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_path
*path
,
1320 struct btrfs_extent_inline_ref
*iref
)
1322 struct btrfs_key key
;
1323 struct extent_buffer
*leaf
;
1324 struct btrfs_extent_data_ref
*ref1
;
1325 struct btrfs_shared_data_ref
*ref2
;
1328 leaf
= path
->nodes
[0];
1329 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1331 if (btrfs_extent_inline_ref_type(leaf
, iref
) ==
1332 BTRFS_EXTENT_DATA_REF_KEY
) {
1333 ref1
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1334 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1336 ref2
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1337 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1339 } else if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1340 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1341 struct btrfs_extent_data_ref
);
1342 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1343 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1344 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1345 struct btrfs_shared_data_ref
);
1346 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1347 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1348 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1349 struct btrfs_extent_ref_v0
*ref0
;
1350 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1351 struct btrfs_extent_ref_v0
);
1352 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1360 static noinline
int lookup_tree_block_ref(struct btrfs_trans_handle
*trans
,
1361 struct btrfs_root
*root
,
1362 struct btrfs_path
*path
,
1363 u64 bytenr
, u64 parent
,
1366 struct btrfs_key key
;
1369 key
.objectid
= bytenr
;
1371 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1372 key
.offset
= parent
;
1374 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1375 key
.offset
= root_objectid
;
1378 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1381 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1382 if (ret
== -ENOENT
&& parent
) {
1383 btrfs_release_path(path
);
1384 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1385 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1393 static noinline
int insert_tree_block_ref(struct btrfs_trans_handle
*trans
,
1394 struct btrfs_root
*root
,
1395 struct btrfs_path
*path
,
1396 u64 bytenr
, u64 parent
,
1399 struct btrfs_key key
;
1402 key
.objectid
= bytenr
;
1404 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1405 key
.offset
= parent
;
1407 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1408 key
.offset
= root_objectid
;
1411 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
1412 btrfs_release_path(path
);
1416 static inline int extent_ref_type(u64 parent
, u64 owner
)
1419 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1421 type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1423 type
= BTRFS_TREE_BLOCK_REF_KEY
;
1426 type
= BTRFS_SHARED_DATA_REF_KEY
;
1428 type
= BTRFS_EXTENT_DATA_REF_KEY
;
1433 static int find_next_key(struct btrfs_path
*path
, int level
,
1434 struct btrfs_key
*key
)
1437 for (; level
< BTRFS_MAX_LEVEL
; level
++) {
1438 if (!path
->nodes
[level
])
1440 if (path
->slots
[level
] + 1 >=
1441 btrfs_header_nritems(path
->nodes
[level
]))
1444 btrfs_item_key_to_cpu(path
->nodes
[level
], key
,
1445 path
->slots
[level
] + 1);
1447 btrfs_node_key_to_cpu(path
->nodes
[level
], key
,
1448 path
->slots
[level
] + 1);
1455 * look for inline back ref. if back ref is found, *ref_ret is set
1456 * to the address of inline back ref, and 0 is returned.
1458 * if back ref isn't found, *ref_ret is set to the address where it
1459 * should be inserted, and -ENOENT is returned.
1461 * if insert is true and there are too many inline back refs, the path
1462 * points to the extent item, and -EAGAIN is returned.
1464 * NOTE: inline back refs are ordered in the same way that back ref
1465 * items in the tree are ordered.
1467 static noinline_for_stack
1468 int lookup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1469 struct btrfs_root
*root
,
1470 struct btrfs_path
*path
,
1471 struct btrfs_extent_inline_ref
**ref_ret
,
1472 u64 bytenr
, u64 num_bytes
,
1473 u64 parent
, u64 root_objectid
,
1474 u64 owner
, u64 offset
, int insert
)
1476 struct btrfs_key key
;
1477 struct extent_buffer
*leaf
;
1478 struct btrfs_extent_item
*ei
;
1479 struct btrfs_extent_inline_ref
*iref
;
1489 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
1492 key
.objectid
= bytenr
;
1493 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1494 key
.offset
= num_bytes
;
1496 want
= extent_ref_type(parent
, owner
);
1498 extra_size
= btrfs_extent_inline_ref_size(want
);
1499 path
->keep_locks
= 1;
1504 * Owner is our parent level, so we can just add one to get the level
1505 * for the block we are interested in.
1507 if (skinny_metadata
&& owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1508 key
.type
= BTRFS_METADATA_ITEM_KEY
;
1513 ret
= btrfs_search_slot(trans
, root
, &key
, path
, extra_size
, 1);
1520 * We may be a newly converted file system which still has the old fat
1521 * extent entries for metadata, so try and see if we have one of those.
1523 if (ret
> 0 && skinny_metadata
) {
1524 skinny_metadata
= false;
1525 if (path
->slots
[0]) {
1527 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
1529 if (key
.objectid
== bytenr
&&
1530 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
1531 key
.offset
== num_bytes
)
1535 key
.objectid
= bytenr
;
1536 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1537 key
.offset
= num_bytes
;
1538 btrfs_release_path(path
);
1543 if (ret
&& !insert
) {
1546 } else if (WARN_ON(ret
)) {
1551 leaf
= path
->nodes
[0];
1552 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1553 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1554 if (item_size
< sizeof(*ei
)) {
1559 ret
= convert_extent_item_v0(trans
, root
, path
, owner
,
1565 leaf
= path
->nodes
[0];
1566 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1569 BUG_ON(item_size
< sizeof(*ei
));
1571 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1572 flags
= btrfs_extent_flags(leaf
, ei
);
1574 ptr
= (unsigned long)(ei
+ 1);
1575 end
= (unsigned long)ei
+ item_size
;
1577 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
&& !skinny_metadata
) {
1578 ptr
+= sizeof(struct btrfs_tree_block_info
);
1588 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1589 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1593 ptr
+= btrfs_extent_inline_ref_size(type
);
1597 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1598 struct btrfs_extent_data_ref
*dref
;
1599 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1600 if (match_extent_data_ref(leaf
, dref
, root_objectid
,
1605 if (hash_extent_data_ref_item(leaf
, dref
) <
1606 hash_extent_data_ref(root_objectid
, owner
, offset
))
1610 ref_offset
= btrfs_extent_inline_ref_offset(leaf
, iref
);
1612 if (parent
== ref_offset
) {
1616 if (ref_offset
< parent
)
1619 if (root_objectid
== ref_offset
) {
1623 if (ref_offset
< root_objectid
)
1627 ptr
+= btrfs_extent_inline_ref_size(type
);
1629 if (err
== -ENOENT
&& insert
) {
1630 if (item_size
+ extra_size
>=
1631 BTRFS_MAX_EXTENT_ITEM_SIZE(root
)) {
1636 * To add new inline back ref, we have to make sure
1637 * there is no corresponding back ref item.
1638 * For simplicity, we just do not add new inline back
1639 * ref if there is any kind of item for this block
1641 if (find_next_key(path
, 0, &key
) == 0 &&
1642 key
.objectid
== bytenr
&&
1643 key
.type
< BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1648 *ref_ret
= (struct btrfs_extent_inline_ref
*)ptr
;
1651 path
->keep_locks
= 0;
1652 btrfs_unlock_up_safe(path
, 1);
1658 * helper to add new inline back ref
1660 static noinline_for_stack
1661 void setup_inline_extent_backref(struct btrfs_root
*root
,
1662 struct btrfs_path
*path
,
1663 struct btrfs_extent_inline_ref
*iref
,
1664 u64 parent
, u64 root_objectid
,
1665 u64 owner
, u64 offset
, int refs_to_add
,
1666 struct btrfs_delayed_extent_op
*extent_op
)
1668 struct extent_buffer
*leaf
;
1669 struct btrfs_extent_item
*ei
;
1672 unsigned long item_offset
;
1677 leaf
= path
->nodes
[0];
1678 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1679 item_offset
= (unsigned long)iref
- (unsigned long)ei
;
1681 type
= extent_ref_type(parent
, owner
);
1682 size
= btrfs_extent_inline_ref_size(type
);
1684 btrfs_extend_item(root
, path
, size
);
1686 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1687 refs
= btrfs_extent_refs(leaf
, ei
);
1688 refs
+= refs_to_add
;
1689 btrfs_set_extent_refs(leaf
, ei
, refs
);
1691 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1693 ptr
= (unsigned long)ei
+ item_offset
;
1694 end
= (unsigned long)ei
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
1695 if (ptr
< end
- size
)
1696 memmove_extent_buffer(leaf
, ptr
+ size
, ptr
,
1699 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1700 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
1701 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1702 struct btrfs_extent_data_ref
*dref
;
1703 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1704 btrfs_set_extent_data_ref_root(leaf
, dref
, root_objectid
);
1705 btrfs_set_extent_data_ref_objectid(leaf
, dref
, owner
);
1706 btrfs_set_extent_data_ref_offset(leaf
, dref
, offset
);
1707 btrfs_set_extent_data_ref_count(leaf
, dref
, refs_to_add
);
1708 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1709 struct btrfs_shared_data_ref
*sref
;
1710 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1711 btrfs_set_shared_data_ref_count(leaf
, sref
, refs_to_add
);
1712 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1713 } else if (type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
1714 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1716 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
1718 btrfs_mark_buffer_dirty(leaf
);
1721 static int lookup_extent_backref(struct btrfs_trans_handle
*trans
,
1722 struct btrfs_root
*root
,
1723 struct btrfs_path
*path
,
1724 struct btrfs_extent_inline_ref
**ref_ret
,
1725 u64 bytenr
, u64 num_bytes
, u64 parent
,
1726 u64 root_objectid
, u64 owner
, u64 offset
)
1730 ret
= lookup_inline_extent_backref(trans
, root
, path
, ref_ret
,
1731 bytenr
, num_bytes
, parent
,
1732 root_objectid
, owner
, offset
, 0);
1736 btrfs_release_path(path
);
1739 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1740 ret
= lookup_tree_block_ref(trans
, root
, path
, bytenr
, parent
,
1743 ret
= lookup_extent_data_ref(trans
, root
, path
, bytenr
, parent
,
1744 root_objectid
, owner
, offset
);
1750 * helper to update/remove inline back ref
1752 static noinline_for_stack
1753 void update_inline_extent_backref(struct btrfs_root
*root
,
1754 struct btrfs_path
*path
,
1755 struct btrfs_extent_inline_ref
*iref
,
1757 struct btrfs_delayed_extent_op
*extent_op
,
1760 struct extent_buffer
*leaf
;
1761 struct btrfs_extent_item
*ei
;
1762 struct btrfs_extent_data_ref
*dref
= NULL
;
1763 struct btrfs_shared_data_ref
*sref
= NULL
;
1771 leaf
= path
->nodes
[0];
1772 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1773 refs
= btrfs_extent_refs(leaf
, ei
);
1774 WARN_ON(refs_to_mod
< 0 && refs
+ refs_to_mod
<= 0);
1775 refs
+= refs_to_mod
;
1776 btrfs_set_extent_refs(leaf
, ei
, refs
);
1778 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1780 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1782 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1783 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1784 refs
= btrfs_extent_data_ref_count(leaf
, dref
);
1785 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1786 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1787 refs
= btrfs_shared_data_ref_count(leaf
, sref
);
1790 BUG_ON(refs_to_mod
!= -1);
1793 BUG_ON(refs_to_mod
< 0 && refs
< -refs_to_mod
);
1794 refs
+= refs_to_mod
;
1797 if (type
== BTRFS_EXTENT_DATA_REF_KEY
)
1798 btrfs_set_extent_data_ref_count(leaf
, dref
, refs
);
1800 btrfs_set_shared_data_ref_count(leaf
, sref
, refs
);
1803 size
= btrfs_extent_inline_ref_size(type
);
1804 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1805 ptr
= (unsigned long)iref
;
1806 end
= (unsigned long)ei
+ item_size
;
1807 if (ptr
+ size
< end
)
1808 memmove_extent_buffer(leaf
, ptr
, ptr
+ size
,
1811 btrfs_truncate_item(root
, path
, item_size
, 1);
1813 btrfs_mark_buffer_dirty(leaf
);
1816 static noinline_for_stack
1817 int insert_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1818 struct btrfs_root
*root
,
1819 struct btrfs_path
*path
,
1820 u64 bytenr
, u64 num_bytes
, u64 parent
,
1821 u64 root_objectid
, u64 owner
,
1822 u64 offset
, int refs_to_add
,
1823 struct btrfs_delayed_extent_op
*extent_op
)
1825 struct btrfs_extent_inline_ref
*iref
;
1828 ret
= lookup_inline_extent_backref(trans
, root
, path
, &iref
,
1829 bytenr
, num_bytes
, parent
,
1830 root_objectid
, owner
, offset
, 1);
1832 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
);
1833 update_inline_extent_backref(root
, path
, iref
,
1834 refs_to_add
, extent_op
, NULL
);
1835 } else if (ret
== -ENOENT
) {
1836 setup_inline_extent_backref(root
, path
, iref
, parent
,
1837 root_objectid
, owner
, offset
,
1838 refs_to_add
, extent_op
);
1844 static int insert_extent_backref(struct btrfs_trans_handle
*trans
,
1845 struct btrfs_root
*root
,
1846 struct btrfs_path
*path
,
1847 u64 bytenr
, u64 parent
, u64 root_objectid
,
1848 u64 owner
, u64 offset
, int refs_to_add
)
1851 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1852 BUG_ON(refs_to_add
!= 1);
1853 ret
= insert_tree_block_ref(trans
, root
, path
, bytenr
,
1854 parent
, root_objectid
);
1856 ret
= insert_extent_data_ref(trans
, root
, path
, bytenr
,
1857 parent
, root_objectid
,
1858 owner
, offset
, refs_to_add
);
1863 static int remove_extent_backref(struct btrfs_trans_handle
*trans
,
1864 struct btrfs_root
*root
,
1865 struct btrfs_path
*path
,
1866 struct btrfs_extent_inline_ref
*iref
,
1867 int refs_to_drop
, int is_data
, int *last_ref
)
1871 BUG_ON(!is_data
&& refs_to_drop
!= 1);
1873 update_inline_extent_backref(root
, path
, iref
,
1874 -refs_to_drop
, NULL
, last_ref
);
1875 } else if (is_data
) {
1876 ret
= remove_extent_data_ref(trans
, root
, path
, refs_to_drop
,
1880 ret
= btrfs_del_item(trans
, root
, path
);
1885 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
1886 static int btrfs_issue_discard(struct block_device
*bdev
, u64 start
, u64 len
,
1887 u64
*discarded_bytes
)
1890 u64 bytes_left
, end
;
1891 u64 aligned_start
= ALIGN(start
, 1 << 9);
1893 if (WARN_ON(start
!= aligned_start
)) {
1894 len
-= aligned_start
- start
;
1895 len
= round_down(len
, 1 << 9);
1896 start
= aligned_start
;
1899 *discarded_bytes
= 0;
1907 /* Skip any superblocks on this device. */
1908 for (j
= 0; j
< BTRFS_SUPER_MIRROR_MAX
; j
++) {
1909 u64 sb_start
= btrfs_sb_offset(j
);
1910 u64 sb_end
= sb_start
+ BTRFS_SUPER_INFO_SIZE
;
1911 u64 size
= sb_start
- start
;
1913 if (!in_range(sb_start
, start
, bytes_left
) &&
1914 !in_range(sb_end
, start
, bytes_left
) &&
1915 !in_range(start
, sb_start
, BTRFS_SUPER_INFO_SIZE
))
1919 * Superblock spans beginning of range. Adjust start and
1922 if (sb_start
<= start
) {
1923 start
+= sb_end
- start
;
1928 bytes_left
= end
- start
;
1933 ret
= blkdev_issue_discard(bdev
, start
>> 9, size
>> 9,
1936 *discarded_bytes
+= size
;
1937 else if (ret
!= -EOPNOTSUPP
)
1946 bytes_left
= end
- start
;
1950 ret
= blkdev_issue_discard(bdev
, start
>> 9, bytes_left
>> 9,
1953 *discarded_bytes
+= bytes_left
;
1958 int btrfs_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
1959 u64 num_bytes
, u64
*actual_bytes
)
1962 u64 discarded_bytes
= 0;
1963 struct btrfs_bio
*bbio
= NULL
;
1966 /* Tell the block device(s) that the sectors can be discarded */
1967 ret
= btrfs_map_block(root
->fs_info
, REQ_DISCARD
,
1968 bytenr
, &num_bytes
, &bbio
, 0);
1969 /* Error condition is -ENOMEM */
1971 struct btrfs_bio_stripe
*stripe
= bbio
->stripes
;
1975 for (i
= 0; i
< bbio
->num_stripes
; i
++, stripe
++) {
1977 if (!stripe
->dev
->can_discard
)
1980 ret
= btrfs_issue_discard(stripe
->dev
->bdev
,
1985 discarded_bytes
+= bytes
;
1986 else if (ret
!= -EOPNOTSUPP
)
1987 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1990 * Just in case we get back EOPNOTSUPP for some reason,
1991 * just ignore the return value so we don't screw up
1992 * people calling discard_extent.
1996 btrfs_put_bbio(bbio
);
2000 *actual_bytes
= discarded_bytes
;
2003 if (ret
== -EOPNOTSUPP
)
2008 /* Can return -ENOMEM */
2009 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
2010 struct btrfs_root
*root
,
2011 u64 bytenr
, u64 num_bytes
, u64 parent
,
2012 u64 root_objectid
, u64 owner
, u64 offset
,
2016 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2018 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
&&
2019 root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
2021 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
2022 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
2024 parent
, root_objectid
, (int)owner
,
2025 BTRFS_ADD_DELAYED_REF
, NULL
, no_quota
);
2027 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
2029 parent
, root_objectid
, owner
, offset
,
2030 BTRFS_ADD_DELAYED_REF
, NULL
, no_quota
);
2035 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
2036 struct btrfs_root
*root
,
2037 struct btrfs_delayed_ref_node
*node
,
2038 u64 parent
, u64 root_objectid
,
2039 u64 owner
, u64 offset
, int refs_to_add
,
2040 struct btrfs_delayed_extent_op
*extent_op
)
2042 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2043 struct btrfs_path
*path
;
2044 struct extent_buffer
*leaf
;
2045 struct btrfs_extent_item
*item
;
2046 struct btrfs_key key
;
2047 u64 bytenr
= node
->bytenr
;
2048 u64 num_bytes
= node
->num_bytes
;
2051 int no_quota
= node
->no_quota
;
2053 path
= btrfs_alloc_path();
2057 if (!is_fstree(root_objectid
) || !root
->fs_info
->quota_enabled
)
2061 path
->leave_spinning
= 1;
2062 /* this will setup the path even if it fails to insert the back ref */
2063 ret
= insert_inline_extent_backref(trans
, fs_info
->extent_root
, path
,
2064 bytenr
, num_bytes
, parent
,
2065 root_objectid
, owner
, offset
,
2066 refs_to_add
, extent_op
);
2067 if ((ret
< 0 && ret
!= -EAGAIN
) || !ret
)
2071 * Ok we had -EAGAIN which means we didn't have space to insert and
2072 * inline extent ref, so just update the reference count and add a
2075 leaf
= path
->nodes
[0];
2076 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2077 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2078 refs
= btrfs_extent_refs(leaf
, item
);
2079 btrfs_set_extent_refs(leaf
, item
, refs
+ refs_to_add
);
2081 __run_delayed_extent_op(extent_op
, leaf
, item
);
2083 btrfs_mark_buffer_dirty(leaf
);
2084 btrfs_release_path(path
);
2087 path
->leave_spinning
= 1;
2088 /* now insert the actual backref */
2089 ret
= insert_extent_backref(trans
, root
->fs_info
->extent_root
,
2090 path
, bytenr
, parent
, root_objectid
,
2091 owner
, offset
, refs_to_add
);
2093 btrfs_abort_transaction(trans
, root
, ret
);
2095 btrfs_free_path(path
);
2099 static int run_delayed_data_ref(struct btrfs_trans_handle
*trans
,
2100 struct btrfs_root
*root
,
2101 struct btrfs_delayed_ref_node
*node
,
2102 struct btrfs_delayed_extent_op
*extent_op
,
2103 int insert_reserved
)
2106 struct btrfs_delayed_data_ref
*ref
;
2107 struct btrfs_key ins
;
2112 ins
.objectid
= node
->bytenr
;
2113 ins
.offset
= node
->num_bytes
;
2114 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2116 ref
= btrfs_delayed_node_to_data_ref(node
);
2117 trace_run_delayed_data_ref(node
, ref
, node
->action
);
2119 if (node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2120 parent
= ref
->parent
;
2121 ref_root
= ref
->root
;
2123 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2125 flags
|= extent_op
->flags_to_set
;
2126 ret
= alloc_reserved_file_extent(trans
, root
,
2127 parent
, ref_root
, flags
,
2128 ref
->objectid
, ref
->offset
,
2129 &ins
, node
->ref_mod
);
2130 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2131 ret
= __btrfs_inc_extent_ref(trans
, root
, node
, parent
,
2132 ref_root
, ref
->objectid
,
2133 ref
->offset
, node
->ref_mod
,
2135 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2136 ret
= __btrfs_free_extent(trans
, root
, node
, parent
,
2137 ref_root
, ref
->objectid
,
2138 ref
->offset
, node
->ref_mod
,
2146 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
2147 struct extent_buffer
*leaf
,
2148 struct btrfs_extent_item
*ei
)
2150 u64 flags
= btrfs_extent_flags(leaf
, ei
);
2151 if (extent_op
->update_flags
) {
2152 flags
|= extent_op
->flags_to_set
;
2153 btrfs_set_extent_flags(leaf
, ei
, flags
);
2156 if (extent_op
->update_key
) {
2157 struct btrfs_tree_block_info
*bi
;
2158 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
));
2159 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
2160 btrfs_set_tree_block_key(leaf
, bi
, &extent_op
->key
);
2164 static int run_delayed_extent_op(struct btrfs_trans_handle
*trans
,
2165 struct btrfs_root
*root
,
2166 struct btrfs_delayed_ref_node
*node
,
2167 struct btrfs_delayed_extent_op
*extent_op
)
2169 struct btrfs_key key
;
2170 struct btrfs_path
*path
;
2171 struct btrfs_extent_item
*ei
;
2172 struct extent_buffer
*leaf
;
2176 int metadata
= !extent_op
->is_data
;
2181 if (metadata
&& !btrfs_fs_incompat(root
->fs_info
, SKINNY_METADATA
))
2184 path
= btrfs_alloc_path();
2188 key
.objectid
= node
->bytenr
;
2191 key
.type
= BTRFS_METADATA_ITEM_KEY
;
2192 key
.offset
= extent_op
->level
;
2194 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2195 key
.offset
= node
->num_bytes
;
2200 path
->leave_spinning
= 1;
2201 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
,
2209 if (path
->slots
[0] > 0) {
2211 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
2213 if (key
.objectid
== node
->bytenr
&&
2214 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
2215 key
.offset
== node
->num_bytes
)
2219 btrfs_release_path(path
);
2222 key
.objectid
= node
->bytenr
;
2223 key
.offset
= node
->num_bytes
;
2224 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2233 leaf
= path
->nodes
[0];
2234 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2235 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2236 if (item_size
< sizeof(*ei
)) {
2237 ret
= convert_extent_item_v0(trans
, root
->fs_info
->extent_root
,
2243 leaf
= path
->nodes
[0];
2244 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2247 BUG_ON(item_size
< sizeof(*ei
));
2248 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2249 __run_delayed_extent_op(extent_op
, leaf
, ei
);
2251 btrfs_mark_buffer_dirty(leaf
);
2253 btrfs_free_path(path
);
2257 static int run_delayed_tree_ref(struct btrfs_trans_handle
*trans
,
2258 struct btrfs_root
*root
,
2259 struct btrfs_delayed_ref_node
*node
,
2260 struct btrfs_delayed_extent_op
*extent_op
,
2261 int insert_reserved
)
2264 struct btrfs_delayed_tree_ref
*ref
;
2265 struct btrfs_key ins
;
2268 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
2271 ref
= btrfs_delayed_node_to_tree_ref(node
);
2272 trace_run_delayed_tree_ref(node
, ref
, node
->action
);
2274 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2275 parent
= ref
->parent
;
2276 ref_root
= ref
->root
;
2278 ins
.objectid
= node
->bytenr
;
2279 if (skinny_metadata
) {
2280 ins
.offset
= ref
->level
;
2281 ins
.type
= BTRFS_METADATA_ITEM_KEY
;
2283 ins
.offset
= node
->num_bytes
;
2284 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2287 BUG_ON(node
->ref_mod
!= 1);
2288 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2289 BUG_ON(!extent_op
|| !extent_op
->update_flags
);
2290 ret
= alloc_reserved_tree_block(trans
, root
,
2292 extent_op
->flags_to_set
,
2296 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2297 ret
= __btrfs_inc_extent_ref(trans
, root
, node
,
2301 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2302 ret
= __btrfs_free_extent(trans
, root
, node
,
2304 ref
->level
, 0, 1, extent_op
);
2311 /* helper function to actually process a single delayed ref entry */
2312 static int run_one_delayed_ref(struct btrfs_trans_handle
*trans
,
2313 struct btrfs_root
*root
,
2314 struct btrfs_delayed_ref_node
*node
,
2315 struct btrfs_delayed_extent_op
*extent_op
,
2316 int insert_reserved
)
2320 if (trans
->aborted
) {
2321 if (insert_reserved
)
2322 btrfs_pin_extent(root
, node
->bytenr
,
2323 node
->num_bytes
, 1);
2327 if (btrfs_delayed_ref_is_head(node
)) {
2328 struct btrfs_delayed_ref_head
*head
;
2330 * we've hit the end of the chain and we were supposed
2331 * to insert this extent into the tree. But, it got
2332 * deleted before we ever needed to insert it, so all
2333 * we have to do is clean up the accounting
2336 head
= btrfs_delayed_node_to_head(node
);
2337 trace_run_delayed_ref_head(node
, head
, node
->action
);
2339 if (insert_reserved
) {
2340 btrfs_pin_extent(root
, node
->bytenr
,
2341 node
->num_bytes
, 1);
2342 if (head
->is_data
) {
2343 ret
= btrfs_del_csums(trans
, root
,
2349 /* Also free its reserved qgroup space */
2350 btrfs_qgroup_free_delayed_ref(root
->fs_info
,
2351 head
->qgroup_ref_root
,
2352 head
->qgroup_reserved
);
2356 if (node
->type
== BTRFS_TREE_BLOCK_REF_KEY
||
2357 node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2358 ret
= run_delayed_tree_ref(trans
, root
, node
, extent_op
,
2360 else if (node
->type
== BTRFS_EXTENT_DATA_REF_KEY
||
2361 node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2362 ret
= run_delayed_data_ref(trans
, root
, node
, extent_op
,
2369 static inline struct btrfs_delayed_ref_node
*
2370 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
2372 struct btrfs_delayed_ref_node
*ref
;
2374 if (list_empty(&head
->ref_list
))
2378 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2379 * This is to prevent a ref count from going down to zero, which deletes
2380 * the extent item from the extent tree, when there still are references
2381 * to add, which would fail because they would not find the extent item.
2383 list_for_each_entry(ref
, &head
->ref_list
, list
) {
2384 if (ref
->action
== BTRFS_ADD_DELAYED_REF
)
2388 return list_entry(head
->ref_list
.next
, struct btrfs_delayed_ref_node
,
2393 * Returns 0 on success or if called with an already aborted transaction.
2394 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2396 static noinline
int __btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2397 struct btrfs_root
*root
,
2400 struct btrfs_delayed_ref_root
*delayed_refs
;
2401 struct btrfs_delayed_ref_node
*ref
;
2402 struct btrfs_delayed_ref_head
*locked_ref
= NULL
;
2403 struct btrfs_delayed_extent_op
*extent_op
;
2404 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2405 ktime_t start
= ktime_get();
2407 unsigned long count
= 0;
2408 unsigned long actual_count
= 0;
2409 int must_insert_reserved
= 0;
2411 delayed_refs
= &trans
->transaction
->delayed_refs
;
2417 spin_lock(&delayed_refs
->lock
);
2418 locked_ref
= btrfs_select_ref_head(trans
);
2420 spin_unlock(&delayed_refs
->lock
);
2424 /* grab the lock that says we are going to process
2425 * all the refs for this head */
2426 ret
= btrfs_delayed_ref_lock(trans
, locked_ref
);
2427 spin_unlock(&delayed_refs
->lock
);
2429 * we may have dropped the spin lock to get the head
2430 * mutex lock, and that might have given someone else
2431 * time to free the head. If that's true, it has been
2432 * removed from our list and we can move on.
2434 if (ret
== -EAGAIN
) {
2441 spin_lock(&locked_ref
->lock
);
2444 * locked_ref is the head node, so we have to go one
2445 * node back for any delayed ref updates
2447 ref
= select_delayed_ref(locked_ref
);
2449 if (ref
&& ref
->seq
&&
2450 btrfs_check_delayed_seq(fs_info
, delayed_refs
, ref
->seq
)) {
2451 spin_unlock(&locked_ref
->lock
);
2452 btrfs_delayed_ref_unlock(locked_ref
);
2453 spin_lock(&delayed_refs
->lock
);
2454 locked_ref
->processing
= 0;
2455 delayed_refs
->num_heads_ready
++;
2456 spin_unlock(&delayed_refs
->lock
);
2464 * record the must insert reserved flag before we
2465 * drop the spin lock.
2467 must_insert_reserved
= locked_ref
->must_insert_reserved
;
2468 locked_ref
->must_insert_reserved
= 0;
2470 extent_op
= locked_ref
->extent_op
;
2471 locked_ref
->extent_op
= NULL
;
2476 /* All delayed refs have been processed, Go ahead
2477 * and send the head node to run_one_delayed_ref,
2478 * so that any accounting fixes can happen
2480 ref
= &locked_ref
->node
;
2482 if (extent_op
&& must_insert_reserved
) {
2483 btrfs_free_delayed_extent_op(extent_op
);
2488 spin_unlock(&locked_ref
->lock
);
2489 ret
= run_delayed_extent_op(trans
, root
,
2491 btrfs_free_delayed_extent_op(extent_op
);
2495 * Need to reset must_insert_reserved if
2496 * there was an error so the abort stuff
2497 * can cleanup the reserved space
2500 if (must_insert_reserved
)
2501 locked_ref
->must_insert_reserved
= 1;
2502 locked_ref
->processing
= 0;
2503 btrfs_debug(fs_info
, "run_delayed_extent_op returned %d", ret
);
2504 btrfs_delayed_ref_unlock(locked_ref
);
2511 * Need to drop our head ref lock and re-aqcuire the
2512 * delayed ref lock and then re-check to make sure
2515 spin_unlock(&locked_ref
->lock
);
2516 spin_lock(&delayed_refs
->lock
);
2517 spin_lock(&locked_ref
->lock
);
2518 if (!list_empty(&locked_ref
->ref_list
) ||
2519 locked_ref
->extent_op
) {
2520 spin_unlock(&locked_ref
->lock
);
2521 spin_unlock(&delayed_refs
->lock
);
2525 delayed_refs
->num_heads
--;
2526 rb_erase(&locked_ref
->href_node
,
2527 &delayed_refs
->href_root
);
2528 spin_unlock(&delayed_refs
->lock
);
2532 list_del(&ref
->list
);
2534 atomic_dec(&delayed_refs
->num_entries
);
2536 if (!btrfs_delayed_ref_is_head(ref
)) {
2538 * when we play the delayed ref, also correct the
2541 switch (ref
->action
) {
2542 case BTRFS_ADD_DELAYED_REF
:
2543 case BTRFS_ADD_DELAYED_EXTENT
:
2544 locked_ref
->node
.ref_mod
-= ref
->ref_mod
;
2546 case BTRFS_DROP_DELAYED_REF
:
2547 locked_ref
->node
.ref_mod
+= ref
->ref_mod
;
2553 spin_unlock(&locked_ref
->lock
);
2555 ret
= run_one_delayed_ref(trans
, root
, ref
, extent_op
,
2556 must_insert_reserved
);
2558 btrfs_free_delayed_extent_op(extent_op
);
2560 locked_ref
->processing
= 0;
2561 btrfs_delayed_ref_unlock(locked_ref
);
2562 btrfs_put_delayed_ref(ref
);
2563 btrfs_debug(fs_info
, "run_one_delayed_ref returned %d", ret
);
2568 * If this node is a head, that means all the refs in this head
2569 * have been dealt with, and we will pick the next head to deal
2570 * with, so we must unlock the head and drop it from the cluster
2571 * list before we release it.
2573 if (btrfs_delayed_ref_is_head(ref
)) {
2574 if (locked_ref
->is_data
&&
2575 locked_ref
->total_ref_mod
< 0) {
2576 spin_lock(&delayed_refs
->lock
);
2577 delayed_refs
->pending_csums
-= ref
->num_bytes
;
2578 spin_unlock(&delayed_refs
->lock
);
2580 btrfs_delayed_ref_unlock(locked_ref
);
2583 btrfs_put_delayed_ref(ref
);
2589 * We don't want to include ref heads since we can have empty ref heads
2590 * and those will drastically skew our runtime down since we just do
2591 * accounting, no actual extent tree updates.
2593 if (actual_count
> 0) {
2594 u64 runtime
= ktime_to_ns(ktime_sub(ktime_get(), start
));
2598 * We weigh the current average higher than our current runtime
2599 * to avoid large swings in the average.
2601 spin_lock(&delayed_refs
->lock
);
2602 avg
= fs_info
->avg_delayed_ref_runtime
* 3 + runtime
;
2603 fs_info
->avg_delayed_ref_runtime
= avg
>> 2; /* div by 4 */
2604 spin_unlock(&delayed_refs
->lock
);
2609 #ifdef SCRAMBLE_DELAYED_REFS
2611 * Normally delayed refs get processed in ascending bytenr order. This
2612 * correlates in most cases to the order added. To expose dependencies on this
2613 * order, we start to process the tree in the middle instead of the beginning
2615 static u64
find_middle(struct rb_root
*root
)
2617 struct rb_node
*n
= root
->rb_node
;
2618 struct btrfs_delayed_ref_node
*entry
;
2621 u64 first
= 0, last
= 0;
2625 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2626 first
= entry
->bytenr
;
2630 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2631 last
= entry
->bytenr
;
2636 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2637 WARN_ON(!entry
->in_tree
);
2639 middle
= entry
->bytenr
;
2652 static inline u64
heads_to_leaves(struct btrfs_root
*root
, u64 heads
)
2656 num_bytes
= heads
* (sizeof(struct btrfs_extent_item
) +
2657 sizeof(struct btrfs_extent_inline_ref
));
2658 if (!btrfs_fs_incompat(root
->fs_info
, SKINNY_METADATA
))
2659 num_bytes
+= heads
* sizeof(struct btrfs_tree_block_info
);
2662 * We don't ever fill up leaves all the way so multiply by 2 just to be
2663 * closer to what we're really going to want to ouse.
2665 return div_u64(num_bytes
, BTRFS_LEAF_DATA_SIZE(root
));
2669 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2670 * would require to store the csums for that many bytes.
2672 u64
btrfs_csum_bytes_to_leaves(struct btrfs_root
*root
, u64 csum_bytes
)
2675 u64 num_csums_per_leaf
;
2678 csum_size
= BTRFS_LEAF_DATA_SIZE(root
) - sizeof(struct btrfs_item
);
2679 num_csums_per_leaf
= div64_u64(csum_size
,
2680 (u64
)btrfs_super_csum_size(root
->fs_info
->super_copy
));
2681 num_csums
= div64_u64(csum_bytes
, root
->sectorsize
);
2682 num_csums
+= num_csums_per_leaf
- 1;
2683 num_csums
= div64_u64(num_csums
, num_csums_per_leaf
);
2687 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle
*trans
,
2688 struct btrfs_root
*root
)
2690 struct btrfs_block_rsv
*global_rsv
;
2691 u64 num_heads
= trans
->transaction
->delayed_refs
.num_heads_ready
;
2692 u64 csum_bytes
= trans
->transaction
->delayed_refs
.pending_csums
;
2693 u64 num_dirty_bgs
= trans
->transaction
->num_dirty_bgs
;
2694 u64 num_bytes
, num_dirty_bgs_bytes
;
2697 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
2698 num_heads
= heads_to_leaves(root
, num_heads
);
2700 num_bytes
+= (num_heads
- 1) * root
->nodesize
;
2702 num_bytes
+= btrfs_csum_bytes_to_leaves(root
, csum_bytes
) * root
->nodesize
;
2703 num_dirty_bgs_bytes
= btrfs_calc_trans_metadata_size(root
,
2705 global_rsv
= &root
->fs_info
->global_block_rsv
;
2708 * If we can't allocate any more chunks lets make sure we have _lots_ of
2709 * wiggle room since running delayed refs can create more delayed refs.
2711 if (global_rsv
->space_info
->full
) {
2712 num_dirty_bgs_bytes
<<= 1;
2716 spin_lock(&global_rsv
->lock
);
2717 if (global_rsv
->reserved
<= num_bytes
+ num_dirty_bgs_bytes
)
2719 spin_unlock(&global_rsv
->lock
);
2723 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle
*trans
,
2724 struct btrfs_root
*root
)
2726 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2728 atomic_read(&trans
->transaction
->delayed_refs
.num_entries
);
2733 avg_runtime
= fs_info
->avg_delayed_ref_runtime
;
2734 val
= num_entries
* avg_runtime
;
2735 if (num_entries
* avg_runtime
>= NSEC_PER_SEC
)
2737 if (val
>= NSEC_PER_SEC
/ 2)
2740 return btrfs_check_space_for_delayed_refs(trans
, root
);
2743 struct async_delayed_refs
{
2744 struct btrfs_root
*root
;
2748 struct completion wait
;
2749 struct btrfs_work work
;
2752 static void delayed_ref_async_start(struct btrfs_work
*work
)
2754 struct async_delayed_refs
*async
;
2755 struct btrfs_trans_handle
*trans
;
2758 async
= container_of(work
, struct async_delayed_refs
, work
);
2760 trans
= btrfs_join_transaction(async
->root
);
2761 if (IS_ERR(trans
)) {
2762 async
->error
= PTR_ERR(trans
);
2767 * trans->sync means that when we call end_transaciton, we won't
2768 * wait on delayed refs
2771 ret
= btrfs_run_delayed_refs(trans
, async
->root
, async
->count
);
2775 ret
= btrfs_end_transaction(trans
, async
->root
);
2776 if (ret
&& !async
->error
)
2780 complete(&async
->wait
);
2785 int btrfs_async_run_delayed_refs(struct btrfs_root
*root
,
2786 unsigned long count
, int wait
)
2788 struct async_delayed_refs
*async
;
2791 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
2795 async
->root
= root
->fs_info
->tree_root
;
2796 async
->count
= count
;
2802 init_completion(&async
->wait
);
2804 btrfs_init_work(&async
->work
, btrfs_extent_refs_helper
,
2805 delayed_ref_async_start
, NULL
, NULL
);
2807 btrfs_queue_work(root
->fs_info
->extent_workers
, &async
->work
);
2810 wait_for_completion(&async
->wait
);
2819 * this starts processing the delayed reference count updates and
2820 * extent insertions we have queued up so far. count can be
2821 * 0, which means to process everything in the tree at the start
2822 * of the run (but not newly added entries), or it can be some target
2823 * number you'd like to process.
2825 * Returns 0 on success or if called with an aborted transaction
2826 * Returns <0 on error and aborts the transaction
2828 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2829 struct btrfs_root
*root
, unsigned long count
)
2831 struct rb_node
*node
;
2832 struct btrfs_delayed_ref_root
*delayed_refs
;
2833 struct btrfs_delayed_ref_head
*head
;
2835 int run_all
= count
== (unsigned long)-1;
2836 bool can_flush_pending_bgs
= trans
->can_flush_pending_bgs
;
2838 /* We'll clean this up in btrfs_cleanup_transaction */
2842 if (root
== root
->fs_info
->extent_root
)
2843 root
= root
->fs_info
->tree_root
;
2845 delayed_refs
= &trans
->transaction
->delayed_refs
;
2847 count
= atomic_read(&delayed_refs
->num_entries
) * 2;
2850 #ifdef SCRAMBLE_DELAYED_REFS
2851 delayed_refs
->run_delayed_start
= find_middle(&delayed_refs
->root
);
2853 trans
->can_flush_pending_bgs
= false;
2854 ret
= __btrfs_run_delayed_refs(trans
, root
, count
);
2856 btrfs_abort_transaction(trans
, root
, ret
);
2861 if (!list_empty(&trans
->new_bgs
))
2862 btrfs_create_pending_block_groups(trans
, root
);
2864 spin_lock(&delayed_refs
->lock
);
2865 node
= rb_first(&delayed_refs
->href_root
);
2867 spin_unlock(&delayed_refs
->lock
);
2870 count
= (unsigned long)-1;
2873 head
= rb_entry(node
, struct btrfs_delayed_ref_head
,
2875 if (btrfs_delayed_ref_is_head(&head
->node
)) {
2876 struct btrfs_delayed_ref_node
*ref
;
2879 atomic_inc(&ref
->refs
);
2881 spin_unlock(&delayed_refs
->lock
);
2883 * Mutex was contended, block until it's
2884 * released and try again
2886 mutex_lock(&head
->mutex
);
2887 mutex_unlock(&head
->mutex
);
2889 btrfs_put_delayed_ref(ref
);
2895 node
= rb_next(node
);
2897 spin_unlock(&delayed_refs
->lock
);
2902 assert_qgroups_uptodate(trans
);
2903 trans
->can_flush_pending_bgs
= can_flush_pending_bgs
;
2907 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
2908 struct btrfs_root
*root
,
2909 u64 bytenr
, u64 num_bytes
, u64 flags
,
2910 int level
, int is_data
)
2912 struct btrfs_delayed_extent_op
*extent_op
;
2915 extent_op
= btrfs_alloc_delayed_extent_op();
2919 extent_op
->flags_to_set
= flags
;
2920 extent_op
->update_flags
= 1;
2921 extent_op
->update_key
= 0;
2922 extent_op
->is_data
= is_data
? 1 : 0;
2923 extent_op
->level
= level
;
2925 ret
= btrfs_add_delayed_extent_op(root
->fs_info
, trans
, bytenr
,
2926 num_bytes
, extent_op
);
2928 btrfs_free_delayed_extent_op(extent_op
);
2932 static noinline
int check_delayed_ref(struct btrfs_trans_handle
*trans
,
2933 struct btrfs_root
*root
,
2934 struct btrfs_path
*path
,
2935 u64 objectid
, u64 offset
, u64 bytenr
)
2937 struct btrfs_delayed_ref_head
*head
;
2938 struct btrfs_delayed_ref_node
*ref
;
2939 struct btrfs_delayed_data_ref
*data_ref
;
2940 struct btrfs_delayed_ref_root
*delayed_refs
;
2943 delayed_refs
= &trans
->transaction
->delayed_refs
;
2944 spin_lock(&delayed_refs
->lock
);
2945 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
2947 spin_unlock(&delayed_refs
->lock
);
2951 if (!mutex_trylock(&head
->mutex
)) {
2952 atomic_inc(&head
->node
.refs
);
2953 spin_unlock(&delayed_refs
->lock
);
2955 btrfs_release_path(path
);
2958 * Mutex was contended, block until it's released and let
2961 mutex_lock(&head
->mutex
);
2962 mutex_unlock(&head
->mutex
);
2963 btrfs_put_delayed_ref(&head
->node
);
2966 spin_unlock(&delayed_refs
->lock
);
2968 spin_lock(&head
->lock
);
2969 list_for_each_entry(ref
, &head
->ref_list
, list
) {
2970 /* If it's a shared ref we know a cross reference exists */
2971 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
) {
2976 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
2979 * If our ref doesn't match the one we're currently looking at
2980 * then we have a cross reference.
2982 if (data_ref
->root
!= root
->root_key
.objectid
||
2983 data_ref
->objectid
!= objectid
||
2984 data_ref
->offset
!= offset
) {
2989 spin_unlock(&head
->lock
);
2990 mutex_unlock(&head
->mutex
);
2994 static noinline
int check_committed_ref(struct btrfs_trans_handle
*trans
,
2995 struct btrfs_root
*root
,
2996 struct btrfs_path
*path
,
2997 u64 objectid
, u64 offset
, u64 bytenr
)
2999 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
3000 struct extent_buffer
*leaf
;
3001 struct btrfs_extent_data_ref
*ref
;
3002 struct btrfs_extent_inline_ref
*iref
;
3003 struct btrfs_extent_item
*ei
;
3004 struct btrfs_key key
;
3008 key
.objectid
= bytenr
;
3009 key
.offset
= (u64
)-1;
3010 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
3012 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
3015 BUG_ON(ret
== 0); /* Corruption */
3018 if (path
->slots
[0] == 0)
3022 leaf
= path
->nodes
[0];
3023 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
3025 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
3029 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
3030 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3031 if (item_size
< sizeof(*ei
)) {
3032 WARN_ON(item_size
!= sizeof(struct btrfs_extent_item_v0
));
3036 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
3038 if (item_size
!= sizeof(*ei
) +
3039 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
3042 if (btrfs_extent_generation(leaf
, ei
) <=
3043 btrfs_root_last_snapshot(&root
->root_item
))
3046 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
3047 if (btrfs_extent_inline_ref_type(leaf
, iref
) !=
3048 BTRFS_EXTENT_DATA_REF_KEY
)
3051 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
3052 if (btrfs_extent_refs(leaf
, ei
) !=
3053 btrfs_extent_data_ref_count(leaf
, ref
) ||
3054 btrfs_extent_data_ref_root(leaf
, ref
) !=
3055 root
->root_key
.objectid
||
3056 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
3057 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
3065 int btrfs_cross_ref_exist(struct btrfs_trans_handle
*trans
,
3066 struct btrfs_root
*root
,
3067 u64 objectid
, u64 offset
, u64 bytenr
)
3069 struct btrfs_path
*path
;
3073 path
= btrfs_alloc_path();
3078 ret
= check_committed_ref(trans
, root
, path
, objectid
,
3080 if (ret
&& ret
!= -ENOENT
)
3083 ret2
= check_delayed_ref(trans
, root
, path
, objectid
,
3085 } while (ret2
== -EAGAIN
);
3087 if (ret2
&& ret2
!= -ENOENT
) {
3092 if (ret
!= -ENOENT
|| ret2
!= -ENOENT
)
3095 btrfs_free_path(path
);
3096 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
3101 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
3102 struct btrfs_root
*root
,
3103 struct extent_buffer
*buf
,
3104 int full_backref
, int inc
)
3111 struct btrfs_key key
;
3112 struct btrfs_file_extent_item
*fi
;
3116 int (*process_func
)(struct btrfs_trans_handle
*, struct btrfs_root
*,
3117 u64
, u64
, u64
, u64
, u64
, u64
, int);
3120 if (btrfs_test_is_dummy_root(root
))
3123 ref_root
= btrfs_header_owner(buf
);
3124 nritems
= btrfs_header_nritems(buf
);
3125 level
= btrfs_header_level(buf
);
3127 if (!test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) && level
== 0)
3131 process_func
= btrfs_inc_extent_ref
;
3133 process_func
= btrfs_free_extent
;
3136 parent
= buf
->start
;
3140 for (i
= 0; i
< nritems
; i
++) {
3142 btrfs_item_key_to_cpu(buf
, &key
, i
);
3143 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
3145 fi
= btrfs_item_ptr(buf
, i
,
3146 struct btrfs_file_extent_item
);
3147 if (btrfs_file_extent_type(buf
, fi
) ==
3148 BTRFS_FILE_EXTENT_INLINE
)
3150 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
3154 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
3155 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
3156 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
3157 parent
, ref_root
, key
.objectid
,
3162 bytenr
= btrfs_node_blockptr(buf
, i
);
3163 num_bytes
= root
->nodesize
;
3164 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
3165 parent
, ref_root
, level
- 1, 0,
3176 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3177 struct extent_buffer
*buf
, int full_backref
)
3179 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1);
3182 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3183 struct extent_buffer
*buf
, int full_backref
)
3185 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0);
3188 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
3189 struct btrfs_root
*root
,
3190 struct btrfs_path
*path
,
3191 struct btrfs_block_group_cache
*cache
)
3194 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
3196 struct extent_buffer
*leaf
;
3198 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
3205 leaf
= path
->nodes
[0];
3206 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
3207 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
3208 btrfs_mark_buffer_dirty(leaf
);
3210 btrfs_release_path(path
);
3215 static struct btrfs_block_group_cache
*
3216 next_block_group(struct btrfs_root
*root
,
3217 struct btrfs_block_group_cache
*cache
)
3219 struct rb_node
*node
;
3221 spin_lock(&root
->fs_info
->block_group_cache_lock
);
3223 /* If our block group was removed, we need a full search. */
3224 if (RB_EMPTY_NODE(&cache
->cache_node
)) {
3225 const u64 next_bytenr
= cache
->key
.objectid
+ cache
->key
.offset
;
3227 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
3228 btrfs_put_block_group(cache
);
3229 cache
= btrfs_lookup_first_block_group(root
->fs_info
,
3233 node
= rb_next(&cache
->cache_node
);
3234 btrfs_put_block_group(cache
);
3236 cache
= rb_entry(node
, struct btrfs_block_group_cache
,
3238 btrfs_get_block_group(cache
);
3241 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
3245 static int cache_save_setup(struct btrfs_block_group_cache
*block_group
,
3246 struct btrfs_trans_handle
*trans
,
3247 struct btrfs_path
*path
)
3249 struct btrfs_root
*root
= block_group
->fs_info
->tree_root
;
3250 struct inode
*inode
= NULL
;
3252 int dcs
= BTRFS_DC_ERROR
;
3258 * If this block group is smaller than 100 megs don't bother caching the
3261 if (block_group
->key
.offset
< (100 * 1024 * 1024)) {
3262 spin_lock(&block_group
->lock
);
3263 block_group
->disk_cache_state
= BTRFS_DC_WRITTEN
;
3264 spin_unlock(&block_group
->lock
);
3271 inode
= lookup_free_space_inode(root
, block_group
, path
);
3272 if (IS_ERR(inode
) && PTR_ERR(inode
) != -ENOENT
) {
3273 ret
= PTR_ERR(inode
);
3274 btrfs_release_path(path
);
3278 if (IS_ERR(inode
)) {
3282 if (block_group
->ro
)
3285 ret
= create_free_space_inode(root
, trans
, block_group
, path
);
3291 /* We've already setup this transaction, go ahead and exit */
3292 if (block_group
->cache_generation
== trans
->transid
&&
3293 i_size_read(inode
)) {
3294 dcs
= BTRFS_DC_SETUP
;
3299 * We want to set the generation to 0, that way if anything goes wrong
3300 * from here on out we know not to trust this cache when we load up next
3303 BTRFS_I(inode
)->generation
= 0;
3304 ret
= btrfs_update_inode(trans
, root
, inode
);
3307 * So theoretically we could recover from this, simply set the
3308 * super cache generation to 0 so we know to invalidate the
3309 * cache, but then we'd have to keep track of the block groups
3310 * that fail this way so we know we _have_ to reset this cache
3311 * before the next commit or risk reading stale cache. So to
3312 * limit our exposure to horrible edge cases lets just abort the
3313 * transaction, this only happens in really bad situations
3316 btrfs_abort_transaction(trans
, root
, ret
);
3321 if (i_size_read(inode
) > 0) {
3322 ret
= btrfs_check_trunc_cache_free_space(root
,
3323 &root
->fs_info
->global_block_rsv
);
3327 ret
= btrfs_truncate_free_space_cache(root
, trans
, NULL
, inode
);
3332 spin_lock(&block_group
->lock
);
3333 if (block_group
->cached
!= BTRFS_CACHE_FINISHED
||
3334 !btrfs_test_opt(root
, SPACE_CACHE
)) {
3336 * don't bother trying to write stuff out _if_
3337 * a) we're not cached,
3338 * b) we're with nospace_cache mount option.
3340 dcs
= BTRFS_DC_WRITTEN
;
3341 spin_unlock(&block_group
->lock
);
3344 spin_unlock(&block_group
->lock
);
3347 * Try to preallocate enough space based on how big the block group is.
3348 * Keep in mind this has to include any pinned space which could end up
3349 * taking up quite a bit since it's not folded into the other space
3352 num_pages
= div_u64(block_group
->key
.offset
, 256 * 1024 * 1024);
3357 num_pages
*= PAGE_CACHE_SIZE
;
3359 ret
= btrfs_check_data_free_space(inode
, num_pages
, num_pages
);
3363 ret
= btrfs_prealloc_file_range_trans(inode
, trans
, 0, 0, num_pages
,
3364 num_pages
, num_pages
,
3367 dcs
= BTRFS_DC_SETUP
;
3368 btrfs_free_reserved_data_space(inode
, num_pages
);
3373 btrfs_release_path(path
);
3375 spin_lock(&block_group
->lock
);
3376 if (!ret
&& dcs
== BTRFS_DC_SETUP
)
3377 block_group
->cache_generation
= trans
->transid
;
3378 block_group
->disk_cache_state
= dcs
;
3379 spin_unlock(&block_group
->lock
);
3384 int btrfs_setup_space_cache(struct btrfs_trans_handle
*trans
,
3385 struct btrfs_root
*root
)
3387 struct btrfs_block_group_cache
*cache
, *tmp
;
3388 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3389 struct btrfs_path
*path
;
3391 if (list_empty(&cur_trans
->dirty_bgs
) ||
3392 !btrfs_test_opt(root
, SPACE_CACHE
))
3395 path
= btrfs_alloc_path();
3399 /* Could add new block groups, use _safe just in case */
3400 list_for_each_entry_safe(cache
, tmp
, &cur_trans
->dirty_bgs
,
3402 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
)
3403 cache_save_setup(cache
, trans
, path
);
3406 btrfs_free_path(path
);
3411 * transaction commit does final block group cache writeback during a
3412 * critical section where nothing is allowed to change the FS. This is
3413 * required in order for the cache to actually match the block group,
3414 * but can introduce a lot of latency into the commit.
3416 * So, btrfs_start_dirty_block_groups is here to kick off block group
3417 * cache IO. There's a chance we'll have to redo some of it if the
3418 * block group changes again during the commit, but it greatly reduces
3419 * the commit latency by getting rid of the easy block groups while
3420 * we're still allowing others to join the commit.
3422 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle
*trans
,
3423 struct btrfs_root
*root
)
3425 struct btrfs_block_group_cache
*cache
;
3426 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3429 struct btrfs_path
*path
= NULL
;
3431 struct list_head
*io
= &cur_trans
->io_bgs
;
3432 int num_started
= 0;
3435 spin_lock(&cur_trans
->dirty_bgs_lock
);
3436 if (list_empty(&cur_trans
->dirty_bgs
)) {
3437 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3440 list_splice_init(&cur_trans
->dirty_bgs
, &dirty
);
3441 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3445 * make sure all the block groups on our dirty list actually
3448 btrfs_create_pending_block_groups(trans
, root
);
3451 path
= btrfs_alloc_path();
3457 * cache_write_mutex is here only to save us from balance or automatic
3458 * removal of empty block groups deleting this block group while we are
3459 * writing out the cache
3461 mutex_lock(&trans
->transaction
->cache_write_mutex
);
3462 while (!list_empty(&dirty
)) {
3463 cache
= list_first_entry(&dirty
,
3464 struct btrfs_block_group_cache
,
3467 * this can happen if something re-dirties a block
3468 * group that is already under IO. Just wait for it to
3469 * finish and then do it all again
3471 if (!list_empty(&cache
->io_list
)) {
3472 list_del_init(&cache
->io_list
);
3473 btrfs_wait_cache_io(root
, trans
, cache
,
3474 &cache
->io_ctl
, path
,
3475 cache
->key
.objectid
);
3476 btrfs_put_block_group(cache
);
3481 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3482 * if it should update the cache_state. Don't delete
3483 * until after we wait.
3485 * Since we're not running in the commit critical section
3486 * we need the dirty_bgs_lock to protect from update_block_group
3488 spin_lock(&cur_trans
->dirty_bgs_lock
);
3489 list_del_init(&cache
->dirty_list
);
3490 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3494 cache_save_setup(cache
, trans
, path
);
3496 if (cache
->disk_cache_state
== BTRFS_DC_SETUP
) {
3497 cache
->io_ctl
.inode
= NULL
;
3498 ret
= btrfs_write_out_cache(root
, trans
, cache
, path
);
3499 if (ret
== 0 && cache
->io_ctl
.inode
) {
3504 * the cache_write_mutex is protecting
3507 list_add_tail(&cache
->io_list
, io
);
3510 * if we failed to write the cache, the
3511 * generation will be bad and life goes on
3517 ret
= write_one_cache_group(trans
, root
, path
, cache
);
3519 * Our block group might still be attached to the list
3520 * of new block groups in the transaction handle of some
3521 * other task (struct btrfs_trans_handle->new_bgs). This
3522 * means its block group item isn't yet in the extent
3523 * tree. If this happens ignore the error, as we will
3524 * try again later in the critical section of the
3525 * transaction commit.
3527 if (ret
== -ENOENT
) {
3529 spin_lock(&cur_trans
->dirty_bgs_lock
);
3530 if (list_empty(&cache
->dirty_list
)) {
3531 list_add_tail(&cache
->dirty_list
,
3532 &cur_trans
->dirty_bgs
);
3533 btrfs_get_block_group(cache
);
3535 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3537 btrfs_abort_transaction(trans
, root
, ret
);
3541 /* if its not on the io list, we need to put the block group */
3543 btrfs_put_block_group(cache
);
3549 * Avoid blocking other tasks for too long. It might even save
3550 * us from writing caches for block groups that are going to be
3553 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
3554 mutex_lock(&trans
->transaction
->cache_write_mutex
);
3556 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
3559 * go through delayed refs for all the stuff we've just kicked off
3560 * and then loop back (just once)
3562 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
3563 if (!ret
&& loops
== 0) {
3565 spin_lock(&cur_trans
->dirty_bgs_lock
);
3566 list_splice_init(&cur_trans
->dirty_bgs
, &dirty
);
3568 * dirty_bgs_lock protects us from concurrent block group
3569 * deletes too (not just cache_write_mutex).
3571 if (!list_empty(&dirty
)) {
3572 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3575 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3578 btrfs_free_path(path
);
3582 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
3583 struct btrfs_root
*root
)
3585 struct btrfs_block_group_cache
*cache
;
3586 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3589 struct btrfs_path
*path
;
3590 struct list_head
*io
= &cur_trans
->io_bgs
;
3591 int num_started
= 0;
3593 path
= btrfs_alloc_path();
3598 * We don't need the lock here since we are protected by the transaction
3599 * commit. We want to do the cache_save_setup first and then run the
3600 * delayed refs to make sure we have the best chance at doing this all
3603 while (!list_empty(&cur_trans
->dirty_bgs
)) {
3604 cache
= list_first_entry(&cur_trans
->dirty_bgs
,
3605 struct btrfs_block_group_cache
,
3609 * this can happen if cache_save_setup re-dirties a block
3610 * group that is already under IO. Just wait for it to
3611 * finish and then do it all again
3613 if (!list_empty(&cache
->io_list
)) {
3614 list_del_init(&cache
->io_list
);
3615 btrfs_wait_cache_io(root
, trans
, cache
,
3616 &cache
->io_ctl
, path
,
3617 cache
->key
.objectid
);
3618 btrfs_put_block_group(cache
);
3622 * don't remove from the dirty list until after we've waited
3625 list_del_init(&cache
->dirty_list
);
3628 cache_save_setup(cache
, trans
, path
);
3631 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long) -1);
3633 if (!ret
&& cache
->disk_cache_state
== BTRFS_DC_SETUP
) {
3634 cache
->io_ctl
.inode
= NULL
;
3635 ret
= btrfs_write_out_cache(root
, trans
, cache
, path
);
3636 if (ret
== 0 && cache
->io_ctl
.inode
) {
3639 list_add_tail(&cache
->io_list
, io
);
3642 * if we failed to write the cache, the
3643 * generation will be bad and life goes on
3649 ret
= write_one_cache_group(trans
, root
, path
, cache
);
3651 btrfs_abort_transaction(trans
, root
, ret
);
3654 /* if its not on the io list, we need to put the block group */
3656 btrfs_put_block_group(cache
);
3659 while (!list_empty(io
)) {
3660 cache
= list_first_entry(io
, struct btrfs_block_group_cache
,
3662 list_del_init(&cache
->io_list
);
3663 btrfs_wait_cache_io(root
, trans
, cache
,
3664 &cache
->io_ctl
, path
, cache
->key
.objectid
);
3665 btrfs_put_block_group(cache
);
3668 btrfs_free_path(path
);
3672 int btrfs_extent_readonly(struct btrfs_root
*root
, u64 bytenr
)
3674 struct btrfs_block_group_cache
*block_group
;
3677 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
3678 if (!block_group
|| block_group
->ro
)
3681 btrfs_put_block_group(block_group
);
3685 static const char *alloc_name(u64 flags
)
3688 case BTRFS_BLOCK_GROUP_METADATA
|BTRFS_BLOCK_GROUP_DATA
:
3690 case BTRFS_BLOCK_GROUP_METADATA
:
3692 case BTRFS_BLOCK_GROUP_DATA
:
3694 case BTRFS_BLOCK_GROUP_SYSTEM
:
3698 return "invalid-combination";
3702 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
3703 u64 total_bytes
, u64 bytes_used
,
3704 struct btrfs_space_info
**space_info
)
3706 struct btrfs_space_info
*found
;
3711 if (flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3712 BTRFS_BLOCK_GROUP_RAID10
))
3717 found
= __find_space_info(info
, flags
);
3719 spin_lock(&found
->lock
);
3720 found
->total_bytes
+= total_bytes
;
3721 found
->disk_total
+= total_bytes
* factor
;
3722 found
->bytes_used
+= bytes_used
;
3723 found
->disk_used
+= bytes_used
* factor
;
3724 if (total_bytes
> 0)
3726 spin_unlock(&found
->lock
);
3727 *space_info
= found
;
3730 found
= kzalloc(sizeof(*found
), GFP_NOFS
);
3734 ret
= percpu_counter_init(&found
->total_bytes_pinned
, 0, GFP_KERNEL
);
3740 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
3741 INIT_LIST_HEAD(&found
->block_groups
[i
]);
3742 init_rwsem(&found
->groups_sem
);
3743 spin_lock_init(&found
->lock
);
3744 found
->flags
= flags
& BTRFS_BLOCK_GROUP_TYPE_MASK
;
3745 found
->total_bytes
= total_bytes
;
3746 found
->disk_total
= total_bytes
* factor
;
3747 found
->bytes_used
= bytes_used
;
3748 found
->disk_used
= bytes_used
* factor
;
3749 found
->bytes_pinned
= 0;
3750 found
->bytes_reserved
= 0;
3751 found
->bytes_readonly
= 0;
3752 found
->bytes_may_use
= 0;
3754 found
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3755 found
->chunk_alloc
= 0;
3757 init_waitqueue_head(&found
->wait
);
3758 INIT_LIST_HEAD(&found
->ro_bgs
);
3760 ret
= kobject_init_and_add(&found
->kobj
, &space_info_ktype
,
3761 info
->space_info_kobj
, "%s",
3762 alloc_name(found
->flags
));
3768 *space_info
= found
;
3769 list_add_rcu(&found
->list
, &info
->space_info
);
3770 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3771 info
->data_sinfo
= found
;
3776 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
3778 u64 extra_flags
= chunk_to_extended(flags
) &
3779 BTRFS_EXTENDED_PROFILE_MASK
;
3781 write_seqlock(&fs_info
->profiles_lock
);
3782 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3783 fs_info
->avail_data_alloc_bits
|= extra_flags
;
3784 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3785 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
3786 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3787 fs_info
->avail_system_alloc_bits
|= extra_flags
;
3788 write_sequnlock(&fs_info
->profiles_lock
);
3792 * returns target flags in extended format or 0 if restripe for this
3793 * chunk_type is not in progress
3795 * should be called with either volume_mutex or balance_lock held
3797 static u64
get_restripe_target(struct btrfs_fs_info
*fs_info
, u64 flags
)
3799 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3805 if (flags
& BTRFS_BLOCK_GROUP_DATA
&&
3806 bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3807 target
= BTRFS_BLOCK_GROUP_DATA
| bctl
->data
.target
;
3808 } else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
&&
3809 bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3810 target
= BTRFS_BLOCK_GROUP_SYSTEM
| bctl
->sys
.target
;
3811 } else if (flags
& BTRFS_BLOCK_GROUP_METADATA
&&
3812 bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3813 target
= BTRFS_BLOCK_GROUP_METADATA
| bctl
->meta
.target
;
3820 * @flags: available profiles in extended format (see ctree.h)
3822 * Returns reduced profile in chunk format. If profile changing is in
3823 * progress (either running or paused) picks the target profile (if it's
3824 * already available), otherwise falls back to plain reducing.
3826 static u64
btrfs_reduce_alloc_profile(struct btrfs_root
*root
, u64 flags
)
3828 u64 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
3834 * see if restripe for this chunk_type is in progress, if so
3835 * try to reduce to the target profile
3837 spin_lock(&root
->fs_info
->balance_lock
);
3838 target
= get_restripe_target(root
->fs_info
, flags
);
3840 /* pick target profile only if it's already available */
3841 if ((flags
& target
) & BTRFS_EXTENDED_PROFILE_MASK
) {
3842 spin_unlock(&root
->fs_info
->balance_lock
);
3843 return extended_to_chunk(target
);
3846 spin_unlock(&root
->fs_info
->balance_lock
);
3848 /* First, mask out the RAID levels which aren't possible */
3849 for (raid_type
= 0; raid_type
< BTRFS_NR_RAID_TYPES
; raid_type
++) {
3850 if (num_devices
>= btrfs_raid_array
[raid_type
].devs_min
)
3851 allowed
|= btrfs_raid_group
[raid_type
];
3855 if (allowed
& BTRFS_BLOCK_GROUP_RAID6
)
3856 allowed
= BTRFS_BLOCK_GROUP_RAID6
;
3857 else if (allowed
& BTRFS_BLOCK_GROUP_RAID5
)
3858 allowed
= BTRFS_BLOCK_GROUP_RAID5
;
3859 else if (allowed
& BTRFS_BLOCK_GROUP_RAID10
)
3860 allowed
= BTRFS_BLOCK_GROUP_RAID10
;
3861 else if (allowed
& BTRFS_BLOCK_GROUP_RAID1
)
3862 allowed
= BTRFS_BLOCK_GROUP_RAID1
;
3863 else if (allowed
& BTRFS_BLOCK_GROUP_RAID0
)
3864 allowed
= BTRFS_BLOCK_GROUP_RAID0
;
3866 flags
&= ~BTRFS_BLOCK_GROUP_PROFILE_MASK
;
3868 return extended_to_chunk(flags
| allowed
);
3871 static u64
get_alloc_profile(struct btrfs_root
*root
, u64 orig_flags
)
3878 seq
= read_seqbegin(&root
->fs_info
->profiles_lock
);
3880 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3881 flags
|= root
->fs_info
->avail_data_alloc_bits
;
3882 else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3883 flags
|= root
->fs_info
->avail_system_alloc_bits
;
3884 else if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3885 flags
|= root
->fs_info
->avail_metadata_alloc_bits
;
3886 } while (read_seqretry(&root
->fs_info
->profiles_lock
, seq
));
3888 return btrfs_reduce_alloc_profile(root
, flags
);
3891 u64
btrfs_get_alloc_profile(struct btrfs_root
*root
, int data
)
3897 flags
= BTRFS_BLOCK_GROUP_DATA
;
3898 else if (root
== root
->fs_info
->chunk_root
)
3899 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
3901 flags
= BTRFS_BLOCK_GROUP_METADATA
;
3903 ret
= get_alloc_profile(root
, flags
);
3908 * This will check the space that the inode allocates from to make sure we have
3909 * enough space for bytes.
3911 int btrfs_check_data_free_space(struct inode
*inode
, u64 bytes
, u64 write_bytes
)
3913 struct btrfs_space_info
*data_sinfo
;
3914 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3915 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3918 int need_commit
= 2;
3919 int have_pinned_space
;
3921 /* make sure bytes are sectorsize aligned */
3922 bytes
= ALIGN(bytes
, root
->sectorsize
);
3924 if (btrfs_is_free_space_inode(inode
)) {
3926 ASSERT(current
->journal_info
);
3929 data_sinfo
= fs_info
->data_sinfo
;
3934 /* make sure we have enough space to handle the data first */
3935 spin_lock(&data_sinfo
->lock
);
3936 used
= data_sinfo
->bytes_used
+ data_sinfo
->bytes_reserved
+
3937 data_sinfo
->bytes_pinned
+ data_sinfo
->bytes_readonly
+
3938 data_sinfo
->bytes_may_use
;
3940 if (used
+ bytes
> data_sinfo
->total_bytes
) {
3941 struct btrfs_trans_handle
*trans
;
3944 * if we don't have enough free bytes in this space then we need
3945 * to alloc a new chunk.
3947 if (!data_sinfo
->full
) {
3950 data_sinfo
->force_alloc
= CHUNK_ALLOC_FORCE
;
3951 spin_unlock(&data_sinfo
->lock
);
3953 alloc_target
= btrfs_get_alloc_profile(root
, 1);
3955 * It is ugly that we don't call nolock join
3956 * transaction for the free space inode case here.
3957 * But it is safe because we only do the data space
3958 * reservation for the free space cache in the
3959 * transaction context, the common join transaction
3960 * just increase the counter of the current transaction
3961 * handler, doesn't try to acquire the trans_lock of
3964 trans
= btrfs_join_transaction(root
);
3966 return PTR_ERR(trans
);
3968 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
3970 CHUNK_ALLOC_NO_FORCE
);
3971 btrfs_end_transaction(trans
, root
);
3976 have_pinned_space
= 1;
3982 data_sinfo
= fs_info
->data_sinfo
;
3988 * If we don't have enough pinned space to deal with this
3989 * allocation, and no removed chunk in current transaction,
3990 * don't bother committing the transaction.
3992 have_pinned_space
= percpu_counter_compare(
3993 &data_sinfo
->total_bytes_pinned
,
3994 used
+ bytes
- data_sinfo
->total_bytes
);
3995 spin_unlock(&data_sinfo
->lock
);
3997 /* commit the current transaction and try again */
4000 !atomic_read(&root
->fs_info
->open_ioctl_trans
)) {
4003 if (need_commit
> 0)
4004 btrfs_wait_ordered_roots(fs_info
, -1);
4006 trans
= btrfs_join_transaction(root
);
4008 return PTR_ERR(trans
);
4009 if (have_pinned_space
>= 0 ||
4010 trans
->transaction
->have_free_bgs
||
4012 ret
= btrfs_commit_transaction(trans
, root
);
4016 * make sure that all running delayed iput are
4019 down_write(&root
->fs_info
->delayed_iput_sem
);
4020 up_write(&root
->fs_info
->delayed_iput_sem
);
4023 btrfs_end_transaction(trans
, root
);
4027 trace_btrfs_space_reservation(root
->fs_info
,
4028 "space_info:enospc",
4029 data_sinfo
->flags
, bytes
, 1);
4032 ret
= btrfs_qgroup_reserve(root
, write_bytes
);
4035 data_sinfo
->bytes_may_use
+= bytes
;
4036 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
4037 data_sinfo
->flags
, bytes
, 1);
4039 spin_unlock(&data_sinfo
->lock
);
4045 * Called if we need to clear a data reservation for this inode.
4047 void btrfs_free_reserved_data_space(struct inode
*inode
, u64 bytes
)
4049 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4050 struct btrfs_space_info
*data_sinfo
;
4052 /* make sure bytes are sectorsize aligned */
4053 bytes
= ALIGN(bytes
, root
->sectorsize
);
4055 data_sinfo
= root
->fs_info
->data_sinfo
;
4056 spin_lock(&data_sinfo
->lock
);
4057 WARN_ON(data_sinfo
->bytes_may_use
< bytes
);
4058 data_sinfo
->bytes_may_use
-= bytes
;
4059 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
4060 data_sinfo
->flags
, bytes
, 0);
4061 spin_unlock(&data_sinfo
->lock
);
4064 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
4066 struct list_head
*head
= &info
->space_info
;
4067 struct btrfs_space_info
*found
;
4070 list_for_each_entry_rcu(found
, head
, list
) {
4071 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
4072 found
->force_alloc
= CHUNK_ALLOC_FORCE
;
4077 static inline u64
calc_global_rsv_need_space(struct btrfs_block_rsv
*global
)
4079 return (global
->size
<< 1);
4082 static int should_alloc_chunk(struct btrfs_root
*root
,
4083 struct btrfs_space_info
*sinfo
, int force
)
4085 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
4086 u64 num_bytes
= sinfo
->total_bytes
- sinfo
->bytes_readonly
;
4087 u64 num_allocated
= sinfo
->bytes_used
+ sinfo
->bytes_reserved
;
4090 if (force
== CHUNK_ALLOC_FORCE
)
4094 * We need to take into account the global rsv because for all intents
4095 * and purposes it's used space. Don't worry about locking the
4096 * global_rsv, it doesn't change except when the transaction commits.
4098 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
4099 num_allocated
+= calc_global_rsv_need_space(global_rsv
);
4102 * in limited mode, we want to have some free space up to
4103 * about 1% of the FS size.
4105 if (force
== CHUNK_ALLOC_LIMITED
) {
4106 thresh
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
4107 thresh
= max_t(u64
, 64 * 1024 * 1024,
4108 div_factor_fine(thresh
, 1));
4110 if (num_bytes
- num_allocated
< thresh
)
4114 if (num_allocated
+ 2 * 1024 * 1024 < div_factor(num_bytes
, 8))
4119 static u64
get_profile_num_devs(struct btrfs_root
*root
, u64 type
)
4123 if (type
& (BTRFS_BLOCK_GROUP_RAID10
|
4124 BTRFS_BLOCK_GROUP_RAID0
|
4125 BTRFS_BLOCK_GROUP_RAID5
|
4126 BTRFS_BLOCK_GROUP_RAID6
))
4127 num_dev
= root
->fs_info
->fs_devices
->rw_devices
;
4128 else if (type
& BTRFS_BLOCK_GROUP_RAID1
)
4131 num_dev
= 1; /* DUP or single */
4137 * If @is_allocation is true, reserve space in the system space info necessary
4138 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4141 void check_system_chunk(struct btrfs_trans_handle
*trans
,
4142 struct btrfs_root
*root
,
4145 struct btrfs_space_info
*info
;
4152 * Needed because we can end up allocating a system chunk and for an
4153 * atomic and race free space reservation in the chunk block reserve.
4155 ASSERT(mutex_is_locked(&root
->fs_info
->chunk_mutex
));
4157 info
= __find_space_info(root
->fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
4158 spin_lock(&info
->lock
);
4159 left
= info
->total_bytes
- info
->bytes_used
- info
->bytes_pinned
-
4160 info
->bytes_reserved
- info
->bytes_readonly
-
4161 info
->bytes_may_use
;
4162 spin_unlock(&info
->lock
);
4164 num_devs
= get_profile_num_devs(root
, type
);
4166 /* num_devs device items to update and 1 chunk item to add or remove */
4167 thresh
= btrfs_calc_trunc_metadata_size(root
, num_devs
) +
4168 btrfs_calc_trans_metadata_size(root
, 1);
4170 if (left
< thresh
&& btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
4171 btrfs_info(root
->fs_info
, "left=%llu, need=%llu, flags=%llu",
4172 left
, thresh
, type
);
4173 dump_space_info(info
, 0, 0);
4176 if (left
< thresh
) {
4179 flags
= btrfs_get_alloc_profile(root
->fs_info
->chunk_root
, 0);
4181 * Ignore failure to create system chunk. We might end up not
4182 * needing it, as we might not need to COW all nodes/leafs from
4183 * the paths we visit in the chunk tree (they were already COWed
4184 * or created in the current transaction for example).
4186 ret
= btrfs_alloc_chunk(trans
, root
, flags
);
4190 ret
= btrfs_block_rsv_add(root
->fs_info
->chunk_root
,
4191 &root
->fs_info
->chunk_block_rsv
,
4192 thresh
, BTRFS_RESERVE_NO_FLUSH
);
4194 trans
->chunk_bytes_reserved
+= thresh
;
4198 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
4199 struct btrfs_root
*extent_root
, u64 flags
, int force
)
4201 struct btrfs_space_info
*space_info
;
4202 struct btrfs_fs_info
*fs_info
= extent_root
->fs_info
;
4203 int wait_for_alloc
= 0;
4206 /* Don't re-enter if we're already allocating a chunk */
4207 if (trans
->allocating_chunk
)
4210 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
4212 ret
= update_space_info(extent_root
->fs_info
, flags
,
4214 BUG_ON(ret
); /* -ENOMEM */
4216 BUG_ON(!space_info
); /* Logic error */
4219 spin_lock(&space_info
->lock
);
4220 if (force
< space_info
->force_alloc
)
4221 force
= space_info
->force_alloc
;
4222 if (space_info
->full
) {
4223 if (should_alloc_chunk(extent_root
, space_info
, force
))
4227 spin_unlock(&space_info
->lock
);
4231 if (!should_alloc_chunk(extent_root
, space_info
, force
)) {
4232 spin_unlock(&space_info
->lock
);
4234 } else if (space_info
->chunk_alloc
) {
4237 space_info
->chunk_alloc
= 1;
4240 spin_unlock(&space_info
->lock
);
4242 mutex_lock(&fs_info
->chunk_mutex
);
4245 * The chunk_mutex is held throughout the entirety of a chunk
4246 * allocation, so once we've acquired the chunk_mutex we know that the
4247 * other guy is done and we need to recheck and see if we should
4250 if (wait_for_alloc
) {
4251 mutex_unlock(&fs_info
->chunk_mutex
);
4256 trans
->allocating_chunk
= true;
4259 * If we have mixed data/metadata chunks we want to make sure we keep
4260 * allocating mixed chunks instead of individual chunks.
4262 if (btrfs_mixed_space_info(space_info
))
4263 flags
|= (BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
);
4266 * if we're doing a data chunk, go ahead and make sure that
4267 * we keep a reasonable number of metadata chunks allocated in the
4270 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
4271 fs_info
->data_chunk_allocations
++;
4272 if (!(fs_info
->data_chunk_allocations
%
4273 fs_info
->metadata_ratio
))
4274 force_metadata_allocation(fs_info
);
4278 * Check if we have enough space in SYSTEM chunk because we may need
4279 * to update devices.
4281 check_system_chunk(trans
, extent_root
, flags
);
4283 ret
= btrfs_alloc_chunk(trans
, extent_root
, flags
);
4284 trans
->allocating_chunk
= false;
4286 spin_lock(&space_info
->lock
);
4287 if (ret
< 0 && ret
!= -ENOSPC
)
4290 space_info
->full
= 1;
4294 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
4296 space_info
->chunk_alloc
= 0;
4297 spin_unlock(&space_info
->lock
);
4298 mutex_unlock(&fs_info
->chunk_mutex
);
4300 * When we allocate a new chunk we reserve space in the chunk block
4301 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4302 * add new nodes/leafs to it if we end up needing to do it when
4303 * inserting the chunk item and updating device items as part of the
4304 * second phase of chunk allocation, performed by
4305 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4306 * large number of new block groups to create in our transaction
4307 * handle's new_bgs list to avoid exhausting the chunk block reserve
4308 * in extreme cases - like having a single transaction create many new
4309 * block groups when starting to write out the free space caches of all
4310 * the block groups that were made dirty during the lifetime of the
4313 if (trans
->can_flush_pending_bgs
&&
4314 trans
->chunk_bytes_reserved
>= (2 * 1024 * 1024ull)) {
4315 btrfs_create_pending_block_groups(trans
, trans
->root
);
4316 btrfs_trans_release_chunk_metadata(trans
);
4321 static int can_overcommit(struct btrfs_root
*root
,
4322 struct btrfs_space_info
*space_info
, u64 bytes
,
4323 enum btrfs_reserve_flush_enum flush
)
4325 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
4326 u64 profile
= btrfs_get_alloc_profile(root
, 0);
4331 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4332 space_info
->bytes_pinned
+ space_info
->bytes_readonly
;
4335 * We only want to allow over committing if we have lots of actual space
4336 * free, but if we don't have enough space to handle the global reserve
4337 * space then we could end up having a real enospc problem when trying
4338 * to allocate a chunk or some other such important allocation.
4340 spin_lock(&global_rsv
->lock
);
4341 space_size
= calc_global_rsv_need_space(global_rsv
);
4342 spin_unlock(&global_rsv
->lock
);
4343 if (used
+ space_size
>= space_info
->total_bytes
)
4346 used
+= space_info
->bytes_may_use
;
4348 spin_lock(&root
->fs_info
->free_chunk_lock
);
4349 avail
= root
->fs_info
->free_chunk_space
;
4350 spin_unlock(&root
->fs_info
->free_chunk_lock
);
4353 * If we have dup, raid1 or raid10 then only half of the free
4354 * space is actually useable. For raid56, the space info used
4355 * doesn't include the parity drive, so we don't have to
4358 if (profile
& (BTRFS_BLOCK_GROUP_DUP
|
4359 BTRFS_BLOCK_GROUP_RAID1
|
4360 BTRFS_BLOCK_GROUP_RAID10
))
4364 * If we aren't flushing all things, let us overcommit up to
4365 * 1/2th of the space. If we can flush, don't let us overcommit
4366 * too much, let it overcommit up to 1/8 of the space.
4368 if (flush
== BTRFS_RESERVE_FLUSH_ALL
)
4373 if (used
+ bytes
< space_info
->total_bytes
+ avail
)
4378 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root
*root
,
4379 unsigned long nr_pages
, int nr_items
)
4381 struct super_block
*sb
= root
->fs_info
->sb
;
4383 if (down_read_trylock(&sb
->s_umount
)) {
4384 writeback_inodes_sb_nr(sb
, nr_pages
, WB_REASON_FS_FREE_SPACE
);
4385 up_read(&sb
->s_umount
);
4388 * We needn't worry the filesystem going from r/w to r/o though
4389 * we don't acquire ->s_umount mutex, because the filesystem
4390 * should guarantee the delalloc inodes list be empty after
4391 * the filesystem is readonly(all dirty pages are written to
4394 btrfs_start_delalloc_roots(root
->fs_info
, 0, nr_items
);
4395 if (!current
->journal_info
)
4396 btrfs_wait_ordered_roots(root
->fs_info
, nr_items
);
4400 static inline int calc_reclaim_items_nr(struct btrfs_root
*root
, u64 to_reclaim
)
4405 bytes
= btrfs_calc_trans_metadata_size(root
, 1);
4406 nr
= (int)div64_u64(to_reclaim
, bytes
);
4412 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4415 * shrink metadata reservation for delalloc
4417 static void shrink_delalloc(struct btrfs_root
*root
, u64 to_reclaim
, u64 orig
,
4420 struct btrfs_block_rsv
*block_rsv
;
4421 struct btrfs_space_info
*space_info
;
4422 struct btrfs_trans_handle
*trans
;
4426 unsigned long nr_pages
;
4429 enum btrfs_reserve_flush_enum flush
;
4431 /* Calc the number of the pages we need flush for space reservation */
4432 items
= calc_reclaim_items_nr(root
, to_reclaim
);
4433 to_reclaim
= items
* EXTENT_SIZE_PER_ITEM
;
4435 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4436 block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
4437 space_info
= block_rsv
->space_info
;
4439 delalloc_bytes
= percpu_counter_sum_positive(
4440 &root
->fs_info
->delalloc_bytes
);
4441 if (delalloc_bytes
== 0) {
4445 btrfs_wait_ordered_roots(root
->fs_info
, items
);
4450 while (delalloc_bytes
&& loops
< 3) {
4451 max_reclaim
= min(delalloc_bytes
, to_reclaim
);
4452 nr_pages
= max_reclaim
>> PAGE_CACHE_SHIFT
;
4453 btrfs_writeback_inodes_sb_nr(root
, nr_pages
, items
);
4455 * We need to wait for the async pages to actually start before
4458 max_reclaim
= atomic_read(&root
->fs_info
->async_delalloc_pages
);
4462 if (max_reclaim
<= nr_pages
)
4465 max_reclaim
-= nr_pages
;
4467 wait_event(root
->fs_info
->async_submit_wait
,
4468 atomic_read(&root
->fs_info
->async_delalloc_pages
) <=
4472 flush
= BTRFS_RESERVE_FLUSH_ALL
;
4474 flush
= BTRFS_RESERVE_NO_FLUSH
;
4475 spin_lock(&space_info
->lock
);
4476 if (can_overcommit(root
, space_info
, orig
, flush
)) {
4477 spin_unlock(&space_info
->lock
);
4480 spin_unlock(&space_info
->lock
);
4483 if (wait_ordered
&& !trans
) {
4484 btrfs_wait_ordered_roots(root
->fs_info
, items
);
4486 time_left
= schedule_timeout_killable(1);
4490 delalloc_bytes
= percpu_counter_sum_positive(
4491 &root
->fs_info
->delalloc_bytes
);
4496 * maybe_commit_transaction - possibly commit the transaction if its ok to
4497 * @root - the root we're allocating for
4498 * @bytes - the number of bytes we want to reserve
4499 * @force - force the commit
4501 * This will check to make sure that committing the transaction will actually
4502 * get us somewhere and then commit the transaction if it does. Otherwise it
4503 * will return -ENOSPC.
4505 static int may_commit_transaction(struct btrfs_root
*root
,
4506 struct btrfs_space_info
*space_info
,
4507 u64 bytes
, int force
)
4509 struct btrfs_block_rsv
*delayed_rsv
= &root
->fs_info
->delayed_block_rsv
;
4510 struct btrfs_trans_handle
*trans
;
4512 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4519 /* See if there is enough pinned space to make this reservation */
4520 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4525 * See if there is some space in the delayed insertion reservation for
4528 if (space_info
!= delayed_rsv
->space_info
)
4531 spin_lock(&delayed_rsv
->lock
);
4532 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4533 bytes
- delayed_rsv
->size
) >= 0) {
4534 spin_unlock(&delayed_rsv
->lock
);
4537 spin_unlock(&delayed_rsv
->lock
);
4540 trans
= btrfs_join_transaction(root
);
4544 return btrfs_commit_transaction(trans
, root
);
4548 FLUSH_DELAYED_ITEMS_NR
= 1,
4549 FLUSH_DELAYED_ITEMS
= 2,
4551 FLUSH_DELALLOC_WAIT
= 4,
4556 static int flush_space(struct btrfs_root
*root
,
4557 struct btrfs_space_info
*space_info
, u64 num_bytes
,
4558 u64 orig_bytes
, int state
)
4560 struct btrfs_trans_handle
*trans
;
4565 case FLUSH_DELAYED_ITEMS_NR
:
4566 case FLUSH_DELAYED_ITEMS
:
4567 if (state
== FLUSH_DELAYED_ITEMS_NR
)
4568 nr
= calc_reclaim_items_nr(root
, num_bytes
) * 2;
4572 trans
= btrfs_join_transaction(root
);
4573 if (IS_ERR(trans
)) {
4574 ret
= PTR_ERR(trans
);
4577 ret
= btrfs_run_delayed_items_nr(trans
, root
, nr
);
4578 btrfs_end_transaction(trans
, root
);
4580 case FLUSH_DELALLOC
:
4581 case FLUSH_DELALLOC_WAIT
:
4582 shrink_delalloc(root
, num_bytes
* 2, orig_bytes
,
4583 state
== FLUSH_DELALLOC_WAIT
);
4586 trans
= btrfs_join_transaction(root
);
4587 if (IS_ERR(trans
)) {
4588 ret
= PTR_ERR(trans
);
4591 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
4592 btrfs_get_alloc_profile(root
, 0),
4593 CHUNK_ALLOC_NO_FORCE
);
4594 btrfs_end_transaction(trans
, root
);
4599 ret
= may_commit_transaction(root
, space_info
, orig_bytes
, 0);
4610 btrfs_calc_reclaim_metadata_size(struct btrfs_root
*root
,
4611 struct btrfs_space_info
*space_info
)
4617 to_reclaim
= min_t(u64
, num_online_cpus() * 1024 * 1024,
4619 spin_lock(&space_info
->lock
);
4620 if (can_overcommit(root
, space_info
, to_reclaim
,
4621 BTRFS_RESERVE_FLUSH_ALL
)) {
4626 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4627 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4628 space_info
->bytes_may_use
;
4629 if (can_overcommit(root
, space_info
, 1024 * 1024,
4630 BTRFS_RESERVE_FLUSH_ALL
))
4631 expected
= div_factor_fine(space_info
->total_bytes
, 95);
4633 expected
= div_factor_fine(space_info
->total_bytes
, 90);
4635 if (used
> expected
)
4636 to_reclaim
= used
- expected
;
4639 to_reclaim
= min(to_reclaim
, space_info
->bytes_may_use
+
4640 space_info
->bytes_reserved
);
4642 spin_unlock(&space_info
->lock
);
4647 static inline int need_do_async_reclaim(struct btrfs_space_info
*space_info
,
4648 struct btrfs_fs_info
*fs_info
, u64 used
)
4650 u64 thresh
= div_factor_fine(space_info
->total_bytes
, 98);
4652 /* If we're just plain full then async reclaim just slows us down. */
4653 if (space_info
->bytes_used
>= thresh
)
4656 return (used
>= thresh
&& !btrfs_fs_closing(fs_info
) &&
4657 !test_bit(BTRFS_FS_STATE_REMOUNTING
, &fs_info
->fs_state
));
4660 static int btrfs_need_do_async_reclaim(struct btrfs_space_info
*space_info
,
4661 struct btrfs_fs_info
*fs_info
,
4666 spin_lock(&space_info
->lock
);
4668 * We run out of space and have not got any free space via flush_space,
4669 * so don't bother doing async reclaim.
4671 if (flush_state
> COMMIT_TRANS
&& space_info
->full
) {
4672 spin_unlock(&space_info
->lock
);
4676 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4677 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4678 space_info
->bytes_may_use
;
4679 if (need_do_async_reclaim(space_info
, fs_info
, used
)) {
4680 spin_unlock(&space_info
->lock
);
4683 spin_unlock(&space_info
->lock
);
4688 static void btrfs_async_reclaim_metadata_space(struct work_struct
*work
)
4690 struct btrfs_fs_info
*fs_info
;
4691 struct btrfs_space_info
*space_info
;
4695 fs_info
= container_of(work
, struct btrfs_fs_info
, async_reclaim_work
);
4696 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4698 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
->fs_root
,
4703 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
4705 flush_space(fs_info
->fs_root
, space_info
, to_reclaim
,
4706 to_reclaim
, flush_state
);
4708 if (!btrfs_need_do_async_reclaim(space_info
, fs_info
,
4711 } while (flush_state
< COMMIT_TRANS
);
4714 void btrfs_init_async_reclaim_work(struct work_struct
*work
)
4716 INIT_WORK(work
, btrfs_async_reclaim_metadata_space
);
4720 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4721 * @root - the root we're allocating for
4722 * @block_rsv - the block_rsv we're allocating for
4723 * @orig_bytes - the number of bytes we want
4724 * @flush - whether or not we can flush to make our reservation
4726 * This will reserve orgi_bytes number of bytes from the space info associated
4727 * with the block_rsv. If there is not enough space it will make an attempt to
4728 * flush out space to make room. It will do this by flushing delalloc if
4729 * possible or committing the transaction. If flush is 0 then no attempts to
4730 * regain reservations will be made and this will fail if there is not enough
4733 static int reserve_metadata_bytes(struct btrfs_root
*root
,
4734 struct btrfs_block_rsv
*block_rsv
,
4736 enum btrfs_reserve_flush_enum flush
)
4738 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
4740 u64 num_bytes
= orig_bytes
;
4741 int flush_state
= FLUSH_DELAYED_ITEMS_NR
;
4743 bool flushing
= false;
4747 spin_lock(&space_info
->lock
);
4749 * We only want to wait if somebody other than us is flushing and we
4750 * are actually allowed to flush all things.
4752 while (flush
== BTRFS_RESERVE_FLUSH_ALL
&& !flushing
&&
4753 space_info
->flush
) {
4754 spin_unlock(&space_info
->lock
);
4756 * If we have a trans handle we can't wait because the flusher
4757 * may have to commit the transaction, which would mean we would
4758 * deadlock since we are waiting for the flusher to finish, but
4759 * hold the current transaction open.
4761 if (current
->journal_info
)
4763 ret
= wait_event_killable(space_info
->wait
, !space_info
->flush
);
4764 /* Must have been killed, return */
4768 spin_lock(&space_info
->lock
);
4772 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4773 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4774 space_info
->bytes_may_use
;
4777 * The idea here is that we've not already over-reserved the block group
4778 * then we can go ahead and save our reservation first and then start
4779 * flushing if we need to. Otherwise if we've already overcommitted
4780 * lets start flushing stuff first and then come back and try to make
4783 if (used
<= space_info
->total_bytes
) {
4784 if (used
+ orig_bytes
<= space_info
->total_bytes
) {
4785 space_info
->bytes_may_use
+= orig_bytes
;
4786 trace_btrfs_space_reservation(root
->fs_info
,
4787 "space_info", space_info
->flags
, orig_bytes
, 1);
4791 * Ok set num_bytes to orig_bytes since we aren't
4792 * overocmmitted, this way we only try and reclaim what
4795 num_bytes
= orig_bytes
;
4799 * Ok we're over committed, set num_bytes to the overcommitted
4800 * amount plus the amount of bytes that we need for this
4803 num_bytes
= used
- space_info
->total_bytes
+
4807 if (ret
&& can_overcommit(root
, space_info
, orig_bytes
, flush
)) {
4808 space_info
->bytes_may_use
+= orig_bytes
;
4809 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
4810 space_info
->flags
, orig_bytes
,
4816 * Couldn't make our reservation, save our place so while we're trying
4817 * to reclaim space we can actually use it instead of somebody else
4818 * stealing it from us.
4820 * We make the other tasks wait for the flush only when we can flush
4823 if (ret
&& flush
!= BTRFS_RESERVE_NO_FLUSH
) {
4825 space_info
->flush
= 1;
4826 } else if (!ret
&& space_info
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
4829 * We will do the space reservation dance during log replay,
4830 * which means we won't have fs_info->fs_root set, so don't do
4831 * the async reclaim as we will panic.
4833 if (!root
->fs_info
->log_root_recovering
&&
4834 need_do_async_reclaim(space_info
, root
->fs_info
, used
) &&
4835 !work_busy(&root
->fs_info
->async_reclaim_work
))
4836 queue_work(system_unbound_wq
,
4837 &root
->fs_info
->async_reclaim_work
);
4839 spin_unlock(&space_info
->lock
);
4841 if (!ret
|| flush
== BTRFS_RESERVE_NO_FLUSH
)
4844 ret
= flush_space(root
, space_info
, num_bytes
, orig_bytes
,
4849 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4850 * would happen. So skip delalloc flush.
4852 if (flush
== BTRFS_RESERVE_FLUSH_LIMIT
&&
4853 (flush_state
== FLUSH_DELALLOC
||
4854 flush_state
== FLUSH_DELALLOC_WAIT
))
4855 flush_state
= ALLOC_CHUNK
;
4859 else if (flush
== BTRFS_RESERVE_FLUSH_LIMIT
&&
4860 flush_state
< COMMIT_TRANS
)
4862 else if (flush
== BTRFS_RESERVE_FLUSH_ALL
&&
4863 flush_state
<= COMMIT_TRANS
)
4867 if (ret
== -ENOSPC
&&
4868 unlikely(root
->orphan_cleanup_state
== ORPHAN_CLEANUP_STARTED
)) {
4869 struct btrfs_block_rsv
*global_rsv
=
4870 &root
->fs_info
->global_block_rsv
;
4872 if (block_rsv
!= global_rsv
&&
4873 !block_rsv_use_bytes(global_rsv
, orig_bytes
))
4877 trace_btrfs_space_reservation(root
->fs_info
,
4878 "space_info:enospc",
4879 space_info
->flags
, orig_bytes
, 1);
4881 spin_lock(&space_info
->lock
);
4882 space_info
->flush
= 0;
4883 wake_up_all(&space_info
->wait
);
4884 spin_unlock(&space_info
->lock
);
4889 static struct btrfs_block_rsv
*get_block_rsv(
4890 const struct btrfs_trans_handle
*trans
,
4891 const struct btrfs_root
*root
)
4893 struct btrfs_block_rsv
*block_rsv
= NULL
;
4895 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
4896 (root
== root
->fs_info
->csum_root
&& trans
->adding_csums
) ||
4897 (root
== root
->fs_info
->uuid_root
))
4898 block_rsv
= trans
->block_rsv
;
4901 block_rsv
= root
->block_rsv
;
4904 block_rsv
= &root
->fs_info
->empty_block_rsv
;
4909 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
4913 spin_lock(&block_rsv
->lock
);
4914 if (block_rsv
->reserved
>= num_bytes
) {
4915 block_rsv
->reserved
-= num_bytes
;
4916 if (block_rsv
->reserved
< block_rsv
->size
)
4917 block_rsv
->full
= 0;
4920 spin_unlock(&block_rsv
->lock
);
4924 static void block_rsv_add_bytes(struct btrfs_block_rsv
*block_rsv
,
4925 u64 num_bytes
, int update_size
)
4927 spin_lock(&block_rsv
->lock
);
4928 block_rsv
->reserved
+= num_bytes
;
4930 block_rsv
->size
+= num_bytes
;
4931 else if (block_rsv
->reserved
>= block_rsv
->size
)
4932 block_rsv
->full
= 1;
4933 spin_unlock(&block_rsv
->lock
);
4936 int btrfs_cond_migrate_bytes(struct btrfs_fs_info
*fs_info
,
4937 struct btrfs_block_rsv
*dest
, u64 num_bytes
,
4940 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
4943 if (global_rsv
->space_info
!= dest
->space_info
)
4946 spin_lock(&global_rsv
->lock
);
4947 min_bytes
= div_factor(global_rsv
->size
, min_factor
);
4948 if (global_rsv
->reserved
< min_bytes
+ num_bytes
) {
4949 spin_unlock(&global_rsv
->lock
);
4952 global_rsv
->reserved
-= num_bytes
;
4953 if (global_rsv
->reserved
< global_rsv
->size
)
4954 global_rsv
->full
= 0;
4955 spin_unlock(&global_rsv
->lock
);
4957 block_rsv_add_bytes(dest
, num_bytes
, 1);
4961 static void block_rsv_release_bytes(struct btrfs_fs_info
*fs_info
,
4962 struct btrfs_block_rsv
*block_rsv
,
4963 struct btrfs_block_rsv
*dest
, u64 num_bytes
)
4965 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
4967 spin_lock(&block_rsv
->lock
);
4968 if (num_bytes
== (u64
)-1)
4969 num_bytes
= block_rsv
->size
;
4970 block_rsv
->size
-= num_bytes
;
4971 if (block_rsv
->reserved
>= block_rsv
->size
) {
4972 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
4973 block_rsv
->reserved
= block_rsv
->size
;
4974 block_rsv
->full
= 1;
4978 spin_unlock(&block_rsv
->lock
);
4980 if (num_bytes
> 0) {
4982 spin_lock(&dest
->lock
);
4986 bytes_to_add
= dest
->size
- dest
->reserved
;
4987 bytes_to_add
= min(num_bytes
, bytes_to_add
);
4988 dest
->reserved
+= bytes_to_add
;
4989 if (dest
->reserved
>= dest
->size
)
4991 num_bytes
-= bytes_to_add
;
4993 spin_unlock(&dest
->lock
);
4996 spin_lock(&space_info
->lock
);
4997 space_info
->bytes_may_use
-= num_bytes
;
4998 trace_btrfs_space_reservation(fs_info
, "space_info",
4999 space_info
->flags
, num_bytes
, 0);
5000 spin_unlock(&space_info
->lock
);
5005 static int block_rsv_migrate_bytes(struct btrfs_block_rsv
*src
,
5006 struct btrfs_block_rsv
*dst
, u64 num_bytes
)
5010 ret
= block_rsv_use_bytes(src
, num_bytes
);
5014 block_rsv_add_bytes(dst
, num_bytes
, 1);
5018 void btrfs_init_block_rsv(struct btrfs_block_rsv
*rsv
, unsigned short type
)
5020 memset(rsv
, 0, sizeof(*rsv
));
5021 spin_lock_init(&rsv
->lock
);
5025 struct btrfs_block_rsv
*btrfs_alloc_block_rsv(struct btrfs_root
*root
,
5026 unsigned short type
)
5028 struct btrfs_block_rsv
*block_rsv
;
5029 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5031 block_rsv
= kmalloc(sizeof(*block_rsv
), GFP_NOFS
);
5035 btrfs_init_block_rsv(block_rsv
, type
);
5036 block_rsv
->space_info
= __find_space_info(fs_info
,
5037 BTRFS_BLOCK_GROUP_METADATA
);
5041 void btrfs_free_block_rsv(struct btrfs_root
*root
,
5042 struct btrfs_block_rsv
*rsv
)
5046 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
5050 void __btrfs_free_block_rsv(struct btrfs_block_rsv
*rsv
)
5055 int btrfs_block_rsv_add(struct btrfs_root
*root
,
5056 struct btrfs_block_rsv
*block_rsv
, u64 num_bytes
,
5057 enum btrfs_reserve_flush_enum flush
)
5064 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
5066 block_rsv_add_bytes(block_rsv
, num_bytes
, 1);
5073 int btrfs_block_rsv_check(struct btrfs_root
*root
,
5074 struct btrfs_block_rsv
*block_rsv
, int min_factor
)
5082 spin_lock(&block_rsv
->lock
);
5083 num_bytes
= div_factor(block_rsv
->size
, min_factor
);
5084 if (block_rsv
->reserved
>= num_bytes
)
5086 spin_unlock(&block_rsv
->lock
);
5091 int btrfs_block_rsv_refill(struct btrfs_root
*root
,
5092 struct btrfs_block_rsv
*block_rsv
, u64 min_reserved
,
5093 enum btrfs_reserve_flush_enum flush
)
5101 spin_lock(&block_rsv
->lock
);
5102 num_bytes
= min_reserved
;
5103 if (block_rsv
->reserved
>= num_bytes
)
5106 num_bytes
-= block_rsv
->reserved
;
5107 spin_unlock(&block_rsv
->lock
);
5112 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
5114 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
5121 int btrfs_block_rsv_migrate(struct btrfs_block_rsv
*src_rsv
,
5122 struct btrfs_block_rsv
*dst_rsv
,
5125 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
5128 void btrfs_block_rsv_release(struct btrfs_root
*root
,
5129 struct btrfs_block_rsv
*block_rsv
,
5132 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
5133 if (global_rsv
== block_rsv
||
5134 block_rsv
->space_info
!= global_rsv
->space_info
)
5136 block_rsv_release_bytes(root
->fs_info
, block_rsv
, global_rsv
,
5141 * helper to calculate size of global block reservation.
5142 * the desired value is sum of space used by extent tree,
5143 * checksum tree and root tree
5145 static u64
calc_global_metadata_size(struct btrfs_fs_info
*fs_info
)
5147 struct btrfs_space_info
*sinfo
;
5151 int csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
5153 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_DATA
);
5154 spin_lock(&sinfo
->lock
);
5155 data_used
= sinfo
->bytes_used
;
5156 spin_unlock(&sinfo
->lock
);
5158 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
5159 spin_lock(&sinfo
->lock
);
5160 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_DATA
)
5162 meta_used
= sinfo
->bytes_used
;
5163 spin_unlock(&sinfo
->lock
);
5165 num_bytes
= (data_used
>> fs_info
->sb
->s_blocksize_bits
) *
5167 num_bytes
+= div_u64(data_used
+ meta_used
, 50);
5169 if (num_bytes
* 3 > meta_used
)
5170 num_bytes
= div_u64(meta_used
, 3);
5172 return ALIGN(num_bytes
, fs_info
->extent_root
->nodesize
<< 10);
5175 static void update_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5177 struct btrfs_block_rsv
*block_rsv
= &fs_info
->global_block_rsv
;
5178 struct btrfs_space_info
*sinfo
= block_rsv
->space_info
;
5181 num_bytes
= calc_global_metadata_size(fs_info
);
5183 spin_lock(&sinfo
->lock
);
5184 spin_lock(&block_rsv
->lock
);
5186 block_rsv
->size
= min_t(u64
, num_bytes
, 512 * 1024 * 1024);
5188 num_bytes
= sinfo
->bytes_used
+ sinfo
->bytes_pinned
+
5189 sinfo
->bytes_reserved
+ sinfo
->bytes_readonly
+
5190 sinfo
->bytes_may_use
;
5192 if (sinfo
->total_bytes
> num_bytes
) {
5193 num_bytes
= sinfo
->total_bytes
- num_bytes
;
5194 block_rsv
->reserved
+= num_bytes
;
5195 sinfo
->bytes_may_use
+= num_bytes
;
5196 trace_btrfs_space_reservation(fs_info
, "space_info",
5197 sinfo
->flags
, num_bytes
, 1);
5200 if (block_rsv
->reserved
>= block_rsv
->size
) {
5201 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
5202 sinfo
->bytes_may_use
-= num_bytes
;
5203 trace_btrfs_space_reservation(fs_info
, "space_info",
5204 sinfo
->flags
, num_bytes
, 0);
5205 block_rsv
->reserved
= block_rsv
->size
;
5206 block_rsv
->full
= 1;
5209 spin_unlock(&block_rsv
->lock
);
5210 spin_unlock(&sinfo
->lock
);
5213 static void init_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5215 struct btrfs_space_info
*space_info
;
5217 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
5218 fs_info
->chunk_block_rsv
.space_info
= space_info
;
5220 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
5221 fs_info
->global_block_rsv
.space_info
= space_info
;
5222 fs_info
->delalloc_block_rsv
.space_info
= space_info
;
5223 fs_info
->trans_block_rsv
.space_info
= space_info
;
5224 fs_info
->empty_block_rsv
.space_info
= space_info
;
5225 fs_info
->delayed_block_rsv
.space_info
= space_info
;
5227 fs_info
->extent_root
->block_rsv
= &fs_info
->global_block_rsv
;
5228 fs_info
->csum_root
->block_rsv
= &fs_info
->global_block_rsv
;
5229 fs_info
->dev_root
->block_rsv
= &fs_info
->global_block_rsv
;
5230 fs_info
->tree_root
->block_rsv
= &fs_info
->global_block_rsv
;
5231 if (fs_info
->quota_root
)
5232 fs_info
->quota_root
->block_rsv
= &fs_info
->global_block_rsv
;
5233 fs_info
->chunk_root
->block_rsv
= &fs_info
->chunk_block_rsv
;
5235 update_global_block_rsv(fs_info
);
5238 static void release_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5240 block_rsv_release_bytes(fs_info
, &fs_info
->global_block_rsv
, NULL
,
5242 WARN_ON(fs_info
->delalloc_block_rsv
.size
> 0);
5243 WARN_ON(fs_info
->delalloc_block_rsv
.reserved
> 0);
5244 WARN_ON(fs_info
->trans_block_rsv
.size
> 0);
5245 WARN_ON(fs_info
->trans_block_rsv
.reserved
> 0);
5246 WARN_ON(fs_info
->chunk_block_rsv
.size
> 0);
5247 WARN_ON(fs_info
->chunk_block_rsv
.reserved
> 0);
5248 WARN_ON(fs_info
->delayed_block_rsv
.size
> 0);
5249 WARN_ON(fs_info
->delayed_block_rsv
.reserved
> 0);
5252 void btrfs_trans_release_metadata(struct btrfs_trans_handle
*trans
,
5253 struct btrfs_root
*root
)
5255 if (!trans
->block_rsv
)
5258 if (!trans
->bytes_reserved
)
5261 trace_btrfs_space_reservation(root
->fs_info
, "transaction",
5262 trans
->transid
, trans
->bytes_reserved
, 0);
5263 btrfs_block_rsv_release(root
, trans
->block_rsv
, trans
->bytes_reserved
);
5264 trans
->bytes_reserved
= 0;
5268 * To be called after all the new block groups attached to the transaction
5269 * handle have been created (btrfs_create_pending_block_groups()).
5271 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle
*trans
)
5273 struct btrfs_fs_info
*fs_info
= trans
->root
->fs_info
;
5275 if (!trans
->chunk_bytes_reserved
)
5278 WARN_ON_ONCE(!list_empty(&trans
->new_bgs
));
5280 block_rsv_release_bytes(fs_info
, &fs_info
->chunk_block_rsv
, NULL
,
5281 trans
->chunk_bytes_reserved
);
5282 trans
->chunk_bytes_reserved
= 0;
5285 /* Can only return 0 or -ENOSPC */
5286 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle
*trans
,
5287 struct inode
*inode
)
5289 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5290 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
5291 struct btrfs_block_rsv
*dst_rsv
= root
->orphan_block_rsv
;
5294 * We need to hold space in order to delete our orphan item once we've
5295 * added it, so this takes the reservation so we can release it later
5296 * when we are truly done with the orphan item.
5298 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
5299 trace_btrfs_space_reservation(root
->fs_info
, "orphan",
5300 btrfs_ino(inode
), num_bytes
, 1);
5301 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
5304 void btrfs_orphan_release_metadata(struct inode
*inode
)
5306 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5307 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
5308 trace_btrfs_space_reservation(root
->fs_info
, "orphan",
5309 btrfs_ino(inode
), num_bytes
, 0);
5310 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
, num_bytes
);
5314 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5315 * root: the root of the parent directory
5316 * rsv: block reservation
5317 * items: the number of items that we need do reservation
5318 * qgroup_reserved: used to return the reserved size in qgroup
5320 * This function is used to reserve the space for snapshot/subvolume
5321 * creation and deletion. Those operations are different with the
5322 * common file/directory operations, they change two fs/file trees
5323 * and root tree, the number of items that the qgroup reserves is
5324 * different with the free space reservation. So we can not use
5325 * the space reseravtion mechanism in start_transaction().
5327 int btrfs_subvolume_reserve_metadata(struct btrfs_root
*root
,
5328 struct btrfs_block_rsv
*rsv
,
5330 u64
*qgroup_reserved
,
5331 bool use_global_rsv
)
5335 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
5337 if (root
->fs_info
->quota_enabled
) {
5338 /* One for parent inode, two for dir entries */
5339 num_bytes
= 3 * root
->nodesize
;
5340 ret
= btrfs_qgroup_reserve_meta(root
, num_bytes
);
5347 *qgroup_reserved
= num_bytes
;
5349 num_bytes
= btrfs_calc_trans_metadata_size(root
, items
);
5350 rsv
->space_info
= __find_space_info(root
->fs_info
,
5351 BTRFS_BLOCK_GROUP_METADATA
);
5352 ret
= btrfs_block_rsv_add(root
, rsv
, num_bytes
,
5353 BTRFS_RESERVE_FLUSH_ALL
);
5355 if (ret
== -ENOSPC
&& use_global_rsv
)
5356 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
, num_bytes
);
5358 if (ret
&& *qgroup_reserved
)
5359 btrfs_qgroup_free_meta(root
, *qgroup_reserved
);
5364 void btrfs_subvolume_release_metadata(struct btrfs_root
*root
,
5365 struct btrfs_block_rsv
*rsv
,
5366 u64 qgroup_reserved
)
5368 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
5372 * drop_outstanding_extent - drop an outstanding extent
5373 * @inode: the inode we're dropping the extent for
5374 * @num_bytes: the number of bytes we're relaseing.
5376 * This is called when we are freeing up an outstanding extent, either called
5377 * after an error or after an extent is written. This will return the number of
5378 * reserved extents that need to be freed. This must be called with
5379 * BTRFS_I(inode)->lock held.
5381 static unsigned drop_outstanding_extent(struct inode
*inode
, u64 num_bytes
)
5383 unsigned drop_inode_space
= 0;
5384 unsigned dropped_extents
= 0;
5385 unsigned num_extents
= 0;
5387 num_extents
= (unsigned)div64_u64(num_bytes
+
5388 BTRFS_MAX_EXTENT_SIZE
- 1,
5389 BTRFS_MAX_EXTENT_SIZE
);
5390 ASSERT(num_extents
);
5391 ASSERT(BTRFS_I(inode
)->outstanding_extents
>= num_extents
);
5392 BTRFS_I(inode
)->outstanding_extents
-= num_extents
;
5394 if (BTRFS_I(inode
)->outstanding_extents
== 0 &&
5395 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5396 &BTRFS_I(inode
)->runtime_flags
))
5397 drop_inode_space
= 1;
5400 * If we have more or the same amount of outsanding extents than we have
5401 * reserved then we need to leave the reserved extents count alone.
5403 if (BTRFS_I(inode
)->outstanding_extents
>=
5404 BTRFS_I(inode
)->reserved_extents
)
5405 return drop_inode_space
;
5407 dropped_extents
= BTRFS_I(inode
)->reserved_extents
-
5408 BTRFS_I(inode
)->outstanding_extents
;
5409 BTRFS_I(inode
)->reserved_extents
-= dropped_extents
;
5410 return dropped_extents
+ drop_inode_space
;
5414 * calc_csum_metadata_size - return the amount of metada space that must be
5415 * reserved/free'd for the given bytes.
5416 * @inode: the inode we're manipulating
5417 * @num_bytes: the number of bytes in question
5418 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5420 * This adjusts the number of csum_bytes in the inode and then returns the
5421 * correct amount of metadata that must either be reserved or freed. We
5422 * calculate how many checksums we can fit into one leaf and then divide the
5423 * number of bytes that will need to be checksumed by this value to figure out
5424 * how many checksums will be required. If we are adding bytes then the number
5425 * may go up and we will return the number of additional bytes that must be
5426 * reserved. If it is going down we will return the number of bytes that must
5429 * This must be called with BTRFS_I(inode)->lock held.
5431 static u64
calc_csum_metadata_size(struct inode
*inode
, u64 num_bytes
,
5434 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5435 u64 old_csums
, num_csums
;
5437 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
&&
5438 BTRFS_I(inode
)->csum_bytes
== 0)
5441 old_csums
= btrfs_csum_bytes_to_leaves(root
, BTRFS_I(inode
)->csum_bytes
);
5443 BTRFS_I(inode
)->csum_bytes
+= num_bytes
;
5445 BTRFS_I(inode
)->csum_bytes
-= num_bytes
;
5446 num_csums
= btrfs_csum_bytes_to_leaves(root
, BTRFS_I(inode
)->csum_bytes
);
5448 /* No change, no need to reserve more */
5449 if (old_csums
== num_csums
)
5453 return btrfs_calc_trans_metadata_size(root
,
5454 num_csums
- old_csums
);
5456 return btrfs_calc_trans_metadata_size(root
, old_csums
- num_csums
);
5459 int btrfs_delalloc_reserve_metadata(struct inode
*inode
, u64 num_bytes
)
5461 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5462 struct btrfs_block_rsv
*block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
5465 unsigned nr_extents
= 0;
5466 int extra_reserve
= 0;
5467 enum btrfs_reserve_flush_enum flush
= BTRFS_RESERVE_FLUSH_ALL
;
5469 bool delalloc_lock
= true;
5473 /* If we are a free space inode we need to not flush since we will be in
5474 * the middle of a transaction commit. We also don't need the delalloc
5475 * mutex since we won't race with anybody. We need this mostly to make
5476 * lockdep shut its filthy mouth.
5478 if (btrfs_is_free_space_inode(inode
)) {
5479 flush
= BTRFS_RESERVE_NO_FLUSH
;
5480 delalloc_lock
= false;
5483 if (flush
!= BTRFS_RESERVE_NO_FLUSH
&&
5484 btrfs_transaction_in_commit(root
->fs_info
))
5485 schedule_timeout(1);
5488 mutex_lock(&BTRFS_I(inode
)->delalloc_mutex
);
5490 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
5492 spin_lock(&BTRFS_I(inode
)->lock
);
5493 nr_extents
= (unsigned)div64_u64(num_bytes
+
5494 BTRFS_MAX_EXTENT_SIZE
- 1,
5495 BTRFS_MAX_EXTENT_SIZE
);
5496 BTRFS_I(inode
)->outstanding_extents
+= nr_extents
;
5499 if (BTRFS_I(inode
)->outstanding_extents
>
5500 BTRFS_I(inode
)->reserved_extents
)
5501 nr_extents
= BTRFS_I(inode
)->outstanding_extents
-
5502 BTRFS_I(inode
)->reserved_extents
;
5505 * Add an item to reserve for updating the inode when we complete the
5508 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5509 &BTRFS_I(inode
)->runtime_flags
)) {
5514 to_reserve
= btrfs_calc_trans_metadata_size(root
, nr_extents
);
5515 to_reserve
+= calc_csum_metadata_size(inode
, num_bytes
, 1);
5516 csum_bytes
= BTRFS_I(inode
)->csum_bytes
;
5517 spin_unlock(&BTRFS_I(inode
)->lock
);
5519 if (root
->fs_info
->quota_enabled
) {
5520 ret
= btrfs_qgroup_reserve_meta(root
,
5521 nr_extents
* root
->nodesize
);
5526 ret
= reserve_metadata_bytes(root
, block_rsv
, to_reserve
, flush
);
5527 if (unlikely(ret
)) {
5528 btrfs_qgroup_free_meta(root
, nr_extents
* root
->nodesize
);
5532 spin_lock(&BTRFS_I(inode
)->lock
);
5533 if (extra_reserve
) {
5534 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5535 &BTRFS_I(inode
)->runtime_flags
);
5538 BTRFS_I(inode
)->reserved_extents
+= nr_extents
;
5539 spin_unlock(&BTRFS_I(inode
)->lock
);
5542 mutex_unlock(&BTRFS_I(inode
)->delalloc_mutex
);
5545 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5546 btrfs_ino(inode
), to_reserve
, 1);
5547 block_rsv_add_bytes(block_rsv
, to_reserve
, 1);
5552 spin_lock(&BTRFS_I(inode
)->lock
);
5553 dropped
= drop_outstanding_extent(inode
, num_bytes
);
5555 * If the inodes csum_bytes is the same as the original
5556 * csum_bytes then we know we haven't raced with any free()ers
5557 * so we can just reduce our inodes csum bytes and carry on.
5559 if (BTRFS_I(inode
)->csum_bytes
== csum_bytes
) {
5560 calc_csum_metadata_size(inode
, num_bytes
, 0);
5562 u64 orig_csum_bytes
= BTRFS_I(inode
)->csum_bytes
;
5566 * This is tricky, but first we need to figure out how much we
5567 * free'd from any free-ers that occured during this
5568 * reservation, so we reset ->csum_bytes to the csum_bytes
5569 * before we dropped our lock, and then call the free for the
5570 * number of bytes that were freed while we were trying our
5573 bytes
= csum_bytes
- BTRFS_I(inode
)->csum_bytes
;
5574 BTRFS_I(inode
)->csum_bytes
= csum_bytes
;
5575 to_free
= calc_csum_metadata_size(inode
, bytes
, 0);
5579 * Now we need to see how much we would have freed had we not
5580 * been making this reservation and our ->csum_bytes were not
5581 * artificially inflated.
5583 BTRFS_I(inode
)->csum_bytes
= csum_bytes
- num_bytes
;
5584 bytes
= csum_bytes
- orig_csum_bytes
;
5585 bytes
= calc_csum_metadata_size(inode
, bytes
, 0);
5588 * Now reset ->csum_bytes to what it should be. If bytes is
5589 * more than to_free then we would have free'd more space had we
5590 * not had an artificially high ->csum_bytes, so we need to free
5591 * the remainder. If bytes is the same or less then we don't
5592 * need to do anything, the other free-ers did the correct
5595 BTRFS_I(inode
)->csum_bytes
= orig_csum_bytes
- num_bytes
;
5596 if (bytes
> to_free
)
5597 to_free
= bytes
- to_free
;
5601 spin_unlock(&BTRFS_I(inode
)->lock
);
5603 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
5606 btrfs_block_rsv_release(root
, block_rsv
, to_free
);
5607 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5608 btrfs_ino(inode
), to_free
, 0);
5611 mutex_unlock(&BTRFS_I(inode
)->delalloc_mutex
);
5616 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5617 * @inode: the inode to release the reservation for
5618 * @num_bytes: the number of bytes we're releasing
5620 * This will release the metadata reservation for an inode. This can be called
5621 * once we complete IO for a given set of bytes to release their metadata
5624 void btrfs_delalloc_release_metadata(struct inode
*inode
, u64 num_bytes
)
5626 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5630 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
5631 spin_lock(&BTRFS_I(inode
)->lock
);
5632 dropped
= drop_outstanding_extent(inode
, num_bytes
);
5635 to_free
= calc_csum_metadata_size(inode
, num_bytes
, 0);
5636 spin_unlock(&BTRFS_I(inode
)->lock
);
5638 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
5640 if (btrfs_test_is_dummy_root(root
))
5643 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5644 btrfs_ino(inode
), to_free
, 0);
5646 btrfs_block_rsv_release(root
, &root
->fs_info
->delalloc_block_rsv
,
5651 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5652 * @inode: inode we're writing to
5653 * @num_bytes: the number of bytes we want to allocate
5655 * This will do the following things
5657 * o reserve space in the data space info for num_bytes
5658 * o reserve space in the metadata space info based on number of outstanding
5659 * extents and how much csums will be needed
5660 * o add to the inodes ->delalloc_bytes
5661 * o add it to the fs_info's delalloc inodes list.
5663 * This will return 0 for success and -ENOSPC if there is no space left.
5665 int btrfs_delalloc_reserve_space(struct inode
*inode
, u64 num_bytes
)
5669 ret
= btrfs_check_data_free_space(inode
, num_bytes
, num_bytes
);
5673 ret
= btrfs_delalloc_reserve_metadata(inode
, num_bytes
);
5675 btrfs_free_reserved_data_space(inode
, num_bytes
);
5683 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5684 * @inode: inode we're releasing space for
5685 * @num_bytes: the number of bytes we want to free up
5687 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5688 * called in the case that we don't need the metadata AND data reservations
5689 * anymore. So if there is an error or we insert an inline extent.
5691 * This function will release the metadata space that was not used and will
5692 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5693 * list if there are no delalloc bytes left.
5695 void btrfs_delalloc_release_space(struct inode
*inode
, u64 num_bytes
)
5697 btrfs_delalloc_release_metadata(inode
, num_bytes
);
5698 btrfs_free_reserved_data_space(inode
, num_bytes
);
5701 static int update_block_group(struct btrfs_trans_handle
*trans
,
5702 struct btrfs_root
*root
, u64 bytenr
,
5703 u64 num_bytes
, int alloc
)
5705 struct btrfs_block_group_cache
*cache
= NULL
;
5706 struct btrfs_fs_info
*info
= root
->fs_info
;
5707 u64 total
= num_bytes
;
5712 /* block accounting for super block */
5713 spin_lock(&info
->delalloc_root_lock
);
5714 old_val
= btrfs_super_bytes_used(info
->super_copy
);
5716 old_val
+= num_bytes
;
5718 old_val
-= num_bytes
;
5719 btrfs_set_super_bytes_used(info
->super_copy
, old_val
);
5720 spin_unlock(&info
->delalloc_root_lock
);
5723 cache
= btrfs_lookup_block_group(info
, bytenr
);
5726 if (cache
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
5727 BTRFS_BLOCK_GROUP_RAID1
|
5728 BTRFS_BLOCK_GROUP_RAID10
))
5733 * If this block group has free space cache written out, we
5734 * need to make sure to load it if we are removing space. This
5735 * is because we need the unpinning stage to actually add the
5736 * space back to the block group, otherwise we will leak space.
5738 if (!alloc
&& cache
->cached
== BTRFS_CACHE_NO
)
5739 cache_block_group(cache
, 1);
5741 byte_in_group
= bytenr
- cache
->key
.objectid
;
5742 WARN_ON(byte_in_group
> cache
->key
.offset
);
5744 spin_lock(&cache
->space_info
->lock
);
5745 spin_lock(&cache
->lock
);
5747 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
5748 cache
->disk_cache_state
< BTRFS_DC_CLEAR
)
5749 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
5751 old_val
= btrfs_block_group_used(&cache
->item
);
5752 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
5754 old_val
+= num_bytes
;
5755 btrfs_set_block_group_used(&cache
->item
, old_val
);
5756 cache
->reserved
-= num_bytes
;
5757 cache
->space_info
->bytes_reserved
-= num_bytes
;
5758 cache
->space_info
->bytes_used
+= num_bytes
;
5759 cache
->space_info
->disk_used
+= num_bytes
* factor
;
5760 spin_unlock(&cache
->lock
);
5761 spin_unlock(&cache
->space_info
->lock
);
5763 old_val
-= num_bytes
;
5764 btrfs_set_block_group_used(&cache
->item
, old_val
);
5765 cache
->pinned
+= num_bytes
;
5766 cache
->space_info
->bytes_pinned
+= num_bytes
;
5767 cache
->space_info
->bytes_used
-= num_bytes
;
5768 cache
->space_info
->disk_used
-= num_bytes
* factor
;
5769 spin_unlock(&cache
->lock
);
5770 spin_unlock(&cache
->space_info
->lock
);
5772 set_extent_dirty(info
->pinned_extents
,
5773 bytenr
, bytenr
+ num_bytes
- 1,
5774 GFP_NOFS
| __GFP_NOFAIL
);
5776 * No longer have used bytes in this block group, queue
5780 spin_lock(&info
->unused_bgs_lock
);
5781 if (list_empty(&cache
->bg_list
)) {
5782 btrfs_get_block_group(cache
);
5783 list_add_tail(&cache
->bg_list
,
5786 spin_unlock(&info
->unused_bgs_lock
);
5790 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
5791 if (list_empty(&cache
->dirty_list
)) {
5792 list_add_tail(&cache
->dirty_list
,
5793 &trans
->transaction
->dirty_bgs
);
5794 trans
->transaction
->num_dirty_bgs
++;
5795 btrfs_get_block_group(cache
);
5797 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
5799 btrfs_put_block_group(cache
);
5801 bytenr
+= num_bytes
;
5806 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
5808 struct btrfs_block_group_cache
*cache
;
5811 spin_lock(&root
->fs_info
->block_group_cache_lock
);
5812 bytenr
= root
->fs_info
->first_logical_byte
;
5813 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
5815 if (bytenr
< (u64
)-1)
5818 cache
= btrfs_lookup_first_block_group(root
->fs_info
, search_start
);
5822 bytenr
= cache
->key
.objectid
;
5823 btrfs_put_block_group(cache
);
5828 static int pin_down_extent(struct btrfs_root
*root
,
5829 struct btrfs_block_group_cache
*cache
,
5830 u64 bytenr
, u64 num_bytes
, int reserved
)
5832 spin_lock(&cache
->space_info
->lock
);
5833 spin_lock(&cache
->lock
);
5834 cache
->pinned
+= num_bytes
;
5835 cache
->space_info
->bytes_pinned
+= num_bytes
;
5837 cache
->reserved
-= num_bytes
;
5838 cache
->space_info
->bytes_reserved
-= num_bytes
;
5840 spin_unlock(&cache
->lock
);
5841 spin_unlock(&cache
->space_info
->lock
);
5843 set_extent_dirty(root
->fs_info
->pinned_extents
, bytenr
,
5844 bytenr
+ num_bytes
- 1, GFP_NOFS
| __GFP_NOFAIL
);
5846 trace_btrfs_reserved_extent_free(root
, bytenr
, num_bytes
);
5851 * this function must be called within transaction
5853 int btrfs_pin_extent(struct btrfs_root
*root
,
5854 u64 bytenr
, u64 num_bytes
, int reserved
)
5856 struct btrfs_block_group_cache
*cache
;
5858 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
5859 BUG_ON(!cache
); /* Logic error */
5861 pin_down_extent(root
, cache
, bytenr
, num_bytes
, reserved
);
5863 btrfs_put_block_group(cache
);
5868 * this function must be called within transaction
5870 int btrfs_pin_extent_for_log_replay(struct btrfs_root
*root
,
5871 u64 bytenr
, u64 num_bytes
)
5873 struct btrfs_block_group_cache
*cache
;
5876 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
5881 * pull in the free space cache (if any) so that our pin
5882 * removes the free space from the cache. We have load_only set
5883 * to one because the slow code to read in the free extents does check
5884 * the pinned extents.
5886 cache_block_group(cache
, 1);
5888 pin_down_extent(root
, cache
, bytenr
, num_bytes
, 0);
5890 /* remove us from the free space cache (if we're there at all) */
5891 ret
= btrfs_remove_free_space(cache
, bytenr
, num_bytes
);
5892 btrfs_put_block_group(cache
);
5896 static int __exclude_logged_extent(struct btrfs_root
*root
, u64 start
, u64 num_bytes
)
5899 struct btrfs_block_group_cache
*block_group
;
5900 struct btrfs_caching_control
*caching_ctl
;
5902 block_group
= btrfs_lookup_block_group(root
->fs_info
, start
);
5906 cache_block_group(block_group
, 0);
5907 caching_ctl
= get_caching_control(block_group
);
5911 BUG_ON(!block_group_cache_done(block_group
));
5912 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
5914 mutex_lock(&caching_ctl
->mutex
);
5916 if (start
>= caching_ctl
->progress
) {
5917 ret
= add_excluded_extent(root
, start
, num_bytes
);
5918 } else if (start
+ num_bytes
<= caching_ctl
->progress
) {
5919 ret
= btrfs_remove_free_space(block_group
,
5922 num_bytes
= caching_ctl
->progress
- start
;
5923 ret
= btrfs_remove_free_space(block_group
,
5928 num_bytes
= (start
+ num_bytes
) -
5929 caching_ctl
->progress
;
5930 start
= caching_ctl
->progress
;
5931 ret
= add_excluded_extent(root
, start
, num_bytes
);
5934 mutex_unlock(&caching_ctl
->mutex
);
5935 put_caching_control(caching_ctl
);
5937 btrfs_put_block_group(block_group
);
5941 int btrfs_exclude_logged_extents(struct btrfs_root
*log
,
5942 struct extent_buffer
*eb
)
5944 struct btrfs_file_extent_item
*item
;
5945 struct btrfs_key key
;
5949 if (!btrfs_fs_incompat(log
->fs_info
, MIXED_GROUPS
))
5952 for (i
= 0; i
< btrfs_header_nritems(eb
); i
++) {
5953 btrfs_item_key_to_cpu(eb
, &key
, i
);
5954 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
5956 item
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
5957 found_type
= btrfs_file_extent_type(eb
, item
);
5958 if (found_type
== BTRFS_FILE_EXTENT_INLINE
)
5960 if (btrfs_file_extent_disk_bytenr(eb
, item
) == 0)
5962 key
.objectid
= btrfs_file_extent_disk_bytenr(eb
, item
);
5963 key
.offset
= btrfs_file_extent_disk_num_bytes(eb
, item
);
5964 __exclude_logged_extent(log
, key
.objectid
, key
.offset
);
5971 * btrfs_update_reserved_bytes - update the block_group and space info counters
5972 * @cache: The cache we are manipulating
5973 * @num_bytes: The number of bytes in question
5974 * @reserve: One of the reservation enums
5975 * @delalloc: The blocks are allocated for the delalloc write
5977 * This is called by the allocator when it reserves space, or by somebody who is
5978 * freeing space that was never actually used on disk. For example if you
5979 * reserve some space for a new leaf in transaction A and before transaction A
5980 * commits you free that leaf, you call this with reserve set to 0 in order to
5981 * clear the reservation.
5983 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5984 * ENOSPC accounting. For data we handle the reservation through clearing the
5985 * delalloc bits in the io_tree. We have to do this since we could end up
5986 * allocating less disk space for the amount of data we have reserved in the
5987 * case of compression.
5989 * If this is a reservation and the block group has become read only we cannot
5990 * make the reservation and return -EAGAIN, otherwise this function always
5993 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
5994 u64 num_bytes
, int reserve
, int delalloc
)
5996 struct btrfs_space_info
*space_info
= cache
->space_info
;
5999 spin_lock(&space_info
->lock
);
6000 spin_lock(&cache
->lock
);
6001 if (reserve
!= RESERVE_FREE
) {
6005 cache
->reserved
+= num_bytes
;
6006 space_info
->bytes_reserved
+= num_bytes
;
6007 if (reserve
== RESERVE_ALLOC
) {
6008 trace_btrfs_space_reservation(cache
->fs_info
,
6009 "space_info", space_info
->flags
,
6011 space_info
->bytes_may_use
-= num_bytes
;
6015 cache
->delalloc_bytes
+= num_bytes
;
6019 space_info
->bytes_readonly
+= num_bytes
;
6020 cache
->reserved
-= num_bytes
;
6021 space_info
->bytes_reserved
-= num_bytes
;
6024 cache
->delalloc_bytes
-= num_bytes
;
6026 spin_unlock(&cache
->lock
);
6027 spin_unlock(&space_info
->lock
);
6031 void btrfs_prepare_extent_commit(struct btrfs_trans_handle
*trans
,
6032 struct btrfs_root
*root
)
6034 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6035 struct btrfs_caching_control
*next
;
6036 struct btrfs_caching_control
*caching_ctl
;
6037 struct btrfs_block_group_cache
*cache
;
6039 down_write(&fs_info
->commit_root_sem
);
6041 list_for_each_entry_safe(caching_ctl
, next
,
6042 &fs_info
->caching_block_groups
, list
) {
6043 cache
= caching_ctl
->block_group
;
6044 if (block_group_cache_done(cache
)) {
6045 cache
->last_byte_to_unpin
= (u64
)-1;
6046 list_del_init(&caching_ctl
->list
);
6047 put_caching_control(caching_ctl
);
6049 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
6053 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
6054 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
6056 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
6058 up_write(&fs_info
->commit_root_sem
);
6060 update_global_block_rsv(fs_info
);
6063 static int unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
,
6064 const bool return_free_space
)
6066 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6067 struct btrfs_block_group_cache
*cache
= NULL
;
6068 struct btrfs_space_info
*space_info
;
6069 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
6073 while (start
<= end
) {
6076 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
6078 btrfs_put_block_group(cache
);
6079 cache
= btrfs_lookup_block_group(fs_info
, start
);
6080 BUG_ON(!cache
); /* Logic error */
6083 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
6084 len
= min(len
, end
+ 1 - start
);
6086 if (start
< cache
->last_byte_to_unpin
) {
6087 len
= min(len
, cache
->last_byte_to_unpin
- start
);
6088 if (return_free_space
)
6089 btrfs_add_free_space(cache
, start
, len
);
6093 space_info
= cache
->space_info
;
6095 spin_lock(&space_info
->lock
);
6096 spin_lock(&cache
->lock
);
6097 cache
->pinned
-= len
;
6098 space_info
->bytes_pinned
-= len
;
6099 percpu_counter_add(&space_info
->total_bytes_pinned
, -len
);
6101 space_info
->bytes_readonly
+= len
;
6104 spin_unlock(&cache
->lock
);
6105 if (!readonly
&& global_rsv
->space_info
== space_info
) {
6106 spin_lock(&global_rsv
->lock
);
6107 if (!global_rsv
->full
) {
6108 len
= min(len
, global_rsv
->size
-
6109 global_rsv
->reserved
);
6110 global_rsv
->reserved
+= len
;
6111 space_info
->bytes_may_use
+= len
;
6112 if (global_rsv
->reserved
>= global_rsv
->size
)
6113 global_rsv
->full
= 1;
6115 spin_unlock(&global_rsv
->lock
);
6117 spin_unlock(&space_info
->lock
);
6121 btrfs_put_block_group(cache
);
6125 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
6126 struct btrfs_root
*root
)
6128 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6129 struct btrfs_block_group_cache
*block_group
, *tmp
;
6130 struct list_head
*deleted_bgs
;
6131 struct extent_io_tree
*unpin
;
6136 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
6137 unpin
= &fs_info
->freed_extents
[1];
6139 unpin
= &fs_info
->freed_extents
[0];
6141 while (!trans
->aborted
) {
6142 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
6143 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
6144 EXTENT_DIRTY
, NULL
);
6146 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
6150 if (btrfs_test_opt(root
, DISCARD
))
6151 ret
= btrfs_discard_extent(root
, start
,
6152 end
+ 1 - start
, NULL
);
6154 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
6155 unpin_extent_range(root
, start
, end
, true);
6156 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
6161 * Transaction is finished. We don't need the lock anymore. We
6162 * do need to clean up the block groups in case of a transaction
6165 deleted_bgs
= &trans
->transaction
->deleted_bgs
;
6166 list_for_each_entry_safe(block_group
, tmp
, deleted_bgs
, bg_list
) {
6170 if (!trans
->aborted
)
6171 ret
= btrfs_discard_extent(root
,
6172 block_group
->key
.objectid
,
6173 block_group
->key
.offset
,
6176 list_del_init(&block_group
->bg_list
);
6177 btrfs_put_block_group_trimming(block_group
);
6178 btrfs_put_block_group(block_group
);
6181 const char *errstr
= btrfs_decode_error(ret
);
6183 "Discard failed while removing blockgroup: errno=%d %s\n",
6191 static void add_pinned_bytes(struct btrfs_fs_info
*fs_info
, u64 num_bytes
,
6192 u64 owner
, u64 root_objectid
)
6194 struct btrfs_space_info
*space_info
;
6197 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
6198 if (root_objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
6199 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
6201 flags
= BTRFS_BLOCK_GROUP_METADATA
;
6203 flags
= BTRFS_BLOCK_GROUP_DATA
;
6206 space_info
= __find_space_info(fs_info
, flags
);
6207 BUG_ON(!space_info
); /* Logic bug */
6208 percpu_counter_add(&space_info
->total_bytes_pinned
, num_bytes
);
6212 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
6213 struct btrfs_root
*root
,
6214 struct btrfs_delayed_ref_node
*node
, u64 parent
,
6215 u64 root_objectid
, u64 owner_objectid
,
6216 u64 owner_offset
, int refs_to_drop
,
6217 struct btrfs_delayed_extent_op
*extent_op
)
6219 struct btrfs_key key
;
6220 struct btrfs_path
*path
;
6221 struct btrfs_fs_info
*info
= root
->fs_info
;
6222 struct btrfs_root
*extent_root
= info
->extent_root
;
6223 struct extent_buffer
*leaf
;
6224 struct btrfs_extent_item
*ei
;
6225 struct btrfs_extent_inline_ref
*iref
;
6228 int extent_slot
= 0;
6229 int found_extent
= 0;
6231 int no_quota
= node
->no_quota
;
6234 u64 bytenr
= node
->bytenr
;
6235 u64 num_bytes
= node
->num_bytes
;
6237 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
6240 if (!info
->quota_enabled
|| !is_fstree(root_objectid
))
6243 path
= btrfs_alloc_path();
6248 path
->leave_spinning
= 1;
6250 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
6251 BUG_ON(!is_data
&& refs_to_drop
!= 1);
6254 skinny_metadata
= 0;
6256 ret
= lookup_extent_backref(trans
, extent_root
, path
, &iref
,
6257 bytenr
, num_bytes
, parent
,
6258 root_objectid
, owner_objectid
,
6261 extent_slot
= path
->slots
[0];
6262 while (extent_slot
>= 0) {
6263 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
6265 if (key
.objectid
!= bytenr
)
6267 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
6268 key
.offset
== num_bytes
) {
6272 if (key
.type
== BTRFS_METADATA_ITEM_KEY
&&
6273 key
.offset
== owner_objectid
) {
6277 if (path
->slots
[0] - extent_slot
> 5)
6281 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6282 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
6283 if (found_extent
&& item_size
< sizeof(*ei
))
6286 if (!found_extent
) {
6288 ret
= remove_extent_backref(trans
, extent_root
, path
,
6290 is_data
, &last_ref
);
6292 btrfs_abort_transaction(trans
, extent_root
, ret
);
6295 btrfs_release_path(path
);
6296 path
->leave_spinning
= 1;
6298 key
.objectid
= bytenr
;
6299 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6300 key
.offset
= num_bytes
;
6302 if (!is_data
&& skinny_metadata
) {
6303 key
.type
= BTRFS_METADATA_ITEM_KEY
;
6304 key
.offset
= owner_objectid
;
6307 ret
= btrfs_search_slot(trans
, extent_root
,
6309 if (ret
> 0 && skinny_metadata
&& path
->slots
[0]) {
6311 * Couldn't find our skinny metadata item,
6312 * see if we have ye olde extent item.
6315 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
6317 if (key
.objectid
== bytenr
&&
6318 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
6319 key
.offset
== num_bytes
)
6323 if (ret
> 0 && skinny_metadata
) {
6324 skinny_metadata
= false;
6325 key
.objectid
= bytenr
;
6326 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6327 key
.offset
= num_bytes
;
6328 btrfs_release_path(path
);
6329 ret
= btrfs_search_slot(trans
, extent_root
,
6334 btrfs_err(info
, "umm, got %d back from search, was looking for %llu",
6337 btrfs_print_leaf(extent_root
,
6341 btrfs_abort_transaction(trans
, extent_root
, ret
);
6344 extent_slot
= path
->slots
[0];
6346 } else if (WARN_ON(ret
== -ENOENT
)) {
6347 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
6349 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6350 bytenr
, parent
, root_objectid
, owner_objectid
,
6352 btrfs_abort_transaction(trans
, extent_root
, ret
);
6355 btrfs_abort_transaction(trans
, extent_root
, ret
);
6359 leaf
= path
->nodes
[0];
6360 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
6361 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6362 if (item_size
< sizeof(*ei
)) {
6363 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
6364 ret
= convert_extent_item_v0(trans
, extent_root
, path
,
6367 btrfs_abort_transaction(trans
, extent_root
, ret
);
6371 btrfs_release_path(path
);
6372 path
->leave_spinning
= 1;
6374 key
.objectid
= bytenr
;
6375 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6376 key
.offset
= num_bytes
;
6378 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
6381 btrfs_err(info
, "umm, got %d back from search, was looking for %llu",
6383 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
6386 btrfs_abort_transaction(trans
, extent_root
, ret
);
6390 extent_slot
= path
->slots
[0];
6391 leaf
= path
->nodes
[0];
6392 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
6395 BUG_ON(item_size
< sizeof(*ei
));
6396 ei
= btrfs_item_ptr(leaf
, extent_slot
,
6397 struct btrfs_extent_item
);
6398 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
6399 key
.type
== BTRFS_EXTENT_ITEM_KEY
) {
6400 struct btrfs_tree_block_info
*bi
;
6401 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
6402 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
6403 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
6406 refs
= btrfs_extent_refs(leaf
, ei
);
6407 if (refs
< refs_to_drop
) {
6408 btrfs_err(info
, "trying to drop %d refs but we only have %Lu "
6409 "for bytenr %Lu", refs_to_drop
, refs
, bytenr
);
6411 btrfs_abort_transaction(trans
, extent_root
, ret
);
6414 refs
-= refs_to_drop
;
6418 __run_delayed_extent_op(extent_op
, leaf
, ei
);
6420 * In the case of inline back ref, reference count will
6421 * be updated by remove_extent_backref
6424 BUG_ON(!found_extent
);
6426 btrfs_set_extent_refs(leaf
, ei
, refs
);
6427 btrfs_mark_buffer_dirty(leaf
);
6430 ret
= remove_extent_backref(trans
, extent_root
, path
,
6432 is_data
, &last_ref
);
6434 btrfs_abort_transaction(trans
, extent_root
, ret
);
6438 add_pinned_bytes(root
->fs_info
, -num_bytes
, owner_objectid
,
6442 BUG_ON(is_data
&& refs_to_drop
!=
6443 extent_data_ref_count(path
, iref
));
6445 BUG_ON(path
->slots
[0] != extent_slot
);
6447 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
6448 path
->slots
[0] = extent_slot
;
6454 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
6457 btrfs_abort_transaction(trans
, extent_root
, ret
);
6460 btrfs_release_path(path
);
6463 ret
= btrfs_del_csums(trans
, root
, bytenr
, num_bytes
);
6465 btrfs_abort_transaction(trans
, extent_root
, ret
);
6470 ret
= update_block_group(trans
, root
, bytenr
, num_bytes
, 0);
6472 btrfs_abort_transaction(trans
, extent_root
, ret
);
6476 btrfs_release_path(path
);
6479 btrfs_free_path(path
);
6484 * when we free an block, it is possible (and likely) that we free the last
6485 * delayed ref for that extent as well. This searches the delayed ref tree for
6486 * a given extent, and if there are no other delayed refs to be processed, it
6487 * removes it from the tree.
6489 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
6490 struct btrfs_root
*root
, u64 bytenr
)
6492 struct btrfs_delayed_ref_head
*head
;
6493 struct btrfs_delayed_ref_root
*delayed_refs
;
6496 delayed_refs
= &trans
->transaction
->delayed_refs
;
6497 spin_lock(&delayed_refs
->lock
);
6498 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
6500 goto out_delayed_unlock
;
6502 spin_lock(&head
->lock
);
6503 if (!list_empty(&head
->ref_list
))
6506 if (head
->extent_op
) {
6507 if (!head
->must_insert_reserved
)
6509 btrfs_free_delayed_extent_op(head
->extent_op
);
6510 head
->extent_op
= NULL
;
6514 * waiting for the lock here would deadlock. If someone else has it
6515 * locked they are already in the process of dropping it anyway
6517 if (!mutex_trylock(&head
->mutex
))
6521 * at this point we have a head with no other entries. Go
6522 * ahead and process it.
6524 head
->node
.in_tree
= 0;
6525 rb_erase(&head
->href_node
, &delayed_refs
->href_root
);
6527 atomic_dec(&delayed_refs
->num_entries
);
6530 * we don't take a ref on the node because we're removing it from the
6531 * tree, so we just steal the ref the tree was holding.
6533 delayed_refs
->num_heads
--;
6534 if (head
->processing
== 0)
6535 delayed_refs
->num_heads_ready
--;
6536 head
->processing
= 0;
6537 spin_unlock(&head
->lock
);
6538 spin_unlock(&delayed_refs
->lock
);
6540 BUG_ON(head
->extent_op
);
6541 if (head
->must_insert_reserved
)
6544 mutex_unlock(&head
->mutex
);
6545 btrfs_put_delayed_ref(&head
->node
);
6548 spin_unlock(&head
->lock
);
6551 spin_unlock(&delayed_refs
->lock
);
6555 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
6556 struct btrfs_root
*root
,
6557 struct extent_buffer
*buf
,
6558 u64 parent
, int last_ref
)
6563 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
6564 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
6565 buf
->start
, buf
->len
,
6566 parent
, root
->root_key
.objectid
,
6567 btrfs_header_level(buf
),
6568 BTRFS_DROP_DELAYED_REF
, NULL
, 0);
6569 BUG_ON(ret
); /* -ENOMEM */
6575 if (btrfs_header_generation(buf
) == trans
->transid
) {
6576 struct btrfs_block_group_cache
*cache
;
6578 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
6579 ret
= check_ref_cleanup(trans
, root
, buf
->start
);
6584 cache
= btrfs_lookup_block_group(root
->fs_info
, buf
->start
);
6586 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
6587 pin_down_extent(root
, cache
, buf
->start
, buf
->len
, 1);
6588 btrfs_put_block_group(cache
);
6592 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
6594 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
6595 btrfs_update_reserved_bytes(cache
, buf
->len
, RESERVE_FREE
, 0);
6596 btrfs_put_block_group(cache
);
6597 trace_btrfs_reserved_extent_free(root
, buf
->start
, buf
->len
);
6602 add_pinned_bytes(root
->fs_info
, buf
->len
,
6603 btrfs_header_level(buf
),
6604 root
->root_key
.objectid
);
6607 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6610 clear_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
);
6613 /* Can return -ENOMEM */
6614 int btrfs_free_extent(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
6615 u64 bytenr
, u64 num_bytes
, u64 parent
, u64 root_objectid
,
6616 u64 owner
, u64 offset
, int no_quota
)
6619 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6621 if (btrfs_test_is_dummy_root(root
))
6624 add_pinned_bytes(root
->fs_info
, num_bytes
, owner
, root_objectid
);
6627 * tree log blocks never actually go into the extent allocation
6628 * tree, just update pinning info and exit early.
6630 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
6631 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
6632 /* unlocks the pinned mutex */
6633 btrfs_pin_extent(root
, bytenr
, num_bytes
, 1);
6635 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
6636 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
6638 parent
, root_objectid
, (int)owner
,
6639 BTRFS_DROP_DELAYED_REF
, NULL
, no_quota
);
6641 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
6643 parent
, root_objectid
, owner
,
6644 offset
, BTRFS_DROP_DELAYED_REF
,
6651 * when we wait for progress in the block group caching, its because
6652 * our allocation attempt failed at least once. So, we must sleep
6653 * and let some progress happen before we try again.
6655 * This function will sleep at least once waiting for new free space to
6656 * show up, and then it will check the block group free space numbers
6657 * for our min num_bytes. Another option is to have it go ahead
6658 * and look in the rbtree for a free extent of a given size, but this
6661 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6662 * any of the information in this block group.
6664 static noinline
void
6665 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
6668 struct btrfs_caching_control
*caching_ctl
;
6670 caching_ctl
= get_caching_control(cache
);
6674 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
6675 (cache
->free_space_ctl
->free_space
>= num_bytes
));
6677 put_caching_control(caching_ctl
);
6681 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
6683 struct btrfs_caching_control
*caching_ctl
;
6686 caching_ctl
= get_caching_control(cache
);
6688 return (cache
->cached
== BTRFS_CACHE_ERROR
) ? -EIO
: 0;
6690 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
6691 if (cache
->cached
== BTRFS_CACHE_ERROR
)
6693 put_caching_control(caching_ctl
);
6697 int __get_raid_index(u64 flags
)
6699 if (flags
& BTRFS_BLOCK_GROUP_RAID10
)
6700 return BTRFS_RAID_RAID10
;
6701 else if (flags
& BTRFS_BLOCK_GROUP_RAID1
)
6702 return BTRFS_RAID_RAID1
;
6703 else if (flags
& BTRFS_BLOCK_GROUP_DUP
)
6704 return BTRFS_RAID_DUP
;
6705 else if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
6706 return BTRFS_RAID_RAID0
;
6707 else if (flags
& BTRFS_BLOCK_GROUP_RAID5
)
6708 return BTRFS_RAID_RAID5
;
6709 else if (flags
& BTRFS_BLOCK_GROUP_RAID6
)
6710 return BTRFS_RAID_RAID6
;
6712 return BTRFS_RAID_SINGLE
; /* BTRFS_BLOCK_GROUP_SINGLE */
6715 int get_block_group_index(struct btrfs_block_group_cache
*cache
)
6717 return __get_raid_index(cache
->flags
);
6720 static const char *btrfs_raid_type_names
[BTRFS_NR_RAID_TYPES
] = {
6721 [BTRFS_RAID_RAID10
] = "raid10",
6722 [BTRFS_RAID_RAID1
] = "raid1",
6723 [BTRFS_RAID_DUP
] = "dup",
6724 [BTRFS_RAID_RAID0
] = "raid0",
6725 [BTRFS_RAID_SINGLE
] = "single",
6726 [BTRFS_RAID_RAID5
] = "raid5",
6727 [BTRFS_RAID_RAID6
] = "raid6",
6730 static const char *get_raid_name(enum btrfs_raid_types type
)
6732 if (type
>= BTRFS_NR_RAID_TYPES
)
6735 return btrfs_raid_type_names
[type
];
6738 enum btrfs_loop_type
{
6739 LOOP_CACHING_NOWAIT
= 0,
6740 LOOP_CACHING_WAIT
= 1,
6741 LOOP_ALLOC_CHUNK
= 2,
6742 LOOP_NO_EMPTY_SIZE
= 3,
6746 btrfs_lock_block_group(struct btrfs_block_group_cache
*cache
,
6750 down_read(&cache
->data_rwsem
);
6754 btrfs_grab_block_group(struct btrfs_block_group_cache
*cache
,
6757 btrfs_get_block_group(cache
);
6759 down_read(&cache
->data_rwsem
);
6762 static struct btrfs_block_group_cache
*
6763 btrfs_lock_cluster(struct btrfs_block_group_cache
*block_group
,
6764 struct btrfs_free_cluster
*cluster
,
6767 struct btrfs_block_group_cache
*used_bg
;
6768 bool locked
= false;
6770 spin_lock(&cluster
->refill_lock
);
6772 if (used_bg
== cluster
->block_group
)
6775 up_read(&used_bg
->data_rwsem
);
6776 btrfs_put_block_group(used_bg
);
6779 used_bg
= cluster
->block_group
;
6783 if (used_bg
== block_group
)
6786 btrfs_get_block_group(used_bg
);
6791 if (down_read_trylock(&used_bg
->data_rwsem
))
6794 spin_unlock(&cluster
->refill_lock
);
6795 down_read(&used_bg
->data_rwsem
);
6801 btrfs_release_block_group(struct btrfs_block_group_cache
*cache
,
6805 up_read(&cache
->data_rwsem
);
6806 btrfs_put_block_group(cache
);
6810 * walks the btree of allocated extents and find a hole of a given size.
6811 * The key ins is changed to record the hole:
6812 * ins->objectid == start position
6813 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6814 * ins->offset == the size of the hole.
6815 * Any available blocks before search_start are skipped.
6817 * If there is no suitable free space, we will record the max size of
6818 * the free space extent currently.
6820 static noinline
int find_free_extent(struct btrfs_root
*orig_root
,
6821 u64 num_bytes
, u64 empty_size
,
6822 u64 hint_byte
, struct btrfs_key
*ins
,
6823 u64 flags
, int delalloc
)
6826 struct btrfs_root
*root
= orig_root
->fs_info
->extent_root
;
6827 struct btrfs_free_cluster
*last_ptr
= NULL
;
6828 struct btrfs_block_group_cache
*block_group
= NULL
;
6829 u64 search_start
= 0;
6830 u64 max_extent_size
= 0;
6831 int empty_cluster
= 2 * 1024 * 1024;
6832 struct btrfs_space_info
*space_info
;
6834 int index
= __get_raid_index(flags
);
6835 int alloc_type
= (flags
& BTRFS_BLOCK_GROUP_DATA
) ?
6836 RESERVE_ALLOC_NO_ACCOUNT
: RESERVE_ALLOC
;
6837 bool failed_cluster_refill
= false;
6838 bool failed_alloc
= false;
6839 bool use_cluster
= true;
6840 bool have_caching_bg
= false;
6842 WARN_ON(num_bytes
< root
->sectorsize
);
6843 ins
->type
= BTRFS_EXTENT_ITEM_KEY
;
6847 trace_find_free_extent(orig_root
, num_bytes
, empty_size
, flags
);
6849 space_info
= __find_space_info(root
->fs_info
, flags
);
6851 btrfs_err(root
->fs_info
, "No space info for %llu", flags
);
6856 * If the space info is for both data and metadata it means we have a
6857 * small filesystem and we can't use the clustering stuff.
6859 if (btrfs_mixed_space_info(space_info
))
6860 use_cluster
= false;
6862 if (flags
& BTRFS_BLOCK_GROUP_METADATA
&& use_cluster
) {
6863 last_ptr
= &root
->fs_info
->meta_alloc_cluster
;
6864 if (!btrfs_test_opt(root
, SSD
))
6865 empty_cluster
= 64 * 1024;
6868 if ((flags
& BTRFS_BLOCK_GROUP_DATA
) && use_cluster
&&
6869 btrfs_test_opt(root
, SSD
)) {
6870 last_ptr
= &root
->fs_info
->data_alloc_cluster
;
6874 spin_lock(&last_ptr
->lock
);
6875 if (last_ptr
->block_group
)
6876 hint_byte
= last_ptr
->window_start
;
6877 spin_unlock(&last_ptr
->lock
);
6880 search_start
= max(search_start
, first_logical_byte(root
, 0));
6881 search_start
= max(search_start
, hint_byte
);
6886 if (search_start
== hint_byte
) {
6887 block_group
= btrfs_lookup_block_group(root
->fs_info
,
6890 * we don't want to use the block group if it doesn't match our
6891 * allocation bits, or if its not cached.
6893 * However if we are re-searching with an ideal block group
6894 * picked out then we don't care that the block group is cached.
6896 if (block_group
&& block_group_bits(block_group
, flags
) &&
6897 block_group
->cached
!= BTRFS_CACHE_NO
) {
6898 down_read(&space_info
->groups_sem
);
6899 if (list_empty(&block_group
->list
) ||
6902 * someone is removing this block group,
6903 * we can't jump into the have_block_group
6904 * target because our list pointers are not
6907 btrfs_put_block_group(block_group
);
6908 up_read(&space_info
->groups_sem
);
6910 index
= get_block_group_index(block_group
);
6911 btrfs_lock_block_group(block_group
, delalloc
);
6912 goto have_block_group
;
6914 } else if (block_group
) {
6915 btrfs_put_block_group(block_group
);
6919 have_caching_bg
= false;
6920 down_read(&space_info
->groups_sem
);
6921 list_for_each_entry(block_group
, &space_info
->block_groups
[index
],
6926 btrfs_grab_block_group(block_group
, delalloc
);
6927 search_start
= block_group
->key
.objectid
;
6930 * this can happen if we end up cycling through all the
6931 * raid types, but we want to make sure we only allocate
6932 * for the proper type.
6934 if (!block_group_bits(block_group
, flags
)) {
6935 u64 extra
= BTRFS_BLOCK_GROUP_DUP
|
6936 BTRFS_BLOCK_GROUP_RAID1
|
6937 BTRFS_BLOCK_GROUP_RAID5
|
6938 BTRFS_BLOCK_GROUP_RAID6
|
6939 BTRFS_BLOCK_GROUP_RAID10
;
6942 * if they asked for extra copies and this block group
6943 * doesn't provide them, bail. This does allow us to
6944 * fill raid0 from raid1.
6946 if ((flags
& extra
) && !(block_group
->flags
& extra
))
6951 cached
= block_group_cache_done(block_group
);
6952 if (unlikely(!cached
)) {
6953 ret
= cache_block_group(block_group
, 0);
6958 if (unlikely(block_group
->cached
== BTRFS_CACHE_ERROR
))
6960 if (unlikely(block_group
->ro
))
6964 * Ok we want to try and use the cluster allocator, so
6968 struct btrfs_block_group_cache
*used_block_group
;
6969 unsigned long aligned_cluster
;
6971 * the refill lock keeps out other
6972 * people trying to start a new cluster
6974 used_block_group
= btrfs_lock_cluster(block_group
,
6977 if (!used_block_group
)
6978 goto refill_cluster
;
6980 if (used_block_group
!= block_group
&&
6981 (used_block_group
->ro
||
6982 !block_group_bits(used_block_group
, flags
)))
6983 goto release_cluster
;
6985 offset
= btrfs_alloc_from_cluster(used_block_group
,
6988 used_block_group
->key
.objectid
,
6991 /* we have a block, we're done */
6992 spin_unlock(&last_ptr
->refill_lock
);
6993 trace_btrfs_reserve_extent_cluster(root
,
6995 search_start
, num_bytes
);
6996 if (used_block_group
!= block_group
) {
6997 btrfs_release_block_group(block_group
,
6999 block_group
= used_block_group
;
7004 WARN_ON(last_ptr
->block_group
!= used_block_group
);
7006 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7007 * set up a new clusters, so lets just skip it
7008 * and let the allocator find whatever block
7009 * it can find. If we reach this point, we
7010 * will have tried the cluster allocator
7011 * plenty of times and not have found
7012 * anything, so we are likely way too
7013 * fragmented for the clustering stuff to find
7016 * However, if the cluster is taken from the
7017 * current block group, release the cluster
7018 * first, so that we stand a better chance of
7019 * succeeding in the unclustered
7021 if (loop
>= LOOP_NO_EMPTY_SIZE
&&
7022 used_block_group
!= block_group
) {
7023 spin_unlock(&last_ptr
->refill_lock
);
7024 btrfs_release_block_group(used_block_group
,
7026 goto unclustered_alloc
;
7030 * this cluster didn't work out, free it and
7033 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
7035 if (used_block_group
!= block_group
)
7036 btrfs_release_block_group(used_block_group
,
7039 if (loop
>= LOOP_NO_EMPTY_SIZE
) {
7040 spin_unlock(&last_ptr
->refill_lock
);
7041 goto unclustered_alloc
;
7044 aligned_cluster
= max_t(unsigned long,
7045 empty_cluster
+ empty_size
,
7046 block_group
->full_stripe_len
);
7048 /* allocate a cluster in this block group */
7049 ret
= btrfs_find_space_cluster(root
, block_group
,
7050 last_ptr
, search_start
,
7055 * now pull our allocation out of this
7058 offset
= btrfs_alloc_from_cluster(block_group
,
7064 /* we found one, proceed */
7065 spin_unlock(&last_ptr
->refill_lock
);
7066 trace_btrfs_reserve_extent_cluster(root
,
7067 block_group
, search_start
,
7071 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
7072 && !failed_cluster_refill
) {
7073 spin_unlock(&last_ptr
->refill_lock
);
7075 failed_cluster_refill
= true;
7076 wait_block_group_cache_progress(block_group
,
7077 num_bytes
+ empty_cluster
+ empty_size
);
7078 goto have_block_group
;
7082 * at this point we either didn't find a cluster
7083 * or we weren't able to allocate a block from our
7084 * cluster. Free the cluster we've been trying
7085 * to use, and go to the next block group
7087 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
7088 spin_unlock(&last_ptr
->refill_lock
);
7093 spin_lock(&block_group
->free_space_ctl
->tree_lock
);
7095 block_group
->free_space_ctl
->free_space
<
7096 num_bytes
+ empty_cluster
+ empty_size
) {
7097 if (block_group
->free_space_ctl
->free_space
>
7100 block_group
->free_space_ctl
->free_space
;
7101 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
7104 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
7106 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
7107 num_bytes
, empty_size
,
7110 * If we didn't find a chunk, and we haven't failed on this
7111 * block group before, and this block group is in the middle of
7112 * caching and we are ok with waiting, then go ahead and wait
7113 * for progress to be made, and set failed_alloc to true.
7115 * If failed_alloc is true then we've already waited on this
7116 * block group once and should move on to the next block group.
7118 if (!offset
&& !failed_alloc
&& !cached
&&
7119 loop
> LOOP_CACHING_NOWAIT
) {
7120 wait_block_group_cache_progress(block_group
,
7121 num_bytes
+ empty_size
);
7122 failed_alloc
= true;
7123 goto have_block_group
;
7124 } else if (!offset
) {
7126 have_caching_bg
= true;
7130 search_start
= ALIGN(offset
, root
->stripesize
);
7132 /* move on to the next group */
7133 if (search_start
+ num_bytes
>
7134 block_group
->key
.objectid
+ block_group
->key
.offset
) {
7135 btrfs_add_free_space(block_group
, offset
, num_bytes
);
7139 if (offset
< search_start
)
7140 btrfs_add_free_space(block_group
, offset
,
7141 search_start
- offset
);
7142 BUG_ON(offset
> search_start
);
7144 ret
= btrfs_update_reserved_bytes(block_group
, num_bytes
,
7145 alloc_type
, delalloc
);
7146 if (ret
== -EAGAIN
) {
7147 btrfs_add_free_space(block_group
, offset
, num_bytes
);
7151 /* we are all good, lets return */
7152 ins
->objectid
= search_start
;
7153 ins
->offset
= num_bytes
;
7155 trace_btrfs_reserve_extent(orig_root
, block_group
,
7156 search_start
, num_bytes
);
7157 btrfs_release_block_group(block_group
, delalloc
);
7160 failed_cluster_refill
= false;
7161 failed_alloc
= false;
7162 BUG_ON(index
!= get_block_group_index(block_group
));
7163 btrfs_release_block_group(block_group
, delalloc
);
7165 up_read(&space_info
->groups_sem
);
7167 if (!ins
->objectid
&& loop
>= LOOP_CACHING_WAIT
&& have_caching_bg
)
7170 if (!ins
->objectid
&& ++index
< BTRFS_NR_RAID_TYPES
)
7174 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7175 * caching kthreads as we move along
7176 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7177 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7178 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7181 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
) {
7184 if (loop
== LOOP_ALLOC_CHUNK
) {
7185 struct btrfs_trans_handle
*trans
;
7188 trans
= current
->journal_info
;
7192 trans
= btrfs_join_transaction(root
);
7194 if (IS_ERR(trans
)) {
7195 ret
= PTR_ERR(trans
);
7199 ret
= do_chunk_alloc(trans
, root
, flags
,
7202 * Do not bail out on ENOSPC since we
7203 * can do more things.
7205 if (ret
< 0 && ret
!= -ENOSPC
)
7206 btrfs_abort_transaction(trans
,
7211 btrfs_end_transaction(trans
, root
);
7216 if (loop
== LOOP_NO_EMPTY_SIZE
) {
7222 } else if (!ins
->objectid
) {
7224 } else if (ins
->objectid
) {
7229 ins
->offset
= max_extent_size
;
7233 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
7234 int dump_block_groups
)
7236 struct btrfs_block_group_cache
*cache
;
7239 spin_lock(&info
->lock
);
7240 printk(KERN_INFO
"BTRFS: space_info %llu has %llu free, is %sfull\n",
7242 info
->total_bytes
- info
->bytes_used
- info
->bytes_pinned
-
7243 info
->bytes_reserved
- info
->bytes_readonly
,
7244 (info
->full
) ? "" : "not ");
7245 printk(KERN_INFO
"BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7246 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7247 info
->total_bytes
, info
->bytes_used
, info
->bytes_pinned
,
7248 info
->bytes_reserved
, info
->bytes_may_use
,
7249 info
->bytes_readonly
);
7250 spin_unlock(&info
->lock
);
7252 if (!dump_block_groups
)
7255 down_read(&info
->groups_sem
);
7257 list_for_each_entry(cache
, &info
->block_groups
[index
], list
) {
7258 spin_lock(&cache
->lock
);
7259 printk(KERN_INFO
"BTRFS: "
7260 "block group %llu has %llu bytes, "
7261 "%llu used %llu pinned %llu reserved %s\n",
7262 cache
->key
.objectid
, cache
->key
.offset
,
7263 btrfs_block_group_used(&cache
->item
), cache
->pinned
,
7264 cache
->reserved
, cache
->ro
? "[readonly]" : "");
7265 btrfs_dump_free_space(cache
, bytes
);
7266 spin_unlock(&cache
->lock
);
7268 if (++index
< BTRFS_NR_RAID_TYPES
)
7270 up_read(&info
->groups_sem
);
7273 int btrfs_reserve_extent(struct btrfs_root
*root
,
7274 u64 num_bytes
, u64 min_alloc_size
,
7275 u64 empty_size
, u64 hint_byte
,
7276 struct btrfs_key
*ins
, int is_data
, int delalloc
)
7278 bool final_tried
= false;
7282 flags
= btrfs_get_alloc_profile(root
, is_data
);
7284 WARN_ON(num_bytes
< root
->sectorsize
);
7285 ret
= find_free_extent(root
, num_bytes
, empty_size
, hint_byte
, ins
,
7288 if (ret
== -ENOSPC
) {
7289 if (!final_tried
&& ins
->offset
) {
7290 num_bytes
= min(num_bytes
>> 1, ins
->offset
);
7291 num_bytes
= round_down(num_bytes
, root
->sectorsize
);
7292 num_bytes
= max(num_bytes
, min_alloc_size
);
7293 if (num_bytes
== min_alloc_size
)
7296 } else if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
7297 struct btrfs_space_info
*sinfo
;
7299 sinfo
= __find_space_info(root
->fs_info
, flags
);
7300 btrfs_err(root
->fs_info
, "allocation failed flags %llu, wanted %llu",
7303 dump_space_info(sinfo
, num_bytes
, 1);
7310 static int __btrfs_free_reserved_extent(struct btrfs_root
*root
,
7312 int pin
, int delalloc
)
7314 struct btrfs_block_group_cache
*cache
;
7317 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
7319 btrfs_err(root
->fs_info
, "Unable to find block group for %llu",
7325 pin_down_extent(root
, cache
, start
, len
, 1);
7327 if (btrfs_test_opt(root
, DISCARD
))
7328 ret
= btrfs_discard_extent(root
, start
, len
, NULL
);
7329 btrfs_add_free_space(cache
, start
, len
);
7330 btrfs_update_reserved_bytes(cache
, len
, RESERVE_FREE
, delalloc
);
7333 btrfs_put_block_group(cache
);
7335 trace_btrfs_reserved_extent_free(root
, start
, len
);
7340 int btrfs_free_reserved_extent(struct btrfs_root
*root
,
7341 u64 start
, u64 len
, int delalloc
)
7343 return __btrfs_free_reserved_extent(root
, start
, len
, 0, delalloc
);
7346 int btrfs_free_and_pin_reserved_extent(struct btrfs_root
*root
,
7349 return __btrfs_free_reserved_extent(root
, start
, len
, 1, 0);
7352 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
7353 struct btrfs_root
*root
,
7354 u64 parent
, u64 root_objectid
,
7355 u64 flags
, u64 owner
, u64 offset
,
7356 struct btrfs_key
*ins
, int ref_mod
)
7359 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7360 struct btrfs_extent_item
*extent_item
;
7361 struct btrfs_extent_inline_ref
*iref
;
7362 struct btrfs_path
*path
;
7363 struct extent_buffer
*leaf
;
7368 type
= BTRFS_SHARED_DATA_REF_KEY
;
7370 type
= BTRFS_EXTENT_DATA_REF_KEY
;
7372 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
7374 path
= btrfs_alloc_path();
7378 path
->leave_spinning
= 1;
7379 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
7382 btrfs_free_path(path
);
7386 leaf
= path
->nodes
[0];
7387 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
7388 struct btrfs_extent_item
);
7389 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
7390 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
7391 btrfs_set_extent_flags(leaf
, extent_item
,
7392 flags
| BTRFS_EXTENT_FLAG_DATA
);
7394 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
7395 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
7397 struct btrfs_shared_data_ref
*ref
;
7398 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
7399 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
7400 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
7402 struct btrfs_extent_data_ref
*ref
;
7403 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
7404 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
7405 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
7406 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
7407 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
7410 btrfs_mark_buffer_dirty(path
->nodes
[0]);
7411 btrfs_free_path(path
);
7413 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1);
7414 if (ret
) { /* -ENOENT, logic error */
7415 btrfs_err(fs_info
, "update block group failed for %llu %llu",
7416 ins
->objectid
, ins
->offset
);
7419 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, ins
->offset
);
7423 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
7424 struct btrfs_root
*root
,
7425 u64 parent
, u64 root_objectid
,
7426 u64 flags
, struct btrfs_disk_key
*key
,
7427 int level
, struct btrfs_key
*ins
,
7431 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7432 struct btrfs_extent_item
*extent_item
;
7433 struct btrfs_tree_block_info
*block_info
;
7434 struct btrfs_extent_inline_ref
*iref
;
7435 struct btrfs_path
*path
;
7436 struct extent_buffer
*leaf
;
7437 u32 size
= sizeof(*extent_item
) + sizeof(*iref
);
7438 u64 num_bytes
= ins
->offset
;
7439 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
7442 if (!skinny_metadata
)
7443 size
+= sizeof(*block_info
);
7445 path
= btrfs_alloc_path();
7447 btrfs_free_and_pin_reserved_extent(root
, ins
->objectid
,
7452 path
->leave_spinning
= 1;
7453 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
7456 btrfs_free_path(path
);
7457 btrfs_free_and_pin_reserved_extent(root
, ins
->objectid
,
7462 leaf
= path
->nodes
[0];
7463 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
7464 struct btrfs_extent_item
);
7465 btrfs_set_extent_refs(leaf
, extent_item
, 1);
7466 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
7467 btrfs_set_extent_flags(leaf
, extent_item
,
7468 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
7470 if (skinny_metadata
) {
7471 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
7472 num_bytes
= root
->nodesize
;
7474 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
7475 btrfs_set_tree_block_key(leaf
, block_info
, key
);
7476 btrfs_set_tree_block_level(leaf
, block_info
, level
);
7477 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
7481 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
7482 btrfs_set_extent_inline_ref_type(leaf
, iref
,
7483 BTRFS_SHARED_BLOCK_REF_KEY
);
7484 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
7486 btrfs_set_extent_inline_ref_type(leaf
, iref
,
7487 BTRFS_TREE_BLOCK_REF_KEY
);
7488 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
7491 btrfs_mark_buffer_dirty(leaf
);
7492 btrfs_free_path(path
);
7494 ret
= update_block_group(trans
, root
, ins
->objectid
, root
->nodesize
,
7496 if (ret
) { /* -ENOENT, logic error */
7497 btrfs_err(fs_info
, "update block group failed for %llu %llu",
7498 ins
->objectid
, ins
->offset
);
7502 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, root
->nodesize
);
7506 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
7507 struct btrfs_root
*root
,
7508 u64 root_objectid
, u64 owner
,
7509 u64 offset
, struct btrfs_key
*ins
)
7513 BUG_ON(root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
7515 ret
= btrfs_add_delayed_data_ref(root
->fs_info
, trans
, ins
->objectid
,
7517 root_objectid
, owner
, offset
,
7518 BTRFS_ADD_DELAYED_EXTENT
, NULL
, 0);
7523 * this is used by the tree logging recovery code. It records that
7524 * an extent has been allocated and makes sure to clear the free
7525 * space cache bits as well
7527 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
7528 struct btrfs_root
*root
,
7529 u64 root_objectid
, u64 owner
, u64 offset
,
7530 struct btrfs_key
*ins
)
7533 struct btrfs_block_group_cache
*block_group
;
7536 * Mixed block groups will exclude before processing the log so we only
7537 * need to do the exlude dance if this fs isn't mixed.
7539 if (!btrfs_fs_incompat(root
->fs_info
, MIXED_GROUPS
)) {
7540 ret
= __exclude_logged_extent(root
, ins
->objectid
, ins
->offset
);
7545 block_group
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
7549 ret
= btrfs_update_reserved_bytes(block_group
, ins
->offset
,
7550 RESERVE_ALLOC_NO_ACCOUNT
, 0);
7551 BUG_ON(ret
); /* logic error */
7552 ret
= alloc_reserved_file_extent(trans
, root
, 0, root_objectid
,
7553 0, owner
, offset
, ins
, 1);
7554 btrfs_put_block_group(block_group
);
7558 static struct extent_buffer
*
7559 btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
7560 u64 bytenr
, int level
)
7562 struct extent_buffer
*buf
;
7564 buf
= btrfs_find_create_tree_block(root
, bytenr
);
7566 return ERR_PTR(-ENOMEM
);
7567 btrfs_set_header_generation(buf
, trans
->transid
);
7568 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, buf
, level
);
7569 btrfs_tree_lock(buf
);
7570 clean_tree_block(trans
, root
->fs_info
, buf
);
7571 clear_bit(EXTENT_BUFFER_STALE
, &buf
->bflags
);
7573 btrfs_set_lock_blocking(buf
);
7574 btrfs_set_buffer_uptodate(buf
);
7576 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
7577 buf
->log_index
= root
->log_transid
% 2;
7579 * we allow two log transactions at a time, use different
7580 * EXENT bit to differentiate dirty pages.
7582 if (buf
->log_index
== 0)
7583 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
7584 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
7586 set_extent_new(&root
->dirty_log_pages
, buf
->start
,
7587 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
7589 buf
->log_index
= -1;
7590 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
7591 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
7593 trans
->blocks_used
++;
7594 /* this returns a buffer locked for blocking */
7598 static struct btrfs_block_rsv
*
7599 use_block_rsv(struct btrfs_trans_handle
*trans
,
7600 struct btrfs_root
*root
, u32 blocksize
)
7602 struct btrfs_block_rsv
*block_rsv
;
7603 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
7605 bool global_updated
= false;
7607 block_rsv
= get_block_rsv(trans
, root
);
7609 if (unlikely(block_rsv
->size
== 0))
7612 ret
= block_rsv_use_bytes(block_rsv
, blocksize
);
7616 if (block_rsv
->failfast
)
7617 return ERR_PTR(ret
);
7619 if (block_rsv
->type
== BTRFS_BLOCK_RSV_GLOBAL
&& !global_updated
) {
7620 global_updated
= true;
7621 update_global_block_rsv(root
->fs_info
);
7625 if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
7626 static DEFINE_RATELIMIT_STATE(_rs
,
7627 DEFAULT_RATELIMIT_INTERVAL
* 10,
7628 /*DEFAULT_RATELIMIT_BURST*/ 1);
7629 if (__ratelimit(&_rs
))
7631 "BTRFS: block rsv returned %d\n", ret
);
7634 ret
= reserve_metadata_bytes(root
, block_rsv
, blocksize
,
7635 BTRFS_RESERVE_NO_FLUSH
);
7639 * If we couldn't reserve metadata bytes try and use some from
7640 * the global reserve if its space type is the same as the global
7643 if (block_rsv
->type
!= BTRFS_BLOCK_RSV_GLOBAL
&&
7644 block_rsv
->space_info
== global_rsv
->space_info
) {
7645 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
7649 return ERR_PTR(ret
);
7652 static void unuse_block_rsv(struct btrfs_fs_info
*fs_info
,
7653 struct btrfs_block_rsv
*block_rsv
, u32 blocksize
)
7655 block_rsv_add_bytes(block_rsv
, blocksize
, 0);
7656 block_rsv_release_bytes(fs_info
, block_rsv
, NULL
, 0);
7660 * finds a free extent and does all the dirty work required for allocation
7661 * returns the tree buffer or an ERR_PTR on error.
7663 struct extent_buffer
*btrfs_alloc_tree_block(struct btrfs_trans_handle
*trans
,
7664 struct btrfs_root
*root
,
7665 u64 parent
, u64 root_objectid
,
7666 struct btrfs_disk_key
*key
, int level
,
7667 u64 hint
, u64 empty_size
)
7669 struct btrfs_key ins
;
7670 struct btrfs_block_rsv
*block_rsv
;
7671 struct extent_buffer
*buf
;
7672 struct btrfs_delayed_extent_op
*extent_op
;
7675 u32 blocksize
= root
->nodesize
;
7676 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
7679 if (btrfs_test_is_dummy_root(root
)) {
7680 buf
= btrfs_init_new_buffer(trans
, root
, root
->alloc_bytenr
,
7683 root
->alloc_bytenr
+= blocksize
;
7687 block_rsv
= use_block_rsv(trans
, root
, blocksize
);
7688 if (IS_ERR(block_rsv
))
7689 return ERR_CAST(block_rsv
);
7691 ret
= btrfs_reserve_extent(root
, blocksize
, blocksize
,
7692 empty_size
, hint
, &ins
, 0, 0);
7696 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
, level
);
7699 goto out_free_reserved
;
7702 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
7704 parent
= ins
.objectid
;
7705 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
7709 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
7710 extent_op
= btrfs_alloc_delayed_extent_op();
7716 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
7718 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
7719 extent_op
->flags_to_set
= flags
;
7720 if (skinny_metadata
)
7721 extent_op
->update_key
= 0;
7723 extent_op
->update_key
= 1;
7724 extent_op
->update_flags
= 1;
7725 extent_op
->is_data
= 0;
7726 extent_op
->level
= level
;
7728 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
7729 ins
.objectid
, ins
.offset
,
7730 parent
, root_objectid
, level
,
7731 BTRFS_ADD_DELAYED_EXTENT
,
7734 goto out_free_delayed
;
7739 btrfs_free_delayed_extent_op(extent_op
);
7741 free_extent_buffer(buf
);
7743 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
, 0);
7745 unuse_block_rsv(root
->fs_info
, block_rsv
, blocksize
);
7746 return ERR_PTR(ret
);
7749 struct walk_control
{
7750 u64 refs
[BTRFS_MAX_LEVEL
];
7751 u64 flags
[BTRFS_MAX_LEVEL
];
7752 struct btrfs_key update_progress
;
7763 #define DROP_REFERENCE 1
7764 #define UPDATE_BACKREF 2
7766 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
7767 struct btrfs_root
*root
,
7768 struct walk_control
*wc
,
7769 struct btrfs_path
*path
)
7777 struct btrfs_key key
;
7778 struct extent_buffer
*eb
;
7783 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
7784 wc
->reada_count
= wc
->reada_count
* 2 / 3;
7785 wc
->reada_count
= max(wc
->reada_count
, 2);
7787 wc
->reada_count
= wc
->reada_count
* 3 / 2;
7788 wc
->reada_count
= min_t(int, wc
->reada_count
,
7789 BTRFS_NODEPTRS_PER_BLOCK(root
));
7792 eb
= path
->nodes
[wc
->level
];
7793 nritems
= btrfs_header_nritems(eb
);
7794 blocksize
= root
->nodesize
;
7796 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
7797 if (nread
>= wc
->reada_count
)
7801 bytenr
= btrfs_node_blockptr(eb
, slot
);
7802 generation
= btrfs_node_ptr_generation(eb
, slot
);
7804 if (slot
== path
->slots
[wc
->level
])
7807 if (wc
->stage
== UPDATE_BACKREF
&&
7808 generation
<= root
->root_key
.offset
)
7811 /* We don't lock the tree block, it's OK to be racy here */
7812 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
,
7813 wc
->level
- 1, 1, &refs
,
7815 /* We don't care about errors in readahead. */
7820 if (wc
->stage
== DROP_REFERENCE
) {
7824 if (wc
->level
== 1 &&
7825 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
7827 if (!wc
->update_ref
||
7828 generation
<= root
->root_key
.offset
)
7830 btrfs_node_key_to_cpu(eb
, &key
, slot
);
7831 ret
= btrfs_comp_cpu_keys(&key
,
7832 &wc
->update_progress
);
7836 if (wc
->level
== 1 &&
7837 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
7841 readahead_tree_block(root
, bytenr
);
7844 wc
->reada_slot
= slot
;
7848 * TODO: Modify related function to add related node/leaf to dirty_extent_root,
7849 * for later qgroup accounting.
7851 * Current, this function does nothing.
7853 static int account_leaf_items(struct btrfs_trans_handle
*trans
,
7854 struct btrfs_root
*root
,
7855 struct extent_buffer
*eb
)
7857 int nr
= btrfs_header_nritems(eb
);
7859 struct btrfs_key key
;
7860 struct btrfs_file_extent_item
*fi
;
7861 u64 bytenr
, num_bytes
;
7863 for (i
= 0; i
< nr
; i
++) {
7864 btrfs_item_key_to_cpu(eb
, &key
, i
);
7866 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
7869 fi
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
7870 /* filter out non qgroup-accountable extents */
7871 extent_type
= btrfs_file_extent_type(eb
, fi
);
7873 if (extent_type
== BTRFS_FILE_EXTENT_INLINE
)
7876 bytenr
= btrfs_file_extent_disk_bytenr(eb
, fi
);
7880 num_bytes
= btrfs_file_extent_disk_num_bytes(eb
, fi
);
7886 * Walk up the tree from the bottom, freeing leaves and any interior
7887 * nodes which have had all slots visited. If a node (leaf or
7888 * interior) is freed, the node above it will have it's slot
7889 * incremented. The root node will never be freed.
7891 * At the end of this function, we should have a path which has all
7892 * slots incremented to the next position for a search. If we need to
7893 * read a new node it will be NULL and the node above it will have the
7894 * correct slot selected for a later read.
7896 * If we increment the root nodes slot counter past the number of
7897 * elements, 1 is returned to signal completion of the search.
7899 static int adjust_slots_upwards(struct btrfs_root
*root
,
7900 struct btrfs_path
*path
, int root_level
)
7904 struct extent_buffer
*eb
;
7906 if (root_level
== 0)
7909 while (level
<= root_level
) {
7910 eb
= path
->nodes
[level
];
7911 nr
= btrfs_header_nritems(eb
);
7912 path
->slots
[level
]++;
7913 slot
= path
->slots
[level
];
7914 if (slot
>= nr
|| level
== 0) {
7916 * Don't free the root - we will detect this
7917 * condition after our loop and return a
7918 * positive value for caller to stop walking the tree.
7920 if (level
!= root_level
) {
7921 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7922 path
->locks
[level
] = 0;
7924 free_extent_buffer(eb
);
7925 path
->nodes
[level
] = NULL
;
7926 path
->slots
[level
] = 0;
7930 * We have a valid slot to walk back down
7931 * from. Stop here so caller can process these
7940 eb
= path
->nodes
[root_level
];
7941 if (path
->slots
[root_level
] >= btrfs_header_nritems(eb
))
7948 * root_eb is the subtree root and is locked before this function is called.
7949 * TODO: Modify this function to mark all (including complete shared node)
7950 * to dirty_extent_root to allow it get accounted in qgroup.
7952 static int account_shared_subtree(struct btrfs_trans_handle
*trans
,
7953 struct btrfs_root
*root
,
7954 struct extent_buffer
*root_eb
,
7960 struct extent_buffer
*eb
= root_eb
;
7961 struct btrfs_path
*path
= NULL
;
7963 BUG_ON(root_level
< 0 || root_level
> BTRFS_MAX_LEVEL
);
7964 BUG_ON(root_eb
== NULL
);
7966 if (!root
->fs_info
->quota_enabled
)
7969 if (!extent_buffer_uptodate(root_eb
)) {
7970 ret
= btrfs_read_buffer(root_eb
, root_gen
);
7975 if (root_level
== 0) {
7976 ret
= account_leaf_items(trans
, root
, root_eb
);
7980 path
= btrfs_alloc_path();
7985 * Walk down the tree. Missing extent blocks are filled in as
7986 * we go. Metadata is accounted every time we read a new
7989 * When we reach a leaf, we account for file extent items in it,
7990 * walk back up the tree (adjusting slot pointers as we go)
7991 * and restart the search process.
7993 extent_buffer_get(root_eb
); /* For path */
7994 path
->nodes
[root_level
] = root_eb
;
7995 path
->slots
[root_level
] = 0;
7996 path
->locks
[root_level
] = 0; /* so release_path doesn't try to unlock */
7999 while (level
>= 0) {
8000 if (path
->nodes
[level
] == NULL
) {
8005 /* We need to get child blockptr/gen from
8006 * parent before we can read it. */
8007 eb
= path
->nodes
[level
+ 1];
8008 parent_slot
= path
->slots
[level
+ 1];
8009 child_bytenr
= btrfs_node_blockptr(eb
, parent_slot
);
8010 child_gen
= btrfs_node_ptr_generation(eb
, parent_slot
);
8012 eb
= read_tree_block(root
, child_bytenr
, child_gen
);
8016 } else if (!extent_buffer_uptodate(eb
)) {
8017 free_extent_buffer(eb
);
8022 path
->nodes
[level
] = eb
;
8023 path
->slots
[level
] = 0;
8025 btrfs_tree_read_lock(eb
);
8026 btrfs_set_lock_blocking_rw(eb
, BTRFS_READ_LOCK
);
8027 path
->locks
[level
] = BTRFS_READ_LOCK_BLOCKING
;
8031 ret
= account_leaf_items(trans
, root
, path
->nodes
[level
]);
8035 /* Nonzero return here means we completed our search */
8036 ret
= adjust_slots_upwards(root
, path
, root_level
);
8040 /* Restart search with new slots */
8049 btrfs_free_path(path
);
8055 * helper to process tree block while walking down the tree.
8057 * when wc->stage == UPDATE_BACKREF, this function updates
8058 * back refs for pointers in the block.
8060 * NOTE: return value 1 means we should stop walking down.
8062 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
8063 struct btrfs_root
*root
,
8064 struct btrfs_path
*path
,
8065 struct walk_control
*wc
, int lookup_info
)
8067 int level
= wc
->level
;
8068 struct extent_buffer
*eb
= path
->nodes
[level
];
8069 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
8072 if (wc
->stage
== UPDATE_BACKREF
&&
8073 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
8077 * when reference count of tree block is 1, it won't increase
8078 * again. once full backref flag is set, we never clear it.
8081 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
8082 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
8083 BUG_ON(!path
->locks
[level
]);
8084 ret
= btrfs_lookup_extent_info(trans
, root
,
8085 eb
->start
, level
, 1,
8088 BUG_ON(ret
== -ENOMEM
);
8091 BUG_ON(wc
->refs
[level
] == 0);
8094 if (wc
->stage
== DROP_REFERENCE
) {
8095 if (wc
->refs
[level
] > 1)
8098 if (path
->locks
[level
] && !wc
->keep_locks
) {
8099 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8100 path
->locks
[level
] = 0;
8105 /* wc->stage == UPDATE_BACKREF */
8106 if (!(wc
->flags
[level
] & flag
)) {
8107 BUG_ON(!path
->locks
[level
]);
8108 ret
= btrfs_inc_ref(trans
, root
, eb
, 1);
8109 BUG_ON(ret
); /* -ENOMEM */
8110 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
8111 BUG_ON(ret
); /* -ENOMEM */
8112 ret
= btrfs_set_disk_extent_flags(trans
, root
, eb
->start
,
8114 btrfs_header_level(eb
), 0);
8115 BUG_ON(ret
); /* -ENOMEM */
8116 wc
->flags
[level
] |= flag
;
8120 * the block is shared by multiple trees, so it's not good to
8121 * keep the tree lock
8123 if (path
->locks
[level
] && level
> 0) {
8124 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8125 path
->locks
[level
] = 0;
8131 * helper to process tree block pointer.
8133 * when wc->stage == DROP_REFERENCE, this function checks
8134 * reference count of the block pointed to. if the block
8135 * is shared and we need update back refs for the subtree
8136 * rooted at the block, this function changes wc->stage to
8137 * UPDATE_BACKREF. if the block is shared and there is no
8138 * need to update back, this function drops the reference
8141 * NOTE: return value 1 means we should stop walking down.
8143 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
8144 struct btrfs_root
*root
,
8145 struct btrfs_path
*path
,
8146 struct walk_control
*wc
, int *lookup_info
)
8152 struct btrfs_key key
;
8153 struct extent_buffer
*next
;
8154 int level
= wc
->level
;
8157 bool need_account
= false;
8159 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
8160 path
->slots
[level
]);
8162 * if the lower level block was created before the snapshot
8163 * was created, we know there is no need to update back refs
8166 if (wc
->stage
== UPDATE_BACKREF
&&
8167 generation
<= root
->root_key
.offset
) {
8172 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
8173 blocksize
= root
->nodesize
;
8175 next
= btrfs_find_tree_block(root
->fs_info
, bytenr
);
8177 next
= btrfs_find_create_tree_block(root
, bytenr
);
8180 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, next
,
8184 btrfs_tree_lock(next
);
8185 btrfs_set_lock_blocking(next
);
8187 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, level
- 1, 1,
8188 &wc
->refs
[level
- 1],
8189 &wc
->flags
[level
- 1]);
8191 btrfs_tree_unlock(next
);
8195 if (unlikely(wc
->refs
[level
- 1] == 0)) {
8196 btrfs_err(root
->fs_info
, "Missing references.");
8201 if (wc
->stage
== DROP_REFERENCE
) {
8202 if (wc
->refs
[level
- 1] > 1) {
8203 need_account
= true;
8205 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8208 if (!wc
->update_ref
||
8209 generation
<= root
->root_key
.offset
)
8212 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
8213 path
->slots
[level
]);
8214 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
8218 wc
->stage
= UPDATE_BACKREF
;
8219 wc
->shared_level
= level
- 1;
8223 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8227 if (!btrfs_buffer_uptodate(next
, generation
, 0)) {
8228 btrfs_tree_unlock(next
);
8229 free_extent_buffer(next
);
8235 if (reada
&& level
== 1)
8236 reada_walk_down(trans
, root
, wc
, path
);
8237 next
= read_tree_block(root
, bytenr
, generation
);
8239 return PTR_ERR(next
);
8240 } else if (!extent_buffer_uptodate(next
)) {
8241 free_extent_buffer(next
);
8244 btrfs_tree_lock(next
);
8245 btrfs_set_lock_blocking(next
);
8249 BUG_ON(level
!= btrfs_header_level(next
));
8250 path
->nodes
[level
] = next
;
8251 path
->slots
[level
] = 0;
8252 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8258 wc
->refs
[level
- 1] = 0;
8259 wc
->flags
[level
- 1] = 0;
8260 if (wc
->stage
== DROP_REFERENCE
) {
8261 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
8262 parent
= path
->nodes
[level
]->start
;
8264 BUG_ON(root
->root_key
.objectid
!=
8265 btrfs_header_owner(path
->nodes
[level
]));
8270 ret
= account_shared_subtree(trans
, root
, next
,
8271 generation
, level
- 1);
8273 btrfs_err_rl(root
->fs_info
,
8275 "%d accounting shared subtree. Quota "
8276 "is out of sync, rescan required.",
8280 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
, parent
,
8281 root
->root_key
.objectid
, level
- 1, 0, 0);
8282 BUG_ON(ret
); /* -ENOMEM */
8284 btrfs_tree_unlock(next
);
8285 free_extent_buffer(next
);
8291 * helper to process tree block while walking up the tree.
8293 * when wc->stage == DROP_REFERENCE, this function drops
8294 * reference count on the block.
8296 * when wc->stage == UPDATE_BACKREF, this function changes
8297 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8298 * to UPDATE_BACKREF previously while processing the block.
8300 * NOTE: return value 1 means we should stop walking up.
8302 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
8303 struct btrfs_root
*root
,
8304 struct btrfs_path
*path
,
8305 struct walk_control
*wc
)
8308 int level
= wc
->level
;
8309 struct extent_buffer
*eb
= path
->nodes
[level
];
8312 if (wc
->stage
== UPDATE_BACKREF
) {
8313 BUG_ON(wc
->shared_level
< level
);
8314 if (level
< wc
->shared_level
)
8317 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
8321 wc
->stage
= DROP_REFERENCE
;
8322 wc
->shared_level
= -1;
8323 path
->slots
[level
] = 0;
8326 * check reference count again if the block isn't locked.
8327 * we should start walking down the tree again if reference
8330 if (!path
->locks
[level
]) {
8332 btrfs_tree_lock(eb
);
8333 btrfs_set_lock_blocking(eb
);
8334 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8336 ret
= btrfs_lookup_extent_info(trans
, root
,
8337 eb
->start
, level
, 1,
8341 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8342 path
->locks
[level
] = 0;
8345 BUG_ON(wc
->refs
[level
] == 0);
8346 if (wc
->refs
[level
] == 1) {
8347 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8348 path
->locks
[level
] = 0;
8354 /* wc->stage == DROP_REFERENCE */
8355 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
8357 if (wc
->refs
[level
] == 1) {
8359 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8360 ret
= btrfs_dec_ref(trans
, root
, eb
, 1);
8362 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
8363 BUG_ON(ret
); /* -ENOMEM */
8364 ret
= account_leaf_items(trans
, root
, eb
);
8366 btrfs_err_rl(root
->fs_info
,
8368 "%d accounting leaf items. Quota "
8369 "is out of sync, rescan required.",
8373 /* make block locked assertion in clean_tree_block happy */
8374 if (!path
->locks
[level
] &&
8375 btrfs_header_generation(eb
) == trans
->transid
) {
8376 btrfs_tree_lock(eb
);
8377 btrfs_set_lock_blocking(eb
);
8378 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8380 clean_tree_block(trans
, root
->fs_info
, eb
);
8383 if (eb
== root
->node
) {
8384 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8387 BUG_ON(root
->root_key
.objectid
!=
8388 btrfs_header_owner(eb
));
8390 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8391 parent
= path
->nodes
[level
+ 1]->start
;
8393 BUG_ON(root
->root_key
.objectid
!=
8394 btrfs_header_owner(path
->nodes
[level
+ 1]));
8397 btrfs_free_tree_block(trans
, root
, eb
, parent
, wc
->refs
[level
] == 1);
8399 wc
->refs
[level
] = 0;
8400 wc
->flags
[level
] = 0;
8404 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
8405 struct btrfs_root
*root
,
8406 struct btrfs_path
*path
,
8407 struct walk_control
*wc
)
8409 int level
= wc
->level
;
8410 int lookup_info
= 1;
8413 while (level
>= 0) {
8414 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
8421 if (path
->slots
[level
] >=
8422 btrfs_header_nritems(path
->nodes
[level
]))
8425 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
8427 path
->slots
[level
]++;
8436 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
8437 struct btrfs_root
*root
,
8438 struct btrfs_path
*path
,
8439 struct walk_control
*wc
, int max_level
)
8441 int level
= wc
->level
;
8444 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
8445 while (level
< max_level
&& path
->nodes
[level
]) {
8447 if (path
->slots
[level
] + 1 <
8448 btrfs_header_nritems(path
->nodes
[level
])) {
8449 path
->slots
[level
]++;
8452 ret
= walk_up_proc(trans
, root
, path
, wc
);
8456 if (path
->locks
[level
]) {
8457 btrfs_tree_unlock_rw(path
->nodes
[level
],
8458 path
->locks
[level
]);
8459 path
->locks
[level
] = 0;
8461 free_extent_buffer(path
->nodes
[level
]);
8462 path
->nodes
[level
] = NULL
;
8470 * drop a subvolume tree.
8472 * this function traverses the tree freeing any blocks that only
8473 * referenced by the tree.
8475 * when a shared tree block is found. this function decreases its
8476 * reference count by one. if update_ref is true, this function
8477 * also make sure backrefs for the shared block and all lower level
8478 * blocks are properly updated.
8480 * If called with for_reloc == 0, may exit early with -EAGAIN
8482 int btrfs_drop_snapshot(struct btrfs_root
*root
,
8483 struct btrfs_block_rsv
*block_rsv
, int update_ref
,
8486 struct btrfs_path
*path
;
8487 struct btrfs_trans_handle
*trans
;
8488 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
8489 struct btrfs_root_item
*root_item
= &root
->root_item
;
8490 struct walk_control
*wc
;
8491 struct btrfs_key key
;
8495 bool root_dropped
= false;
8497 btrfs_debug(root
->fs_info
, "Drop subvolume %llu", root
->objectid
);
8499 path
= btrfs_alloc_path();
8505 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
8507 btrfs_free_path(path
);
8512 trans
= btrfs_start_transaction(tree_root
, 0);
8513 if (IS_ERR(trans
)) {
8514 err
= PTR_ERR(trans
);
8519 trans
->block_rsv
= block_rsv
;
8521 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
8522 level
= btrfs_header_level(root
->node
);
8523 path
->nodes
[level
] = btrfs_lock_root_node(root
);
8524 btrfs_set_lock_blocking(path
->nodes
[level
]);
8525 path
->slots
[level
] = 0;
8526 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8527 memset(&wc
->update_progress
, 0,
8528 sizeof(wc
->update_progress
));
8530 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
8531 memcpy(&wc
->update_progress
, &key
,
8532 sizeof(wc
->update_progress
));
8534 level
= root_item
->drop_level
;
8536 path
->lowest_level
= level
;
8537 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
8538 path
->lowest_level
= 0;
8546 * unlock our path, this is safe because only this
8547 * function is allowed to delete this snapshot
8549 btrfs_unlock_up_safe(path
, 0);
8551 level
= btrfs_header_level(root
->node
);
8553 btrfs_tree_lock(path
->nodes
[level
]);
8554 btrfs_set_lock_blocking(path
->nodes
[level
]);
8555 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8557 ret
= btrfs_lookup_extent_info(trans
, root
,
8558 path
->nodes
[level
]->start
,
8559 level
, 1, &wc
->refs
[level
],
8565 BUG_ON(wc
->refs
[level
] == 0);
8567 if (level
== root_item
->drop_level
)
8570 btrfs_tree_unlock(path
->nodes
[level
]);
8571 path
->locks
[level
] = 0;
8572 WARN_ON(wc
->refs
[level
] != 1);
8578 wc
->shared_level
= -1;
8579 wc
->stage
= DROP_REFERENCE
;
8580 wc
->update_ref
= update_ref
;
8582 wc
->for_reloc
= for_reloc
;
8583 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
8587 ret
= walk_down_tree(trans
, root
, path
, wc
);
8593 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
8600 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
8604 if (wc
->stage
== DROP_REFERENCE
) {
8606 btrfs_node_key(path
->nodes
[level
],
8607 &root_item
->drop_progress
,
8608 path
->slots
[level
]);
8609 root_item
->drop_level
= level
;
8612 BUG_ON(wc
->level
== 0);
8613 if (btrfs_should_end_transaction(trans
, tree_root
) ||
8614 (!for_reloc
&& btrfs_need_cleaner_sleep(root
))) {
8615 ret
= btrfs_update_root(trans
, tree_root
,
8619 btrfs_abort_transaction(trans
, tree_root
, ret
);
8624 btrfs_end_transaction_throttle(trans
, tree_root
);
8625 if (!for_reloc
&& btrfs_need_cleaner_sleep(root
)) {
8626 pr_debug("BTRFS: drop snapshot early exit\n");
8631 trans
= btrfs_start_transaction(tree_root
, 0);
8632 if (IS_ERR(trans
)) {
8633 err
= PTR_ERR(trans
);
8637 trans
->block_rsv
= block_rsv
;
8640 btrfs_release_path(path
);
8644 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
8646 btrfs_abort_transaction(trans
, tree_root
, ret
);
8650 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
8651 ret
= btrfs_find_root(tree_root
, &root
->root_key
, path
,
8654 btrfs_abort_transaction(trans
, tree_root
, ret
);
8657 } else if (ret
> 0) {
8658 /* if we fail to delete the orphan item this time
8659 * around, it'll get picked up the next time.
8661 * The most common failure here is just -ENOENT.
8663 btrfs_del_orphan_item(trans
, tree_root
,
8664 root
->root_key
.objectid
);
8668 if (test_bit(BTRFS_ROOT_IN_RADIX
, &root
->state
)) {
8669 btrfs_add_dropped_root(trans
, root
);
8671 free_extent_buffer(root
->node
);
8672 free_extent_buffer(root
->commit_root
);
8673 btrfs_put_fs_root(root
);
8675 root_dropped
= true;
8677 btrfs_end_transaction_throttle(trans
, tree_root
);
8680 btrfs_free_path(path
);
8683 * So if we need to stop dropping the snapshot for whatever reason we
8684 * need to make sure to add it back to the dead root list so that we
8685 * keep trying to do the work later. This also cleans up roots if we
8686 * don't have it in the radix (like when we recover after a power fail
8687 * or unmount) so we don't leak memory.
8689 if (!for_reloc
&& root_dropped
== false)
8690 btrfs_add_dead_root(root
);
8691 if (err
&& err
!= -EAGAIN
)
8692 btrfs_std_error(root
->fs_info
, err
, NULL
);
8697 * drop subtree rooted at tree block 'node'.
8699 * NOTE: this function will unlock and release tree block 'node'
8700 * only used by relocation code
8702 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
8703 struct btrfs_root
*root
,
8704 struct extent_buffer
*node
,
8705 struct extent_buffer
*parent
)
8707 struct btrfs_path
*path
;
8708 struct walk_control
*wc
;
8714 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
8716 path
= btrfs_alloc_path();
8720 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
8722 btrfs_free_path(path
);
8726 btrfs_assert_tree_locked(parent
);
8727 parent_level
= btrfs_header_level(parent
);
8728 extent_buffer_get(parent
);
8729 path
->nodes
[parent_level
] = parent
;
8730 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
8732 btrfs_assert_tree_locked(node
);
8733 level
= btrfs_header_level(node
);
8734 path
->nodes
[level
] = node
;
8735 path
->slots
[level
] = 0;
8736 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8738 wc
->refs
[parent_level
] = 1;
8739 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
8741 wc
->shared_level
= -1;
8742 wc
->stage
= DROP_REFERENCE
;
8746 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
8749 wret
= walk_down_tree(trans
, root
, path
, wc
);
8755 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
8763 btrfs_free_path(path
);
8767 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
8773 * if restripe for this chunk_type is on pick target profile and
8774 * return, otherwise do the usual balance
8776 stripped
= get_restripe_target(root
->fs_info
, flags
);
8778 return extended_to_chunk(stripped
);
8780 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
8782 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
8783 BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
|
8784 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
8786 if (num_devices
== 1) {
8787 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
8788 stripped
= flags
& ~stripped
;
8790 /* turn raid0 into single device chunks */
8791 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
8794 /* turn mirroring into duplication */
8795 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
8796 BTRFS_BLOCK_GROUP_RAID10
))
8797 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
8799 /* they already had raid on here, just return */
8800 if (flags
& stripped
)
8803 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
8804 stripped
= flags
& ~stripped
;
8806 /* switch duplicated blocks with raid1 */
8807 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
8808 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
8810 /* this is drive concat, leave it alone */
8816 static int inc_block_group_ro(struct btrfs_block_group_cache
*cache
, int force
)
8818 struct btrfs_space_info
*sinfo
= cache
->space_info
;
8820 u64 min_allocable_bytes
;
8824 * We need some metadata space and system metadata space for
8825 * allocating chunks in some corner cases until we force to set
8826 * it to be readonly.
8829 (BTRFS_BLOCK_GROUP_SYSTEM
| BTRFS_BLOCK_GROUP_METADATA
)) &&
8831 min_allocable_bytes
= 1 * 1024 * 1024;
8833 min_allocable_bytes
= 0;
8835 spin_lock(&sinfo
->lock
);
8836 spin_lock(&cache
->lock
);
8844 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
8845 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
8847 if (sinfo
->bytes_used
+ sinfo
->bytes_reserved
+ sinfo
->bytes_pinned
+
8848 sinfo
->bytes_may_use
+ sinfo
->bytes_readonly
+ num_bytes
+
8849 min_allocable_bytes
<= sinfo
->total_bytes
) {
8850 sinfo
->bytes_readonly
+= num_bytes
;
8852 list_add_tail(&cache
->ro_list
, &sinfo
->ro_bgs
);
8856 spin_unlock(&cache
->lock
);
8857 spin_unlock(&sinfo
->lock
);
8861 int btrfs_inc_block_group_ro(struct btrfs_root
*root
,
8862 struct btrfs_block_group_cache
*cache
)
8865 struct btrfs_trans_handle
*trans
;
8870 trans
= btrfs_join_transaction(root
);
8872 return PTR_ERR(trans
);
8875 * we're not allowed to set block groups readonly after the dirty
8876 * block groups cache has started writing. If it already started,
8877 * back off and let this transaction commit
8879 mutex_lock(&root
->fs_info
->ro_block_group_mutex
);
8880 if (trans
->transaction
->dirty_bg_run
) {
8881 u64 transid
= trans
->transid
;
8883 mutex_unlock(&root
->fs_info
->ro_block_group_mutex
);
8884 btrfs_end_transaction(trans
, root
);
8886 ret
= btrfs_wait_for_commit(root
, transid
);
8893 * if we are changing raid levels, try to allocate a corresponding
8894 * block group with the new raid level.
8896 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
8897 if (alloc_flags
!= cache
->flags
) {
8898 ret
= do_chunk_alloc(trans
, root
, alloc_flags
,
8901 * ENOSPC is allowed here, we may have enough space
8902 * already allocated at the new raid level to
8911 ret
= inc_block_group_ro(cache
, 0);
8914 alloc_flags
= get_alloc_profile(root
, cache
->space_info
->flags
);
8915 ret
= do_chunk_alloc(trans
, root
, alloc_flags
,
8919 ret
= inc_block_group_ro(cache
, 0);
8921 if (cache
->flags
& BTRFS_BLOCK_GROUP_SYSTEM
) {
8922 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
8923 lock_chunks(root
->fs_info
->chunk_root
);
8924 check_system_chunk(trans
, root
, alloc_flags
);
8925 unlock_chunks(root
->fs_info
->chunk_root
);
8927 mutex_unlock(&root
->fs_info
->ro_block_group_mutex
);
8929 btrfs_end_transaction(trans
, root
);
8933 int btrfs_force_chunk_alloc(struct btrfs_trans_handle
*trans
,
8934 struct btrfs_root
*root
, u64 type
)
8936 u64 alloc_flags
= get_alloc_profile(root
, type
);
8937 return do_chunk_alloc(trans
, root
, alloc_flags
,
8942 * helper to account the unused space of all the readonly block group in the
8943 * space_info. takes mirrors into account.
8945 u64
btrfs_account_ro_block_groups_free_space(struct btrfs_space_info
*sinfo
)
8947 struct btrfs_block_group_cache
*block_group
;
8951 /* It's df, we don't care if it's racey */
8952 if (list_empty(&sinfo
->ro_bgs
))
8955 spin_lock(&sinfo
->lock
);
8956 list_for_each_entry(block_group
, &sinfo
->ro_bgs
, ro_list
) {
8957 spin_lock(&block_group
->lock
);
8959 if (!block_group
->ro
) {
8960 spin_unlock(&block_group
->lock
);
8964 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_RAID1
|
8965 BTRFS_BLOCK_GROUP_RAID10
|
8966 BTRFS_BLOCK_GROUP_DUP
))
8971 free_bytes
+= (block_group
->key
.offset
-
8972 btrfs_block_group_used(&block_group
->item
)) *
8975 spin_unlock(&block_group
->lock
);
8977 spin_unlock(&sinfo
->lock
);
8982 void btrfs_dec_block_group_ro(struct btrfs_root
*root
,
8983 struct btrfs_block_group_cache
*cache
)
8985 struct btrfs_space_info
*sinfo
= cache
->space_info
;
8990 spin_lock(&sinfo
->lock
);
8991 spin_lock(&cache
->lock
);
8993 num_bytes
= cache
->key
.offset
- cache
->reserved
-
8994 cache
->pinned
- cache
->bytes_super
-
8995 btrfs_block_group_used(&cache
->item
);
8996 sinfo
->bytes_readonly
-= num_bytes
;
8997 list_del_init(&cache
->ro_list
);
8999 spin_unlock(&cache
->lock
);
9000 spin_unlock(&sinfo
->lock
);
9004 * checks to see if its even possible to relocate this block group.
9006 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9007 * ok to go ahead and try.
9009 int btrfs_can_relocate(struct btrfs_root
*root
, u64 bytenr
)
9011 struct btrfs_block_group_cache
*block_group
;
9012 struct btrfs_space_info
*space_info
;
9013 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
9014 struct btrfs_device
*device
;
9015 struct btrfs_trans_handle
*trans
;
9024 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
9026 /* odd, couldn't find the block group, leave it alone */
9030 min_free
= btrfs_block_group_used(&block_group
->item
);
9032 /* no bytes used, we're good */
9036 space_info
= block_group
->space_info
;
9037 spin_lock(&space_info
->lock
);
9039 full
= space_info
->full
;
9042 * if this is the last block group we have in this space, we can't
9043 * relocate it unless we're able to allocate a new chunk below.
9045 * Otherwise, we need to make sure we have room in the space to handle
9046 * all of the extents from this block group. If we can, we're good
9048 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
9049 (space_info
->bytes_used
+ space_info
->bytes_reserved
+
9050 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
9051 min_free
< space_info
->total_bytes
)) {
9052 spin_unlock(&space_info
->lock
);
9055 spin_unlock(&space_info
->lock
);
9058 * ok we don't have enough space, but maybe we have free space on our
9059 * devices to allocate new chunks for relocation, so loop through our
9060 * alloc devices and guess if we have enough space. if this block
9061 * group is going to be restriped, run checks against the target
9062 * profile instead of the current one.
9074 target
= get_restripe_target(root
->fs_info
, block_group
->flags
);
9076 index
= __get_raid_index(extended_to_chunk(target
));
9079 * this is just a balance, so if we were marked as full
9080 * we know there is no space for a new chunk
9085 index
= get_block_group_index(block_group
);
9088 if (index
== BTRFS_RAID_RAID10
) {
9092 } else if (index
== BTRFS_RAID_RAID1
) {
9094 } else if (index
== BTRFS_RAID_DUP
) {
9097 } else if (index
== BTRFS_RAID_RAID0
) {
9098 dev_min
= fs_devices
->rw_devices
;
9099 min_free
= div64_u64(min_free
, dev_min
);
9102 /* We need to do this so that we can look at pending chunks */
9103 trans
= btrfs_join_transaction(root
);
9104 if (IS_ERR(trans
)) {
9105 ret
= PTR_ERR(trans
);
9109 mutex_lock(&root
->fs_info
->chunk_mutex
);
9110 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
9114 * check to make sure we can actually find a chunk with enough
9115 * space to fit our block group in.
9117 if (device
->total_bytes
> device
->bytes_used
+ min_free
&&
9118 !device
->is_tgtdev_for_dev_replace
) {
9119 ret
= find_free_dev_extent(trans
, device
, min_free
,
9124 if (dev_nr
>= dev_min
)
9130 mutex_unlock(&root
->fs_info
->chunk_mutex
);
9131 btrfs_end_transaction(trans
, root
);
9133 btrfs_put_block_group(block_group
);
9137 static int find_first_block_group(struct btrfs_root
*root
,
9138 struct btrfs_path
*path
, struct btrfs_key
*key
)
9141 struct btrfs_key found_key
;
9142 struct extent_buffer
*leaf
;
9145 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
9150 slot
= path
->slots
[0];
9151 leaf
= path
->nodes
[0];
9152 if (slot
>= btrfs_header_nritems(leaf
)) {
9153 ret
= btrfs_next_leaf(root
, path
);
9160 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
9162 if (found_key
.objectid
>= key
->objectid
&&
9163 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
9173 void btrfs_put_block_group_cache(struct btrfs_fs_info
*info
)
9175 struct btrfs_block_group_cache
*block_group
;
9179 struct inode
*inode
;
9181 block_group
= btrfs_lookup_first_block_group(info
, last
);
9182 while (block_group
) {
9183 spin_lock(&block_group
->lock
);
9184 if (block_group
->iref
)
9186 spin_unlock(&block_group
->lock
);
9187 block_group
= next_block_group(info
->tree_root
,
9197 inode
= block_group
->inode
;
9198 block_group
->iref
= 0;
9199 block_group
->inode
= NULL
;
9200 spin_unlock(&block_group
->lock
);
9202 last
= block_group
->key
.objectid
+ block_group
->key
.offset
;
9203 btrfs_put_block_group(block_group
);
9207 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
9209 struct btrfs_block_group_cache
*block_group
;
9210 struct btrfs_space_info
*space_info
;
9211 struct btrfs_caching_control
*caching_ctl
;
9214 down_write(&info
->commit_root_sem
);
9215 while (!list_empty(&info
->caching_block_groups
)) {
9216 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
9217 struct btrfs_caching_control
, list
);
9218 list_del(&caching_ctl
->list
);
9219 put_caching_control(caching_ctl
);
9221 up_write(&info
->commit_root_sem
);
9223 spin_lock(&info
->unused_bgs_lock
);
9224 while (!list_empty(&info
->unused_bgs
)) {
9225 block_group
= list_first_entry(&info
->unused_bgs
,
9226 struct btrfs_block_group_cache
,
9228 list_del_init(&block_group
->bg_list
);
9229 btrfs_put_block_group(block_group
);
9231 spin_unlock(&info
->unused_bgs_lock
);
9233 spin_lock(&info
->block_group_cache_lock
);
9234 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
9235 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
9237 rb_erase(&block_group
->cache_node
,
9238 &info
->block_group_cache_tree
);
9239 RB_CLEAR_NODE(&block_group
->cache_node
);
9240 spin_unlock(&info
->block_group_cache_lock
);
9242 down_write(&block_group
->space_info
->groups_sem
);
9243 list_del(&block_group
->list
);
9244 up_write(&block_group
->space_info
->groups_sem
);
9246 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
9247 wait_block_group_cache_done(block_group
);
9250 * We haven't cached this block group, which means we could
9251 * possibly have excluded extents on this block group.
9253 if (block_group
->cached
== BTRFS_CACHE_NO
||
9254 block_group
->cached
== BTRFS_CACHE_ERROR
)
9255 free_excluded_extents(info
->extent_root
, block_group
);
9257 btrfs_remove_free_space_cache(block_group
);
9258 btrfs_put_block_group(block_group
);
9260 spin_lock(&info
->block_group_cache_lock
);
9262 spin_unlock(&info
->block_group_cache_lock
);
9264 /* now that all the block groups are freed, go through and
9265 * free all the space_info structs. This is only called during
9266 * the final stages of unmount, and so we know nobody is
9267 * using them. We call synchronize_rcu() once before we start,
9268 * just to be on the safe side.
9272 release_global_block_rsv(info
);
9274 while (!list_empty(&info
->space_info
)) {
9277 space_info
= list_entry(info
->space_info
.next
,
9278 struct btrfs_space_info
,
9280 if (btrfs_test_opt(info
->tree_root
, ENOSPC_DEBUG
)) {
9281 if (WARN_ON(space_info
->bytes_pinned
> 0 ||
9282 space_info
->bytes_reserved
> 0 ||
9283 space_info
->bytes_may_use
> 0)) {
9284 dump_space_info(space_info
, 0, 0);
9287 list_del(&space_info
->list
);
9288 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++) {
9289 struct kobject
*kobj
;
9290 kobj
= space_info
->block_group_kobjs
[i
];
9291 space_info
->block_group_kobjs
[i
] = NULL
;
9297 kobject_del(&space_info
->kobj
);
9298 kobject_put(&space_info
->kobj
);
9303 static void __link_block_group(struct btrfs_space_info
*space_info
,
9304 struct btrfs_block_group_cache
*cache
)
9306 int index
= get_block_group_index(cache
);
9309 down_write(&space_info
->groups_sem
);
9310 if (list_empty(&space_info
->block_groups
[index
]))
9312 list_add_tail(&cache
->list
, &space_info
->block_groups
[index
]);
9313 up_write(&space_info
->groups_sem
);
9316 struct raid_kobject
*rkobj
;
9319 rkobj
= kzalloc(sizeof(*rkobj
), GFP_NOFS
);
9322 rkobj
->raid_type
= index
;
9323 kobject_init(&rkobj
->kobj
, &btrfs_raid_ktype
);
9324 ret
= kobject_add(&rkobj
->kobj
, &space_info
->kobj
,
9325 "%s", get_raid_name(index
));
9327 kobject_put(&rkobj
->kobj
);
9330 space_info
->block_group_kobjs
[index
] = &rkobj
->kobj
;
9335 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
9338 static struct btrfs_block_group_cache
*
9339 btrfs_create_block_group_cache(struct btrfs_root
*root
, u64 start
, u64 size
)
9341 struct btrfs_block_group_cache
*cache
;
9343 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
9347 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
9349 if (!cache
->free_space_ctl
) {
9354 cache
->key
.objectid
= start
;
9355 cache
->key
.offset
= size
;
9356 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
9358 cache
->sectorsize
= root
->sectorsize
;
9359 cache
->fs_info
= root
->fs_info
;
9360 cache
->full_stripe_len
= btrfs_full_stripe_len(root
,
9361 &root
->fs_info
->mapping_tree
,
9363 atomic_set(&cache
->count
, 1);
9364 spin_lock_init(&cache
->lock
);
9365 init_rwsem(&cache
->data_rwsem
);
9366 INIT_LIST_HEAD(&cache
->list
);
9367 INIT_LIST_HEAD(&cache
->cluster_list
);
9368 INIT_LIST_HEAD(&cache
->bg_list
);
9369 INIT_LIST_HEAD(&cache
->ro_list
);
9370 INIT_LIST_HEAD(&cache
->dirty_list
);
9371 INIT_LIST_HEAD(&cache
->io_list
);
9372 btrfs_init_free_space_ctl(cache
);
9373 atomic_set(&cache
->trimming
, 0);
9378 int btrfs_read_block_groups(struct btrfs_root
*root
)
9380 struct btrfs_path
*path
;
9382 struct btrfs_block_group_cache
*cache
;
9383 struct btrfs_fs_info
*info
= root
->fs_info
;
9384 struct btrfs_space_info
*space_info
;
9385 struct btrfs_key key
;
9386 struct btrfs_key found_key
;
9387 struct extent_buffer
*leaf
;
9391 root
= info
->extent_root
;
9394 key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
9395 path
= btrfs_alloc_path();
9400 cache_gen
= btrfs_super_cache_generation(root
->fs_info
->super_copy
);
9401 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
9402 btrfs_super_generation(root
->fs_info
->super_copy
) != cache_gen
)
9404 if (btrfs_test_opt(root
, CLEAR_CACHE
))
9408 ret
= find_first_block_group(root
, path
, &key
);
9414 leaf
= path
->nodes
[0];
9415 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
9417 cache
= btrfs_create_block_group_cache(root
, found_key
.objectid
,
9426 * When we mount with old space cache, we need to
9427 * set BTRFS_DC_CLEAR and set dirty flag.
9429 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9430 * truncate the old free space cache inode and
9432 * b) Setting 'dirty flag' makes sure that we flush
9433 * the new space cache info onto disk.
9435 if (btrfs_test_opt(root
, SPACE_CACHE
))
9436 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
9439 read_extent_buffer(leaf
, &cache
->item
,
9440 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
9441 sizeof(cache
->item
));
9442 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
9444 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
9445 btrfs_release_path(path
);
9448 * We need to exclude the super stripes now so that the space
9449 * info has super bytes accounted for, otherwise we'll think
9450 * we have more space than we actually do.
9452 ret
= exclude_super_stripes(root
, cache
);
9455 * We may have excluded something, so call this just in
9458 free_excluded_extents(root
, cache
);
9459 btrfs_put_block_group(cache
);
9464 * check for two cases, either we are full, and therefore
9465 * don't need to bother with the caching work since we won't
9466 * find any space, or we are empty, and we can just add all
9467 * the space in and be done with it. This saves us _alot_ of
9468 * time, particularly in the full case.
9470 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
9471 cache
->last_byte_to_unpin
= (u64
)-1;
9472 cache
->cached
= BTRFS_CACHE_FINISHED
;
9473 free_excluded_extents(root
, cache
);
9474 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
9475 cache
->last_byte_to_unpin
= (u64
)-1;
9476 cache
->cached
= BTRFS_CACHE_FINISHED
;
9477 add_new_free_space(cache
, root
->fs_info
,
9479 found_key
.objectid
+
9481 free_excluded_extents(root
, cache
);
9484 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
9486 btrfs_remove_free_space_cache(cache
);
9487 btrfs_put_block_group(cache
);
9491 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
9492 btrfs_block_group_used(&cache
->item
),
9495 btrfs_remove_free_space_cache(cache
);
9496 spin_lock(&info
->block_group_cache_lock
);
9497 rb_erase(&cache
->cache_node
,
9498 &info
->block_group_cache_tree
);
9499 RB_CLEAR_NODE(&cache
->cache_node
);
9500 spin_unlock(&info
->block_group_cache_lock
);
9501 btrfs_put_block_group(cache
);
9505 cache
->space_info
= space_info
;
9506 spin_lock(&cache
->space_info
->lock
);
9507 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
9508 spin_unlock(&cache
->space_info
->lock
);
9510 __link_block_group(space_info
, cache
);
9512 set_avail_alloc_bits(root
->fs_info
, cache
->flags
);
9513 if (btrfs_chunk_readonly(root
, cache
->key
.objectid
)) {
9514 inc_block_group_ro(cache
, 1);
9515 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
9516 spin_lock(&info
->unused_bgs_lock
);
9517 /* Should always be true but just in case. */
9518 if (list_empty(&cache
->bg_list
)) {
9519 btrfs_get_block_group(cache
);
9520 list_add_tail(&cache
->bg_list
,
9523 spin_unlock(&info
->unused_bgs_lock
);
9527 list_for_each_entry_rcu(space_info
, &root
->fs_info
->space_info
, list
) {
9528 if (!(get_alloc_profile(root
, space_info
->flags
) &
9529 (BTRFS_BLOCK_GROUP_RAID10
|
9530 BTRFS_BLOCK_GROUP_RAID1
|
9531 BTRFS_BLOCK_GROUP_RAID5
|
9532 BTRFS_BLOCK_GROUP_RAID6
|
9533 BTRFS_BLOCK_GROUP_DUP
)))
9536 * avoid allocating from un-mirrored block group if there are
9537 * mirrored block groups.
9539 list_for_each_entry(cache
,
9540 &space_info
->block_groups
[BTRFS_RAID_RAID0
],
9542 inc_block_group_ro(cache
, 1);
9543 list_for_each_entry(cache
,
9544 &space_info
->block_groups
[BTRFS_RAID_SINGLE
],
9546 inc_block_group_ro(cache
, 1);
9549 init_global_block_rsv(info
);
9552 btrfs_free_path(path
);
9556 void btrfs_create_pending_block_groups(struct btrfs_trans_handle
*trans
,
9557 struct btrfs_root
*root
)
9559 struct btrfs_block_group_cache
*block_group
, *tmp
;
9560 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
9561 struct btrfs_block_group_item item
;
9562 struct btrfs_key key
;
9564 bool can_flush_pending_bgs
= trans
->can_flush_pending_bgs
;
9566 trans
->can_flush_pending_bgs
= false;
9567 list_for_each_entry_safe(block_group
, tmp
, &trans
->new_bgs
, bg_list
) {
9571 spin_lock(&block_group
->lock
);
9572 memcpy(&item
, &block_group
->item
, sizeof(item
));
9573 memcpy(&key
, &block_group
->key
, sizeof(key
));
9574 spin_unlock(&block_group
->lock
);
9576 ret
= btrfs_insert_item(trans
, extent_root
, &key
, &item
,
9579 btrfs_abort_transaction(trans
, extent_root
, ret
);
9580 ret
= btrfs_finish_chunk_alloc(trans
, extent_root
,
9581 key
.objectid
, key
.offset
);
9583 btrfs_abort_transaction(trans
, extent_root
, ret
);
9585 list_del_init(&block_group
->bg_list
);
9587 trans
->can_flush_pending_bgs
= can_flush_pending_bgs
;
9590 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
9591 struct btrfs_root
*root
, u64 bytes_used
,
9592 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
9596 struct btrfs_root
*extent_root
;
9597 struct btrfs_block_group_cache
*cache
;
9599 extent_root
= root
->fs_info
->extent_root
;
9601 btrfs_set_log_full_commit(root
->fs_info
, trans
);
9603 cache
= btrfs_create_block_group_cache(root
, chunk_offset
, size
);
9607 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
9608 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
9609 btrfs_set_block_group_flags(&cache
->item
, type
);
9611 cache
->flags
= type
;
9612 cache
->last_byte_to_unpin
= (u64
)-1;
9613 cache
->cached
= BTRFS_CACHE_FINISHED
;
9614 ret
= exclude_super_stripes(root
, cache
);
9617 * We may have excluded something, so call this just in
9620 free_excluded_extents(root
, cache
);
9621 btrfs_put_block_group(cache
);
9625 add_new_free_space(cache
, root
->fs_info
, chunk_offset
,
9626 chunk_offset
+ size
);
9628 free_excluded_extents(root
, cache
);
9631 * Call to ensure the corresponding space_info object is created and
9632 * assigned to our block group, but don't update its counters just yet.
9633 * We want our bg to be added to the rbtree with its ->space_info set.
9635 ret
= update_space_info(root
->fs_info
, cache
->flags
, 0, 0,
9636 &cache
->space_info
);
9638 btrfs_remove_free_space_cache(cache
);
9639 btrfs_put_block_group(cache
);
9643 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
9645 btrfs_remove_free_space_cache(cache
);
9646 btrfs_put_block_group(cache
);
9651 * Now that our block group has its ->space_info set and is inserted in
9652 * the rbtree, update the space info's counters.
9654 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
9655 &cache
->space_info
);
9657 btrfs_remove_free_space_cache(cache
);
9658 spin_lock(&root
->fs_info
->block_group_cache_lock
);
9659 rb_erase(&cache
->cache_node
,
9660 &root
->fs_info
->block_group_cache_tree
);
9661 RB_CLEAR_NODE(&cache
->cache_node
);
9662 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
9663 btrfs_put_block_group(cache
);
9666 update_global_block_rsv(root
->fs_info
);
9668 spin_lock(&cache
->space_info
->lock
);
9669 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
9670 spin_unlock(&cache
->space_info
->lock
);
9672 __link_block_group(cache
->space_info
, cache
);
9674 list_add_tail(&cache
->bg_list
, &trans
->new_bgs
);
9676 set_avail_alloc_bits(extent_root
->fs_info
, type
);
9681 static void clear_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
9683 u64 extra_flags
= chunk_to_extended(flags
) &
9684 BTRFS_EXTENDED_PROFILE_MASK
;
9686 write_seqlock(&fs_info
->profiles_lock
);
9687 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
9688 fs_info
->avail_data_alloc_bits
&= ~extra_flags
;
9689 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
9690 fs_info
->avail_metadata_alloc_bits
&= ~extra_flags
;
9691 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
9692 fs_info
->avail_system_alloc_bits
&= ~extra_flags
;
9693 write_sequnlock(&fs_info
->profiles_lock
);
9696 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
9697 struct btrfs_root
*root
, u64 group_start
,
9698 struct extent_map
*em
)
9700 struct btrfs_path
*path
;
9701 struct btrfs_block_group_cache
*block_group
;
9702 struct btrfs_free_cluster
*cluster
;
9703 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
9704 struct btrfs_key key
;
9705 struct inode
*inode
;
9706 struct kobject
*kobj
= NULL
;
9710 struct btrfs_caching_control
*caching_ctl
= NULL
;
9713 root
= root
->fs_info
->extent_root
;
9715 block_group
= btrfs_lookup_block_group(root
->fs_info
, group_start
);
9716 BUG_ON(!block_group
);
9717 BUG_ON(!block_group
->ro
);
9720 * Free the reserved super bytes from this block group before
9723 free_excluded_extents(root
, block_group
);
9725 memcpy(&key
, &block_group
->key
, sizeof(key
));
9726 index
= get_block_group_index(block_group
);
9727 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
9728 BTRFS_BLOCK_GROUP_RAID1
|
9729 BTRFS_BLOCK_GROUP_RAID10
))
9734 /* make sure this block group isn't part of an allocation cluster */
9735 cluster
= &root
->fs_info
->data_alloc_cluster
;
9736 spin_lock(&cluster
->refill_lock
);
9737 btrfs_return_cluster_to_free_space(block_group
, cluster
);
9738 spin_unlock(&cluster
->refill_lock
);
9741 * make sure this block group isn't part of a metadata
9742 * allocation cluster
9744 cluster
= &root
->fs_info
->meta_alloc_cluster
;
9745 spin_lock(&cluster
->refill_lock
);
9746 btrfs_return_cluster_to_free_space(block_group
, cluster
);
9747 spin_unlock(&cluster
->refill_lock
);
9749 path
= btrfs_alloc_path();
9756 * get the inode first so any iput calls done for the io_list
9757 * aren't the final iput (no unlinks allowed now)
9759 inode
= lookup_free_space_inode(tree_root
, block_group
, path
);
9761 mutex_lock(&trans
->transaction
->cache_write_mutex
);
9763 * make sure our free spache cache IO is done before remove the
9766 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
9767 if (!list_empty(&block_group
->io_list
)) {
9768 list_del_init(&block_group
->io_list
);
9770 WARN_ON(!IS_ERR(inode
) && inode
!= block_group
->io_ctl
.inode
);
9772 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
9773 btrfs_wait_cache_io(root
, trans
, block_group
,
9774 &block_group
->io_ctl
, path
,
9775 block_group
->key
.objectid
);
9776 btrfs_put_block_group(block_group
);
9777 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
9780 if (!list_empty(&block_group
->dirty_list
)) {
9781 list_del_init(&block_group
->dirty_list
);
9782 btrfs_put_block_group(block_group
);
9784 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
9785 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
9787 if (!IS_ERR(inode
)) {
9788 ret
= btrfs_orphan_add(trans
, inode
);
9790 btrfs_add_delayed_iput(inode
);
9794 /* One for the block groups ref */
9795 spin_lock(&block_group
->lock
);
9796 if (block_group
->iref
) {
9797 block_group
->iref
= 0;
9798 block_group
->inode
= NULL
;
9799 spin_unlock(&block_group
->lock
);
9802 spin_unlock(&block_group
->lock
);
9804 /* One for our lookup ref */
9805 btrfs_add_delayed_iput(inode
);
9808 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
9809 key
.offset
= block_group
->key
.objectid
;
9812 ret
= btrfs_search_slot(trans
, tree_root
, &key
, path
, -1, 1);
9816 btrfs_release_path(path
);
9818 ret
= btrfs_del_item(trans
, tree_root
, path
);
9821 btrfs_release_path(path
);
9824 spin_lock(&root
->fs_info
->block_group_cache_lock
);
9825 rb_erase(&block_group
->cache_node
,
9826 &root
->fs_info
->block_group_cache_tree
);
9827 RB_CLEAR_NODE(&block_group
->cache_node
);
9829 if (root
->fs_info
->first_logical_byte
== block_group
->key
.objectid
)
9830 root
->fs_info
->first_logical_byte
= (u64
)-1;
9831 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
9833 down_write(&block_group
->space_info
->groups_sem
);
9835 * we must use list_del_init so people can check to see if they
9836 * are still on the list after taking the semaphore
9838 list_del_init(&block_group
->list
);
9839 if (list_empty(&block_group
->space_info
->block_groups
[index
])) {
9840 kobj
= block_group
->space_info
->block_group_kobjs
[index
];
9841 block_group
->space_info
->block_group_kobjs
[index
] = NULL
;
9842 clear_avail_alloc_bits(root
->fs_info
, block_group
->flags
);
9844 up_write(&block_group
->space_info
->groups_sem
);
9850 if (block_group
->has_caching_ctl
)
9851 caching_ctl
= get_caching_control(block_group
);
9852 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
9853 wait_block_group_cache_done(block_group
);
9854 if (block_group
->has_caching_ctl
) {
9855 down_write(&root
->fs_info
->commit_root_sem
);
9857 struct btrfs_caching_control
*ctl
;
9859 list_for_each_entry(ctl
,
9860 &root
->fs_info
->caching_block_groups
, list
)
9861 if (ctl
->block_group
== block_group
) {
9863 atomic_inc(&caching_ctl
->count
);
9868 list_del_init(&caching_ctl
->list
);
9869 up_write(&root
->fs_info
->commit_root_sem
);
9871 /* Once for the caching bgs list and once for us. */
9872 put_caching_control(caching_ctl
);
9873 put_caching_control(caching_ctl
);
9877 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
9878 if (!list_empty(&block_group
->dirty_list
)) {
9881 if (!list_empty(&block_group
->io_list
)) {
9884 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
9885 btrfs_remove_free_space_cache(block_group
);
9887 spin_lock(&block_group
->space_info
->lock
);
9888 list_del_init(&block_group
->ro_list
);
9890 if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
9891 WARN_ON(block_group
->space_info
->total_bytes
9892 < block_group
->key
.offset
);
9893 WARN_ON(block_group
->space_info
->bytes_readonly
9894 < block_group
->key
.offset
);
9895 WARN_ON(block_group
->space_info
->disk_total
9896 < block_group
->key
.offset
* factor
);
9898 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
9899 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
9900 block_group
->space_info
->disk_total
-= block_group
->key
.offset
* factor
;
9902 spin_unlock(&block_group
->space_info
->lock
);
9904 memcpy(&key
, &block_group
->key
, sizeof(key
));
9907 if (!list_empty(&em
->list
)) {
9908 /* We're in the transaction->pending_chunks list. */
9909 free_extent_map(em
);
9911 spin_lock(&block_group
->lock
);
9912 block_group
->removed
= 1;
9914 * At this point trimming can't start on this block group, because we
9915 * removed the block group from the tree fs_info->block_group_cache_tree
9916 * so no one can't find it anymore and even if someone already got this
9917 * block group before we removed it from the rbtree, they have already
9918 * incremented block_group->trimming - if they didn't, they won't find
9919 * any free space entries because we already removed them all when we
9920 * called btrfs_remove_free_space_cache().
9922 * And we must not remove the extent map from the fs_info->mapping_tree
9923 * to prevent the same logical address range and physical device space
9924 * ranges from being reused for a new block group. This is because our
9925 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
9926 * completely transactionless, so while it is trimming a range the
9927 * currently running transaction might finish and a new one start,
9928 * allowing for new block groups to be created that can reuse the same
9929 * physical device locations unless we take this special care.
9931 * There may also be an implicit trim operation if the file system
9932 * is mounted with -odiscard. The same protections must remain
9933 * in place until the extents have been discarded completely when
9934 * the transaction commit has completed.
9936 remove_em
= (atomic_read(&block_group
->trimming
) == 0);
9938 * Make sure a trimmer task always sees the em in the pinned_chunks list
9939 * if it sees block_group->removed == 1 (needs to lock block_group->lock
9940 * before checking block_group->removed).
9944 * Our em might be in trans->transaction->pending_chunks which
9945 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
9946 * and so is the fs_info->pinned_chunks list.
9948 * So at this point we must be holding the chunk_mutex to avoid
9949 * any races with chunk allocation (more specifically at
9950 * volumes.c:contains_pending_extent()), to ensure it always
9951 * sees the em, either in the pending_chunks list or in the
9952 * pinned_chunks list.
9954 list_move_tail(&em
->list
, &root
->fs_info
->pinned_chunks
);
9956 spin_unlock(&block_group
->lock
);
9959 struct extent_map_tree
*em_tree
;
9961 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
9962 write_lock(&em_tree
->lock
);
9964 * The em might be in the pending_chunks list, so make sure the
9965 * chunk mutex is locked, since remove_extent_mapping() will
9966 * delete us from that list.
9968 remove_extent_mapping(em_tree
, em
);
9969 write_unlock(&em_tree
->lock
);
9970 /* once for the tree */
9971 free_extent_map(em
);
9974 unlock_chunks(root
);
9976 btrfs_put_block_group(block_group
);
9977 btrfs_put_block_group(block_group
);
9979 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
9985 ret
= btrfs_del_item(trans
, root
, path
);
9987 btrfs_free_path(path
);
9992 * Process the unused_bgs list and remove any that don't have any allocated
9993 * space inside of them.
9995 void btrfs_delete_unused_bgs(struct btrfs_fs_info
*fs_info
)
9997 struct btrfs_block_group_cache
*block_group
;
9998 struct btrfs_space_info
*space_info
;
9999 struct btrfs_root
*root
= fs_info
->extent_root
;
10000 struct btrfs_trans_handle
*trans
;
10003 if (!fs_info
->open
)
10006 spin_lock(&fs_info
->unused_bgs_lock
);
10007 while (!list_empty(&fs_info
->unused_bgs
)) {
10011 block_group
= list_first_entry(&fs_info
->unused_bgs
,
10012 struct btrfs_block_group_cache
,
10014 space_info
= block_group
->space_info
;
10015 list_del_init(&block_group
->bg_list
);
10016 if (ret
|| btrfs_mixed_space_info(space_info
)) {
10017 btrfs_put_block_group(block_group
);
10020 spin_unlock(&fs_info
->unused_bgs_lock
);
10022 mutex_lock(&root
->fs_info
->delete_unused_bgs_mutex
);
10024 /* Don't want to race with allocators so take the groups_sem */
10025 down_write(&space_info
->groups_sem
);
10026 spin_lock(&block_group
->lock
);
10027 if (block_group
->reserved
||
10028 btrfs_block_group_used(&block_group
->item
) ||
10031 * We want to bail if we made new allocations or have
10032 * outstanding allocations in this block group. We do
10033 * the ro check in case balance is currently acting on
10034 * this block group.
10036 spin_unlock(&block_group
->lock
);
10037 up_write(&space_info
->groups_sem
);
10040 spin_unlock(&block_group
->lock
);
10042 /* We don't want to force the issue, only flip if it's ok. */
10043 ret
= inc_block_group_ro(block_group
, 0);
10044 up_write(&space_info
->groups_sem
);
10051 * Want to do this before we do anything else so we can recover
10052 * properly if we fail to join the transaction.
10054 /* 1 for btrfs_orphan_reserve_metadata() */
10055 trans
= btrfs_start_transaction(root
, 1);
10056 if (IS_ERR(trans
)) {
10057 btrfs_dec_block_group_ro(root
, block_group
);
10058 ret
= PTR_ERR(trans
);
10063 * We could have pending pinned extents for this block group,
10064 * just delete them, we don't care about them anymore.
10066 start
= block_group
->key
.objectid
;
10067 end
= start
+ block_group
->key
.offset
- 1;
10069 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10070 * btrfs_finish_extent_commit(). If we are at transaction N,
10071 * another task might be running finish_extent_commit() for the
10072 * previous transaction N - 1, and have seen a range belonging
10073 * to the block group in freed_extents[] before we were able to
10074 * clear the whole block group range from freed_extents[]. This
10075 * means that task can lookup for the block group after we
10076 * unpinned it from freed_extents[] and removed it, leading to
10077 * a BUG_ON() at btrfs_unpin_extent_range().
10079 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
10080 ret
= clear_extent_bits(&fs_info
->freed_extents
[0], start
, end
,
10081 EXTENT_DIRTY
, GFP_NOFS
);
10083 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10084 btrfs_dec_block_group_ro(root
, block_group
);
10087 ret
= clear_extent_bits(&fs_info
->freed_extents
[1], start
, end
,
10088 EXTENT_DIRTY
, GFP_NOFS
);
10090 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10091 btrfs_dec_block_group_ro(root
, block_group
);
10094 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10096 /* Reset pinned so btrfs_put_block_group doesn't complain */
10097 spin_lock(&space_info
->lock
);
10098 spin_lock(&block_group
->lock
);
10100 space_info
->bytes_pinned
-= block_group
->pinned
;
10101 space_info
->bytes_readonly
+= block_group
->pinned
;
10102 percpu_counter_add(&space_info
->total_bytes_pinned
,
10103 -block_group
->pinned
);
10104 block_group
->pinned
= 0;
10106 spin_unlock(&block_group
->lock
);
10107 spin_unlock(&space_info
->lock
);
10109 /* DISCARD can flip during remount */
10110 trimming
= btrfs_test_opt(root
, DISCARD
);
10112 /* Implicit trim during transaction commit. */
10114 btrfs_get_block_group_trimming(block_group
);
10117 * Btrfs_remove_chunk will abort the transaction if things go
10120 ret
= btrfs_remove_chunk(trans
, root
,
10121 block_group
->key
.objectid
);
10125 btrfs_put_block_group_trimming(block_group
);
10130 * If we're not mounted with -odiscard, we can just forget
10131 * about this block group. Otherwise we'll need to wait
10132 * until transaction commit to do the actual discard.
10135 WARN_ON(!list_empty(&block_group
->bg_list
));
10136 spin_lock(&trans
->transaction
->deleted_bgs_lock
);
10137 list_move(&block_group
->bg_list
,
10138 &trans
->transaction
->deleted_bgs
);
10139 spin_unlock(&trans
->transaction
->deleted_bgs_lock
);
10140 btrfs_get_block_group(block_group
);
10143 btrfs_end_transaction(trans
, root
);
10145 mutex_unlock(&root
->fs_info
->delete_unused_bgs_mutex
);
10146 btrfs_put_block_group(block_group
);
10147 spin_lock(&fs_info
->unused_bgs_lock
);
10149 spin_unlock(&fs_info
->unused_bgs_lock
);
10152 int btrfs_init_space_info(struct btrfs_fs_info
*fs_info
)
10154 struct btrfs_space_info
*space_info
;
10155 struct btrfs_super_block
*disk_super
;
10161 disk_super
= fs_info
->super_copy
;
10162 if (!btrfs_super_root(disk_super
))
10165 features
= btrfs_super_incompat_flags(disk_super
);
10166 if (features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
10169 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
10170 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
10175 flags
= BTRFS_BLOCK_GROUP_METADATA
| BTRFS_BLOCK_GROUP_DATA
;
10176 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
10178 flags
= BTRFS_BLOCK_GROUP_METADATA
;
10179 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
10183 flags
= BTRFS_BLOCK_GROUP_DATA
;
10184 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
10190 int btrfs_error_unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
10192 return unpin_extent_range(root
, start
, end
, false);
10196 * It used to be that old block groups would be left around forever.
10197 * Iterating over them would be enough to trim unused space. Since we
10198 * now automatically remove them, we also need to iterate over unallocated
10201 * We don't want a transaction for this since the discard may take a
10202 * substantial amount of time. We don't require that a transaction be
10203 * running, but we do need to take a running transaction into account
10204 * to ensure that we're not discarding chunks that were released in
10205 * the current transaction.
10207 * Holding the chunks lock will prevent other threads from allocating
10208 * or releasing chunks, but it won't prevent a running transaction
10209 * from committing and releasing the memory that the pending chunks
10210 * list head uses. For that, we need to take a reference to the
10213 static int btrfs_trim_free_extents(struct btrfs_device
*device
,
10214 u64 minlen
, u64
*trimmed
)
10216 u64 start
= 0, len
= 0;
10221 /* Not writeable = nothing to do. */
10222 if (!device
->writeable
)
10225 /* No free space = nothing to do. */
10226 if (device
->total_bytes
<= device
->bytes_used
)
10232 struct btrfs_fs_info
*fs_info
= device
->dev_root
->fs_info
;
10233 struct btrfs_transaction
*trans
;
10236 ret
= mutex_lock_interruptible(&fs_info
->chunk_mutex
);
10240 down_read(&fs_info
->commit_root_sem
);
10242 spin_lock(&fs_info
->trans_lock
);
10243 trans
= fs_info
->running_transaction
;
10245 atomic_inc(&trans
->use_count
);
10246 spin_unlock(&fs_info
->trans_lock
);
10248 ret
= find_free_dev_extent_start(trans
, device
, minlen
, start
,
10251 btrfs_put_transaction(trans
);
10254 up_read(&fs_info
->commit_root_sem
);
10255 mutex_unlock(&fs_info
->chunk_mutex
);
10256 if (ret
== -ENOSPC
)
10261 ret
= btrfs_issue_discard(device
->bdev
, start
, len
, &bytes
);
10262 up_read(&fs_info
->commit_root_sem
);
10263 mutex_unlock(&fs_info
->chunk_mutex
);
10271 if (fatal_signal_pending(current
)) {
10272 ret
= -ERESTARTSYS
;
10282 int btrfs_trim_fs(struct btrfs_root
*root
, struct fstrim_range
*range
)
10284 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
10285 struct btrfs_block_group_cache
*cache
= NULL
;
10286 struct btrfs_device
*device
;
10287 struct list_head
*devices
;
10292 u64 total_bytes
= btrfs_super_total_bytes(fs_info
->super_copy
);
10296 * try to trim all FS space, our block group may start from non-zero.
10298 if (range
->len
== total_bytes
)
10299 cache
= btrfs_lookup_first_block_group(fs_info
, range
->start
);
10301 cache
= btrfs_lookup_block_group(fs_info
, range
->start
);
10304 if (cache
->key
.objectid
>= (range
->start
+ range
->len
)) {
10305 btrfs_put_block_group(cache
);
10309 start
= max(range
->start
, cache
->key
.objectid
);
10310 end
= min(range
->start
+ range
->len
,
10311 cache
->key
.objectid
+ cache
->key
.offset
);
10313 if (end
- start
>= range
->minlen
) {
10314 if (!block_group_cache_done(cache
)) {
10315 ret
= cache_block_group(cache
, 0);
10317 btrfs_put_block_group(cache
);
10320 ret
= wait_block_group_cache_done(cache
);
10322 btrfs_put_block_group(cache
);
10326 ret
= btrfs_trim_block_group(cache
,
10332 trimmed
+= group_trimmed
;
10334 btrfs_put_block_group(cache
);
10339 cache
= next_block_group(fs_info
->tree_root
, cache
);
10342 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
10343 devices
= &root
->fs_info
->fs_devices
->alloc_list
;
10344 list_for_each_entry(device
, devices
, dev_alloc_list
) {
10345 ret
= btrfs_trim_free_extents(device
, range
->minlen
,
10350 trimmed
+= group_trimmed
;
10352 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
10354 range
->len
= trimmed
;
10359 * btrfs_{start,end}_write_no_snapshoting() are similar to
10360 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10361 * data into the page cache through nocow before the subvolume is snapshoted,
10362 * but flush the data into disk after the snapshot creation, or to prevent
10363 * operations while snapshoting is ongoing and that cause the snapshot to be
10364 * inconsistent (writes followed by expanding truncates for example).
10366 void btrfs_end_write_no_snapshoting(struct btrfs_root
*root
)
10368 percpu_counter_dec(&root
->subv_writers
->counter
);
10370 * Make sure counter is updated before we wake up waiters.
10373 if (waitqueue_active(&root
->subv_writers
->wait
))
10374 wake_up(&root
->subv_writers
->wait
);
10377 int btrfs_start_write_no_snapshoting(struct btrfs_root
*root
)
10379 if (atomic_read(&root
->will_be_snapshoted
))
10382 percpu_counter_inc(&root
->subv_writers
->counter
);
10384 * Make sure counter is updated before we check for snapshot creation.
10387 if (atomic_read(&root
->will_be_snapshoted
)) {
10388 btrfs_end_write_no_snapshoting(root
);