(trans
, root
);
4447 FLUSH_DELAYED_ITEMS_NR
= 1,
4448 FLUSH_DELAYED_ITEMS
= 2,
4450 FLUSH_DELALLOC_WAIT
= 4,
4455 static int flush_space(struct btrfs_root
*root
,
4456 struct btrfs_space_info
*space_info
, u64 num_bytes
,
4457 u64 orig_bytes
, int state
)
4459 struct btrfs_trans_handle
*trans
;
4464 case FLUSH_DELAYED_ITEMS_NR
:
4465 case FLUSH_DELAYED_ITEMS
:
4466 if (state
== FLUSH_DELAYED_ITEMS_NR
)
4467 nr
= calc_reclaim_items_nr(root
, num_bytes
) * 2;
4471 trans
= btrfs_join_transaction(root
);
4472 if (IS_ERR(trans
)) {
4473 ret
= PTR_ERR(trans
);
4476 ret
= btrfs_run_delayed_items_nr(trans
, root
, nr
);
4477 btrfs_end_transaction(trans
, root
);
4479 case FLUSH_DELALLOC
:
4480 case FLUSH_DELALLOC_WAIT
:
4481 shrink_delalloc(root
, num_bytes
* 2, orig_bytes
,
4482 state
== FLUSH_DELALLOC_WAIT
);
4485 trans
= btrfs_join_transaction(root
);
4486 if (IS_ERR(trans
)) {
4487 ret
= PTR_ERR(trans
);
4490 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
4491 btrfs_get_alloc_profile(root
, 0),
4492 CHUNK_ALLOC_NO_FORCE
);
4493 btrfs_end_transaction(trans
, root
);
4498 ret
= may_commit_transaction(root
, space_info
, orig_bytes
, 0);
4509 btrfs_calc_reclaim_metadata_size(struct btrfs_root
*root
,
4510 struct btrfs_space_info
*space_info
)
4516 to_reclaim
= min_t(u64
, num_online_cpus() * 1024 * 1024,
4518 spin_lock(&space_info
->lock
);
4519 if (can_overcommit(root
, space_info
, to_reclaim
,
4520 BTRFS_RESERVE_FLUSH_ALL
)) {
4525 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4526 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4527 space_info
->bytes_may_use
;
4528 if (can_overcommit(root
, space_info
, 1024 * 1024,
4529 BTRFS_RESERVE_FLUSH_ALL
))
4530 expected
= div_factor_fine(space_info
->total_bytes
, 95);
4532 expected
= div_factor_fine(space_info
->total_bytes
, 90);
4534 if (used
> expected
)
4535 to_reclaim
= used
- expected
;
4538 to_reclaim
= min(to_reclaim
, space_info
->bytes_may_use
+
4539 space_info
->bytes_reserved
);
4541 spin_unlock(&space_info
->lock
);
4546 static inline int need_do_async_reclaim(struct btrfs_space_info
*space_info
,
4547 struct btrfs_fs_info
*fs_info
, u64 used
)
4549 u64 thresh
= div_factor_fine(space_info
->total_bytes
, 98);
4551 /* If we're just plain full then async reclaim just slows us down. */
4552 if (space_info
->bytes_used
>= thresh
)
4555 return (used
>= thresh
&& !btrfs_fs_closing(fs_info
) &&
4556 !test_bit(BTRFS_FS_STATE_REMOUNTING
, &fs_info
->fs_state
));
4559 static int btrfs_need_do_async_reclaim(struct btrfs_space_info
*space_info
,
4560 struct btrfs_fs_info
*fs_info
,
4565 spin_lock(&space_info
->lock
);
4567 * We run out of space and have not got any free space via flush_space,
4568 * so don't bother doing async reclaim.
4570 if (flush_state
> COMMIT_TRANS
&& space_info
->full
) {
4571 spin_unlock(&space_info
->lock
);
4575 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4576 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4577 space_info
->bytes_may_use
;
4578 if (need_do_async_reclaim(space_info
, fs_info
, used
)) {
4579 spin_unlock(&space_info
->lock
);
4582 spin_unlock(&space_info
->lock
);
4587 static void btrfs_async_reclaim_metadata_space(struct work_struct
*work
)
4589 struct btrfs_fs_info
*fs_info
;
4590 struct btrfs_space_info
*space_info
;
4594 fs_info
= container_of(work
, struct btrfs_fs_info
, async_reclaim_work
);
4595 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4597 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
->fs_root
,
4602 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
4604 flush_space(fs_info
->fs_root
, space_info
, to_reclaim
,
4605 to_reclaim
, flush_state
);
4607 if (!btrfs_need_do_async_reclaim(space_info
, fs_info
,
4610 } while (flush_state
< COMMIT_TRANS
);
4613 void btrfs_init_async_reclaim_work(struct work_struct
*work
)
4615 INIT_WORK(work
, btrfs_async_reclaim_metadata_space
);
4619 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4620 * @root - the root we're allocating for
4621 * @block_rsv - the block_rsv we're allocating for
4622 * @orig_bytes - the number of bytes we want
4623 * @flush - whether or not we can flush to make our reservation
4625 * This will reserve orgi_bytes number of bytes from the space info associated
4626 * with the block_rsv. If there is not enough space it will make an attempt to
4627 * flush out space to make room. It will do this by flushing delalloc if
4628 * possible or committing the transaction. If flush is 0 then no attempts to
4629 * regain reservations will be made and this will fail if there is not enough
4632 static int reserve_metadata_bytes(struct btrfs_root
*root
,
4633 struct btrfs_block_rsv
*block_rsv
,
4635 enum btrfs_reserve_flush_enum flush
)
4637 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
4639 u64 num_bytes
= orig_bytes
;
4640 int flush_state
= FLUSH_DELAYED_ITEMS_NR
;
4642 bool flushing
= false;
4646 spin_lock(&space_info
->lock
);
4648 * We only want to wait if somebody other than us is flushing and we
4649 * are actually allowed to flush all things.
4651 while (flush
== BTRFS_RESERVE_FLUSH_ALL
&& !flushing
&&
4652 space_info
->flush
) {
4653 spin_unlock(&space_info
->lock
);
4655 * If we have a trans handle we can't wait because the flusher
4656 * may have to commit the transaction, which would mean we would
4657 * deadlock since we are waiting for the flusher to finish, but
4658 * hold the current transaction open.
4660 if (current
->journal_info
)
4662 ret
= wait_event_killable(space_info
->wait
, !space_info
->flush
);
4663 /* Must have been killed, return */
4667 spin_lock(&space_info
->lock
);
4671 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4672 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4673 space_info
->bytes_may_use
;
4676 * The idea here is that we've not already over-reserved the block group
4677 * then we can go ahead and save our reservation first and then start
4678 * flushing if we need to. Otherwise if we've already overcommitted
4679 * lets start flushing stuff first and then come back and try to make
4682 if (used
<= space_info
->total_bytes
) {
4683 if (used
+ orig_bytes
<= space_info
->total_bytes
) {
4684 space_info
->bytes_may_use
+= orig_bytes
;
4685 trace_btrfs_space_reservation(root
->fs_info
,
4686 "space_info", space_info
->flags
, orig_bytes
, 1);
4690 * Ok set num_bytes to orig_bytes since we aren't
4691 * overocmmitted, this way we only try and reclaim what
4694 num_bytes
= orig_bytes
;
4698 * Ok we're over committed, set num_bytes to the overcommitted
4699 * amount plus the amount of bytes that we need for this
4702 num_bytes
= used
- space_info
->total_bytes
+
4706 if (ret
&& can_overcommit(root
, space_info
, orig_bytes
, flush
)) {
4707 space_info
->bytes_may_use
+= orig_bytes
;
4708 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
4709 space_info
->flags
, orig_bytes
,
4715 * Couldn't make our reservation, save our place so while we're trying
4716 * to reclaim space we can actually use it instead of somebody else
4717 * stealing it from us.
4719 * We make the other tasks wait for the flush only when we can flush
4722 if (ret
&& flush
!= BTRFS_RESERVE_NO_FLUSH
) {
4724 space_info
->flush
= 1;
4725 } else if (!ret
&& space_info
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
4728 * We will do the space reservation dance during log replay,
4729 * which means we won't have fs_info->fs_root set, so don't do
4730 * the async reclaim as we will panic.
4732 if (!root
->fs_info
->log_root_recovering
&&
4733 need_do_async_reclaim(space_info
, root
->fs_info
, used
) &&
4734 !work_busy(&root
->fs_info
->async_reclaim_work
))
4735 queue_work(system_unbound_wq
,
4736 &root
->fs_info
->async_reclaim_work
);
4738 spin_unlock(&space_info
->lock
);
4740 if (!ret
|| flush
== BTRFS_RESERVE_NO_FLUSH
)
4743 ret
= flush_space(root
, space_info
, num_bytes
, orig_bytes
,
4748 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4749 * would happen. So skip delalloc flush.
4751 if (flush
== BTRFS_RESERVE_FLUSH_LIMIT
&&
4752 (flush_state
== FLUSH_DELALLOC
||
4753 flush_state
== FLUSH_DELALLOC_WAIT
))
4754 flush_state
= ALLOC_CHUNK
;
4758 else if (flush
== BTRFS_RESERVE_FLUSH_LIMIT
&&
4759 flush_state
< COMMIT_TRANS
)
4761 else if (flush
== BTRFS_RESERVE_FLUSH_ALL
&&
4762 flush_state
<= COMMIT_TRANS
)
4766 if (ret
== -ENOSPC
&&
4767 unlikely(root
->orphan_cleanup_state
== ORPHAN_CLEANUP_STARTED
)) {
4768 struct btrfs_block_rsv
*global_rsv
=
4769 &root
->fs_info
->global_block_rsv
;
4771 if (block_rsv
!= global_rsv
&&
4772 !block_rsv_use_bytes(global_rsv
, orig_bytes
))
4776 trace_btrfs_space_reservation(root
->fs_info
,
4777 "space_info:enospc",
4778 space_info
->flags
, orig_bytes
, 1);
4780 spin_lock(&space_info
->lock
);
4781 space_info
->flush
= 0;
4782 wake_up_all(&space_info
->wait
);
4783 spin_unlock(&space_info
->lock
);
4788 static struct btrfs_block_rsv
*get_block_rsv(
4789 const struct btrfs_trans_handle
*trans
,
4790 const struct btrfs_root
*root
)
4792 struct btrfs_block_rsv
*block_rsv
= NULL
;
4794 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
))
4795 block_rsv
= trans
->block_rsv
;
4797 if (root
== root
->fs_info
->csum_root
&& trans
->adding_csums
)
4798 block_rsv
= trans
->block_rsv
;
4800 if (root
== root
->fs_info
->uuid_root
)
4801 block_rsv
= trans
->block_rsv
;
4804 block_rsv
= root
->block_rsv
;
4807 block_rsv
= &root
->fs_info
->empty_block_rsv
;
4812 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
4816 spin_lock(&block_rsv
->lock
);
4817 if (block_rsv
->reserved
>= num_bytes
) {
4818 block_rsv
->reserved
-= num_bytes
;
4819 if (block_rsv
->reserved
< block_rsv
->size
)
4820 block_rsv
->full
= 0;
4823 spin_unlock(&block_rsv
->lock
);
4827 static void block_rsv_add_bytes(struct btrfs_block_rsv
*block_rsv
,
4828 u64 num_bytes
, int update_size
)
4830 spin_lock(&block_rsv
->lock
);
4831 block_rsv
->reserved
+= num_bytes
;
4833 block_rsv
->size
+= num_bytes
;
4834 else if (block_rsv
->reserved
>= block_rsv
->size
)
4835 block_rsv
->full
= 1;
4836 spin_unlock(&block_rsv
->lock
);
4839 int btrfs_cond_migrate_bytes(struct btrfs_fs_info
*fs_info
,
4840 struct btrfs_block_rsv
*dest
, u64 num_bytes
,
4843 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
4846 if (global_rsv
->space_info
!= dest
->space_info
)
4849 spin_lock(&global_rsv
->lock
);
4850 min_bytes
= div_factor(global_rsv
->size
, min_factor
);
4851 if (global_rsv
->reserved
< min_bytes
+ num_bytes
) {
4852 spin_unlock(&global_rsv
->lock
);
4855 global_rsv
->reserved
-= num_bytes
;
4856 if (global_rsv
->reserved
< global_rsv
->size
)
4857 global_rsv
->full
= 0;
4858 spin_unlock(&global_rsv
->lock
);
4860 block_rsv_add_bytes(dest
, num_bytes
, 1);
4864 static void block_rsv_release_bytes(struct btrfs_fs_info
*fs_info
,
4865 struct btrfs_block_rsv
*block_rsv
,
4866 struct btrfs_block_rsv
*dest
, u64 num_bytes
)
4868 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
4870 spin_lock(&block_rsv
->lock
);
4871 if (num_bytes
== (u64
)-1)
4872 num_bytes
= block_rsv
->size
;
4873 block_rsv
->size
-= num_bytes
;
4874 if (block_rsv
->reserved
>= block_rsv
->size
) {
4875 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
4876 block_rsv
->reserved
= block_rsv
->size
;
4877 block_rsv
->full
= 1;
4881 spin_unlock(&block_rsv
->lock
);
4883 if (num_bytes
> 0) {
4885 spin_lock(&dest
->lock
);
4889 bytes_to_add
= dest
->size
- dest
->reserved
;
4890 bytes_to_add
= min(num_bytes
, bytes_to_add
);
4891 dest
->reserved
+= bytes_to_add
;
4892 if (dest
->reserved
>= dest
->size
)
4894 num_bytes
-= bytes_to_add
;
4896 spin_unlock(&dest
->lock
);
4899 spin_lock(&space_info
->lock
);
4900 space_info
->bytes_may_use
-= num_bytes
;
4901 trace_btrfs_space_reservation(fs_info
, "space_info",
4902 space_info
->flags
, num_bytes
, 0);
4903 spin_unlock(&space_info
->lock
);
4908 static int block_rsv_migrate_bytes(struct btrfs_block_rsv
*src
,
4909 struct btrfs_block_rsv
*dst
, u64 num_bytes
)
4913 ret
= block_rsv_use_bytes(src
, num_bytes
);
4917 block_rsv_add_bytes(dst
, num_bytes
, 1);
4921 void btrfs_init_block_rsv(struct btrfs_block_rsv
*rsv
, unsigned short type
)
4923 memset(rsv
, 0, sizeof(*rsv
));
4924 spin_lock_init(&rsv
->lock
);
4928 struct btrfs_block_rsv
*btrfs_alloc_block_rsv(struct btrfs_root
*root
,
4929 unsigned short type
)
4931 struct btrfs_block_rsv
*block_rsv
;
4932 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4934 block_rsv
= kmalloc(sizeof(*block_rsv
), GFP_NOFS
);
4938 btrfs_init_block_rsv(block_rsv
, type
);
4939 block_rsv
->space_info
= __find_space_info(fs_info
,
4940 BTRFS_BLOCK_GROUP_METADATA
);
4944 void btrfs_free_block_rsv(struct btrfs_root
*root
,
4945 struct btrfs_block_rsv
*rsv
)
4949 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
4953 void __btrfs_free_block_rsv(struct btrfs_block_rsv
*rsv
)
4958 int btrfs_block_rsv_add(struct btrfs_root
*root
,
4959 struct btrfs_block_rsv
*block_rsv
, u64 num_bytes
,
4960 enum btrfs_reserve_flush_enum flush
)
4967 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
4969 block_rsv_add_bytes(block_rsv
, num_bytes
, 1);
4976 int btrfs_block_rsv_check(struct btrfs_root
*root
,
4977 struct btrfs_block_rsv
*block_rsv
, int min_factor
)
4985 spin_lock(&block_rsv
->lock
);
4986 num_bytes
= div_factor(block_rsv
->size
, min_factor
);
4987 if (block_rsv
->reserved
>= num_bytes
)
4989 spin_unlock(&block_rsv
->lock
);
4994 int btrfs_block_rsv_refill(struct btrfs_root
*root
,
4995 struct btrfs_block_rsv
*block_rsv
, u64 min_reserved
,
4996 enum btrfs_reserve_flush_enum flush
)
5004 spin_lock(&block_rsv
->lock
);
5005 num_bytes
= min_reserved
;
5006 if (block_rsv
->reserved
>= num_bytes
)
5009 num_bytes
-= block_rsv
->reserved
;
5010 spin_unlock(&block_rsv
->lock
);
5015 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
5017 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
5024 int btrfs_block_rsv_migrate(struct btrfs_block_rsv
*src_rsv
,
5025 struct btrfs_block_rsv
*dst_rsv
,
5028 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
5031 void btrfs_block_rsv_release(struct btrfs_root
*root
,
5032 struct btrfs_block_rsv
*block_rsv
,
5035 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
5036 if (global_rsv
== block_rsv
||
5037 block_rsv
->space_info
!= global_rsv
->space_info
)
5039 block_rsv_release_bytes(root
->fs_info
, block_rsv
, global_rsv
,
5044 * helper to calculate size of global block reservation.
5045 * the desired value is sum of space used by extent tree,
5046 * checksum tree and root tree
5048 static u64
calc_global_metadata_size(struct btrfs_fs_info
*fs_info
)
5050 struct btrfs_space_info
*sinfo
;
5054 int csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
5056 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_DATA
);
5057 spin_lock(&sinfo
->lock
);
5058 data_used
= sinfo
->bytes_used
;
5059 spin_unlock(&sinfo
->lock
);
5061 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
5062 spin_lock(&sinfo
->lock
);
5063 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_DATA
)
5065 meta_used
= sinfo
->bytes_used
;
5066 spin_unlock(&sinfo
->lock
);
5068 num_bytes
= (data_used
>> fs_info
->sb
->s_blocksize_bits
) *
5070 num_bytes
+= div_u64(data_used
+ meta_used
, 50);
5072 if (num_bytes
* 3 > meta_used
)
5073 num_bytes
= div_u64(meta_used
, 3);
5075 return ALIGN(num_bytes
, fs_info
->extent_root
->nodesize
<< 10);
5078 static void update_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5080 struct btrfs_block_rsv
*block_rsv
= &fs_info
->global_block_rsv
;
5081 struct btrfs_space_info
*sinfo
= block_rsv
->space_info
;
5084 num_bytes
= calc_global_metadata_size(fs_info
);
5086 spin_lock(&sinfo
->lock
);
5087 spin_lock(&block_rsv
->lock
);
5089 block_rsv
->size
= min_t(u64
, num_bytes
, 512 * 1024 * 1024);
5091 num_bytes
= sinfo
->bytes_used
+ sinfo
->bytes_pinned
+
5092 sinfo
->bytes_reserved
+ sinfo
->bytes_readonly
+
5093 sinfo
->bytes_may_use
;
5095 if (sinfo
->total_bytes
> num_bytes
) {
5096 num_bytes
= sinfo
->total_bytes
- num_bytes
;
5097 block_rsv
->reserved
+= num_bytes
;
5098 sinfo
->bytes_may_use
+= num_bytes
;
5099 trace_btrfs_space_reservation(fs_info
, "space_info",
5100 sinfo
->flags
, num_bytes
, 1);
5103 if (block_rsv
->reserved
>= block_rsv
->size
) {
5104 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
5105 sinfo
->bytes_may_use
-= num_bytes
;
5106 trace_btrfs_space_reservation(fs_info
, "space_info",
5107 sinfo
->flags
, num_bytes
, 0);
5108 block_rsv
->reserved
= block_rsv
->size
;
5109 block_rsv
->full
= 1;
5112 spin_unlock(&block_rsv
->lock
);
5113 spin_unlock(&sinfo
->lock
);
5116 static void init_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5118 struct btrfs_space_info
*space_info
;
5120 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
5121 fs_info
->chunk_block_rsv
.space_info
= space_info
;
5123 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
5124 fs_info
->global_block_rsv
.space_info
= space_info
;
5125 fs_info
->delalloc_block_rsv
.space_info
= space_info
;
5126 fs_info
->trans_block_rsv
.space_info
= space_info
;
5127 fs_info
->empty_block_rsv
.space_info
= space_info
;
5128 fs_info
->delayed_block_rsv
.space_info
= space_info
;
5130 fs_info
->extent_root
->block_rsv
= &fs_info
->global_block_rsv
;
5131 fs_info
->csum_root
->block_rsv
= &fs_info
->global_block_rsv
;
5132 fs_info
->dev_root
->block_rsv
= &fs_info
->global_block_rsv
;
5133 fs_info
->tree_root
->block_rsv
= &fs_info
->global_block_rsv
;
5134 if (fs_info
->quota_root
)
5135 fs_info
->quota_root
->block_rsv
= &fs_info
->global_block_rsv
;
5136 fs_info
->chunk_root
->block_rsv
= &fs_info
->chunk_block_rsv
;
5138 update_global_block_rsv(fs_info
);
5141 static void release_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5143 block_rsv_release_bytes(fs_info
, &fs_info
->global_block_rsv
, NULL
,
5145 WARN_ON(fs_info
->delalloc_block_rsv
.size
> 0);
5146 WARN_ON(fs_info
->delalloc_block_rsv
.reserved
> 0);
5147 WARN_ON(fs_info
->trans_block_rsv
.size
> 0);
5148 WARN_ON(fs_info
->trans_block_rsv
.reserved
> 0);
5149 WARN_ON(fs_info
->chunk_block_rsv
.size
> 0);
5150 WARN_ON(fs_info
->chunk_block_rsv
.reserved
> 0);
5151 WARN_ON(fs_info
->delayed_block_rsv
.size
> 0);
5152 WARN_ON(fs_info
->delayed_block_rsv
.reserved
> 0);
5155 void btrfs_trans_release_metadata(struct btrfs_trans_handle
*trans
,
5156 struct btrfs_root
*root
)
5158 if (!trans
->block_rsv
)
5161 if (!trans
->bytes_reserved
)
5164 trace_btrfs_space_reservation(root
->fs_info
, "transaction",
5165 trans
->transid
, trans
->bytes_reserved
, 0);
5166 btrfs_block_rsv_release(root
, trans
->block_rsv
, trans
->bytes_reserved
);
5167 trans
->bytes_reserved
= 0;
5171 * To be called after all the new block groups attached to the transaction
5172 * handle have been created (btrfs_create_pending_block_groups()).
5174 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle
*trans
)
5176 struct btrfs_fs_info
*fs_info
= trans
->root
->fs_info
;
5178 if (!trans
->chunk_bytes_reserved
)
5181 WARN_ON_ONCE(!list_empty(&trans
->new_bgs
));
5183 block_rsv_release_bytes(fs_info
, &fs_info
->chunk_block_rsv
, NULL
,
5184 trans
->chunk_bytes_reserved
);
5185 trans
->chunk_bytes_reserved
= 0;
5188 /* Can only return 0 or -ENOSPC */
5189 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle
*trans
,
5190 struct inode
*inode
)
5192 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5193 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
5194 struct btrfs_block_rsv
*dst_rsv
= root
->orphan_block_rsv
;
5197 * We need to hold space in order to delete our orphan item once we've
5198 * added it, so this takes the reservation so we can release it later
5199 * when we are truly done with the orphan item.
5201 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
5202 trace_btrfs_space_reservation(root
->fs_info
, "orphan",
5203 btrfs_ino(inode
), num_bytes
, 1);
5204 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
5207 void btrfs_orphan_release_metadata(struct inode
*inode
)
5209 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5210 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
5211 trace_btrfs_space_reservation(root
->fs_info
, "orphan",
5212 btrfs_ino(inode
), num_bytes
, 0);
5213 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
, num_bytes
);
5217 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5218 * root: the root of the parent directory
5219 * rsv: block reservation
5220 * items: the number of items that we need do reservation
5221 * qgroup_reserved: used to return the reserved size in qgroup
5223 * This function is used to reserve the space for snapshot/subvolume
5224 * creation and deletion. Those operations are different with the
5225 * common file/directory operations, they change two fs/file trees
5226 * and root tree, the number of items that the qgroup reserves is
5227 * different with the free space reservation. So we can not use
5228 * the space reseravtion mechanism in start_transaction().
5230 int btrfs_subvolume_reserve_metadata(struct btrfs_root
*root
,
5231 struct btrfs_block_rsv
*rsv
,
5233 u64
*qgroup_reserved
,
5234 bool use_global_rsv
)
5238 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
5240 if (root
->fs_info
->quota_enabled
) {
5241 /* One for parent inode, two for dir entries */
5242 num_bytes
= 3 * root
->nodesize
;
5243 ret
= btrfs_qgroup_reserve(root
, num_bytes
);
5250 *qgroup_reserved
= num_bytes
;
5252 num_bytes
= btrfs_calc_trans_metadata_size(root
, items
);
5253 rsv
->space_info
= __find_space_info(root
->fs_info
,
5254 BTRFS_BLOCK_GROUP_METADATA
);
5255 ret
= btrfs_block_rsv_add(root
, rsv
, num_bytes
,
5256 BTRFS_RESERVE_FLUSH_ALL
);
5258 if (ret
== -ENOSPC
&& use_global_rsv
)
5259 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
, num_bytes
);
5262 if (*qgroup_reserved
)
5263 btrfs_qgroup_free(root
, *qgroup_reserved
);
5269 void btrfs_subvolume_release_metadata(struct btrfs_root
*root
,
5270 struct btrfs_block_rsv
*rsv
,
5271 u64 qgroup_reserved
)
5273 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
5277 * drop_outstanding_extent - drop an outstanding extent
5278 * @inode: the inode we're dropping the extent for
5279 * @num_bytes: the number of bytes we're relaseing.
5281 * This is called when we are freeing up an outstanding extent, either called
5282 * after an error or after an extent is written. This will return the number of
5283 * reserved extents that need to be freed. This must be called with
5284 * BTRFS_I(inode)->lock held.
5286 static unsigned drop_outstanding_extent(struct inode
*inode
, u64 num_bytes
)
5288 unsigned drop_inode_space
= 0;
5289 unsigned dropped_extents
= 0;
5290 unsigned num_extents
= 0;
5292 num_extents
= (unsigned)div64_u64(num_bytes
+
5293 BTRFS_MAX_EXTENT_SIZE
- 1,
5294 BTRFS_MAX_EXTENT_SIZE
);
5295 ASSERT(num_extents
);
5296 ASSERT(BTRFS_I(inode
)->outstanding_extents
>= num_extents
);
5297 BTRFS_I(inode
)->outstanding_extents
-= num_extents
;
5299 if (BTRFS_I(inode
)->outstanding_extents
== 0 &&
5300 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5301 &BTRFS_I(inode
)->runtime_flags
))
5302 drop_inode_space
= 1;
5305 * If we have more or the same amount of outsanding extents than we have
5306 * reserved then we need to leave the reserved extents count alone.
5308 if (BTRFS_I(inode
)->outstanding_extents
>=
5309 BTRFS_I(inode
)->reserved_extents
)
5310 return drop_inode_space
;
5312 dropped_extents
= BTRFS_I(inode
)->reserved_extents
-
5313 BTRFS_I(inode
)->outstanding_extents
;
5314 BTRFS_I(inode
)->reserved_extents
-= dropped_extents
;
5315 return dropped_extents
+ drop_inode_space
;
5319 * calc_csum_metadata_size - return the amount of metada space that must be
5320 * reserved/free'd for the given bytes.
5321 * @inode: the inode we're manipulating
5322 * @num_bytes: the number of bytes in question
5323 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5325 * This adjusts the number of csum_bytes in the inode and then returns the
5326 * correct amount of metadata that must either be reserved or freed. We
5327 * calculate how many checksums we can fit into one leaf and then divide the
5328 * number of bytes that will need to be checksumed by this value to figure out
5329 * how many checksums will be required. If we are adding bytes then the number
5330 * may go up and we will return the number of additional bytes that must be
5331 * reserved. If it is going down we will return the number of bytes that must
5334 * This must be called with BTRFS_I(inode)->lock held.
5336 static u64
calc_csum_metadata_size(struct inode
*inode
, u64 num_bytes
,
5339 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5340 u64 old_csums
, num_csums
;
5342 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
&&
5343 BTRFS_I(inode
)->csum_bytes
== 0)
5346 old_csums
= btrfs_csum_bytes_to_leaves(root
, BTRFS_I(inode
)->csum_bytes
);
5348 BTRFS_I(inode
)->csum_bytes
+= num_bytes
;
5350 BTRFS_I(inode
)->csum_bytes
-= num_bytes
;
5351 num_csums
= btrfs_csum_bytes_to_leaves(root
, BTRFS_I(inode
)->csum_bytes
);
5353 /* No change, no need to reserve more */
5354 if (old_csums
== num_csums
)
5358 return btrfs_calc_trans_metadata_size(root
,
5359 num_csums
- old_csums
);
5361 return btrfs_calc_trans_metadata_size(root
, old_csums
- num_csums
);
5364 int btrfs_delalloc_reserve_metadata(struct inode
*inode
, u64 num_bytes
)
5366 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5367 struct btrfs_block_rsv
*block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
5370 unsigned nr_extents
= 0;
5371 int extra_reserve
= 0;
5372 enum btrfs_reserve_flush_enum flush
= BTRFS_RESERVE_FLUSH_ALL
;
5374 bool delalloc_lock
= true;
5378 /* If we are a free space inode we need to not flush since we will be in
5379 * the middle of a transaction commit. We also don't need the delalloc
5380 * mutex since we won't race with anybody. We need this mostly to make
5381 * lockdep shut its filthy mouth.
5383 if (btrfs_is_free_space_inode(inode
)) {
5384 flush
= BTRFS_RESERVE_NO_FLUSH
;
5385 delalloc_lock
= false;
5388 if (flush
!= BTRFS_RESERVE_NO_FLUSH
&&
5389 btrfs_transaction_in_commit(root
->fs_info
))
5390 schedule_timeout(1);
5393 mutex_lock(&BTRFS_I(inode
)->delalloc_mutex
);
5395 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
5397 spin_lock(&BTRFS_I(inode
)->lock
);
5398 nr_extents
= (unsigned)div64_u64(num_bytes
+
5399 BTRFS_MAX_EXTENT_SIZE
- 1,
5400 BTRFS_MAX_EXTENT_SIZE
);
5401 BTRFS_I(inode
)->outstanding_extents
+= nr_extents
;
5404 if (BTRFS_I(inode
)->outstanding_extents
>
5405 BTRFS_I(inode
)->reserved_extents
)
5406 nr_extents
= BTRFS_I(inode
)->outstanding_extents
-
5407 BTRFS_I(inode
)->reserved_extents
;
5410 * Add an item to reserve for updating the inode when we complete the
5413 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5414 &BTRFS_I(inode
)->runtime_flags
)) {
5419 to_reserve
= btrfs_calc_trans_metadata_size(root
, nr_extents
);
5420 to_reserve
+= calc_csum_metadata_size(inode
, num_bytes
, 1);
5421 csum_bytes
= BTRFS_I(inode
)->csum_bytes
;
5422 spin_unlock(&BTRFS_I(inode
)->lock
);
5424 if (root
->fs_info
->quota_enabled
) {
5425 ret
= btrfs_qgroup_reserve(root
, nr_extents
* root
->nodesize
);
5430 ret
= reserve_metadata_bytes(root
, block_rsv
, to_reserve
, flush
);
5431 if (unlikely(ret
)) {
5432 if (root
->fs_info
->quota_enabled
)
5433 btrfs_qgroup_free(root
, nr_extents
* root
->nodesize
);
5437 spin_lock(&BTRFS_I(inode
)->lock
);
5438 if (extra_reserve
) {
5439 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5440 &BTRFS_I(inode
)->runtime_flags
);
5443 BTRFS_I(inode
)->reserved_extents
+= nr_extents
;
5444 spin_unlock(&BTRFS_I(inode
)->lock
);
5447 mutex_unlock(&BTRFS_I(inode
)->delalloc_mutex
);
5450 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5451 btrfs_ino(inode
), to_reserve
, 1);
5452 block_rsv_add_bytes(block_rsv
, to_reserve
, 1);
5457 spin_lock(&BTRFS_I(inode
)->lock
);
5458 dropped
= drop_outstanding_extent(inode
, num_bytes
);
5460 * If the inodes csum_bytes is the same as the original
5461 * csum_bytes then we know we haven't raced with any free()ers
5462 * so we can just reduce our inodes csum bytes and carry on.
5464 if (BTRFS_I(inode
)->csum_bytes
== csum_bytes
) {
5465 calc_csum_metadata_size(inode
, num_bytes
, 0);
5467 u64 orig_csum_bytes
= BTRFS_I(inode
)->csum_bytes
;
5471 * This is tricky, but first we need to figure out how much we
5472 * free'd from any free-ers that occured during this
5473 * reservation, so we reset ->csum_bytes to the csum_bytes
5474 * before we dropped our lock, and then call the free for the
5475 * number of bytes that were freed while we were trying our
5478 bytes
= csum_bytes
- BTRFS_I(inode
)->csum_bytes
;
5479 BTRFS_I(inode
)->csum_bytes
= csum_bytes
;
5480 to_free
= calc_csum_metadata_size(inode
, bytes
, 0);
5484 * Now we need to see how much we would have freed had we not
5485 * been making this reservation and our ->csum_bytes were not
5486 * artificially inflated.
5488 BTRFS_I(inode
)->csum_bytes
= csum_bytes
- num_bytes
;
5489 bytes
= csum_bytes
- orig_csum_bytes
;
5490 bytes
= calc_csum_metadata_size(inode
, bytes
, 0);
5493 * Now reset ->csum_bytes to what it should be. If bytes is
5494 * more than to_free then we would have free'd more space had we
5495 * not had an artificially high ->csum_bytes, so we need to free
5496 * the remainder. If bytes is the same or less then we don't
5497 * need to do anything, the other free-ers did the correct
5500 BTRFS_I(inode
)->csum_bytes
= orig_csum_bytes
- num_bytes
;
5501 if (bytes
> to_free
)
5502 to_free
= bytes
- to_free
;
5506 spin_unlock(&BTRFS_I(inode
)->lock
);
5508 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
5511 btrfs_block_rsv_release(root
, block_rsv
, to_free
);
5512 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5513 btrfs_ino(inode
), to_free
, 0);
5516 mutex_unlock(&BTRFS_I(inode
)->delalloc_mutex
);
5521 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5522 * @inode: the inode to release the reservation for
5523 * @num_bytes: the number of bytes we're releasing
5525 * This will release the metadata reservation for an inode. This can be called
5526 * once we complete IO for a given set of bytes to release their metadata
5529 void btrfs_delalloc_release_metadata(struct inode
*inode
, u64 num_bytes
)
5531 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5535 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
5536 spin_lock(&BTRFS_I(inode
)->lock
);
5537 dropped
= drop_outstanding_extent(inode
, num_bytes
);
5540 to_free
= calc_csum_metadata_size(inode
, num_bytes
, 0);
5541 spin_unlock(&BTRFS_I(inode
)->lock
);
5543 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
5545 if (btrfs_test_is_dummy_root(root
))
5548 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5549 btrfs_ino(inode
), to_free
, 0);
5551 btrfs_block_rsv_release(root
, &root
->fs_info
->delalloc_block_rsv
,
5556 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5557 * @inode: inode we're writing to
5558 * @num_bytes: the number of bytes we want to allocate
5560 * This will do the following things
5562 * o reserve space in the data space info for num_bytes
5563 * o reserve space in the metadata space info based on number of outstanding
5564 * extents and how much csums will be needed
5565 * o add to the inodes ->delalloc_bytes
5566 * o add it to the fs_info's delalloc inodes list.
5568 * This will return 0 for success and -ENOSPC if there is no space left.
5570 int btrfs_delalloc_reserve_space(struct inode
*inode
, u64 num_bytes
)
5574 ret
= btrfs_check_data_free_space(inode
, num_bytes
, num_bytes
);
5578 ret
= btrfs_delalloc_reserve_metadata(inode
, num_bytes
);
5580 btrfs_free_reserved_data_space(inode
, num_bytes
);
5588 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5589 * @inode: inode we're releasing space for
5590 * @num_bytes: the number of bytes we want to free up
5592 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5593 * called in the case that we don't need the metadata AND data reservations
5594 * anymore. So if there is an error or we insert an inline extent.
5596 * This function will release the metadata space that was not used and will
5597 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5598 * list if there are no delalloc bytes left.
5600 void btrfs_delalloc_release_space(struct inode
*inode
, u64 num_bytes
)
5602 btrfs_delalloc_release_metadata(inode
, num_bytes
);
5603 btrfs_free_reserved_data_space(inode
, num_bytes
);
5606 static int update_block_group(struct btrfs_trans_handle
*trans
,
5607 struct btrfs_root
*root
, u64 bytenr
,
5608 u64 num_bytes
, int alloc
)
5610 struct btrfs_block_group_cache
*cache
= NULL
;
5611 struct btrfs_fs_info
*info
= root
->fs_info
;
5612 u64 total
= num_bytes
;
5617 /* block accounting for super block */
5618 spin_lock(&info
->delalloc_root_lock
);
5619 old_val
= btrfs_super_bytes_used(info
->super_copy
);
5621 old_val
+= num_bytes
;
5623 old_val
-= num_bytes
;
5624 btrfs_set_super_bytes_used(info
->super_copy
, old_val
);
5625 spin_unlock(&info
->delalloc_root_lock
);
5628 cache
= btrfs_lookup_block_group(info
, bytenr
);
5631 if (cache
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
5632 BTRFS_BLOCK_GROUP_RAID1
|
5633 BTRFS_BLOCK_GROUP_RAID10
))
5638 * If this block group has free space cache written out, we
5639 * need to make sure to load it if we are removing space. This
5640 * is because we need the unpinning stage to actually add the
5641 * space back to the block group, otherwise we will leak space.
5643 if (!alloc
&& cache
->cached
== BTRFS_CACHE_NO
)
5644 cache_block_group(cache
, 1);
5646 byte_in_group
= bytenr
- cache
->key
.objectid
;
5647 WARN_ON(byte_in_group
> cache
->key
.offset
);
5649 spin_lock(&cache
->space_info
->lock
);
5650 spin_lock(&cache
->lock
);
5652 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
5653 cache
->disk_cache_state
< BTRFS_DC_CLEAR
)
5654 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
5656 old_val
= btrfs_block_group_used(&cache
->item
);
5657 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
5659 old_val
+= num_bytes
;
5660 btrfs_set_block_group_used(&cache
->item
, old_val
);
5661 cache
->reserved
-= num_bytes
;
5662 cache
->space_info
->bytes_reserved
-= num_bytes
;
5663 cache
->space_info
->bytes_used
+= num_bytes
;
5664 cache
->space_info
->disk_used
+= num_bytes
* factor
;
5665 spin_unlock(&cache
->lock
);
5666 spin_unlock(&cache
->space_info
->lock
);
5668 old_val
-= num_bytes
;
5669 btrfs_set_block_group_used(&cache
->item
, old_val
);
5670 cache
->pinned
+= num_bytes
;
5671 cache
->space_info
->bytes_pinned
+= num_bytes
;
5672 cache
->space_info
->bytes_used
-= num_bytes
;
5673 cache
->space_info
->disk_used
-= num_bytes
* factor
;
5674 spin_unlock(&cache
->lock
);
5675 spin_unlock(&cache
->space_info
->lock
);
5677 set_extent_dirty(info
->pinned_extents
,
5678 bytenr
, bytenr
+ num_bytes
- 1,
5679 GFP_NOFS
| __GFP_NOFAIL
);
5681 * No longer have used bytes in this block group, queue
5685 spin_lock(&info
->unused_bgs_lock
);
5686 if (list_empty(&cache
->bg_list
)) {
5687 btrfs_get_block_group(cache
);
5688 list_add_tail(&cache
->bg_list
,
5691 spin_unlock(&info
->unused_bgs_lock
);
5695 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
5696 if (list_empty(&cache
->dirty_list
)) {
5697 list_add_tail(&cache
->dirty_list
,
5698 &trans
->transaction
->dirty_bgs
);
5699 trans
->transaction
->num_dirty_bgs
++;
5700 btrfs_get_block_group(cache
);
5702 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
5704 btrfs_put_block_group(cache
);
5706 bytenr
+= num_bytes
;
5711 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
5713 struct btrfs_block_group_cache
*cache
;
5716 spin_lock(&root
->fs_info
->block_group_cache_lock
);
5717 bytenr
= root
->fs_info
->first_logical_byte
;
5718 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
5720 if (bytenr
< (u64
)-1)
5723 cache
= btrfs_lookup_first_block_group(root
->fs_info
, search_start
);
5727 bytenr
= cache
->key
.objectid
;
5728 btrfs_put_block_group(cache
);
5733 static int pin_down_extent(struct btrfs_root
*root
,
5734 struct btrfs_block_group_cache
*cache
,
5735 u64 bytenr
, u64 num_bytes
, int reserved
)
5737 spin_lock(&cache
->space_info
->lock
);
5738 spin_lock(&cache
->lock
);
5739 cache
->pinned
+= num_bytes
;
5740 cache
->space_info
->bytes_pinned
+= num_bytes
;
5742 cache
->reserved
-= num_bytes
;
5743 cache
->space_info
->bytes_reserved
-= num_bytes
;
5745 spin_unlock(&cache
->lock
);
5746 spin_unlock(&cache
->space_info
->lock
);
5748 set_extent_dirty(root
->fs_info
->pinned_extents
, bytenr
,
5749 bytenr
+ num_bytes
- 1, GFP_NOFS
| __GFP_NOFAIL
);
5751 trace_btrfs_reserved_extent_free(root
, bytenr
, num_bytes
);
5756 * this function must be called within transaction
5758 int btrfs_pin_extent(struct btrfs_root
*root
,
5759 u64 bytenr
, u64 num_bytes
, int reserved
)
5761 struct btrfs_block_group_cache
*cache
;
5763 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
5764 BUG_ON(!cache
); /* Logic error */
5766 pin_down_extent(root
, cache
, bytenr
, num_bytes
, reserved
);
5768 btrfs_put_block_group(cache
);
5773 * this function must be called within transaction
5775 int btrfs_pin_extent_for_log_replay(struct btrfs_root
*root
,
5776 u64 bytenr
, u64 num_bytes
)
5778 struct btrfs_block_group_cache
*cache
;
5781 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
5786 * pull in the free space cache (if any) so that our pin
5787 * removes the free space from the cache. We have load_only set
5788 * to one because the slow code to read in the free extents does check
5789 * the pinned extents.
5791 cache_block_group(cache
, 1);
5793 pin_down_extent(root
, cache
, bytenr
, num_bytes
, 0);
5795 /* remove us from the free space cache (if we're there at all) */
5796 ret
= btrfs_remove_free_space(cache
, bytenr
, num_bytes
);
5797 btrfs_put_block_group(cache
);
5801 static int __exclude_logged_extent(struct btrfs_root
*root
, u64 start
, u64 num_bytes
)
5804 struct btrfs_block_group_cache
*block_group
;
5805 struct btrfs_caching_control
*caching_ctl
;
5807 block_group
= btrfs_lookup_block_group(root
->fs_info
, start
);
5811 cache_block_group(block_group
, 0);
5812 caching_ctl
= get_caching_control(block_group
);
5816 BUG_ON(!block_group_cache_done(block_group
));
5817 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
5819 mutex_lock(&caching_ctl
->mutex
);
5821 if (start
>= caching_ctl
->progress
) {
5822 ret
= add_excluded_extent(root
, start
, num_bytes
);
5823 } else if (start
+ num_bytes
<= caching_ctl
->progress
) {
5824 ret
= btrfs_remove_free_space(block_group
,
5827 num_bytes
= caching_ctl
->progress
- start
;
5828 ret
= btrfs_remove_free_space(block_group
,
5833 num_bytes
= (start
+ num_bytes
) -
5834 caching_ctl
->progress
;
5835 start
= caching_ctl
->progress
;
5836 ret
= add_excluded_extent(root
, start
, num_bytes
);
5839 mutex_unlock(&caching_ctl
->mutex
);
5840 put_caching_control(caching_ctl
);
5842 btrfs_put_block_group(block_group
);
5846 int btrfs_exclude_logged_extents(struct btrfs_root
*log
,
5847 struct extent_buffer
*eb
)
5849 struct btrfs_file_extent_item
*item
;
5850 struct btrfs_key key
;
5854 if (!btrfs_fs_incompat(log
->fs_info
, MIXED_GROUPS
))
5857 for (i
= 0; i
< btrfs_header_nritems(eb
); i
++) {
5858 btrfs_item_key_to_cpu(eb
, &key
, i
);
5859 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
5861 item
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
5862 found_type
= btrfs_file_extent_type(eb
, item
);
5863 if (found_type
== BTRFS_FILE_EXTENT_INLINE
)
5865 if (btrfs_file_extent_disk_bytenr(eb
, item
) == 0)
5867 key
.objectid
= btrfs_file_extent_disk_bytenr(eb
, item
);
5868 key
.offset
= btrfs_file_extent_disk_num_bytes(eb
, item
);
5869 __exclude_logged_extent(log
, key
.objectid
, key
.offset
);
5876 * btrfs_update_reserved_bytes - update the block_group and space info counters
5877 * @cache: The cache we are manipulating
5878 * @num_bytes: The number of bytes in question
5879 * @reserve: One of the reservation enums
5880 * @delalloc: The blocks are allocated for the delalloc write
5882 * This is called by the allocator when it reserves space, or by somebody who is
5883 * freeing space that was never actually used on disk. For example if you
5884 * reserve some space for a new leaf in transaction A and before transaction A
5885 * commits you free that leaf, you call this with reserve set to 0 in order to
5886 * clear the reservation.
5888 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5889 * ENOSPC accounting. For data we handle the reservation through clearing the
5890 * delalloc bits in the io_tree. We have to do this since we could end up
5891 * allocating less disk space for the amount of data we have reserved in the
5892 * case of compression.
5894 * If this is a reservation and the block group has become read only we cannot
5895 * make the reservation and return -EAGAIN, otherwise this function always
5898 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
5899 u64 num_bytes
, int reserve
, int delalloc
)
5901 struct btrfs_space_info
*space_info
= cache
->space_info
;
5904 spin_lock(&space_info
->lock
);
5905 spin_lock(&cache
->lock
);
5906 if (reserve
!= RESERVE_FREE
) {
5910 cache
->reserved
+= num_bytes
;
5911 space_info
->bytes_reserved
+= num_bytes
;
5912 if (reserve
== RESERVE_ALLOC
) {
5913 trace_btrfs_space_reservation(cache
->fs_info
,
5914 "space_info", space_info
->flags
,
5916 space_info
->bytes_may_use
-= num_bytes
;
5920 cache
->delalloc_bytes
+= num_bytes
;
5924 space_info
->bytes_readonly
+= num_bytes
;
5925 cache
->reserved
-= num_bytes
;
5926 space_info
->bytes_reserved
-= num_bytes
;
5929 cache
->delalloc_bytes
-= num_bytes
;
5931 spin_unlock(&cache
->lock
);
5932 spin_unlock(&space_info
->lock
);
5936 void btrfs_prepare_extent_commit(struct btrfs_trans_handle
*trans
,
5937 struct btrfs_root
*root
)
5939 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5940 struct btrfs_caching_control
*next
;
5941 struct btrfs_caching_control
*caching_ctl
;
5942 struct btrfs_block_group_cache
*cache
;
5944 down_write(&fs_info
->commit_root_sem
);
5946 list_for_each_entry_safe(caching_ctl
, next
,
5947 &fs_info
->caching_block_groups
, list
) {
5948 cache
= caching_ctl
->block_group
;
5949 if (block_group_cache_done(cache
)) {
5950 cache
->last_byte_to_unpin
= (u64
)-1;
5951 list_del_init(&caching_ctl
->list
);
5952 put_caching_control(caching_ctl
);
5954 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
5958 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
5959 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
5961 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
5963 up_write(&fs_info
->commit_root_sem
);
5965 update_global_block_rsv(fs_info
);
5968 static int unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
,
5969 const bool return_free_space
)
5971 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5972 struct btrfs_block_group_cache
*cache
= NULL
;
5973 struct btrfs_space_info
*space_info
;
5974 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5978 while (start
<= end
) {
5981 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
5983 btrfs_put_block_group(cache
);
5984 cache
= btrfs_lookup_block_group(fs_info
, start
);
5985 BUG_ON(!cache
); /* Logic error */
5988 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
5989 len
= min(len
, end
+ 1 - start
);
5991 if (start
< cache
->last_byte_to_unpin
) {
5992 len
= min(len
, cache
->last_byte_to_unpin
- start
);
5993 if (return_free_space
)
5994 btrfs_add_free_space(cache
, start
, len
);
5998 space_info
= cache
->space_info
;
6000 spin_lock(&space_info
->lock
);
6001 spin_lock(&cache
->lock
);
6002 cache
->pinned
-= len
;
6003 space_info
->bytes_pinned
-= len
;
6004 percpu_counter_add(&space_info
->total_bytes_pinned
, -len
);
6006 space_info
->bytes_readonly
+= len
;
6009 spin_unlock(&cache
->lock
);
6010 if (!readonly
&& global_rsv
->space_info
== space_info
) {
6011 spin_lock(&global_rsv
->lock
);
6012 if (!global_rsv
->full
) {
6013 len
= min(len
, global_rsv
->size
-
6014 global_rsv
->reserved
);
6015 global_rsv
->reserved
+= len
;
6016 space_info
->bytes_may_use
+= len
;
6017 if (global_rsv
->reserved
>= global_rsv
->size
)
6018 global_rsv
->full
= 1;
6020 spin_unlock(&global_rsv
->lock
);
6022 spin_unlock(&space_info
->lock
);
6026 btrfs_put_block_group(cache
);
6030 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
6031 struct btrfs_root
*root
)
6033 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6034 struct extent_io_tree
*unpin
;
6042 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
6043 unpin
= &fs_info
->freed_extents
[1];
6045 unpin
= &fs_info
->freed_extents
[0];
6048 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
6049 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
6050 EXTENT_DIRTY
, NULL
);
6052 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
6056 if (btrfs_test_opt(root
, DISCARD
))
6057 ret
= btrfs_discard_extent(root
, start
,
6058 end
+ 1 - start
, NULL
);
6060 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
6061 unpin_extent_range(root
, start
, end
, true);
6062 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
6069 static void add_pinned_bytes(struct btrfs_fs_info
*fs_info
, u64 num_bytes
,
6070 u64 owner
, u64 root_objectid
)
6072 struct btrfs_space_info
*space_info
;
6075 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
6076 if (root_objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
6077 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
6079 flags
= BTRFS_BLOCK_GROUP_METADATA
;
6081 flags
= BTRFS_BLOCK_GROUP_DATA
;
6084 space_info
= __find_space_info(fs_info
, flags
);
6085 BUG_ON(!space_info
); /* Logic bug */
6086 percpu_counter_add(&space_info
->total_bytes_pinned
, num_bytes
);
6090 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
6091 struct btrfs_root
*root
,
6092 struct btrfs_delayed_ref_node
*node
, u64 parent
,
6093 u64 root_objectid
, u64 owner_objectid
,
6094 u64 owner_offset
, int refs_to_drop
,
6095 struct btrfs_delayed_extent_op
*extent_op
)
6097 struct btrfs_key key
;
6098 struct btrfs_path
*path
;
6099 struct btrfs_fs_info
*info
= root
->fs_info
;
6100 struct btrfs_root
*extent_root
= info
->extent_root
;
6101 struct extent_buffer
*leaf
;
6102 struct btrfs_extent_item
*ei
;
6103 struct btrfs_extent_inline_ref
*iref
;
6106 int extent_slot
= 0;
6107 int found_extent
= 0;
6109 int no_quota
= node
->no_quota
;
6112 u64 bytenr
= node
->bytenr
;
6113 u64 num_bytes
= node
->num_bytes
;
6115 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
6118 if (!info
->quota_enabled
|| !is_fstree(root_objectid
))
6121 path
= btrfs_alloc_path();
6126 path
->leave_spinning
= 1;
6128 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
6129 BUG_ON(!is_data
&& refs_to_drop
!= 1);
6132 skinny_metadata
= 0;
6134 ret
= lookup_extent_backref(trans
, extent_root
, path
, &iref
,
6135 bytenr
, num_bytes
, parent
,
6136 root_objectid
, owner_objectid
,
6139 extent_slot
= path
->slots
[0];
6140 while (extent_slot
>= 0) {
6141 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
6143 if (key
.objectid
!= bytenr
)
6145 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
6146 key
.offset
== num_bytes
) {
6150 if (key
.type
== BTRFS_METADATA_ITEM_KEY
&&
6151 key
.offset
== owner_objectid
) {
6155 if (path
->slots
[0] - extent_slot
> 5)
6159 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6160 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
6161 if (found_extent
&& item_size
< sizeof(*ei
))
6164 if (!found_extent
) {
6166 ret
= remove_extent_backref(trans
, extent_root
, path
,
6168 is_data
, &last_ref
);
6170 btrfs_abort_transaction(trans
, extent_root
, ret
);
6173 btrfs_release_path(path
);
6174 path
->leave_spinning
= 1;
6176 key
.objectid
= bytenr
;
6177 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6178 key
.offset
= num_bytes
;
6180 if (!is_data
&& skinny_metadata
) {
6181 key
.type
= BTRFS_METADATA_ITEM_KEY
;
6182 key
.offset
= owner_objectid
;
6185 ret
= btrfs_search_slot(trans
, extent_root
,
6187 if (ret
> 0 && skinny_metadata
&& path
->slots
[0]) {
6189 * Couldn't find our skinny metadata item,
6190 * see if we have ye olde extent item.
6193 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
6195 if (key
.objectid
== bytenr
&&
6196 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
6197 key
.offset
== num_bytes
)
6201 if (ret
> 0 && skinny_metadata
) {
6202 skinny_metadata
= false;
6203 key
.objectid
= bytenr
;
6204 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6205 key
.offset
= num_bytes
;
6206 btrfs_release_path(path
);
6207 ret
= btrfs_search_slot(trans
, extent_root
,
6212 btrfs_err(info
, "umm, got %d back from search, was looking for %llu",
6215 btrfs_print_leaf(extent_root
,
6219 btrfs_abort_transaction(trans
, extent_root
, ret
);
6222 extent_slot
= path
->slots
[0];
6224 } else if (WARN_ON(ret
== -ENOENT
)) {
6225 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
6227 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6228 bytenr
, parent
, root_objectid
, owner_objectid
,
6230 btrfs_abort_transaction(trans
, extent_root
, ret
);
6233 btrfs_abort_transaction(trans
, extent_root
, ret
);
6237 leaf
= path
->nodes
[0];
6238 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
6239 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6240 if (item_size
< sizeof(*ei
)) {
6241 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
6242 ret
= convert_extent_item_v0(trans
, extent_root
, path
,
6245 btrfs_abort_transaction(trans
, extent_root
, ret
);
6249 btrfs_release_path(path
);
6250 path
->leave_spinning
= 1;
6252 key
.objectid
= bytenr
;
6253 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6254 key
.offset
= num_bytes
;
6256 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
6259 btrfs_err(info
, "umm, got %d back from search, was looking for %llu",
6261 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
6264 btrfs_abort_transaction(trans
, extent_root
, ret
);
6268 extent_slot
= path
->slots
[0];
6269 leaf
= path
->nodes
[0];
6270 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
6273 BUG_ON(item_size
< sizeof(*ei
));
6274 ei
= btrfs_item_ptr(leaf
, extent_slot
,
6275 struct btrfs_extent_item
);
6276 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
6277 key
.type
== BTRFS_EXTENT_ITEM_KEY
) {
6278 struct btrfs_tree_block_info
*bi
;
6279 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
6280 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
6281 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
6284 refs
= btrfs_extent_refs(leaf
, ei
);
6285 if (refs
< refs_to_drop
) {
6286 btrfs_err(info
, "trying to drop %d refs but we only have %Lu "
6287 "for bytenr %Lu", refs_to_drop
, refs
, bytenr
);
6289 btrfs_abort_transaction(trans
, extent_root
, ret
);
6292 refs
-= refs_to_drop
;
6296 __run_delayed_extent_op(extent_op
, leaf
, ei
);
6298 * In the case of inline back ref, reference count will
6299 * be updated by remove_extent_backref
6302 BUG_ON(!found_extent
);
6304 btrfs_set_extent_refs(leaf
, ei
, refs
);
6305 btrfs_mark_buffer_dirty(leaf
);
6308 ret
= remove_extent_backref(trans
, extent_root
, path
,
6310 is_data
, &last_ref
);
6312 btrfs_abort_transaction(trans
, extent_root
, ret
);
6316 add_pinned_bytes(root
->fs_info
, -num_bytes
, owner_objectid
,
6320 BUG_ON(is_data
&& refs_to_drop
!=
6321 extent_data_ref_count(root
, path
, iref
));
6323 BUG_ON(path
->slots
[0] != extent_slot
);
6325 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
6326 path
->slots
[0] = extent_slot
;
6332 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
6335 btrfs_abort_transaction(trans
, extent_root
, ret
);
6338 btrfs_release_path(path
);
6341 ret
= btrfs_del_csums(trans
, root
, bytenr
, num_bytes
);
6343 btrfs_abort_transaction(trans
, extent_root
, ret
);
6348 ret
= update_block_group(trans
, root
, bytenr
, num_bytes
, 0);
6350 btrfs_abort_transaction(trans
, extent_root
, ret
);
6354 btrfs_release_path(path
);
6357 btrfs_free_path(path
);
6362 * when we free an block, it is possible (and likely) that we free the last
6363 * delayed ref for that extent as well. This searches the delayed ref tree for
6364 * a given extent, and if there are no other delayed refs to be processed, it
6365 * removes it from the tree.
6367 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
6368 struct btrfs_root
*root
, u64 bytenr
)
6370 struct btrfs_delayed_ref_head
*head
;
6371 struct btrfs_delayed_ref_root
*delayed_refs
;
6374 delayed_refs
= &trans
->transaction
->delayed_refs
;
6375 spin_lock(&delayed_refs
->lock
);
6376 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
6378 goto out_delayed_unlock
;
6380 spin_lock(&head
->lock
);
6381 if (!list_empty(&head
->ref_list
))
6384 if (head
->extent_op
) {
6385 if (!head
->must_insert_reserved
)
6387 btrfs_free_delayed_extent_op(head
->extent_op
);
6388 head
->extent_op
= NULL
;
6392 * waiting for the lock here would deadlock. If someone else has it
6393 * locked they are already in the process of dropping it anyway
6395 if (!mutex_trylock(&head
->mutex
))
6399 * at this point we have a head with no other entries. Go
6400 * ahead and process it.
6402 head
->node
.in_tree
= 0;
6403 rb_erase(&head
->href_node
, &delayed_refs
->href_root
);
6405 atomic_dec(&delayed_refs
->num_entries
);
6408 * we don't take a ref on the node because we're removing it from the
6409 * tree, so we just steal the ref the tree was holding.
6411 delayed_refs
->num_heads
--;
6412 if (head
->processing
== 0)
6413 delayed_refs
->num_heads_ready
--;
6414 head
->processing
= 0;
6415 spin_unlock(&head
->lock
);
6416 spin_unlock(&delayed_refs
->lock
);
6418 BUG_ON(head
->extent_op
);
6419 if (head
->must_insert_reserved
)
6422 mutex_unlock(&head
->mutex
);
6423 btrfs_put_delayed_ref(&head
->node
);
6426 spin_unlock(&head
->lock
);
6429 spin_unlock(&delayed_refs
->lock
);
6433 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
6434 struct btrfs_root
*root
,
6435 struct extent_buffer
*buf
,
6436 u64 parent
, int last_ref
)
6441 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
6442 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
6443 buf
->start
, buf
->len
,
6444 parent
, root
->root_key
.objectid
,
6445 btrfs_header_level(buf
),
6446 BTRFS_DROP_DELAYED_REF
, NULL
, 0);
6447 BUG_ON(ret
); /* -ENOMEM */
6453 if (btrfs_header_generation(buf
) == trans
->transid
) {
6454 struct btrfs_block_group_cache
*cache
;
6456 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
6457 ret
= check_ref_cleanup(trans
, root
, buf
->start
);
6462 cache
= btrfs_lookup_block_group(root
->fs_info
, buf
->start
);
6464 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
6465 pin_down_extent(root
, cache
, buf
->start
, buf
->len
, 1);
6466 btrfs_put_block_group(cache
);
6470 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
6472 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
6473 btrfs_update_reserved_bytes(cache
, buf
->len
, RESERVE_FREE
, 0);
6474 btrfs_put_block_group(cache
);
6475 trace_btrfs_reserved_extent_free(root
, buf
->start
, buf
->len
);
6480 add_pinned_bytes(root
->fs_info
, buf
->len
,
6481 btrfs_header_level(buf
),
6482 root
->root_key
.objectid
);
6485 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6488 clear_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
);
6491 /* Can return -ENOMEM */
6492 int btrfs_free_extent(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
6493 u64 bytenr
, u64 num_bytes
, u64 parent
, u64 root_objectid
,
6494 u64 owner
, u64 offset
, int no_quota
)
6497 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6499 if (btrfs_test_is_dummy_root(root
))
6502 add_pinned_bytes(root
->fs_info
, num_bytes
, owner
, root_objectid
);
6505 * tree log blocks never actually go into the extent allocation
6506 * tree, just update pinning info and exit early.
6508 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
6509 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
6510 /* unlocks the pinned mutex */
6511 btrfs_pin_extent(root
, bytenr
, num_bytes
, 1);
6513 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
6514 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
6516 parent
, root_objectid
, (int)owner
,
6517 BTRFS_DROP_DELAYED_REF
, NULL
, no_quota
);
6519 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
6521 parent
, root_objectid
, owner
,
6522 offset
, BTRFS_DROP_DELAYED_REF
,
6529 * when we wait for progress in the block group caching, its because
6530 * our allocation attempt failed at least once. So, we must sleep
6531 * and let some progress happen before we try again.
6533 * This function will sleep at least once waiting for new free space to
6534 * show up, and then it will check the block group free space numbers
6535 * for our min num_bytes. Another option is to have it go ahead
6536 * and look in the rbtree for a free extent of a given size, but this
6539 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6540 * any of the information in this block group.
6542 static noinline
void
6543 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
6546 struct btrfs_caching_control
*caching_ctl
;
6548 caching_ctl
= get_caching_control(cache
);
6552 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
6553 (cache
->free_space_ctl
->free_space
>= num_bytes
));
6555 put_caching_control(caching_ctl
);
6559 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
6561 struct btrfs_caching_control
*caching_ctl
;
6564 caching_ctl
= get_caching_control(cache
);
6566 return (cache
->cached
== BTRFS_CACHE_ERROR
) ? -EIO
: 0;
6568 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
6569 if (cache
->cached
== BTRFS_CACHE_ERROR
)
6571 put_caching_control(caching_ctl
);
6575 int __get_raid_index(u64 flags
)
6577 if (flags
& BTRFS_BLOCK_GROUP_RAID10
)
6578 return BTRFS_RAID_RAID10
;
6579 else if (flags
& BTRFS_BLOCK_GROUP_RAID1
)
6580 return BTRFS_RAID_RAID1
;
6581 else if (flags
& BTRFS_BLOCK_GROUP_DUP
)
6582 return BTRFS_RAID_DUP
;
6583 else if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
6584 return BTRFS_RAID_RAID0
;
6585 else if (flags
& BTRFS_BLOCK_GROUP_RAID5
)
6586 return BTRFS_RAID_RAID5
;
6587 else if (flags
& BTRFS_BLOCK_GROUP_RAID6
)
6588 return BTRFS_RAID_RAID6
;
6590 return BTRFS_RAID_SINGLE
; /* BTRFS_BLOCK_GROUP_SINGLE */
6593 int get_block_group_index(struct btrfs_block_group_cache
*cache
)
6595 return __get_raid_index(cache
->flags
);
6598 static const char *btrfs_raid_type_names
[BTRFS_NR_RAID_TYPES
] = {
6599 [BTRFS_RAID_RAID10
] = "raid10",
6600 [BTRFS_RAID_RAID1
] = "raid1",
6601 [BTRFS_RAID_DUP
] = "dup",
6602 [BTRFS_RAID_RAID0
] = "raid0",
6603 [BTRFS_RAID_SINGLE
] = "single",
6604 [BTRFS_RAID_RAID5
] = "raid5",
6605 [BTRFS_RAID_RAID6
] = "raid6",
6608 static const char *get_raid_name(enum btrfs_raid_types type
)
6610 if (type
>= BTRFS_NR_RAID_TYPES
)
6613 return btrfs_raid_type_names
[type
];
6616 enum btrfs_loop_type
{
6617 LOOP_CACHING_NOWAIT
= 0,
6618 LOOP_CACHING_WAIT
= 1,
6619 LOOP_ALLOC_CHUNK
= 2,
6620 LOOP_NO_EMPTY_SIZE
= 3,
6624 btrfs_lock_block_group(struct btrfs_block_group_cache
*cache
,
6628 down_read(&cache
->data_rwsem
);
6632 btrfs_grab_block_group(struct btrfs_block_group_cache
*cache
,
6635 btrfs_get_block_group(cache
);
6637 down_read(&cache
->data_rwsem
);
6640 static struct btrfs_block_group_cache
*
6641 btrfs_lock_cluster(struct btrfs_block_group_cache
*block_group
,
6642 struct btrfs_free_cluster
*cluster
,
6645 struct btrfs_block_group_cache
*used_bg
;
6646 bool locked
= false;
6648 spin_lock(&cluster
->refill_lock
);
6650 if (used_bg
== cluster
->block_group
)
6653 up_read(&used_bg
->data_rwsem
);
6654 btrfs_put_block_group(used_bg
);
6657 used_bg
= cluster
->block_group
;
6661 if (used_bg
== block_group
)
6664 btrfs_get_block_group(used_bg
);
6669 if (down_read_trylock(&used_bg
->data_rwsem
))
6672 spin_unlock(&cluster
->refill_lock
);
6673 down_read(&used_bg
->data_rwsem
);
6679 btrfs_release_block_group(struct btrfs_block_group_cache
*cache
,
6683 up_read(&cache
->data_rwsem
);
6684 btrfs_put_block_group(cache
);
6688 * walks the btree of allocated extents and find a hole of a given size.
6689 * The key ins is changed to record the hole:
6690 * ins->objectid == start position
6691 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6692 * ins->offset == the size of the hole.
6693 * Any available blocks before search_start are skipped.
6695 * If there is no suitable free space, we will record the max size of
6696 * the free space extent currently.
6698 static noinline
int find_free_extent(struct btrfs_root
*orig_root
,
6699 u64 num_bytes
, u64 empty_size
,
6700 u64 hint_byte
, struct btrfs_key
*ins
,
6701 u64 flags
, int delalloc
)
6704 struct btrfs_root
*root
= orig_root
->fs_info
->extent_root
;
6705 struct btrfs_free_cluster
*last_ptr
= NULL
;
6706 struct btrfs_block_group_cache
*block_group
= NULL
;
6707 u64 search_start
= 0;
6708 u64 max_extent_size
= 0;
6709 int empty_cluster
= 2 * 1024 * 1024;
6710 struct btrfs_space_info
*space_info
;
6712 int index
= __get_raid_index(flags
);
6713 int alloc_type
= (flags
& BTRFS_BLOCK_GROUP_DATA
) ?
6714 RESERVE_ALLOC_NO_ACCOUNT
: RESERVE_ALLOC
;
6715 bool failed_cluster_refill
= false;
6716 bool failed_alloc
= false;
6717 bool use_cluster
= true;
6718 bool have_caching_bg
= false;
6720 WARN_ON(num_bytes
< root
->sectorsize
);
6721 ins
->type
= BTRFS_EXTENT_ITEM_KEY
;
6725 trace_find_free_extent(orig_root
, num_bytes
, empty_size
, flags
);
6727 space_info
= __find_space_info(root
->fs_info
, flags
);
6729 btrfs_err(root
->fs_info
, "No space info for %llu", flags
);
6734 * If the space info is for both data and metadata it means we have a
6735 * small filesystem and we can't use the clustering stuff.
6737 if (btrfs_mixed_space_info(space_info
))
6738 use_cluster
= false;
6740 if (flags
& BTRFS_BLOCK_GROUP_METADATA
&& use_cluster
) {
6741 last_ptr
= &root
->fs_info
->meta_alloc_cluster
;
6742 if (!btrfs_test_opt(root
, SSD
))
6743 empty_cluster
= 64 * 1024;
6746 if ((flags
& BTRFS_BLOCK_GROUP_DATA
) && use_cluster
&&
6747 btrfs_test_opt(root
, SSD
)) {
6748 last_ptr
= &root
->fs_info
->data_alloc_cluster
;
6752 spin_lock(&last_ptr
->lock
);
6753 if (last_ptr
->block_group
)
6754 hint_byte
= last_ptr
->window_start
;
6755 spin_unlock(&last_ptr
->lock
);
6758 search_start
= max(search_start
, first_logical_byte(root
, 0));
6759 search_start
= max(search_start
, hint_byte
);
6764 if (search_start
== hint_byte
) {
6765 block_group
= btrfs_lookup_block_group(root
->fs_info
,
6768 * we don't want to use the block group if it doesn't match our
6769 * allocation bits, or if its not cached.
6771 * However if we are re-searching with an ideal block group
6772 * picked out then we don't care that the block group is cached.
6774 if (block_group
&& block_group_bits(block_group
, flags
) &&
6775 block_group
->cached
!= BTRFS_CACHE_NO
) {
6776 down_read(&space_info
->groups_sem
);
6777 if (list_empty(&block_group
->list
) ||
6780 * someone is removing this block group,
6781 * we can't jump into the have_block_group
6782 * target because our list pointers are not
6785 btrfs_put_block_group(block_group
);
6786 up_read(&space_info
->groups_sem
);
6788 index
= get_block_group_index(block_group
);
6789 btrfs_lock_block_group(block_group
, delalloc
);
6790 goto have_block_group
;
6792 } else if (block_group
) {
6793 btrfs_put_block_group(block_group
);
6797 have_caching_bg
= false;
6798 down_read(&space_info
->groups_sem
);
6799 list_for_each_entry(block_group
, &space_info
->block_groups
[index
],
6804 btrfs_grab_block_group(block_group
, delalloc
);
6805 search_start
= block_group
->key
.objectid
;
6808 * this can happen if we end up cycling through all the
6809 * raid types, but we want to make sure we only allocate
6810 * for the proper type.
6812 if (!block_group_bits(block_group
, flags
)) {
6813 u64 extra
= BTRFS_BLOCK_GROUP_DUP
|
6814 BTRFS_BLOCK_GROUP_RAID1
|
6815 BTRFS_BLOCK_GROUP_RAID5
|
6816 BTRFS_BLOCK_GROUP_RAID6
|
6817 BTRFS_BLOCK_GROUP_RAID10
;
6820 * if they asked for extra copies and this block group
6821 * doesn't provide them, bail. This does allow us to
6822 * fill raid0 from raid1.
6824 if ((flags
& extra
) && !(block_group
->flags
& extra
))
6829 cached
= block_group_cache_done(block_group
);
6830 if (unlikely(!cached
)) {
6831 ret
= cache_block_group(block_group
, 0);
6836 if (unlikely(block_group
->cached
== BTRFS_CACHE_ERROR
))
6838 if (unlikely(block_group
->ro
))
6842 * Ok we want to try and use the cluster allocator, so
6846 struct btrfs_block_group_cache
*used_block_group
;
6847 unsigned long aligned_cluster
;
6849 * the refill lock keeps out other
6850 * people trying to start a new cluster
6852 used_block_group
= btrfs_lock_cluster(block_group
,
6855 if (!used_block_group
)
6856 goto refill_cluster
;
6858 if (used_block_group
!= block_group
&&
6859 (used_block_group
->ro
||
6860 !block_group_bits(used_block_group
, flags
)))
6861 goto release_cluster
;
6863 offset
= btrfs_alloc_from_cluster(used_block_group
,
6866 used_block_group
->key
.objectid
,
6869 /* we have a block, we're done */
6870 spin_unlock(&last_ptr
->refill_lock
);
6871 trace_btrfs_reserve_extent_cluster(root
,
6873 search_start
, num_bytes
);
6874 if (used_block_group
!= block_group
) {
6875 btrfs_release_block_group(block_group
,
6877 block_group
= used_block_group
;
6882 WARN_ON(last_ptr
->block_group
!= used_block_group
);
6884 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6885 * set up a new clusters, so lets just skip it
6886 * and let the allocator find whatever block
6887 * it can find. If we reach this point, we
6888 * will have tried the cluster allocator
6889 * plenty of times and not have found
6890 * anything, so we are likely way too
6891 * fragmented for the clustering stuff to find
6894 * However, if the cluster is taken from the
6895 * current block group, release the cluster
6896 * first, so that we stand a better chance of
6897 * succeeding in the unclustered
6899 if (loop
>= LOOP_NO_EMPTY_SIZE
&&
6900 used_block_group
!= block_group
) {
6901 spin_unlock(&last_ptr
->refill_lock
);
6902 btrfs_release_block_group(used_block_group
,
6904 goto unclustered_alloc
;
6908 * this cluster didn't work out, free it and
6911 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
6913 if (used_block_group
!= block_group
)
6914 btrfs_release_block_group(used_block_group
,
6917 if (loop
>= LOOP_NO_EMPTY_SIZE
) {
6918 spin_unlock(&last_ptr
->refill_lock
);
6919 goto unclustered_alloc
;
6922 aligned_cluster
= max_t(unsigned long,
6923 empty_cluster
+ empty_size
,
6924 block_group
->full_stripe_len
);
6926 /* allocate a cluster in this block group */
6927 ret
= btrfs_find_space_cluster(root
, block_group
,
6928 last_ptr
, search_start
,
6933 * now pull our allocation out of this
6936 offset
= btrfs_alloc_from_cluster(block_group
,
6942 /* we found one, proceed */
6943 spin_unlock(&last_ptr
->refill_lock
);
6944 trace_btrfs_reserve_extent_cluster(root
,
6945 block_group
, search_start
,
6949 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
6950 && !failed_cluster_refill
) {
6951 spin_unlock(&last_ptr
->refill_lock
);
6953 failed_cluster_refill
= true;
6954 wait_block_group_cache_progress(block_group
,
6955 num_bytes
+ empty_cluster
+ empty_size
);
6956 goto have_block_group
;
6960 * at this point we either didn't find a cluster
6961 * or we weren't able to allocate a block from our
6962 * cluster. Free the cluster we've been trying
6963 * to use, and go to the next block group
6965 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
6966 spin_unlock(&last_ptr
->refill_lock
);
6971 spin_lock(&block_group
->free_space_ctl
->tree_lock
);
6973 block_group
->free_space_ctl
->free_space
<
6974 num_bytes
+ empty_cluster
+ empty_size
) {
6975 if (block_group
->free_space_ctl
->free_space
>
6978 block_group
->free_space_ctl
->free_space
;
6979 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
6982 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
6984 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
6985 num_bytes
, empty_size
,
6988 * If we didn't find a chunk, and we haven't failed on this
6989 * block group before, and this block group is in the middle of
6990 * caching and we are ok with waiting, then go ahead and wait
6991 * for progress to be made, and set failed_alloc to true.
6993 * If failed_alloc is true then we've already waited on this
6994 * block group once and should move on to the next block group.
6996 if (!offset
&& !failed_alloc
&& !cached
&&
6997 loop
> LOOP_CACHING_NOWAIT
) {
6998 wait_block_group_cache_progress(block_group
,
6999 num_bytes
+ empty_size
);
7000 failed_alloc
= true;
7001 goto have_block_group
;
7002 } else if (!offset
) {
7004 have_caching_bg
= true;
7008 search_start
= ALIGN(offset
, root
->stripesize
);
7010 /* move on to the next group */
7011 if (search_start
+ num_bytes
>
7012 block_group
->key
.objectid
+ block_group
->key
.offset
) {
7013 btrfs_add_free_space(block_group
, offset
, num_bytes
);
7017 if (offset
< search_start
)
7018 btrfs_add_free_space(block_group
, offset
,
7019 search_start
- offset
);
7020 BUG_ON(offset
> search_start
);
7022 ret
= btrfs_update_reserved_bytes(block_group
, num_bytes
,
7023 alloc_type
, delalloc
);
7024 if (ret
== -EAGAIN
) {
7025 btrfs_add_free_space(block_group
, offset
, num_bytes
);
7029 /* we are all good, lets return */
7030 ins
->objectid
= search_start
;
7031 ins
->offset
= num_bytes
;
7033 trace_btrfs_reserve_extent(orig_root
, block_group
,
7034 search_start
, num_bytes
);
7035 btrfs_release_block_group(block_group
, delalloc
);
7038 failed_cluster_refill
= false;
7039 failed_alloc
= false;
7040 BUG_ON(index
!= get_block_group_index(block_group
));
7041 btrfs_release_block_group(block_group
, delalloc
);
7043 up_read(&space_info
->groups_sem
);
7045 if (!ins
->objectid
&& loop
>= LOOP_CACHING_WAIT
&& have_caching_bg
)
7048 if (!ins
->objectid
&& ++index
< BTRFS_NR_RAID_TYPES
)
7052 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7053 * caching kthreads as we move along
7054 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7055 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7056 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7059 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
) {
7062 if (loop
== LOOP_ALLOC_CHUNK
) {
7063 struct btrfs_trans_handle
*trans
;
7066 trans
= current
->journal_info
;
7070 trans
= btrfs_join_transaction(root
);
7072 if (IS_ERR(trans
)) {
7073 ret
= PTR_ERR(trans
);
7077 ret
= do_chunk_alloc(trans
, root
, flags
,
7080 * Do not bail out on ENOSPC since we
7081 * can do more things.
7083 if (ret
< 0 && ret
!= -ENOSPC
)
7084 btrfs_abort_transaction(trans
,
7089 btrfs_end_transaction(trans
, root
);
7094 if (loop
== LOOP_NO_EMPTY_SIZE
) {
7100 } else if (!ins
->objectid
) {
7102 } else if (ins
->objectid
) {
7107 ins
->offset
= max_extent_size
;
7111 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
7112 int dump_block_groups
)
7114 struct btrfs_block_group_cache
*cache
;
7117 spin_lock(&info
->lock
);
7118 printk(KERN_INFO
"BTRFS: space_info %llu has %llu free, is %sfull\n",
7120 info
->total_bytes
- info
->bytes_used
- info
->bytes_pinned
-
7121 info
->bytes_reserved
- info
->bytes_readonly
,
7122 (info
->full
) ? "" : "not ");
7123 printk(KERN_INFO
"BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7124 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7125 info
->total_bytes
, info
->bytes_used
, info
->bytes_pinned
,
7126 info
->bytes_reserved
, info
->bytes_may_use
,
7127 info
->bytes_readonly
);
7128 spin_unlock(&info
->lock
);
7130 if (!dump_block_groups
)
7133 down_read(&info
->groups_sem
);
7135 list_for_each_entry(cache
, &info
->block_groups
[index
], list
) {
7136 spin_lock(&cache
->lock
);
7137 printk(KERN_INFO
"BTRFS: "
7138 "block group %llu has %llu bytes, "
7139 "%llu used %llu pinned %llu reserved %s\n",
7140 cache
->key
.objectid
, cache
->key
.offset
,
7141 btrfs_block_group_used(&cache
->item
), cache
->pinned
,
7142 cache
->reserved
, cache
->ro
? "[readonly]" : "");
7143 btrfs_dump_free_space(cache
, bytes
);
7144 spin_unlock(&cache
->lock
);
7146 if (++index
< BTRFS_NR_RAID_TYPES
)
7148 up_read(&info
->groups_sem
);
7151 int btrfs_reserve_extent(struct btrfs_root
*root
,
7152 u64 num_bytes
, u64 min_alloc_size
,
7153 u64 empty_size
, u64 hint_byte
,
7154 struct btrfs_key
*ins
, int is_data
, int delalloc
)
7156 bool final_tried
= false;
7160 flags
= btrfs_get_alloc_profile(root
, is_data
);
7162 WARN_ON(num_bytes
< root
->sectorsize
);
7163 ret
= find_free_extent(root
, num_bytes
, empty_size
, hint_byte
, ins
,
7166 if (ret
== -ENOSPC
) {
7167 if (!final_tried
&& ins
->offset
) {
7168 num_bytes
= min(num_bytes
>> 1, ins
->offset
);
7169 num_bytes
= round_down(num_bytes
, root
->sectorsize
);
7170 num_bytes
= max(num_bytes
, min_alloc_size
);
7171 if (num_bytes
== min_alloc_size
)
7174 } else if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
7175 struct btrfs_space_info
*sinfo
;
7177 sinfo
= __find_space_info(root
->fs_info
, flags
);
7178 btrfs_err(root
->fs_info
, "allocation failed flags %llu, wanted %llu",
7181 dump_space_info(sinfo
, num_bytes
, 1);
7188 static int __btrfs_free_reserved_extent(struct btrfs_root
*root
,
7190 int pin
, int delalloc
)
7192 struct btrfs_block_group_cache
*cache
;
7195 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
7197 btrfs_err(root
->fs_info
, "Unable to find block group for %llu",
7203 pin_down_extent(root
, cache
, start
, len
, 1);
7205 if (btrfs_test_opt(root
, DISCARD
))
7206 ret
= btrfs_discard_extent(root
, start
, len
, NULL
);
7207 btrfs_add_free_space(cache
, start
, len
);
7208 btrfs_update_reserved_bytes(cache
, len
, RESERVE_FREE
, delalloc
);
7211 btrfs_put_block_group(cache
);
7213 trace_btrfs_reserved_extent_free(root
, start
, len
);
7218 int btrfs_free_reserved_extent(struct btrfs_root
*root
,
7219 u64 start
, u64 len
, int delalloc
)
7221 return __btrfs_free_reserved_extent(root
, start
, len
, 0, delalloc
);
7224 int btrfs_free_and_pin_reserved_extent(struct btrfs_root
*root
,
7227 return __btrfs_free_reserved_extent(root
, start
, len
, 1, 0);
7230 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
7231 struct btrfs_root
*root
,
7232 u64 parent
, u64 root_objectid
,
7233 u64 flags
, u64 owner
, u64 offset
,
7234 struct btrfs_key
*ins
, int ref_mod
)
7237 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7238 struct btrfs_extent_item
*extent_item
;
7239 struct btrfs_extent_inline_ref
*iref
;
7240 struct btrfs_path
*path
;
7241 struct extent_buffer
*leaf
;
7246 type
= BTRFS_SHARED_DATA_REF_KEY
;
7248 type
= BTRFS_EXTENT_DATA_REF_KEY
;
7250 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
7252 path
= btrfs_alloc_path();
7256 path
->leave_spinning
= 1;
7257 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
7260 btrfs_free_path(path
);
7264 leaf
= path
->nodes
[0];
7265 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
7266 struct btrfs_extent_item
);
7267 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
7268 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
7269 btrfs_set_extent_flags(leaf
, extent_item
,
7270 flags
| BTRFS_EXTENT_FLAG_DATA
);
7272 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
7273 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
7275 struct btrfs_shared_data_ref
*ref
;
7276 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
7277 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
7278 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
7280 struct btrfs_extent_data_ref
*ref
;
7281 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
7282 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
7283 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
7284 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
7285 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
7288 btrfs_mark_buffer_dirty(path
->nodes
[0]);
7289 btrfs_free_path(path
);
7291 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1);
7292 if (ret
) { /* -ENOENT, logic error */
7293 btrfs_err(fs_info
, "update block group failed for %llu %llu",
7294 ins
->objectid
, ins
->offset
);
7297 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, ins
->offset
);
7301 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
7302 struct btrfs_root
*root
,
7303 u64 parent
, u64 root_objectid
,
7304 u64 flags
, struct btrfs_disk_key
*key
,
7305 int level
, struct btrfs_key
*ins
,
7309 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7310 struct btrfs_extent_item
*extent_item
;
7311 struct btrfs_tree_block_info
*block_info
;
7312 struct btrfs_extent_inline_ref
*iref
;
7313 struct btrfs_path
*path
;
7314 struct extent_buffer
*leaf
;
7315 u32 size
= sizeof(*extent_item
) + sizeof(*iref
);
7316 u64 num_bytes
= ins
->offset
;
7317 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
7320 if (!skinny_metadata
)
7321 size
+= sizeof(*block_info
);
7323 path
= btrfs_alloc_path();
7325 btrfs_free_and_pin_reserved_extent(root
, ins
->objectid
,
7330 path
->leave_spinning
= 1;
7331 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
7334 btrfs_free_path(path
);
7335 btrfs_free_and_pin_reserved_extent(root
, ins
->objectid
,
7340 leaf
= path
->nodes
[0];
7341 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
7342 struct btrfs_extent_item
);
7343 btrfs_set_extent_refs(leaf
, extent_item
, 1);
7344 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
7345 btrfs_set_extent_flags(leaf
, extent_item
,
7346 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
7348 if (skinny_metadata
) {
7349 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
7350 num_bytes
= root
->nodesize
;
7352 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
7353 btrfs_set_tree_block_key(leaf
, block_info
, key
);
7354 btrfs_set_tree_block_level(leaf
, block_info
, level
);
7355 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
7359 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
7360 btrfs_set_extent_inline_ref_type(leaf
, iref
,
7361 BTRFS_SHARED_BLOCK_REF_KEY
);
7362 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
7364 btrfs_set_extent_inline_ref_type(leaf
, iref
,
7365 BTRFS_TREE_BLOCK_REF_KEY
);
7366 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
7369 btrfs_mark_buffer_dirty(leaf
);
7370 btrfs_free_path(path
);
7372 ret
= update_block_group(trans
, root
, ins
->objectid
, root
->nodesize
,
7374 if (ret
) { /* -ENOENT, logic error */
7375 btrfs_err(fs_info
, "update block group failed for %llu %llu",
7376 ins
->objectid
, ins
->offset
);
7380 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, root
->nodesize
);
7384 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
7385 struct btrfs_root
*root
,
7386 u64 root_objectid
, u64 owner
,
7387 u64 offset
, struct btrfs_key
*ins
)
7391 BUG_ON(root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
7393 ret
= btrfs_add_delayed_data_ref(root
->fs_info
, trans
, ins
->objectid
,
7395 root_objectid
, owner
, offset
,
7396 BTRFS_ADD_DELAYED_EXTENT
, NULL
, 0);
7401 * this is used by the tree logging recovery code. It records that
7402 * an extent has been allocated and makes sure to clear the free
7403 * space cache bits as well
7405 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
7406 struct btrfs_root
*root
,
7407 u64 root_objectid
, u64 owner
, u64 offset
,
7408 struct btrfs_key
*ins
)
7411 struct btrfs_block_group_cache
*block_group
;
7414 * Mixed block groups will exclude before processing the log so we only
7415 * need to do the exlude dance if this fs isn't mixed.
7417 if (!btrfs_fs_incompat(root
->fs_info
, MIXED_GROUPS
)) {
7418 ret
= __exclude_logged_extent(root
, ins
->objectid
, ins
->offset
);
7423 block_group
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
7427 ret
= btrfs_update_reserved_bytes(block_group
, ins
->offset
,
7428 RESERVE_ALLOC_NO_ACCOUNT
, 0);
7429 BUG_ON(ret
); /* logic error */
7430 ret
= alloc_reserved_file_extent(trans
, root
, 0, root_objectid
,
7431 0, owner
, offset
, ins
, 1);
7432 btrfs_put_block_group(block_group
);
7436 static struct extent_buffer
*
7437 btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
7438 u64 bytenr
, int level
)
7440 struct extent_buffer
*buf
;
7442 buf
= btrfs_find_create_tree_block(root
, bytenr
);
7444 return ERR_PTR(-ENOMEM
);
7445 btrfs_set_header_generation(buf
, trans
->transid
);
7446 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, buf
, level
);
7447 btrfs_tree_lock(buf
);
7448 clean_tree_block(trans
, root
->fs_info
, buf
);
7449 clear_bit(EXTENT_BUFFER_STALE
, &buf
->bflags
);
7451 btrfs_set_lock_blocking(buf
);
7452 btrfs_set_buffer_uptodate(buf
);
7454 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
7455 buf
->log_index
= root
->log_transid
% 2;
7457 * we allow two log transactions at a time, use different
7458 * EXENT bit to differentiate dirty pages.
7460 if (buf
->log_index
== 0)
7461 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
7462 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
7464 set_extent_new(&root
->dirty_log_pages
, buf
->start
,
7465 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
7467 buf
->log_index
= -1;
7468 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
7469 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
7471 trans
->blocks_used
++;
7472 /* this returns a buffer locked for blocking */
7476 static struct btrfs_block_rsv
*
7477 use_block_rsv(struct btrfs_trans_handle
*trans
,
7478 struct btrfs_root
*root
, u32 blocksize
)
7480 struct btrfs_block_rsv
*block_rsv
;
7481 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
7483 bool global_updated
= false;
7485 block_rsv
= get_block_rsv(trans
, root
);
7487 if (unlikely(block_rsv
->size
== 0))
7490 ret
= block_rsv_use_bytes(block_rsv
, blocksize
);
7494 if (block_rsv
->failfast
)
7495 return ERR_PTR(ret
);
7497 if (block_rsv
->type
== BTRFS_BLOCK_RSV_GLOBAL
&& !global_updated
) {
7498 global_updated
= true;
7499 update_global_block_rsv(root
->fs_info
);
7503 if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
7504 static DEFINE_RATELIMIT_STATE(_rs
,
7505 DEFAULT_RATELIMIT_INTERVAL
* 10,
7506 /*DEFAULT_RATELIMIT_BURST*/ 1);
7507 if (__ratelimit(&_rs
))
7509 "BTRFS: block rsv returned %d\n", ret
);
7512 ret
= reserve_metadata_bytes(root
, block_rsv
, blocksize
,
7513 BTRFS_RESERVE_NO_FLUSH
);
7517 * If we couldn't reserve metadata bytes try and use some from
7518 * the global reserve if its space type is the same as the global
7521 if (block_rsv
->type
!= BTRFS_BLOCK_RSV_GLOBAL
&&
7522 block_rsv
->space_info
== global_rsv
->space_info
) {
7523 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
7527 return ERR_PTR(ret
);
7530 static void unuse_block_rsv(struct btrfs_fs_info
*fs_info
,
7531 struct btrfs_block_rsv
*block_rsv
, u32 blocksize
)
7533 block_rsv_add_bytes(block_rsv
, blocksize
, 0);
7534 block_rsv_release_bytes(fs_info
, block_rsv
, NULL
, 0);
7538 * finds a free extent and does all the dirty work required for allocation
7539 * returns the key for the extent through ins, and a tree buffer for
7540 * the first block of the extent through buf.
7542 * returns the tree buffer or an ERR_PTR on error.
7544 struct extent_buffer
*btrfs_alloc_tree_block(struct btrfs_trans_handle
*trans
,
7545 struct btrfs_root
*root
,
7546 u64 parent
, u64 root_objectid
,
7547 struct btrfs_disk_key
*key
, int level
,
7548 u64 hint
, u64 empty_size
)
7550 struct btrfs_key ins
;
7551 struct btrfs_block_rsv
*block_rsv
;
7552 struct extent_buffer
*buf
;
7553 struct btrfs_delayed_extent_op
*extent_op
;
7556 u32 blocksize
= root
->nodesize
;
7557 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
7560 if (btrfs_test_is_dummy_root(root
)) {
7561 buf
= btrfs_init_new_buffer(trans
, root
, root
->alloc_bytenr
,
7564 root
->alloc_bytenr
+= blocksize
;
7568 block_rsv
= use_block_rsv(trans
, root
, blocksize
);
7569 if (IS_ERR(block_rsv
))
7570 return ERR_CAST(block_rsv
);
7572 ret
= btrfs_reserve_extent(root
, blocksize
, blocksize
,
7573 empty_size
, hint
, &ins
, 0, 0);
7577 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
, level
);
7580 goto out_free_reserved
;
7583 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
7585 parent
= ins
.objectid
;
7586 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
7590 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
7591 extent_op
= btrfs_alloc_delayed_extent_op();
7597 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
7599 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
7600 extent_op
->flags_to_set
= flags
;
7601 if (skinny_metadata
)
7602 extent_op
->update_key
= 0;
7604 extent_op
->update_key
= 1;
7605 extent_op
->update_flags
= 1;
7606 extent_op
->is_data
= 0;
7607 extent_op
->level
= level
;
7609 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
7610 ins
.objectid
, ins
.offset
,
7611 parent
, root_objectid
, level
,
7612 BTRFS_ADD_DELAYED_EXTENT
,
7615 goto out_free_delayed
;
7620 btrfs_free_delayed_extent_op(extent_op
);
7622 free_extent_buffer(buf
);
7624 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
, 0);
7626 unuse_block_rsv(root
->fs_info
, block_rsv
, blocksize
);
7627 return ERR_PTR(ret
);
7630 struct walk_control
{
7631 u64 refs
[BTRFS_MAX_LEVEL
];
7632 u64 flags
[BTRFS_MAX_LEVEL
];
7633 struct btrfs_key update_progress
;
7644 #define DROP_REFERENCE 1
7645 #define UPDATE_BACKREF 2
7647 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
7648 struct btrfs_root
*root
,
7649 struct walk_control
*wc
,
7650 struct btrfs_path
*path
)
7658 struct btrfs_key key
;
7659 struct extent_buffer
*eb
;
7664 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
7665 wc
->reada_count
= wc
->reada_count
* 2 / 3;
7666 wc
->reada_count
= max(wc
->reada_count
, 2);
7668 wc
->reada_count
= wc
->reada_count
* 3 / 2;
7669 wc
->reada_count
= min_t(int, wc
->reada_count
,
7670 BTRFS_NODEPTRS_PER_BLOCK(root
));
7673 eb
= path
->nodes
[wc
->level
];
7674 nritems
= btrfs_header_nritems(eb
);
7675 blocksize
= root
->nodesize
;
7677 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
7678 if (nread
>= wc
->reada_count
)
7682 bytenr
= btrfs_node_blockptr(eb
, slot
);
7683 generation
= btrfs_node_ptr_generation(eb
, slot
);
7685 if (slot
== path
->slots
[wc
->level
])
7688 if (wc
->stage
== UPDATE_BACKREF
&&
7689 generation
<= root
->root_key
.offset
)
7692 /* We don't lock the tree block, it's OK to be racy here */
7693 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
,
7694 wc
->level
- 1, 1, &refs
,
7696 /* We don't care about errors in readahead. */
7701 if (wc
->stage
== DROP_REFERENCE
) {
7705 if (wc
->level
== 1 &&
7706 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
7708 if (!wc
->update_ref
||
7709 generation
<= root
->root_key
.offset
)
7711 btrfs_node_key_to_cpu(eb
, &key
, slot
);
7712 ret
= btrfs_comp_cpu_keys(&key
,
7713 &wc
->update_progress
);
7717 if (wc
->level
== 1 &&
7718 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
7722 readahead_tree_block(root
, bytenr
);
7725 wc
->reada_slot
= slot
;
7729 * TODO: Modify related function to add related node/leaf to dirty_extent_root,
7730 * for later qgroup accounting.
7732 * Current, this function does nothing.
7734 static int account_leaf_items(struct btrfs_trans_handle
*trans
,
7735 struct btrfs_root
*root
,
7736 struct extent_buffer
*eb
)
7738 int nr
= btrfs_header_nritems(eb
);
7740 struct btrfs_key key
;
7741 struct btrfs_file_extent_item
*fi
;
7742 u64 bytenr
, num_bytes
;
7744 for (i
= 0; i
< nr
; i
++) {
7745 btrfs_item_key_to_cpu(eb
, &key
, i
);
7747 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
7750 fi
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
7751 /* filter out non qgroup-accountable extents */
7752 extent_type
= btrfs_file_extent_type(eb
, fi
);
7754 if (extent_type
== BTRFS_FILE_EXTENT_INLINE
)
7757 bytenr
= btrfs_file_extent_disk_bytenr(eb
, fi
);
7761 num_bytes
= btrfs_file_extent_disk_num_bytes(eb
, fi
);
7767 * Walk up the tree from the bottom, freeing leaves and any interior
7768 * nodes which have had all slots visited. If a node (leaf or
7769 * interior) is freed, the node above it will have it's slot
7770 * incremented. The root node will never be freed.
7772 * At the end of this function, we should have a path which has all
7773 * slots incremented to the next position for a search. If we need to
7774 * read a new node it will be NULL and the node above it will have the
7775 * correct slot selected for a later read.
7777 * If we increment the root nodes slot counter past the number of
7778 * elements, 1 is returned to signal completion of the search.
7780 static int adjust_slots_upwards(struct btrfs_root
*root
,
7781 struct btrfs_path
*path
, int root_level
)
7785 struct extent_buffer
*eb
;
7787 if (root_level
== 0)
7790 while (level
<= root_level
) {
7791 eb
= path
->nodes
[level
];
7792 nr
= btrfs_header_nritems(eb
);
7793 path
->slots
[level
]++;
7794 slot
= path
->slots
[level
];
7795 if (slot
>= nr
|| level
== 0) {
7797 * Don't free the root - we will detect this
7798 * condition after our loop and return a
7799 * positive value for caller to stop walking the tree.
7801 if (level
!= root_level
) {
7802 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7803 path
->locks
[level
] = 0;
7805 free_extent_buffer(eb
);
7806 path
->nodes
[level
] = NULL
;
7807 path
->slots
[level
] = 0;
7811 * We have a valid slot to walk back down
7812 * from. Stop here so caller can process these
7821 eb
= path
->nodes
[root_level
];
7822 if (path
->slots
[root_level
] >= btrfs_header_nritems(eb
))
7829 * root_eb is the subtree root and is locked before this function is called.
7830 * TODO: Modify this function to mark all (including complete shared node)
7831 * to dirty_extent_root to allow it get accounted in qgroup.
7833 static int account_shared_subtree(struct btrfs_trans_handle
*trans
,
7834 struct btrfs_root
*root
,
7835 struct extent_buffer
*root_eb
,
7841 struct extent_buffer
*eb
= root_eb
;
7842 struct btrfs_path
*path
= NULL
;
7844 BUG_ON(root_level
< 0 || root_level
> BTRFS_MAX_LEVEL
);
7845 BUG_ON(root_eb
== NULL
);
7847 if (!root
->fs_info
->quota_enabled
)
7850 if (!extent_buffer_uptodate(root_eb
)) {
7851 ret
= btrfs_read_buffer(root_eb
, root_gen
);
7856 if (root_level
== 0) {
7857 ret
= account_leaf_items(trans
, root
, root_eb
);
7861 path
= btrfs_alloc_path();
7866 * Walk down the tree. Missing extent blocks are filled in as
7867 * we go. Metadata is accounted every time we read a new
7870 * When we reach a leaf, we account for file extent items in it,
7871 * walk back up the tree (adjusting slot pointers as we go)
7872 * and restart the search process.
7874 extent_buffer_get(root_eb
); /* For path */
7875 path
->nodes
[root_level
] = root_eb
;
7876 path
->slots
[root_level
] = 0;
7877 path
->locks
[root_level
] = 0; /* so release_path doesn't try to unlock */
7880 while (level
>= 0) {
7881 if (path
->nodes
[level
] == NULL
) {
7886 /* We need to get child blockptr/gen from
7887 * parent before we can read it. */
7888 eb
= path
->nodes
[level
+ 1];
7889 parent_slot
= path
->slots
[level
+ 1];
7890 child_bytenr
= btrfs_node_blockptr(eb
, parent_slot
);
7891 child_gen
= btrfs_node_ptr_generation(eb
, parent_slot
);
7893 eb
= read_tree_block(root
, child_bytenr
, child_gen
);
7897 } else if (!extent_buffer_uptodate(eb
)) {
7898 free_extent_buffer(eb
);
7903 path
->nodes
[level
] = eb
;
7904 path
->slots
[level
] = 0;
7906 btrfs_tree_read_lock(eb
);
7907 btrfs_set_lock_blocking_rw(eb
, BTRFS_READ_LOCK
);
7908 path
->locks
[level
] = BTRFS_READ_LOCK_BLOCKING
;
7912 ret
= account_leaf_items(trans
, root
, path
->nodes
[level
]);
7916 /* Nonzero return here means we completed our search */
7917 ret
= adjust_slots_upwards(root
, path
, root_level
);
7921 /* Restart search with new slots */
7930 btrfs_free_path(path
);
7936 * helper to process tree block while walking down the tree.
7938 * when wc->stage == UPDATE_BACKREF, this function updates
7939 * back refs for pointers in the block.
7941 * NOTE: return value 1 means we should stop walking down.
7943 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
7944 struct btrfs_root
*root
,
7945 struct btrfs_path
*path
,
7946 struct walk_control
*wc
, int lookup_info
)
7948 int level
= wc
->level
;
7949 struct extent_buffer
*eb
= path
->nodes
[level
];
7950 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
7953 if (wc
->stage
== UPDATE_BACKREF
&&
7954 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
7958 * when reference count of tree block is 1, it won't increase
7959 * again. once full backref flag is set, we never clear it.
7962 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
7963 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
7964 BUG_ON(!path
->locks
[level
]);
7965 ret
= btrfs_lookup_extent_info(trans
, root
,
7966 eb
->start
, level
, 1,
7969 BUG_ON(ret
== -ENOMEM
);
7972 BUG_ON(wc
->refs
[level
] == 0);
7975 if (wc
->stage
== DROP_REFERENCE
) {
7976 if (wc
->refs
[level
] > 1)
7979 if (path
->locks
[level
] && !wc
->keep_locks
) {
7980 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7981 path
->locks
[level
] = 0;
7986 /* wc->stage == UPDATE_BACKREF */
7987 if (!(wc
->flags
[level
] & flag
)) {
7988 BUG_ON(!path
->locks
[level
]);
7989 ret
= btrfs_inc_ref(trans
, root
, eb
, 1);
7990 BUG_ON(ret
); /* -ENOMEM */
7991 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
7992 BUG_ON(ret
); /* -ENOMEM */
7993 ret
= btrfs_set_disk_extent_flags(trans
, root
, eb
->start
,
7995 btrfs_header_level(eb
), 0);
7996 BUG_ON(ret
); /* -ENOMEM */
7997 wc
->flags
[level
] |= flag
;
8001 * the block is shared by multiple trees, so it's not good to
8002 * keep the tree lock
8004 if (path
->locks
[level
] && level
> 0) {
8005 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8006 path
->locks
[level
] = 0;
8012 * helper to process tree block pointer.
8014 * when wc->stage == DROP_REFERENCE, this function checks
8015 * reference count of the block pointed to. if the block
8016 * is shared and we need update back refs for the subtree
8017 * rooted at the block, this function changes wc->stage to
8018 * UPDATE_BACKREF. if the block is shared and there is no
8019 * need to update back, this function drops the reference
8022 * NOTE: return value 1 means we should stop walking down.
8024 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
8025 struct btrfs_root
*root
,
8026 struct btrfs_path
*path
,
8027 struct walk_control
*wc
, int *lookup_info
)
8033 struct btrfs_key key
;
8034 struct extent_buffer
*next
;
8035 int level
= wc
->level
;
8038 bool need_account
= false;
8040 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
8041 path
->slots
[level
]);
8043 * if the lower level block was created before the snapshot
8044 * was created, we know there is no need to update back refs
8047 if (wc
->stage
== UPDATE_BACKREF
&&
8048 generation
<= root
->root_key
.offset
) {
8053 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
8054 blocksize
= root
->nodesize
;
8056 next
= btrfs_find_tree_block(root
->fs_info
, bytenr
);
8058 next
= btrfs_find_create_tree_block(root
, bytenr
);
8061 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, next
,
8065 btrfs_tree_lock(next
);
8066 btrfs_set_lock_blocking(next
);
8068 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, level
- 1, 1,
8069 &wc
->refs
[level
- 1],
8070 &wc
->flags
[level
- 1]);
8072 btrfs_tree_unlock(next
);
8076 if (unlikely(wc
->refs
[level
- 1] == 0)) {
8077 btrfs_err(root
->fs_info
, "Missing references.");
8082 if (wc
->stage
== DROP_REFERENCE
) {
8083 if (wc
->refs
[level
- 1] > 1) {
8084 need_account
= true;
8086 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8089 if (!wc
->update_ref
||
8090 generation
<= root
->root_key
.offset
)
8093 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
8094 path
->slots
[level
]);
8095 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
8099 wc
->stage
= UPDATE_BACKREF
;
8100 wc
->shared_level
= level
- 1;
8104 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8108 if (!btrfs_buffer_uptodate(next
, generation
, 0)) {
8109 btrfs_tree_unlock(next
);
8110 free_extent_buffer(next
);
8116 if (reada
&& level
== 1)
8117 reada_walk_down(trans
, root
, wc
, path
);
8118 next
= read_tree_block(root
, bytenr
, generation
);
8120 return PTR_ERR(next
);
8121 } else if (!extent_buffer_uptodate(next
)) {
8122 free_extent_buffer(next
);
8125 btrfs_tree_lock(next
);
8126 btrfs_set_lock_blocking(next
);
8130 BUG_ON(level
!= btrfs_header_level(next
));
8131 path
->nodes
[level
] = next
;
8132 path
->slots
[level
] = 0;
8133 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8139 wc
->refs
[level
- 1] = 0;
8140 wc
->flags
[level
- 1] = 0;
8141 if (wc
->stage
== DROP_REFERENCE
) {
8142 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
8143 parent
= path
->nodes
[level
]->start
;
8145 BUG_ON(root
->root_key
.objectid
!=
8146 btrfs_header_owner(path
->nodes
[level
]));
8151 ret
= account_shared_subtree(trans
, root
, next
,
8152 generation
, level
- 1);
8154 printk_ratelimited(KERN_ERR
"BTRFS: %s Error "
8155 "%d accounting shared subtree. Quota "
8156 "is out of sync, rescan required.\n",
8157 root
->fs_info
->sb
->s_id
, ret
);
8160 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
, parent
,
8161 root
->root_key
.objectid
, level
- 1, 0, 0);
8162 BUG_ON(ret
); /* -ENOMEM */
8164 btrfs_tree_unlock(next
);
8165 free_extent_buffer(next
);
8171 * helper to process tree block while walking up the tree.
8173 * when wc->stage == DROP_REFERENCE, this function drops
8174 * reference count on the block.
8176 * when wc->stage == UPDATE_BACKREF, this function changes
8177 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8178 * to UPDATE_BACKREF previously while processing the block.
8180 * NOTE: return value 1 means we should stop walking up.
8182 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
8183 struct btrfs_root
*root
,
8184 struct btrfs_path
*path
,
8185 struct walk_control
*wc
)
8188 int level
= wc
->level
;
8189 struct extent_buffer
*eb
= path
->nodes
[level
];
8192 if (wc
->stage
== UPDATE_BACKREF
) {
8193 BUG_ON(wc
->shared_level
< level
);
8194 if (level
< wc
->shared_level
)
8197 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
8201 wc
->stage
= DROP_REFERENCE
;
8202 wc
->shared_level
= -1;
8203 path
->slots
[level
] = 0;
8206 * check reference count again if the block isn't locked.
8207 * we should start walking down the tree again if reference
8210 if (!path
->locks
[level
]) {
8212 btrfs_tree_lock(eb
);
8213 btrfs_set_lock_blocking(eb
);
8214 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8216 ret
= btrfs_lookup_extent_info(trans
, root
,
8217 eb
->start
, level
, 1,
8221 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8222 path
->locks
[level
] = 0;
8225 BUG_ON(wc
->refs
[level
] == 0);
8226 if (wc
->refs
[level
] == 1) {
8227 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8228 path
->locks
[level
] = 0;
8234 /* wc->stage == DROP_REFERENCE */
8235 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
8237 if (wc
->refs
[level
] == 1) {
8239 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8240 ret
= btrfs_dec_ref(trans
, root
, eb
, 1);
8242 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
8243 BUG_ON(ret
); /* -ENOMEM */
8244 ret
= account_leaf_items(trans
, root
, eb
);
8246 printk_ratelimited(KERN_ERR
"BTRFS: %s Error "
8247 "%d accounting leaf items. Quota "
8248 "is out of sync, rescan required.\n",
8249 root
->fs_info
->sb
->s_id
, ret
);
8252 /* make block locked assertion in clean_tree_block happy */
8253 if (!path
->locks
[level
] &&
8254 btrfs_header_generation(eb
) == trans
->transid
) {
8255 btrfs_tree_lock(eb
);
8256 btrfs_set_lock_blocking(eb
);
8257 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8259 clean_tree_block(trans
, root
->fs_info
, eb
);
8262 if (eb
== root
->node
) {
8263 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8266 BUG_ON(root
->root_key
.objectid
!=
8267 btrfs_header_owner(eb
));
8269 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8270 parent
= path
->nodes
[level
+ 1]->start
;
8272 BUG_ON(root
->root_key
.objectid
!=
8273 btrfs_header_owner(path
->nodes
[level
+ 1]));
8276 btrfs_free_tree_block(trans
, root
, eb
, parent
, wc
->refs
[level
] == 1);
8278 wc
->refs
[level
] = 0;
8279 wc
->flags
[level
] = 0;
8283 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
8284 struct btrfs_root
*root
,
8285 struct btrfs_path
*path
,
8286 struct walk_control
*wc
)
8288 int level
= wc
->level
;
8289 int lookup_info
= 1;
8292 while (level
>= 0) {
8293 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
8300 if (path
->slots
[level
] >=
8301 btrfs_header_nritems(path
->nodes
[level
]))
8304 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
8306 path
->slots
[level
]++;
8315 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
8316 struct btrfs_root
*root
,
8317 struct btrfs_path
*path
,
8318 struct walk_control
*wc
, int max_level
)
8320 int level
= wc
->level
;
8323 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
8324 while (level
< max_level
&& path
->nodes
[level
]) {
8326 if (path
->slots
[level
] + 1 <
8327 btrfs_header_nritems(path
->nodes
[level
])) {
8328 path
->slots
[level
]++;
8331 ret
= walk_up_proc(trans
, root
, path
, wc
);
8335 if (path
->locks
[level
]) {
8336 btrfs_tree_unlock_rw(path
->nodes
[level
],
8337 path
->locks
[level
]);
8338 path
->locks
[level
] = 0;
8340 free_extent_buffer(path
->nodes
[level
]);
8341 path
->nodes
[level
] = NULL
;
8349 * drop a subvolume tree.
8351 * this function traverses the tree freeing any blocks that only
8352 * referenced by the tree.
8354 * when a shared tree block is found. this function decreases its
8355 * reference count by one. if update_ref is true, this function
8356 * also make sure backrefs for the shared block and all lower level
8357 * blocks are properly updated.
8359 * If called with for_reloc == 0, may exit early with -EAGAIN
8361 int btrfs_drop_snapshot(struct btrfs_root
*root
,
8362 struct btrfs_block_rsv
*block_rsv
, int update_ref
,
8365 struct btrfs_path
*path
;
8366 struct btrfs_trans_handle
*trans
;
8367 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
8368 struct btrfs_root_item
*root_item
= &root
->root_item
;
8369 struct walk_control
*wc
;
8370 struct btrfs_key key
;
8374 bool root_dropped
= false;
8376 btrfs_debug(root
->fs_info
, "Drop subvolume %llu", root
->objectid
);
8378 path
= btrfs_alloc_path();
8384 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
8386 btrfs_free_path(path
);
8391 trans
= btrfs_start_transaction(tree_root
, 0);
8392 if (IS_ERR(trans
)) {
8393 err
= PTR_ERR(trans
);
8398 trans
->block_rsv
= block_rsv
;
8400 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
8401 level
= btrfs_header_level(root
->node
);
8402 path
->nodes
[level
] = btrfs_lock_root_node(root
);
8403 btrfs_set_lock_blocking(path
->nodes
[level
]);
8404 path
->slots
[level
] = 0;
8405 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8406 memset(&wc
->update_progress
, 0,
8407 sizeof(wc
->update_progress
));
8409 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
8410 memcpy(&wc
->update_progress
, &key
,
8411 sizeof(wc
->update_progress
));
8413 level
= root_item
->drop_level
;
8415 path
->lowest_level
= level
;
8416 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
8417 path
->lowest_level
= 0;
8425 * unlock our path, this is safe because only this
8426 * function is allowed to delete this snapshot
8428 btrfs_unlock_up_safe(path
, 0);
8430 level
= btrfs_header_level(root
->node
);
8432 btrfs_tree_lock(path
->nodes
[level
]);
8433 btrfs_set_lock_blocking(path
->nodes
[level
]);
8434 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8436 ret
= btrfs_lookup_extent_info(trans
, root
,
8437 path
->nodes
[level
]->start
,
8438 level
, 1, &wc
->refs
[level
],
8444 BUG_ON(wc
->refs
[level
] == 0);
8446 if (level
== root_item
->drop_level
)
8449 btrfs_tree_unlock(path
->nodes
[level
]);
8450 path
->locks
[level
] = 0;
8451 WARN_ON(wc
->refs
[level
] != 1);
8457 wc
->shared_level
= -1;
8458 wc
->stage
= DROP_REFERENCE
;
8459 wc
->update_ref
= update_ref
;
8461 wc
->for_reloc
= for_reloc
;
8462 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
8466 ret
= walk_down_tree(trans
, root
, path
, wc
);
8472 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
8479 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
8483 if (wc
->stage
== DROP_REFERENCE
) {
8485 btrfs_node_key(path
->nodes
[level
],
8486 &root_item
->drop_progress
,
8487 path
->slots
[level
]);
8488 root_item
->drop_level
= level
;
8491 BUG_ON(wc
->level
== 0);
8492 if (btrfs_should_end_transaction(trans
, tree_root
) ||
8493 (!for_reloc
&& btrfs_need_cleaner_sleep(root
))) {
8494 ret
= btrfs_update_root(trans
, tree_root
,
8498 btrfs_abort_transaction(trans
, tree_root
, ret
);
8503 btrfs_end_transaction_throttle(trans
, tree_root
);
8504 if (!for_reloc
&& btrfs_need_cleaner_sleep(root
)) {
8505 pr_debug("BTRFS: drop snapshot early exit\n");
8510 trans
= btrfs_start_transaction(tree_root
, 0);
8511 if (IS_ERR(trans
)) {
8512 err
= PTR_ERR(trans
);
8516 trans
->block_rsv
= block_rsv
;
8519 btrfs_release_path(path
);
8523 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
8525 btrfs_abort_transaction(trans
, tree_root
, ret
);
8529 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
8530 ret
= btrfs_find_root(tree_root
, &root
->root_key
, path
,
8533 btrfs_abort_transaction(trans
, tree_root
, ret
);
8536 } else if (ret
> 0) {
8537 /* if we fail to delete the orphan item this time
8538 * around, it'll get picked up the next time.
8540 * The most common failure here is just -ENOENT.
8542 btrfs_del_orphan_item(trans
, tree_root
,
8543 root
->root_key
.objectid
);
8547 if (test_bit(BTRFS_ROOT_IN_RADIX
, &root
->state
)) {
8548 btrfs_drop_and_free_fs_root(tree_root
->fs_info
, root
);
8550 free_extent_buffer(root
->node
);
8551 free_extent_buffer(root
->commit_root
);
8552 btrfs_put_fs_root(root
);
8554 root_dropped
= true;
8556 btrfs_end_transaction_throttle(trans
, tree_root
);
8559 btrfs_free_path(path
);
8562 * So if we need to stop dropping the snapshot for whatever reason we
8563 * need to make sure to add it back to the dead root list so that we
8564 * keep trying to do the work later. This also cleans up roots if we
8565 * don't have it in the radix (like when we recover after a power fail
8566 * or unmount) so we don't leak memory.
8568 if (!for_reloc
&& root_dropped
== false)
8569 btrfs_add_dead_root(root
);
8570 if (err
&& err
!= -EAGAIN
)
8571 btrfs_std_error(root
->fs_info
, err
);
8576 * drop subtree rooted at tree block 'node'.
8578 * NOTE: this function will unlock and release tree block 'node'
8579 * only used by relocation code
8581 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
8582 struct btrfs_root
*root
,
8583 struct extent_buffer
*node
,
8584 struct extent_buffer
*parent
)
8586 struct btrfs_path
*path
;
8587 struct walk_control
*wc
;
8593 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
8595 path
= btrfs_alloc_path();
8599 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
8601 btrfs_free_path(path
);
8605 btrfs_assert_tree_locked(parent
);
8606 parent_level
= btrfs_header_level(parent
);
8607 extent_buffer_get(parent
);
8608 path
->nodes
[parent_level
] = parent
;
8609 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
8611 btrfs_assert_tree_locked(node
);
8612 level
= btrfs_header_level(node
);
8613 path
->nodes
[level
] = node
;
8614 path
->slots
[level
] = 0;
8615 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8617 wc
->refs
[parent_level
] = 1;
8618 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
8620 wc
->shared_level
= -1;
8621 wc
->stage
= DROP_REFERENCE
;
8625 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
8628 wret
= walk_down_tree(trans
, root
, path
, wc
);
8634 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
8642 btrfs_free_path(path
);
8646 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
8652 * if restripe for this chunk_type is on pick target profile and
8653 * return, otherwise do the usual balance
8655 stripped
= get_restripe_target(root
->fs_info
, flags
);
8657 return extended_to_chunk(stripped
);
8659 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
8661 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
8662 BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
|
8663 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
8665 if (num_devices
== 1) {
8666 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
8667 stripped
= flags
& ~stripped
;
8669 /* turn raid0 into single device chunks */
8670 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
8673 /* turn mirroring into duplication */
8674 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
8675 BTRFS_BLOCK_GROUP_RAID10
))
8676 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
8678 /* they already had raid on here, just return */
8679 if (flags
& stripped
)
8682 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
8683 stripped
= flags
& ~stripped
;
8685 /* switch duplicated blocks with raid1 */
8686 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
8687 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
8689 /* this is drive concat, leave it alone */
8695 static int set_block_group_ro(struct btrfs_block_group_cache
*cache
, int force
)
8697 struct btrfs_space_info
*sinfo
= cache
->space_info
;
8699 u64 min_allocable_bytes
;
8704 * We need some metadata space and system metadata space for
8705 * allocating chunks in some corner cases until we force to set
8706 * it to be readonly.
8709 (BTRFS_BLOCK_GROUP_SYSTEM
| BTRFS_BLOCK_GROUP_METADATA
)) &&
8711 min_allocable_bytes
= 1 * 1024 * 1024;
8713 min_allocable_bytes
= 0;
8715 spin_lock(&sinfo
->lock
);
8716 spin_lock(&cache
->lock
);
8723 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
8724 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
8726 if (sinfo
->bytes_used
+ sinfo
->bytes_reserved
+ sinfo
->bytes_pinned
+
8727 sinfo
->bytes_may_use
+ sinfo
->bytes_readonly
+ num_bytes
+
8728 min_allocable_bytes
<= sinfo
->total_bytes
) {
8729 sinfo
->bytes_readonly
+= num_bytes
;
8731 list_add_tail(&cache
->ro_list
, &sinfo
->ro_bgs
);
8735 spin_unlock(&cache
->lock
);
8736 spin_unlock(&sinfo
->lock
);
8740 int btrfs_set_block_group_ro(struct btrfs_root
*root
,
8741 struct btrfs_block_group_cache
*cache
)
8744 struct btrfs_trans_handle
*trans
;
8751 trans
= btrfs_join_transaction(root
);
8753 return PTR_ERR(trans
);
8756 * we're not allowed to set block groups readonly after the dirty
8757 * block groups cache has started writing. If it already started,
8758 * back off and let this transaction commit
8760 mutex_lock(&root
->fs_info
->ro_block_group_mutex
);
8761 if (trans
->transaction
->dirty_bg_run
) {
8762 u64 transid
= trans
->transid
;
8764 mutex_unlock(&root
->fs_info
->ro_block_group_mutex
);
8765 btrfs_end_transaction(trans
, root
);
8767 ret
= btrfs_wait_for_commit(root
, transid
);
8774 * if we are changing raid levels, try to allocate a corresponding
8775 * block group with the new raid level.
8777 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
8778 if (alloc_flags
!= cache
->flags
) {
8779 ret
= do_chunk_alloc(trans
, root
, alloc_flags
,
8782 * ENOSPC is allowed here, we may have enough space
8783 * already allocated at the new raid level to
8792 ret
= set_block_group_ro(cache
, 0);
8795 alloc_flags
= get_alloc_profile(root
, cache
->space_info
->flags
);
8796 ret
= do_chunk_alloc(trans
, root
, alloc_flags
,
8800 ret
= set_block_group_ro(cache
, 0);
8802 if (cache
->flags
& BTRFS_BLOCK_GROUP_SYSTEM
) {
8803 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
8804 lock_chunks(root
->fs_info
->chunk_root
);
8805 check_system_chunk(trans
, root
, alloc_flags
);
8806 unlock_chunks(root
->fs_info
->chunk_root
);
8808 mutex_unlock(&root
->fs_info
->ro_block_group_mutex
);
8810 btrfs_end_transaction(trans
, root
);
8814 int btrfs_force_chunk_alloc(struct btrfs_trans_handle
*trans
,
8815 struct btrfs_root
*root
, u64 type
)
8817 u64 alloc_flags
= get_alloc_profile(root
, type
);
8818 return do_chunk_alloc(trans
, root
, alloc_flags
,
8823 * helper to account the unused space of all the readonly block group in the
8824 * space_info. takes mirrors into account.
8826 u64
btrfs_account_ro_block_groups_free_space(struct btrfs_space_info
*sinfo
)
8828 struct btrfs_block_group_cache
*block_group
;
8832 /* It's df, we don't care if it's racey */
8833 if (list_empty(&sinfo
->ro_bgs
))
8836 spin_lock(&sinfo
->lock
);
8837 list_for_each_entry(block_group
, &sinfo
->ro_bgs
, ro_list
) {
8838 spin_lock(&block_group
->lock
);
8840 if (!block_group
->ro
) {
8841 spin_unlock(&block_group
->lock
);
8845 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_RAID1
|
8846 BTRFS_BLOCK_GROUP_RAID10
|
8847 BTRFS_BLOCK_GROUP_DUP
))
8852 free_bytes
+= (block_group
->key
.offset
-
8853 btrfs_block_group_used(&block_group
->item
)) *
8856 spin_unlock(&block_group
->lock
);
8858 spin_unlock(&sinfo
->lock
);
8863 void btrfs_set_block_group_rw(struct btrfs_root
*root
,
8864 struct btrfs_block_group_cache
*cache
)
8866 struct btrfs_space_info
*sinfo
= cache
->space_info
;
8871 spin_lock(&sinfo
->lock
);
8872 spin_lock(&cache
->lock
);
8873 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
8874 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
8875 sinfo
->bytes_readonly
-= num_bytes
;
8877 list_del_init(&cache
->ro_list
);
8878 spin_unlock(&cache
->lock
);
8879 spin_unlock(&sinfo
->lock
);
8883 * checks to see if its even possible to relocate this block group.
8885 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8886 * ok to go ahead and try.
8888 int btrfs_can_relocate(struct btrfs_root
*root
, u64 bytenr
)
8890 struct btrfs_block_group_cache
*block_group
;
8891 struct btrfs_space_info
*space_info
;
8892 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
8893 struct btrfs_device
*device
;
8894 struct btrfs_trans_handle
*trans
;
8903 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
8905 /* odd, couldn't find the block group, leave it alone */
8909 min_free
= btrfs_block_group_used(&block_group
->item
);
8911 /* no bytes used, we're good */
8915 space_info
= block_group
->space_info
;
8916 spin_lock(&space_info
->lock
);
8918 full
= space_info
->full
;
8921 * if this is the last block group we have in this space, we can't
8922 * relocate it unless we're able to allocate a new chunk below.
8924 * Otherwise, we need to make sure we have room in the space to handle
8925 * all of the extents from this block group. If we can, we're good
8927 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
8928 (space_info
->bytes_used
+ space_info
->bytes_reserved
+
8929 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
8930 min_free
< space_info
->total_bytes
)) {
8931 spin_unlock(&space_info
->lock
);
8934 spin_unlock(&space_info
->lock
);
8937 * ok we don't have enough space, but maybe we have free space on our
8938 * devices to allocate new chunks for relocation, so loop through our
8939 * alloc devices and guess if we have enough space. if this block
8940 * group is going to be restriped, run checks against the target
8941 * profile instead of the current one.
8953 target
= get_restripe_target(root
->fs_info
, block_group
->flags
);
8955 index
= __get_raid_index(extended_to_chunk(target
));
8958 * this is just a balance, so if we were marked as full
8959 * we know there is no space for a new chunk
8964 index
= get_block_group_index(block_group
);
8967 if (index
== BTRFS_RAID_RAID10
) {
8971 } else if (index
== BTRFS_RAID_RAID1
) {
8973 } else if (index
== BTRFS_RAID_DUP
) {
8976 } else if (index
== BTRFS_RAID_RAID0
) {
8977 dev_min
= fs_devices
->rw_devices
;
8978 min_free
= div64_u64(min_free
, dev_min
);
8981 /* We need to do this so that we can look at pending chunks */
8982 trans
= btrfs_join_transaction(root
);
8983 if (IS_ERR(trans
)) {
8984 ret
= PTR_ERR(trans
);
8988 mutex_lock(&root
->fs_info
->chunk_mutex
);
8989 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
8993 * check to make sure we can actually find a chunk with enough
8994 * space to fit our block group in.
8996 if (device
->total_bytes
> device
->bytes_used
+ min_free
&&
8997 !device
->is_tgtdev_for_dev_replace
) {
8998 ret
= find_free_dev_extent(trans
, device
, min_free
,
9003 if (dev_nr
>= dev_min
)
9009 mutex_unlock(&root
->fs_info
->chunk_mutex
);
9010 btrfs_end_transaction(trans
, root
);
9012 btrfs_put_block_group(block_group
);
9016 static int find_first_block_group(struct btrfs_root
*root
,
9017 struct btrfs_path
*path
, struct btrfs_key
*key
)
9020 struct btrfs_key found_key
;
9021 struct extent_buffer
*leaf
;
9024 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
9029 slot
= path
->slots
[0];
9030 leaf
= path
->nodes
[0];
9031 if (slot
>= btrfs_header_nritems(leaf
)) {
9032 ret
= btrfs_next_leaf(root
, path
);
9039 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
9041 if (found_key
.objectid
>= key
->objectid
&&
9042 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
9052 void btrfs_put_block_group_cache(struct btrfs_fs_info
*info
)
9054 struct btrfs_block_group_cache
*block_group
;
9058 struct inode
*inode
;
9060 block_group
= btrfs_lookup_first_block_group(info
, last
);
9061 while (block_group
) {
9062 spin_lock(&block_group
->lock
);
9063 if (block_group
->iref
)
9065 spin_unlock(&block_group
->lock
);
9066 block_group
= next_block_group(info
->tree_root
,
9076 inode
= block_group
->inode
;
9077 block_group
->iref
= 0;
9078 block_group
->inode
= NULL
;
9079 spin_unlock(&block_group
->lock
);
9081 last
= block_group
->key
.objectid
+ block_group
->key
.offset
;
9082 btrfs_put_block_group(block_group
);
9086 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
9088 struct btrfs_block_group_cache
*block_group
;
9089 struct btrfs_space_info
*space_info
;
9090 struct btrfs_caching_control
*caching_ctl
;
9093 down_write(&info
->commit_root_sem
);
9094 while (!list_empty(&info
->caching_block_groups
)) {
9095 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
9096 struct btrfs_caching_control
, list
);
9097 list_del(&caching_ctl
->list
);
9098 put_caching_control(caching_ctl
);
9100 up_write(&info
->commit_root_sem
);
9102 spin_lock(&info
->unused_bgs_lock
);
9103 while (!list_empty(&info
->unused_bgs
)) {
9104 block_group
= list_first_entry(&info
->unused_bgs
,
9105 struct btrfs_block_group_cache
,
9107 list_del_init(&block_group
->bg_list
);
9108 btrfs_put_block_group(block_group
);
9110 spin_unlock(&info
->unused_bgs_lock
);
9112 spin_lock(&info
->block_group_cache_lock
);
9113 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
9114 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
9116 rb_erase(&block_group
->cache_node
,
9117 &info
->block_group_cache_tree
);
9118 RB_CLEAR_NODE(&block_group
->cache_node
);
9119 spin_unlock(&info
->block_group_cache_lock
);
9121 down_write(&block_group
->space_info
->groups_sem
);
9122 list_del(&block_group
->list
);
9123 up_write(&block_group
->space_info
->groups_sem
);
9125 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
9126 wait_block_group_cache_done(block_group
);
9129 * We haven't cached this block group, which means we could
9130 * possibly have excluded extents on this block group.
9132 if (block_group
->cached
== BTRFS_CACHE_NO
||
9133 block_group
->cached
== BTRFS_CACHE_ERROR
)
9134 free_excluded_extents(info
->extent_root
, block_group
);
9136 btrfs_remove_free_space_cache(block_group
);
9137 btrfs_put_block_group(block_group
);
9139 spin_lock(&info
->block_group_cache_lock
);
9141 spin_unlock(&info
->block_group_cache_lock
);
9143 /* now that all the block groups are freed, go through and
9144 * free all the space_info structs. This is only called during
9145 * the final stages of unmount, and so we know nobody is
9146 * using them. We call synchronize_rcu() once before we start,
9147 * just to be on the safe side.
9151 release_global_block_rsv(info
);
9153 while (!list_empty(&info
->space_info
)) {
9156 space_info
= list_entry(info
->space_info
.next
,
9157 struct btrfs_space_info
,
9159 if (btrfs_test_opt(info
->tree_root
, ENOSPC_DEBUG
)) {
9160 if (WARN_ON(space_info
->bytes_pinned
> 0 ||
9161 space_info
->bytes_reserved
> 0 ||
9162 space_info
->bytes_may_use
> 0)) {
9163 dump_space_info(space_info
, 0, 0);
9166 list_del(&space_info
->list
);
9167 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++) {
9168 struct kobject
*kobj
;
9169 kobj
= space_info
->block_group_kobjs
[i
];
9170 space_info
->block_group_kobjs
[i
] = NULL
;
9176 kobject_del(&space_info
->kobj
);
9177 kobject_put(&space_info
->kobj
);
9182 static void __link_block_group(struct btrfs_space_info
*space_info
,
9183 struct btrfs_block_group_cache
*cache
)
9185 int index
= get_block_group_index(cache
);
9188 down_write(&space_info
->groups_sem
);
9189 if (list_empty(&space_info
->block_groups
[index
]))
9191 list_add_tail(&cache
->list
, &space_info
->block_groups
[index
]);
9192 up_write(&space_info
->groups_sem
);
9195 struct raid_kobject
*rkobj
;
9198 rkobj
= kzalloc(sizeof(*rkobj
), GFP_NOFS
);
9201 rkobj
->raid_type
= index
;
9202 kobject_init(&rkobj
->kobj
, &btrfs_raid_ktype
);
9203 ret
= kobject_add(&rkobj
->kobj
, &space_info
->kobj
,
9204 "%s", get_raid_name(index
));
9206 kobject_put(&rkobj
->kobj
);
9209 space_info
->block_group_kobjs
[index
] = &rkobj
->kobj
;
9214 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
9217 static struct btrfs_block_group_cache
*
9218 btrfs_create_block_group_cache(struct btrfs_root
*root
, u64 start
, u64 size
)
9220 struct btrfs_block_group_cache
*cache
;
9222 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
9226 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
9228 if (!cache
->free_space_ctl
) {
9233 cache
->key
.objectid
= start
;
9234 cache
->key
.offset
= size
;
9235 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
9237 cache
->sectorsize
= root
->sectorsize
;
9238 cache
->fs_info
= root
->fs_info
;
9239 cache
->full_stripe_len
= btrfs_full_stripe_len(root
,
9240 &root
->fs_info
->mapping_tree
,
9242 atomic_set(&cache
->count
, 1);
9243 spin_lock_init(&cache
->lock
);
9244 init_rwsem(&cache
->data_rwsem
);
9245 INIT_LIST_HEAD(&cache
->list
);
9246 INIT_LIST_HEAD(&cache
->cluster_list
);
9247 INIT_LIST_HEAD(&cache
->bg_list
);
9248 INIT_LIST_HEAD(&cache
->ro_list
);
9249 INIT_LIST_HEAD(&cache
->dirty_list
);
9250 INIT_LIST_HEAD(&cache
->io_list
);
9251 btrfs_init_free_space_ctl(cache
);
9252 atomic_set(&cache
->trimming
, 0);
9257 int btrfs_read_block_groups(struct btrfs_root
*root
)
9259 struct btrfs_path
*path
;
9261 struct btrfs_block_group_cache
*cache
;
9262 struct btrfs_fs_info
*info
= root
->fs_info
;
9263 struct btrfs_space_info
*space_info
;
9264 struct btrfs_key key
;
9265 struct btrfs_key found_key
;
9266 struct extent_buffer
*leaf
;
9270 root
= info
->extent_root
;
9273 key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
9274 path
= btrfs_alloc_path();
9279 cache_gen
= btrfs_super_cache_generation(root
->fs_info
->super_copy
);
9280 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
9281 btrfs_super_generation(root
->fs_info
->super_copy
) != cache_gen
)
9283 if (btrfs_test_opt(root
, CLEAR_CACHE
))
9287 ret
= find_first_block_group(root
, path
, &key
);
9293 leaf
= path
->nodes
[0];
9294 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
9296 cache
= btrfs_create_block_group_cache(root
, found_key
.objectid
,
9305 * When we mount with old space cache, we need to
9306 * set BTRFS_DC_CLEAR and set dirty flag.
9308 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9309 * truncate the old free space cache inode and
9311 * b) Setting 'dirty flag' makes sure that we flush
9312 * the new space cache info onto disk.
9314 if (btrfs_test_opt(root
, SPACE_CACHE
))
9315 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
9318 read_extent_buffer(leaf
, &cache
->item
,
9319 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
9320 sizeof(cache
->item
));
9321 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
9323 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
9324 btrfs_release_path(path
);
9327 * We need to exclude the super stripes now so that the space
9328 * info has super bytes accounted for, otherwise we'll think
9329 * we have more space than we actually do.
9331 ret
= exclude_super_stripes(root
, cache
);
9334 * We may have excluded something, so call this just in
9337 free_excluded_extents(root
, cache
);
9338 btrfs_put_block_group(cache
);
9343 * check for two cases, either we are full, and therefore
9344 * don't need to bother with the caching work since we won't
9345 * find any space, or we are empty, and we can just add all
9346 * the space in and be done with it. This saves us _alot_ of
9347 * time, particularly in the full case.
9349 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
9350 cache
->last_byte_to_unpin
= (u64
)-1;
9351 cache
->cached
= BTRFS_CACHE_FINISHED
;
9352 free_excluded_extents(root
, cache
);
9353 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
9354 cache
->last_byte_to_unpin
= (u64
)-1;
9355 cache
->cached
= BTRFS_CACHE_FINISHED
;
9356 add_new_free_space(cache
, root
->fs_info
,
9358 found_key
.objectid
+
9360 free_excluded_extents(root
, cache
);
9363 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
9365 btrfs_remove_free_space_cache(cache
);
9366 btrfs_put_block_group(cache
);
9370 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
9371 btrfs_block_group_used(&cache
->item
),
9374 btrfs_remove_free_space_cache(cache
);
9375 spin_lock(&info
->block_group_cache_lock
);
9376 rb_erase(&cache
->cache_node
,
9377 &info
->block_group_cache_tree
);
9378 RB_CLEAR_NODE(&cache
->cache_node
);
9379 spin_unlock(&info
->block_group_cache_lock
);
9380 btrfs_put_block_group(cache
);
9384 cache
->space_info
= space_info
;
9385 spin_lock(&cache
->space_info
->lock
);
9386 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
9387 spin_unlock(&cache
->space_info
->lock
);
9389 __link_block_group(space_info
, cache
);
9391 set_avail_alloc_bits(root
->fs_info
, cache
->flags
);
9392 if (btrfs_chunk_readonly(root
, cache
->key
.objectid
)) {
9393 set_block_group_ro(cache
, 1);
9394 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
9395 spin_lock(&info
->unused_bgs_lock
);
9396 /* Should always be true but just in case. */
9397 if (list_empty(&cache
->bg_list
)) {
9398 btrfs_get_block_group(cache
);
9399 list_add_tail(&cache
->bg_list
,
9402 spin_unlock(&info
->unused_bgs_lock
);
9406 list_for_each_entry_rcu(space_info
, &root
->fs_info
->space_info
, list
) {
9407 if (!(get_alloc_profile(root
, space_info
->flags
) &
9408 (BTRFS_BLOCK_GROUP_RAID10
|
9409 BTRFS_BLOCK_GROUP_RAID1
|
9410 BTRFS_BLOCK_GROUP_RAID5
|
9411 BTRFS_BLOCK_GROUP_RAID6
|
9412 BTRFS_BLOCK_GROUP_DUP
)))
9415 * avoid allocating from un-mirrored block group if there are
9416 * mirrored block groups.
9418 list_for_each_entry(cache
,
9419 &space_info
->block_groups
[BTRFS_RAID_RAID0
],
9421 set_block_group_ro(cache
, 1);
9422 list_for_each_entry(cache
,
9423 &space_info
->block_groups
[BTRFS_RAID_SINGLE
],
9425 set_block_group_ro(cache
, 1);
9428 init_global_block_rsv(info
);
9431 btrfs_free_path(path
);
9435 void btrfs_create_pending_block_groups(struct btrfs_trans_handle
*trans
,
9436 struct btrfs_root
*root
)
9438 struct btrfs_block_group_cache
*block_group
, *tmp
;
9439 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
9440 struct btrfs_block_group_item item
;
9441 struct btrfs_key key
;
9444 list_for_each_entry_safe(block_group
, tmp
, &trans
->new_bgs
, bg_list
) {
9448 spin_lock(&block_group
->lock
);
9449 memcpy(&item
, &block_group
->item
, sizeof(item
));
9450 memcpy(&key
, &block_group
->key
, sizeof(key
));
9451 spin_unlock(&block_group
->lock
);
9453 ret
= btrfs_insert_item(trans
, extent_root
, &key
, &item
,
9456 btrfs_abort_transaction(trans
, extent_root
, ret
);
9457 ret
= btrfs_finish_chunk_alloc(trans
, extent_root
,
9458 key
.objectid
, key
.offset
);
9460 btrfs_abort_transaction(trans
, extent_root
, ret
);
9462 list_del_init(&block_group
->bg_list
);
9466 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
9467 struct btrfs_root
*root
, u64 bytes_used
,
9468 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
9472 struct btrfs_root
*extent_root
;
9473 struct btrfs_block_group_cache
*cache
;
9475 extent_root
= root
->fs_info
->extent_root
;
9477 btrfs_set_log_full_commit(root
->fs_info
, trans
);
9479 cache
= btrfs_create_block_group_cache(root
, chunk_offset
, size
);
9483 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
9484 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
9485 btrfs_set_block_group_flags(&cache
->item
, type
);
9487 cache
->flags
= type
;
9488 cache
->last_byte_to_unpin
= (u64
)-1;
9489 cache
->cached
= BTRFS_CACHE_FINISHED
;
9490 ret
= exclude_super_stripes(root
, cache
);
9493 * We may have excluded something, so call this just in
9496 free_excluded_extents(root
, cache
);
9497 btrfs_put_block_group(cache
);
9501 add_new_free_space(cache
, root
->fs_info
, chunk_offset
,
9502 chunk_offset
+ size
);
9504 free_excluded_extents(root
, cache
);
9507 * Call to ensure the corresponding space_info object is created and
9508 * assigned to our block group, but don't update its counters just yet.
9509 * We want our bg to be added to the rbtree with its ->space_info set.
9511 ret
= update_space_info(root
->fs_info
, cache
->flags
, 0, 0,
9512 &cache
->space_info
);
9514 btrfs_remove_free_space_cache(cache
);
9515 btrfs_put_block_group(cache
);
9519 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
9521 btrfs_remove_free_space_cache(cache
);
9522 btrfs_put_block_group(cache
);
9527 * Now that our block group has its ->space_info set and is inserted in
9528 * the rbtree, update the space info's counters.
9530 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
9531 &cache
->space_info
);
9533 btrfs_remove_free_space_cache(cache
);
9534 spin_lock(&root
->fs_info
->block_group_cache_lock
);
9535 rb_erase(&cache
->cache_node
,
9536 &root
->fs_info
->block_group_cache_tree
);
9537 RB_CLEAR_NODE(&cache
->cache_node
);
9538 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
9539 btrfs_put_block_group(cache
);
9542 update_global_block_rsv(root
->fs_info
);
9544 spin_lock(&cache
->space_info
->lock
);
9545 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
9546 spin_unlock(&cache
->space_info
->lock
);
9548 __link_block_group(cache
->space_info
, cache
);
9550 list_add_tail(&cache
->bg_list
, &trans
->new_bgs
);
9552 set_avail_alloc_bits(extent_root
->fs_info
, type
);
9557 static void clear_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
9559 u64 extra_flags
= chunk_to_extended(flags
) &
9560 BTRFS_EXTENDED_PROFILE_MASK
;
9562 write_seqlock(&fs_info
->profiles_lock
);
9563 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
9564 fs_info
->avail_data_alloc_bits
&= ~extra_flags
;
9565 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
9566 fs_info
->avail_metadata_alloc_bits
&= ~extra_flags
;
9567 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
9568 fs_info
->avail_system_alloc_bits
&= ~extra_flags
;
9569 write_sequnlock(&fs_info
->profiles_lock
);
9572 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
9573 struct btrfs_root
*root
, u64 group_start
,
9574 struct extent_map
*em
)
9576 struct btrfs_path
*path
;
9577 struct btrfs_block_group_cache
*block_group
;
9578 struct btrfs_free_cluster
*cluster
;
9579 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
9580 struct btrfs_key key
;
9581 struct inode
*inode
;
9582 struct kobject
*kobj
= NULL
;
9586 struct btrfs_caching_control
*caching_ctl
= NULL
;
9589 root
= root
->fs_info
->extent_root
;
9591 block_group
= btrfs_lookup_block_group(root
->fs_info
, group_start
);
9592 BUG_ON(!block_group
);
9593 BUG_ON(!block_group
->ro
);
9596 * Free the reserved super bytes from this block group before
9599 free_excluded_extents(root
, block_group
);
9601 memcpy(&key
, &block_group
->key
, sizeof(key
));
9602 index
= get_block_group_index(block_group
);
9603 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
9604 BTRFS_BLOCK_GROUP_RAID1
|
9605 BTRFS_BLOCK_GROUP_RAID10
))
9610 /* make sure this block group isn't part of an allocation cluster */
9611 cluster
= &root
->fs_info
->data_alloc_cluster
;
9612 spin_lock(&cluster
->refill_lock
);
9613 btrfs_return_cluster_to_free_space(block_group
, cluster
);
9614 spin_unlock(&cluster
->refill_lock
);
9617 * make sure this block group isn't part of a metadata
9618 * allocation cluster
9620 cluster
= &root
->fs_info
->meta_alloc_cluster
;
9621 spin_lock(&cluster
->refill_lock
);
9622 btrfs_return_cluster_to_free_space(block_group
, cluster
);
9623 spin_unlock(&cluster
->refill_lock
);
9625 path
= btrfs_alloc_path();
9632 * get the inode first so any iput calls done for the io_list
9633 * aren't the final iput (no unlinks allowed now)
9635 inode
= lookup_free_space_inode(tree_root
, block_group
, path
);
9637 mutex_lock(&trans
->transaction
->cache_write_mutex
);
9639 * make sure our free spache cache IO is done before remove the
9642 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
9643 if (!list_empty(&block_group
->io_list
)) {
9644 list_del_init(&block_group
->io_list
);
9646 WARN_ON(!IS_ERR(inode
) && inode
!= block_group
->io_ctl
.inode
);
9648 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
9649 btrfs_wait_cache_io(root
, trans
, block_group
,
9650 &block_group
->io_ctl
, path
,
9651 block_group
->key
.objectid
);
9652 btrfs_put_block_group(block_group
);
9653 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
9656 if (!list_empty(&block_group
->dirty_list
)) {
9657 list_del_init(&block_group
->dirty_list
);
9658 btrfs_put_block_group(block_group
);
9660 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
9661 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
9663 if (!IS_ERR(inode
)) {
9664 ret
= btrfs_orphan_add(trans
, inode
);
9666 btrfs_add_delayed_iput(inode
);
9670 /* One for the block groups ref */
9671 spin_lock(&block_group
->lock
);
9672 if (block_group
->iref
) {
9673 block_group
->iref
= 0;
9674 block_group
->inode
= NULL
;
9675 spin_unlock(&block_group
->lock
);
9678 spin_unlock(&block_group
->lock
);
9680 /* One for our lookup ref */
9681 btrfs_add_delayed_iput(inode
);
9684 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
9685 key
.offset
= block_group
->key
.objectid
;
9688 ret
= btrfs_search_slot(trans
, tree_root
, &key
, path
, -1, 1);
9692 btrfs_release_path(path
);
9694 ret
= btrfs_del_item(trans
, tree_root
, path
);
9697 btrfs_release_path(path
);
9700 spin_lock(&root
->fs_info
->block_group_cache_lock
);
9701 rb_erase(&block_group
->cache_node
,
9702 &root
->fs_info
->block_group_cache_tree
);
9703 RB_CLEAR_NODE(&block_group
->cache_node
);
9705 if (root
->fs_info
->first_logical_byte
== block_group
->key
.objectid
)
9706 root
->fs_info
->first_logical_byte
= (u64
)-1;
9707 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
9709 down_write(&block_group
->space_info
->groups_sem
);
9711 * we must use list_del_init so people can check to see if they
9712 * are still on the list after taking the semaphore
9714 list_del_init(&block_group
->list
);
9715 if (list_empty(&block_group
->space_info
->block_groups
[index
])) {
9716 kobj
= block_group
->space_info
->block_group_kobjs
[index
];
9717 block_group
->space_info
->block_group_kobjs
[index
] = NULL
;
9718 clear_avail_alloc_bits(root
->fs_info
, block_group
->flags
);
9720 up_write(&block_group
->space_info
->groups_sem
);
9726 if (block_group
->has_caching_ctl
)
9727 caching_ctl
= get_caching_control(block_group
);
9728 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
9729 wait_block_group_cache_done(block_group
);
9730 if (block_group
->has_caching_ctl
) {
9731 down_write(&root
->fs_info
->commit_root_sem
);
9733 struct btrfs_caching_control
*ctl
;
9735 list_for_each_entry(ctl
,
9736 &root
->fs_info
->caching_block_groups
, list
)
9737 if (ctl
->block_group
== block_group
) {
9739 atomic_inc(&caching_ctl
->count
);
9744 list_del_init(&caching_ctl
->list
);
9745 up_write(&root
->fs_info
->commit_root_sem
);
9747 /* Once for the caching bgs list and once for us. */
9748 put_caching_control(caching_ctl
);
9749 put_caching_control(caching_ctl
);
9753 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
9754 if (!list_empty(&block_group
->dirty_list
)) {
9757 if (!list_empty(&block_group
->io_list
)) {
9760 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
9761 btrfs_remove_free_space_cache(block_group
);
9763 spin_lock(&block_group
->space_info
->lock
);
9764 list_del_init(&block_group
->ro_list
);
9766 if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
9767 WARN_ON(block_group
->space_info
->total_bytes
9768 < block_group
->key
.offset
);
9769 WARN_ON(block_group
->space_info
->bytes_readonly
9770 < block_group
->key
.offset
);
9771 WARN_ON(block_group
->space_info
->disk_total
9772 < block_group
->key
.offset
* factor
);
9774 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
9775 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
9776 block_group
->space_info
->disk_total
-= block_group
->key
.offset
* factor
;
9778 spin_unlock(&block_group
->space_info
->lock
);
9780 memcpy(&key
, &block_group
->key
, sizeof(key
));
9783 if (!list_empty(&em
->list
)) {
9784 /* We're in the transaction->pending_chunks list. */
9785 free_extent_map(em
);
9787 spin_lock(&block_group
->lock
);
9788 block_group
->removed
= 1;
9790 * At this point trimming can't start on this block group, because we
9791 * removed the block group from the tree fs_info->block_group_cache_tree
9792 * so no one can't find it anymore and even if someone already got this
9793 * block group before we removed it from the rbtree, they have already
9794 * incremented block_group->trimming - if they didn't, they won't find
9795 * any free space entries because we already removed them all when we
9796 * called btrfs_remove_free_space_cache().
9798 * And we must not remove the extent map from the fs_info->mapping_tree
9799 * to prevent the same logical address range and physical device space
9800 * ranges from being reused for a new block group. This is because our
9801 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
9802 * completely transactionless, so while it is trimming a range the
9803 * currently running transaction might finish and a new one start,
9804 * allowing for new block groups to be created that can reuse the same
9805 * physical device locations unless we take this special care.
9807 remove_em
= (atomic_read(&block_group
->trimming
) == 0);
9809 * Make sure a trimmer task always sees the em in the pinned_chunks list
9810 * if it sees block_group->removed == 1 (needs to lock block_group->lock
9811 * before checking block_group->removed).
9815 * Our em might be in trans->transaction->pending_chunks which
9816 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
9817 * and so is the fs_info->pinned_chunks list.
9819 * So at this point we must be holding the chunk_mutex to avoid
9820 * any races with chunk allocation (more specifically at
9821 * volumes.c:contains_pending_extent()), to ensure it always
9822 * sees the em, either in the pending_chunks list or in the
9823 * pinned_chunks list.
9825 list_move_tail(&em
->list
, &root
->fs_info
->pinned_chunks
);
9827 spin_unlock(&block_group
->lock
);
9830 struct extent_map_tree
*em_tree
;
9832 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
9833 write_lock(&em_tree
->lock
);
9835 * The em might be in the pending_chunks list, so make sure the
9836 * chunk mutex is locked, since remove_extent_mapping() will
9837 * delete us from that list.
9839 remove_extent_mapping(em_tree
, em
);
9840 write_unlock(&em_tree
->lock
);
9841 /* once for the tree */
9842 free_extent_map(em
);
9845 unlock_chunks(root
);
9847 btrfs_put_block_group(block_group
);
9848 btrfs_put_block_group(block_group
);
9850 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
9856 ret
= btrfs_del_item(trans
, root
, path
);
9858 btrfs_free_path(path
);
9863 * Process the unused_bgs list and remove any that don't have any allocated
9864 * space inside of them.
9866 void btrfs_delete_unused_bgs(struct btrfs_fs_info
*fs_info
)
9868 struct btrfs_block_group_cache
*block_group
;
9869 struct btrfs_space_info
*space_info
;
9870 struct btrfs_root
*root
= fs_info
->extent_root
;
9871 struct btrfs_trans_handle
*trans
;
9877 spin_lock(&fs_info
->unused_bgs_lock
);
9878 while (!list_empty(&fs_info
->unused_bgs
)) {
9881 block_group
= list_first_entry(&fs_info
->unused_bgs
,
9882 struct btrfs_block_group_cache
,
9884 space_info
= block_group
->space_info
;
9885 list_del_init(&block_group
->bg_list
);
9886 if (ret
|| btrfs_mixed_space_info(space_info
)) {
9887 btrfs_put_block_group(block_group
);
9890 spin_unlock(&fs_info
->unused_bgs_lock
);
9892 mutex_lock(&root
->fs_info
->delete_unused_bgs_mutex
);
9894 /* Don't want to race with allocators so take the groups_sem */
9895 down_write(&space_info
->groups_sem
);
9896 spin_lock(&block_group
->lock
);
9897 if (block_group
->reserved
||
9898 btrfs_block_group_used(&block_group
->item
) ||
9901 * We want to bail if we made new allocations or have
9902 * outstanding allocations in this block group. We do
9903 * the ro check in case balance is currently acting on
9906 spin_unlock(&block_group
->lock
);
9907 up_write(&space_info
->groups_sem
);
9910 spin_unlock(&block_group
->lock
);
9912 /* We don't want to force the issue, only flip if it's ok. */
9913 ret
= set_block_group_ro(block_group
, 0);
9914 up_write(&space_info
->groups_sem
);
9921 * Want to do this before we do anything else so we can recover
9922 * properly if we fail to join the transaction.
9924 /* 1 for btrfs_orphan_reserve_metadata() */
9925 trans
= btrfs_start_transaction(root
, 1);
9926 if (IS_ERR(trans
)) {
9927 btrfs_set_block_group_rw(root
, block_group
);
9928 ret
= PTR_ERR(trans
);
9933 * We could have pending pinned extents for this block group,
9934 * just delete them, we don't care about them anymore.
9936 start
= block_group
->key
.objectid
;
9937 end
= start
+ block_group
->key
.offset
- 1;
9939 * Hold the unused_bg_unpin_mutex lock to avoid racing with
9940 * btrfs_finish_extent_commit(). If we are at transaction N,
9941 * another task might be running finish_extent_commit() for the
9942 * previous transaction N - 1, and have seen a range belonging
9943 * to the block group in freed_extents[] before we were able to
9944 * clear the whole block group range from freed_extents[]. This
9945 * means that task can lookup for the block group after we
9946 * unpinned it from freed_extents[] and removed it, leading to
9947 * a BUG_ON() at btrfs_unpin_extent_range().
9949 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
9950 ret
= clear_extent_bits(&fs_info
->freed_extents
[0], start
, end
,
9951 EXTENT_DIRTY
, GFP_NOFS
);
9953 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
9954 btrfs_set_block_group_rw(root
, block_group
);
9957 ret
= clear_extent_bits(&fs_info
->freed_extents
[1], start
, end
,
9958 EXTENT_DIRTY
, GFP_NOFS
);
9960 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
9961 btrfs_set_block_group_rw(root
, block_group
);
9964 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
9966 /* Reset pinned so btrfs_put_block_group doesn't complain */
9967 spin_lock(&space_info
->lock
);
9968 spin_lock(&block_group
->lock
);
9970 space_info
->bytes_pinned
-= block_group
->pinned
;
9971 space_info
->bytes_readonly
+= block_group
->pinned
;
9972 percpu_counter_add(&space_info
->total_bytes_pinned
,
9973 -block_group
->pinned
);
9974 block_group
->pinned
= 0;
9976 spin_unlock(&block_group
->lock
);
9977 spin_unlock(&space_info
->lock
);
9980 * Btrfs_remove_chunk will abort the transaction if things go
9983 ret
= btrfs_remove_chunk(trans
, root
,
9984 block_group
->key
.objectid
);
9986 btrfs_end_transaction(trans
, root
);
9988 mutex_unlock(&root
->fs_info
->delete_unused_bgs_mutex
);
9989 btrfs_put_block_group(block_group
);
9990 spin_lock(&fs_info
->unused_bgs_lock
);
9992 spin_unlock(&fs_info
->unused_bgs_lock
);
9995 int btrfs_init_space_info(struct btrfs_fs_info
*fs_info
)
9997 struct btrfs_space_info
*space_info
;
9998 struct btrfs_super_block
*disk_super
;
10004 disk_super
= fs_info
->super_copy
;
10005 if (!btrfs_super_root(disk_super
))
10008 features
= btrfs_super_incompat_flags(disk_super
);
10009 if (features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
10012 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
10013 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
10018 flags
= BTRFS_BLOCK_GROUP_METADATA
| BTRFS_BLOCK_GROUP_DATA
;
10019 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
10021 flags
= BTRFS_BLOCK_GROUP_METADATA
;
10022 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
10026 flags
= BTRFS_BLOCK_GROUP_DATA
;
10027 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
10033 int btrfs_error_unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
10035 return unpin_extent_range(root
, start
, end
, false);
10038 int btrfs_trim_fs(struct btrfs_root
*root
, struct fstrim_range
*range
)
10040 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
10041 struct btrfs_block_group_cache
*cache
= NULL
;
10046 u64 total_bytes
= btrfs_super_total_bytes(fs_info
->super_copy
);
10050 * try to trim all FS space, our block group may start from non-zero.
10052 if (range
->len
== total_bytes
)
10053 cache
= btrfs_lookup_first_block_group(fs_info
, range
->start
);
10055 cache
= btrfs_lookup_block_group(fs_info
, range
->start
);
10058 if (cache
->key
.objectid
>= (range
->start
+ range
->len
)) {
10059 btrfs_put_block_group(cache
);
10063 start
= max(range
->start
, cache
->key
.objectid
);
10064 end
= min(range
->start
+ range
->len
,
10065 cache
->key
.objectid
+ cache
->key
.offset
);
10067 if (end
- start
>= range
->minlen
) {
10068 if (!block_group_cache_done(cache
)) {
10069 ret
= cache_block_group(cache
, 0);
10071 btrfs_put_block_group(cache
);
10074 ret
= wait_block_group_cache_done(cache
);
10076 btrfs_put_block_group(cache
);
10080 ret
= btrfs_trim_block_group(cache
,
10086 trimmed
+= group_trimmed
;
10088 btrfs_put_block_group(cache
);
10093 cache
= next_block_group(fs_info
->tree_root
, cache
);
10096 range
->len
= trimmed
;
10101 * btrfs_{start,end}_write_no_snapshoting() are similar to
10102 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10103 * data into the page cache through nocow before the subvolume is snapshoted,
10104 * but flush the data into disk after the snapshot creation, or to prevent
10105 * operations while snapshoting is ongoing and that cause the snapshot to be
10106 * inconsistent (writes followed by expanding truncates for example).
10108 void btrfs_end_write_no_snapshoting(struct btrfs_root
*root
)
10110 percpu_counter_dec(&root
->subv_writers
->counter
);
10112 * Make sure counter is updated before we wake up
10116 if (waitqueue_active(&root
->subv_writers
->wait
))
10117 wake_up(&root
->subv_writers
->wait
);
10120 int btrfs_start_write_no_snapshoting(struct btrfs_root
*root
)
10122 if (atomic_read(&root
->will_be_snapshoted
))
10125 percpu_counter_inc(&root
->subv_writers
->counter
);
10127 * Make sure counter is updated before we check for snapshot creation.
10130 if (atomic_read(&root
->will_be_snapshoted
)) {
10131 btrfs_end_write_no_snapshoting(root
);