2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include <linux/slab.h>
26 #include <linux/ratelimit.h>
27 #include <linux/percpu_counter.h>
31 #include "print-tree.h"
35 #include "free-space-cache.h"
40 #undef SCRAMBLE_DELAYED_REFS
43 * control flags for do_chunk_alloc's force field
44 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
45 * if we really need one.
47 * CHUNK_ALLOC_LIMITED means to only try and allocate one
48 * if we have very few chunks already allocated. This is
49 * used as part of the clustering code to help make sure
50 * we have a good pool of storage to cluster in, without
51 * filling the FS with empty chunks
53 * CHUNK_ALLOC_FORCE means it must try to allocate one
57 CHUNK_ALLOC_NO_FORCE
= 0,
58 CHUNK_ALLOC_LIMITED
= 1,
59 CHUNK_ALLOC_FORCE
= 2,
63 * Control how reservations are dealt with.
65 * RESERVE_FREE - freeing a reservation.
66 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
68 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
69 * bytes_may_use as the ENOSPC accounting is done elsewhere
74 RESERVE_ALLOC_NO_ACCOUNT
= 2,
77 static int update_block_group(struct btrfs_trans_handle
*trans
,
78 struct btrfs_root
*root
, u64 bytenr
,
79 u64 num_bytes
, int alloc
);
80 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
81 struct btrfs_root
*root
,
82 struct btrfs_delayed_ref_node
*node
, u64 parent
,
83 u64 root_objectid
, u64 owner_objectid
,
84 u64 owner_offset
, int refs_to_drop
,
85 struct btrfs_delayed_extent_op
*extra_op
);
86 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
87 struct extent_buffer
*leaf
,
88 struct btrfs_extent_item
*ei
);
89 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
90 struct btrfs_root
*root
,
91 u64 parent
, u64 root_objectid
,
92 u64 flags
, u64 owner
, u64 offset
,
93 struct btrfs_key
*ins
, int ref_mod
);
94 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
95 struct btrfs_root
*root
,
96 u64 parent
, u64 root_objectid
,
97 u64 flags
, struct btrfs_disk_key
*key
,
98 int level
, struct btrfs_key
*ins
,
100 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
101 struct btrfs_root
*extent_root
, u64 flags
,
103 static int find_next_key(struct btrfs_path
*path
, int level
,
104 struct btrfs_key
*key
);
105 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
106 int dump_block_groups
);
107 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
108 u64 num_bytes
, int reserve
,
110 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
112 int btrfs_pin_extent(struct btrfs_root
*root
,
113 u64 bytenr
, u64 num_bytes
, int reserved
);
116 block_group_cache_done(struct btrfs_block_group_cache
*cache
)
119 return cache
->cached
== BTRFS_CACHE_FINISHED
||
120 cache
->cached
== BTRFS_CACHE_ERROR
;
123 static int block_group_bits(struct btrfs_block_group_cache
*cache
, u64 bits
)
125 return (cache
->flags
& bits
) == bits
;
128 static void btrfs_get_block_group(struct btrfs_block_group_cache
*cache
)
130 atomic_inc(&cache
->count
);
133 void btrfs_put_block_group(struct btrfs_block_group_cache
*cache
)
135 if (atomic_dec_and_test(&cache
->count
)) {
136 WARN_ON(cache
->pinned
> 0);
137 WARN_ON(cache
->reserved
> 0);
138 kfree(cache
->free_space_ctl
);
144 * this adds the block group to the fs_info rb tree for the block group
147 static int btrfs_add_block_group_cache(struct btrfs_fs_info
*info
,
148 struct btrfs_block_group_cache
*block_group
)
151 struct rb_node
*parent
= NULL
;
152 struct btrfs_block_group_cache
*cache
;
154 spin_lock(&info
->block_group_cache_lock
);
155 p
= &info
->block_group_cache_tree
.rb_node
;
159 cache
= rb_entry(parent
, struct btrfs_block_group_cache
,
161 if (block_group
->key
.objectid
< cache
->key
.objectid
) {
163 } else if (block_group
->key
.objectid
> cache
->key
.objectid
) {
166 spin_unlock(&info
->block_group_cache_lock
);
171 rb_link_node(&block_group
->cache_node
, parent
, p
);
172 rb_insert_color(&block_group
->cache_node
,
173 &info
->block_group_cache_tree
);
175 if (info
->first_logical_byte
> block_group
->key
.objectid
)
176 info
->first_logical_byte
= block_group
->key
.objectid
;
178 spin_unlock(&info
->block_group_cache_lock
);
184 * This will return the block group at or after bytenr if contains is 0, else
185 * it will return the block group that contains the bytenr
187 static struct btrfs_block_group_cache
*
188 block_group_cache_tree_search(struct btrfs_fs_info
*info
, u64 bytenr
,
191 struct btrfs_block_group_cache
*cache
, *ret
= NULL
;
195 spin_lock(&info
->block_group_cache_lock
);
196 n
= info
->block_group_cache_tree
.rb_node
;
199 cache
= rb_entry(n
, struct btrfs_block_group_cache
,
201 end
= cache
->key
.objectid
+ cache
->key
.offset
- 1;
202 start
= cache
->key
.objectid
;
204 if (bytenr
< start
) {
205 if (!contains
&& (!ret
|| start
< ret
->key
.objectid
))
208 } else if (bytenr
> start
) {
209 if (contains
&& bytenr
<= end
) {
220 btrfs_get_block_group(ret
);
221 if (bytenr
== 0 && info
->first_logical_byte
> ret
->key
.objectid
)
222 info
->first_logical_byte
= ret
->key
.objectid
;
224 spin_unlock(&info
->block_group_cache_lock
);
229 static int add_excluded_extent(struct btrfs_root
*root
,
230 u64 start
, u64 num_bytes
)
232 u64 end
= start
+ num_bytes
- 1;
233 set_extent_bits(&root
->fs_info
->freed_extents
[0],
234 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
235 set_extent_bits(&root
->fs_info
->freed_extents
[1],
236 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
240 static void free_excluded_extents(struct btrfs_root
*root
,
241 struct btrfs_block_group_cache
*cache
)
245 start
= cache
->key
.objectid
;
246 end
= start
+ cache
->key
.offset
- 1;
248 clear_extent_bits(&root
->fs_info
->freed_extents
[0],
249 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
250 clear_extent_bits(&root
->fs_info
->freed_extents
[1],
251 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
254 static int exclude_super_stripes(struct btrfs_root
*root
,
255 struct btrfs_block_group_cache
*cache
)
262 if (cache
->key
.objectid
< BTRFS_SUPER_INFO_OFFSET
) {
263 stripe_len
= BTRFS_SUPER_INFO_OFFSET
- cache
->key
.objectid
;
264 cache
->bytes_super
+= stripe_len
;
265 ret
= add_excluded_extent(root
, cache
->key
.objectid
,
271 for (i
= 0; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
272 bytenr
= btrfs_sb_offset(i
);
273 ret
= btrfs_rmap_block(&root
->fs_info
->mapping_tree
,
274 cache
->key
.objectid
, bytenr
,
275 0, &logical
, &nr
, &stripe_len
);
282 if (logical
[nr
] > cache
->key
.objectid
+
286 if (logical
[nr
] + stripe_len
<= cache
->key
.objectid
)
290 if (start
< cache
->key
.objectid
) {
291 start
= cache
->key
.objectid
;
292 len
= (logical
[nr
] + stripe_len
) - start
;
294 len
= min_t(u64
, stripe_len
,
295 cache
->key
.objectid
+
296 cache
->key
.offset
- start
);
299 cache
->bytes_super
+= len
;
300 ret
= add_excluded_extent(root
, start
, len
);
312 static struct btrfs_caching_control
*
313 get_caching_control(struct btrfs_block_group_cache
*cache
)
315 struct btrfs_caching_control
*ctl
;
317 spin_lock(&cache
->lock
);
318 if (!cache
->caching_ctl
) {
319 spin_unlock(&cache
->lock
);
323 ctl
= cache
->caching_ctl
;
324 atomic_inc(&ctl
->count
);
325 spin_unlock(&cache
->lock
);
329 static void put_caching_control(struct btrfs_caching_control
*ctl
)
331 if (atomic_dec_and_test(&ctl
->count
))
336 * this is only called by cache_block_group, since we could have freed extents
337 * we need to check the pinned_extents for any extents that can't be used yet
338 * since their free space will be released as soon as the transaction commits.
340 static u64
add_new_free_space(struct btrfs_block_group_cache
*block_group
,
341 struct btrfs_fs_info
*info
, u64 start
, u64 end
)
343 u64 extent_start
, extent_end
, size
, total_added
= 0;
346 while (start
< end
) {
347 ret
= find_first_extent_bit(info
->pinned_extents
, start
,
348 &extent_start
, &extent_end
,
349 EXTENT_DIRTY
| EXTENT_UPTODATE
,
354 if (extent_start
<= start
) {
355 start
= extent_end
+ 1;
356 } else if (extent_start
> start
&& extent_start
< end
) {
357 size
= extent_start
- start
;
359 ret
= btrfs_add_free_space(block_group
, start
,
361 BUG_ON(ret
); /* -ENOMEM or logic error */
362 start
= extent_end
+ 1;
371 ret
= btrfs_add_free_space(block_group
, start
, size
);
372 BUG_ON(ret
); /* -ENOMEM or logic error */
378 static noinline
void caching_thread(struct btrfs_work
*work
)
380 struct btrfs_block_group_cache
*block_group
;
381 struct btrfs_fs_info
*fs_info
;
382 struct btrfs_caching_control
*caching_ctl
;
383 struct btrfs_root
*extent_root
;
384 struct btrfs_path
*path
;
385 struct extent_buffer
*leaf
;
386 struct btrfs_key key
;
392 caching_ctl
= container_of(work
, struct btrfs_caching_control
, work
);
393 block_group
= caching_ctl
->block_group
;
394 fs_info
= block_group
->fs_info
;
395 extent_root
= fs_info
->extent_root
;
397 path
= btrfs_alloc_path();
401 last
= max_t(u64
, block_group
->key
.objectid
, BTRFS_SUPER_INFO_OFFSET
);
404 * We don't want to deadlock with somebody trying to allocate a new
405 * extent for the extent root while also trying to search the extent
406 * root to add free space. So we skip locking and search the commit
407 * root, since its read-only
409 path
->skip_locking
= 1;
410 path
->search_commit_root
= 1;
415 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
417 mutex_lock(&caching_ctl
->mutex
);
418 /* need to make sure the commit_root doesn't disappear */
419 down_read(&fs_info
->commit_root_sem
);
422 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
426 leaf
= path
->nodes
[0];
427 nritems
= btrfs_header_nritems(leaf
);
430 if (btrfs_fs_closing(fs_info
) > 1) {
435 if (path
->slots
[0] < nritems
) {
436 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
438 ret
= find_next_key(path
, 0, &key
);
442 if (need_resched() ||
443 rwsem_is_contended(&fs_info
->commit_root_sem
)) {
444 caching_ctl
->progress
= last
;
445 btrfs_release_path(path
);
446 up_read(&fs_info
->commit_root_sem
);
447 mutex_unlock(&caching_ctl
->mutex
);
452 ret
= btrfs_next_leaf(extent_root
, path
);
457 leaf
= path
->nodes
[0];
458 nritems
= btrfs_header_nritems(leaf
);
462 if (key
.objectid
< last
) {
465 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
467 caching_ctl
->progress
= last
;
468 btrfs_release_path(path
);
472 if (key
.objectid
< block_group
->key
.objectid
) {
477 if (key
.objectid
>= block_group
->key
.objectid
+
478 block_group
->key
.offset
)
481 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
||
482 key
.type
== BTRFS_METADATA_ITEM_KEY
) {
483 total_found
+= add_new_free_space(block_group
,
486 if (key
.type
== BTRFS_METADATA_ITEM_KEY
)
487 last
= key
.objectid
+
488 fs_info
->tree_root
->nodesize
;
490 last
= key
.objectid
+ key
.offset
;
492 if (total_found
> (1024 * 1024 * 2)) {
494 wake_up(&caching_ctl
->wait
);
501 total_found
+= add_new_free_space(block_group
, fs_info
, last
,
502 block_group
->key
.objectid
+
503 block_group
->key
.offset
);
504 caching_ctl
->progress
= (u64
)-1;
506 spin_lock(&block_group
->lock
);
507 block_group
->caching_ctl
= NULL
;
508 block_group
->cached
= BTRFS_CACHE_FINISHED
;
509 spin_unlock(&block_group
->lock
);
512 btrfs_free_path(path
);
513 up_read(&fs_info
->commit_root_sem
);
515 free_excluded_extents(extent_root
, block_group
);
517 mutex_unlock(&caching_ctl
->mutex
);
520 spin_lock(&block_group
->lock
);
521 block_group
->caching_ctl
= NULL
;
522 block_group
->cached
= BTRFS_CACHE_ERROR
;
523 spin_unlock(&block_group
->lock
);
525 wake_up(&caching_ctl
->wait
);
527 put_caching_control(caching_ctl
);
528 btrfs_put_block_group(block_group
);
531 static int cache_block_group(struct btrfs_block_group_cache
*cache
,
535 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
536 struct btrfs_caching_control
*caching_ctl
;
539 caching_ctl
= kzalloc(sizeof(*caching_ctl
), GFP_NOFS
);
543 INIT_LIST_HEAD(&caching_ctl
->list
);
544 mutex_init(&caching_ctl
->mutex
);
545 init_waitqueue_head(&caching_ctl
->wait
);
546 caching_ctl
->block_group
= cache
;
547 caching_ctl
->progress
= cache
->key
.objectid
;
548 atomic_set(&caching_ctl
->count
, 1);
549 btrfs_init_work(&caching_ctl
->work
, btrfs_cache_helper
,
550 caching_thread
, NULL
, NULL
);
552 spin_lock(&cache
->lock
);
554 * This should be a rare occasion, but this could happen I think in the
555 * case where one thread starts to load the space cache info, and then
556 * some other thread starts a transaction commit which tries to do an
557 * allocation while the other thread is still loading the space cache
558 * info. The previous loop should have kept us from choosing this block
559 * group, but if we've moved to the state where we will wait on caching
560 * block groups we need to first check if we're doing a fast load here,
561 * so we can wait for it to finish, otherwise we could end up allocating
562 * from a block group who's cache gets evicted for one reason or
565 while (cache
->cached
== BTRFS_CACHE_FAST
) {
566 struct btrfs_caching_control
*ctl
;
568 ctl
= cache
->caching_ctl
;
569 atomic_inc(&ctl
->count
);
570 prepare_to_wait(&ctl
->wait
, &wait
, TASK_UNINTERRUPTIBLE
);
571 spin_unlock(&cache
->lock
);
575 finish_wait(&ctl
->wait
, &wait
);
576 put_caching_control(ctl
);
577 spin_lock(&cache
->lock
);
580 if (cache
->cached
!= BTRFS_CACHE_NO
) {
581 spin_unlock(&cache
->lock
);
585 WARN_ON(cache
->caching_ctl
);
586 cache
->caching_ctl
= caching_ctl
;
587 cache
->cached
= BTRFS_CACHE_FAST
;
588 spin_unlock(&cache
->lock
);
590 if (fs_info
->mount_opt
& BTRFS_MOUNT_SPACE_CACHE
) {
591 mutex_lock(&caching_ctl
->mutex
);
592 ret
= load_free_space_cache(fs_info
, cache
);
594 spin_lock(&cache
->lock
);
596 cache
->caching_ctl
= NULL
;
597 cache
->cached
= BTRFS_CACHE_FINISHED
;
598 cache
->last_byte_to_unpin
= (u64
)-1;
599 caching_ctl
->progress
= (u64
)-1;
601 if (load_cache_only
) {
602 cache
->caching_ctl
= NULL
;
603 cache
->cached
= BTRFS_CACHE_NO
;
605 cache
->cached
= BTRFS_CACHE_STARTED
;
606 cache
->has_caching_ctl
= 1;
609 spin_unlock(&cache
->lock
);
610 mutex_unlock(&caching_ctl
->mutex
);
612 wake_up(&caching_ctl
->wait
);
614 put_caching_control(caching_ctl
);
615 free_excluded_extents(fs_info
->extent_root
, cache
);
620 * We are not going to do the fast caching, set cached to the
621 * appropriate value and wakeup any waiters.
623 spin_lock(&cache
->lock
);
624 if (load_cache_only
) {
625 cache
->caching_ctl
= NULL
;
626 cache
->cached
= BTRFS_CACHE_NO
;
628 cache
->cached
= BTRFS_CACHE_STARTED
;
629 cache
->has_caching_ctl
= 1;
631 spin_unlock(&cache
->lock
);
632 wake_up(&caching_ctl
->wait
);
635 if (load_cache_only
) {
636 put_caching_control(caching_ctl
);
640 down_write(&fs_info
->commit_root_sem
);
641 atomic_inc(&caching_ctl
->count
);
642 list_add_tail(&caching_ctl
->list
, &fs_info
->caching_block_groups
);
643 up_write(&fs_info
->commit_root_sem
);
645 btrfs_get_block_group(cache
);
647 btrfs_queue_work(fs_info
->caching_workers
, &caching_ctl
->work
);
653 * return the block group that starts at or after bytenr
655 static struct btrfs_block_group_cache
*
656 btrfs_lookup_first_block_group(struct btrfs_fs_info
*info
, u64 bytenr
)
658 struct btrfs_block_group_cache
*cache
;
660 cache
= block_group_cache_tree_search(info
, bytenr
, 0);
666 * return the block group that contains the given bytenr
668 struct btrfs_block_group_cache
*btrfs_lookup_block_group(
669 struct btrfs_fs_info
*info
,
672 struct btrfs_block_group_cache
*cache
;
674 cache
= block_group_cache_tree_search(info
, bytenr
, 1);
679 static struct btrfs_space_info
*__find_space_info(struct btrfs_fs_info
*info
,
682 struct list_head
*head
= &info
->space_info
;
683 struct btrfs_space_info
*found
;
685 flags
&= BTRFS_BLOCK_GROUP_TYPE_MASK
;
688 list_for_each_entry_rcu(found
, head
, list
) {
689 if (found
->flags
& flags
) {
699 * after adding space to the filesystem, we need to clear the full flags
700 * on all the space infos.
702 void btrfs_clear_space_info_full(struct btrfs_fs_info
*info
)
704 struct list_head
*head
= &info
->space_info
;
705 struct btrfs_space_info
*found
;
708 list_for_each_entry_rcu(found
, head
, list
)
713 /* simple helper to search for an existing data extent at a given offset */
714 int btrfs_lookup_data_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
717 struct btrfs_key key
;
718 struct btrfs_path
*path
;
720 path
= btrfs_alloc_path();
724 key
.objectid
= start
;
726 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
727 ret
= btrfs_search_slot(NULL
, root
->fs_info
->extent_root
, &key
, path
,
729 btrfs_free_path(path
);
734 * helper function to lookup reference count and flags of a tree block.
736 * the head node for delayed ref is used to store the sum of all the
737 * reference count modifications queued up in the rbtree. the head
738 * node may also store the extent flags to set. This way you can check
739 * to see what the reference count and extent flags would be if all of
740 * the delayed refs are not processed.
742 int btrfs_lookup_extent_info(struct btrfs_trans_handle
*trans
,
743 struct btrfs_root
*root
, u64 bytenr
,
744 u64 offset
, int metadata
, u64
*refs
, u64
*flags
)
746 struct btrfs_delayed_ref_head
*head
;
747 struct btrfs_delayed_ref_root
*delayed_refs
;
748 struct btrfs_path
*path
;
749 struct btrfs_extent_item
*ei
;
750 struct extent_buffer
*leaf
;
751 struct btrfs_key key
;
758 * If we don't have skinny metadata, don't bother doing anything
761 if (metadata
&& !btrfs_fs_incompat(root
->fs_info
, SKINNY_METADATA
)) {
762 offset
= root
->nodesize
;
766 path
= btrfs_alloc_path();
771 path
->skip_locking
= 1;
772 path
->search_commit_root
= 1;
776 key
.objectid
= bytenr
;
779 key
.type
= BTRFS_METADATA_ITEM_KEY
;
781 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
783 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
,
788 if (ret
> 0 && metadata
&& key
.type
== BTRFS_METADATA_ITEM_KEY
) {
789 if (path
->slots
[0]) {
791 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
793 if (key
.objectid
== bytenr
&&
794 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
795 key
.offset
== root
->nodesize
)
801 leaf
= path
->nodes
[0];
802 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
803 if (item_size
>= sizeof(*ei
)) {
804 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
805 struct btrfs_extent_item
);
806 num_refs
= btrfs_extent_refs(leaf
, ei
);
807 extent_flags
= btrfs_extent_flags(leaf
, ei
);
809 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
810 struct btrfs_extent_item_v0
*ei0
;
811 BUG_ON(item_size
!= sizeof(*ei0
));
812 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
813 struct btrfs_extent_item_v0
);
814 num_refs
= btrfs_extent_refs_v0(leaf
, ei0
);
815 /* FIXME: this isn't correct for data */
816 extent_flags
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
821 BUG_ON(num_refs
== 0);
831 delayed_refs
= &trans
->transaction
->delayed_refs
;
832 spin_lock(&delayed_refs
->lock
);
833 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
835 if (!mutex_trylock(&head
->mutex
)) {
836 atomic_inc(&head
->node
.refs
);
837 spin_unlock(&delayed_refs
->lock
);
839 btrfs_release_path(path
);
842 * Mutex was contended, block until it's released and try
845 mutex_lock(&head
->mutex
);
846 mutex_unlock(&head
->mutex
);
847 btrfs_put_delayed_ref(&head
->node
);
850 spin_lock(&head
->lock
);
851 if (head
->extent_op
&& head
->extent_op
->update_flags
)
852 extent_flags
|= head
->extent_op
->flags_to_set
;
854 BUG_ON(num_refs
== 0);
856 num_refs
+= head
->node
.ref_mod
;
857 spin_unlock(&head
->lock
);
858 mutex_unlock(&head
->mutex
);
860 spin_unlock(&delayed_refs
->lock
);
862 WARN_ON(num_refs
== 0);
866 *flags
= extent_flags
;
868 btrfs_free_path(path
);
873 * Back reference rules. Back refs have three main goals:
875 * 1) differentiate between all holders of references to an extent so that
876 * when a reference is dropped we can make sure it was a valid reference
877 * before freeing the extent.
879 * 2) Provide enough information to quickly find the holders of an extent
880 * if we notice a given block is corrupted or bad.
882 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
883 * maintenance. This is actually the same as #2, but with a slightly
884 * different use case.
886 * There are two kinds of back refs. The implicit back refs is optimized
887 * for pointers in non-shared tree blocks. For a given pointer in a block,
888 * back refs of this kind provide information about the block's owner tree
889 * and the pointer's key. These information allow us to find the block by
890 * b-tree searching. The full back refs is for pointers in tree blocks not
891 * referenced by their owner trees. The location of tree block is recorded
892 * in the back refs. Actually the full back refs is generic, and can be
893 * used in all cases the implicit back refs is used. The major shortcoming
894 * of the full back refs is its overhead. Every time a tree block gets
895 * COWed, we have to update back refs entry for all pointers in it.
897 * For a newly allocated tree block, we use implicit back refs for
898 * pointers in it. This means most tree related operations only involve
899 * implicit back refs. For a tree block created in old transaction, the
900 * only way to drop a reference to it is COW it. So we can detect the
901 * event that tree block loses its owner tree's reference and do the
902 * back refs conversion.
904 * When a tree block is COW'd through a tree, there are four cases:
906 * The reference count of the block is one and the tree is the block's
907 * owner tree. Nothing to do in this case.
909 * The reference count of the block is one and the tree is not the
910 * block's owner tree. In this case, full back refs is used for pointers
911 * in the block. Remove these full back refs, add implicit back refs for
912 * every pointers in the new block.
914 * The reference count of the block is greater than one and the tree is
915 * the block's owner tree. In this case, implicit back refs is used for
916 * pointers in the block. Add full back refs for every pointers in the
917 * block, increase lower level extents' reference counts. The original
918 * implicit back refs are entailed to the new block.
920 * The reference count of the block is greater than one and the tree is
921 * not the block's owner tree. Add implicit back refs for every pointer in
922 * the new block, increase lower level extents' reference count.
924 * Back Reference Key composing:
926 * The key objectid corresponds to the first byte in the extent,
927 * The key type is used to differentiate between types of back refs.
928 * There are different meanings of the key offset for different types
931 * File extents can be referenced by:
933 * - multiple snapshots, subvolumes, or different generations in one subvol
934 * - different files inside a single subvolume
935 * - different offsets inside a file (bookend extents in file.c)
937 * The extent ref structure for the implicit back refs has fields for:
939 * - Objectid of the subvolume root
940 * - objectid of the file holding the reference
941 * - original offset in the file
942 * - how many bookend extents
944 * The key offset for the implicit back refs is hash of the first
947 * The extent ref structure for the full back refs has field for:
949 * - number of pointers in the tree leaf
951 * The key offset for the implicit back refs is the first byte of
954 * When a file extent is allocated, The implicit back refs is used.
955 * the fields are filled in:
957 * (root_key.objectid, inode objectid, offset in file, 1)
959 * When a file extent is removed file truncation, we find the
960 * corresponding implicit back refs and check the following fields:
962 * (btrfs_header_owner(leaf), inode objectid, offset in file)
964 * Btree extents can be referenced by:
966 * - Different subvolumes
968 * Both the implicit back refs and the full back refs for tree blocks
969 * only consist of key. The key offset for the implicit back refs is
970 * objectid of block's owner tree. The key offset for the full back refs
971 * is the first byte of parent block.
973 * When implicit back refs is used, information about the lowest key and
974 * level of the tree block are required. These information are stored in
975 * tree block info structure.
978 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
979 static int convert_extent_item_v0(struct btrfs_trans_handle
*trans
,
980 struct btrfs_root
*root
,
981 struct btrfs_path
*path
,
982 u64 owner
, u32 extra_size
)
984 struct btrfs_extent_item
*item
;
985 struct btrfs_extent_item_v0
*ei0
;
986 struct btrfs_extent_ref_v0
*ref0
;
987 struct btrfs_tree_block_info
*bi
;
988 struct extent_buffer
*leaf
;
989 struct btrfs_key key
;
990 struct btrfs_key found_key
;
991 u32 new_size
= sizeof(*item
);
995 leaf
= path
->nodes
[0];
996 BUG_ON(btrfs_item_size_nr(leaf
, path
->slots
[0]) != sizeof(*ei0
));
998 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
999 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1000 struct btrfs_extent_item_v0
);
1001 refs
= btrfs_extent_refs_v0(leaf
, ei0
);
1003 if (owner
== (u64
)-1) {
1005 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1006 ret
= btrfs_next_leaf(root
, path
);
1009 BUG_ON(ret
> 0); /* Corruption */
1010 leaf
= path
->nodes
[0];
1012 btrfs_item_key_to_cpu(leaf
, &found_key
,
1014 BUG_ON(key
.objectid
!= found_key
.objectid
);
1015 if (found_key
.type
!= BTRFS_EXTENT_REF_V0_KEY
) {
1019 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1020 struct btrfs_extent_ref_v0
);
1021 owner
= btrfs_ref_objectid_v0(leaf
, ref0
);
1025 btrfs_release_path(path
);
1027 if (owner
< BTRFS_FIRST_FREE_OBJECTID
)
1028 new_size
+= sizeof(*bi
);
1030 new_size
-= sizeof(*ei0
);
1031 ret
= btrfs_search_slot(trans
, root
, &key
, path
,
1032 new_size
+ extra_size
, 1);
1035 BUG_ON(ret
); /* Corruption */
1037 btrfs_extend_item(root
, path
, new_size
);
1039 leaf
= path
->nodes
[0];
1040 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1041 btrfs_set_extent_refs(leaf
, item
, refs
);
1042 /* FIXME: get real generation */
1043 btrfs_set_extent_generation(leaf
, item
, 0);
1044 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1045 btrfs_set_extent_flags(leaf
, item
,
1046 BTRFS_EXTENT_FLAG_TREE_BLOCK
|
1047 BTRFS_BLOCK_FLAG_FULL_BACKREF
);
1048 bi
= (struct btrfs_tree_block_info
*)(item
+ 1);
1049 /* FIXME: get first key of the block */
1050 memset_extent_buffer(leaf
, 0, (unsigned long)bi
, sizeof(*bi
));
1051 btrfs_set_tree_block_level(leaf
, bi
, (int)owner
);
1053 btrfs_set_extent_flags(leaf
, item
, BTRFS_EXTENT_FLAG_DATA
);
1055 btrfs_mark_buffer_dirty(leaf
);
1060 static u64
hash_extent_data_ref(u64 root_objectid
, u64 owner
, u64 offset
)
1062 u32 high_crc
= ~(u32
)0;
1063 u32 low_crc
= ~(u32
)0;
1066 lenum
= cpu_to_le64(root_objectid
);
1067 high_crc
= btrfs_crc32c(high_crc
, &lenum
, sizeof(lenum
));
1068 lenum
= cpu_to_le64(owner
);
1069 low_crc
= btrfs_crc32c(low_crc
, &lenum
, sizeof(lenum
));
1070 lenum
= cpu_to_le64(offset
);
1071 low_crc
= btrfs_crc32c(low_crc
, &lenum
, sizeof(lenum
));
1073 return ((u64
)high_crc
<< 31) ^ (u64
)low_crc
;
1076 static u64
hash_extent_data_ref_item(struct extent_buffer
*leaf
,
1077 struct btrfs_extent_data_ref
*ref
)
1079 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf
, ref
),
1080 btrfs_extent_data_ref_objectid(leaf
, ref
),
1081 btrfs_extent_data_ref_offset(leaf
, ref
));
1084 static int match_extent_data_ref(struct extent_buffer
*leaf
,
1085 struct btrfs_extent_data_ref
*ref
,
1086 u64 root_objectid
, u64 owner
, u64 offset
)
1088 if (btrfs_extent_data_ref_root(leaf
, ref
) != root_objectid
||
1089 btrfs_extent_data_ref_objectid(leaf
, ref
) != owner
||
1090 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
1095 static noinline
int lookup_extent_data_ref(struct btrfs_trans_handle
*trans
,
1096 struct btrfs_root
*root
,
1097 struct btrfs_path
*path
,
1098 u64 bytenr
, u64 parent
,
1100 u64 owner
, u64 offset
)
1102 struct btrfs_key key
;
1103 struct btrfs_extent_data_ref
*ref
;
1104 struct extent_buffer
*leaf
;
1110 key
.objectid
= bytenr
;
1112 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1113 key
.offset
= parent
;
1115 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1116 key
.offset
= hash_extent_data_ref(root_objectid
,
1121 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1130 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1131 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1132 btrfs_release_path(path
);
1133 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1144 leaf
= path
->nodes
[0];
1145 nritems
= btrfs_header_nritems(leaf
);
1147 if (path
->slots
[0] >= nritems
) {
1148 ret
= btrfs_next_leaf(root
, path
);
1154 leaf
= path
->nodes
[0];
1155 nritems
= btrfs_header_nritems(leaf
);
1159 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1160 if (key
.objectid
!= bytenr
||
1161 key
.type
!= BTRFS_EXTENT_DATA_REF_KEY
)
1164 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1165 struct btrfs_extent_data_ref
);
1167 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1170 btrfs_release_path(path
);
1182 static noinline
int insert_extent_data_ref(struct btrfs_trans_handle
*trans
,
1183 struct btrfs_root
*root
,
1184 struct btrfs_path
*path
,
1185 u64 bytenr
, u64 parent
,
1186 u64 root_objectid
, u64 owner
,
1187 u64 offset
, int refs_to_add
)
1189 struct btrfs_key key
;
1190 struct extent_buffer
*leaf
;
1195 key
.objectid
= bytenr
;
1197 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1198 key
.offset
= parent
;
1199 size
= sizeof(struct btrfs_shared_data_ref
);
1201 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1202 key
.offset
= hash_extent_data_ref(root_objectid
,
1204 size
= sizeof(struct btrfs_extent_data_ref
);
1207 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, size
);
1208 if (ret
&& ret
!= -EEXIST
)
1211 leaf
= path
->nodes
[0];
1213 struct btrfs_shared_data_ref
*ref
;
1214 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1215 struct btrfs_shared_data_ref
);
1217 btrfs_set_shared_data_ref_count(leaf
, ref
, refs_to_add
);
1219 num_refs
= btrfs_shared_data_ref_count(leaf
, ref
);
1220 num_refs
+= refs_to_add
;
1221 btrfs_set_shared_data_ref_count(leaf
, ref
, num_refs
);
1224 struct btrfs_extent_data_ref
*ref
;
1225 while (ret
== -EEXIST
) {
1226 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1227 struct btrfs_extent_data_ref
);
1228 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1231 btrfs_release_path(path
);
1233 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1235 if (ret
&& ret
!= -EEXIST
)
1238 leaf
= path
->nodes
[0];
1240 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1241 struct btrfs_extent_data_ref
);
1243 btrfs_set_extent_data_ref_root(leaf
, ref
,
1245 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
1246 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
1247 btrfs_set_extent_data_ref_count(leaf
, ref
, refs_to_add
);
1249 num_refs
= btrfs_extent_data_ref_count(leaf
, ref
);
1250 num_refs
+= refs_to_add
;
1251 btrfs_set_extent_data_ref_count(leaf
, ref
, num_refs
);
1254 btrfs_mark_buffer_dirty(leaf
);
1257 btrfs_release_path(path
);
1261 static noinline
int remove_extent_data_ref(struct btrfs_trans_handle
*trans
,
1262 struct btrfs_root
*root
,
1263 struct btrfs_path
*path
,
1264 int refs_to_drop
, int *last_ref
)
1266 struct btrfs_key key
;
1267 struct btrfs_extent_data_ref
*ref1
= NULL
;
1268 struct btrfs_shared_data_ref
*ref2
= NULL
;
1269 struct extent_buffer
*leaf
;
1273 leaf
= path
->nodes
[0];
1274 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1276 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1277 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1278 struct btrfs_extent_data_ref
);
1279 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1280 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1281 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1282 struct btrfs_shared_data_ref
);
1283 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1284 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1285 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1286 struct btrfs_extent_ref_v0
*ref0
;
1287 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1288 struct btrfs_extent_ref_v0
);
1289 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1295 BUG_ON(num_refs
< refs_to_drop
);
1296 num_refs
-= refs_to_drop
;
1298 if (num_refs
== 0) {
1299 ret
= btrfs_del_item(trans
, root
, path
);
1302 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
)
1303 btrfs_set_extent_data_ref_count(leaf
, ref1
, num_refs
);
1304 else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
)
1305 btrfs_set_shared_data_ref_count(leaf
, ref2
, num_refs
);
1306 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1308 struct btrfs_extent_ref_v0
*ref0
;
1309 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1310 struct btrfs_extent_ref_v0
);
1311 btrfs_set_ref_count_v0(leaf
, ref0
, num_refs
);
1314 btrfs_mark_buffer_dirty(leaf
);
1319 static noinline u32
extent_data_ref_count(struct btrfs_root
*root
,
1320 struct btrfs_path
*path
,
1321 struct btrfs_extent_inline_ref
*iref
)
1323 struct btrfs_key key
;
1324 struct extent_buffer
*leaf
;
1325 struct btrfs_extent_data_ref
*ref1
;
1326 struct btrfs_shared_data_ref
*ref2
;
1329 leaf
= path
->nodes
[0];
1330 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1332 if (btrfs_extent_inline_ref_type(leaf
, iref
) ==
1333 BTRFS_EXTENT_DATA_REF_KEY
) {
1334 ref1
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1335 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1337 ref2
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1338 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1340 } else if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1341 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1342 struct btrfs_extent_data_ref
);
1343 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1344 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1345 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1346 struct btrfs_shared_data_ref
);
1347 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1348 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1349 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1350 struct btrfs_extent_ref_v0
*ref0
;
1351 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1352 struct btrfs_extent_ref_v0
);
1353 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1361 static noinline
int lookup_tree_block_ref(struct btrfs_trans_handle
*trans
,
1362 struct btrfs_root
*root
,
1363 struct btrfs_path
*path
,
1364 u64 bytenr
, u64 parent
,
1367 struct btrfs_key key
;
1370 key
.objectid
= bytenr
;
1372 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1373 key
.offset
= parent
;
1375 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1376 key
.offset
= root_objectid
;
1379 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1382 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1383 if (ret
== -ENOENT
&& parent
) {
1384 btrfs_release_path(path
);
1385 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1386 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1394 static noinline
int insert_tree_block_ref(struct btrfs_trans_handle
*trans
,
1395 struct btrfs_root
*root
,
1396 struct btrfs_path
*path
,
1397 u64 bytenr
, u64 parent
,
1400 struct btrfs_key key
;
1403 key
.objectid
= bytenr
;
1405 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1406 key
.offset
= parent
;
1408 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1409 key
.offset
= root_objectid
;
1412 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
1413 btrfs_release_path(path
);
1417 static inline int extent_ref_type(u64 parent
, u64 owner
)
1420 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1422 type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1424 type
= BTRFS_TREE_BLOCK_REF_KEY
;
1427 type
= BTRFS_SHARED_DATA_REF_KEY
;
1429 type
= BTRFS_EXTENT_DATA_REF_KEY
;
1434 static int find_next_key(struct btrfs_path
*path
, int level
,
1435 struct btrfs_key
*key
)
1438 for (; level
< BTRFS_MAX_LEVEL
; level
++) {
1439 if (!path
->nodes
[level
])
1441 if (path
->slots
[level
] + 1 >=
1442 btrfs_header_nritems(path
->nodes
[level
]))
1445 btrfs_item_key_to_cpu(path
->nodes
[level
], key
,
1446 path
->slots
[level
] + 1);
1448 btrfs_node_key_to_cpu(path
->nodes
[level
], key
,
1449 path
->slots
[level
] + 1);
1456 * look for inline back ref. if back ref is found, *ref_ret is set
1457 * to the address of inline back ref, and 0 is returned.
1459 * if back ref isn't found, *ref_ret is set to the address where it
1460 * should be inserted, and -ENOENT is returned.
1462 * if insert is true and there are too many inline back refs, the path
1463 * points to the extent item, and -EAGAIN is returned.
1465 * NOTE: inline back refs are ordered in the same way that back ref
1466 * items in the tree are ordered.
1468 static noinline_for_stack
1469 int lookup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1470 struct btrfs_root
*root
,
1471 struct btrfs_path
*path
,
1472 struct btrfs_extent_inline_ref
**ref_ret
,
1473 u64 bytenr
, u64 num_bytes
,
1474 u64 parent
, u64 root_objectid
,
1475 u64 owner
, u64 offset
, int insert
)
1477 struct btrfs_key key
;
1478 struct extent_buffer
*leaf
;
1479 struct btrfs_extent_item
*ei
;
1480 struct btrfs_extent_inline_ref
*iref
;
1490 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
1493 key
.objectid
= bytenr
;
1494 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1495 key
.offset
= num_bytes
;
1497 want
= extent_ref_type(parent
, owner
);
1499 extra_size
= btrfs_extent_inline_ref_size(want
);
1500 path
->keep_locks
= 1;
1505 * Owner is our parent level, so we can just add one to get the level
1506 * for the block we are interested in.
1508 if (skinny_metadata
&& owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1509 key
.type
= BTRFS_METADATA_ITEM_KEY
;
1514 ret
= btrfs_search_slot(trans
, root
, &key
, path
, extra_size
, 1);
1521 * We may be a newly converted file system which still has the old fat
1522 * extent entries for metadata, so try and see if we have one of those.
1524 if (ret
> 0 && skinny_metadata
) {
1525 skinny_metadata
= false;
1526 if (path
->slots
[0]) {
1528 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
1530 if (key
.objectid
== bytenr
&&
1531 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
1532 key
.offset
== num_bytes
)
1536 key
.objectid
= bytenr
;
1537 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1538 key
.offset
= num_bytes
;
1539 btrfs_release_path(path
);
1544 if (ret
&& !insert
) {
1547 } else if (WARN_ON(ret
)) {
1552 leaf
= path
->nodes
[0];
1553 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1554 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1555 if (item_size
< sizeof(*ei
)) {
1560 ret
= convert_extent_item_v0(trans
, root
, path
, owner
,
1566 leaf
= path
->nodes
[0];
1567 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1570 BUG_ON(item_size
< sizeof(*ei
));
1572 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1573 flags
= btrfs_extent_flags(leaf
, ei
);
1575 ptr
= (unsigned long)(ei
+ 1);
1576 end
= (unsigned long)ei
+ item_size
;
1578 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
&& !skinny_metadata
) {
1579 ptr
+= sizeof(struct btrfs_tree_block_info
);
1589 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1590 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1594 ptr
+= btrfs_extent_inline_ref_size(type
);
1598 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1599 struct btrfs_extent_data_ref
*dref
;
1600 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1601 if (match_extent_data_ref(leaf
, dref
, root_objectid
,
1606 if (hash_extent_data_ref_item(leaf
, dref
) <
1607 hash_extent_data_ref(root_objectid
, owner
, offset
))
1611 ref_offset
= btrfs_extent_inline_ref_offset(leaf
, iref
);
1613 if (parent
== ref_offset
) {
1617 if (ref_offset
< parent
)
1620 if (root_objectid
== ref_offset
) {
1624 if (ref_offset
< root_objectid
)
1628 ptr
+= btrfs_extent_inline_ref_size(type
);
1630 if (err
== -ENOENT
&& insert
) {
1631 if (item_size
+ extra_size
>=
1632 BTRFS_MAX_EXTENT_ITEM_SIZE(root
)) {
1637 * To add new inline back ref, we have to make sure
1638 * there is no corresponding back ref item.
1639 * For simplicity, we just do not add new inline back
1640 * ref if there is any kind of item for this block
1642 if (find_next_key(path
, 0, &key
) == 0 &&
1643 key
.objectid
== bytenr
&&
1644 key
.type
< BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1649 *ref_ret
= (struct btrfs_extent_inline_ref
*)ptr
;
1652 path
->keep_locks
= 0;
1653 btrfs_unlock_up_safe(path
, 1);
1659 * helper to add new inline back ref
1661 static noinline_for_stack
1662 void setup_inline_extent_backref(struct btrfs_root
*root
,
1663 struct btrfs_path
*path
,
1664 struct btrfs_extent_inline_ref
*iref
,
1665 u64 parent
, u64 root_objectid
,
1666 u64 owner
, u64 offset
, int refs_to_add
,
1667 struct btrfs_delayed_extent_op
*extent_op
)
1669 struct extent_buffer
*leaf
;
1670 struct btrfs_extent_item
*ei
;
1673 unsigned long item_offset
;
1678 leaf
= path
->nodes
[0];
1679 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1680 item_offset
= (unsigned long)iref
- (unsigned long)ei
;
1682 type
= extent_ref_type(parent
, owner
);
1683 size
= btrfs_extent_inline_ref_size(type
);
1685 btrfs_extend_item(root
, path
, size
);
1687 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1688 refs
= btrfs_extent_refs(leaf
, ei
);
1689 refs
+= refs_to_add
;
1690 btrfs_set_extent_refs(leaf
, ei
, refs
);
1692 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1694 ptr
= (unsigned long)ei
+ item_offset
;
1695 end
= (unsigned long)ei
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
1696 if (ptr
< end
- size
)
1697 memmove_extent_buffer(leaf
, ptr
+ size
, ptr
,
1700 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1701 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
1702 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1703 struct btrfs_extent_data_ref
*dref
;
1704 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1705 btrfs_set_extent_data_ref_root(leaf
, dref
, root_objectid
);
1706 btrfs_set_extent_data_ref_objectid(leaf
, dref
, owner
);
1707 btrfs_set_extent_data_ref_offset(leaf
, dref
, offset
);
1708 btrfs_set_extent_data_ref_count(leaf
, dref
, refs_to_add
);
1709 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1710 struct btrfs_shared_data_ref
*sref
;
1711 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1712 btrfs_set_shared_data_ref_count(leaf
, sref
, refs_to_add
);
1713 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1714 } else if (type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
1715 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1717 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
1719 btrfs_mark_buffer_dirty(leaf
);
1722 static int lookup_extent_backref(struct btrfs_trans_handle
*trans
,
1723 struct btrfs_root
*root
,
1724 struct btrfs_path
*path
,
1725 struct btrfs_extent_inline_ref
**ref_ret
,
1726 u64 bytenr
, u64 num_bytes
, u64 parent
,
1727 u64 root_objectid
, u64 owner
, u64 offset
)
1731 ret
= lookup_inline_extent_backref(trans
, root
, path
, ref_ret
,
1732 bytenr
, num_bytes
, parent
,
1733 root_objectid
, owner
, offset
, 0);
1737 btrfs_release_path(path
);
1740 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1741 ret
= lookup_tree_block_ref(trans
, root
, path
, bytenr
, parent
,
1744 ret
= lookup_extent_data_ref(trans
, root
, path
, bytenr
, parent
,
1745 root_objectid
, owner
, offset
);
1751 * helper to update/remove inline back ref
1753 static noinline_for_stack
1754 void update_inline_extent_backref(struct btrfs_root
*root
,
1755 struct btrfs_path
*path
,
1756 struct btrfs_extent_inline_ref
*iref
,
1758 struct btrfs_delayed_extent_op
*extent_op
,
1761 struct extent_buffer
*leaf
;
1762 struct btrfs_extent_item
*ei
;
1763 struct btrfs_extent_data_ref
*dref
= NULL
;
1764 struct btrfs_shared_data_ref
*sref
= NULL
;
1772 leaf
= path
->nodes
[0];
1773 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1774 refs
= btrfs_extent_refs(leaf
, ei
);
1775 WARN_ON(refs_to_mod
< 0 && refs
+ refs_to_mod
<= 0);
1776 refs
+= refs_to_mod
;
1777 btrfs_set_extent_refs(leaf
, ei
, refs
);
1779 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1781 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1783 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1784 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1785 refs
= btrfs_extent_data_ref_count(leaf
, dref
);
1786 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1787 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1788 refs
= btrfs_shared_data_ref_count(leaf
, sref
);
1791 BUG_ON(refs_to_mod
!= -1);
1794 BUG_ON(refs_to_mod
< 0 && refs
< -refs_to_mod
);
1795 refs
+= refs_to_mod
;
1798 if (type
== BTRFS_EXTENT_DATA_REF_KEY
)
1799 btrfs_set_extent_data_ref_count(leaf
, dref
, refs
);
1801 btrfs_set_shared_data_ref_count(leaf
, sref
, refs
);
1804 size
= btrfs_extent_inline_ref_size(type
);
1805 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1806 ptr
= (unsigned long)iref
;
1807 end
= (unsigned long)ei
+ item_size
;
1808 if (ptr
+ size
< end
)
1809 memmove_extent_buffer(leaf
, ptr
, ptr
+ size
,
1812 btrfs_truncate_item(root
, path
, item_size
, 1);
1814 btrfs_mark_buffer_dirty(leaf
);
1817 static noinline_for_stack
1818 int insert_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1819 struct btrfs_root
*root
,
1820 struct btrfs_path
*path
,
1821 u64 bytenr
, u64 num_bytes
, u64 parent
,
1822 u64 root_objectid
, u64 owner
,
1823 u64 offset
, int refs_to_add
,
1824 struct btrfs_delayed_extent_op
*extent_op
)
1826 struct btrfs_extent_inline_ref
*iref
;
1829 ret
= lookup_inline_extent_backref(trans
, root
, path
, &iref
,
1830 bytenr
, num_bytes
, parent
,
1831 root_objectid
, owner
, offset
, 1);
1833 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
);
1834 update_inline_extent_backref(root
, path
, iref
,
1835 refs_to_add
, extent_op
, NULL
);
1836 } else if (ret
== -ENOENT
) {
1837 setup_inline_extent_backref(root
, path
, iref
, parent
,
1838 root_objectid
, owner
, offset
,
1839 refs_to_add
, extent_op
);
1845 static int insert_extent_backref(struct btrfs_trans_handle
*trans
,
1846 struct btrfs_root
*root
,
1847 struct btrfs_path
*path
,
1848 u64 bytenr
, u64 parent
, u64 root_objectid
,
1849 u64 owner
, u64 offset
, int refs_to_add
)
1852 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1853 BUG_ON(refs_to_add
!= 1);
1854 ret
= insert_tree_block_ref(trans
, root
, path
, bytenr
,
1855 parent
, root_objectid
);
1857 ret
= insert_extent_data_ref(trans
, root
, path
, bytenr
,
1858 parent
, root_objectid
,
1859 owner
, offset
, refs_to_add
);
1864 static int remove_extent_backref(struct btrfs_trans_handle
*trans
,
1865 struct btrfs_root
*root
,
1866 struct btrfs_path
*path
,
1867 struct btrfs_extent_inline_ref
*iref
,
1868 int refs_to_drop
, int is_data
, int *last_ref
)
1872 BUG_ON(!is_data
&& refs_to_drop
!= 1);
1874 update_inline_extent_backref(root
, path
, iref
,
1875 -refs_to_drop
, NULL
, last_ref
);
1876 } else if (is_data
) {
1877 ret
= remove_extent_data_ref(trans
, root
, path
, refs_to_drop
,
1881 ret
= btrfs_del_item(trans
, root
, path
);
1886 static int btrfs_issue_discard(struct block_device
*bdev
,
1889 return blkdev_issue_discard(bdev
, start
>> 9, len
>> 9, GFP_NOFS
, 0);
1892 int btrfs_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
1893 u64 num_bytes
, u64
*actual_bytes
)
1896 u64 discarded_bytes
= 0;
1897 struct btrfs_bio
*bbio
= NULL
;
1900 /* Tell the block device(s) that the sectors can be discarded */
1901 ret
= btrfs_map_block(root
->fs_info
, REQ_DISCARD
,
1902 bytenr
, &num_bytes
, &bbio
, 0);
1903 /* Error condition is -ENOMEM */
1905 struct btrfs_bio_stripe
*stripe
= bbio
->stripes
;
1909 for (i
= 0; i
< bbio
->num_stripes
; i
++, stripe
++) {
1910 if (!stripe
->dev
->can_discard
)
1913 ret
= btrfs_issue_discard(stripe
->dev
->bdev
,
1917 discarded_bytes
+= stripe
->length
;
1918 else if (ret
!= -EOPNOTSUPP
)
1919 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1922 * Just in case we get back EOPNOTSUPP for some reason,
1923 * just ignore the return value so we don't screw up
1924 * people calling discard_extent.
1928 btrfs_put_bbio(bbio
);
1932 *actual_bytes
= discarded_bytes
;
1935 if (ret
== -EOPNOTSUPP
)
1940 /* Can return -ENOMEM */
1941 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1942 struct btrfs_root
*root
,
1943 u64 bytenr
, u64 num_bytes
, u64 parent
,
1944 u64 root_objectid
, u64 owner
, u64 offset
,
1948 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1950 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
&&
1951 root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
1953 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1954 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
1956 parent
, root_objectid
, (int)owner
,
1957 BTRFS_ADD_DELAYED_REF
, NULL
, no_quota
);
1959 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
1961 parent
, root_objectid
, owner
, offset
,
1962 BTRFS_ADD_DELAYED_REF
, NULL
, no_quota
);
1967 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1968 struct btrfs_root
*root
,
1969 struct btrfs_delayed_ref_node
*node
,
1970 u64 parent
, u64 root_objectid
,
1971 u64 owner
, u64 offset
, int refs_to_add
,
1972 struct btrfs_delayed_extent_op
*extent_op
)
1974 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1975 struct btrfs_path
*path
;
1976 struct extent_buffer
*leaf
;
1977 struct btrfs_extent_item
*item
;
1978 struct btrfs_key key
;
1979 u64 bytenr
= node
->bytenr
;
1980 u64 num_bytes
= node
->num_bytes
;
1983 int no_quota
= node
->no_quota
;
1984 enum btrfs_qgroup_operation_type type
= BTRFS_QGROUP_OPER_ADD_EXCL
;
1986 path
= btrfs_alloc_path();
1990 if (!is_fstree(root_objectid
) || !root
->fs_info
->quota_enabled
)
1994 path
->leave_spinning
= 1;
1995 /* this will setup the path even if it fails to insert the back ref */
1996 ret
= insert_inline_extent_backref(trans
, fs_info
->extent_root
, path
,
1997 bytenr
, num_bytes
, parent
,
1998 root_objectid
, owner
, offset
,
1999 refs_to_add
, extent_op
);
2000 if ((ret
< 0 && ret
!= -EAGAIN
) || (!ret
&& no_quota
))
2003 * Ok we were able to insert an inline extent and it appears to be a new
2004 * reference, deal with the qgroup accounting.
2006 if (!ret
&& !no_quota
) {
2007 ASSERT(root
->fs_info
->quota_enabled
);
2008 leaf
= path
->nodes
[0];
2009 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2010 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2011 struct btrfs_extent_item
);
2012 if (btrfs_extent_refs(leaf
, item
) > (u64
)refs_to_add
)
2013 type
= BTRFS_QGROUP_OPER_ADD_SHARED
;
2014 btrfs_release_path(path
);
2016 ret
= btrfs_qgroup_record_ref(trans
, fs_info
, root_objectid
,
2017 bytenr
, num_bytes
, type
, 0);
2022 * Ok we had -EAGAIN which means we didn't have space to insert and
2023 * inline extent ref, so just update the reference count and add a
2026 leaf
= path
->nodes
[0];
2027 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2028 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2029 refs
= btrfs_extent_refs(leaf
, item
);
2031 type
= BTRFS_QGROUP_OPER_ADD_SHARED
;
2032 btrfs_set_extent_refs(leaf
, item
, refs
+ refs_to_add
);
2034 __run_delayed_extent_op(extent_op
, leaf
, item
);
2036 btrfs_mark_buffer_dirty(leaf
);
2037 btrfs_release_path(path
);
2040 ret
= btrfs_qgroup_record_ref(trans
, fs_info
, root_objectid
,
2041 bytenr
, num_bytes
, type
, 0);
2047 path
->leave_spinning
= 1;
2048 /* now insert the actual backref */
2049 ret
= insert_extent_backref(trans
, root
->fs_info
->extent_root
,
2050 path
, bytenr
, parent
, root_objectid
,
2051 owner
, offset
, refs_to_add
);
2053 btrfs_abort_transaction(trans
, root
, ret
);
2055 btrfs_free_path(path
);
2059 static int run_delayed_data_ref(struct btrfs_trans_handle
*trans
,
2060 struct btrfs_root
*root
,
2061 struct btrfs_delayed_ref_node
*node
,
2062 struct btrfs_delayed_extent_op
*extent_op
,
2063 int insert_reserved
)
2066 struct btrfs_delayed_data_ref
*ref
;
2067 struct btrfs_key ins
;
2072 ins
.objectid
= node
->bytenr
;
2073 ins
.offset
= node
->num_bytes
;
2074 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2076 ref
= btrfs_delayed_node_to_data_ref(node
);
2077 trace_run_delayed_data_ref(node
, ref
, node
->action
);
2079 if (node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2080 parent
= ref
->parent
;
2081 ref_root
= ref
->root
;
2083 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2085 flags
|= extent_op
->flags_to_set
;
2086 ret
= alloc_reserved_file_extent(trans
, root
,
2087 parent
, ref_root
, flags
,
2088 ref
->objectid
, ref
->offset
,
2089 &ins
, node
->ref_mod
);
2090 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2091 ret
= __btrfs_inc_extent_ref(trans
, root
, node
, parent
,
2092 ref_root
, ref
->objectid
,
2093 ref
->offset
, node
->ref_mod
,
2095 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2096 ret
= __btrfs_free_extent(trans
, root
, node
, parent
,
2097 ref_root
, ref
->objectid
,
2098 ref
->offset
, node
->ref_mod
,
2106 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
2107 struct extent_buffer
*leaf
,
2108 struct btrfs_extent_item
*ei
)
2110 u64 flags
= btrfs_extent_flags(leaf
, ei
);
2111 if (extent_op
->update_flags
) {
2112 flags
|= extent_op
->flags_to_set
;
2113 btrfs_set_extent_flags(leaf
, ei
, flags
);
2116 if (extent_op
->update_key
) {
2117 struct btrfs_tree_block_info
*bi
;
2118 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
));
2119 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
2120 btrfs_set_tree_block_key(leaf
, bi
, &extent_op
->key
);
2124 static int run_delayed_extent_op(struct btrfs_trans_handle
*trans
,
2125 struct btrfs_root
*root
,
2126 struct btrfs_delayed_ref_node
*node
,
2127 struct btrfs_delayed_extent_op
*extent_op
)
2129 struct btrfs_key key
;
2130 struct btrfs_path
*path
;
2131 struct btrfs_extent_item
*ei
;
2132 struct extent_buffer
*leaf
;
2136 int metadata
= !extent_op
->is_data
;
2141 if (metadata
&& !btrfs_fs_incompat(root
->fs_info
, SKINNY_METADATA
))
2144 path
= btrfs_alloc_path();
2148 key
.objectid
= node
->bytenr
;
2151 key
.type
= BTRFS_METADATA_ITEM_KEY
;
2152 key
.offset
= extent_op
->level
;
2154 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2155 key
.offset
= node
->num_bytes
;
2160 path
->leave_spinning
= 1;
2161 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
,
2169 if (path
->slots
[0] > 0) {
2171 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
2173 if (key
.objectid
== node
->bytenr
&&
2174 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
2175 key
.offset
== node
->num_bytes
)
2179 btrfs_release_path(path
);
2182 key
.objectid
= node
->bytenr
;
2183 key
.offset
= node
->num_bytes
;
2184 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2193 leaf
= path
->nodes
[0];
2194 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2195 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2196 if (item_size
< sizeof(*ei
)) {
2197 ret
= convert_extent_item_v0(trans
, root
->fs_info
->extent_root
,
2203 leaf
= path
->nodes
[0];
2204 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2207 BUG_ON(item_size
< sizeof(*ei
));
2208 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2209 __run_delayed_extent_op(extent_op
, leaf
, ei
);
2211 btrfs_mark_buffer_dirty(leaf
);
2213 btrfs_free_path(path
);
2217 static int run_delayed_tree_ref(struct btrfs_trans_handle
*trans
,
2218 struct btrfs_root
*root
,
2219 struct btrfs_delayed_ref_node
*node
,
2220 struct btrfs_delayed_extent_op
*extent_op
,
2221 int insert_reserved
)
2224 struct btrfs_delayed_tree_ref
*ref
;
2225 struct btrfs_key ins
;
2228 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
2231 ref
= btrfs_delayed_node_to_tree_ref(node
);
2232 trace_run_delayed_tree_ref(node
, ref
, node
->action
);
2234 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2235 parent
= ref
->parent
;
2236 ref_root
= ref
->root
;
2238 ins
.objectid
= node
->bytenr
;
2239 if (skinny_metadata
) {
2240 ins
.offset
= ref
->level
;
2241 ins
.type
= BTRFS_METADATA_ITEM_KEY
;
2243 ins
.offset
= node
->num_bytes
;
2244 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2247 BUG_ON(node
->ref_mod
!= 1);
2248 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2249 BUG_ON(!extent_op
|| !extent_op
->update_flags
);
2250 ret
= alloc_reserved_tree_block(trans
, root
,
2252 extent_op
->flags_to_set
,
2256 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2257 ret
= __btrfs_inc_extent_ref(trans
, root
, node
,
2261 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2262 ret
= __btrfs_free_extent(trans
, root
, node
,
2264 ref
->level
, 0, 1, extent_op
);
2271 /* helper function to actually process a single delayed ref entry */
2272 static int run_one_delayed_ref(struct btrfs_trans_handle
*trans
,
2273 struct btrfs_root
*root
,
2274 struct btrfs_delayed_ref_node
*node
,
2275 struct btrfs_delayed_extent_op
*extent_op
,
2276 int insert_reserved
)
2280 if (trans
->aborted
) {
2281 if (insert_reserved
)
2282 btrfs_pin_extent(root
, node
->bytenr
,
2283 node
->num_bytes
, 1);
2287 if (btrfs_delayed_ref_is_head(node
)) {
2288 struct btrfs_delayed_ref_head
*head
;
2290 * we've hit the end of the chain and we were supposed
2291 * to insert this extent into the tree. But, it got
2292 * deleted before we ever needed to insert it, so all
2293 * we have to do is clean up the accounting
2296 head
= btrfs_delayed_node_to_head(node
);
2297 trace_run_delayed_ref_head(node
, head
, node
->action
);
2299 if (insert_reserved
) {
2300 btrfs_pin_extent(root
, node
->bytenr
,
2301 node
->num_bytes
, 1);
2302 if (head
->is_data
) {
2303 ret
= btrfs_del_csums(trans
, root
,
2311 if (node
->type
== BTRFS_TREE_BLOCK_REF_KEY
||
2312 node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2313 ret
= run_delayed_tree_ref(trans
, root
, node
, extent_op
,
2315 else if (node
->type
== BTRFS_EXTENT_DATA_REF_KEY
||
2316 node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2317 ret
= run_delayed_data_ref(trans
, root
, node
, extent_op
,
2324 static inline struct btrfs_delayed_ref_node
*
2325 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
2327 if (list_empty(&head
->ref_list
))
2330 return list_entry(head
->ref_list
.next
, struct btrfs_delayed_ref_node
,
2335 * Returns 0 on success or if called with an already aborted transaction.
2336 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2338 static noinline
int __btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2339 struct btrfs_root
*root
,
2342 struct btrfs_delayed_ref_root
*delayed_refs
;
2343 struct btrfs_delayed_ref_node
*ref
;
2344 struct btrfs_delayed_ref_head
*locked_ref
= NULL
;
2345 struct btrfs_delayed_extent_op
*extent_op
;
2346 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2347 ktime_t start
= ktime_get();
2349 unsigned long count
= 0;
2350 unsigned long actual_count
= 0;
2351 int must_insert_reserved
= 0;
2353 delayed_refs
= &trans
->transaction
->delayed_refs
;
2359 spin_lock(&delayed_refs
->lock
);
2360 locked_ref
= btrfs_select_ref_head(trans
);
2362 spin_unlock(&delayed_refs
->lock
);
2366 /* grab the lock that says we are going to process
2367 * all the refs for this head */
2368 ret
= btrfs_delayed_ref_lock(trans
, locked_ref
);
2369 spin_unlock(&delayed_refs
->lock
);
2371 * we may have dropped the spin lock to get the head
2372 * mutex lock, and that might have given someone else
2373 * time to free the head. If that's true, it has been
2374 * removed from our list and we can move on.
2376 if (ret
== -EAGAIN
) {
2383 spin_lock(&locked_ref
->lock
);
2386 * locked_ref is the head node, so we have to go one
2387 * node back for any delayed ref updates
2389 ref
= select_delayed_ref(locked_ref
);
2391 if (ref
&& ref
->seq
&&
2392 btrfs_check_delayed_seq(fs_info
, delayed_refs
, ref
->seq
)) {
2393 spin_unlock(&locked_ref
->lock
);
2394 btrfs_delayed_ref_unlock(locked_ref
);
2395 spin_lock(&delayed_refs
->lock
);
2396 locked_ref
->processing
= 0;
2397 delayed_refs
->num_heads_ready
++;
2398 spin_unlock(&delayed_refs
->lock
);
2406 * record the must insert reserved flag before we
2407 * drop the spin lock.
2409 must_insert_reserved
= locked_ref
->must_insert_reserved
;
2410 locked_ref
->must_insert_reserved
= 0;
2412 extent_op
= locked_ref
->extent_op
;
2413 locked_ref
->extent_op
= NULL
;
2418 /* All delayed refs have been processed, Go ahead
2419 * and send the head node to run_one_delayed_ref,
2420 * so that any accounting fixes can happen
2422 ref
= &locked_ref
->node
;
2424 if (extent_op
&& must_insert_reserved
) {
2425 btrfs_free_delayed_extent_op(extent_op
);
2430 spin_unlock(&locked_ref
->lock
);
2431 ret
= run_delayed_extent_op(trans
, root
,
2433 btrfs_free_delayed_extent_op(extent_op
);
2437 * Need to reset must_insert_reserved if
2438 * there was an error so the abort stuff
2439 * can cleanup the reserved space
2442 if (must_insert_reserved
)
2443 locked_ref
->must_insert_reserved
= 1;
2444 locked_ref
->processing
= 0;
2445 btrfs_debug(fs_info
, "run_delayed_extent_op returned %d", ret
);
2446 btrfs_delayed_ref_unlock(locked_ref
);
2453 * Need to drop our head ref lock and re-aqcuire the
2454 * delayed ref lock and then re-check to make sure
2457 spin_unlock(&locked_ref
->lock
);
2458 spin_lock(&delayed_refs
->lock
);
2459 spin_lock(&locked_ref
->lock
);
2460 if (!list_empty(&locked_ref
->ref_list
) ||
2461 locked_ref
->extent_op
) {
2462 spin_unlock(&locked_ref
->lock
);
2463 spin_unlock(&delayed_refs
->lock
);
2467 delayed_refs
->num_heads
--;
2468 rb_erase(&locked_ref
->href_node
,
2469 &delayed_refs
->href_root
);
2470 spin_unlock(&delayed_refs
->lock
);
2474 list_del(&ref
->list
);
2476 atomic_dec(&delayed_refs
->num_entries
);
2478 if (!btrfs_delayed_ref_is_head(ref
)) {
2480 * when we play the delayed ref, also correct the
2483 switch (ref
->action
) {
2484 case BTRFS_ADD_DELAYED_REF
:
2485 case BTRFS_ADD_DELAYED_EXTENT
:
2486 locked_ref
->node
.ref_mod
-= ref
->ref_mod
;
2488 case BTRFS_DROP_DELAYED_REF
:
2489 locked_ref
->node
.ref_mod
+= ref
->ref_mod
;
2495 spin_unlock(&locked_ref
->lock
);
2497 ret
= run_one_delayed_ref(trans
, root
, ref
, extent_op
,
2498 must_insert_reserved
);
2500 btrfs_free_delayed_extent_op(extent_op
);
2502 locked_ref
->processing
= 0;
2503 btrfs_delayed_ref_unlock(locked_ref
);
2504 btrfs_put_delayed_ref(ref
);
2505 btrfs_debug(fs_info
, "run_one_delayed_ref returned %d", ret
);
2510 * If this node is a head, that means all the refs in this head
2511 * have been dealt with, and we will pick the next head to deal
2512 * with, so we must unlock the head and drop it from the cluster
2513 * list before we release it.
2515 if (btrfs_delayed_ref_is_head(ref
)) {
2516 if (locked_ref
->is_data
&&
2517 locked_ref
->total_ref_mod
< 0) {
2518 spin_lock(&delayed_refs
->lock
);
2519 delayed_refs
->pending_csums
-= ref
->num_bytes
;
2520 spin_unlock(&delayed_refs
->lock
);
2522 btrfs_delayed_ref_unlock(locked_ref
);
2525 btrfs_put_delayed_ref(ref
);
2531 * We don't want to include ref heads since we can have empty ref heads
2532 * and those will drastically skew our runtime down since we just do
2533 * accounting, no actual extent tree updates.
2535 if (actual_count
> 0) {
2536 u64 runtime
= ktime_to_ns(ktime_sub(ktime_get(), start
));
2540 * We weigh the current average higher than our current runtime
2541 * to avoid large swings in the average.
2543 spin_lock(&delayed_refs
->lock
);
2544 avg
= fs_info
->avg_delayed_ref_runtime
* 3 + runtime
;
2545 fs_info
->avg_delayed_ref_runtime
= avg
>> 2; /* div by 4 */
2546 spin_unlock(&delayed_refs
->lock
);
2551 #ifdef SCRAMBLE_DELAYED_REFS
2553 * Normally delayed refs get processed in ascending bytenr order. This
2554 * correlates in most cases to the order added. To expose dependencies on this
2555 * order, we start to process the tree in the middle instead of the beginning
2557 static u64
find_middle(struct rb_root
*root
)
2559 struct rb_node
*n
= root
->rb_node
;
2560 struct btrfs_delayed_ref_node
*entry
;
2563 u64 first
= 0, last
= 0;
2567 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2568 first
= entry
->bytenr
;
2572 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2573 last
= entry
->bytenr
;
2578 entry
= rb_entry(n
, struct btrfs_delayed_ref_node
, rb_node
);
2579 WARN_ON(!entry
->in_tree
);
2581 middle
= entry
->bytenr
;
2594 static inline u64
heads_to_leaves(struct btrfs_root
*root
, u64 heads
)
2598 num_bytes
= heads
* (sizeof(struct btrfs_extent_item
) +
2599 sizeof(struct btrfs_extent_inline_ref
));
2600 if (!btrfs_fs_incompat(root
->fs_info
, SKINNY_METADATA
))
2601 num_bytes
+= heads
* sizeof(struct btrfs_tree_block_info
);
2604 * We don't ever fill up leaves all the way so multiply by 2 just to be
2605 * closer to what we're really going to want to ouse.
2607 return div_u64(num_bytes
, BTRFS_LEAF_DATA_SIZE(root
));
2611 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2612 * would require to store the csums for that many bytes.
2614 u64
btrfs_csum_bytes_to_leaves(struct btrfs_root
*root
, u64 csum_bytes
)
2617 u64 num_csums_per_leaf
;
2620 csum_size
= BTRFS_LEAF_DATA_SIZE(root
) - sizeof(struct btrfs_item
);
2621 num_csums_per_leaf
= div64_u64(csum_size
,
2622 (u64
)btrfs_super_csum_size(root
->fs_info
->super_copy
));
2623 num_csums
= div64_u64(csum_bytes
, root
->sectorsize
);
2624 num_csums
+= num_csums_per_leaf
- 1;
2625 num_csums
= div64_u64(num_csums
, num_csums_per_leaf
);
2629 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle
*trans
,
2630 struct btrfs_root
*root
)
2632 struct btrfs_block_rsv
*global_rsv
;
2633 u64 num_heads
= trans
->transaction
->delayed_refs
.num_heads_ready
;
2634 u64 csum_bytes
= trans
->transaction
->delayed_refs
.pending_csums
;
2635 u64 num_dirty_bgs
= trans
->transaction
->num_dirty_bgs
;
2636 u64 num_bytes
, num_dirty_bgs_bytes
;
2639 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
2640 num_heads
= heads_to_leaves(root
, num_heads
);
2642 num_bytes
+= (num_heads
- 1) * root
->nodesize
;
2644 num_bytes
+= btrfs_csum_bytes_to_leaves(root
, csum_bytes
) * root
->nodesize
;
2645 num_dirty_bgs_bytes
= btrfs_calc_trans_metadata_size(root
,
2647 global_rsv
= &root
->fs_info
->global_block_rsv
;
2650 * If we can't allocate any more chunks lets make sure we have _lots_ of
2651 * wiggle room since running delayed refs can create more delayed refs.
2653 if (global_rsv
->space_info
->full
) {
2654 num_dirty_bgs_bytes
<<= 1;
2658 spin_lock(&global_rsv
->lock
);
2659 if (global_rsv
->reserved
<= num_bytes
+ num_dirty_bgs_bytes
)
2661 spin_unlock(&global_rsv
->lock
);
2665 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle
*trans
,
2666 struct btrfs_root
*root
)
2668 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2670 atomic_read(&trans
->transaction
->delayed_refs
.num_entries
);
2675 avg_runtime
= fs_info
->avg_delayed_ref_runtime
;
2676 val
= num_entries
* avg_runtime
;
2677 if (num_entries
* avg_runtime
>= NSEC_PER_SEC
)
2679 if (val
>= NSEC_PER_SEC
/ 2)
2682 return btrfs_check_space_for_delayed_refs(trans
, root
);
2685 struct async_delayed_refs
{
2686 struct btrfs_root
*root
;
2690 struct completion wait
;
2691 struct btrfs_work work
;
2694 static void delayed_ref_async_start(struct btrfs_work
*work
)
2696 struct async_delayed_refs
*async
;
2697 struct btrfs_trans_handle
*trans
;
2700 async
= container_of(work
, struct async_delayed_refs
, work
);
2702 trans
= btrfs_join_transaction(async
->root
);
2703 if (IS_ERR(trans
)) {
2704 async
->error
= PTR_ERR(trans
);
2709 * trans->sync means that when we call end_transaciton, we won't
2710 * wait on delayed refs
2713 ret
= btrfs_run_delayed_refs(trans
, async
->root
, async
->count
);
2717 ret
= btrfs_end_transaction(trans
, async
->root
);
2718 if (ret
&& !async
->error
)
2722 complete(&async
->wait
);
2727 int btrfs_async_run_delayed_refs(struct btrfs_root
*root
,
2728 unsigned long count
, int wait
)
2730 struct async_delayed_refs
*async
;
2733 async
= kmalloc(sizeof(*async
), GFP_NOFS
);
2737 async
->root
= root
->fs_info
->tree_root
;
2738 async
->count
= count
;
2744 init_completion(&async
->wait
);
2746 btrfs_init_work(&async
->work
, btrfs_extent_refs_helper
,
2747 delayed_ref_async_start
, NULL
, NULL
);
2749 btrfs_queue_work(root
->fs_info
->extent_workers
, &async
->work
);
2752 wait_for_completion(&async
->wait
);
2761 * this starts processing the delayed reference count updates and
2762 * extent insertions we have queued up so far. count can be
2763 * 0, which means to process everything in the tree at the start
2764 * of the run (but not newly added entries), or it can be some target
2765 * number you'd like to process.
2767 * Returns 0 on success or if called with an aborted transaction
2768 * Returns <0 on error and aborts the transaction
2770 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2771 struct btrfs_root
*root
, unsigned long count
)
2773 struct rb_node
*node
;
2774 struct btrfs_delayed_ref_root
*delayed_refs
;
2775 struct btrfs_delayed_ref_head
*head
;
2777 int run_all
= count
== (unsigned long)-1;
2779 /* We'll clean this up in btrfs_cleanup_transaction */
2783 if (root
== root
->fs_info
->extent_root
)
2784 root
= root
->fs_info
->tree_root
;
2786 delayed_refs
= &trans
->transaction
->delayed_refs
;
2788 count
= atomic_read(&delayed_refs
->num_entries
) * 2;
2791 #ifdef SCRAMBLE_DELAYED_REFS
2792 delayed_refs
->run_delayed_start
= find_middle(&delayed_refs
->root
);
2794 ret
= __btrfs_run_delayed_refs(trans
, root
, count
);
2796 btrfs_abort_transaction(trans
, root
, ret
);
2801 if (!list_empty(&trans
->new_bgs
))
2802 btrfs_create_pending_block_groups(trans
, root
);
2804 spin_lock(&delayed_refs
->lock
);
2805 node
= rb_first(&delayed_refs
->href_root
);
2807 spin_unlock(&delayed_refs
->lock
);
2810 count
= (unsigned long)-1;
2813 head
= rb_entry(node
, struct btrfs_delayed_ref_head
,
2815 if (btrfs_delayed_ref_is_head(&head
->node
)) {
2816 struct btrfs_delayed_ref_node
*ref
;
2819 atomic_inc(&ref
->refs
);
2821 spin_unlock(&delayed_refs
->lock
);
2823 * Mutex was contended, block until it's
2824 * released and try again
2826 mutex_lock(&head
->mutex
);
2827 mutex_unlock(&head
->mutex
);
2829 btrfs_put_delayed_ref(ref
);
2835 node
= rb_next(node
);
2837 spin_unlock(&delayed_refs
->lock
);
2842 ret
= btrfs_delayed_qgroup_accounting(trans
, root
->fs_info
);
2845 assert_qgroups_uptodate(trans
);
2849 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
2850 struct btrfs_root
*root
,
2851 u64 bytenr
, u64 num_bytes
, u64 flags
,
2852 int level
, int is_data
)
2854 struct btrfs_delayed_extent_op
*extent_op
;
2857 extent_op
= btrfs_alloc_delayed_extent_op();
2861 extent_op
->flags_to_set
= flags
;
2862 extent_op
->update_flags
= 1;
2863 extent_op
->update_key
= 0;
2864 extent_op
->is_data
= is_data
? 1 : 0;
2865 extent_op
->level
= level
;
2867 ret
= btrfs_add_delayed_extent_op(root
->fs_info
, trans
, bytenr
,
2868 num_bytes
, extent_op
);
2870 btrfs_free_delayed_extent_op(extent_op
);
2874 static noinline
int check_delayed_ref(struct btrfs_trans_handle
*trans
,
2875 struct btrfs_root
*root
,
2876 struct btrfs_path
*path
,
2877 u64 objectid
, u64 offset
, u64 bytenr
)
2879 struct btrfs_delayed_ref_head
*head
;
2880 struct btrfs_delayed_ref_node
*ref
;
2881 struct btrfs_delayed_data_ref
*data_ref
;
2882 struct btrfs_delayed_ref_root
*delayed_refs
;
2885 delayed_refs
= &trans
->transaction
->delayed_refs
;
2886 spin_lock(&delayed_refs
->lock
);
2887 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
2889 spin_unlock(&delayed_refs
->lock
);
2893 if (!mutex_trylock(&head
->mutex
)) {
2894 atomic_inc(&head
->node
.refs
);
2895 spin_unlock(&delayed_refs
->lock
);
2897 btrfs_release_path(path
);
2900 * Mutex was contended, block until it's released and let
2903 mutex_lock(&head
->mutex
);
2904 mutex_unlock(&head
->mutex
);
2905 btrfs_put_delayed_ref(&head
->node
);
2908 spin_unlock(&delayed_refs
->lock
);
2910 spin_lock(&head
->lock
);
2911 list_for_each_entry(ref
, &head
->ref_list
, list
) {
2912 /* If it's a shared ref we know a cross reference exists */
2913 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
) {
2918 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
2921 * If our ref doesn't match the one we're currently looking at
2922 * then we have a cross reference.
2924 if (data_ref
->root
!= root
->root_key
.objectid
||
2925 data_ref
->objectid
!= objectid
||
2926 data_ref
->offset
!= offset
) {
2931 spin_unlock(&head
->lock
);
2932 mutex_unlock(&head
->mutex
);
2936 static noinline
int check_committed_ref(struct btrfs_trans_handle
*trans
,
2937 struct btrfs_root
*root
,
2938 struct btrfs_path
*path
,
2939 u64 objectid
, u64 offset
, u64 bytenr
)
2941 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2942 struct extent_buffer
*leaf
;
2943 struct btrfs_extent_data_ref
*ref
;
2944 struct btrfs_extent_inline_ref
*iref
;
2945 struct btrfs_extent_item
*ei
;
2946 struct btrfs_key key
;
2950 key
.objectid
= bytenr
;
2951 key
.offset
= (u64
)-1;
2952 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2954 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
2957 BUG_ON(ret
== 0); /* Corruption */
2960 if (path
->slots
[0] == 0)
2964 leaf
= path
->nodes
[0];
2965 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2967 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
2971 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2972 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2973 if (item_size
< sizeof(*ei
)) {
2974 WARN_ON(item_size
!= sizeof(struct btrfs_extent_item_v0
));
2978 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2980 if (item_size
!= sizeof(*ei
) +
2981 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
2984 if (btrfs_extent_generation(leaf
, ei
) <=
2985 btrfs_root_last_snapshot(&root
->root_item
))
2988 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
2989 if (btrfs_extent_inline_ref_type(leaf
, iref
) !=
2990 BTRFS_EXTENT_DATA_REF_KEY
)
2993 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
2994 if (btrfs_extent_refs(leaf
, ei
) !=
2995 btrfs_extent_data_ref_count(leaf
, ref
) ||
2996 btrfs_extent_data_ref_root(leaf
, ref
) !=
2997 root
->root_key
.objectid
||
2998 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
2999 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
3007 int btrfs_cross_ref_exist(struct btrfs_trans_handle
*trans
,
3008 struct btrfs_root
*root
,
3009 u64 objectid
, u64 offset
, u64 bytenr
)
3011 struct btrfs_path
*path
;
3015 path
= btrfs_alloc_path();
3020 ret
= check_committed_ref(trans
, root
, path
, objectid
,
3022 if (ret
&& ret
!= -ENOENT
)
3025 ret2
= check_delayed_ref(trans
, root
, path
, objectid
,
3027 } while (ret2
== -EAGAIN
);
3029 if (ret2
&& ret2
!= -ENOENT
) {
3034 if (ret
!= -ENOENT
|| ret2
!= -ENOENT
)
3037 btrfs_free_path(path
);
3038 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
3043 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
3044 struct btrfs_root
*root
,
3045 struct extent_buffer
*buf
,
3046 int full_backref
, int inc
)
3053 struct btrfs_key key
;
3054 struct btrfs_file_extent_item
*fi
;
3058 int (*process_func
)(struct btrfs_trans_handle
*, struct btrfs_root
*,
3059 u64
, u64
, u64
, u64
, u64
, u64
, int);
3062 if (btrfs_test_is_dummy_root(root
))
3065 ref_root
= btrfs_header_owner(buf
);
3066 nritems
= btrfs_header_nritems(buf
);
3067 level
= btrfs_header_level(buf
);
3069 if (!test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) && level
== 0)
3073 process_func
= btrfs_inc_extent_ref
;
3075 process_func
= btrfs_free_extent
;
3078 parent
= buf
->start
;
3082 for (i
= 0; i
< nritems
; i
++) {
3084 btrfs_item_key_to_cpu(buf
, &key
, i
);
3085 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
3087 fi
= btrfs_item_ptr(buf
, i
,
3088 struct btrfs_file_extent_item
);
3089 if (btrfs_file_extent_type(buf
, fi
) ==
3090 BTRFS_FILE_EXTENT_INLINE
)
3092 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
3096 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
3097 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
3098 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
3099 parent
, ref_root
, key
.objectid
,
3104 bytenr
= btrfs_node_blockptr(buf
, i
);
3105 num_bytes
= root
->nodesize
;
3106 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
3107 parent
, ref_root
, level
- 1, 0,
3118 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3119 struct extent_buffer
*buf
, int full_backref
)
3121 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1);
3124 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3125 struct extent_buffer
*buf
, int full_backref
)
3127 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0);
3130 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
3131 struct btrfs_root
*root
,
3132 struct btrfs_path
*path
,
3133 struct btrfs_block_group_cache
*cache
)
3136 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
3138 struct extent_buffer
*leaf
;
3140 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
3147 leaf
= path
->nodes
[0];
3148 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
3149 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
3150 btrfs_mark_buffer_dirty(leaf
);
3152 btrfs_release_path(path
);
3157 static struct btrfs_block_group_cache
*
3158 next_block_group(struct btrfs_root
*root
,
3159 struct btrfs_block_group_cache
*cache
)
3161 struct rb_node
*node
;
3163 spin_lock(&root
->fs_info
->block_group_cache_lock
);
3165 /* If our block group was removed, we need a full search. */
3166 if (RB_EMPTY_NODE(&cache
->cache_node
)) {
3167 const u64 next_bytenr
= cache
->key
.objectid
+ cache
->key
.offset
;
3169 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
3170 btrfs_put_block_group(cache
);
3171 cache
= btrfs_lookup_first_block_group(root
->fs_info
,
3175 node
= rb_next(&cache
->cache_node
);
3176 btrfs_put_block_group(cache
);
3178 cache
= rb_entry(node
, struct btrfs_block_group_cache
,
3180 btrfs_get_block_group(cache
);
3183 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
3187 static int cache_save_setup(struct btrfs_block_group_cache
*block_group
,
3188 struct btrfs_trans_handle
*trans
,
3189 struct btrfs_path
*path
)
3191 struct btrfs_root
*root
= block_group
->fs_info
->tree_root
;
3192 struct inode
*inode
= NULL
;
3194 int dcs
= BTRFS_DC_ERROR
;
3200 * If this block group is smaller than 100 megs don't bother caching the
3203 if (block_group
->key
.offset
< (100 * 1024 * 1024)) {
3204 spin_lock(&block_group
->lock
);
3205 block_group
->disk_cache_state
= BTRFS_DC_WRITTEN
;
3206 spin_unlock(&block_group
->lock
);
3213 inode
= lookup_free_space_inode(root
, block_group
, path
);
3214 if (IS_ERR(inode
) && PTR_ERR(inode
) != -ENOENT
) {
3215 ret
= PTR_ERR(inode
);
3216 btrfs_release_path(path
);
3220 if (IS_ERR(inode
)) {
3224 if (block_group
->ro
)
3227 ret
= create_free_space_inode(root
, trans
, block_group
, path
);
3233 /* We've already setup this transaction, go ahead and exit */
3234 if (block_group
->cache_generation
== trans
->transid
&&
3235 i_size_read(inode
)) {
3236 dcs
= BTRFS_DC_SETUP
;
3241 * We want to set the generation to 0, that way if anything goes wrong
3242 * from here on out we know not to trust this cache when we load up next
3245 BTRFS_I(inode
)->generation
= 0;
3246 ret
= btrfs_update_inode(trans
, root
, inode
);
3249 * So theoretically we could recover from this, simply set the
3250 * super cache generation to 0 so we know to invalidate the
3251 * cache, but then we'd have to keep track of the block groups
3252 * that fail this way so we know we _have_ to reset this cache
3253 * before the next commit or risk reading stale cache. So to
3254 * limit our exposure to horrible edge cases lets just abort the
3255 * transaction, this only happens in really bad situations
3258 btrfs_abort_transaction(trans
, root
, ret
);
3263 if (i_size_read(inode
) > 0) {
3264 ret
= btrfs_check_trunc_cache_free_space(root
,
3265 &root
->fs_info
->global_block_rsv
);
3269 ret
= btrfs_truncate_free_space_cache(root
, trans
, NULL
, inode
);
3274 spin_lock(&block_group
->lock
);
3275 if (block_group
->cached
!= BTRFS_CACHE_FINISHED
||
3276 !btrfs_test_opt(root
, SPACE_CACHE
)) {
3278 * don't bother trying to write stuff out _if_
3279 * a) we're not cached,
3280 * b) we're with nospace_cache mount option.
3282 dcs
= BTRFS_DC_WRITTEN
;
3283 spin_unlock(&block_group
->lock
);
3286 spin_unlock(&block_group
->lock
);
3289 * Try to preallocate enough space based on how big the block group is.
3290 * Keep in mind this has to include any pinned space which could end up
3291 * taking up quite a bit since it's not folded into the other space
3294 num_pages
= div_u64(block_group
->key
.offset
, 256 * 1024 * 1024);
3299 num_pages
*= PAGE_CACHE_SIZE
;
3301 ret
= btrfs_check_data_free_space(inode
, num_pages
, num_pages
);
3305 ret
= btrfs_prealloc_file_range_trans(inode
, trans
, 0, 0, num_pages
,
3306 num_pages
, num_pages
,
3309 dcs
= BTRFS_DC_SETUP
;
3310 btrfs_free_reserved_data_space(inode
, num_pages
);
3315 btrfs_release_path(path
);
3317 spin_lock(&block_group
->lock
);
3318 if (!ret
&& dcs
== BTRFS_DC_SETUP
)
3319 block_group
->cache_generation
= trans
->transid
;
3320 block_group
->disk_cache_state
= dcs
;
3321 spin_unlock(&block_group
->lock
);
3326 int btrfs_setup_space_cache(struct btrfs_trans_handle
*trans
,
3327 struct btrfs_root
*root
)
3329 struct btrfs_block_group_cache
*cache
, *tmp
;
3330 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3331 struct btrfs_path
*path
;
3333 if (list_empty(&cur_trans
->dirty_bgs
) ||
3334 !btrfs_test_opt(root
, SPACE_CACHE
))
3337 path
= btrfs_alloc_path();
3341 /* Could add new block groups, use _safe just in case */
3342 list_for_each_entry_safe(cache
, tmp
, &cur_trans
->dirty_bgs
,
3344 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
)
3345 cache_save_setup(cache
, trans
, path
);
3348 btrfs_free_path(path
);
3353 * transaction commit does final block group cache writeback during a
3354 * critical section where nothing is allowed to change the FS. This is
3355 * required in order for the cache to actually match the block group,
3356 * but can introduce a lot of latency into the commit.
3358 * So, btrfs_start_dirty_block_groups is here to kick off block group
3359 * cache IO. There's a chance we'll have to redo some of it if the
3360 * block group changes again during the commit, but it greatly reduces
3361 * the commit latency by getting rid of the easy block groups while
3362 * we're still allowing others to join the commit.
3364 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle
*trans
,
3365 struct btrfs_root
*root
)
3367 struct btrfs_block_group_cache
*cache
;
3368 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3371 struct btrfs_path
*path
= NULL
;
3373 struct list_head
*io
= &cur_trans
->io_bgs
;
3374 int num_started
= 0;
3377 spin_lock(&cur_trans
->dirty_bgs_lock
);
3378 if (list_empty(&cur_trans
->dirty_bgs
)) {
3379 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3382 list_splice_init(&cur_trans
->dirty_bgs
, &dirty
);
3383 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3387 * make sure all the block groups on our dirty list actually
3390 btrfs_create_pending_block_groups(trans
, root
);
3393 path
= btrfs_alloc_path();
3399 * cache_write_mutex is here only to save us from balance or automatic
3400 * removal of empty block groups deleting this block group while we are
3401 * writing out the cache
3403 mutex_lock(&trans
->transaction
->cache_write_mutex
);
3404 while (!list_empty(&dirty
)) {
3405 cache
= list_first_entry(&dirty
,
3406 struct btrfs_block_group_cache
,
3409 * this can happen if something re-dirties a block
3410 * group that is already under IO. Just wait for it to
3411 * finish and then do it all again
3413 if (!list_empty(&cache
->io_list
)) {
3414 list_del_init(&cache
->io_list
);
3415 btrfs_wait_cache_io(root
, trans
, cache
,
3416 &cache
->io_ctl
, path
,
3417 cache
->key
.objectid
);
3418 btrfs_put_block_group(cache
);
3423 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3424 * if it should update the cache_state. Don't delete
3425 * until after we wait.
3427 * Since we're not running in the commit critical section
3428 * we need the dirty_bgs_lock to protect from update_block_group
3430 spin_lock(&cur_trans
->dirty_bgs_lock
);
3431 list_del_init(&cache
->dirty_list
);
3432 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3436 cache_save_setup(cache
, trans
, path
);
3438 if (cache
->disk_cache_state
== BTRFS_DC_SETUP
) {
3439 cache
->io_ctl
.inode
= NULL
;
3440 ret
= btrfs_write_out_cache(root
, trans
, cache
, path
);
3441 if (ret
== 0 && cache
->io_ctl
.inode
) {
3446 * the cache_write_mutex is protecting
3449 list_add_tail(&cache
->io_list
, io
);
3452 * if we failed to write the cache, the
3453 * generation will be bad and life goes on
3459 ret
= write_one_cache_group(trans
, root
, path
, cache
);
3461 * Our block group might still be attached to the list
3462 * of new block groups in the transaction handle of some
3463 * other task (struct btrfs_trans_handle->new_bgs). This
3464 * means its block group item isn't yet in the extent
3465 * tree. If this happens ignore the error, as we will
3466 * try again later in the critical section of the
3467 * transaction commit.
3469 if (ret
== -ENOENT
) {
3471 spin_lock(&cur_trans
->dirty_bgs_lock
);
3472 if (list_empty(&cache
->dirty_list
)) {
3473 list_add_tail(&cache
->dirty_list
,
3474 &cur_trans
->dirty_bgs
);
3475 btrfs_get_block_group(cache
);
3477 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3479 btrfs_abort_transaction(trans
, root
, ret
);
3483 /* if its not on the io list, we need to put the block group */
3485 btrfs_put_block_group(cache
);
3491 * Avoid blocking other tasks for too long. It might even save
3492 * us from writing caches for block groups that are going to be
3495 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
3496 mutex_lock(&trans
->transaction
->cache_write_mutex
);
3498 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
3501 * go through delayed refs for all the stuff we've just kicked off
3502 * and then loop back (just once)
3504 ret
= btrfs_run_delayed_refs(trans
, root
, 0);
3505 if (!ret
&& loops
== 0) {
3507 spin_lock(&cur_trans
->dirty_bgs_lock
);
3508 list_splice_init(&cur_trans
->dirty_bgs
, &dirty
);
3510 * dirty_bgs_lock protects us from concurrent block group
3511 * deletes too (not just cache_write_mutex).
3513 if (!list_empty(&dirty
)) {
3514 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3517 spin_unlock(&cur_trans
->dirty_bgs_lock
);
3520 btrfs_free_path(path
);
3524 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
3525 struct btrfs_root
*root
)
3527 struct btrfs_block_group_cache
*cache
;
3528 struct btrfs_transaction
*cur_trans
= trans
->transaction
;
3531 struct btrfs_path
*path
;
3532 struct list_head
*io
= &cur_trans
->io_bgs
;
3533 int num_started
= 0;
3535 path
= btrfs_alloc_path();
3540 * We don't need the lock here since we are protected by the transaction
3541 * commit. We want to do the cache_save_setup first and then run the
3542 * delayed refs to make sure we have the best chance at doing this all
3545 while (!list_empty(&cur_trans
->dirty_bgs
)) {
3546 cache
= list_first_entry(&cur_trans
->dirty_bgs
,
3547 struct btrfs_block_group_cache
,
3551 * this can happen if cache_save_setup re-dirties a block
3552 * group that is already under IO. Just wait for it to
3553 * finish and then do it all again
3555 if (!list_empty(&cache
->io_list
)) {
3556 list_del_init(&cache
->io_list
);
3557 btrfs_wait_cache_io(root
, trans
, cache
,
3558 &cache
->io_ctl
, path
,
3559 cache
->key
.objectid
);
3560 btrfs_put_block_group(cache
);
3564 * don't remove from the dirty list until after we've waited
3567 list_del_init(&cache
->dirty_list
);
3570 cache_save_setup(cache
, trans
, path
);
3573 ret
= btrfs_run_delayed_refs(trans
, root
, (unsigned long) -1);
3575 if (!ret
&& cache
->disk_cache_state
== BTRFS_DC_SETUP
) {
3576 cache
->io_ctl
.inode
= NULL
;
3577 ret
= btrfs_write_out_cache(root
, trans
, cache
, path
);
3578 if (ret
== 0 && cache
->io_ctl
.inode
) {
3581 list_add_tail(&cache
->io_list
, io
);
3584 * if we failed to write the cache, the
3585 * generation will be bad and life goes on
3591 ret
= write_one_cache_group(trans
, root
, path
, cache
);
3593 btrfs_abort_transaction(trans
, root
, ret
);
3596 /* if its not on the io list, we need to put the block group */
3598 btrfs_put_block_group(cache
);
3601 while (!list_empty(io
)) {
3602 cache
= list_first_entry(io
, struct btrfs_block_group_cache
,
3604 list_del_init(&cache
->io_list
);
3605 btrfs_wait_cache_io(root
, trans
, cache
,
3606 &cache
->io_ctl
, path
, cache
->key
.objectid
);
3607 btrfs_put_block_group(cache
);
3610 btrfs_free_path(path
);
3614 int btrfs_extent_readonly(struct btrfs_root
*root
, u64 bytenr
)
3616 struct btrfs_block_group_cache
*block_group
;
3619 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
3620 if (!block_group
|| block_group
->ro
)
3623 btrfs_put_block_group(block_group
);
3627 static const char *alloc_name(u64 flags
)
3630 case BTRFS_BLOCK_GROUP_METADATA
|BTRFS_BLOCK_GROUP_DATA
:
3632 case BTRFS_BLOCK_GROUP_METADATA
:
3634 case BTRFS_BLOCK_GROUP_DATA
:
3636 case BTRFS_BLOCK_GROUP_SYSTEM
:
3640 return "invalid-combination";
3644 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
3645 u64 total_bytes
, u64 bytes_used
,
3646 struct btrfs_space_info
**space_info
)
3648 struct btrfs_space_info
*found
;
3653 if (flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3654 BTRFS_BLOCK_GROUP_RAID10
))
3659 found
= __find_space_info(info
, flags
);
3661 spin_lock(&found
->lock
);
3662 found
->total_bytes
+= total_bytes
;
3663 found
->disk_total
+= total_bytes
* factor
;
3664 found
->bytes_used
+= bytes_used
;
3665 found
->disk_used
+= bytes_used
* factor
;
3666 if (total_bytes
> 0)
3668 spin_unlock(&found
->lock
);
3669 *space_info
= found
;
3672 found
= kzalloc(sizeof(*found
), GFP_NOFS
);
3676 ret
= percpu_counter_init(&found
->total_bytes_pinned
, 0, GFP_KERNEL
);
3682 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
3683 INIT_LIST_HEAD(&found
->block_groups
[i
]);
3684 init_rwsem(&found
->groups_sem
);
3685 spin_lock_init(&found
->lock
);
3686 found
->flags
= flags
& BTRFS_BLOCK_GROUP_TYPE_MASK
;
3687 found
->total_bytes
= total_bytes
;
3688 found
->disk_total
= total_bytes
* factor
;
3689 found
->bytes_used
= bytes_used
;
3690 found
->disk_used
= bytes_used
* factor
;
3691 found
->bytes_pinned
= 0;
3692 found
->bytes_reserved
= 0;
3693 found
->bytes_readonly
= 0;
3694 found
->bytes_may_use
= 0;
3695 if (total_bytes
> 0)
3699 found
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3700 found
->chunk_alloc
= 0;
3702 init_waitqueue_head(&found
->wait
);
3703 INIT_LIST_HEAD(&found
->ro_bgs
);
3705 ret
= kobject_init_and_add(&found
->kobj
, &space_info_ktype
,
3706 info
->space_info_kobj
, "%s",
3707 alloc_name(found
->flags
));
3713 *space_info
= found
;
3714 list_add_rcu(&found
->list
, &info
->space_info
);
3715 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3716 info
->data_sinfo
= found
;
3721 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
3723 u64 extra_flags
= chunk_to_extended(flags
) &
3724 BTRFS_EXTENDED_PROFILE_MASK
;
3726 write_seqlock(&fs_info
->profiles_lock
);
3727 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3728 fs_info
->avail_data_alloc_bits
|= extra_flags
;
3729 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3730 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
3731 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3732 fs_info
->avail_system_alloc_bits
|= extra_flags
;
3733 write_sequnlock(&fs_info
->profiles_lock
);
3737 * returns target flags in extended format or 0 if restripe for this
3738 * chunk_type is not in progress
3740 * should be called with either volume_mutex or balance_lock held
3742 static u64
get_restripe_target(struct btrfs_fs_info
*fs_info
, u64 flags
)
3744 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3750 if (flags
& BTRFS_BLOCK_GROUP_DATA
&&
3751 bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3752 target
= BTRFS_BLOCK_GROUP_DATA
| bctl
->data
.target
;
3753 } else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
&&
3754 bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3755 target
= BTRFS_BLOCK_GROUP_SYSTEM
| bctl
->sys
.target
;
3756 } else if (flags
& BTRFS_BLOCK_GROUP_METADATA
&&
3757 bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3758 target
= BTRFS_BLOCK_GROUP_METADATA
| bctl
->meta
.target
;
3765 * @flags: available profiles in extended format (see ctree.h)
3767 * Returns reduced profile in chunk format. If profile changing is in
3768 * progress (either running or paused) picks the target profile (if it's
3769 * already available), otherwise falls back to plain reducing.
3771 static u64
btrfs_reduce_alloc_profile(struct btrfs_root
*root
, u64 flags
)
3773 u64 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
3778 * see if restripe for this chunk_type is in progress, if so
3779 * try to reduce to the target profile
3781 spin_lock(&root
->fs_info
->balance_lock
);
3782 target
= get_restripe_target(root
->fs_info
, flags
);
3784 /* pick target profile only if it's already available */
3785 if ((flags
& target
) & BTRFS_EXTENDED_PROFILE_MASK
) {
3786 spin_unlock(&root
->fs_info
->balance_lock
);
3787 return extended_to_chunk(target
);
3790 spin_unlock(&root
->fs_info
->balance_lock
);
3792 /* First, mask out the RAID levels which aren't possible */
3793 if (num_devices
== 1)
3794 flags
&= ~(BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID0
|
3795 BTRFS_BLOCK_GROUP_RAID5
);
3796 if (num_devices
< 3)
3797 flags
&= ~BTRFS_BLOCK_GROUP_RAID6
;
3798 if (num_devices
< 4)
3799 flags
&= ~BTRFS_BLOCK_GROUP_RAID10
;
3801 tmp
= flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID0
|
3802 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID5
|
3803 BTRFS_BLOCK_GROUP_RAID6
| BTRFS_BLOCK_GROUP_RAID10
);
3806 if (tmp
& BTRFS_BLOCK_GROUP_RAID6
)
3807 tmp
= BTRFS_BLOCK_GROUP_RAID6
;
3808 else if (tmp
& BTRFS_BLOCK_GROUP_RAID5
)
3809 tmp
= BTRFS_BLOCK_GROUP_RAID5
;
3810 else if (tmp
& BTRFS_BLOCK_GROUP_RAID10
)
3811 tmp
= BTRFS_BLOCK_GROUP_RAID10
;
3812 else if (tmp
& BTRFS_BLOCK_GROUP_RAID1
)
3813 tmp
= BTRFS_BLOCK_GROUP_RAID1
;
3814 else if (tmp
& BTRFS_BLOCK_GROUP_RAID0
)
3815 tmp
= BTRFS_BLOCK_GROUP_RAID0
;
3817 return extended_to_chunk(flags
| tmp
);
3820 static u64
get_alloc_profile(struct btrfs_root
*root
, u64 orig_flags
)
3827 seq
= read_seqbegin(&root
->fs_info
->profiles_lock
);
3829 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3830 flags
|= root
->fs_info
->avail_data_alloc_bits
;
3831 else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3832 flags
|= root
->fs_info
->avail_system_alloc_bits
;
3833 else if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3834 flags
|= root
->fs_info
->avail_metadata_alloc_bits
;
3835 } while (read_seqretry(&root
->fs_info
->profiles_lock
, seq
));
3837 return btrfs_reduce_alloc_profile(root
, flags
);
3840 u64
btrfs_get_alloc_profile(struct btrfs_root
*root
, int data
)
3846 flags
= BTRFS_BLOCK_GROUP_DATA
;
3847 else if (root
== root
->fs_info
->chunk_root
)
3848 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
3850 flags
= BTRFS_BLOCK_GROUP_METADATA
;
3852 ret
= get_alloc_profile(root
, flags
);
3857 * This will check the space that the inode allocates from to make sure we have
3858 * enough space for bytes.
3860 int btrfs_check_data_free_space(struct inode
*inode
, u64 bytes
, u64 write_bytes
)
3862 struct btrfs_space_info
*data_sinfo
;
3863 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3864 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3867 int need_commit
= 2;
3868 int have_pinned_space
;
3870 /* make sure bytes are sectorsize aligned */
3871 bytes
= ALIGN(bytes
, root
->sectorsize
);
3873 if (btrfs_is_free_space_inode(inode
)) {
3875 ASSERT(current
->journal_info
);
3878 data_sinfo
= fs_info
->data_sinfo
;
3883 /* make sure we have enough space to handle the data first */
3884 spin_lock(&data_sinfo
->lock
);
3885 used
= data_sinfo
->bytes_used
+ data_sinfo
->bytes_reserved
+
3886 data_sinfo
->bytes_pinned
+ data_sinfo
->bytes_readonly
+
3887 data_sinfo
->bytes_may_use
;
3889 if (used
+ bytes
> data_sinfo
->total_bytes
) {
3890 struct btrfs_trans_handle
*trans
;
3893 * if we don't have enough free bytes in this space then we need
3894 * to alloc a new chunk.
3896 if (!data_sinfo
->full
) {
3899 data_sinfo
->force_alloc
= CHUNK_ALLOC_FORCE
;
3900 spin_unlock(&data_sinfo
->lock
);
3902 alloc_target
= btrfs_get_alloc_profile(root
, 1);
3904 * It is ugly that we don't call nolock join
3905 * transaction for the free space inode case here.
3906 * But it is safe because we only do the data space
3907 * reservation for the free space cache in the
3908 * transaction context, the common join transaction
3909 * just increase the counter of the current transaction
3910 * handler, doesn't try to acquire the trans_lock of
3913 trans
= btrfs_join_transaction(root
);
3915 return PTR_ERR(trans
);
3917 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
3919 CHUNK_ALLOC_NO_FORCE
);
3920 btrfs_end_transaction(trans
, root
);
3925 have_pinned_space
= 1;
3931 data_sinfo
= fs_info
->data_sinfo
;
3937 * If we don't have enough pinned space to deal with this
3938 * allocation, and no removed chunk in current transaction,
3939 * don't bother committing the transaction.
3941 have_pinned_space
= percpu_counter_compare(
3942 &data_sinfo
->total_bytes_pinned
,
3943 used
+ bytes
- data_sinfo
->total_bytes
);
3944 spin_unlock(&data_sinfo
->lock
);
3946 /* commit the current transaction and try again */
3949 !atomic_read(&root
->fs_info
->open_ioctl_trans
)) {
3952 trans
= btrfs_join_transaction(root
);
3954 return PTR_ERR(trans
);
3955 if (have_pinned_space
>= 0 ||
3956 trans
->transaction
->have_free_bgs
||
3958 ret
= btrfs_commit_transaction(trans
, root
);
3962 * make sure that all running delayed iput are
3965 down_write(&root
->fs_info
->delayed_iput_sem
);
3966 up_write(&root
->fs_info
->delayed_iput_sem
);
3969 btrfs_end_transaction(trans
, root
);
3973 trace_btrfs_space_reservation(root
->fs_info
,
3974 "space_info:enospc",
3975 data_sinfo
->flags
, bytes
, 1);
3978 ret
= btrfs_qgroup_reserve(root
, write_bytes
);
3981 data_sinfo
->bytes_may_use
+= bytes
;
3982 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
3983 data_sinfo
->flags
, bytes
, 1);
3985 spin_unlock(&data_sinfo
->lock
);
3991 * Called if we need to clear a data reservation for this inode.
3993 void btrfs_free_reserved_data_space(struct inode
*inode
, u64 bytes
)
3995 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3996 struct btrfs_space_info
*data_sinfo
;
3998 /* make sure bytes are sectorsize aligned */
3999 bytes
= ALIGN(bytes
, root
->sectorsize
);
4001 data_sinfo
= root
->fs_info
->data_sinfo
;
4002 spin_lock(&data_sinfo
->lock
);
4003 WARN_ON(data_sinfo
->bytes_may_use
< bytes
);
4004 data_sinfo
->bytes_may_use
-= bytes
;
4005 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
4006 data_sinfo
->flags
, bytes
, 0);
4007 spin_unlock(&data_sinfo
->lock
);
4010 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
4012 struct list_head
*head
= &info
->space_info
;
4013 struct btrfs_space_info
*found
;
4016 list_for_each_entry_rcu(found
, head
, list
) {
4017 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
4018 found
->force_alloc
= CHUNK_ALLOC_FORCE
;
4023 static inline u64
calc_global_rsv_need_space(struct btrfs_block_rsv
*global
)
4025 return (global
->size
<< 1);
4028 static int should_alloc_chunk(struct btrfs_root
*root
,
4029 struct btrfs_space_info
*sinfo
, int force
)
4031 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
4032 u64 num_bytes
= sinfo
->total_bytes
- sinfo
->bytes_readonly
;
4033 u64 num_allocated
= sinfo
->bytes_used
+ sinfo
->bytes_reserved
;
4036 if (force
== CHUNK_ALLOC_FORCE
)
4040 * We need to take into account the global rsv because for all intents
4041 * and purposes it's used space. Don't worry about locking the
4042 * global_rsv, it doesn't change except when the transaction commits.
4044 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
4045 num_allocated
+= calc_global_rsv_need_space(global_rsv
);
4048 * in limited mode, we want to have some free space up to
4049 * about 1% of the FS size.
4051 if (force
== CHUNK_ALLOC_LIMITED
) {
4052 thresh
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
4053 thresh
= max_t(u64
, 64 * 1024 * 1024,
4054 div_factor_fine(thresh
, 1));
4056 if (num_bytes
- num_allocated
< thresh
)
4060 if (num_allocated
+ 2 * 1024 * 1024 < div_factor(num_bytes
, 8))
4065 static u64
get_profile_num_devs(struct btrfs_root
*root
, u64 type
)
4069 if (type
& (BTRFS_BLOCK_GROUP_RAID10
|
4070 BTRFS_BLOCK_GROUP_RAID0
|
4071 BTRFS_BLOCK_GROUP_RAID5
|
4072 BTRFS_BLOCK_GROUP_RAID6
))
4073 num_dev
= root
->fs_info
->fs_devices
->rw_devices
;
4074 else if (type
& BTRFS_BLOCK_GROUP_RAID1
)
4077 num_dev
= 1; /* DUP or single */
4083 * If @is_allocation is true, reserve space in the system space info necessary
4084 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4087 void check_system_chunk(struct btrfs_trans_handle
*trans
,
4088 struct btrfs_root
*root
,
4091 struct btrfs_space_info
*info
;
4098 * Needed because we can end up allocating a system chunk and for an
4099 * atomic and race free space reservation in the chunk block reserve.
4101 ASSERT(mutex_is_locked(&root
->fs_info
->chunk_mutex
));
4103 info
= __find_space_info(root
->fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
4104 spin_lock(&info
->lock
);
4105 left
= info
->total_bytes
- info
->bytes_used
- info
->bytes_pinned
-
4106 info
->bytes_reserved
- info
->bytes_readonly
-
4107 info
->bytes_may_use
;
4108 spin_unlock(&info
->lock
);
4110 num_devs
= get_profile_num_devs(root
, type
);
4112 /* num_devs device items to update and 1 chunk item to add or remove */
4113 thresh
= btrfs_calc_trunc_metadata_size(root
, num_devs
) +
4114 btrfs_calc_trans_metadata_size(root
, 1);
4116 if (left
< thresh
&& btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
4117 btrfs_info(root
->fs_info
, "left=%llu, need=%llu, flags=%llu",
4118 left
, thresh
, type
);
4119 dump_space_info(info
, 0, 0);
4122 if (left
< thresh
) {
4125 flags
= btrfs_get_alloc_profile(root
->fs_info
->chunk_root
, 0);
4127 * Ignore failure to create system chunk. We might end up not
4128 * needing it, as we might not need to COW all nodes/leafs from
4129 * the paths we visit in the chunk tree (they were already COWed
4130 * or created in the current transaction for example).
4132 ret
= btrfs_alloc_chunk(trans
, root
, flags
);
4136 ret
= btrfs_block_rsv_add(root
->fs_info
->chunk_root
,
4137 &root
->fs_info
->chunk_block_rsv
,
4138 thresh
, BTRFS_RESERVE_NO_FLUSH
);
4140 trans
->chunk_bytes_reserved
+= thresh
;
4144 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
4145 struct btrfs_root
*extent_root
, u64 flags
, int force
)
4147 struct btrfs_space_info
*space_info
;
4148 struct btrfs_fs_info
*fs_info
= extent_root
->fs_info
;
4149 int wait_for_alloc
= 0;
4152 /* Don't re-enter if we're already allocating a chunk */
4153 if (trans
->allocating_chunk
)
4156 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
4158 ret
= update_space_info(extent_root
->fs_info
, flags
,
4160 BUG_ON(ret
); /* -ENOMEM */
4162 BUG_ON(!space_info
); /* Logic error */
4165 spin_lock(&space_info
->lock
);
4166 if (force
< space_info
->force_alloc
)
4167 force
= space_info
->force_alloc
;
4168 if (space_info
->full
) {
4169 if (should_alloc_chunk(extent_root
, space_info
, force
))
4173 spin_unlock(&space_info
->lock
);
4177 if (!should_alloc_chunk(extent_root
, space_info
, force
)) {
4178 spin_unlock(&space_info
->lock
);
4180 } else if (space_info
->chunk_alloc
) {
4183 space_info
->chunk_alloc
= 1;
4186 spin_unlock(&space_info
->lock
);
4188 mutex_lock(&fs_info
->chunk_mutex
);
4191 * The chunk_mutex is held throughout the entirety of a chunk
4192 * allocation, so once we've acquired the chunk_mutex we know that the
4193 * other guy is done and we need to recheck and see if we should
4196 if (wait_for_alloc
) {
4197 mutex_unlock(&fs_info
->chunk_mutex
);
4202 trans
->allocating_chunk
= true;
4205 * If we have mixed data/metadata chunks we want to make sure we keep
4206 * allocating mixed chunks instead of individual chunks.
4208 if (btrfs_mixed_space_info(space_info
))
4209 flags
|= (BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
);
4212 * if we're doing a data chunk, go ahead and make sure that
4213 * we keep a reasonable number of metadata chunks allocated in the
4216 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
4217 fs_info
->data_chunk_allocations
++;
4218 if (!(fs_info
->data_chunk_allocations
%
4219 fs_info
->metadata_ratio
))
4220 force_metadata_allocation(fs_info
);
4224 * Check if we have enough space in SYSTEM chunk because we may need
4225 * to update devices.
4227 check_system_chunk(trans
, extent_root
, flags
);
4229 ret
= btrfs_alloc_chunk(trans
, extent_root
, flags
);
4230 trans
->allocating_chunk
= false;
4232 spin_lock(&space_info
->lock
);
4233 if (ret
< 0 && ret
!= -ENOSPC
)
4236 space_info
->full
= 1;
4240 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
4242 space_info
->chunk_alloc
= 0;
4243 spin_unlock(&space_info
->lock
);
4244 mutex_unlock(&fs_info
->chunk_mutex
);
4248 static int can_overcommit(struct btrfs_root
*root
,
4249 struct btrfs_space_info
*space_info
, u64 bytes
,
4250 enum btrfs_reserve_flush_enum flush
)
4252 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
4253 u64 profile
= btrfs_get_alloc_profile(root
, 0);
4258 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4259 space_info
->bytes_pinned
+ space_info
->bytes_readonly
;
4262 * We only want to allow over committing if we have lots of actual space
4263 * free, but if we don't have enough space to handle the global reserve
4264 * space then we could end up having a real enospc problem when trying
4265 * to allocate a chunk or some other such important allocation.
4267 spin_lock(&global_rsv
->lock
);
4268 space_size
= calc_global_rsv_need_space(global_rsv
);
4269 spin_unlock(&global_rsv
->lock
);
4270 if (used
+ space_size
>= space_info
->total_bytes
)
4273 used
+= space_info
->bytes_may_use
;
4275 spin_lock(&root
->fs_info
->free_chunk_lock
);
4276 avail
= root
->fs_info
->free_chunk_space
;
4277 spin_unlock(&root
->fs_info
->free_chunk_lock
);
4280 * If we have dup, raid1 or raid10 then only half of the free
4281 * space is actually useable. For raid56, the space info used
4282 * doesn't include the parity drive, so we don't have to
4285 if (profile
& (BTRFS_BLOCK_GROUP_DUP
|
4286 BTRFS_BLOCK_GROUP_RAID1
|
4287 BTRFS_BLOCK_GROUP_RAID10
))
4291 * If we aren't flushing all things, let us overcommit up to
4292 * 1/2th of the space. If we can flush, don't let us overcommit
4293 * too much, let it overcommit up to 1/8 of the space.
4295 if (flush
== BTRFS_RESERVE_FLUSH_ALL
)
4300 if (used
+ bytes
< space_info
->total_bytes
+ avail
)
4305 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root
*root
,
4306 unsigned long nr_pages
, int nr_items
)
4308 struct super_block
*sb
= root
->fs_info
->sb
;
4310 if (down_read_trylock(&sb
->s_umount
)) {
4311 writeback_inodes_sb_nr(sb
, nr_pages
, WB_REASON_FS_FREE_SPACE
);
4312 up_read(&sb
->s_umount
);
4315 * We needn't worry the filesystem going from r/w to r/o though
4316 * we don't acquire ->s_umount mutex, because the filesystem
4317 * should guarantee the delalloc inodes list be empty after
4318 * the filesystem is readonly(all dirty pages are written to
4321 btrfs_start_delalloc_roots(root
->fs_info
, 0, nr_items
);
4322 if (!current
->journal_info
)
4323 btrfs_wait_ordered_roots(root
->fs_info
, nr_items
);
4327 static inline int calc_reclaim_items_nr(struct btrfs_root
*root
, u64 to_reclaim
)
4332 bytes
= btrfs_calc_trans_metadata_size(root
, 1);
4333 nr
= (int)div64_u64(to_reclaim
, bytes
);
4339 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4342 * shrink metadata reservation for delalloc
4344 static void shrink_delalloc(struct btrfs_root
*root
, u64 to_reclaim
, u64 orig
,
4347 struct btrfs_block_rsv
*block_rsv
;
4348 struct btrfs_space_info
*space_info
;
4349 struct btrfs_trans_handle
*trans
;
4353 unsigned long nr_pages
;
4356 enum btrfs_reserve_flush_enum flush
;
4358 /* Calc the number of the pages we need flush for space reservation */
4359 items
= calc_reclaim_items_nr(root
, to_reclaim
);
4360 to_reclaim
= items
* EXTENT_SIZE_PER_ITEM
;
4362 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4363 block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
4364 space_info
= block_rsv
->space_info
;
4366 delalloc_bytes
= percpu_counter_sum_positive(
4367 &root
->fs_info
->delalloc_bytes
);
4368 if (delalloc_bytes
== 0) {
4372 btrfs_wait_ordered_roots(root
->fs_info
, items
);
4377 while (delalloc_bytes
&& loops
< 3) {
4378 max_reclaim
= min(delalloc_bytes
, to_reclaim
);
4379 nr_pages
= max_reclaim
>> PAGE_CACHE_SHIFT
;
4380 btrfs_writeback_inodes_sb_nr(root
, nr_pages
, items
);
4382 * We need to wait for the async pages to actually start before
4385 max_reclaim
= atomic_read(&root
->fs_info
->async_delalloc_pages
);
4389 if (max_reclaim
<= nr_pages
)
4392 max_reclaim
-= nr_pages
;
4394 wait_event(root
->fs_info
->async_submit_wait
,
4395 atomic_read(&root
->fs_info
->async_delalloc_pages
) <=
4399 flush
= BTRFS_RESERVE_FLUSH_ALL
;
4401 flush
= BTRFS_RESERVE_NO_FLUSH
;
4402 spin_lock(&space_info
->lock
);
4403 if (can_overcommit(root
, space_info
, orig
, flush
)) {
4404 spin_unlock(&space_info
->lock
);
4407 spin_unlock(&space_info
->lock
);
4410 if (wait_ordered
&& !trans
) {
4411 btrfs_wait_ordered_roots(root
->fs_info
, items
);
4413 time_left
= schedule_timeout_killable(1);
4417 delalloc_bytes
= percpu_counter_sum_positive(
4418 &root
->fs_info
->delalloc_bytes
);
4423 * maybe_commit_transaction - possibly commit the transaction if its ok to
4424 * @root - the root we're allocating for
4425 * @bytes - the number of bytes we want to reserve
4426 * @force - force the commit
4428 * This will check to make sure that committing the transaction will actually
4429 * get us somewhere and then commit the transaction if it does. Otherwise it
4430 * will return -ENOSPC.
4432 static int may_commit_transaction(struct btrfs_root
*root
,
4433 struct btrfs_space_info
*space_info
,
4434 u64 bytes
, int force
)
4436 struct btrfs_block_rsv
*delayed_rsv
= &root
->fs_info
->delayed_block_rsv
;
4437 struct btrfs_trans_handle
*trans
;
4439 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
4446 /* See if there is enough pinned space to make this reservation */
4447 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4452 * See if there is some space in the delayed insertion reservation for
4455 if (space_info
!= delayed_rsv
->space_info
)
4458 spin_lock(&delayed_rsv
->lock
);
4459 if (percpu_counter_compare(&space_info
->total_bytes_pinned
,
4460 bytes
- delayed_rsv
->size
) >= 0) {
4461 spin_unlock(&delayed_rsv
->lock
);
4464 spin_unlock(&delayed_rsv
->lock
);
4467 trans
= btrfs_join_transaction(root
);
4471 return btrfs_commit_transaction(trans
, root
);
4475 FLUSH_DELAYED_ITEMS_NR
= 1,
4476 FLUSH_DELAYED_ITEMS
= 2,
4478 FLUSH_DELALLOC_WAIT
= 4,
4483 static int flush_space(struct btrfs_root
*root
,
4484 struct btrfs_space_info
*space_info
, u64 num_bytes
,
4485 u64 orig_bytes
, int state
)
4487 struct btrfs_trans_handle
*trans
;
4492 case FLUSH_DELAYED_ITEMS_NR
:
4493 case FLUSH_DELAYED_ITEMS
:
4494 if (state
== FLUSH_DELAYED_ITEMS_NR
)
4495 nr
= calc_reclaim_items_nr(root
, num_bytes
) * 2;
4499 trans
= btrfs_join_transaction(root
);
4500 if (IS_ERR(trans
)) {
4501 ret
= PTR_ERR(trans
);
4504 ret
= btrfs_run_delayed_items_nr(trans
, root
, nr
);
4505 btrfs_end_transaction(trans
, root
);
4507 case FLUSH_DELALLOC
:
4508 case FLUSH_DELALLOC_WAIT
:
4509 shrink_delalloc(root
, num_bytes
* 2, orig_bytes
,
4510 state
== FLUSH_DELALLOC_WAIT
);
4513 trans
= btrfs_join_transaction(root
);
4514 if (IS_ERR(trans
)) {
4515 ret
= PTR_ERR(trans
);
4518 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
4519 btrfs_get_alloc_profile(root
, 0),
4520 CHUNK_ALLOC_NO_FORCE
);
4521 btrfs_end_transaction(trans
, root
);
4526 ret
= may_commit_transaction(root
, space_info
, orig_bytes
, 0);
4537 btrfs_calc_reclaim_metadata_size(struct btrfs_root
*root
,
4538 struct btrfs_space_info
*space_info
)
4544 to_reclaim
= min_t(u64
, num_online_cpus() * 1024 * 1024,
4546 spin_lock(&space_info
->lock
);
4547 if (can_overcommit(root
, space_info
, to_reclaim
,
4548 BTRFS_RESERVE_FLUSH_ALL
)) {
4553 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4554 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4555 space_info
->bytes_may_use
;
4556 if (can_overcommit(root
, space_info
, 1024 * 1024,
4557 BTRFS_RESERVE_FLUSH_ALL
))
4558 expected
= div_factor_fine(space_info
->total_bytes
, 95);
4560 expected
= div_factor_fine(space_info
->total_bytes
, 90);
4562 if (used
> expected
)
4563 to_reclaim
= used
- expected
;
4566 to_reclaim
= min(to_reclaim
, space_info
->bytes_may_use
+
4567 space_info
->bytes_reserved
);
4569 spin_unlock(&space_info
->lock
);
4574 static inline int need_do_async_reclaim(struct btrfs_space_info
*space_info
,
4575 struct btrfs_fs_info
*fs_info
, u64 used
)
4577 u64 thresh
= div_factor_fine(space_info
->total_bytes
, 98);
4579 /* If we're just plain full then async reclaim just slows us down. */
4580 if (space_info
->bytes_used
>= thresh
)
4583 return (used
>= thresh
&& !btrfs_fs_closing(fs_info
) &&
4584 !test_bit(BTRFS_FS_STATE_REMOUNTING
, &fs_info
->fs_state
));
4587 static int btrfs_need_do_async_reclaim(struct btrfs_space_info
*space_info
,
4588 struct btrfs_fs_info
*fs_info
,
4593 spin_lock(&space_info
->lock
);
4595 * We run out of space and have not got any free space via flush_space,
4596 * so don't bother doing async reclaim.
4598 if (flush_state
> COMMIT_TRANS
&& space_info
->full
) {
4599 spin_unlock(&space_info
->lock
);
4603 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4604 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4605 space_info
->bytes_may_use
;
4606 if (need_do_async_reclaim(space_info
, fs_info
, used
)) {
4607 spin_unlock(&space_info
->lock
);
4610 spin_unlock(&space_info
->lock
);
4615 static void btrfs_async_reclaim_metadata_space(struct work_struct
*work
)
4617 struct btrfs_fs_info
*fs_info
;
4618 struct btrfs_space_info
*space_info
;
4622 fs_info
= container_of(work
, struct btrfs_fs_info
, async_reclaim_work
);
4623 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4625 to_reclaim
= btrfs_calc_reclaim_metadata_size(fs_info
->fs_root
,
4630 flush_state
= FLUSH_DELAYED_ITEMS_NR
;
4632 flush_space(fs_info
->fs_root
, space_info
, to_reclaim
,
4633 to_reclaim
, flush_state
);
4635 if (!btrfs_need_do_async_reclaim(space_info
, fs_info
,
4638 } while (flush_state
< COMMIT_TRANS
);
4641 void btrfs_init_async_reclaim_work(struct work_struct
*work
)
4643 INIT_WORK(work
, btrfs_async_reclaim_metadata_space
);
4647 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4648 * @root - the root we're allocating for
4649 * @block_rsv - the block_rsv we're allocating for
4650 * @orig_bytes - the number of bytes we want
4651 * @flush - whether or not we can flush to make our reservation
4653 * This will reserve orgi_bytes number of bytes from the space info associated
4654 * with the block_rsv. If there is not enough space it will make an attempt to
4655 * flush out space to make room. It will do this by flushing delalloc if
4656 * possible or committing the transaction. If flush is 0 then no attempts to
4657 * regain reservations will be made and this will fail if there is not enough
4660 static int reserve_metadata_bytes(struct btrfs_root
*root
,
4661 struct btrfs_block_rsv
*block_rsv
,
4663 enum btrfs_reserve_flush_enum flush
)
4665 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
4667 u64 num_bytes
= orig_bytes
;
4668 int flush_state
= FLUSH_DELAYED_ITEMS_NR
;
4670 bool flushing
= false;
4674 spin_lock(&space_info
->lock
);
4676 * We only want to wait if somebody other than us is flushing and we
4677 * are actually allowed to flush all things.
4679 while (flush
== BTRFS_RESERVE_FLUSH_ALL
&& !flushing
&&
4680 space_info
->flush
) {
4681 spin_unlock(&space_info
->lock
);
4683 * If we have a trans handle we can't wait because the flusher
4684 * may have to commit the transaction, which would mean we would
4685 * deadlock since we are waiting for the flusher to finish, but
4686 * hold the current transaction open.
4688 if (current
->journal_info
)
4690 ret
= wait_event_killable(space_info
->wait
, !space_info
->flush
);
4691 /* Must have been killed, return */
4695 spin_lock(&space_info
->lock
);
4699 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
4700 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
4701 space_info
->bytes_may_use
;
4704 * The idea here is that we've not already over-reserved the block group
4705 * then we can go ahead and save our reservation first and then start
4706 * flushing if we need to. Otherwise if we've already overcommitted
4707 * lets start flushing stuff first and then come back and try to make
4710 if (used
<= space_info
->total_bytes
) {
4711 if (used
+ orig_bytes
<= space_info
->total_bytes
) {
4712 space_info
->bytes_may_use
+= orig_bytes
;
4713 trace_btrfs_space_reservation(root
->fs_info
,
4714 "space_info", space_info
->flags
, orig_bytes
, 1);
4718 * Ok set num_bytes to orig_bytes since we aren't
4719 * overocmmitted, this way we only try and reclaim what
4722 num_bytes
= orig_bytes
;
4726 * Ok we're over committed, set num_bytes to the overcommitted
4727 * amount plus the amount of bytes that we need for this
4730 num_bytes
= used
- space_info
->total_bytes
+
4734 if (ret
&& can_overcommit(root
, space_info
, orig_bytes
, flush
)) {
4735 space_info
->bytes_may_use
+= orig_bytes
;
4736 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
4737 space_info
->flags
, orig_bytes
,
4743 * Couldn't make our reservation, save our place so while we're trying
4744 * to reclaim space we can actually use it instead of somebody else
4745 * stealing it from us.
4747 * We make the other tasks wait for the flush only when we can flush
4750 if (ret
&& flush
!= BTRFS_RESERVE_NO_FLUSH
) {
4752 space_info
->flush
= 1;
4753 } else if (!ret
&& space_info
->flags
& BTRFS_BLOCK_GROUP_METADATA
) {
4756 * We will do the space reservation dance during log replay,
4757 * which means we won't have fs_info->fs_root set, so don't do
4758 * the async reclaim as we will panic.
4760 if (!root
->fs_info
->log_root_recovering
&&
4761 need_do_async_reclaim(space_info
, root
->fs_info
, used
) &&
4762 !work_busy(&root
->fs_info
->async_reclaim_work
))
4763 queue_work(system_unbound_wq
,
4764 &root
->fs_info
->async_reclaim_work
);
4766 spin_unlock(&space_info
->lock
);
4768 if (!ret
|| flush
== BTRFS_RESERVE_NO_FLUSH
)
4771 ret
= flush_space(root
, space_info
, num_bytes
, orig_bytes
,
4776 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4777 * would happen. So skip delalloc flush.
4779 if (flush
== BTRFS_RESERVE_FLUSH_LIMIT
&&
4780 (flush_state
== FLUSH_DELALLOC
||
4781 flush_state
== FLUSH_DELALLOC_WAIT
))
4782 flush_state
= ALLOC_CHUNK
;
4786 else if (flush
== BTRFS_RESERVE_FLUSH_LIMIT
&&
4787 flush_state
< COMMIT_TRANS
)
4789 else if (flush
== BTRFS_RESERVE_FLUSH_ALL
&&
4790 flush_state
<= COMMIT_TRANS
)
4794 if (ret
== -ENOSPC
&&
4795 unlikely(root
->orphan_cleanup_state
== ORPHAN_CLEANUP_STARTED
)) {
4796 struct btrfs_block_rsv
*global_rsv
=
4797 &root
->fs_info
->global_block_rsv
;
4799 if (block_rsv
!= global_rsv
&&
4800 !block_rsv_use_bytes(global_rsv
, orig_bytes
))
4804 trace_btrfs_space_reservation(root
->fs_info
,
4805 "space_info:enospc",
4806 space_info
->flags
, orig_bytes
, 1);
4808 spin_lock(&space_info
->lock
);
4809 space_info
->flush
= 0;
4810 wake_up_all(&space_info
->wait
);
4811 spin_unlock(&space_info
->lock
);
4816 static struct btrfs_block_rsv
*get_block_rsv(
4817 const struct btrfs_trans_handle
*trans
,
4818 const struct btrfs_root
*root
)
4820 struct btrfs_block_rsv
*block_rsv
= NULL
;
4822 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
))
4823 block_rsv
= trans
->block_rsv
;
4825 if (root
== root
->fs_info
->csum_root
&& trans
->adding_csums
)
4826 block_rsv
= trans
->block_rsv
;
4828 if (root
== root
->fs_info
->uuid_root
)
4829 block_rsv
= trans
->block_rsv
;
4832 block_rsv
= root
->block_rsv
;
4835 block_rsv
= &root
->fs_info
->empty_block_rsv
;
4840 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
4844 spin_lock(&block_rsv
->lock
);
4845 if (block_rsv
->reserved
>= num_bytes
) {
4846 block_rsv
->reserved
-= num_bytes
;
4847 if (block_rsv
->reserved
< block_rsv
->size
)
4848 block_rsv
->full
= 0;
4851 spin_unlock(&block_rsv
->lock
);
4855 static void block_rsv_add_bytes(struct btrfs_block_rsv
*block_rsv
,
4856 u64 num_bytes
, int update_size
)
4858 spin_lock(&block_rsv
->lock
);
4859 block_rsv
->reserved
+= num_bytes
;
4861 block_rsv
->size
+= num_bytes
;
4862 else if (block_rsv
->reserved
>= block_rsv
->size
)
4863 block_rsv
->full
= 1;
4864 spin_unlock(&block_rsv
->lock
);
4867 int btrfs_cond_migrate_bytes(struct btrfs_fs_info
*fs_info
,
4868 struct btrfs_block_rsv
*dest
, u64 num_bytes
,
4871 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
4874 if (global_rsv
->space_info
!= dest
->space_info
)
4877 spin_lock(&global_rsv
->lock
);
4878 min_bytes
= div_factor(global_rsv
->size
, min_factor
);
4879 if (global_rsv
->reserved
< min_bytes
+ num_bytes
) {
4880 spin_unlock(&global_rsv
->lock
);
4883 global_rsv
->reserved
-= num_bytes
;
4884 if (global_rsv
->reserved
< global_rsv
->size
)
4885 global_rsv
->full
= 0;
4886 spin_unlock(&global_rsv
->lock
);
4888 block_rsv_add_bytes(dest
, num_bytes
, 1);
4892 static void block_rsv_release_bytes(struct btrfs_fs_info
*fs_info
,
4893 struct btrfs_block_rsv
*block_rsv
,
4894 struct btrfs_block_rsv
*dest
, u64 num_bytes
)
4896 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
4898 spin_lock(&block_rsv
->lock
);
4899 if (num_bytes
== (u64
)-1)
4900 num_bytes
= block_rsv
->size
;
4901 block_rsv
->size
-= num_bytes
;
4902 if (block_rsv
->reserved
>= block_rsv
->size
) {
4903 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
4904 block_rsv
->reserved
= block_rsv
->size
;
4905 block_rsv
->full
= 1;
4909 spin_unlock(&block_rsv
->lock
);
4911 if (num_bytes
> 0) {
4913 spin_lock(&dest
->lock
);
4917 bytes_to_add
= dest
->size
- dest
->reserved
;
4918 bytes_to_add
= min(num_bytes
, bytes_to_add
);
4919 dest
->reserved
+= bytes_to_add
;
4920 if (dest
->reserved
>= dest
->size
)
4922 num_bytes
-= bytes_to_add
;
4924 spin_unlock(&dest
->lock
);
4927 spin_lock(&space_info
->lock
);
4928 space_info
->bytes_may_use
-= num_bytes
;
4929 trace_btrfs_space_reservation(fs_info
, "space_info",
4930 space_info
->flags
, num_bytes
, 0);
4931 spin_unlock(&space_info
->lock
);
4936 static int block_rsv_migrate_bytes(struct btrfs_block_rsv
*src
,
4937 struct btrfs_block_rsv
*dst
, u64 num_bytes
)
4941 ret
= block_rsv_use_bytes(src
, num_bytes
);
4945 block_rsv_add_bytes(dst
, num_bytes
, 1);
4949 void btrfs_init_block_rsv(struct btrfs_block_rsv
*rsv
, unsigned short type
)
4951 memset(rsv
, 0, sizeof(*rsv
));
4952 spin_lock_init(&rsv
->lock
);
4956 struct btrfs_block_rsv
*btrfs_alloc_block_rsv(struct btrfs_root
*root
,
4957 unsigned short type
)
4959 struct btrfs_block_rsv
*block_rsv
;
4960 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4962 block_rsv
= kmalloc(sizeof(*block_rsv
), GFP_NOFS
);
4966 btrfs_init_block_rsv(block_rsv
, type
);
4967 block_rsv
->space_info
= __find_space_info(fs_info
,
4968 BTRFS_BLOCK_GROUP_METADATA
);
4972 void btrfs_free_block_rsv(struct btrfs_root
*root
,
4973 struct btrfs_block_rsv
*rsv
)
4977 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
4981 void __btrfs_free_block_rsv(struct btrfs_block_rsv
*rsv
)
4986 int btrfs_block_rsv_add(struct btrfs_root
*root
,
4987 struct btrfs_block_rsv
*block_rsv
, u64 num_bytes
,
4988 enum btrfs_reserve_flush_enum flush
)
4995 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
4997 block_rsv_add_bytes(block_rsv
, num_bytes
, 1);
5004 int btrfs_block_rsv_check(struct btrfs_root
*root
,
5005 struct btrfs_block_rsv
*block_rsv
, int min_factor
)
5013 spin_lock(&block_rsv
->lock
);
5014 num_bytes
= div_factor(block_rsv
->size
, min_factor
);
5015 if (block_rsv
->reserved
>= num_bytes
)
5017 spin_unlock(&block_rsv
->lock
);
5022 int btrfs_block_rsv_refill(struct btrfs_root
*root
,
5023 struct btrfs_block_rsv
*block_rsv
, u64 min_reserved
,
5024 enum btrfs_reserve_flush_enum flush
)
5032 spin_lock(&block_rsv
->lock
);
5033 num_bytes
= min_reserved
;
5034 if (block_rsv
->reserved
>= num_bytes
)
5037 num_bytes
-= block_rsv
->reserved
;
5038 spin_unlock(&block_rsv
->lock
);
5043 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
5045 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
5052 int btrfs_block_rsv_migrate(struct btrfs_block_rsv
*src_rsv
,
5053 struct btrfs_block_rsv
*dst_rsv
,
5056 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
5059 void btrfs_block_rsv_release(struct btrfs_root
*root
,
5060 struct btrfs_block_rsv
*block_rsv
,
5063 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
5064 if (global_rsv
== block_rsv
||
5065 block_rsv
->space_info
!= global_rsv
->space_info
)
5067 block_rsv_release_bytes(root
->fs_info
, block_rsv
, global_rsv
,
5072 * helper to calculate size of global block reservation.
5073 * the desired value is sum of space used by extent tree,
5074 * checksum tree and root tree
5076 static u64
calc_global_metadata_size(struct btrfs_fs_info
*fs_info
)
5078 struct btrfs_space_info
*sinfo
;
5082 int csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
5084 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_DATA
);
5085 spin_lock(&sinfo
->lock
);
5086 data_used
= sinfo
->bytes_used
;
5087 spin_unlock(&sinfo
->lock
);
5089 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
5090 spin_lock(&sinfo
->lock
);
5091 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_DATA
)
5093 meta_used
= sinfo
->bytes_used
;
5094 spin_unlock(&sinfo
->lock
);
5096 num_bytes
= (data_used
>> fs_info
->sb
->s_blocksize_bits
) *
5098 num_bytes
+= div_u64(data_used
+ meta_used
, 50);
5100 if (num_bytes
* 3 > meta_used
)
5101 num_bytes
= div_u64(meta_used
, 3);
5103 return ALIGN(num_bytes
, fs_info
->extent_root
->nodesize
<< 10);
5106 static void update_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5108 struct btrfs_block_rsv
*block_rsv
= &fs_info
->global_block_rsv
;
5109 struct btrfs_space_info
*sinfo
= block_rsv
->space_info
;
5112 num_bytes
= calc_global_metadata_size(fs_info
);
5114 spin_lock(&sinfo
->lock
);
5115 spin_lock(&block_rsv
->lock
);
5117 block_rsv
->size
= min_t(u64
, num_bytes
, 512 * 1024 * 1024);
5119 num_bytes
= sinfo
->bytes_used
+ sinfo
->bytes_pinned
+
5120 sinfo
->bytes_reserved
+ sinfo
->bytes_readonly
+
5121 sinfo
->bytes_may_use
;
5123 if (sinfo
->total_bytes
> num_bytes
) {
5124 num_bytes
= sinfo
->total_bytes
- num_bytes
;
5125 block_rsv
->reserved
+= num_bytes
;
5126 sinfo
->bytes_may_use
+= num_bytes
;
5127 trace_btrfs_space_reservation(fs_info
, "space_info",
5128 sinfo
->flags
, num_bytes
, 1);
5131 if (block_rsv
->reserved
>= block_rsv
->size
) {
5132 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
5133 sinfo
->bytes_may_use
-= num_bytes
;
5134 trace_btrfs_space_reservation(fs_info
, "space_info",
5135 sinfo
->flags
, num_bytes
, 0);
5136 block_rsv
->reserved
= block_rsv
->size
;
5137 block_rsv
->full
= 1;
5140 spin_unlock(&block_rsv
->lock
);
5141 spin_unlock(&sinfo
->lock
);
5144 static void init_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5146 struct btrfs_space_info
*space_info
;
5148 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
5149 fs_info
->chunk_block_rsv
.space_info
= space_info
;
5151 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
5152 fs_info
->global_block_rsv
.space_info
= space_info
;
5153 fs_info
->delalloc_block_rsv
.space_info
= space_info
;
5154 fs_info
->trans_block_rsv
.space_info
= space_info
;
5155 fs_info
->empty_block_rsv
.space_info
= space_info
;
5156 fs_info
->delayed_block_rsv
.space_info
= space_info
;
5158 fs_info
->extent_root
->block_rsv
= &fs_info
->global_block_rsv
;
5159 fs_info
->csum_root
->block_rsv
= &fs_info
->global_block_rsv
;
5160 fs_info
->dev_root
->block_rsv
= &fs_info
->global_block_rsv
;
5161 fs_info
->tree_root
->block_rsv
= &fs_info
->global_block_rsv
;
5162 if (fs_info
->quota_root
)
5163 fs_info
->quota_root
->block_rsv
= &fs_info
->global_block_rsv
;
5164 fs_info
->chunk_root
->block_rsv
= &fs_info
->chunk_block_rsv
;
5166 update_global_block_rsv(fs_info
);
5169 static void release_global_block_rsv(struct btrfs_fs_info
*fs_info
)
5171 block_rsv_release_bytes(fs_info
, &fs_info
->global_block_rsv
, NULL
,
5173 WARN_ON(fs_info
->delalloc_block_rsv
.size
> 0);
5174 WARN_ON(fs_info
->delalloc_block_rsv
.reserved
> 0);
5175 WARN_ON(fs_info
->trans_block_rsv
.size
> 0);
5176 WARN_ON(fs_info
->trans_block_rsv
.reserved
> 0);
5177 WARN_ON(fs_info
->chunk_block_rsv
.size
> 0);
5178 WARN_ON(fs_info
->chunk_block_rsv
.reserved
> 0);
5179 WARN_ON(fs_info
->delayed_block_rsv
.size
> 0);
5180 WARN_ON(fs_info
->delayed_block_rsv
.reserved
> 0);
5183 void btrfs_trans_release_metadata(struct btrfs_trans_handle
*trans
,
5184 struct btrfs_root
*root
)
5186 if (!trans
->block_rsv
)
5189 if (!trans
->bytes_reserved
)
5192 trace_btrfs_space_reservation(root
->fs_info
, "transaction",
5193 trans
->transid
, trans
->bytes_reserved
, 0);
5194 btrfs_block_rsv_release(root
, trans
->block_rsv
, trans
->bytes_reserved
);
5195 trans
->bytes_reserved
= 0;
5199 * To be called after all the new block groups attached to the transaction
5200 * handle have been created (btrfs_create_pending_block_groups()).
5202 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle
*trans
)
5204 struct btrfs_fs_info
*fs_info
= trans
->root
->fs_info
;
5206 if (!trans
->chunk_bytes_reserved
)
5209 WARN_ON_ONCE(!list_empty(&trans
->new_bgs
));
5211 block_rsv_release_bytes(fs_info
, &fs_info
->chunk_block_rsv
, NULL
,
5212 trans
->chunk_bytes_reserved
);
5213 trans
->chunk_bytes_reserved
= 0;
5216 /* Can only return 0 or -ENOSPC */
5217 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle
*trans
,
5218 struct inode
*inode
)
5220 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5221 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
5222 struct btrfs_block_rsv
*dst_rsv
= root
->orphan_block_rsv
;
5225 * We need to hold space in order to delete our orphan item once we've
5226 * added it, so this takes the reservation so we can release it later
5227 * when we are truly done with the orphan item.
5229 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
5230 trace_btrfs_space_reservation(root
->fs_info
, "orphan",
5231 btrfs_ino(inode
), num_bytes
, 1);
5232 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
5235 void btrfs_orphan_release_metadata(struct inode
*inode
)
5237 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5238 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
5239 trace_btrfs_space_reservation(root
->fs_info
, "orphan",
5240 btrfs_ino(inode
), num_bytes
, 0);
5241 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
, num_bytes
);
5245 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5246 * root: the root of the parent directory
5247 * rsv: block reservation
5248 * items: the number of items that we need do reservation
5249 * qgroup_reserved: used to return the reserved size in qgroup
5251 * This function is used to reserve the space for snapshot/subvolume
5252 * creation and deletion. Those operations are different with the
5253 * common file/directory operations, they change two fs/file trees
5254 * and root tree, the number of items that the qgroup reserves is
5255 * different with the free space reservation. So we can not use
5256 * the space reseravtion mechanism in start_transaction().
5258 int btrfs_subvolume_reserve_metadata(struct btrfs_root
*root
,
5259 struct btrfs_block_rsv
*rsv
,
5261 u64
*qgroup_reserved
,
5262 bool use_global_rsv
)
5266 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
5268 if (root
->fs_info
->quota_enabled
) {
5269 /* One for parent inode, two for dir entries */
5270 num_bytes
= 3 * root
->nodesize
;
5271 ret
= btrfs_qgroup_reserve(root
, num_bytes
);
5278 *qgroup_reserved
= num_bytes
;
5280 num_bytes
= btrfs_calc_trans_metadata_size(root
, items
);
5281 rsv
->space_info
= __find_space_info(root
->fs_info
,
5282 BTRFS_BLOCK_GROUP_METADATA
);
5283 ret
= btrfs_block_rsv_add(root
, rsv
, num_bytes
,
5284 BTRFS_RESERVE_FLUSH_ALL
);
5286 if (ret
== -ENOSPC
&& use_global_rsv
)
5287 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
, num_bytes
);
5290 if (*qgroup_reserved
)
5291 btrfs_qgroup_free(root
, *qgroup_reserved
);
5297 void btrfs_subvolume_release_metadata(struct btrfs_root
*root
,
5298 struct btrfs_block_rsv
*rsv
,
5299 u64 qgroup_reserved
)
5301 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
5305 * drop_outstanding_extent - drop an outstanding extent
5306 * @inode: the inode we're dropping the extent for
5307 * @num_bytes: the number of bytes we're relaseing.
5309 * This is called when we are freeing up an outstanding extent, either called
5310 * after an error or after an extent is written. This will return the number of
5311 * reserved extents that need to be freed. This must be called with
5312 * BTRFS_I(inode)->lock held.
5314 static unsigned drop_outstanding_extent(struct inode
*inode
, u64 num_bytes
)
5316 unsigned drop_inode_space
= 0;
5317 unsigned dropped_extents
= 0;
5318 unsigned num_extents
= 0;
5320 num_extents
= (unsigned)div64_u64(num_bytes
+
5321 BTRFS_MAX_EXTENT_SIZE
- 1,
5322 BTRFS_MAX_EXTENT_SIZE
);
5323 ASSERT(num_extents
);
5324 ASSERT(BTRFS_I(inode
)->outstanding_extents
>= num_extents
);
5325 BTRFS_I(inode
)->outstanding_extents
-= num_extents
;
5327 if (BTRFS_I(inode
)->outstanding_extents
== 0 &&
5328 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5329 &BTRFS_I(inode
)->runtime_flags
))
5330 drop_inode_space
= 1;
5333 * If we have more or the same amount of outsanding extents than we have
5334 * reserved then we need to leave the reserved extents count alone.
5336 if (BTRFS_I(inode
)->outstanding_extents
>=
5337 BTRFS_I(inode
)->reserved_extents
)
5338 return drop_inode_space
;
5340 dropped_extents
= BTRFS_I(inode
)->reserved_extents
-
5341 BTRFS_I(inode
)->outstanding_extents
;
5342 BTRFS_I(inode
)->reserved_extents
-= dropped_extents
;
5343 return dropped_extents
+ drop_inode_space
;
5347 * calc_csum_metadata_size - return the amount of metada space that must be
5348 * reserved/free'd for the given bytes.
5349 * @inode: the inode we're manipulating
5350 * @num_bytes: the number of bytes in question
5351 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5353 * This adjusts the number of csum_bytes in the inode and then returns the
5354 * correct amount of metadata that must either be reserved or freed. We
5355 * calculate how many checksums we can fit into one leaf and then divide the
5356 * number of bytes that will need to be checksumed by this value to figure out
5357 * how many checksums will be required. If we are adding bytes then the number
5358 * may go up and we will return the number of additional bytes that must be
5359 * reserved. If it is going down we will return the number of bytes that must
5362 * This must be called with BTRFS_I(inode)->lock held.
5364 static u64
calc_csum_metadata_size(struct inode
*inode
, u64 num_bytes
,
5367 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5368 u64 old_csums
, num_csums
;
5370 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
&&
5371 BTRFS_I(inode
)->csum_bytes
== 0)
5374 old_csums
= btrfs_csum_bytes_to_leaves(root
, BTRFS_I(inode
)->csum_bytes
);
5376 BTRFS_I(inode
)->csum_bytes
+= num_bytes
;
5378 BTRFS_I(inode
)->csum_bytes
-= num_bytes
;
5379 num_csums
= btrfs_csum_bytes_to_leaves(root
, BTRFS_I(inode
)->csum_bytes
);
5381 /* No change, no need to reserve more */
5382 if (old_csums
== num_csums
)
5386 return btrfs_calc_trans_metadata_size(root
,
5387 num_csums
- old_csums
);
5389 return btrfs_calc_trans_metadata_size(root
, old_csums
- num_csums
);
5392 int btrfs_delalloc_reserve_metadata(struct inode
*inode
, u64 num_bytes
)
5394 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5395 struct btrfs_block_rsv
*block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
5398 unsigned nr_extents
= 0;
5399 int extra_reserve
= 0;
5400 enum btrfs_reserve_flush_enum flush
= BTRFS_RESERVE_FLUSH_ALL
;
5402 bool delalloc_lock
= true;
5406 /* If we are a free space inode we need to not flush since we will be in
5407 * the middle of a transaction commit. We also don't need the delalloc
5408 * mutex since we won't race with anybody. We need this mostly to make
5409 * lockdep shut its filthy mouth.
5411 if (btrfs_is_free_space_inode(inode
)) {
5412 flush
= BTRFS_RESERVE_NO_FLUSH
;
5413 delalloc_lock
= false;
5416 if (flush
!= BTRFS_RESERVE_NO_FLUSH
&&
5417 btrfs_transaction_in_commit(root
->fs_info
))
5418 schedule_timeout(1);
5421 mutex_lock(&BTRFS_I(inode
)->delalloc_mutex
);
5423 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
5425 spin_lock(&BTRFS_I(inode
)->lock
);
5426 nr_extents
= (unsigned)div64_u64(num_bytes
+
5427 BTRFS_MAX_EXTENT_SIZE
- 1,
5428 BTRFS_MAX_EXTENT_SIZE
);
5429 BTRFS_I(inode
)->outstanding_extents
+= nr_extents
;
5432 if (BTRFS_I(inode
)->outstanding_extents
>
5433 BTRFS_I(inode
)->reserved_extents
)
5434 nr_extents
= BTRFS_I(inode
)->outstanding_extents
-
5435 BTRFS_I(inode
)->reserved_extents
;
5438 * Add an item to reserve for updating the inode when we complete the
5441 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5442 &BTRFS_I(inode
)->runtime_flags
)) {
5447 to_reserve
= btrfs_calc_trans_metadata_size(root
, nr_extents
);
5448 to_reserve
+= calc_csum_metadata_size(inode
, num_bytes
, 1);
5449 csum_bytes
= BTRFS_I(inode
)->csum_bytes
;
5450 spin_unlock(&BTRFS_I(inode
)->lock
);
5452 if (root
->fs_info
->quota_enabled
) {
5453 ret
= btrfs_qgroup_reserve(root
, nr_extents
* root
->nodesize
);
5458 ret
= reserve_metadata_bytes(root
, block_rsv
, to_reserve
, flush
);
5459 if (unlikely(ret
)) {
5460 if (root
->fs_info
->quota_enabled
)
5461 btrfs_qgroup_free(root
, nr_extents
* root
->nodesize
);
5465 spin_lock(&BTRFS_I(inode
)->lock
);
5466 if (extra_reserve
) {
5467 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED
,
5468 &BTRFS_I(inode
)->runtime_flags
);
5471 BTRFS_I(inode
)->reserved_extents
+= nr_extents
;
5472 spin_unlock(&BTRFS_I(inode
)->lock
);
5475 mutex_unlock(&BTRFS_I(inode
)->delalloc_mutex
);
5478 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5479 btrfs_ino(inode
), to_reserve
, 1);
5480 block_rsv_add_bytes(block_rsv
, to_reserve
, 1);
5485 spin_lock(&BTRFS_I(inode
)->lock
);
5486 dropped
= drop_outstanding_extent(inode
, num_bytes
);
5488 * If the inodes csum_bytes is the same as the original
5489 * csum_bytes then we know we haven't raced with any free()ers
5490 * so we can just reduce our inodes csum bytes and carry on.
5492 if (BTRFS_I(inode
)->csum_bytes
== csum_bytes
) {
5493 calc_csum_metadata_size(inode
, num_bytes
, 0);
5495 u64 orig_csum_bytes
= BTRFS_I(inode
)->csum_bytes
;
5499 * This is tricky, but first we need to figure out how much we
5500 * free'd from any free-ers that occured during this
5501 * reservation, so we reset ->csum_bytes to the csum_bytes
5502 * before we dropped our lock, and then call the free for the
5503 * number of bytes that were freed while we were trying our
5506 bytes
= csum_bytes
- BTRFS_I(inode
)->csum_bytes
;
5507 BTRFS_I(inode
)->csum_bytes
= csum_bytes
;
5508 to_free
= calc_csum_metadata_size(inode
, bytes
, 0);
5512 * Now we need to see how much we would have freed had we not
5513 * been making this reservation and our ->csum_bytes were not
5514 * artificially inflated.
5516 BTRFS_I(inode
)->csum_bytes
= csum_bytes
- num_bytes
;
5517 bytes
= csum_bytes
- orig_csum_bytes
;
5518 bytes
= calc_csum_metadata_size(inode
, bytes
, 0);
5521 * Now reset ->csum_bytes to what it should be. If bytes is
5522 * more than to_free then we would have free'd more space had we
5523 * not had an artificially high ->csum_bytes, so we need to free
5524 * the remainder. If bytes is the same or less then we don't
5525 * need to do anything, the other free-ers did the correct
5528 BTRFS_I(inode
)->csum_bytes
= orig_csum_bytes
- num_bytes
;
5529 if (bytes
> to_free
)
5530 to_free
= bytes
- to_free
;
5534 spin_unlock(&BTRFS_I(inode
)->lock
);
5536 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
5539 btrfs_block_rsv_release(root
, block_rsv
, to_free
);
5540 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5541 btrfs_ino(inode
), to_free
, 0);
5544 mutex_unlock(&BTRFS_I(inode
)->delalloc_mutex
);
5549 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5550 * @inode: the inode to release the reservation for
5551 * @num_bytes: the number of bytes we're releasing
5553 * This will release the metadata reservation for an inode. This can be called
5554 * once we complete IO for a given set of bytes to release their metadata
5557 void btrfs_delalloc_release_metadata(struct inode
*inode
, u64 num_bytes
)
5559 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5563 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
5564 spin_lock(&BTRFS_I(inode
)->lock
);
5565 dropped
= drop_outstanding_extent(inode
, num_bytes
);
5568 to_free
= calc_csum_metadata_size(inode
, num_bytes
, 0);
5569 spin_unlock(&BTRFS_I(inode
)->lock
);
5571 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
5573 if (btrfs_test_is_dummy_root(root
))
5576 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
5577 btrfs_ino(inode
), to_free
, 0);
5579 btrfs_block_rsv_release(root
, &root
->fs_info
->delalloc_block_rsv
,
5584 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5585 * @inode: inode we're writing to
5586 * @num_bytes: the number of bytes we want to allocate
5588 * This will do the following things
5590 * o reserve space in the data space info for num_bytes
5591 * o reserve space in the metadata space info based on number of outstanding
5592 * extents and how much csums will be needed
5593 * o add to the inodes ->delalloc_bytes
5594 * o add it to the fs_info's delalloc inodes list.
5596 * This will return 0 for success and -ENOSPC if there is no space left.
5598 int btrfs_delalloc_reserve_space(struct inode
*inode
, u64 num_bytes
)
5602 ret
= btrfs_check_data_free_space(inode
, num_bytes
, num_bytes
);
5606 ret
= btrfs_delalloc_reserve_metadata(inode
, num_bytes
);
5608 btrfs_free_reserved_data_space(inode
, num_bytes
);
5616 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5617 * @inode: inode we're releasing space for
5618 * @num_bytes: the number of bytes we want to free up
5620 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5621 * called in the case that we don't need the metadata AND data reservations
5622 * anymore. So if there is an error or we insert an inline extent.
5624 * This function will release the metadata space that was not used and will
5625 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5626 * list if there are no delalloc bytes left.
5628 void btrfs_delalloc_release_space(struct inode
*inode
, u64 num_bytes
)
5630 btrfs_delalloc_release_metadata(inode
, num_bytes
);
5631 btrfs_free_reserved_data_space(inode
, num_bytes
);
5634 static int update_block_group(struct btrfs_trans_handle
*trans
,
5635 struct btrfs_root
*root
, u64 bytenr
,
5636 u64 num_bytes
, int alloc
)
5638 struct btrfs_block_group_cache
*cache
= NULL
;
5639 struct btrfs_fs_info
*info
= root
->fs_info
;
5640 u64 total
= num_bytes
;
5645 /* block accounting for super block */
5646 spin_lock(&info
->delalloc_root_lock
);
5647 old_val
= btrfs_super_bytes_used(info
->super_copy
);
5649 old_val
+= num_bytes
;
5651 old_val
-= num_bytes
;
5652 btrfs_set_super_bytes_used(info
->super_copy
, old_val
);
5653 spin_unlock(&info
->delalloc_root_lock
);
5656 cache
= btrfs_lookup_block_group(info
, bytenr
);
5659 if (cache
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
5660 BTRFS_BLOCK_GROUP_RAID1
|
5661 BTRFS_BLOCK_GROUP_RAID10
))
5666 * If this block group has free space cache written out, we
5667 * need to make sure to load it if we are removing space. This
5668 * is because we need the unpinning stage to actually add the
5669 * space back to the block group, otherwise we will leak space.
5671 if (!alloc
&& cache
->cached
== BTRFS_CACHE_NO
)
5672 cache_block_group(cache
, 1);
5674 byte_in_group
= bytenr
- cache
->key
.objectid
;
5675 WARN_ON(byte_in_group
> cache
->key
.offset
);
5677 spin_lock(&cache
->space_info
->lock
);
5678 spin_lock(&cache
->lock
);
5680 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
5681 cache
->disk_cache_state
< BTRFS_DC_CLEAR
)
5682 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
5684 old_val
= btrfs_block_group_used(&cache
->item
);
5685 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
5687 old_val
+= num_bytes
;
5688 btrfs_set_block_group_used(&cache
->item
, old_val
);
5689 cache
->reserved
-= num_bytes
;
5690 cache
->space_info
->bytes_reserved
-= num_bytes
;
5691 cache
->space_info
->bytes_used
+= num_bytes
;
5692 cache
->space_info
->disk_used
+= num_bytes
* factor
;
5693 spin_unlock(&cache
->lock
);
5694 spin_unlock(&cache
->space_info
->lock
);
5696 old_val
-= num_bytes
;
5697 btrfs_set_block_group_used(&cache
->item
, old_val
);
5698 cache
->pinned
+= num_bytes
;
5699 cache
->space_info
->bytes_pinned
+= num_bytes
;
5700 cache
->space_info
->bytes_used
-= num_bytes
;
5701 cache
->space_info
->disk_used
-= num_bytes
* factor
;
5702 spin_unlock(&cache
->lock
);
5703 spin_unlock(&cache
->space_info
->lock
);
5705 set_extent_dirty(info
->pinned_extents
,
5706 bytenr
, bytenr
+ num_bytes
- 1,
5707 GFP_NOFS
| __GFP_NOFAIL
);
5709 * No longer have used bytes in this block group, queue
5713 spin_lock(&info
->unused_bgs_lock
);
5714 if (list_empty(&cache
->bg_list
)) {
5715 btrfs_get_block_group(cache
);
5716 list_add_tail(&cache
->bg_list
,
5719 spin_unlock(&info
->unused_bgs_lock
);
5723 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
5724 if (list_empty(&cache
->dirty_list
)) {
5725 list_add_tail(&cache
->dirty_list
,
5726 &trans
->transaction
->dirty_bgs
);
5727 trans
->transaction
->num_dirty_bgs
++;
5728 btrfs_get_block_group(cache
);
5730 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
5732 btrfs_put_block_group(cache
);
5734 bytenr
+= num_bytes
;
5739 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
5741 struct btrfs_block_group_cache
*cache
;
5744 spin_lock(&root
->fs_info
->block_group_cache_lock
);
5745 bytenr
= root
->fs_info
->first_logical_byte
;
5746 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
5748 if (bytenr
< (u64
)-1)
5751 cache
= btrfs_lookup_first_block_group(root
->fs_info
, search_start
);
5755 bytenr
= cache
->key
.objectid
;
5756 btrfs_put_block_group(cache
);
5761 static int pin_down_extent(struct btrfs_root
*root
,
5762 struct btrfs_block_group_cache
*cache
,
5763 u64 bytenr
, u64 num_bytes
, int reserved
)
5765 spin_lock(&cache
->space_info
->lock
);
5766 spin_lock(&cache
->lock
);
5767 cache
->pinned
+= num_bytes
;
5768 cache
->space_info
->bytes_pinned
+= num_bytes
;
5770 cache
->reserved
-= num_bytes
;
5771 cache
->space_info
->bytes_reserved
-= num_bytes
;
5773 spin_unlock(&cache
->lock
);
5774 spin_unlock(&cache
->space_info
->lock
);
5776 set_extent_dirty(root
->fs_info
->pinned_extents
, bytenr
,
5777 bytenr
+ num_bytes
- 1, GFP_NOFS
| __GFP_NOFAIL
);
5779 trace_btrfs_reserved_extent_free(root
, bytenr
, num_bytes
);
5784 * this function must be called within transaction
5786 int btrfs_pin_extent(struct btrfs_root
*root
,
5787 u64 bytenr
, u64 num_bytes
, int reserved
)
5789 struct btrfs_block_group_cache
*cache
;
5791 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
5792 BUG_ON(!cache
); /* Logic error */
5794 pin_down_extent(root
, cache
, bytenr
, num_bytes
, reserved
);
5796 btrfs_put_block_group(cache
);
5801 * this function must be called within transaction
5803 int btrfs_pin_extent_for_log_replay(struct btrfs_root
*root
,
5804 u64 bytenr
, u64 num_bytes
)
5806 struct btrfs_block_group_cache
*cache
;
5809 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
5814 * pull in the free space cache (if any) so that our pin
5815 * removes the free space from the cache. We have load_only set
5816 * to one because the slow code to read in the free extents does check
5817 * the pinned extents.
5819 cache_block_group(cache
, 1);
5821 pin_down_extent(root
, cache
, bytenr
, num_bytes
, 0);
5823 /* remove us from the free space cache (if we're there at all) */
5824 ret
= btrfs_remove_free_space(cache
, bytenr
, num_bytes
);
5825 btrfs_put_block_group(cache
);
5829 static int __exclude_logged_extent(struct btrfs_root
*root
, u64 start
, u64 num_bytes
)
5832 struct btrfs_block_group_cache
*block_group
;
5833 struct btrfs_caching_control
*caching_ctl
;
5835 block_group
= btrfs_lookup_block_group(root
->fs_info
, start
);
5839 cache_block_group(block_group
, 0);
5840 caching_ctl
= get_caching_control(block_group
);
5844 BUG_ON(!block_group_cache_done(block_group
));
5845 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
5847 mutex_lock(&caching_ctl
->mutex
);
5849 if (start
>= caching_ctl
->progress
) {
5850 ret
= add_excluded_extent(root
, start
, num_bytes
);
5851 } else if (start
+ num_bytes
<= caching_ctl
->progress
) {
5852 ret
= btrfs_remove_free_space(block_group
,
5855 num_bytes
= caching_ctl
->progress
- start
;
5856 ret
= btrfs_remove_free_space(block_group
,
5861 num_bytes
= (start
+ num_bytes
) -
5862 caching_ctl
->progress
;
5863 start
= caching_ctl
->progress
;
5864 ret
= add_excluded_extent(root
, start
, num_bytes
);
5867 mutex_unlock(&caching_ctl
->mutex
);
5868 put_caching_control(caching_ctl
);
5870 btrfs_put_block_group(block_group
);
5874 int btrfs_exclude_logged_extents(struct btrfs_root
*log
,
5875 struct extent_buffer
*eb
)
5877 struct btrfs_file_extent_item
*item
;
5878 struct btrfs_key key
;
5882 if (!btrfs_fs_incompat(log
->fs_info
, MIXED_GROUPS
))
5885 for (i
= 0; i
< btrfs_header_nritems(eb
); i
++) {
5886 btrfs_item_key_to_cpu(eb
, &key
, i
);
5887 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
5889 item
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
5890 found_type
= btrfs_file_extent_type(eb
, item
);
5891 if (found_type
== BTRFS_FILE_EXTENT_INLINE
)
5893 if (btrfs_file_extent_disk_bytenr(eb
, item
) == 0)
5895 key
.objectid
= btrfs_file_extent_disk_bytenr(eb
, item
);
5896 key
.offset
= btrfs_file_extent_disk_num_bytes(eb
, item
);
5897 __exclude_logged_extent(log
, key
.objectid
, key
.offset
);
5904 * btrfs_update_reserved_bytes - update the block_group and space info counters
5905 * @cache: The cache we are manipulating
5906 * @num_bytes: The number of bytes in question
5907 * @reserve: One of the reservation enums
5908 * @delalloc: The blocks are allocated for the delalloc write
5910 * This is called by the allocator when it reserves space, or by somebody who is
5911 * freeing space that was never actually used on disk. For example if you
5912 * reserve some space for a new leaf in transaction A and before transaction A
5913 * commits you free that leaf, you call this with reserve set to 0 in order to
5914 * clear the reservation.
5916 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5917 * ENOSPC accounting. For data we handle the reservation through clearing the
5918 * delalloc bits in the io_tree. We have to do this since we could end up
5919 * allocating less disk space for the amount of data we have reserved in the
5920 * case of compression.
5922 * If this is a reservation and the block group has become read only we cannot
5923 * make the reservation and return -EAGAIN, otherwise this function always
5926 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
5927 u64 num_bytes
, int reserve
, int delalloc
)
5929 struct btrfs_space_info
*space_info
= cache
->space_info
;
5932 spin_lock(&space_info
->lock
);
5933 spin_lock(&cache
->lock
);
5934 if (reserve
!= RESERVE_FREE
) {
5938 cache
->reserved
+= num_bytes
;
5939 space_info
->bytes_reserved
+= num_bytes
;
5940 if (reserve
== RESERVE_ALLOC
) {
5941 trace_btrfs_space_reservation(cache
->fs_info
,
5942 "space_info", space_info
->flags
,
5944 space_info
->bytes_may_use
-= num_bytes
;
5948 cache
->delalloc_bytes
+= num_bytes
;
5952 space_info
->bytes_readonly
+= num_bytes
;
5953 cache
->reserved
-= num_bytes
;
5954 space_info
->bytes_reserved
-= num_bytes
;
5957 cache
->delalloc_bytes
-= num_bytes
;
5959 spin_unlock(&cache
->lock
);
5960 spin_unlock(&space_info
->lock
);
5964 void btrfs_prepare_extent_commit(struct btrfs_trans_handle
*trans
,
5965 struct btrfs_root
*root
)
5967 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5968 struct btrfs_caching_control
*next
;
5969 struct btrfs_caching_control
*caching_ctl
;
5970 struct btrfs_block_group_cache
*cache
;
5972 down_write(&fs_info
->commit_root_sem
);
5974 list_for_each_entry_safe(caching_ctl
, next
,
5975 &fs_info
->caching_block_groups
, list
) {
5976 cache
= caching_ctl
->block_group
;
5977 if (block_group_cache_done(cache
)) {
5978 cache
->last_byte_to_unpin
= (u64
)-1;
5979 list_del_init(&caching_ctl
->list
);
5980 put_caching_control(caching_ctl
);
5982 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
5986 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
5987 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
5989 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
5991 up_write(&fs_info
->commit_root_sem
);
5993 update_global_block_rsv(fs_info
);
5996 static int unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
,
5997 const bool return_free_space
)
5999 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6000 struct btrfs_block_group_cache
*cache
= NULL
;
6001 struct btrfs_space_info
*space_info
;
6002 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
6006 while (start
<= end
) {
6009 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
6011 btrfs_put_block_group(cache
);
6012 cache
= btrfs_lookup_block_group(fs_info
, start
);
6013 BUG_ON(!cache
); /* Logic error */
6016 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
6017 len
= min(len
, end
+ 1 - start
);
6019 if (start
< cache
->last_byte_to_unpin
) {
6020 len
= min(len
, cache
->last_byte_to_unpin
- start
);
6021 if (return_free_space
)
6022 btrfs_add_free_space(cache
, start
, len
);
6026 space_info
= cache
->space_info
;
6028 spin_lock(&space_info
->lock
);
6029 spin_lock(&cache
->lock
);
6030 cache
->pinned
-= len
;
6031 space_info
->bytes_pinned
-= len
;
6032 percpu_counter_add(&space_info
->total_bytes_pinned
, -len
);
6034 space_info
->bytes_readonly
+= len
;
6037 spin_unlock(&cache
->lock
);
6038 if (!readonly
&& global_rsv
->space_info
== space_info
) {
6039 spin_lock(&global_rsv
->lock
);
6040 if (!global_rsv
->full
) {
6041 len
= min(len
, global_rsv
->size
-
6042 global_rsv
->reserved
);
6043 global_rsv
->reserved
+= len
;
6044 space_info
->bytes_may_use
+= len
;
6045 if (global_rsv
->reserved
>= global_rsv
->size
)
6046 global_rsv
->full
= 1;
6048 spin_unlock(&global_rsv
->lock
);
6050 spin_unlock(&space_info
->lock
);
6054 btrfs_put_block_group(cache
);
6058 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
6059 struct btrfs_root
*root
)
6061 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6062 struct extent_io_tree
*unpin
;
6070 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
6071 unpin
= &fs_info
->freed_extents
[1];
6073 unpin
= &fs_info
->freed_extents
[0];
6076 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
6077 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
6078 EXTENT_DIRTY
, NULL
);
6080 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
6084 if (btrfs_test_opt(root
, DISCARD
))
6085 ret
= btrfs_discard_extent(root
, start
,
6086 end
+ 1 - start
, NULL
);
6088 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
6089 unpin_extent_range(root
, start
, end
, true);
6090 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
6097 static void add_pinned_bytes(struct btrfs_fs_info
*fs_info
, u64 num_bytes
,
6098 u64 owner
, u64 root_objectid
)
6100 struct btrfs_space_info
*space_info
;
6103 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
6104 if (root_objectid
== BTRFS_CHUNK_TREE_OBJECTID
)
6105 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
6107 flags
= BTRFS_BLOCK_GROUP_METADATA
;
6109 flags
= BTRFS_BLOCK_GROUP_DATA
;
6112 space_info
= __find_space_info(fs_info
, flags
);
6113 BUG_ON(!space_info
); /* Logic bug */
6114 percpu_counter_add(&space_info
->total_bytes_pinned
, num_bytes
);
6118 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
6119 struct btrfs_root
*root
,
6120 struct btrfs_delayed_ref_node
*node
, u64 parent
,
6121 u64 root_objectid
, u64 owner_objectid
,
6122 u64 owner_offset
, int refs_to_drop
,
6123 struct btrfs_delayed_extent_op
*extent_op
)
6125 struct btrfs_key key
;
6126 struct btrfs_path
*path
;
6127 struct btrfs_fs_info
*info
= root
->fs_info
;
6128 struct btrfs_root
*extent_root
= info
->extent_root
;
6129 struct extent_buffer
*leaf
;
6130 struct btrfs_extent_item
*ei
;
6131 struct btrfs_extent_inline_ref
*iref
;
6134 int extent_slot
= 0;
6135 int found_extent
= 0;
6137 int no_quota
= node
->no_quota
;
6140 u64 bytenr
= node
->bytenr
;
6141 u64 num_bytes
= node
->num_bytes
;
6143 enum btrfs_qgroup_operation_type type
= BTRFS_QGROUP_OPER_SUB_EXCL
;
6144 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
6147 if (!info
->quota_enabled
|| !is_fstree(root_objectid
))
6150 path
= btrfs_alloc_path();
6155 path
->leave_spinning
= 1;
6157 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
6158 BUG_ON(!is_data
&& refs_to_drop
!= 1);
6161 skinny_metadata
= 0;
6163 ret
= lookup_extent_backref(trans
, extent_root
, path
, &iref
,
6164 bytenr
, num_bytes
, parent
,
6165 root_objectid
, owner_objectid
,
6168 extent_slot
= path
->slots
[0];
6169 while (extent_slot
>= 0) {
6170 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
6172 if (key
.objectid
!= bytenr
)
6174 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
6175 key
.offset
== num_bytes
) {
6179 if (key
.type
== BTRFS_METADATA_ITEM_KEY
&&
6180 key
.offset
== owner_objectid
) {
6184 if (path
->slots
[0] - extent_slot
> 5)
6188 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6189 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
6190 if (found_extent
&& item_size
< sizeof(*ei
))
6193 if (!found_extent
) {
6195 ret
= remove_extent_backref(trans
, extent_root
, path
,
6197 is_data
, &last_ref
);
6199 btrfs_abort_transaction(trans
, extent_root
, ret
);
6202 btrfs_release_path(path
);
6203 path
->leave_spinning
= 1;
6205 key
.objectid
= bytenr
;
6206 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6207 key
.offset
= num_bytes
;
6209 if (!is_data
&& skinny_metadata
) {
6210 key
.type
= BTRFS_METADATA_ITEM_KEY
;
6211 key
.offset
= owner_objectid
;
6214 ret
= btrfs_search_slot(trans
, extent_root
,
6216 if (ret
> 0 && skinny_metadata
&& path
->slots
[0]) {
6218 * Couldn't find our skinny metadata item,
6219 * see if we have ye olde extent item.
6222 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
6224 if (key
.objectid
== bytenr
&&
6225 key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
6226 key
.offset
== num_bytes
)
6230 if (ret
> 0 && skinny_metadata
) {
6231 skinny_metadata
= false;
6232 key
.objectid
= bytenr
;
6233 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6234 key
.offset
= num_bytes
;
6235 btrfs_release_path(path
);
6236 ret
= btrfs_search_slot(trans
, extent_root
,
6241 btrfs_err(info
, "umm, got %d back from search, was looking for %llu",
6244 btrfs_print_leaf(extent_root
,
6248 btrfs_abort_transaction(trans
, extent_root
, ret
);
6251 extent_slot
= path
->slots
[0];
6253 } else if (WARN_ON(ret
== -ENOENT
)) {
6254 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
6256 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6257 bytenr
, parent
, root_objectid
, owner_objectid
,
6259 btrfs_abort_transaction(trans
, extent_root
, ret
);
6262 btrfs_abort_transaction(trans
, extent_root
, ret
);
6266 leaf
= path
->nodes
[0];
6267 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
6268 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6269 if (item_size
< sizeof(*ei
)) {
6270 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
6271 ret
= convert_extent_item_v0(trans
, extent_root
, path
,
6274 btrfs_abort_transaction(trans
, extent_root
, ret
);
6278 btrfs_release_path(path
);
6279 path
->leave_spinning
= 1;
6281 key
.objectid
= bytenr
;
6282 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
6283 key
.offset
= num_bytes
;
6285 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
6288 btrfs_err(info
, "umm, got %d back from search, was looking for %llu",
6290 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
6293 btrfs_abort_transaction(trans
, extent_root
, ret
);
6297 extent_slot
= path
->slots
[0];
6298 leaf
= path
->nodes
[0];
6299 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
6302 BUG_ON(item_size
< sizeof(*ei
));
6303 ei
= btrfs_item_ptr(leaf
, extent_slot
,
6304 struct btrfs_extent_item
);
6305 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
6306 key
.type
== BTRFS_EXTENT_ITEM_KEY
) {
6307 struct btrfs_tree_block_info
*bi
;
6308 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
6309 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
6310 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
6313 refs
= btrfs_extent_refs(leaf
, ei
);
6314 if (refs
< refs_to_drop
) {
6315 btrfs_err(info
, "trying to drop %d refs but we only have %Lu "
6316 "for bytenr %Lu", refs_to_drop
, refs
, bytenr
);
6318 btrfs_abort_transaction(trans
, extent_root
, ret
);
6321 refs
-= refs_to_drop
;
6324 type
= BTRFS_QGROUP_OPER_SUB_SHARED
;
6326 __run_delayed_extent_op(extent_op
, leaf
, ei
);
6328 * In the case of inline back ref, reference count will
6329 * be updated by remove_extent_backref
6332 BUG_ON(!found_extent
);
6334 btrfs_set_extent_refs(leaf
, ei
, refs
);
6335 btrfs_mark_buffer_dirty(leaf
);
6338 ret
= remove_extent_backref(trans
, extent_root
, path
,
6340 is_data
, &last_ref
);
6342 btrfs_abort_transaction(trans
, extent_root
, ret
);
6346 add_pinned_bytes(root
->fs_info
, -num_bytes
, owner_objectid
,
6350 BUG_ON(is_data
&& refs_to_drop
!=
6351 extent_data_ref_count(root
, path
, iref
));
6353 BUG_ON(path
->slots
[0] != extent_slot
);
6355 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
6356 path
->slots
[0] = extent_slot
;
6362 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
6365 btrfs_abort_transaction(trans
, extent_root
, ret
);
6368 btrfs_release_path(path
);
6371 ret
= btrfs_del_csums(trans
, root
, bytenr
, num_bytes
);
6373 btrfs_abort_transaction(trans
, extent_root
, ret
);
6378 ret
= update_block_group(trans
, root
, bytenr
, num_bytes
, 0);
6380 btrfs_abort_transaction(trans
, extent_root
, ret
);
6384 btrfs_release_path(path
);
6386 /* Deal with the quota accounting */
6387 if (!ret
&& last_ref
&& !no_quota
) {
6390 if (owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
&&
6391 type
== BTRFS_QGROUP_OPER_SUB_SHARED
)
6394 ret
= btrfs_qgroup_record_ref(trans
, info
, root_objectid
,
6395 bytenr
, num_bytes
, type
,
6399 btrfs_free_path(path
);
6404 * when we free an block, it is possible (and likely) that we free the last
6405 * delayed ref for that extent as well. This searches the delayed ref tree for
6406 * a given extent, and if there are no other delayed refs to be processed, it
6407 * removes it from the tree.
6409 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
6410 struct btrfs_root
*root
, u64 bytenr
)
6412 struct btrfs_delayed_ref_head
*head
;
6413 struct btrfs_delayed_ref_root
*delayed_refs
;
6416 delayed_refs
= &trans
->transaction
->delayed_refs
;
6417 spin_lock(&delayed_refs
->lock
);
6418 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
6420 goto out_delayed_unlock
;
6422 spin_lock(&head
->lock
);
6423 if (!list_empty(&head
->ref_list
))
6426 if (head
->extent_op
) {
6427 if (!head
->must_insert_reserved
)
6429 btrfs_free_delayed_extent_op(head
->extent_op
);
6430 head
->extent_op
= NULL
;
6434 * waiting for the lock here would deadlock. If someone else has it
6435 * locked they are already in the process of dropping it anyway
6437 if (!mutex_trylock(&head
->mutex
))
6441 * at this point we have a head with no other entries. Go
6442 * ahead and process it.
6444 head
->node
.in_tree
= 0;
6445 rb_erase(&head
->href_node
, &delayed_refs
->href_root
);
6447 atomic_dec(&delayed_refs
->num_entries
);
6450 * we don't take a ref on the node because we're removing it from the
6451 * tree, so we just steal the ref the tree was holding.
6453 delayed_refs
->num_heads
--;
6454 if (head
->processing
== 0)
6455 delayed_refs
->num_heads_ready
--;
6456 head
->processing
= 0;
6457 spin_unlock(&head
->lock
);
6458 spin_unlock(&delayed_refs
->lock
);
6460 BUG_ON(head
->extent_op
);
6461 if (head
->must_insert_reserved
)
6464 mutex_unlock(&head
->mutex
);
6465 btrfs_put_delayed_ref(&head
->node
);
6468 spin_unlock(&head
->lock
);
6471 spin_unlock(&delayed_refs
->lock
);
6475 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
6476 struct btrfs_root
*root
,
6477 struct extent_buffer
*buf
,
6478 u64 parent
, int last_ref
)
6483 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
6484 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
6485 buf
->start
, buf
->len
,
6486 parent
, root
->root_key
.objectid
,
6487 btrfs_header_level(buf
),
6488 BTRFS_DROP_DELAYED_REF
, NULL
, 0);
6489 BUG_ON(ret
); /* -ENOMEM */
6495 if (btrfs_header_generation(buf
) == trans
->transid
) {
6496 struct btrfs_block_group_cache
*cache
;
6498 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
6499 ret
= check_ref_cleanup(trans
, root
, buf
->start
);
6504 cache
= btrfs_lookup_block_group(root
->fs_info
, buf
->start
);
6506 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
6507 pin_down_extent(root
, cache
, buf
->start
, buf
->len
, 1);
6508 btrfs_put_block_group(cache
);
6512 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
6514 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
6515 btrfs_update_reserved_bytes(cache
, buf
->len
, RESERVE_FREE
, 0);
6516 btrfs_put_block_group(cache
);
6517 trace_btrfs_reserved_extent_free(root
, buf
->start
, buf
->len
);
6522 add_pinned_bytes(root
->fs_info
, buf
->len
,
6523 btrfs_header_level(buf
),
6524 root
->root_key
.objectid
);
6527 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6530 clear_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
);
6533 /* Can return -ENOMEM */
6534 int btrfs_free_extent(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
6535 u64 bytenr
, u64 num_bytes
, u64 parent
, u64 root_objectid
,
6536 u64 owner
, u64 offset
, int no_quota
)
6539 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6541 if (btrfs_test_is_dummy_root(root
))
6544 add_pinned_bytes(root
->fs_info
, num_bytes
, owner
, root_objectid
);
6547 * tree log blocks never actually go into the extent allocation
6548 * tree, just update pinning info and exit early.
6550 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
6551 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
6552 /* unlocks the pinned mutex */
6553 btrfs_pin_extent(root
, bytenr
, num_bytes
, 1);
6555 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
6556 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
6558 parent
, root_objectid
, (int)owner
,
6559 BTRFS_DROP_DELAYED_REF
, NULL
, no_quota
);
6561 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
6563 parent
, root_objectid
, owner
,
6564 offset
, BTRFS_DROP_DELAYED_REF
,
6571 * when we wait for progress in the block group caching, its because
6572 * our allocation attempt failed at least once. So, we must sleep
6573 * and let some progress happen before we try again.
6575 * This function will sleep at least once waiting for new free space to
6576 * show up, and then it will check the block group free space numbers
6577 * for our min num_bytes. Another option is to have it go ahead
6578 * and look in the rbtree for a free extent of a given size, but this
6581 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6582 * any of the information in this block group.
6584 static noinline
void
6585 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
6588 struct btrfs_caching_control
*caching_ctl
;
6590 caching_ctl
= get_caching_control(cache
);
6594 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
6595 (cache
->free_space_ctl
->free_space
>= num_bytes
));
6597 put_caching_control(caching_ctl
);
6601 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
6603 struct btrfs_caching_control
*caching_ctl
;
6606 caching_ctl
= get_caching_control(cache
);
6608 return (cache
->cached
== BTRFS_CACHE_ERROR
) ? -EIO
: 0;
6610 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
6611 if (cache
->cached
== BTRFS_CACHE_ERROR
)
6613 put_caching_control(caching_ctl
);
6617 int __get_raid_index(u64 flags
)
6619 if (flags
& BTRFS_BLOCK_GROUP_RAID10
)
6620 return BTRFS_RAID_RAID10
;
6621 else if (flags
& BTRFS_BLOCK_GROUP_RAID1
)
6622 return BTRFS_RAID_RAID1
;
6623 else if (flags
& BTRFS_BLOCK_GROUP_DUP
)
6624 return BTRFS_RAID_DUP
;
6625 else if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
6626 return BTRFS_RAID_RAID0
;
6627 else if (flags
& BTRFS_BLOCK_GROUP_RAID5
)
6628 return BTRFS_RAID_RAID5
;
6629 else if (flags
& BTRFS_BLOCK_GROUP_RAID6
)
6630 return BTRFS_RAID_RAID6
;
6632 return BTRFS_RAID_SINGLE
; /* BTRFS_BLOCK_GROUP_SINGLE */
6635 int get_block_group_index(struct btrfs_block_group_cache
*cache
)
6637 return __get_raid_index(cache
->flags
);
6640 static const char *btrfs_raid_type_names
[BTRFS_NR_RAID_TYPES
] = {
6641 [BTRFS_RAID_RAID10
] = "raid10",
6642 [BTRFS_RAID_RAID1
] = "raid1",
6643 [BTRFS_RAID_DUP
] = "dup",
6644 [BTRFS_RAID_RAID0
] = "raid0",
6645 [BTRFS_RAID_SINGLE
] = "single",
6646 [BTRFS_RAID_RAID5
] = "raid5",
6647 [BTRFS_RAID_RAID6
] = "raid6",
6650 static const char *get_raid_name(enum btrfs_raid_types type
)
6652 if (type
>= BTRFS_NR_RAID_TYPES
)
6655 return btrfs_raid_type_names
[type
];
6658 enum btrfs_loop_type
{
6659 LOOP_CACHING_NOWAIT
= 0,
6660 LOOP_CACHING_WAIT
= 1,
6661 LOOP_ALLOC_CHUNK
= 2,
6662 LOOP_NO_EMPTY_SIZE
= 3,
6666 btrfs_lock_block_group(struct btrfs_block_group_cache
*cache
,
6670 down_read(&cache
->data_rwsem
);
6674 btrfs_grab_block_group(struct btrfs_block_group_cache
*cache
,
6677 btrfs_get_block_group(cache
);
6679 down_read(&cache
->data_rwsem
);
6682 static struct btrfs_block_group_cache
*
6683 btrfs_lock_cluster(struct btrfs_block_group_cache
*block_group
,
6684 struct btrfs_free_cluster
*cluster
,
6687 struct btrfs_block_group_cache
*used_bg
;
6688 bool locked
= false;
6690 spin_lock(&cluster
->refill_lock
);
6692 if (used_bg
== cluster
->block_group
)
6695 up_read(&used_bg
->data_rwsem
);
6696 btrfs_put_block_group(used_bg
);
6699 used_bg
= cluster
->block_group
;
6703 if (used_bg
== block_group
)
6706 btrfs_get_block_group(used_bg
);
6711 if (down_read_trylock(&used_bg
->data_rwsem
))
6714 spin_unlock(&cluster
->refill_lock
);
6715 down_read(&used_bg
->data_rwsem
);
6721 btrfs_release_block_group(struct btrfs_block_group_cache
*cache
,
6725 up_read(&cache
->data_rwsem
);
6726 btrfs_put_block_group(cache
);
6730 * walks the btree of allocated extents and find a hole of a given size.
6731 * The key ins is changed to record the hole:
6732 * ins->objectid == start position
6733 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6734 * ins->offset == the size of the hole.
6735 * Any available blocks before search_start are skipped.
6737 * If there is no suitable free space, we will record the max size of
6738 * the free space extent currently.
6740 static noinline
int find_free_extent(struct btrfs_root
*orig_root
,
6741 u64 num_bytes
, u64 empty_size
,
6742 u64 hint_byte
, struct btrfs_key
*ins
,
6743 u64 flags
, int delalloc
)
6746 struct btrfs_root
*root
= orig_root
->fs_info
->extent_root
;
6747 struct btrfs_free_cluster
*last_ptr
= NULL
;
6748 struct btrfs_block_group_cache
*block_group
= NULL
;
6749 u64 search_start
= 0;
6750 u64 max_extent_size
= 0;
6751 int empty_cluster
= 2 * 1024 * 1024;
6752 struct btrfs_space_info
*space_info
;
6754 int index
= __get_raid_index(flags
);
6755 int alloc_type
= (flags
& BTRFS_BLOCK_GROUP_DATA
) ?
6756 RESERVE_ALLOC_NO_ACCOUNT
: RESERVE_ALLOC
;
6757 bool failed_cluster_refill
= false;
6758 bool failed_alloc
= false;
6759 bool use_cluster
= true;
6760 bool have_caching_bg
= false;
6762 WARN_ON(num_bytes
< root
->sectorsize
);
6763 ins
->type
= BTRFS_EXTENT_ITEM_KEY
;
6767 trace_find_free_extent(orig_root
, num_bytes
, empty_size
, flags
);
6769 space_info
= __find_space_info(root
->fs_info
, flags
);
6771 btrfs_err(root
->fs_info
, "No space info for %llu", flags
);
6776 * If the space info is for both data and metadata it means we have a
6777 * small filesystem and we can't use the clustering stuff.
6779 if (btrfs_mixed_space_info(space_info
))
6780 use_cluster
= false;
6782 if (flags
& BTRFS_BLOCK_GROUP_METADATA
&& use_cluster
) {
6783 last_ptr
= &root
->fs_info
->meta_alloc_cluster
;
6784 if (!btrfs_test_opt(root
, SSD
))
6785 empty_cluster
= 64 * 1024;
6788 if ((flags
& BTRFS_BLOCK_GROUP_DATA
) && use_cluster
&&
6789 btrfs_test_opt(root
, SSD
)) {
6790 last_ptr
= &root
->fs_info
->data_alloc_cluster
;
6794 spin_lock(&last_ptr
->lock
);
6795 if (last_ptr
->block_group
)
6796 hint_byte
= last_ptr
->window_start
;
6797 spin_unlock(&last_ptr
->lock
);
6800 search_start
= max(search_start
, first_logical_byte(root
, 0));
6801 search_start
= max(search_start
, hint_byte
);
6806 if (search_start
== hint_byte
) {
6807 block_group
= btrfs_lookup_block_group(root
->fs_info
,
6810 * we don't want to use the block group if it doesn't match our
6811 * allocation bits, or if its not cached.
6813 * However if we are re-searching with an ideal block group
6814 * picked out then we don't care that the block group is cached.
6816 if (block_group
&& block_group_bits(block_group
, flags
) &&
6817 block_group
->cached
!= BTRFS_CACHE_NO
) {
6818 down_read(&space_info
->groups_sem
);
6819 if (list_empty(&block_group
->list
) ||
6822 * someone is removing this block group,
6823 * we can't jump into the have_block_group
6824 * target because our list pointers are not
6827 btrfs_put_block_group(block_group
);
6828 up_read(&space_info
->groups_sem
);
6830 index
= get_block_group_index(block_group
);
6831 btrfs_lock_block_group(block_group
, delalloc
);
6832 goto have_block_group
;
6834 } else if (block_group
) {
6835 btrfs_put_block_group(block_group
);
6839 have_caching_bg
= false;
6840 down_read(&space_info
->groups_sem
);
6841 list_for_each_entry(block_group
, &space_info
->block_groups
[index
],
6846 btrfs_grab_block_group(block_group
, delalloc
);
6847 search_start
= block_group
->key
.objectid
;
6850 * this can happen if we end up cycling through all the
6851 * raid types, but we want to make sure we only allocate
6852 * for the proper type.
6854 if (!block_group_bits(block_group
, flags
)) {
6855 u64 extra
= BTRFS_BLOCK_GROUP_DUP
|
6856 BTRFS_BLOCK_GROUP_RAID1
|
6857 BTRFS_BLOCK_GROUP_RAID5
|
6858 BTRFS_BLOCK_GROUP_RAID6
|
6859 BTRFS_BLOCK_GROUP_RAID10
;
6862 * if they asked for extra copies and this block group
6863 * doesn't provide them, bail. This does allow us to
6864 * fill raid0 from raid1.
6866 if ((flags
& extra
) && !(block_group
->flags
& extra
))
6871 cached
= block_group_cache_done(block_group
);
6872 if (unlikely(!cached
)) {
6873 ret
= cache_block_group(block_group
, 0);
6878 if (unlikely(block_group
->cached
== BTRFS_CACHE_ERROR
))
6880 if (unlikely(block_group
->ro
))
6884 * Ok we want to try and use the cluster allocator, so
6888 struct btrfs_block_group_cache
*used_block_group
;
6889 unsigned long aligned_cluster
;
6891 * the refill lock keeps out other
6892 * people trying to start a new cluster
6894 used_block_group
= btrfs_lock_cluster(block_group
,
6897 if (!used_block_group
)
6898 goto refill_cluster
;
6900 if (used_block_group
!= block_group
&&
6901 (used_block_group
->ro
||
6902 !block_group_bits(used_block_group
, flags
)))
6903 goto release_cluster
;
6905 offset
= btrfs_alloc_from_cluster(used_block_group
,
6908 used_block_group
->key
.objectid
,
6911 /* we have a block, we're done */
6912 spin_unlock(&last_ptr
->refill_lock
);
6913 trace_btrfs_reserve_extent_cluster(root
,
6915 search_start
, num_bytes
);
6916 if (used_block_group
!= block_group
) {
6917 btrfs_release_block_group(block_group
,
6919 block_group
= used_block_group
;
6924 WARN_ON(last_ptr
->block_group
!= used_block_group
);
6926 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6927 * set up a new clusters, so lets just skip it
6928 * and let the allocator find whatever block
6929 * it can find. If we reach this point, we
6930 * will have tried the cluster allocator
6931 * plenty of times and not have found
6932 * anything, so we are likely way too
6933 * fragmented for the clustering stuff to find
6936 * However, if the cluster is taken from the
6937 * current block group, release the cluster
6938 * first, so that we stand a better chance of
6939 * succeeding in the unclustered
6941 if (loop
>= LOOP_NO_EMPTY_SIZE
&&
6942 used_block_group
!= block_group
) {
6943 spin_unlock(&last_ptr
->refill_lock
);
6944 btrfs_release_block_group(used_block_group
,
6946 goto unclustered_alloc
;
6950 * this cluster didn't work out, free it and
6953 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
6955 if (used_block_group
!= block_group
)
6956 btrfs_release_block_group(used_block_group
,
6959 if (loop
>= LOOP_NO_EMPTY_SIZE
) {
6960 spin_unlock(&last_ptr
->refill_lock
);
6961 goto unclustered_alloc
;
6964 aligned_cluster
= max_t(unsigned long,
6965 empty_cluster
+ empty_size
,
6966 block_group
->full_stripe_len
);
6968 /* allocate a cluster in this block group */
6969 ret
= btrfs_find_space_cluster(root
, block_group
,
6970 last_ptr
, search_start
,
6975 * now pull our allocation out of this
6978 offset
= btrfs_alloc_from_cluster(block_group
,
6984 /* we found one, proceed */
6985 spin_unlock(&last_ptr
->refill_lock
);
6986 trace_btrfs_reserve_extent_cluster(root
,
6987 block_group
, search_start
,
6991 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
6992 && !failed_cluster_refill
) {
6993 spin_unlock(&last_ptr
->refill_lock
);
6995 failed_cluster_refill
= true;
6996 wait_block_group_cache_progress(block_group
,
6997 num_bytes
+ empty_cluster
+ empty_size
);
6998 goto have_block_group
;
7002 * at this point we either didn't find a cluster
7003 * or we weren't able to allocate a block from our
7004 * cluster. Free the cluster we've been trying
7005 * to use, and go to the next block group
7007 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
7008 spin_unlock(&last_ptr
->refill_lock
);
7013 spin_lock(&block_group
->free_space_ctl
->tree_lock
);
7015 block_group
->free_space_ctl
->free_space
<
7016 num_bytes
+ empty_cluster
+ empty_size
) {
7017 if (block_group
->free_space_ctl
->free_space
>
7020 block_group
->free_space_ctl
->free_space
;
7021 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
7024 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
7026 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
7027 num_bytes
, empty_size
,
7030 * If we didn't find a chunk, and we haven't failed on this
7031 * block group before, and this block group is in the middle of
7032 * caching and we are ok with waiting, then go ahead and wait
7033 * for progress to be made, and set failed_alloc to true.
7035 * If failed_alloc is true then we've already waited on this
7036 * block group once and should move on to the next block group.
7038 if (!offset
&& !failed_alloc
&& !cached
&&
7039 loop
> LOOP_CACHING_NOWAIT
) {
7040 wait_block_group_cache_progress(block_group
,
7041 num_bytes
+ empty_size
);
7042 failed_alloc
= true;
7043 goto have_block_group
;
7044 } else if (!offset
) {
7046 have_caching_bg
= true;
7050 search_start
= ALIGN(offset
, root
->stripesize
);
7052 /* move on to the next group */
7053 if (search_start
+ num_bytes
>
7054 block_group
->key
.objectid
+ block_group
->key
.offset
) {
7055 btrfs_add_free_space(block_group
, offset
, num_bytes
);
7059 if (offset
< search_start
)
7060 btrfs_add_free_space(block_group
, offset
,
7061 search_start
- offset
);
7062 BUG_ON(offset
> search_start
);
7064 ret
= btrfs_update_reserved_bytes(block_group
, num_bytes
,
7065 alloc_type
, delalloc
);
7066 if (ret
== -EAGAIN
) {
7067 btrfs_add_free_space(block_group
, offset
, num_bytes
);
7071 /* we are all good, lets return */
7072 ins
->objectid
= search_start
;
7073 ins
->offset
= num_bytes
;
7075 trace_btrfs_reserve_extent(orig_root
, block_group
,
7076 search_start
, num_bytes
);
7077 btrfs_release_block_group(block_group
, delalloc
);
7080 failed_cluster_refill
= false;
7081 failed_alloc
= false;
7082 BUG_ON(index
!= get_block_group_index(block_group
));
7083 btrfs_release_block_group(block_group
, delalloc
);
7085 up_read(&space_info
->groups_sem
);
7087 if (!ins
->objectid
&& loop
>= LOOP_CACHING_WAIT
&& have_caching_bg
)
7090 if (!ins
->objectid
&& ++index
< BTRFS_NR_RAID_TYPES
)
7094 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7095 * caching kthreads as we move along
7096 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7097 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7098 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7101 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
) {
7104 if (loop
== LOOP_ALLOC_CHUNK
) {
7105 struct btrfs_trans_handle
*trans
;
7108 trans
= current
->journal_info
;
7112 trans
= btrfs_join_transaction(root
);
7114 if (IS_ERR(trans
)) {
7115 ret
= PTR_ERR(trans
);
7119 ret
= do_chunk_alloc(trans
, root
, flags
,
7122 * Do not bail out on ENOSPC since we
7123 * can do more things.
7125 if (ret
< 0 && ret
!= -ENOSPC
)
7126 btrfs_abort_transaction(trans
,
7131 btrfs_end_transaction(trans
, root
);
7136 if (loop
== LOOP_NO_EMPTY_SIZE
) {
7142 } else if (!ins
->objectid
) {
7144 } else if (ins
->objectid
) {
7149 ins
->offset
= max_extent_size
;
7153 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
7154 int dump_block_groups
)
7156 struct btrfs_block_group_cache
*cache
;
7159 spin_lock(&info
->lock
);
7160 printk(KERN_INFO
"BTRFS: space_info %llu has %llu free, is %sfull\n",
7162 info
->total_bytes
- info
->bytes_used
- info
->bytes_pinned
-
7163 info
->bytes_reserved
- info
->bytes_readonly
,
7164 (info
->full
) ? "" : "not ");
7165 printk(KERN_INFO
"BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7166 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7167 info
->total_bytes
, info
->bytes_used
, info
->bytes_pinned
,
7168 info
->bytes_reserved
, info
->bytes_may_use
,
7169 info
->bytes_readonly
);
7170 spin_unlock(&info
->lock
);
7172 if (!dump_block_groups
)
7175 down_read(&info
->groups_sem
);
7177 list_for_each_entry(cache
, &info
->block_groups
[index
], list
) {
7178 spin_lock(&cache
->lock
);
7179 printk(KERN_INFO
"BTRFS: "
7180 "block group %llu has %llu bytes, "
7181 "%llu used %llu pinned %llu reserved %s\n",
7182 cache
->key
.objectid
, cache
->key
.offset
,
7183 btrfs_block_group_used(&cache
->item
), cache
->pinned
,
7184 cache
->reserved
, cache
->ro
? "[readonly]" : "");
7185 btrfs_dump_free_space(cache
, bytes
);
7186 spin_unlock(&cache
->lock
);
7188 if (++index
< BTRFS_NR_RAID_TYPES
)
7190 up_read(&info
->groups_sem
);
7193 int btrfs_reserve_extent(struct btrfs_root
*root
,
7194 u64 num_bytes
, u64 min_alloc_size
,
7195 u64 empty_size
, u64 hint_byte
,
7196 struct btrfs_key
*ins
, int is_data
, int delalloc
)
7198 bool final_tried
= false;
7202 flags
= btrfs_get_alloc_profile(root
, is_data
);
7204 WARN_ON(num_bytes
< root
->sectorsize
);
7205 ret
= find_free_extent(root
, num_bytes
, empty_size
, hint_byte
, ins
,
7208 if (ret
== -ENOSPC
) {
7209 if (!final_tried
&& ins
->offset
) {
7210 num_bytes
= min(num_bytes
>> 1, ins
->offset
);
7211 num_bytes
= round_down(num_bytes
, root
->sectorsize
);
7212 num_bytes
= max(num_bytes
, min_alloc_size
);
7213 if (num_bytes
== min_alloc_size
)
7216 } else if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
7217 struct btrfs_space_info
*sinfo
;
7219 sinfo
= __find_space_info(root
->fs_info
, flags
);
7220 btrfs_err(root
->fs_info
, "allocation failed flags %llu, wanted %llu",
7223 dump_space_info(sinfo
, num_bytes
, 1);
7230 static int __btrfs_free_reserved_extent(struct btrfs_root
*root
,
7232 int pin
, int delalloc
)
7234 struct btrfs_block_group_cache
*cache
;
7237 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
7239 btrfs_err(root
->fs_info
, "Unable to find block group for %llu",
7245 pin_down_extent(root
, cache
, start
, len
, 1);
7247 if (btrfs_test_opt(root
, DISCARD
))
7248 ret
= btrfs_discard_extent(root
, start
, len
, NULL
);
7249 btrfs_add_free_space(cache
, start
, len
);
7250 btrfs_update_reserved_bytes(cache
, len
, RESERVE_FREE
, delalloc
);
7253 btrfs_put_block_group(cache
);
7255 trace_btrfs_reserved_extent_free(root
, start
, len
);
7260 int btrfs_free_reserved_extent(struct btrfs_root
*root
,
7261 u64 start
, u64 len
, int delalloc
)
7263 return __btrfs_free_reserved_extent(root
, start
, len
, 0, delalloc
);
7266 int btrfs_free_and_pin_reserved_extent(struct btrfs_root
*root
,
7269 return __btrfs_free_reserved_extent(root
, start
, len
, 1, 0);
7272 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
7273 struct btrfs_root
*root
,
7274 u64 parent
, u64 root_objectid
,
7275 u64 flags
, u64 owner
, u64 offset
,
7276 struct btrfs_key
*ins
, int ref_mod
)
7279 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7280 struct btrfs_extent_item
*extent_item
;
7281 struct btrfs_extent_inline_ref
*iref
;
7282 struct btrfs_path
*path
;
7283 struct extent_buffer
*leaf
;
7288 type
= BTRFS_SHARED_DATA_REF_KEY
;
7290 type
= BTRFS_EXTENT_DATA_REF_KEY
;
7292 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
7294 path
= btrfs_alloc_path();
7298 path
->leave_spinning
= 1;
7299 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
7302 btrfs_free_path(path
);
7306 leaf
= path
->nodes
[0];
7307 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
7308 struct btrfs_extent_item
);
7309 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
7310 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
7311 btrfs_set_extent_flags(leaf
, extent_item
,
7312 flags
| BTRFS_EXTENT_FLAG_DATA
);
7314 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
7315 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
7317 struct btrfs_shared_data_ref
*ref
;
7318 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
7319 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
7320 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
7322 struct btrfs_extent_data_ref
*ref
;
7323 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
7324 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
7325 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
7326 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
7327 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
7330 btrfs_mark_buffer_dirty(path
->nodes
[0]);
7331 btrfs_free_path(path
);
7333 /* Always set parent to 0 here since its exclusive anyway. */
7334 ret
= btrfs_qgroup_record_ref(trans
, fs_info
, root_objectid
,
7335 ins
->objectid
, ins
->offset
,
7336 BTRFS_QGROUP_OPER_ADD_EXCL
, 0);
7340 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1);
7341 if (ret
) { /* -ENOENT, logic error */
7342 btrfs_err(fs_info
, "update block group failed for %llu %llu",
7343 ins
->objectid
, ins
->offset
);
7346 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, ins
->offset
);
7350 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
7351 struct btrfs_root
*root
,
7352 u64 parent
, u64 root_objectid
,
7353 u64 flags
, struct btrfs_disk_key
*key
,
7354 int level
, struct btrfs_key
*ins
,
7358 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7359 struct btrfs_extent_item
*extent_item
;
7360 struct btrfs_tree_block_info
*block_info
;
7361 struct btrfs_extent_inline_ref
*iref
;
7362 struct btrfs_path
*path
;
7363 struct extent_buffer
*leaf
;
7364 u32 size
= sizeof(*extent_item
) + sizeof(*iref
);
7365 u64 num_bytes
= ins
->offset
;
7366 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
7369 if (!skinny_metadata
)
7370 size
+= sizeof(*block_info
);
7372 path
= btrfs_alloc_path();
7374 btrfs_free_and_pin_reserved_extent(root
, ins
->objectid
,
7379 path
->leave_spinning
= 1;
7380 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
7383 btrfs_free_path(path
);
7384 btrfs_free_and_pin_reserved_extent(root
, ins
->objectid
,
7389 leaf
= path
->nodes
[0];
7390 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
7391 struct btrfs_extent_item
);
7392 btrfs_set_extent_refs(leaf
, extent_item
, 1);
7393 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
7394 btrfs_set_extent_flags(leaf
, extent_item
,
7395 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
7397 if (skinny_metadata
) {
7398 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
7399 num_bytes
= root
->nodesize
;
7401 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
7402 btrfs_set_tree_block_key(leaf
, block_info
, key
);
7403 btrfs_set_tree_block_level(leaf
, block_info
, level
);
7404 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
7408 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
7409 btrfs_set_extent_inline_ref_type(leaf
, iref
,
7410 BTRFS_SHARED_BLOCK_REF_KEY
);
7411 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
7413 btrfs_set_extent_inline_ref_type(leaf
, iref
,
7414 BTRFS_TREE_BLOCK_REF_KEY
);
7415 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
7418 btrfs_mark_buffer_dirty(leaf
);
7419 btrfs_free_path(path
);
7422 ret
= btrfs_qgroup_record_ref(trans
, fs_info
, root_objectid
,
7423 ins
->objectid
, num_bytes
,
7424 BTRFS_QGROUP_OPER_ADD_EXCL
, 0);
7429 ret
= update_block_group(trans
, root
, ins
->objectid
, root
->nodesize
,
7431 if (ret
) { /* -ENOENT, logic error */
7432 btrfs_err(fs_info
, "update block group failed for %llu %llu",
7433 ins
->objectid
, ins
->offset
);
7437 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, root
->nodesize
);
7441 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
7442 struct btrfs_root
*root
,
7443 u64 root_objectid
, u64 owner
,
7444 u64 offset
, struct btrfs_key
*ins
)
7448 BUG_ON(root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
7450 ret
= btrfs_add_delayed_data_ref(root
->fs_info
, trans
, ins
->objectid
,
7452 root_objectid
, owner
, offset
,
7453 BTRFS_ADD_DELAYED_EXTENT
, NULL
, 0);
7458 * this is used by the tree logging recovery code. It records that
7459 * an extent has been allocated and makes sure to clear the free
7460 * space cache bits as well
7462 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
7463 struct btrfs_root
*root
,
7464 u64 root_objectid
, u64 owner
, u64 offset
,
7465 struct btrfs_key
*ins
)
7468 struct btrfs_block_group_cache
*block_group
;
7471 * Mixed block groups will exclude before processing the log so we only
7472 * need to do the exlude dance if this fs isn't mixed.
7474 if (!btrfs_fs_incompat(root
->fs_info
, MIXED_GROUPS
)) {
7475 ret
= __exclude_logged_extent(root
, ins
->objectid
, ins
->offset
);
7480 block_group
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
7484 ret
= btrfs_update_reserved_bytes(block_group
, ins
->offset
,
7485 RESERVE_ALLOC_NO_ACCOUNT
, 0);
7486 BUG_ON(ret
); /* logic error */
7487 ret
= alloc_reserved_file_extent(trans
, root
, 0, root_objectid
,
7488 0, owner
, offset
, ins
, 1);
7489 btrfs_put_block_group(block_group
);
7493 static struct extent_buffer
*
7494 btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
7495 u64 bytenr
, int level
)
7497 struct extent_buffer
*buf
;
7499 buf
= btrfs_find_create_tree_block(root
, bytenr
);
7501 return ERR_PTR(-ENOMEM
);
7502 btrfs_set_header_generation(buf
, trans
->transid
);
7503 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, buf
, level
);
7504 btrfs_tree_lock(buf
);
7505 clean_tree_block(trans
, root
->fs_info
, buf
);
7506 clear_bit(EXTENT_BUFFER_STALE
, &buf
->bflags
);
7508 btrfs_set_lock_blocking(buf
);
7509 btrfs_set_buffer_uptodate(buf
);
7511 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
7512 buf
->log_index
= root
->log_transid
% 2;
7514 * we allow two log transactions at a time, use different
7515 * EXENT bit to differentiate dirty pages.
7517 if (buf
->log_index
== 0)
7518 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
7519 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
7521 set_extent_new(&root
->dirty_log_pages
, buf
->start
,
7522 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
7524 buf
->log_index
= -1;
7525 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
7526 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
7528 trans
->blocks_used
++;
7529 /* this returns a buffer locked for blocking */
7533 static struct btrfs_block_rsv
*
7534 use_block_rsv(struct btrfs_trans_handle
*trans
,
7535 struct btrfs_root
*root
, u32 blocksize
)
7537 struct btrfs_block_rsv
*block_rsv
;
7538 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
7540 bool global_updated
= false;
7542 block_rsv
= get_block_rsv(trans
, root
);
7544 if (unlikely(block_rsv
->size
== 0))
7547 ret
= block_rsv_use_bytes(block_rsv
, blocksize
);
7551 if (block_rsv
->failfast
)
7552 return ERR_PTR(ret
);
7554 if (block_rsv
->type
== BTRFS_BLOCK_RSV_GLOBAL
&& !global_updated
) {
7555 global_updated
= true;
7556 update_global_block_rsv(root
->fs_info
);
7560 if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
7561 static DEFINE_RATELIMIT_STATE(_rs
,
7562 DEFAULT_RATELIMIT_INTERVAL
* 10,
7563 /*DEFAULT_RATELIMIT_BURST*/ 1);
7564 if (__ratelimit(&_rs
))
7566 "BTRFS: block rsv returned %d\n", ret
);
7569 ret
= reserve_metadata_bytes(root
, block_rsv
, blocksize
,
7570 BTRFS_RESERVE_NO_FLUSH
);
7574 * If we couldn't reserve metadata bytes try and use some from
7575 * the global reserve if its space type is the same as the global
7578 if (block_rsv
->type
!= BTRFS_BLOCK_RSV_GLOBAL
&&
7579 block_rsv
->space_info
== global_rsv
->space_info
) {
7580 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
7584 return ERR_PTR(ret
);
7587 static void unuse_block_rsv(struct btrfs_fs_info
*fs_info
,
7588 struct btrfs_block_rsv
*block_rsv
, u32 blocksize
)
7590 block_rsv_add_bytes(block_rsv
, blocksize
, 0);
7591 block_rsv_release_bytes(fs_info
, block_rsv
, NULL
, 0);
7595 * finds a free extent and does all the dirty work required for allocation
7596 * returns the key for the extent through ins, and a tree buffer for
7597 * the first block of the extent through buf.
7599 * returns the tree buffer or an ERR_PTR on error.
7601 struct extent_buffer
*btrfs_alloc_tree_block(struct btrfs_trans_handle
*trans
,
7602 struct btrfs_root
*root
,
7603 u64 parent
, u64 root_objectid
,
7604 struct btrfs_disk_key
*key
, int level
,
7605 u64 hint
, u64 empty_size
)
7607 struct btrfs_key ins
;
7608 struct btrfs_block_rsv
*block_rsv
;
7609 struct extent_buffer
*buf
;
7610 struct btrfs_delayed_extent_op
*extent_op
;
7613 u32 blocksize
= root
->nodesize
;
7614 bool skinny_metadata
= btrfs_fs_incompat(root
->fs_info
,
7617 if (btrfs_test_is_dummy_root(root
)) {
7618 buf
= btrfs_init_new_buffer(trans
, root
, root
->alloc_bytenr
,
7621 root
->alloc_bytenr
+= blocksize
;
7625 block_rsv
= use_block_rsv(trans
, root
, blocksize
);
7626 if (IS_ERR(block_rsv
))
7627 return ERR_CAST(block_rsv
);
7629 ret
= btrfs_reserve_extent(root
, blocksize
, blocksize
,
7630 empty_size
, hint
, &ins
, 0, 0);
7634 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
, level
);
7637 goto out_free_reserved
;
7640 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
7642 parent
= ins
.objectid
;
7643 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
7647 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
7648 extent_op
= btrfs_alloc_delayed_extent_op();
7654 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
7656 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
7657 extent_op
->flags_to_set
= flags
;
7658 if (skinny_metadata
)
7659 extent_op
->update_key
= 0;
7661 extent_op
->update_key
= 1;
7662 extent_op
->update_flags
= 1;
7663 extent_op
->is_data
= 0;
7664 extent_op
->level
= level
;
7666 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
7667 ins
.objectid
, ins
.offset
,
7668 parent
, root_objectid
, level
,
7669 BTRFS_ADD_DELAYED_EXTENT
,
7672 goto out_free_delayed
;
7677 btrfs_free_delayed_extent_op(extent_op
);
7679 free_extent_buffer(buf
);
7681 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
, 0);
7683 unuse_block_rsv(root
->fs_info
, block_rsv
, blocksize
);
7684 return ERR_PTR(ret
);
7687 struct walk_control
{
7688 u64 refs
[BTRFS_MAX_LEVEL
];
7689 u64 flags
[BTRFS_MAX_LEVEL
];
7690 struct btrfs_key update_progress
;
7701 #define DROP_REFERENCE 1
7702 #define UPDATE_BACKREF 2
7704 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
7705 struct btrfs_root
*root
,
7706 struct walk_control
*wc
,
7707 struct btrfs_path
*path
)
7715 struct btrfs_key key
;
7716 struct extent_buffer
*eb
;
7721 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
7722 wc
->reada_count
= wc
->reada_count
* 2 / 3;
7723 wc
->reada_count
= max(wc
->reada_count
, 2);
7725 wc
->reada_count
= wc
->reada_count
* 3 / 2;
7726 wc
->reada_count
= min_t(int, wc
->reada_count
,
7727 BTRFS_NODEPTRS_PER_BLOCK(root
));
7730 eb
= path
->nodes
[wc
->level
];
7731 nritems
= btrfs_header_nritems(eb
);
7732 blocksize
= root
->nodesize
;
7734 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
7735 if (nread
>= wc
->reada_count
)
7739 bytenr
= btrfs_node_blockptr(eb
, slot
);
7740 generation
= btrfs_node_ptr_generation(eb
, slot
);
7742 if (slot
== path
->slots
[wc
->level
])
7745 if (wc
->stage
== UPDATE_BACKREF
&&
7746 generation
<= root
->root_key
.offset
)
7749 /* We don't lock the tree block, it's OK to be racy here */
7750 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
,
7751 wc
->level
- 1, 1, &refs
,
7753 /* We don't care about errors in readahead. */
7758 if (wc
->stage
== DROP_REFERENCE
) {
7762 if (wc
->level
== 1 &&
7763 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
7765 if (!wc
->update_ref
||
7766 generation
<= root
->root_key
.offset
)
7768 btrfs_node_key_to_cpu(eb
, &key
, slot
);
7769 ret
= btrfs_comp_cpu_keys(&key
,
7770 &wc
->update_progress
);
7774 if (wc
->level
== 1 &&
7775 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
7779 readahead_tree_block(root
, bytenr
);
7782 wc
->reada_slot
= slot
;
7785 static int account_leaf_items(struct btrfs_trans_handle
*trans
,
7786 struct btrfs_root
*root
,
7787 struct extent_buffer
*eb
)
7789 int nr
= btrfs_header_nritems(eb
);
7790 int i
, extent_type
, ret
;
7791 struct btrfs_key key
;
7792 struct btrfs_file_extent_item
*fi
;
7793 u64 bytenr
, num_bytes
;
7795 for (i
= 0; i
< nr
; i
++) {
7796 btrfs_item_key_to_cpu(eb
, &key
, i
);
7798 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
7801 fi
= btrfs_item_ptr(eb
, i
, struct btrfs_file_extent_item
);
7802 /* filter out non qgroup-accountable extents */
7803 extent_type
= btrfs_file_extent_type(eb
, fi
);
7805 if (extent_type
== BTRFS_FILE_EXTENT_INLINE
)
7808 bytenr
= btrfs_file_extent_disk_bytenr(eb
, fi
);
7812 num_bytes
= btrfs_file_extent_disk_num_bytes(eb
, fi
);
7814 ret
= btrfs_qgroup_record_ref(trans
, root
->fs_info
,
7817 BTRFS_QGROUP_OPER_SUB_SUBTREE
, 0);
7825 * Walk up the tree from the bottom, freeing leaves and any interior
7826 * nodes which have had all slots visited. If a node (leaf or
7827 * interior) is freed, the node above it will have it's slot
7828 * incremented. The root node will never be freed.
7830 * At the end of this function, we should have a path which has all
7831 * slots incremented to the next position for a search. If we need to
7832 * read a new node it will be NULL and the node above it will have the
7833 * correct slot selected for a later read.
7835 * If we increment the root nodes slot counter past the number of
7836 * elements, 1 is returned to signal completion of the search.
7838 static int adjust_slots_upwards(struct btrfs_root
*root
,
7839 struct btrfs_path
*path
, int root_level
)
7843 struct extent_buffer
*eb
;
7845 if (root_level
== 0)
7848 while (level
<= root_level
) {
7849 eb
= path
->nodes
[level
];
7850 nr
= btrfs_header_nritems(eb
);
7851 path
->slots
[level
]++;
7852 slot
= path
->slots
[level
];
7853 if (slot
>= nr
|| level
== 0) {
7855 * Don't free the root - we will detect this
7856 * condition after our loop and return a
7857 * positive value for caller to stop walking the tree.
7859 if (level
!= root_level
) {
7860 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
7861 path
->locks
[level
] = 0;
7863 free_extent_buffer(eb
);
7864 path
->nodes
[level
] = NULL
;
7865 path
->slots
[level
] = 0;
7869 * We have a valid slot to walk back down
7870 * from. Stop here so caller can process these
7879 eb
= path
->nodes
[root_level
];
7880 if (path
->slots
[root_level
] >= btrfs_header_nritems(eb
))
7887 * root_eb is the subtree root and is locked before this function is called.
7889 static int account_shared_subtree(struct btrfs_trans_handle
*trans
,
7890 struct btrfs_root
*root
,
7891 struct extent_buffer
*root_eb
,
7897 struct extent_buffer
*eb
= root_eb
;
7898 struct btrfs_path
*path
= NULL
;
7900 BUG_ON(root_level
< 0 || root_level
> BTRFS_MAX_LEVEL
);
7901 BUG_ON(root_eb
== NULL
);
7903 if (!root
->fs_info
->quota_enabled
)
7906 if (!extent_buffer_uptodate(root_eb
)) {
7907 ret
= btrfs_read_buffer(root_eb
, root_gen
);
7912 if (root_level
== 0) {
7913 ret
= account_leaf_items(trans
, root
, root_eb
);
7917 path
= btrfs_alloc_path();
7922 * Walk down the tree. Missing extent blocks are filled in as
7923 * we go. Metadata is accounted every time we read a new
7926 * When we reach a leaf, we account for file extent items in it,
7927 * walk back up the tree (adjusting slot pointers as we go)
7928 * and restart the search process.
7930 extent_buffer_get(root_eb
); /* For path */
7931 path
->nodes
[root_level
] = root_eb
;
7932 path
->slots
[root_level
] = 0;
7933 path
->locks
[root_level
] = 0; /* so release_path doesn't try to unlock */
7936 while (level
>= 0) {
7937 if (path
->nodes
[level
] == NULL
) {
7942 /* We need to get child blockptr/gen from
7943 * parent before we can read it. */
7944 eb
= path
->nodes
[level
+ 1];
7945 parent_slot
= path
->slots
[level
+ 1];
7946 child_bytenr
= btrfs_node_blockptr(eb
, parent_slot
);
7947 child_gen
= btrfs_node_ptr_generation(eb
, parent_slot
);
7949 eb
= read_tree_block(root
, child_bytenr
, child_gen
);
7953 } else if (!extent_buffer_uptodate(eb
)) {
7954 free_extent_buffer(eb
);
7959 path
->nodes
[level
] = eb
;
7960 path
->slots
[level
] = 0;
7962 btrfs_tree_read_lock(eb
);
7963 btrfs_set_lock_blocking_rw(eb
, BTRFS_READ_LOCK
);
7964 path
->locks
[level
] = BTRFS_READ_LOCK_BLOCKING
;
7966 ret
= btrfs_qgroup_record_ref(trans
, root
->fs_info
,
7970 BTRFS_QGROUP_OPER_SUB_SUBTREE
,
7978 ret
= account_leaf_items(trans
, root
, path
->nodes
[level
]);
7982 /* Nonzero return here means we completed our search */
7983 ret
= adjust_slots_upwards(root
, path
, root_level
);
7987 /* Restart search with new slots */
7996 btrfs_free_path(path
);
8002 * helper to process tree block while walking down the tree.
8004 * when wc->stage == UPDATE_BACKREF, this function updates
8005 * back refs for pointers in the block.
8007 * NOTE: return value 1 means we should stop walking down.
8009 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
8010 struct btrfs_root
*root
,
8011 struct btrfs_path
*path
,
8012 struct walk_control
*wc
, int lookup_info
)
8014 int level
= wc
->level
;
8015 struct extent_buffer
*eb
= path
->nodes
[level
];
8016 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
8019 if (wc
->stage
== UPDATE_BACKREF
&&
8020 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
8024 * when reference count of tree block is 1, it won't increase
8025 * again. once full backref flag is set, we never clear it.
8028 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
8029 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
8030 BUG_ON(!path
->locks
[level
]);
8031 ret
= btrfs_lookup_extent_info(trans
, root
,
8032 eb
->start
, level
, 1,
8035 BUG_ON(ret
== -ENOMEM
);
8038 BUG_ON(wc
->refs
[level
] == 0);
8041 if (wc
->stage
== DROP_REFERENCE
) {
8042 if (wc
->refs
[level
] > 1)
8045 if (path
->locks
[level
] && !wc
->keep_locks
) {
8046 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8047 path
->locks
[level
] = 0;
8052 /* wc->stage == UPDATE_BACKREF */
8053 if (!(wc
->flags
[level
] & flag
)) {
8054 BUG_ON(!path
->locks
[level
]);
8055 ret
= btrfs_inc_ref(trans
, root
, eb
, 1);
8056 BUG_ON(ret
); /* -ENOMEM */
8057 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
8058 BUG_ON(ret
); /* -ENOMEM */
8059 ret
= btrfs_set_disk_extent_flags(trans
, root
, eb
->start
,
8061 btrfs_header_level(eb
), 0);
8062 BUG_ON(ret
); /* -ENOMEM */
8063 wc
->flags
[level
] |= flag
;
8067 * the block is shared by multiple trees, so it's not good to
8068 * keep the tree lock
8070 if (path
->locks
[level
] && level
> 0) {
8071 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8072 path
->locks
[level
] = 0;
8078 * helper to process tree block pointer.
8080 * when wc->stage == DROP_REFERENCE, this function checks
8081 * reference count of the block pointed to. if the block
8082 * is shared and we need update back refs for the subtree
8083 * rooted at the block, this function changes wc->stage to
8084 * UPDATE_BACKREF. if the block is shared and there is no
8085 * need to update back, this function drops the reference
8088 * NOTE: return value 1 means we should stop walking down.
8090 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
8091 struct btrfs_root
*root
,
8092 struct btrfs_path
*path
,
8093 struct walk_control
*wc
, int *lookup_info
)
8099 struct btrfs_key key
;
8100 struct extent_buffer
*next
;
8101 int level
= wc
->level
;
8104 bool need_account
= false;
8106 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
8107 path
->slots
[level
]);
8109 * if the lower level block was created before the snapshot
8110 * was created, we know there is no need to update back refs
8113 if (wc
->stage
== UPDATE_BACKREF
&&
8114 generation
<= root
->root_key
.offset
) {
8119 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
8120 blocksize
= root
->nodesize
;
8122 next
= btrfs_find_tree_block(root
->fs_info
, bytenr
);
8124 next
= btrfs_find_create_tree_block(root
, bytenr
);
8127 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, next
,
8131 btrfs_tree_lock(next
);
8132 btrfs_set_lock_blocking(next
);
8134 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, level
- 1, 1,
8135 &wc
->refs
[level
- 1],
8136 &wc
->flags
[level
- 1]);
8138 btrfs_tree_unlock(next
);
8142 if (unlikely(wc
->refs
[level
- 1] == 0)) {
8143 btrfs_err(root
->fs_info
, "Missing references.");
8148 if (wc
->stage
== DROP_REFERENCE
) {
8149 if (wc
->refs
[level
- 1] > 1) {
8150 need_account
= true;
8152 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8155 if (!wc
->update_ref
||
8156 generation
<= root
->root_key
.offset
)
8159 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
8160 path
->slots
[level
]);
8161 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
8165 wc
->stage
= UPDATE_BACKREF
;
8166 wc
->shared_level
= level
- 1;
8170 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
8174 if (!btrfs_buffer_uptodate(next
, generation
, 0)) {
8175 btrfs_tree_unlock(next
);
8176 free_extent_buffer(next
);
8182 if (reada
&& level
== 1)
8183 reada_walk_down(trans
, root
, wc
, path
);
8184 next
= read_tree_block(root
, bytenr
, generation
);
8186 return PTR_ERR(next
);
8187 } else if (!extent_buffer_uptodate(next
)) {
8188 free_extent_buffer(next
);
8191 btrfs_tree_lock(next
);
8192 btrfs_set_lock_blocking(next
);
8196 BUG_ON(level
!= btrfs_header_level(next
));
8197 path
->nodes
[level
] = next
;
8198 path
->slots
[level
] = 0;
8199 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8205 wc
->refs
[level
- 1] = 0;
8206 wc
->flags
[level
- 1] = 0;
8207 if (wc
->stage
== DROP_REFERENCE
) {
8208 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
8209 parent
= path
->nodes
[level
]->start
;
8211 BUG_ON(root
->root_key
.objectid
!=
8212 btrfs_header_owner(path
->nodes
[level
]));
8217 ret
= account_shared_subtree(trans
, root
, next
,
8218 generation
, level
- 1);
8220 printk_ratelimited(KERN_ERR
"BTRFS: %s Error "
8221 "%d accounting shared subtree. Quota "
8222 "is out of sync, rescan required.\n",
8223 root
->fs_info
->sb
->s_id
, ret
);
8226 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
, parent
,
8227 root
->root_key
.objectid
, level
- 1, 0, 0);
8228 BUG_ON(ret
); /* -ENOMEM */
8230 btrfs_tree_unlock(next
);
8231 free_extent_buffer(next
);
8237 * helper to process tree block while walking up the tree.
8239 * when wc->stage == DROP_REFERENCE, this function drops
8240 * reference count on the block.
8242 * when wc->stage == UPDATE_BACKREF, this function changes
8243 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8244 * to UPDATE_BACKREF previously while processing the block.
8246 * NOTE: return value 1 means we should stop walking up.
8248 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
8249 struct btrfs_root
*root
,
8250 struct btrfs_path
*path
,
8251 struct walk_control
*wc
)
8254 int level
= wc
->level
;
8255 struct extent_buffer
*eb
= path
->nodes
[level
];
8258 if (wc
->stage
== UPDATE_BACKREF
) {
8259 BUG_ON(wc
->shared_level
< level
);
8260 if (level
< wc
->shared_level
)
8263 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
8267 wc
->stage
= DROP_REFERENCE
;
8268 wc
->shared_level
= -1;
8269 path
->slots
[level
] = 0;
8272 * check reference count again if the block isn't locked.
8273 * we should start walking down the tree again if reference
8276 if (!path
->locks
[level
]) {
8278 btrfs_tree_lock(eb
);
8279 btrfs_set_lock_blocking(eb
);
8280 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8282 ret
= btrfs_lookup_extent_info(trans
, root
,
8283 eb
->start
, level
, 1,
8287 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8288 path
->locks
[level
] = 0;
8291 BUG_ON(wc
->refs
[level
] == 0);
8292 if (wc
->refs
[level
] == 1) {
8293 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
8294 path
->locks
[level
] = 0;
8300 /* wc->stage == DROP_REFERENCE */
8301 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
8303 if (wc
->refs
[level
] == 1) {
8305 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8306 ret
= btrfs_dec_ref(trans
, root
, eb
, 1);
8308 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
8309 BUG_ON(ret
); /* -ENOMEM */
8310 ret
= account_leaf_items(trans
, root
, eb
);
8312 printk_ratelimited(KERN_ERR
"BTRFS: %s Error "
8313 "%d accounting leaf items. Quota "
8314 "is out of sync, rescan required.\n",
8315 root
->fs_info
->sb
->s_id
, ret
);
8318 /* make block locked assertion in clean_tree_block happy */
8319 if (!path
->locks
[level
] &&
8320 btrfs_header_generation(eb
) == trans
->transid
) {
8321 btrfs_tree_lock(eb
);
8322 btrfs_set_lock_blocking(eb
);
8323 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8325 clean_tree_block(trans
, root
->fs_info
, eb
);
8328 if (eb
== root
->node
) {
8329 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8332 BUG_ON(root
->root_key
.objectid
!=
8333 btrfs_header_owner(eb
));
8335 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
8336 parent
= path
->nodes
[level
+ 1]->start
;
8338 BUG_ON(root
->root_key
.objectid
!=
8339 btrfs_header_owner(path
->nodes
[level
+ 1]));
8342 btrfs_free_tree_block(trans
, root
, eb
, parent
, wc
->refs
[level
] == 1);
8344 wc
->refs
[level
] = 0;
8345 wc
->flags
[level
] = 0;
8349 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
8350 struct btrfs_root
*root
,
8351 struct btrfs_path
*path
,
8352 struct walk_control
*wc
)
8354 int level
= wc
->level
;
8355 int lookup_info
= 1;
8358 while (level
>= 0) {
8359 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
8366 if (path
->slots
[level
] >=
8367 btrfs_header_nritems(path
->nodes
[level
]))
8370 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
8372 path
->slots
[level
]++;
8381 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
8382 struct btrfs_root
*root
,
8383 struct btrfs_path
*path
,
8384 struct walk_control
*wc
, int max_level
)
8386 int level
= wc
->level
;
8389 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
8390 while (level
< max_level
&& path
->nodes
[level
]) {
8392 if (path
->slots
[level
] + 1 <
8393 btrfs_header_nritems(path
->nodes
[level
])) {
8394 path
->slots
[level
]++;
8397 ret
= walk_up_proc(trans
, root
, path
, wc
);
8401 if (path
->locks
[level
]) {
8402 btrfs_tree_unlock_rw(path
->nodes
[level
],
8403 path
->locks
[level
]);
8404 path
->locks
[level
] = 0;
8406 free_extent_buffer(path
->nodes
[level
]);
8407 path
->nodes
[level
] = NULL
;
8415 * drop a subvolume tree.
8417 * this function traverses the tree freeing any blocks that only
8418 * referenced by the tree.
8420 * when a shared tree block is found. this function decreases its
8421 * reference count by one. if update_ref is true, this function
8422 * also make sure backrefs for the shared block and all lower level
8423 * blocks are properly updated.
8425 * If called with for_reloc == 0, may exit early with -EAGAIN
8427 int btrfs_drop_snapshot(struct btrfs_root
*root
,
8428 struct btrfs_block_rsv
*block_rsv
, int update_ref
,
8431 struct btrfs_path
*path
;
8432 struct btrfs_trans_handle
*trans
;
8433 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
8434 struct btrfs_root_item
*root_item
= &root
->root_item
;
8435 struct walk_control
*wc
;
8436 struct btrfs_key key
;
8440 bool root_dropped
= false;
8442 btrfs_debug(root
->fs_info
, "Drop subvolume %llu", root
->objectid
);
8444 path
= btrfs_alloc_path();
8450 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
8452 btrfs_free_path(path
);
8457 trans
= btrfs_start_transaction(tree_root
, 0);
8458 if (IS_ERR(trans
)) {
8459 err
= PTR_ERR(trans
);
8464 trans
->block_rsv
= block_rsv
;
8466 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
8467 level
= btrfs_header_level(root
->node
);
8468 path
->nodes
[level
] = btrfs_lock_root_node(root
);
8469 btrfs_set_lock_blocking(path
->nodes
[level
]);
8470 path
->slots
[level
] = 0;
8471 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8472 memset(&wc
->update_progress
, 0,
8473 sizeof(wc
->update_progress
));
8475 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
8476 memcpy(&wc
->update_progress
, &key
,
8477 sizeof(wc
->update_progress
));
8479 level
= root_item
->drop_level
;
8481 path
->lowest_level
= level
;
8482 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
8483 path
->lowest_level
= 0;
8491 * unlock our path, this is safe because only this
8492 * function is allowed to delete this snapshot
8494 btrfs_unlock_up_safe(path
, 0);
8496 level
= btrfs_header_level(root
->node
);
8498 btrfs_tree_lock(path
->nodes
[level
]);
8499 btrfs_set_lock_blocking(path
->nodes
[level
]);
8500 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8502 ret
= btrfs_lookup_extent_info(trans
, root
,
8503 path
->nodes
[level
]->start
,
8504 level
, 1, &wc
->refs
[level
],
8510 BUG_ON(wc
->refs
[level
] == 0);
8512 if (level
== root_item
->drop_level
)
8515 btrfs_tree_unlock(path
->nodes
[level
]);
8516 path
->locks
[level
] = 0;
8517 WARN_ON(wc
->refs
[level
] != 1);
8523 wc
->shared_level
= -1;
8524 wc
->stage
= DROP_REFERENCE
;
8525 wc
->update_ref
= update_ref
;
8527 wc
->for_reloc
= for_reloc
;
8528 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
8532 ret
= walk_down_tree(trans
, root
, path
, wc
);
8538 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
8545 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
8549 if (wc
->stage
== DROP_REFERENCE
) {
8551 btrfs_node_key(path
->nodes
[level
],
8552 &root_item
->drop_progress
,
8553 path
->slots
[level
]);
8554 root_item
->drop_level
= level
;
8557 BUG_ON(wc
->level
== 0);
8558 if (btrfs_should_end_transaction(trans
, tree_root
) ||
8559 (!for_reloc
&& btrfs_need_cleaner_sleep(root
))) {
8560 ret
= btrfs_update_root(trans
, tree_root
,
8564 btrfs_abort_transaction(trans
, tree_root
, ret
);
8570 * Qgroup update accounting is run from
8571 * delayed ref handling. This usually works
8572 * out because delayed refs are normally the
8573 * only way qgroup updates are added. However,
8574 * we may have added updates during our tree
8575 * walk so run qgroups here to make sure we
8576 * don't lose any updates.
8578 ret
= btrfs_delayed_qgroup_accounting(trans
,
8581 printk_ratelimited(KERN_ERR
"BTRFS: Failure %d "
8582 "running qgroup updates "
8583 "during snapshot delete. "
8584 "Quota is out of sync, "
8585 "rescan required.\n", ret
);
8587 btrfs_end_transaction_throttle(trans
, tree_root
);
8588 if (!for_reloc
&& btrfs_need_cleaner_sleep(root
)) {
8589 pr_debug("BTRFS: drop snapshot early exit\n");
8594 trans
= btrfs_start_transaction(tree_root
, 0);
8595 if (IS_ERR(trans
)) {
8596 err
= PTR_ERR(trans
);
8600 trans
->block_rsv
= block_rsv
;
8603 btrfs_release_path(path
);
8607 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
8609 btrfs_abort_transaction(trans
, tree_root
, ret
);
8613 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
8614 ret
= btrfs_find_root(tree_root
, &root
->root_key
, path
,
8617 btrfs_abort_transaction(trans
, tree_root
, ret
);
8620 } else if (ret
> 0) {
8621 /* if we fail to delete the orphan item this time
8622 * around, it'll get picked up the next time.
8624 * The most common failure here is just -ENOENT.
8626 btrfs_del_orphan_item(trans
, tree_root
,
8627 root
->root_key
.objectid
);
8631 if (test_bit(BTRFS_ROOT_IN_RADIX
, &root
->state
)) {
8632 btrfs_drop_and_free_fs_root(tree_root
->fs_info
, root
);
8634 free_extent_buffer(root
->node
);
8635 free_extent_buffer(root
->commit_root
);
8636 btrfs_put_fs_root(root
);
8638 root_dropped
= true;
8640 ret
= btrfs_delayed_qgroup_accounting(trans
, tree_root
->fs_info
);
8642 printk_ratelimited(KERN_ERR
"BTRFS: Failure %d "
8643 "running qgroup updates "
8644 "during snapshot delete. "
8645 "Quota is out of sync, "
8646 "rescan required.\n", ret
);
8648 btrfs_end_transaction_throttle(trans
, tree_root
);
8651 btrfs_free_path(path
);
8654 * So if we need to stop dropping the snapshot for whatever reason we
8655 * need to make sure to add it back to the dead root list so that we
8656 * keep trying to do the work later. This also cleans up roots if we
8657 * don't have it in the radix (like when we recover after a power fail
8658 * or unmount) so we don't leak memory.
8660 if (!for_reloc
&& root_dropped
== false)
8661 btrfs_add_dead_root(root
);
8662 if (err
&& err
!= -EAGAIN
)
8663 btrfs_std_error(root
->fs_info
, err
);
8668 * drop subtree rooted at tree block 'node'.
8670 * NOTE: this function will unlock and release tree block 'node'
8671 * only used by relocation code
8673 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
8674 struct btrfs_root
*root
,
8675 struct extent_buffer
*node
,
8676 struct extent_buffer
*parent
)
8678 struct btrfs_path
*path
;
8679 struct walk_control
*wc
;
8685 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
8687 path
= btrfs_alloc_path();
8691 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
8693 btrfs_free_path(path
);
8697 btrfs_assert_tree_locked(parent
);
8698 parent_level
= btrfs_header_level(parent
);
8699 extent_buffer_get(parent
);
8700 path
->nodes
[parent_level
] = parent
;
8701 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
8703 btrfs_assert_tree_locked(node
);
8704 level
= btrfs_header_level(node
);
8705 path
->nodes
[level
] = node
;
8706 path
->slots
[level
] = 0;
8707 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
8709 wc
->refs
[parent_level
] = 1;
8710 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
8712 wc
->shared_level
= -1;
8713 wc
->stage
= DROP_REFERENCE
;
8717 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
8720 wret
= walk_down_tree(trans
, root
, path
, wc
);
8726 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
8734 btrfs_free_path(path
);
8738 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
8744 * if restripe for this chunk_type is on pick target profile and
8745 * return, otherwise do the usual balance
8747 stripped
= get_restripe_target(root
->fs_info
, flags
);
8749 return extended_to_chunk(stripped
);
8751 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
8753 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
8754 BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
|
8755 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
8757 if (num_devices
== 1) {
8758 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
8759 stripped
= flags
& ~stripped
;
8761 /* turn raid0 into single device chunks */
8762 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
8765 /* turn mirroring into duplication */
8766 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
8767 BTRFS_BLOCK_GROUP_RAID10
))
8768 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
8770 /* they already had raid on here, just return */
8771 if (flags
& stripped
)
8774 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
8775 stripped
= flags
& ~stripped
;
8777 /* switch duplicated blocks with raid1 */
8778 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
8779 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
8781 /* this is drive concat, leave it alone */
8787 static int set_block_group_ro(struct btrfs_block_group_cache
*cache
, int force
)
8789 struct btrfs_space_info
*sinfo
= cache
->space_info
;
8791 u64 min_allocable_bytes
;
8796 * We need some metadata space and system metadata space for
8797 * allocating chunks in some corner cases until we force to set
8798 * it to be readonly.
8801 (BTRFS_BLOCK_GROUP_SYSTEM
| BTRFS_BLOCK_GROUP_METADATA
)) &&
8803 min_allocable_bytes
= 1 * 1024 * 1024;
8805 min_allocable_bytes
= 0;
8807 spin_lock(&sinfo
->lock
);
8808 spin_lock(&cache
->lock
);
8815 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
8816 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
8818 if (sinfo
->bytes_used
+ sinfo
->bytes_reserved
+ sinfo
->bytes_pinned
+
8819 sinfo
->bytes_may_use
+ sinfo
->bytes_readonly
+ num_bytes
+
8820 min_allocable_bytes
<= sinfo
->total_bytes
) {
8821 sinfo
->bytes_readonly
+= num_bytes
;
8823 list_add_tail(&cache
->ro_list
, &sinfo
->ro_bgs
);
8827 spin_unlock(&cache
->lock
);
8828 spin_unlock(&sinfo
->lock
);
8832 int btrfs_set_block_group_ro(struct btrfs_root
*root
,
8833 struct btrfs_block_group_cache
*cache
)
8836 struct btrfs_trans_handle
*trans
;
8843 trans
= btrfs_join_transaction(root
);
8845 return PTR_ERR(trans
);
8848 * we're not allowed to set block groups readonly after the dirty
8849 * block groups cache has started writing. If it already started,
8850 * back off and let this transaction commit
8852 mutex_lock(&root
->fs_info
->ro_block_group_mutex
);
8853 if (trans
->transaction
->dirty_bg_run
) {
8854 u64 transid
= trans
->transid
;
8856 mutex_unlock(&root
->fs_info
->ro_block_group_mutex
);
8857 btrfs_end_transaction(trans
, root
);
8859 ret
= btrfs_wait_for_commit(root
, transid
);
8866 * if we are changing raid levels, try to allocate a corresponding
8867 * block group with the new raid level.
8869 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
8870 if (alloc_flags
!= cache
->flags
) {
8871 ret
= do_chunk_alloc(trans
, root
, alloc_flags
,
8874 * ENOSPC is allowed here, we may have enough space
8875 * already allocated at the new raid level to
8884 ret
= set_block_group_ro(cache
, 0);
8887 alloc_flags
= get_alloc_profile(root
, cache
->space_info
->flags
);
8888 ret
= do_chunk_alloc(trans
, root
, alloc_flags
,
8892 ret
= set_block_group_ro(cache
, 0);
8894 if (cache
->flags
& BTRFS_BLOCK_GROUP_SYSTEM
) {
8895 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
8896 lock_chunks(root
->fs_info
->chunk_root
);
8897 check_system_chunk(trans
, root
, alloc_flags
);
8898 unlock_chunks(root
->fs_info
->chunk_root
);
8900 mutex_unlock(&root
->fs_info
->ro_block_group_mutex
);
8902 btrfs_end_transaction(trans
, root
);
8906 int btrfs_force_chunk_alloc(struct btrfs_trans_handle
*trans
,
8907 struct btrfs_root
*root
, u64 type
)
8909 u64 alloc_flags
= get_alloc_profile(root
, type
);
8910 return do_chunk_alloc(trans
, root
, alloc_flags
,
8915 * helper to account the unused space of all the readonly block group in the
8916 * space_info. takes mirrors into account.
8918 u64
btrfs_account_ro_block_groups_free_space(struct btrfs_space_info
*sinfo
)
8920 struct btrfs_block_group_cache
*block_group
;
8924 /* It's df, we don't care if it's racey */
8925 if (list_empty(&sinfo
->ro_bgs
))
8928 spin_lock(&sinfo
->lock
);
8929 list_for_each_entry(block_group
, &sinfo
->ro_bgs
, ro_list
) {
8930 spin_lock(&block_group
->lock
);
8932 if (!block_group
->ro
) {
8933 spin_unlock(&block_group
->lock
);
8937 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_RAID1
|
8938 BTRFS_BLOCK_GROUP_RAID10
|
8939 BTRFS_BLOCK_GROUP_DUP
))
8944 free_bytes
+= (block_group
->key
.offset
-
8945 btrfs_block_group_used(&block_group
->item
)) *
8948 spin_unlock(&block_group
->lock
);
8950 spin_unlock(&sinfo
->lock
);
8955 void btrfs_set_block_group_rw(struct btrfs_root
*root
,
8956 struct btrfs_block_group_cache
*cache
)
8958 struct btrfs_space_info
*sinfo
= cache
->space_info
;
8963 spin_lock(&sinfo
->lock
);
8964 spin_lock(&cache
->lock
);
8965 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
8966 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
8967 sinfo
->bytes_readonly
-= num_bytes
;
8969 list_del_init(&cache
->ro_list
);
8970 spin_unlock(&cache
->lock
);
8971 spin_unlock(&sinfo
->lock
);
8975 * checks to see if its even possible to relocate this block group.
8977 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8978 * ok to go ahead and try.
8980 int btrfs_can_relocate(struct btrfs_root
*root
, u64 bytenr
)
8982 struct btrfs_block_group_cache
*block_group
;
8983 struct btrfs_space_info
*space_info
;
8984 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
8985 struct btrfs_device
*device
;
8986 struct btrfs_trans_handle
*trans
;
8995 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
8997 /* odd, couldn't find the block group, leave it alone */
9001 min_free
= btrfs_block_group_used(&block_group
->item
);
9003 /* no bytes used, we're good */
9007 space_info
= block_group
->space_info
;
9008 spin_lock(&space_info
->lock
);
9010 full
= space_info
->full
;
9013 * if this is the last block group we have in this space, we can't
9014 * relocate it unless we're able to allocate a new chunk below.
9016 * Otherwise, we need to make sure we have room in the space to handle
9017 * all of the extents from this block group. If we can, we're good
9019 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
9020 (space_info
->bytes_used
+ space_info
->bytes_reserved
+
9021 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
9022 min_free
< space_info
->total_bytes
)) {
9023 spin_unlock(&space_info
->lock
);
9026 spin_unlock(&space_info
->lock
);
9029 * ok we don't have enough space, but maybe we have free space on our
9030 * devices to allocate new chunks for relocation, so loop through our
9031 * alloc devices and guess if we have enough space. if this block
9032 * group is going to be restriped, run checks against the target
9033 * profile instead of the current one.
9045 target
= get_restripe_target(root
->fs_info
, block_group
->flags
);
9047 index
= __get_raid_index(extended_to_chunk(target
));
9050 * this is just a balance, so if we were marked as full
9051 * we know there is no space for a new chunk
9056 index
= get_block_group_index(block_group
);
9059 if (index
== BTRFS_RAID_RAID10
) {
9063 } else if (index
== BTRFS_RAID_RAID1
) {
9065 } else if (index
== BTRFS_RAID_DUP
) {
9068 } else if (index
== BTRFS_RAID_RAID0
) {
9069 dev_min
= fs_devices
->rw_devices
;
9070 min_free
= div64_u64(min_free
, dev_min
);
9073 /* We need to do this so that we can look at pending chunks */
9074 trans
= btrfs_join_transaction(root
);
9075 if (IS_ERR(trans
)) {
9076 ret
= PTR_ERR(trans
);
9080 mutex_lock(&root
->fs_info
->chunk_mutex
);
9081 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
9085 * check to make sure we can actually find a chunk with enough
9086 * space to fit our block group in.
9088 if (device
->total_bytes
> device
->bytes_used
+ min_free
&&
9089 !device
->is_tgtdev_for_dev_replace
) {
9090 ret
= find_free_dev_extent(trans
, device
, min_free
,
9095 if (dev_nr
>= dev_min
)
9101 mutex_unlock(&root
->fs_info
->chunk_mutex
);
9102 btrfs_end_transaction(trans
, root
);
9104 btrfs_put_block_group(block_group
);
9108 static int find_first_block_group(struct btrfs_root
*root
,
9109 struct btrfs_path
*path
, struct btrfs_key
*key
)
9112 struct btrfs_key found_key
;
9113 struct extent_buffer
*leaf
;
9116 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
9121 slot
= path
->slots
[0];
9122 leaf
= path
->nodes
[0];
9123 if (slot
>= btrfs_header_nritems(leaf
)) {
9124 ret
= btrfs_next_leaf(root
, path
);
9131 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
9133 if (found_key
.objectid
>= key
->objectid
&&
9134 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
9144 void btrfs_put_block_group_cache(struct btrfs_fs_info
*info
)
9146 struct btrfs_block_group_cache
*block_group
;
9150 struct inode
*inode
;
9152 block_group
= btrfs_lookup_first_block_group(info
, last
);
9153 while (block_group
) {
9154 spin_lock(&block_group
->lock
);
9155 if (block_group
->iref
)
9157 spin_unlock(&block_group
->lock
);
9158 block_group
= next_block_group(info
->tree_root
,
9168 inode
= block_group
->inode
;
9169 block_group
->iref
= 0;
9170 block_group
->inode
= NULL
;
9171 spin_unlock(&block_group
->lock
);
9173 last
= block_group
->key
.objectid
+ block_group
->key
.offset
;
9174 btrfs_put_block_group(block_group
);
9178 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
9180 struct btrfs_block_group_cache
*block_group
;
9181 struct btrfs_space_info
*space_info
;
9182 struct btrfs_caching_control
*caching_ctl
;
9185 down_write(&info
->commit_root_sem
);
9186 while (!list_empty(&info
->caching_block_groups
)) {
9187 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
9188 struct btrfs_caching_control
, list
);
9189 list_del(&caching_ctl
->list
);
9190 put_caching_control(caching_ctl
);
9192 up_write(&info
->commit_root_sem
);
9194 spin_lock(&info
->unused_bgs_lock
);
9195 while (!list_empty(&info
->unused_bgs
)) {
9196 block_group
= list_first_entry(&info
->unused_bgs
,
9197 struct btrfs_block_group_cache
,
9199 list_del_init(&block_group
->bg_list
);
9200 btrfs_put_block_group(block_group
);
9202 spin_unlock(&info
->unused_bgs_lock
);
9204 spin_lock(&info
->block_group_cache_lock
);
9205 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
9206 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
9208 rb_erase(&block_group
->cache_node
,
9209 &info
->block_group_cache_tree
);
9210 RB_CLEAR_NODE(&block_group
->cache_node
);
9211 spin_unlock(&info
->block_group_cache_lock
);
9213 down_write(&block_group
->space_info
->groups_sem
);
9214 list_del(&block_group
->list
);
9215 up_write(&block_group
->space_info
->groups_sem
);
9217 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
9218 wait_block_group_cache_done(block_group
);
9221 * We haven't cached this block group, which means we could
9222 * possibly have excluded extents on this block group.
9224 if (block_group
->cached
== BTRFS_CACHE_NO
||
9225 block_group
->cached
== BTRFS_CACHE_ERROR
)
9226 free_excluded_extents(info
->extent_root
, block_group
);
9228 btrfs_remove_free_space_cache(block_group
);
9229 btrfs_put_block_group(block_group
);
9231 spin_lock(&info
->block_group_cache_lock
);
9233 spin_unlock(&info
->block_group_cache_lock
);
9235 /* now that all the block groups are freed, go through and
9236 * free all the space_info structs. This is only called during
9237 * the final stages of unmount, and so we know nobody is
9238 * using them. We call synchronize_rcu() once before we start,
9239 * just to be on the safe side.
9243 release_global_block_rsv(info
);
9245 while (!list_empty(&info
->space_info
)) {
9248 space_info
= list_entry(info
->space_info
.next
,
9249 struct btrfs_space_info
,
9251 if (btrfs_test_opt(info
->tree_root
, ENOSPC_DEBUG
)) {
9252 if (WARN_ON(space_info
->bytes_pinned
> 0 ||
9253 space_info
->bytes_reserved
> 0 ||
9254 space_info
->bytes_may_use
> 0)) {
9255 dump_space_info(space_info
, 0, 0);
9258 list_del(&space_info
->list
);
9259 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++) {
9260 struct kobject
*kobj
;
9261 kobj
= space_info
->block_group_kobjs
[i
];
9262 space_info
->block_group_kobjs
[i
] = NULL
;
9268 kobject_del(&space_info
->kobj
);
9269 kobject_put(&space_info
->kobj
);
9274 static void __link_block_group(struct btrfs_space_info
*space_info
,
9275 struct btrfs_block_group_cache
*cache
)
9277 int index
= get_block_group_index(cache
);
9280 down_write(&space_info
->groups_sem
);
9281 if (list_empty(&space_info
->block_groups
[index
]))
9283 list_add_tail(&cache
->list
, &space_info
->block_groups
[index
]);
9284 up_write(&space_info
->groups_sem
);
9287 struct raid_kobject
*rkobj
;
9290 rkobj
= kzalloc(sizeof(*rkobj
), GFP_NOFS
);
9293 rkobj
->raid_type
= index
;
9294 kobject_init(&rkobj
->kobj
, &btrfs_raid_ktype
);
9295 ret
= kobject_add(&rkobj
->kobj
, &space_info
->kobj
,
9296 "%s", get_raid_name(index
));
9298 kobject_put(&rkobj
->kobj
);
9301 space_info
->block_group_kobjs
[index
] = &rkobj
->kobj
;
9306 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
9309 static struct btrfs_block_group_cache
*
9310 btrfs_create_block_group_cache(struct btrfs_root
*root
, u64 start
, u64 size
)
9312 struct btrfs_block_group_cache
*cache
;
9314 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
9318 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
9320 if (!cache
->free_space_ctl
) {
9325 cache
->key
.objectid
= start
;
9326 cache
->key
.offset
= size
;
9327 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
9329 cache
->sectorsize
= root
->sectorsize
;
9330 cache
->fs_info
= root
->fs_info
;
9331 cache
->full_stripe_len
= btrfs_full_stripe_len(root
,
9332 &root
->fs_info
->mapping_tree
,
9334 atomic_set(&cache
->count
, 1);
9335 spin_lock_init(&cache
->lock
);
9336 init_rwsem(&cache
->data_rwsem
);
9337 INIT_LIST_HEAD(&cache
->list
);
9338 INIT_LIST_HEAD(&cache
->cluster_list
);
9339 INIT_LIST_HEAD(&cache
->bg_list
);
9340 INIT_LIST_HEAD(&cache
->ro_list
);
9341 INIT_LIST_HEAD(&cache
->dirty_list
);
9342 INIT_LIST_HEAD(&cache
->io_list
);
9343 btrfs_init_free_space_ctl(cache
);
9344 atomic_set(&cache
->trimming
, 0);
9349 int btrfs_read_block_groups(struct btrfs_root
*root
)
9351 struct btrfs_path
*path
;
9353 struct btrfs_block_group_cache
*cache
;
9354 struct btrfs_fs_info
*info
= root
->fs_info
;
9355 struct btrfs_space_info
*space_info
;
9356 struct btrfs_key key
;
9357 struct btrfs_key found_key
;
9358 struct extent_buffer
*leaf
;
9362 root
= info
->extent_root
;
9365 key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
9366 path
= btrfs_alloc_path();
9371 cache_gen
= btrfs_super_cache_generation(root
->fs_info
->super_copy
);
9372 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
9373 btrfs_super_generation(root
->fs_info
->super_copy
) != cache_gen
)
9375 if (btrfs_test_opt(root
, CLEAR_CACHE
))
9379 ret
= find_first_block_group(root
, path
, &key
);
9385 leaf
= path
->nodes
[0];
9386 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
9388 cache
= btrfs_create_block_group_cache(root
, found_key
.objectid
,
9397 * When we mount with old space cache, we need to
9398 * set BTRFS_DC_CLEAR and set dirty flag.
9400 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9401 * truncate the old free space cache inode and
9403 * b) Setting 'dirty flag' makes sure that we flush
9404 * the new space cache info onto disk.
9406 if (btrfs_test_opt(root
, SPACE_CACHE
))
9407 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
9410 read_extent_buffer(leaf
, &cache
->item
,
9411 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
9412 sizeof(cache
->item
));
9413 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
9415 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
9416 btrfs_release_path(path
);
9419 * We need to exclude the super stripes now so that the space
9420 * info has super bytes accounted for, otherwise we'll think
9421 * we have more space than we actually do.
9423 ret
= exclude_super_stripes(root
, cache
);
9426 * We may have excluded something, so call this just in
9429 free_excluded_extents(root
, cache
);
9430 btrfs_put_block_group(cache
);
9435 * check for two cases, either we are full, and therefore
9436 * don't need to bother with the caching work since we won't
9437 * find any space, or we are empty, and we can just add all
9438 * the space in and be done with it. This saves us _alot_ of
9439 * time, particularly in the full case.
9441 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
9442 cache
->last_byte_to_unpin
= (u64
)-1;
9443 cache
->cached
= BTRFS_CACHE_FINISHED
;
9444 free_excluded_extents(root
, cache
);
9445 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
9446 cache
->last_byte_to_unpin
= (u64
)-1;
9447 cache
->cached
= BTRFS_CACHE_FINISHED
;
9448 add_new_free_space(cache
, root
->fs_info
,
9450 found_key
.objectid
+
9452 free_excluded_extents(root
, cache
);
9455 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
9457 btrfs_remove_free_space_cache(cache
);
9458 btrfs_put_block_group(cache
);
9462 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
9463 btrfs_block_group_used(&cache
->item
),
9466 btrfs_remove_free_space_cache(cache
);
9467 spin_lock(&info
->block_group_cache_lock
);
9468 rb_erase(&cache
->cache_node
,
9469 &info
->block_group_cache_tree
);
9470 RB_CLEAR_NODE(&cache
->cache_node
);
9471 spin_unlock(&info
->block_group_cache_lock
);
9472 btrfs_put_block_group(cache
);
9476 cache
->space_info
= space_info
;
9477 spin_lock(&cache
->space_info
->lock
);
9478 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
9479 spin_unlock(&cache
->space_info
->lock
);
9481 __link_block_group(space_info
, cache
);
9483 set_avail_alloc_bits(root
->fs_info
, cache
->flags
);
9484 if (btrfs_chunk_readonly(root
, cache
->key
.objectid
)) {
9485 set_block_group_ro(cache
, 1);
9486 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
9487 spin_lock(&info
->unused_bgs_lock
);
9488 /* Should always be true but just in case. */
9489 if (list_empty(&cache
->bg_list
)) {
9490 btrfs_get_block_group(cache
);
9491 list_add_tail(&cache
->bg_list
,
9494 spin_unlock(&info
->unused_bgs_lock
);
9498 list_for_each_entry_rcu(space_info
, &root
->fs_info
->space_info
, list
) {
9499 if (!(get_alloc_profile(root
, space_info
->flags
) &
9500 (BTRFS_BLOCK_GROUP_RAID10
|
9501 BTRFS_BLOCK_GROUP_RAID1
|
9502 BTRFS_BLOCK_GROUP_RAID5
|
9503 BTRFS_BLOCK_GROUP_RAID6
|
9504 BTRFS_BLOCK_GROUP_DUP
)))
9507 * avoid allocating from un-mirrored block group if there are
9508 * mirrored block groups.
9510 list_for_each_entry(cache
,
9511 &space_info
->block_groups
[BTRFS_RAID_RAID0
],
9513 set_block_group_ro(cache
, 1);
9514 list_for_each_entry(cache
,
9515 &space_info
->block_groups
[BTRFS_RAID_SINGLE
],
9517 set_block_group_ro(cache
, 1);
9520 init_global_block_rsv(info
);
9523 btrfs_free_path(path
);
9527 void btrfs_create_pending_block_groups(struct btrfs_trans_handle
*trans
,
9528 struct btrfs_root
*root
)
9530 struct btrfs_block_group_cache
*block_group
, *tmp
;
9531 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
9532 struct btrfs_block_group_item item
;
9533 struct btrfs_key key
;
9536 list_for_each_entry_safe(block_group
, tmp
, &trans
->new_bgs
, bg_list
) {
9540 spin_lock(&block_group
->lock
);
9541 memcpy(&item
, &block_group
->item
, sizeof(item
));
9542 memcpy(&key
, &block_group
->key
, sizeof(key
));
9543 spin_unlock(&block_group
->lock
);
9545 ret
= btrfs_insert_item(trans
, extent_root
, &key
, &item
,
9548 btrfs_abort_transaction(trans
, extent_root
, ret
);
9549 ret
= btrfs_finish_chunk_alloc(trans
, extent_root
,
9550 key
.objectid
, key
.offset
);
9552 btrfs_abort_transaction(trans
, extent_root
, ret
);
9554 list_del_init(&block_group
->bg_list
);
9558 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
9559 struct btrfs_root
*root
, u64 bytes_used
,
9560 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
9564 struct btrfs_root
*extent_root
;
9565 struct btrfs_block_group_cache
*cache
;
9567 extent_root
= root
->fs_info
->extent_root
;
9569 btrfs_set_log_full_commit(root
->fs_info
, trans
);
9571 cache
= btrfs_create_block_group_cache(root
, chunk_offset
, size
);
9575 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
9576 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
9577 btrfs_set_block_group_flags(&cache
->item
, type
);
9579 cache
->flags
= type
;
9580 cache
->last_byte_to_unpin
= (u64
)-1;
9581 cache
->cached
= BTRFS_CACHE_FINISHED
;
9582 ret
= exclude_super_stripes(root
, cache
);
9585 * We may have excluded something, so call this just in
9588 free_excluded_extents(root
, cache
);
9589 btrfs_put_block_group(cache
);
9593 add_new_free_space(cache
, root
->fs_info
, chunk_offset
,
9594 chunk_offset
+ size
);
9596 free_excluded_extents(root
, cache
);
9599 * Call to ensure the corresponding space_info object is created and
9600 * assigned to our block group, but don't update its counters just yet.
9601 * We want our bg to be added to the rbtree with its ->space_info set.
9603 ret
= update_space_info(root
->fs_info
, cache
->flags
, 0, 0,
9604 &cache
->space_info
);
9606 btrfs_remove_free_space_cache(cache
);
9607 btrfs_put_block_group(cache
);
9611 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
9613 btrfs_remove_free_space_cache(cache
);
9614 btrfs_put_block_group(cache
);
9619 * Now that our block group has its ->space_info set and is inserted in
9620 * the rbtree, update the space info's counters.
9622 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
9623 &cache
->space_info
);
9625 btrfs_remove_free_space_cache(cache
);
9626 spin_lock(&root
->fs_info
->block_group_cache_lock
);
9627 rb_erase(&cache
->cache_node
,
9628 &root
->fs_info
->block_group_cache_tree
);
9629 RB_CLEAR_NODE(&cache
->cache_node
);
9630 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
9631 btrfs_put_block_group(cache
);
9634 update_global_block_rsv(root
->fs_info
);
9636 spin_lock(&cache
->space_info
->lock
);
9637 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
9638 spin_unlock(&cache
->space_info
->lock
);
9640 __link_block_group(cache
->space_info
, cache
);
9642 list_add_tail(&cache
->bg_list
, &trans
->new_bgs
);
9644 set_avail_alloc_bits(extent_root
->fs_info
, type
);
9649 static void clear_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
9651 u64 extra_flags
= chunk_to_extended(flags
) &
9652 BTRFS_EXTENDED_PROFILE_MASK
;
9654 write_seqlock(&fs_info
->profiles_lock
);
9655 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
9656 fs_info
->avail_data_alloc_bits
&= ~extra_flags
;
9657 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
9658 fs_info
->avail_metadata_alloc_bits
&= ~extra_flags
;
9659 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
9660 fs_info
->avail_system_alloc_bits
&= ~extra_flags
;
9661 write_sequnlock(&fs_info
->profiles_lock
);
9664 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
9665 struct btrfs_root
*root
, u64 group_start
,
9666 struct extent_map
*em
)
9668 struct btrfs_path
*path
;
9669 struct btrfs_block_group_cache
*block_group
;
9670 struct btrfs_free_cluster
*cluster
;
9671 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
9672 struct btrfs_key key
;
9673 struct inode
*inode
;
9674 struct kobject
*kobj
= NULL
;
9678 struct btrfs_caching_control
*caching_ctl
= NULL
;
9681 root
= root
->fs_info
->extent_root
;
9683 block_group
= btrfs_lookup_block_group(root
->fs_info
, group_start
);
9684 BUG_ON(!block_group
);
9685 BUG_ON(!block_group
->ro
);
9688 * Free the reserved super bytes from this block group before
9691 free_excluded_extents(root
, block_group
);
9693 memcpy(&key
, &block_group
->key
, sizeof(key
));
9694 index
= get_block_group_index(block_group
);
9695 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
9696 BTRFS_BLOCK_GROUP_RAID1
|
9697 BTRFS_BLOCK_GROUP_RAID10
))
9702 /* make sure this block group isn't part of an allocation cluster */
9703 cluster
= &root
->fs_info
->data_alloc_cluster
;
9704 spin_lock(&cluster
->refill_lock
);
9705 btrfs_return_cluster_to_free_space(block_group
, cluster
);
9706 spin_unlock(&cluster
->refill_lock
);
9709 * make sure this block group isn't part of a metadata
9710 * allocation cluster
9712 cluster
= &root
->fs_info
->meta_alloc_cluster
;
9713 spin_lock(&cluster
->refill_lock
);
9714 btrfs_return_cluster_to_free_space(block_group
, cluster
);
9715 spin_unlock(&cluster
->refill_lock
);
9717 path
= btrfs_alloc_path();
9724 * get the inode first so any iput calls done for the io_list
9725 * aren't the final iput (no unlinks allowed now)
9727 inode
= lookup_free_space_inode(tree_root
, block_group
, path
);
9729 mutex_lock(&trans
->transaction
->cache_write_mutex
);
9731 * make sure our free spache cache IO is done before remove the
9734 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
9735 if (!list_empty(&block_group
->io_list
)) {
9736 list_del_init(&block_group
->io_list
);
9738 WARN_ON(!IS_ERR(inode
) && inode
!= block_group
->io_ctl
.inode
);
9740 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
9741 btrfs_wait_cache_io(root
, trans
, block_group
,
9742 &block_group
->io_ctl
, path
,
9743 block_group
->key
.objectid
);
9744 btrfs_put_block_group(block_group
);
9745 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
9748 if (!list_empty(&block_group
->dirty_list
)) {
9749 list_del_init(&block_group
->dirty_list
);
9750 btrfs_put_block_group(block_group
);
9752 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
9753 mutex_unlock(&trans
->transaction
->cache_write_mutex
);
9755 if (!IS_ERR(inode
)) {
9756 ret
= btrfs_orphan_add(trans
, inode
);
9758 btrfs_add_delayed_iput(inode
);
9762 /* One for the block groups ref */
9763 spin_lock(&block_group
->lock
);
9764 if (block_group
->iref
) {
9765 block_group
->iref
= 0;
9766 block_group
->inode
= NULL
;
9767 spin_unlock(&block_group
->lock
);
9770 spin_unlock(&block_group
->lock
);
9772 /* One for our lookup ref */
9773 btrfs_add_delayed_iput(inode
);
9776 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
9777 key
.offset
= block_group
->key
.objectid
;
9780 ret
= btrfs_search_slot(trans
, tree_root
, &key
, path
, -1, 1);
9784 btrfs_release_path(path
);
9786 ret
= btrfs_del_item(trans
, tree_root
, path
);
9789 btrfs_release_path(path
);
9792 spin_lock(&root
->fs_info
->block_group_cache_lock
);
9793 rb_erase(&block_group
->cache_node
,
9794 &root
->fs_info
->block_group_cache_tree
);
9795 RB_CLEAR_NODE(&block_group
->cache_node
);
9797 if (root
->fs_info
->first_logical_byte
== block_group
->key
.objectid
)
9798 root
->fs_info
->first_logical_byte
= (u64
)-1;
9799 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
9801 down_write(&block_group
->space_info
->groups_sem
);
9803 * we must use list_del_init so people can check to see if they
9804 * are still on the list after taking the semaphore
9806 list_del_init(&block_group
->list
);
9807 if (list_empty(&block_group
->space_info
->block_groups
[index
])) {
9808 kobj
= block_group
->space_info
->block_group_kobjs
[index
];
9809 block_group
->space_info
->block_group_kobjs
[index
] = NULL
;
9810 clear_avail_alloc_bits(root
->fs_info
, block_group
->flags
);
9812 up_write(&block_group
->space_info
->groups_sem
);
9818 if (block_group
->has_caching_ctl
)
9819 caching_ctl
= get_caching_control(block_group
);
9820 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
9821 wait_block_group_cache_done(block_group
);
9822 if (block_group
->has_caching_ctl
) {
9823 down_write(&root
->fs_info
->commit_root_sem
);
9825 struct btrfs_caching_control
*ctl
;
9827 list_for_each_entry(ctl
,
9828 &root
->fs_info
->caching_block_groups
, list
)
9829 if (ctl
->block_group
== block_group
) {
9831 atomic_inc(&caching_ctl
->count
);
9836 list_del_init(&caching_ctl
->list
);
9837 up_write(&root
->fs_info
->commit_root_sem
);
9839 /* Once for the caching bgs list and once for us. */
9840 put_caching_control(caching_ctl
);
9841 put_caching_control(caching_ctl
);
9845 spin_lock(&trans
->transaction
->dirty_bgs_lock
);
9846 if (!list_empty(&block_group
->dirty_list
)) {
9849 if (!list_empty(&block_group
->io_list
)) {
9852 spin_unlock(&trans
->transaction
->dirty_bgs_lock
);
9853 btrfs_remove_free_space_cache(block_group
);
9855 spin_lock(&block_group
->space_info
->lock
);
9856 list_del_init(&block_group
->ro_list
);
9858 if (btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
9859 WARN_ON(block_group
->space_info
->total_bytes
9860 < block_group
->key
.offset
);
9861 WARN_ON(block_group
->space_info
->bytes_readonly
9862 < block_group
->key
.offset
);
9863 WARN_ON(block_group
->space_info
->disk_total
9864 < block_group
->key
.offset
* factor
);
9866 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
9867 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
9868 block_group
->space_info
->disk_total
-= block_group
->key
.offset
* factor
;
9870 spin_unlock(&block_group
->space_info
->lock
);
9872 memcpy(&key
, &block_group
->key
, sizeof(key
));
9875 if (!list_empty(&em
->list
)) {
9876 /* We're in the transaction->pending_chunks list. */
9877 free_extent_map(em
);
9879 spin_lock(&block_group
->lock
);
9880 block_group
->removed
= 1;
9882 * At this point trimming can't start on this block group, because we
9883 * removed the block group from the tree fs_info->block_group_cache_tree
9884 * so no one can't find it anymore and even if someone already got this
9885 * block group before we removed it from the rbtree, they have already
9886 * incremented block_group->trimming - if they didn't, they won't find
9887 * any free space entries because we already removed them all when we
9888 * called btrfs_remove_free_space_cache().
9890 * And we must not remove the extent map from the fs_info->mapping_tree
9891 * to prevent the same logical address range and physical device space
9892 * ranges from being reused for a new block group. This is because our
9893 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
9894 * completely transactionless, so while it is trimming a range the
9895 * currently running transaction might finish and a new one start,
9896 * allowing for new block groups to be created that can reuse the same
9897 * physical device locations unless we take this special care.
9899 remove_em
= (atomic_read(&block_group
->trimming
) == 0);
9901 * Make sure a trimmer task always sees the em in the pinned_chunks list
9902 * if it sees block_group->removed == 1 (needs to lock block_group->lock
9903 * before checking block_group->removed).
9907 * Our em might be in trans->transaction->pending_chunks which
9908 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
9909 * and so is the fs_info->pinned_chunks list.
9911 * So at this point we must be holding the chunk_mutex to avoid
9912 * any races with chunk allocation (more specifically at
9913 * volumes.c:contains_pending_extent()), to ensure it always
9914 * sees the em, either in the pending_chunks list or in the
9915 * pinned_chunks list.
9917 list_move_tail(&em
->list
, &root
->fs_info
->pinned_chunks
);
9919 spin_unlock(&block_group
->lock
);
9922 struct extent_map_tree
*em_tree
;
9924 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
9925 write_lock(&em_tree
->lock
);
9927 * The em might be in the pending_chunks list, so make sure the
9928 * chunk mutex is locked, since remove_extent_mapping() will
9929 * delete us from that list.
9931 remove_extent_mapping(em_tree
, em
);
9932 write_unlock(&em_tree
->lock
);
9933 /* once for the tree */
9934 free_extent_map(em
);
9937 unlock_chunks(root
);
9939 btrfs_put_block_group(block_group
);
9940 btrfs_put_block_group(block_group
);
9942 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
9948 ret
= btrfs_del_item(trans
, root
, path
);
9950 btrfs_free_path(path
);
9955 * Process the unused_bgs list and remove any that don't have any allocated
9956 * space inside of them.
9958 void btrfs_delete_unused_bgs(struct btrfs_fs_info
*fs_info
)
9960 struct btrfs_block_group_cache
*block_group
;
9961 struct btrfs_space_info
*space_info
;
9962 struct btrfs_root
*root
= fs_info
->extent_root
;
9963 struct btrfs_trans_handle
*trans
;
9969 spin_lock(&fs_info
->unused_bgs_lock
);
9970 while (!list_empty(&fs_info
->unused_bgs
)) {
9973 block_group
= list_first_entry(&fs_info
->unused_bgs
,
9974 struct btrfs_block_group_cache
,
9976 space_info
= block_group
->space_info
;
9977 list_del_init(&block_group
->bg_list
);
9978 if (ret
|| btrfs_mixed_space_info(space_info
)) {
9979 btrfs_put_block_group(block_group
);
9982 spin_unlock(&fs_info
->unused_bgs_lock
);
9984 /* Don't want to race with allocators so take the groups_sem */
9985 down_write(&space_info
->groups_sem
);
9986 spin_lock(&block_group
->lock
);
9987 if (block_group
->reserved
||
9988 btrfs_block_group_used(&block_group
->item
) ||
9991 * We want to bail if we made new allocations or have
9992 * outstanding allocations in this block group. We do
9993 * the ro check in case balance is currently acting on
9996 spin_unlock(&block_group
->lock
);
9997 up_write(&space_info
->groups_sem
);
10000 spin_unlock(&block_group
->lock
);
10002 /* We don't want to force the issue, only flip if it's ok. */
10003 ret
= set_block_group_ro(block_group
, 0);
10004 up_write(&space_info
->groups_sem
);
10011 * Want to do this before we do anything else so we can recover
10012 * properly if we fail to join the transaction.
10014 /* 1 for btrfs_orphan_reserve_metadata() */
10015 trans
= btrfs_start_transaction(root
, 1);
10016 if (IS_ERR(trans
)) {
10017 btrfs_set_block_group_rw(root
, block_group
);
10018 ret
= PTR_ERR(trans
);
10023 * We could have pending pinned extents for this block group,
10024 * just delete them, we don't care about them anymore.
10026 start
= block_group
->key
.objectid
;
10027 end
= start
+ block_group
->key
.offset
- 1;
10029 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10030 * btrfs_finish_extent_commit(). If we are at transaction N,
10031 * another task might be running finish_extent_commit() for the
10032 * previous transaction N - 1, and have seen a range belonging
10033 * to the block group in freed_extents[] before we were able to
10034 * clear the whole block group range from freed_extents[]. This
10035 * means that task can lookup for the block group after we
10036 * unpinned it from freed_extents[] and removed it, leading to
10037 * a BUG_ON() at btrfs_unpin_extent_range().
10039 mutex_lock(&fs_info
->unused_bg_unpin_mutex
);
10040 ret
= clear_extent_bits(&fs_info
->freed_extents
[0], start
, end
,
10041 EXTENT_DIRTY
, GFP_NOFS
);
10043 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10044 btrfs_set_block_group_rw(root
, block_group
);
10047 ret
= clear_extent_bits(&fs_info
->freed_extents
[1], start
, end
,
10048 EXTENT_DIRTY
, GFP_NOFS
);
10050 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10051 btrfs_set_block_group_rw(root
, block_group
);
10054 mutex_unlock(&fs_info
->unused_bg_unpin_mutex
);
10056 /* Reset pinned so btrfs_put_block_group doesn't complain */
10057 spin_lock(&space_info
->lock
);
10058 spin_lock(&block_group
->lock
);
10060 space_info
->bytes_pinned
-= block_group
->pinned
;
10061 space_info
->bytes_readonly
+= block_group
->pinned
;
10062 percpu_counter_add(&space_info
->total_bytes_pinned
,
10063 -block_group
->pinned
);
10064 block_group
->pinned
= 0;
10066 spin_unlock(&block_group
->lock
);
10067 spin_unlock(&space_info
->lock
);
10070 * Btrfs_remove_chunk will abort the transaction if things go
10073 ret
= btrfs_remove_chunk(trans
, root
,
10074 block_group
->key
.objectid
);
10076 btrfs_end_transaction(trans
, root
);
10078 btrfs_put_block_group(block_group
);
10079 spin_lock(&fs_info
->unused_bgs_lock
);
10081 spin_unlock(&fs_info
->unused_bgs_lock
);
10084 int btrfs_init_space_info(struct btrfs_fs_info
*fs_info
)
10086 struct btrfs_space_info
*space_info
;
10087 struct btrfs_super_block
*disk_super
;
10093 disk_super
= fs_info
->super_copy
;
10094 if (!btrfs_super_root(disk_super
))
10097 features
= btrfs_super_incompat_flags(disk_super
);
10098 if (features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
10101 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
10102 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
10107 flags
= BTRFS_BLOCK_GROUP_METADATA
| BTRFS_BLOCK_GROUP_DATA
;
10108 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
10110 flags
= BTRFS_BLOCK_GROUP_METADATA
;
10111 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
10115 flags
= BTRFS_BLOCK_GROUP_DATA
;
10116 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
10122 int btrfs_error_unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
10124 return unpin_extent_range(root
, start
, end
, false);
10127 int btrfs_trim_fs(struct btrfs_root
*root
, struct fstrim_range
*range
)
10129 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
10130 struct btrfs_block_group_cache
*cache
= NULL
;
10135 u64 total_bytes
= btrfs_super_total_bytes(fs_info
->super_copy
);
10139 * try to trim all FS space, our block group may start from non-zero.
10141 if (range
->len
== total_bytes
)
10142 cache
= btrfs_lookup_first_block_group(fs_info
, range
->start
);
10144 cache
= btrfs_lookup_block_group(fs_info
, range
->start
);
10147 if (cache
->key
.objectid
>= (range
->start
+ range
->len
)) {
10148 btrfs_put_block_group(cache
);
10152 start
= max(range
->start
, cache
->key
.objectid
);
10153 end
= min(range
->start
+ range
->len
,
10154 cache
->key
.objectid
+ cache
->key
.offset
);
10156 if (end
- start
>= range
->minlen
) {
10157 if (!block_group_cache_done(cache
)) {
10158 ret
= cache_block_group(cache
, 0);
10160 btrfs_put_block_group(cache
);
10163 ret
= wait_block_group_cache_done(cache
);
10165 btrfs_put_block_group(cache
);
10169 ret
= btrfs_trim_block_group(cache
,
10175 trimmed
+= group_trimmed
;
10177 btrfs_put_block_group(cache
);
10182 cache
= next_block_group(fs_info
->tree_root
, cache
);
10185 range
->len
= trimmed
;
10190 * btrfs_{start,end}_write_no_snapshoting() are similar to
10191 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10192 * data into the page cache through nocow before the subvolume is snapshoted,
10193 * but flush the data into disk after the snapshot creation, or to prevent
10194 * operations while snapshoting is ongoing and that cause the snapshot to be
10195 * inconsistent (writes followed by expanding truncates for example).
10197 void btrfs_end_write_no_snapshoting(struct btrfs_root
*root
)
10199 percpu_counter_dec(&root
->subv_writers
->counter
);
10201 * Make sure counter is updated before we wake up
10205 if (waitqueue_active(&root
->subv_writers
->wait
))
10206 wake_up(&root
->subv_writers
->wait
);
10209 int btrfs_start_write_no_snapshoting(struct btrfs_root
*root
)
10211 if (atomic_read(&root
->will_be_snapshoted
))
10214 percpu_counter_inc(&root
->subv_writers
->counter
);
10216 * Make sure counter is updated before we check for snapshot creation.
10219 if (atomic_read(&root
->will_be_snapshoted
)) {
10220 btrfs_end_write_no_snapshoting(root
);