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>
31 #include "print-tree.h"
32 #include "transaction.h"
35 #include "free-space-cache.h"
37 /* control flags for do_chunk_alloc's force field
38 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
39 * if we really need one.
41 * CHUNK_ALLOC_FORCE means it must try to allocate one
43 * CHUNK_ALLOC_LIMITED means to only try and allocate one
44 * if we have very few chunks already allocated. This is
45 * used as part of the clustering code to help make sure
46 * we have a good pool of storage to cluster in, without
47 * filling the FS with empty chunks
51 CHUNK_ALLOC_NO_FORCE
= 0,
52 CHUNK_ALLOC_FORCE
= 1,
53 CHUNK_ALLOC_LIMITED
= 2,
57 * Control how reservations are dealt with.
59 * RESERVE_FREE - freeing a reservation.
60 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
62 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
63 * bytes_may_use as the ENOSPC accounting is done elsewhere
68 RESERVE_ALLOC_NO_ACCOUNT
= 2,
71 static int update_block_group(struct btrfs_trans_handle
*trans
,
72 struct btrfs_root
*root
,
73 u64 bytenr
, u64 num_bytes
, int alloc
);
74 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
75 struct btrfs_root
*root
,
76 u64 bytenr
, u64 num_bytes
, u64 parent
,
77 u64 root_objectid
, u64 owner_objectid
,
78 u64 owner_offset
, int refs_to_drop
,
79 struct btrfs_delayed_extent_op
*extra_op
);
80 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
81 struct extent_buffer
*leaf
,
82 struct btrfs_extent_item
*ei
);
83 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
84 struct btrfs_root
*root
,
85 u64 parent
, u64 root_objectid
,
86 u64 flags
, u64 owner
, u64 offset
,
87 struct btrfs_key
*ins
, int ref_mod
);
88 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
89 struct btrfs_root
*root
,
90 u64 parent
, u64 root_objectid
,
91 u64 flags
, struct btrfs_disk_key
*key
,
92 int level
, struct btrfs_key
*ins
);
93 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
94 struct btrfs_root
*extent_root
, u64 alloc_bytes
,
95 u64 flags
, int force
);
96 static int find_next_key(struct btrfs_path
*path
, int level
,
97 struct btrfs_key
*key
);
98 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
99 int dump_block_groups
);
100 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
101 u64 num_bytes
, int reserve
);
104 block_group_cache_done(struct btrfs_block_group_cache
*cache
)
107 return cache
->cached
== BTRFS_CACHE_FINISHED
;
110 static int block_group_bits(struct btrfs_block_group_cache
*cache
, u64 bits
)
112 return (cache
->flags
& bits
) == bits
;
115 static void btrfs_get_block_group(struct btrfs_block_group_cache
*cache
)
117 atomic_inc(&cache
->count
);
120 void btrfs_put_block_group(struct btrfs_block_group_cache
*cache
)
122 if (atomic_dec_and_test(&cache
->count
)) {
123 WARN_ON(cache
->pinned
> 0);
124 WARN_ON(cache
->reserved
> 0);
125 kfree(cache
->free_space_ctl
);
131 * this adds the block group to the fs_info rb tree for the block group
134 static int btrfs_add_block_group_cache(struct btrfs_fs_info
*info
,
135 struct btrfs_block_group_cache
*block_group
)
138 struct rb_node
*parent
= NULL
;
139 struct btrfs_block_group_cache
*cache
;
141 spin_lock(&info
->block_group_cache_lock
);
142 p
= &info
->block_group_cache_tree
.rb_node
;
146 cache
= rb_entry(parent
, struct btrfs_block_group_cache
,
148 if (block_group
->key
.objectid
< cache
->key
.objectid
) {
150 } else if (block_group
->key
.objectid
> cache
->key
.objectid
) {
153 spin_unlock(&info
->block_group_cache_lock
);
158 rb_link_node(&block_group
->cache_node
, parent
, p
);
159 rb_insert_color(&block_group
->cache_node
,
160 &info
->block_group_cache_tree
);
161 spin_unlock(&info
->block_group_cache_lock
);
167 * This will return the block group at or after bytenr if contains is 0, else
168 * it will return the block group that contains the bytenr
170 static struct btrfs_block_group_cache
*
171 block_group_cache_tree_search(struct btrfs_fs_info
*info
, u64 bytenr
,
174 struct btrfs_block_group_cache
*cache
, *ret
= NULL
;
178 spin_lock(&info
->block_group_cache_lock
);
179 n
= info
->block_group_cache_tree
.rb_node
;
182 cache
= rb_entry(n
, struct btrfs_block_group_cache
,
184 end
= cache
->key
.objectid
+ cache
->key
.offset
- 1;
185 start
= cache
->key
.objectid
;
187 if (bytenr
< start
) {
188 if (!contains
&& (!ret
|| start
< ret
->key
.objectid
))
191 } else if (bytenr
> start
) {
192 if (contains
&& bytenr
<= end
) {
203 btrfs_get_block_group(ret
);
204 spin_unlock(&info
->block_group_cache_lock
);
209 static int add_excluded_extent(struct btrfs_root
*root
,
210 u64 start
, u64 num_bytes
)
212 u64 end
= start
+ num_bytes
- 1;
213 set_extent_bits(&root
->fs_info
->freed_extents
[0],
214 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
215 set_extent_bits(&root
->fs_info
->freed_extents
[1],
216 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
220 static void free_excluded_extents(struct btrfs_root
*root
,
221 struct btrfs_block_group_cache
*cache
)
225 start
= cache
->key
.objectid
;
226 end
= start
+ cache
->key
.offset
- 1;
228 clear_extent_bits(&root
->fs_info
->freed_extents
[0],
229 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
230 clear_extent_bits(&root
->fs_info
->freed_extents
[1],
231 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
234 static int exclude_super_stripes(struct btrfs_root
*root
,
235 struct btrfs_block_group_cache
*cache
)
242 if (cache
->key
.objectid
< BTRFS_SUPER_INFO_OFFSET
) {
243 stripe_len
= BTRFS_SUPER_INFO_OFFSET
- cache
->key
.objectid
;
244 cache
->bytes_super
+= stripe_len
;
245 ret
= add_excluded_extent(root
, cache
->key
.objectid
,
250 for (i
= 0; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
251 bytenr
= btrfs_sb_offset(i
);
252 ret
= btrfs_rmap_block(&root
->fs_info
->mapping_tree
,
253 cache
->key
.objectid
, bytenr
,
254 0, &logical
, &nr
, &stripe_len
);
258 cache
->bytes_super
+= stripe_len
;
259 ret
= add_excluded_extent(root
, logical
[nr
],
269 static struct btrfs_caching_control
*
270 get_caching_control(struct btrfs_block_group_cache
*cache
)
272 struct btrfs_caching_control
*ctl
;
274 spin_lock(&cache
->lock
);
275 if (cache
->cached
!= BTRFS_CACHE_STARTED
) {
276 spin_unlock(&cache
->lock
);
280 /* We're loading it the fast way, so we don't have a caching_ctl. */
281 if (!cache
->caching_ctl
) {
282 spin_unlock(&cache
->lock
);
286 ctl
= cache
->caching_ctl
;
287 atomic_inc(&ctl
->count
);
288 spin_unlock(&cache
->lock
);
292 static void put_caching_control(struct btrfs_caching_control
*ctl
)
294 if (atomic_dec_and_test(&ctl
->count
))
299 * this is only called by cache_block_group, since we could have freed extents
300 * we need to check the pinned_extents for any extents that can't be used yet
301 * since their free space will be released as soon as the transaction commits.
303 static u64
add_new_free_space(struct btrfs_block_group_cache
*block_group
,
304 struct btrfs_fs_info
*info
, u64 start
, u64 end
)
306 u64 extent_start
, extent_end
, size
, total_added
= 0;
309 while (start
< end
) {
310 ret
= find_first_extent_bit(info
->pinned_extents
, start
,
311 &extent_start
, &extent_end
,
312 EXTENT_DIRTY
| EXTENT_UPTODATE
);
316 if (extent_start
<= start
) {
317 start
= extent_end
+ 1;
318 } else if (extent_start
> start
&& extent_start
< end
) {
319 size
= extent_start
- start
;
321 ret
= btrfs_add_free_space(block_group
, start
,
324 start
= extent_end
+ 1;
333 ret
= btrfs_add_free_space(block_group
, start
, size
);
340 static noinline
void caching_thread(struct btrfs_work
*work
)
342 struct btrfs_block_group_cache
*block_group
;
343 struct btrfs_fs_info
*fs_info
;
344 struct btrfs_caching_control
*caching_ctl
;
345 struct btrfs_root
*extent_root
;
346 struct btrfs_path
*path
;
347 struct extent_buffer
*leaf
;
348 struct btrfs_key key
;
354 caching_ctl
= container_of(work
, struct btrfs_caching_control
, work
);
355 block_group
= caching_ctl
->block_group
;
356 fs_info
= block_group
->fs_info
;
357 extent_root
= fs_info
->extent_root
;
359 path
= btrfs_alloc_path();
363 last
= max_t(u64
, block_group
->key
.objectid
, BTRFS_SUPER_INFO_OFFSET
);
366 * We don't want to deadlock with somebody trying to allocate a new
367 * extent for the extent root while also trying to search the extent
368 * root to add free space. So we skip locking and search the commit
369 * root, since its read-only
371 path
->skip_locking
= 1;
372 path
->search_commit_root
= 1;
377 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
379 mutex_lock(&caching_ctl
->mutex
);
380 /* need to make sure the commit_root doesn't disappear */
381 down_read(&fs_info
->extent_commit_sem
);
383 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
387 leaf
= path
->nodes
[0];
388 nritems
= btrfs_header_nritems(leaf
);
391 if (btrfs_fs_closing(fs_info
) > 1) {
396 if (path
->slots
[0] < nritems
) {
397 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
399 ret
= find_next_key(path
, 0, &key
);
403 if (need_resched() ||
404 btrfs_next_leaf(extent_root
, path
)) {
405 caching_ctl
->progress
= last
;
406 btrfs_release_path(path
);
407 up_read(&fs_info
->extent_commit_sem
);
408 mutex_unlock(&caching_ctl
->mutex
);
412 leaf
= path
->nodes
[0];
413 nritems
= btrfs_header_nritems(leaf
);
417 if (key
.objectid
< block_group
->key
.objectid
) {
422 if (key
.objectid
>= block_group
->key
.objectid
+
423 block_group
->key
.offset
)
426 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
) {
427 total_found
+= add_new_free_space(block_group
,
430 last
= key
.objectid
+ key
.offset
;
432 if (total_found
> (1024 * 1024 * 2)) {
434 wake_up(&caching_ctl
->wait
);
441 total_found
+= add_new_free_space(block_group
, fs_info
, last
,
442 block_group
->key
.objectid
+
443 block_group
->key
.offset
);
444 caching_ctl
->progress
= (u64
)-1;
446 spin_lock(&block_group
->lock
);
447 block_group
->caching_ctl
= NULL
;
448 block_group
->cached
= BTRFS_CACHE_FINISHED
;
449 spin_unlock(&block_group
->lock
);
452 btrfs_free_path(path
);
453 up_read(&fs_info
->extent_commit_sem
);
455 free_excluded_extents(extent_root
, block_group
);
457 mutex_unlock(&caching_ctl
->mutex
);
459 wake_up(&caching_ctl
->wait
);
461 put_caching_control(caching_ctl
);
462 btrfs_put_block_group(block_group
);
465 static int cache_block_group(struct btrfs_block_group_cache
*cache
,
466 struct btrfs_trans_handle
*trans
,
467 struct btrfs_root
*root
,
471 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
472 struct btrfs_caching_control
*caching_ctl
;
475 caching_ctl
= kzalloc(sizeof(*caching_ctl
), GFP_NOFS
);
476 BUG_ON(!caching_ctl
);
478 INIT_LIST_HEAD(&caching_ctl
->list
);
479 mutex_init(&caching_ctl
->mutex
);
480 init_waitqueue_head(&caching_ctl
->wait
);
481 caching_ctl
->block_group
= cache
;
482 caching_ctl
->progress
= cache
->key
.objectid
;
483 atomic_set(&caching_ctl
->count
, 1);
484 caching_ctl
->work
.func
= caching_thread
;
486 spin_lock(&cache
->lock
);
488 * This should be a rare occasion, but this could happen I think in the
489 * case where one thread starts to load the space cache info, and then
490 * some other thread starts a transaction commit which tries to do an
491 * allocation while the other thread is still loading the space cache
492 * info. The previous loop should have kept us from choosing this block
493 * group, but if we've moved to the state where we will wait on caching
494 * block groups we need to first check if we're doing a fast load here,
495 * so we can wait for it to finish, otherwise we could end up allocating
496 * from a block group who's cache gets evicted for one reason or
499 while (cache
->cached
== BTRFS_CACHE_FAST
) {
500 struct btrfs_caching_control
*ctl
;
502 ctl
= cache
->caching_ctl
;
503 atomic_inc(&ctl
->count
);
504 prepare_to_wait(&ctl
->wait
, &wait
, TASK_UNINTERRUPTIBLE
);
505 spin_unlock(&cache
->lock
);
509 finish_wait(&ctl
->wait
, &wait
);
510 put_caching_control(ctl
);
511 spin_lock(&cache
->lock
);
514 if (cache
->cached
!= BTRFS_CACHE_NO
) {
515 spin_unlock(&cache
->lock
);
519 WARN_ON(cache
->caching_ctl
);
520 cache
->caching_ctl
= caching_ctl
;
521 cache
->cached
= BTRFS_CACHE_FAST
;
522 spin_unlock(&cache
->lock
);
525 * We can't do the read from on-disk cache during a commit since we need
526 * to have the normal tree locking. Also if we are currently trying to
527 * allocate blocks for the tree root we can't do the fast caching since
528 * we likely hold important locks.
530 if (trans
&& (!trans
->transaction
->in_commit
) &&
531 (root
&& root
!= root
->fs_info
->tree_root
) &&
532 btrfs_test_opt(root
, SPACE_CACHE
)) {
533 ret
= load_free_space_cache(fs_info
, cache
);
535 spin_lock(&cache
->lock
);
537 cache
->caching_ctl
= NULL
;
538 cache
->cached
= BTRFS_CACHE_FINISHED
;
539 cache
->last_byte_to_unpin
= (u64
)-1;
541 if (load_cache_only
) {
542 cache
->caching_ctl
= NULL
;
543 cache
->cached
= BTRFS_CACHE_NO
;
545 cache
->cached
= BTRFS_CACHE_STARTED
;
548 spin_unlock(&cache
->lock
);
549 wake_up(&caching_ctl
->wait
);
551 put_caching_control(caching_ctl
);
552 free_excluded_extents(fs_info
->extent_root
, cache
);
557 * We are not going to do the fast caching, set cached to the
558 * appropriate value and wakeup any waiters.
560 spin_lock(&cache
->lock
);
561 if (load_cache_only
) {
562 cache
->caching_ctl
= NULL
;
563 cache
->cached
= BTRFS_CACHE_NO
;
565 cache
->cached
= BTRFS_CACHE_STARTED
;
567 spin_unlock(&cache
->lock
);
568 wake_up(&caching_ctl
->wait
);
571 if (load_cache_only
) {
572 put_caching_control(caching_ctl
);
576 down_write(&fs_info
->extent_commit_sem
);
577 atomic_inc(&caching_ctl
->count
);
578 list_add_tail(&caching_ctl
->list
, &fs_info
->caching_block_groups
);
579 up_write(&fs_info
->extent_commit_sem
);
581 btrfs_get_block_group(cache
);
583 btrfs_queue_worker(&fs_info
->caching_workers
, &caching_ctl
->work
);
589 * return the block group that starts at or after bytenr
591 static struct btrfs_block_group_cache
*
592 btrfs_lookup_first_block_group(struct btrfs_fs_info
*info
, u64 bytenr
)
594 struct btrfs_block_group_cache
*cache
;
596 cache
= block_group_cache_tree_search(info
, bytenr
, 0);
602 * return the block group that contains the given bytenr
604 struct btrfs_block_group_cache
*btrfs_lookup_block_group(
605 struct btrfs_fs_info
*info
,
608 struct btrfs_block_group_cache
*cache
;
610 cache
= block_group_cache_tree_search(info
, bytenr
, 1);
615 static struct btrfs_space_info
*__find_space_info(struct btrfs_fs_info
*info
,
618 struct list_head
*head
= &info
->space_info
;
619 struct btrfs_space_info
*found
;
621 flags
&= BTRFS_BLOCK_GROUP_TYPE_MASK
;
624 list_for_each_entry_rcu(found
, head
, list
) {
625 if (found
->flags
& flags
) {
635 * after adding space to the filesystem, we need to clear the full flags
636 * on all the space infos.
638 void btrfs_clear_space_info_full(struct btrfs_fs_info
*info
)
640 struct list_head
*head
= &info
->space_info
;
641 struct btrfs_space_info
*found
;
644 list_for_each_entry_rcu(found
, head
, list
)
649 static u64
div_factor(u64 num
, int factor
)
658 static u64
div_factor_fine(u64 num
, int factor
)
667 u64
btrfs_find_block_group(struct btrfs_root
*root
,
668 u64 search_start
, u64 search_hint
, int owner
)
670 struct btrfs_block_group_cache
*cache
;
672 u64 last
= max(search_hint
, search_start
);
679 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
683 spin_lock(&cache
->lock
);
684 last
= cache
->key
.objectid
+ cache
->key
.offset
;
685 used
= btrfs_block_group_used(&cache
->item
);
687 if ((full_search
|| !cache
->ro
) &&
688 block_group_bits(cache
, BTRFS_BLOCK_GROUP_METADATA
)) {
689 if (used
+ cache
->pinned
+ cache
->reserved
<
690 div_factor(cache
->key
.offset
, factor
)) {
691 group_start
= cache
->key
.objectid
;
692 spin_unlock(&cache
->lock
);
693 btrfs_put_block_group(cache
);
697 spin_unlock(&cache
->lock
);
698 btrfs_put_block_group(cache
);
706 if (!full_search
&& factor
< 10) {
716 /* simple helper to search for an existing extent at a given offset */
717 int btrfs_lookup_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
720 struct btrfs_key key
;
721 struct btrfs_path
*path
;
723 path
= btrfs_alloc_path();
727 key
.objectid
= start
;
729 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
730 ret
= btrfs_search_slot(NULL
, root
->fs_info
->extent_root
, &key
, path
,
732 btrfs_free_path(path
);
737 * helper function to lookup reference count and flags of extent.
739 * the head node for delayed ref is used to store the sum of all the
740 * reference count modifications queued up in the rbtree. the head
741 * node may also store the extent flags to set. This way you can check
742 * to see what the reference count and extent flags would be if all of
743 * the delayed refs are not processed.
745 int btrfs_lookup_extent_info(struct btrfs_trans_handle
*trans
,
746 struct btrfs_root
*root
, u64 bytenr
,
747 u64 num_bytes
, u64
*refs
, u64
*flags
)
749 struct btrfs_delayed_ref_head
*head
;
750 struct btrfs_delayed_ref_root
*delayed_refs
;
751 struct btrfs_path
*path
;
752 struct btrfs_extent_item
*ei
;
753 struct extent_buffer
*leaf
;
754 struct btrfs_key key
;
760 path
= btrfs_alloc_path();
764 key
.objectid
= bytenr
;
765 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
766 key
.offset
= num_bytes
;
768 path
->skip_locking
= 1;
769 path
->search_commit_root
= 1;
772 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
,
778 leaf
= path
->nodes
[0];
779 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
780 if (item_size
>= sizeof(*ei
)) {
781 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
782 struct btrfs_extent_item
);
783 num_refs
= btrfs_extent_refs(leaf
, ei
);
784 extent_flags
= btrfs_extent_flags(leaf
, ei
);
786 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
787 struct btrfs_extent_item_v0
*ei0
;
788 BUG_ON(item_size
!= sizeof(*ei0
));
789 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
790 struct btrfs_extent_item_v0
);
791 num_refs
= btrfs_extent_refs_v0(leaf
, ei0
);
792 /* FIXME: this isn't correct for data */
793 extent_flags
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
798 BUG_ON(num_refs
== 0);
808 delayed_refs
= &trans
->transaction
->delayed_refs
;
809 spin_lock(&delayed_refs
->lock
);
810 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
812 if (!mutex_trylock(&head
->mutex
)) {
813 atomic_inc(&head
->node
.refs
);
814 spin_unlock(&delayed_refs
->lock
);
816 btrfs_release_path(path
);
819 * Mutex was contended, block until it's released and try
822 mutex_lock(&head
->mutex
);
823 mutex_unlock(&head
->mutex
);
824 btrfs_put_delayed_ref(&head
->node
);
827 if (head
->extent_op
&& head
->extent_op
->update_flags
)
828 extent_flags
|= head
->extent_op
->flags_to_set
;
830 BUG_ON(num_refs
== 0);
832 num_refs
+= head
->node
.ref_mod
;
833 mutex_unlock(&head
->mutex
);
835 spin_unlock(&delayed_refs
->lock
);
837 WARN_ON(num_refs
== 0);
841 *flags
= extent_flags
;
843 btrfs_free_path(path
);
848 * Back reference rules. Back refs have three main goals:
850 * 1) differentiate between all holders of references to an extent so that
851 * when a reference is dropped we can make sure it was a valid reference
852 * before freeing the extent.
854 * 2) Provide enough information to quickly find the holders of an extent
855 * if we notice a given block is corrupted or bad.
857 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
858 * maintenance. This is actually the same as #2, but with a slightly
859 * different use case.
861 * There are two kinds of back refs. The implicit back refs is optimized
862 * for pointers in non-shared tree blocks. For a given pointer in a block,
863 * back refs of this kind provide information about the block's owner tree
864 * and the pointer's key. These information allow us to find the block by
865 * b-tree searching. The full back refs is for pointers in tree blocks not
866 * referenced by their owner trees. The location of tree block is recorded
867 * in the back refs. Actually the full back refs is generic, and can be
868 * used in all cases the implicit back refs is used. The major shortcoming
869 * of the full back refs is its overhead. Every time a tree block gets
870 * COWed, we have to update back refs entry for all pointers in it.
872 * For a newly allocated tree block, we use implicit back refs for
873 * pointers in it. This means most tree related operations only involve
874 * implicit back refs. For a tree block created in old transaction, the
875 * only way to drop a reference to it is COW it. So we can detect the
876 * event that tree block loses its owner tree's reference and do the
877 * back refs conversion.
879 * When a tree block is COW'd through a tree, there are four cases:
881 * The reference count of the block is one and the tree is the block's
882 * owner tree. Nothing to do in this case.
884 * The reference count of the block is one and the tree is not the
885 * block's owner tree. In this case, full back refs is used for pointers
886 * in the block. Remove these full back refs, add implicit back refs for
887 * every pointers in the new block.
889 * The reference count of the block is greater than one and the tree is
890 * the block's owner tree. In this case, implicit back refs is used for
891 * pointers in the block. Add full back refs for every pointers in the
892 * block, increase lower level extents' reference counts. The original
893 * implicit back refs are entailed to the new block.
895 * The reference count of the block is greater than one and the tree is
896 * not the block's owner tree. Add implicit back refs for every pointer in
897 * the new block, increase lower level extents' reference count.
899 * Back Reference Key composing:
901 * The key objectid corresponds to the first byte in the extent,
902 * The key type is used to differentiate between types of back refs.
903 * There are different meanings of the key offset for different types
906 * File extents can be referenced by:
908 * - multiple snapshots, subvolumes, or different generations in one subvol
909 * - different files inside a single subvolume
910 * - different offsets inside a file (bookend extents in file.c)
912 * The extent ref structure for the implicit back refs has fields for:
914 * - Objectid of the subvolume root
915 * - objectid of the file holding the reference
916 * - original offset in the file
917 * - how many bookend extents
919 * The key offset for the implicit back refs is hash of the first
922 * The extent ref structure for the full back refs has field for:
924 * - number of pointers in the tree leaf
926 * The key offset for the implicit back refs is the first byte of
929 * When a file extent is allocated, The implicit back refs is used.
930 * the fields are filled in:
932 * (root_key.objectid, inode objectid, offset in file, 1)
934 * When a file extent is removed file truncation, we find the
935 * corresponding implicit back refs and check the following fields:
937 * (btrfs_header_owner(leaf), inode objectid, offset in file)
939 * Btree extents can be referenced by:
941 * - Different subvolumes
943 * Both the implicit back refs and the full back refs for tree blocks
944 * only consist of key. The key offset for the implicit back refs is
945 * objectid of block's owner tree. The key offset for the full back refs
946 * is the first byte of parent block.
948 * When implicit back refs is used, information about the lowest key and
949 * level of the tree block are required. These information are stored in
950 * tree block info structure.
953 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
954 static int convert_extent_item_v0(struct btrfs_trans_handle
*trans
,
955 struct btrfs_root
*root
,
956 struct btrfs_path
*path
,
957 u64 owner
, u32 extra_size
)
959 struct btrfs_extent_item
*item
;
960 struct btrfs_extent_item_v0
*ei0
;
961 struct btrfs_extent_ref_v0
*ref0
;
962 struct btrfs_tree_block_info
*bi
;
963 struct extent_buffer
*leaf
;
964 struct btrfs_key key
;
965 struct btrfs_key found_key
;
966 u32 new_size
= sizeof(*item
);
970 leaf
= path
->nodes
[0];
971 BUG_ON(btrfs_item_size_nr(leaf
, path
->slots
[0]) != sizeof(*ei0
));
973 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
974 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
975 struct btrfs_extent_item_v0
);
976 refs
= btrfs_extent_refs_v0(leaf
, ei0
);
978 if (owner
== (u64
)-1) {
980 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
981 ret
= btrfs_next_leaf(root
, path
);
985 leaf
= path
->nodes
[0];
987 btrfs_item_key_to_cpu(leaf
, &found_key
,
989 BUG_ON(key
.objectid
!= found_key
.objectid
);
990 if (found_key
.type
!= BTRFS_EXTENT_REF_V0_KEY
) {
994 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
995 struct btrfs_extent_ref_v0
);
996 owner
= btrfs_ref_objectid_v0(leaf
, ref0
);
1000 btrfs_release_path(path
);
1002 if (owner
< BTRFS_FIRST_FREE_OBJECTID
)
1003 new_size
+= sizeof(*bi
);
1005 new_size
-= sizeof(*ei0
);
1006 ret
= btrfs_search_slot(trans
, root
, &key
, path
,
1007 new_size
+ extra_size
, 1);
1012 ret
= btrfs_extend_item(trans
, root
, path
, new_size
);
1014 leaf
= path
->nodes
[0];
1015 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1016 btrfs_set_extent_refs(leaf
, item
, refs
);
1017 /* FIXME: get real generation */
1018 btrfs_set_extent_generation(leaf
, item
, 0);
1019 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1020 btrfs_set_extent_flags(leaf
, item
,
1021 BTRFS_EXTENT_FLAG_TREE_BLOCK
|
1022 BTRFS_BLOCK_FLAG_FULL_BACKREF
);
1023 bi
= (struct btrfs_tree_block_info
*)(item
+ 1);
1024 /* FIXME: get first key of the block */
1025 memset_extent_buffer(leaf
, 0, (unsigned long)bi
, sizeof(*bi
));
1026 btrfs_set_tree_block_level(leaf
, bi
, (int)owner
);
1028 btrfs_set_extent_flags(leaf
, item
, BTRFS_EXTENT_FLAG_DATA
);
1030 btrfs_mark_buffer_dirty(leaf
);
1035 static u64
hash_extent_data_ref(u64 root_objectid
, u64 owner
, u64 offset
)
1037 u32 high_crc
= ~(u32
)0;
1038 u32 low_crc
= ~(u32
)0;
1041 lenum
= cpu_to_le64(root_objectid
);
1042 high_crc
= crc32c(high_crc
, &lenum
, sizeof(lenum
));
1043 lenum
= cpu_to_le64(owner
);
1044 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
1045 lenum
= cpu_to_le64(offset
);
1046 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
1048 return ((u64
)high_crc
<< 31) ^ (u64
)low_crc
;
1051 static u64
hash_extent_data_ref_item(struct extent_buffer
*leaf
,
1052 struct btrfs_extent_data_ref
*ref
)
1054 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf
, ref
),
1055 btrfs_extent_data_ref_objectid(leaf
, ref
),
1056 btrfs_extent_data_ref_offset(leaf
, ref
));
1059 static int match_extent_data_ref(struct extent_buffer
*leaf
,
1060 struct btrfs_extent_data_ref
*ref
,
1061 u64 root_objectid
, u64 owner
, u64 offset
)
1063 if (btrfs_extent_data_ref_root(leaf
, ref
) != root_objectid
||
1064 btrfs_extent_data_ref_objectid(leaf
, ref
) != owner
||
1065 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
1070 static noinline
int lookup_extent_data_ref(struct btrfs_trans_handle
*trans
,
1071 struct btrfs_root
*root
,
1072 struct btrfs_path
*path
,
1073 u64 bytenr
, u64 parent
,
1075 u64 owner
, u64 offset
)
1077 struct btrfs_key key
;
1078 struct btrfs_extent_data_ref
*ref
;
1079 struct extent_buffer
*leaf
;
1085 key
.objectid
= bytenr
;
1087 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1088 key
.offset
= parent
;
1090 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1091 key
.offset
= hash_extent_data_ref(root_objectid
,
1096 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1105 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1106 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1107 btrfs_release_path(path
);
1108 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1119 leaf
= path
->nodes
[0];
1120 nritems
= btrfs_header_nritems(leaf
);
1122 if (path
->slots
[0] >= nritems
) {
1123 ret
= btrfs_next_leaf(root
, path
);
1129 leaf
= path
->nodes
[0];
1130 nritems
= btrfs_header_nritems(leaf
);
1134 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1135 if (key
.objectid
!= bytenr
||
1136 key
.type
!= BTRFS_EXTENT_DATA_REF_KEY
)
1139 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1140 struct btrfs_extent_data_ref
);
1142 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1145 btrfs_release_path(path
);
1157 static noinline
int insert_extent_data_ref(struct btrfs_trans_handle
*trans
,
1158 struct btrfs_root
*root
,
1159 struct btrfs_path
*path
,
1160 u64 bytenr
, u64 parent
,
1161 u64 root_objectid
, u64 owner
,
1162 u64 offset
, int refs_to_add
)
1164 struct btrfs_key key
;
1165 struct extent_buffer
*leaf
;
1170 key
.objectid
= bytenr
;
1172 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1173 key
.offset
= parent
;
1174 size
= sizeof(struct btrfs_shared_data_ref
);
1176 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1177 key
.offset
= hash_extent_data_ref(root_objectid
,
1179 size
= sizeof(struct btrfs_extent_data_ref
);
1182 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, size
);
1183 if (ret
&& ret
!= -EEXIST
)
1186 leaf
= path
->nodes
[0];
1188 struct btrfs_shared_data_ref
*ref
;
1189 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1190 struct btrfs_shared_data_ref
);
1192 btrfs_set_shared_data_ref_count(leaf
, ref
, refs_to_add
);
1194 num_refs
= btrfs_shared_data_ref_count(leaf
, ref
);
1195 num_refs
+= refs_to_add
;
1196 btrfs_set_shared_data_ref_count(leaf
, ref
, num_refs
);
1199 struct btrfs_extent_data_ref
*ref
;
1200 while (ret
== -EEXIST
) {
1201 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1202 struct btrfs_extent_data_ref
);
1203 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1206 btrfs_release_path(path
);
1208 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1210 if (ret
&& ret
!= -EEXIST
)
1213 leaf
= path
->nodes
[0];
1215 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1216 struct btrfs_extent_data_ref
);
1218 btrfs_set_extent_data_ref_root(leaf
, ref
,
1220 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
1221 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
1222 btrfs_set_extent_data_ref_count(leaf
, ref
, refs_to_add
);
1224 num_refs
= btrfs_extent_data_ref_count(leaf
, ref
);
1225 num_refs
+= refs_to_add
;
1226 btrfs_set_extent_data_ref_count(leaf
, ref
, num_refs
);
1229 btrfs_mark_buffer_dirty(leaf
);
1232 btrfs_release_path(path
);
1236 static noinline
int remove_extent_data_ref(struct btrfs_trans_handle
*trans
,
1237 struct btrfs_root
*root
,
1238 struct btrfs_path
*path
,
1241 struct btrfs_key key
;
1242 struct btrfs_extent_data_ref
*ref1
= NULL
;
1243 struct btrfs_shared_data_ref
*ref2
= NULL
;
1244 struct extent_buffer
*leaf
;
1248 leaf
= path
->nodes
[0];
1249 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1251 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1252 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1253 struct btrfs_extent_data_ref
);
1254 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1255 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1256 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1257 struct btrfs_shared_data_ref
);
1258 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1259 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1260 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1261 struct btrfs_extent_ref_v0
*ref0
;
1262 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1263 struct btrfs_extent_ref_v0
);
1264 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1270 BUG_ON(num_refs
< refs_to_drop
);
1271 num_refs
-= refs_to_drop
;
1273 if (num_refs
== 0) {
1274 ret
= btrfs_del_item(trans
, root
, path
);
1276 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
)
1277 btrfs_set_extent_data_ref_count(leaf
, ref1
, num_refs
);
1278 else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
)
1279 btrfs_set_shared_data_ref_count(leaf
, ref2
, num_refs
);
1280 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1282 struct btrfs_extent_ref_v0
*ref0
;
1283 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1284 struct btrfs_extent_ref_v0
);
1285 btrfs_set_ref_count_v0(leaf
, ref0
, num_refs
);
1288 btrfs_mark_buffer_dirty(leaf
);
1293 static noinline u32
extent_data_ref_count(struct btrfs_root
*root
,
1294 struct btrfs_path
*path
,
1295 struct btrfs_extent_inline_ref
*iref
)
1297 struct btrfs_key key
;
1298 struct extent_buffer
*leaf
;
1299 struct btrfs_extent_data_ref
*ref1
;
1300 struct btrfs_shared_data_ref
*ref2
;
1303 leaf
= path
->nodes
[0];
1304 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1306 if (btrfs_extent_inline_ref_type(leaf
, iref
) ==
1307 BTRFS_EXTENT_DATA_REF_KEY
) {
1308 ref1
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1309 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1311 ref2
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1312 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1314 } else if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1315 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1316 struct btrfs_extent_data_ref
);
1317 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1318 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1319 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1320 struct btrfs_shared_data_ref
);
1321 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1322 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1323 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1324 struct btrfs_extent_ref_v0
*ref0
;
1325 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1326 struct btrfs_extent_ref_v0
);
1327 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1335 static noinline
int lookup_tree_block_ref(struct btrfs_trans_handle
*trans
,
1336 struct btrfs_root
*root
,
1337 struct btrfs_path
*path
,
1338 u64 bytenr
, u64 parent
,
1341 struct btrfs_key key
;
1344 key
.objectid
= bytenr
;
1346 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1347 key
.offset
= parent
;
1349 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1350 key
.offset
= root_objectid
;
1353 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1356 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1357 if (ret
== -ENOENT
&& parent
) {
1358 btrfs_release_path(path
);
1359 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1360 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1368 static noinline
int insert_tree_block_ref(struct btrfs_trans_handle
*trans
,
1369 struct btrfs_root
*root
,
1370 struct btrfs_path
*path
,
1371 u64 bytenr
, u64 parent
,
1374 struct btrfs_key key
;
1377 key
.objectid
= bytenr
;
1379 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1380 key
.offset
= parent
;
1382 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1383 key
.offset
= root_objectid
;
1386 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
1387 btrfs_release_path(path
);
1391 static inline int extent_ref_type(u64 parent
, u64 owner
)
1394 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1396 type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1398 type
= BTRFS_TREE_BLOCK_REF_KEY
;
1401 type
= BTRFS_SHARED_DATA_REF_KEY
;
1403 type
= BTRFS_EXTENT_DATA_REF_KEY
;
1408 static int find_next_key(struct btrfs_path
*path
, int level
,
1409 struct btrfs_key
*key
)
1412 for (; level
< BTRFS_MAX_LEVEL
; level
++) {
1413 if (!path
->nodes
[level
])
1415 if (path
->slots
[level
] + 1 >=
1416 btrfs_header_nritems(path
->nodes
[level
]))
1419 btrfs_item_key_to_cpu(path
->nodes
[level
], key
,
1420 path
->slots
[level
] + 1);
1422 btrfs_node_key_to_cpu(path
->nodes
[level
], key
,
1423 path
->slots
[level
] + 1);
1430 * look for inline back ref. if back ref is found, *ref_ret is set
1431 * to the address of inline back ref, and 0 is returned.
1433 * if back ref isn't found, *ref_ret is set to the address where it
1434 * should be inserted, and -ENOENT is returned.
1436 * if insert is true and there are too many inline back refs, the path
1437 * points to the extent item, and -EAGAIN is returned.
1439 * NOTE: inline back refs are ordered in the same way that back ref
1440 * items in the tree are ordered.
1442 static noinline_for_stack
1443 int lookup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1444 struct btrfs_root
*root
,
1445 struct btrfs_path
*path
,
1446 struct btrfs_extent_inline_ref
**ref_ret
,
1447 u64 bytenr
, u64 num_bytes
,
1448 u64 parent
, u64 root_objectid
,
1449 u64 owner
, u64 offset
, int insert
)
1451 struct btrfs_key key
;
1452 struct extent_buffer
*leaf
;
1453 struct btrfs_extent_item
*ei
;
1454 struct btrfs_extent_inline_ref
*iref
;
1465 key
.objectid
= bytenr
;
1466 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1467 key
.offset
= num_bytes
;
1469 want
= extent_ref_type(parent
, owner
);
1471 extra_size
= btrfs_extent_inline_ref_size(want
);
1472 path
->keep_locks
= 1;
1475 ret
= btrfs_search_slot(trans
, root
, &key
, path
, extra_size
, 1);
1482 leaf
= path
->nodes
[0];
1483 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1484 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1485 if (item_size
< sizeof(*ei
)) {
1490 ret
= convert_extent_item_v0(trans
, root
, path
, owner
,
1496 leaf
= path
->nodes
[0];
1497 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1500 BUG_ON(item_size
< sizeof(*ei
));
1502 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1503 flags
= btrfs_extent_flags(leaf
, ei
);
1505 ptr
= (unsigned long)(ei
+ 1);
1506 end
= (unsigned long)ei
+ item_size
;
1508 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
) {
1509 ptr
+= sizeof(struct btrfs_tree_block_info
);
1512 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_DATA
));
1521 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1522 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1526 ptr
+= btrfs_extent_inline_ref_size(type
);
1530 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1531 struct btrfs_extent_data_ref
*dref
;
1532 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1533 if (match_extent_data_ref(leaf
, dref
, root_objectid
,
1538 if (hash_extent_data_ref_item(leaf
, dref
) <
1539 hash_extent_data_ref(root_objectid
, owner
, offset
))
1543 ref_offset
= btrfs_extent_inline_ref_offset(leaf
, iref
);
1545 if (parent
== ref_offset
) {
1549 if (ref_offset
< parent
)
1552 if (root_objectid
== ref_offset
) {
1556 if (ref_offset
< root_objectid
)
1560 ptr
+= btrfs_extent_inline_ref_size(type
);
1562 if (err
== -ENOENT
&& insert
) {
1563 if (item_size
+ extra_size
>=
1564 BTRFS_MAX_EXTENT_ITEM_SIZE(root
)) {
1569 * To add new inline back ref, we have to make sure
1570 * there is no corresponding back ref item.
1571 * For simplicity, we just do not add new inline back
1572 * ref if there is any kind of item for this block
1574 if (find_next_key(path
, 0, &key
) == 0 &&
1575 key
.objectid
== bytenr
&&
1576 key
.type
< BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1581 *ref_ret
= (struct btrfs_extent_inline_ref
*)ptr
;
1584 path
->keep_locks
= 0;
1585 btrfs_unlock_up_safe(path
, 1);
1591 * helper to add new inline back ref
1593 static noinline_for_stack
1594 int setup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1595 struct btrfs_root
*root
,
1596 struct btrfs_path
*path
,
1597 struct btrfs_extent_inline_ref
*iref
,
1598 u64 parent
, u64 root_objectid
,
1599 u64 owner
, u64 offset
, int refs_to_add
,
1600 struct btrfs_delayed_extent_op
*extent_op
)
1602 struct extent_buffer
*leaf
;
1603 struct btrfs_extent_item
*ei
;
1606 unsigned long item_offset
;
1612 leaf
= path
->nodes
[0];
1613 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1614 item_offset
= (unsigned long)iref
- (unsigned long)ei
;
1616 type
= extent_ref_type(parent
, owner
);
1617 size
= btrfs_extent_inline_ref_size(type
);
1619 ret
= btrfs_extend_item(trans
, root
, path
, size
);
1621 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1622 refs
= btrfs_extent_refs(leaf
, ei
);
1623 refs
+= refs_to_add
;
1624 btrfs_set_extent_refs(leaf
, ei
, refs
);
1626 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1628 ptr
= (unsigned long)ei
+ item_offset
;
1629 end
= (unsigned long)ei
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
1630 if (ptr
< end
- size
)
1631 memmove_extent_buffer(leaf
, ptr
+ size
, ptr
,
1634 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1635 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
1636 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1637 struct btrfs_extent_data_ref
*dref
;
1638 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1639 btrfs_set_extent_data_ref_root(leaf
, dref
, root_objectid
);
1640 btrfs_set_extent_data_ref_objectid(leaf
, dref
, owner
);
1641 btrfs_set_extent_data_ref_offset(leaf
, dref
, offset
);
1642 btrfs_set_extent_data_ref_count(leaf
, dref
, refs_to_add
);
1643 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1644 struct btrfs_shared_data_ref
*sref
;
1645 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1646 btrfs_set_shared_data_ref_count(leaf
, sref
, refs_to_add
);
1647 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1648 } else if (type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
1649 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1651 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
1653 btrfs_mark_buffer_dirty(leaf
);
1657 static int lookup_extent_backref(struct btrfs_trans_handle
*trans
,
1658 struct btrfs_root
*root
,
1659 struct btrfs_path
*path
,
1660 struct btrfs_extent_inline_ref
**ref_ret
,
1661 u64 bytenr
, u64 num_bytes
, u64 parent
,
1662 u64 root_objectid
, u64 owner
, u64 offset
)
1666 ret
= lookup_inline_extent_backref(trans
, root
, path
, ref_ret
,
1667 bytenr
, num_bytes
, parent
,
1668 root_objectid
, owner
, offset
, 0);
1672 btrfs_release_path(path
);
1675 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1676 ret
= lookup_tree_block_ref(trans
, root
, path
, bytenr
, parent
,
1679 ret
= lookup_extent_data_ref(trans
, root
, path
, bytenr
, parent
,
1680 root_objectid
, owner
, offset
);
1686 * helper to update/remove inline back ref
1688 static noinline_for_stack
1689 int update_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1690 struct btrfs_root
*root
,
1691 struct btrfs_path
*path
,
1692 struct btrfs_extent_inline_ref
*iref
,
1694 struct btrfs_delayed_extent_op
*extent_op
)
1696 struct extent_buffer
*leaf
;
1697 struct btrfs_extent_item
*ei
;
1698 struct btrfs_extent_data_ref
*dref
= NULL
;
1699 struct btrfs_shared_data_ref
*sref
= NULL
;
1708 leaf
= path
->nodes
[0];
1709 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1710 refs
= btrfs_extent_refs(leaf
, ei
);
1711 WARN_ON(refs_to_mod
< 0 && refs
+ refs_to_mod
<= 0);
1712 refs
+= refs_to_mod
;
1713 btrfs_set_extent_refs(leaf
, ei
, refs
);
1715 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1717 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1719 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1720 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1721 refs
= btrfs_extent_data_ref_count(leaf
, dref
);
1722 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1723 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1724 refs
= btrfs_shared_data_ref_count(leaf
, sref
);
1727 BUG_ON(refs_to_mod
!= -1);
1730 BUG_ON(refs_to_mod
< 0 && refs
< -refs_to_mod
);
1731 refs
+= refs_to_mod
;
1734 if (type
== BTRFS_EXTENT_DATA_REF_KEY
)
1735 btrfs_set_extent_data_ref_count(leaf
, dref
, refs
);
1737 btrfs_set_shared_data_ref_count(leaf
, sref
, refs
);
1739 size
= btrfs_extent_inline_ref_size(type
);
1740 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1741 ptr
= (unsigned long)iref
;
1742 end
= (unsigned long)ei
+ item_size
;
1743 if (ptr
+ size
< end
)
1744 memmove_extent_buffer(leaf
, ptr
, ptr
+ size
,
1747 ret
= btrfs_truncate_item(trans
, root
, path
, item_size
, 1);
1749 btrfs_mark_buffer_dirty(leaf
);
1753 static noinline_for_stack
1754 int insert_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1755 struct btrfs_root
*root
,
1756 struct btrfs_path
*path
,
1757 u64 bytenr
, u64 num_bytes
, u64 parent
,
1758 u64 root_objectid
, u64 owner
,
1759 u64 offset
, int refs_to_add
,
1760 struct btrfs_delayed_extent_op
*extent_op
)
1762 struct btrfs_extent_inline_ref
*iref
;
1765 ret
= lookup_inline_extent_backref(trans
, root
, path
, &iref
,
1766 bytenr
, num_bytes
, parent
,
1767 root_objectid
, owner
, offset
, 1);
1769 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
);
1770 ret
= update_inline_extent_backref(trans
, root
, path
, iref
,
1771 refs_to_add
, extent_op
);
1772 } else if (ret
== -ENOENT
) {
1773 ret
= setup_inline_extent_backref(trans
, root
, path
, iref
,
1774 parent
, root_objectid
,
1775 owner
, offset
, refs_to_add
,
1781 static int insert_extent_backref(struct btrfs_trans_handle
*trans
,
1782 struct btrfs_root
*root
,
1783 struct btrfs_path
*path
,
1784 u64 bytenr
, u64 parent
, u64 root_objectid
,
1785 u64 owner
, u64 offset
, int refs_to_add
)
1788 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1789 BUG_ON(refs_to_add
!= 1);
1790 ret
= insert_tree_block_ref(trans
, root
, path
, bytenr
,
1791 parent
, root_objectid
);
1793 ret
= insert_extent_data_ref(trans
, root
, path
, bytenr
,
1794 parent
, root_objectid
,
1795 owner
, offset
, refs_to_add
);
1800 static int remove_extent_backref(struct btrfs_trans_handle
*trans
,
1801 struct btrfs_root
*root
,
1802 struct btrfs_path
*path
,
1803 struct btrfs_extent_inline_ref
*iref
,
1804 int refs_to_drop
, int is_data
)
1808 BUG_ON(!is_data
&& refs_to_drop
!= 1);
1810 ret
= update_inline_extent_backref(trans
, root
, path
, iref
,
1811 -refs_to_drop
, NULL
);
1812 } else if (is_data
) {
1813 ret
= remove_extent_data_ref(trans
, root
, path
, refs_to_drop
);
1815 ret
= btrfs_del_item(trans
, root
, path
);
1820 static int btrfs_issue_discard(struct block_device
*bdev
,
1823 return blkdev_issue_discard(bdev
, start
>> 9, len
>> 9, GFP_NOFS
, 0);
1826 static int btrfs_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
1827 u64 num_bytes
, u64
*actual_bytes
)
1830 u64 discarded_bytes
= 0;
1831 struct btrfs_bio
*bbio
= NULL
;
1834 /* Tell the block device(s) that the sectors can be discarded */
1835 ret
= btrfs_map_block(&root
->fs_info
->mapping_tree
, REQ_DISCARD
,
1836 bytenr
, &num_bytes
, &bbio
, 0);
1838 struct btrfs_bio_stripe
*stripe
= bbio
->stripes
;
1842 for (i
= 0; i
< bbio
->num_stripes
; i
++, stripe
++) {
1843 if (!stripe
->dev
->can_discard
)
1846 ret
= btrfs_issue_discard(stripe
->dev
->bdev
,
1850 discarded_bytes
+= stripe
->length
;
1851 else if (ret
!= -EOPNOTSUPP
)
1855 * Just in case we get back EOPNOTSUPP for some reason,
1856 * just ignore the return value so we don't screw up
1857 * people calling discard_extent.
1865 *actual_bytes
= discarded_bytes
;
1871 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1872 struct btrfs_root
*root
,
1873 u64 bytenr
, u64 num_bytes
, u64 parent
,
1874 u64 root_objectid
, u64 owner
, u64 offset
, int for_cow
)
1877 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1879 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
&&
1880 root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
1882 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1883 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
1885 parent
, root_objectid
, (int)owner
,
1886 BTRFS_ADD_DELAYED_REF
, NULL
, for_cow
);
1888 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
1890 parent
, root_objectid
, owner
, offset
,
1891 BTRFS_ADD_DELAYED_REF
, NULL
, for_cow
);
1896 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1897 struct btrfs_root
*root
,
1898 u64 bytenr
, u64 num_bytes
,
1899 u64 parent
, u64 root_objectid
,
1900 u64 owner
, u64 offset
, int refs_to_add
,
1901 struct btrfs_delayed_extent_op
*extent_op
)
1903 struct btrfs_path
*path
;
1904 struct extent_buffer
*leaf
;
1905 struct btrfs_extent_item
*item
;
1910 path
= btrfs_alloc_path();
1915 path
->leave_spinning
= 1;
1916 /* this will setup the path even if it fails to insert the back ref */
1917 ret
= insert_inline_extent_backref(trans
, root
->fs_info
->extent_root
,
1918 path
, bytenr
, num_bytes
, parent
,
1919 root_objectid
, owner
, offset
,
1920 refs_to_add
, extent_op
);
1924 if (ret
!= -EAGAIN
) {
1929 leaf
= path
->nodes
[0];
1930 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1931 refs
= btrfs_extent_refs(leaf
, item
);
1932 btrfs_set_extent_refs(leaf
, item
, refs
+ refs_to_add
);
1934 __run_delayed_extent_op(extent_op
, leaf
, item
);
1936 btrfs_mark_buffer_dirty(leaf
);
1937 btrfs_release_path(path
);
1940 path
->leave_spinning
= 1;
1942 /* now insert the actual backref */
1943 ret
= insert_extent_backref(trans
, root
->fs_info
->extent_root
,
1944 path
, bytenr
, parent
, root_objectid
,
1945 owner
, offset
, refs_to_add
);
1948 btrfs_free_path(path
);
1952 static int run_delayed_data_ref(struct btrfs_trans_handle
*trans
,
1953 struct btrfs_root
*root
,
1954 struct btrfs_delayed_ref_node
*node
,
1955 struct btrfs_delayed_extent_op
*extent_op
,
1956 int insert_reserved
)
1959 struct btrfs_delayed_data_ref
*ref
;
1960 struct btrfs_key ins
;
1965 ins
.objectid
= node
->bytenr
;
1966 ins
.offset
= node
->num_bytes
;
1967 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1969 ref
= btrfs_delayed_node_to_data_ref(node
);
1970 if (node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
1971 parent
= ref
->parent
;
1973 ref_root
= ref
->root
;
1975 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
1977 BUG_ON(extent_op
->update_key
);
1978 flags
|= extent_op
->flags_to_set
;
1980 ret
= alloc_reserved_file_extent(trans
, root
,
1981 parent
, ref_root
, flags
,
1982 ref
->objectid
, ref
->offset
,
1983 &ins
, node
->ref_mod
);
1984 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
1985 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
1986 node
->num_bytes
, parent
,
1987 ref_root
, ref
->objectid
,
1988 ref
->offset
, node
->ref_mod
,
1990 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
1991 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
1992 node
->num_bytes
, parent
,
1993 ref_root
, ref
->objectid
,
1994 ref
->offset
, node
->ref_mod
,
2002 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
2003 struct extent_buffer
*leaf
,
2004 struct btrfs_extent_item
*ei
)
2006 u64 flags
= btrfs_extent_flags(leaf
, ei
);
2007 if (extent_op
->update_flags
) {
2008 flags
|= extent_op
->flags_to_set
;
2009 btrfs_set_extent_flags(leaf
, ei
, flags
);
2012 if (extent_op
->update_key
) {
2013 struct btrfs_tree_block_info
*bi
;
2014 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
));
2015 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
2016 btrfs_set_tree_block_key(leaf
, bi
, &extent_op
->key
);
2020 static int run_delayed_extent_op(struct btrfs_trans_handle
*trans
,
2021 struct btrfs_root
*root
,
2022 struct btrfs_delayed_ref_node
*node
,
2023 struct btrfs_delayed_extent_op
*extent_op
)
2025 struct btrfs_key key
;
2026 struct btrfs_path
*path
;
2027 struct btrfs_extent_item
*ei
;
2028 struct extent_buffer
*leaf
;
2033 path
= btrfs_alloc_path();
2037 key
.objectid
= node
->bytenr
;
2038 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2039 key
.offset
= node
->num_bytes
;
2042 path
->leave_spinning
= 1;
2043 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
,
2054 leaf
= path
->nodes
[0];
2055 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2056 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2057 if (item_size
< sizeof(*ei
)) {
2058 ret
= convert_extent_item_v0(trans
, root
->fs_info
->extent_root
,
2064 leaf
= path
->nodes
[0];
2065 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2068 BUG_ON(item_size
< sizeof(*ei
));
2069 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2070 __run_delayed_extent_op(extent_op
, leaf
, ei
);
2072 btrfs_mark_buffer_dirty(leaf
);
2074 btrfs_free_path(path
);
2078 static int run_delayed_tree_ref(struct btrfs_trans_handle
*trans
,
2079 struct btrfs_root
*root
,
2080 struct btrfs_delayed_ref_node
*node
,
2081 struct btrfs_delayed_extent_op
*extent_op
,
2082 int insert_reserved
)
2085 struct btrfs_delayed_tree_ref
*ref
;
2086 struct btrfs_key ins
;
2090 ins
.objectid
= node
->bytenr
;
2091 ins
.offset
= node
->num_bytes
;
2092 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2094 ref
= btrfs_delayed_node_to_tree_ref(node
);
2095 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2096 parent
= ref
->parent
;
2098 ref_root
= ref
->root
;
2100 BUG_ON(node
->ref_mod
!= 1);
2101 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2102 BUG_ON(!extent_op
|| !extent_op
->update_flags
||
2103 !extent_op
->update_key
);
2104 ret
= alloc_reserved_tree_block(trans
, root
,
2106 extent_op
->flags_to_set
,
2109 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2110 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
2111 node
->num_bytes
, parent
, ref_root
,
2112 ref
->level
, 0, 1, extent_op
);
2113 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2114 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
2115 node
->num_bytes
, parent
, ref_root
,
2116 ref
->level
, 0, 1, extent_op
);
2123 /* helper function to actually process a single delayed ref entry */
2124 static int run_one_delayed_ref(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
,
2128 int insert_reserved
)
2131 if (btrfs_delayed_ref_is_head(node
)) {
2132 struct btrfs_delayed_ref_head
*head
;
2134 * we've hit the end of the chain and we were supposed
2135 * to insert this extent into the tree. But, it got
2136 * deleted before we ever needed to insert it, so all
2137 * we have to do is clean up the accounting
2140 head
= btrfs_delayed_node_to_head(node
);
2141 if (insert_reserved
) {
2142 btrfs_pin_extent(root
, node
->bytenr
,
2143 node
->num_bytes
, 1);
2144 if (head
->is_data
) {
2145 ret
= btrfs_del_csums(trans
, root
,
2151 mutex_unlock(&head
->mutex
);
2155 if (node
->type
== BTRFS_TREE_BLOCK_REF_KEY
||
2156 node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2157 ret
= run_delayed_tree_ref(trans
, root
, node
, extent_op
,
2159 else if (node
->type
== BTRFS_EXTENT_DATA_REF_KEY
||
2160 node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2161 ret
= run_delayed_data_ref(trans
, root
, node
, extent_op
,
2168 static noinline
struct btrfs_delayed_ref_node
*
2169 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
2171 struct rb_node
*node
;
2172 struct btrfs_delayed_ref_node
*ref
;
2173 int action
= BTRFS_ADD_DELAYED_REF
;
2176 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2177 * this prevents ref count from going down to zero when
2178 * there still are pending delayed ref.
2180 node
= rb_prev(&head
->node
.rb_node
);
2184 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
2186 if (ref
->bytenr
!= head
->node
.bytenr
)
2188 if (ref
->action
== action
)
2190 node
= rb_prev(node
);
2192 if (action
== BTRFS_ADD_DELAYED_REF
) {
2193 action
= BTRFS_DROP_DELAYED_REF
;
2199 static noinline
int run_clustered_refs(struct btrfs_trans_handle
*trans
,
2200 struct btrfs_root
*root
,
2201 struct list_head
*cluster
)
2203 struct btrfs_delayed_ref_root
*delayed_refs
;
2204 struct btrfs_delayed_ref_node
*ref
;
2205 struct btrfs_delayed_ref_head
*locked_ref
= NULL
;
2206 struct btrfs_delayed_extent_op
*extent_op
;
2209 int must_insert_reserved
= 0;
2211 delayed_refs
= &trans
->transaction
->delayed_refs
;
2214 /* pick a new head ref from the cluster list */
2215 if (list_empty(cluster
))
2218 locked_ref
= list_entry(cluster
->next
,
2219 struct btrfs_delayed_ref_head
, cluster
);
2221 /* grab the lock that says we are going to process
2222 * all the refs for this head */
2223 ret
= btrfs_delayed_ref_lock(trans
, locked_ref
);
2226 * we may have dropped the spin lock to get the head
2227 * mutex lock, and that might have given someone else
2228 * time to free the head. If that's true, it has been
2229 * removed from our list and we can move on.
2231 if (ret
== -EAGAIN
) {
2239 * locked_ref is the head node, so we have to go one
2240 * node back for any delayed ref updates
2242 ref
= select_delayed_ref(locked_ref
);
2244 if (ref
&& ref
->seq
&&
2245 btrfs_check_delayed_seq(delayed_refs
, ref
->seq
)) {
2247 * there are still refs with lower seq numbers in the
2248 * process of being added. Don't run this ref yet.
2250 list_del_init(&locked_ref
->cluster
);
2251 mutex_unlock(&locked_ref
->mutex
);
2253 delayed_refs
->num_heads_ready
++;
2254 spin_unlock(&delayed_refs
->lock
);
2256 spin_lock(&delayed_refs
->lock
);
2261 * record the must insert reserved flag before we
2262 * drop the spin lock.
2264 must_insert_reserved
= locked_ref
->must_insert_reserved
;
2265 locked_ref
->must_insert_reserved
= 0;
2267 extent_op
= locked_ref
->extent_op
;
2268 locked_ref
->extent_op
= NULL
;
2271 /* All delayed refs have been processed, Go ahead
2272 * and send the head node to run_one_delayed_ref,
2273 * so that any accounting fixes can happen
2275 ref
= &locked_ref
->node
;
2277 if (extent_op
&& must_insert_reserved
) {
2283 spin_unlock(&delayed_refs
->lock
);
2285 ret
= run_delayed_extent_op(trans
, root
,
2293 list_del_init(&locked_ref
->cluster
);
2298 rb_erase(&ref
->rb_node
, &delayed_refs
->root
);
2299 delayed_refs
->num_entries
--;
2301 * we modified num_entries, but as we're currently running
2302 * delayed refs, skip
2303 * wake_up(&delayed_refs->seq_wait);
2306 spin_unlock(&delayed_refs
->lock
);
2308 ret
= run_one_delayed_ref(trans
, root
, ref
, extent_op
,
2309 must_insert_reserved
);
2312 btrfs_put_delayed_ref(ref
);
2316 do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
2318 btrfs_get_alloc_profile(root
, 0),
2319 CHUNK_ALLOC_NO_FORCE
);
2321 spin_lock(&delayed_refs
->lock
);
2327 static void wait_for_more_refs(struct btrfs_delayed_ref_root
*delayed_refs
,
2328 unsigned long num_refs
)
2330 struct list_head
*first_seq
= delayed_refs
->seq_head
.next
;
2332 spin_unlock(&delayed_refs
->lock
);
2333 pr_debug("waiting for more refs (num %ld, first %p)\n",
2334 num_refs
, first_seq
);
2335 wait_event(delayed_refs
->seq_wait
,
2336 num_refs
!= delayed_refs
->num_entries
||
2337 delayed_refs
->seq_head
.next
!= first_seq
);
2338 pr_debug("done waiting for more refs (num %ld, first %p)\n",
2339 delayed_refs
->num_entries
, delayed_refs
->seq_head
.next
);
2340 spin_lock(&delayed_refs
->lock
);
2344 * this starts processing the delayed reference count updates and
2345 * extent insertions we have queued up so far. count can be
2346 * 0, which means to process everything in the tree at the start
2347 * of the run (but not newly added entries), or it can be some target
2348 * number you'd like to process.
2350 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2351 struct btrfs_root
*root
, unsigned long count
)
2353 struct rb_node
*node
;
2354 struct btrfs_delayed_ref_root
*delayed_refs
;
2355 struct btrfs_delayed_ref_node
*ref
;
2356 struct list_head cluster
;
2359 int run_all
= count
== (unsigned long)-1;
2361 unsigned long num_refs
= 0;
2362 int consider_waiting
;
2364 if (root
== root
->fs_info
->extent_root
)
2365 root
= root
->fs_info
->tree_root
;
2367 do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
2368 2 * 1024 * 1024, btrfs_get_alloc_profile(root
, 0),
2369 CHUNK_ALLOC_NO_FORCE
);
2371 delayed_refs
= &trans
->transaction
->delayed_refs
;
2372 INIT_LIST_HEAD(&cluster
);
2374 consider_waiting
= 0;
2375 spin_lock(&delayed_refs
->lock
);
2377 count
= delayed_refs
->num_entries
* 2;
2381 if (!(run_all
|| run_most
) &&
2382 delayed_refs
->num_heads_ready
< 64)
2386 * go find something we can process in the rbtree. We start at
2387 * the beginning of the tree, and then build a cluster
2388 * of refs to process starting at the first one we are able to
2391 delayed_start
= delayed_refs
->run_delayed_start
;
2392 ret
= btrfs_find_ref_cluster(trans
, &cluster
,
2393 delayed_refs
->run_delayed_start
);
2397 if (delayed_start
>= delayed_refs
->run_delayed_start
) {
2398 if (consider_waiting
== 0) {
2400 * btrfs_find_ref_cluster looped. let's do one
2401 * more cycle. if we don't run any delayed ref
2402 * during that cycle (because we can't because
2403 * all of them are blocked) and if the number of
2404 * refs doesn't change, we avoid busy waiting.
2406 consider_waiting
= 1;
2407 num_refs
= delayed_refs
->num_entries
;
2409 wait_for_more_refs(delayed_refs
, num_refs
);
2411 * after waiting, things have changed. we
2412 * dropped the lock and someone else might have
2413 * run some refs, built new clusters and so on.
2414 * therefore, we restart staleness detection.
2416 consider_waiting
= 0;
2420 ret
= run_clustered_refs(trans
, root
, &cluster
);
2423 count
-= min_t(unsigned long, ret
, count
);
2428 if (ret
|| delayed_refs
->run_delayed_start
== 0) {
2429 /* refs were run, let's reset staleness detection */
2430 consider_waiting
= 0;
2435 node
= rb_first(&delayed_refs
->root
);
2438 count
= (unsigned long)-1;
2441 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
2443 if (btrfs_delayed_ref_is_head(ref
)) {
2444 struct btrfs_delayed_ref_head
*head
;
2446 head
= btrfs_delayed_node_to_head(ref
);
2447 atomic_inc(&ref
->refs
);
2449 spin_unlock(&delayed_refs
->lock
);
2451 * Mutex was contended, block until it's
2452 * released and try again
2454 mutex_lock(&head
->mutex
);
2455 mutex_unlock(&head
->mutex
);
2457 btrfs_put_delayed_ref(ref
);
2461 node
= rb_next(node
);
2463 spin_unlock(&delayed_refs
->lock
);
2464 schedule_timeout(1);
2468 spin_unlock(&delayed_refs
->lock
);
2472 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
2473 struct btrfs_root
*root
,
2474 u64 bytenr
, u64 num_bytes
, u64 flags
,
2477 struct btrfs_delayed_extent_op
*extent_op
;
2480 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
2484 extent_op
->flags_to_set
= flags
;
2485 extent_op
->update_flags
= 1;
2486 extent_op
->update_key
= 0;
2487 extent_op
->is_data
= is_data
? 1 : 0;
2489 ret
= btrfs_add_delayed_extent_op(root
->fs_info
, trans
, bytenr
,
2490 num_bytes
, extent_op
);
2496 static noinline
int check_delayed_ref(struct btrfs_trans_handle
*trans
,
2497 struct btrfs_root
*root
,
2498 struct btrfs_path
*path
,
2499 u64 objectid
, u64 offset
, u64 bytenr
)
2501 struct btrfs_delayed_ref_head
*head
;
2502 struct btrfs_delayed_ref_node
*ref
;
2503 struct btrfs_delayed_data_ref
*data_ref
;
2504 struct btrfs_delayed_ref_root
*delayed_refs
;
2505 struct rb_node
*node
;
2509 delayed_refs
= &trans
->transaction
->delayed_refs
;
2510 spin_lock(&delayed_refs
->lock
);
2511 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
2515 if (!mutex_trylock(&head
->mutex
)) {
2516 atomic_inc(&head
->node
.refs
);
2517 spin_unlock(&delayed_refs
->lock
);
2519 btrfs_release_path(path
);
2522 * Mutex was contended, block until it's released and let
2525 mutex_lock(&head
->mutex
);
2526 mutex_unlock(&head
->mutex
);
2527 btrfs_put_delayed_ref(&head
->node
);
2531 node
= rb_prev(&head
->node
.rb_node
);
2535 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2537 if (ref
->bytenr
!= bytenr
)
2541 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
)
2544 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
2546 node
= rb_prev(node
);
2548 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2549 if (ref
->bytenr
== bytenr
)
2553 if (data_ref
->root
!= root
->root_key
.objectid
||
2554 data_ref
->objectid
!= objectid
|| data_ref
->offset
!= offset
)
2559 mutex_unlock(&head
->mutex
);
2561 spin_unlock(&delayed_refs
->lock
);
2565 static noinline
int check_committed_ref(struct btrfs_trans_handle
*trans
,
2566 struct btrfs_root
*root
,
2567 struct btrfs_path
*path
,
2568 u64 objectid
, u64 offset
, u64 bytenr
)
2570 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2571 struct extent_buffer
*leaf
;
2572 struct btrfs_extent_data_ref
*ref
;
2573 struct btrfs_extent_inline_ref
*iref
;
2574 struct btrfs_extent_item
*ei
;
2575 struct btrfs_key key
;
2579 key
.objectid
= bytenr
;
2580 key
.offset
= (u64
)-1;
2581 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2583 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
2589 if (path
->slots
[0] == 0)
2593 leaf
= path
->nodes
[0];
2594 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2596 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
2600 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2601 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2602 if (item_size
< sizeof(*ei
)) {
2603 WARN_ON(item_size
!= sizeof(struct btrfs_extent_item_v0
));
2607 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2609 if (item_size
!= sizeof(*ei
) +
2610 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
2613 if (btrfs_extent_generation(leaf
, ei
) <=
2614 btrfs_root_last_snapshot(&root
->root_item
))
2617 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
2618 if (btrfs_extent_inline_ref_type(leaf
, iref
) !=
2619 BTRFS_EXTENT_DATA_REF_KEY
)
2622 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
2623 if (btrfs_extent_refs(leaf
, ei
) !=
2624 btrfs_extent_data_ref_count(leaf
, ref
) ||
2625 btrfs_extent_data_ref_root(leaf
, ref
) !=
2626 root
->root_key
.objectid
||
2627 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
2628 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
2636 int btrfs_cross_ref_exist(struct btrfs_trans_handle
*trans
,
2637 struct btrfs_root
*root
,
2638 u64 objectid
, u64 offset
, u64 bytenr
)
2640 struct btrfs_path
*path
;
2644 path
= btrfs_alloc_path();
2649 ret
= check_committed_ref(trans
, root
, path
, objectid
,
2651 if (ret
&& ret
!= -ENOENT
)
2654 ret2
= check_delayed_ref(trans
, root
, path
, objectid
,
2656 } while (ret2
== -EAGAIN
);
2658 if (ret2
&& ret2
!= -ENOENT
) {
2663 if (ret
!= -ENOENT
|| ret2
!= -ENOENT
)
2666 btrfs_free_path(path
);
2667 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
2672 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
2673 struct btrfs_root
*root
,
2674 struct extent_buffer
*buf
,
2675 int full_backref
, int inc
, int for_cow
)
2682 struct btrfs_key key
;
2683 struct btrfs_file_extent_item
*fi
;
2687 int (*process_func
)(struct btrfs_trans_handle
*, struct btrfs_root
*,
2688 u64
, u64
, u64
, u64
, u64
, u64
, int);
2690 ref_root
= btrfs_header_owner(buf
);
2691 nritems
= btrfs_header_nritems(buf
);
2692 level
= btrfs_header_level(buf
);
2694 if (!root
->ref_cows
&& level
== 0)
2698 process_func
= btrfs_inc_extent_ref
;
2700 process_func
= btrfs_free_extent
;
2703 parent
= buf
->start
;
2707 for (i
= 0; i
< nritems
; i
++) {
2709 btrfs_item_key_to_cpu(buf
, &key
, i
);
2710 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
2712 fi
= btrfs_item_ptr(buf
, i
,
2713 struct btrfs_file_extent_item
);
2714 if (btrfs_file_extent_type(buf
, fi
) ==
2715 BTRFS_FILE_EXTENT_INLINE
)
2717 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
2721 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
2722 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
2723 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
2724 parent
, ref_root
, key
.objectid
,
2725 key
.offset
, for_cow
);
2729 bytenr
= btrfs_node_blockptr(buf
, i
);
2730 num_bytes
= btrfs_level_size(root
, level
- 1);
2731 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
2732 parent
, ref_root
, level
- 1, 0,
2744 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2745 struct extent_buffer
*buf
, int full_backref
, int for_cow
)
2747 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1, for_cow
);
2750 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2751 struct extent_buffer
*buf
, int full_backref
, int for_cow
)
2753 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0, for_cow
);
2756 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
2757 struct btrfs_root
*root
,
2758 struct btrfs_path
*path
,
2759 struct btrfs_block_group_cache
*cache
)
2762 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2764 struct extent_buffer
*leaf
;
2766 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
2771 leaf
= path
->nodes
[0];
2772 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
2773 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
2774 btrfs_mark_buffer_dirty(leaf
);
2775 btrfs_release_path(path
);
2783 static struct btrfs_block_group_cache
*
2784 next_block_group(struct btrfs_root
*root
,
2785 struct btrfs_block_group_cache
*cache
)
2787 struct rb_node
*node
;
2788 spin_lock(&root
->fs_info
->block_group_cache_lock
);
2789 node
= rb_next(&cache
->cache_node
);
2790 btrfs_put_block_group(cache
);
2792 cache
= rb_entry(node
, struct btrfs_block_group_cache
,
2794 btrfs_get_block_group(cache
);
2797 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
2801 static int cache_save_setup(struct btrfs_block_group_cache
*block_group
,
2802 struct btrfs_trans_handle
*trans
,
2803 struct btrfs_path
*path
)
2805 struct btrfs_root
*root
= block_group
->fs_info
->tree_root
;
2806 struct inode
*inode
= NULL
;
2808 int dcs
= BTRFS_DC_ERROR
;
2814 * If this block group is smaller than 100 megs don't bother caching the
2817 if (block_group
->key
.offset
< (100 * 1024 * 1024)) {
2818 spin_lock(&block_group
->lock
);
2819 block_group
->disk_cache_state
= BTRFS_DC_WRITTEN
;
2820 spin_unlock(&block_group
->lock
);
2825 inode
= lookup_free_space_inode(root
, block_group
, path
);
2826 if (IS_ERR(inode
) && PTR_ERR(inode
) != -ENOENT
) {
2827 ret
= PTR_ERR(inode
);
2828 btrfs_release_path(path
);
2832 if (IS_ERR(inode
)) {
2836 if (block_group
->ro
)
2839 ret
= create_free_space_inode(root
, trans
, block_group
, path
);
2845 /* We've already setup this transaction, go ahead and exit */
2846 if (block_group
->cache_generation
== trans
->transid
&&
2847 i_size_read(inode
)) {
2848 dcs
= BTRFS_DC_SETUP
;
2853 * We want to set the generation to 0, that way if anything goes wrong
2854 * from here on out we know not to trust this cache when we load up next
2857 BTRFS_I(inode
)->generation
= 0;
2858 ret
= btrfs_update_inode(trans
, root
, inode
);
2861 if (i_size_read(inode
) > 0) {
2862 ret
= btrfs_truncate_free_space_cache(root
, trans
, path
,
2868 spin_lock(&block_group
->lock
);
2869 if (block_group
->cached
!= BTRFS_CACHE_FINISHED
) {
2870 /* We're not cached, don't bother trying to write stuff out */
2871 dcs
= BTRFS_DC_WRITTEN
;
2872 spin_unlock(&block_group
->lock
);
2875 spin_unlock(&block_group
->lock
);
2877 num_pages
= (int)div64_u64(block_group
->key
.offset
, 1024 * 1024 * 1024);
2882 * Just to make absolutely sure we have enough space, we're going to
2883 * preallocate 12 pages worth of space for each block group. In
2884 * practice we ought to use at most 8, but we need extra space so we can
2885 * add our header and have a terminator between the extents and the
2889 num_pages
*= PAGE_CACHE_SIZE
;
2891 ret
= btrfs_check_data_free_space(inode
, num_pages
);
2895 ret
= btrfs_prealloc_file_range_trans(inode
, trans
, 0, 0, num_pages
,
2896 num_pages
, num_pages
,
2899 dcs
= BTRFS_DC_SETUP
;
2900 btrfs_free_reserved_data_space(inode
, num_pages
);
2905 btrfs_release_path(path
);
2907 spin_lock(&block_group
->lock
);
2908 if (!ret
&& dcs
== BTRFS_DC_SETUP
)
2909 block_group
->cache_generation
= trans
->transid
;
2910 block_group
->disk_cache_state
= dcs
;
2911 spin_unlock(&block_group
->lock
);
2916 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
2917 struct btrfs_root
*root
)
2919 struct btrfs_block_group_cache
*cache
;
2921 struct btrfs_path
*path
;
2924 path
= btrfs_alloc_path();
2930 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
2932 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
)
2934 cache
= next_block_group(root
, cache
);
2942 err
= cache_save_setup(cache
, trans
, path
);
2943 last
= cache
->key
.objectid
+ cache
->key
.offset
;
2944 btrfs_put_block_group(cache
);
2949 err
= btrfs_run_delayed_refs(trans
, root
,
2954 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
2956 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
) {
2957 btrfs_put_block_group(cache
);
2963 cache
= next_block_group(root
, cache
);
2972 if (cache
->disk_cache_state
== BTRFS_DC_SETUP
)
2973 cache
->disk_cache_state
= BTRFS_DC_NEED_WRITE
;
2975 last
= cache
->key
.objectid
+ cache
->key
.offset
;
2977 err
= write_one_cache_group(trans
, root
, path
, cache
);
2979 btrfs_put_block_group(cache
);
2984 * I don't think this is needed since we're just marking our
2985 * preallocated extent as written, but just in case it can't
2989 err
= btrfs_run_delayed_refs(trans
, root
,
2994 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
2997 * Really this shouldn't happen, but it could if we
2998 * couldn't write the entire preallocated extent and
2999 * splitting the extent resulted in a new block.
3002 btrfs_put_block_group(cache
);
3005 if (cache
->disk_cache_state
== BTRFS_DC_NEED_WRITE
)
3007 cache
= next_block_group(root
, cache
);
3016 btrfs_write_out_cache(root
, trans
, cache
, path
);
3019 * If we didn't have an error then the cache state is still
3020 * NEED_WRITE, so we can set it to WRITTEN.
3022 if (cache
->disk_cache_state
== BTRFS_DC_NEED_WRITE
)
3023 cache
->disk_cache_state
= BTRFS_DC_WRITTEN
;
3024 last
= cache
->key
.objectid
+ cache
->key
.offset
;
3025 btrfs_put_block_group(cache
);
3028 btrfs_free_path(path
);
3032 int btrfs_extent_readonly(struct btrfs_root
*root
, u64 bytenr
)
3034 struct btrfs_block_group_cache
*block_group
;
3037 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
3038 if (!block_group
|| block_group
->ro
)
3041 btrfs_put_block_group(block_group
);
3045 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
3046 u64 total_bytes
, u64 bytes_used
,
3047 struct btrfs_space_info
**space_info
)
3049 struct btrfs_space_info
*found
;
3053 if (flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3054 BTRFS_BLOCK_GROUP_RAID10
))
3059 found
= __find_space_info(info
, flags
);
3061 spin_lock(&found
->lock
);
3062 found
->total_bytes
+= total_bytes
;
3063 found
->disk_total
+= total_bytes
* factor
;
3064 found
->bytes_used
+= bytes_used
;
3065 found
->disk_used
+= bytes_used
* factor
;
3067 spin_unlock(&found
->lock
);
3068 *space_info
= found
;
3071 found
= kzalloc(sizeof(*found
), GFP_NOFS
);
3075 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
3076 INIT_LIST_HEAD(&found
->block_groups
[i
]);
3077 init_rwsem(&found
->groups_sem
);
3078 spin_lock_init(&found
->lock
);
3079 found
->flags
= flags
& BTRFS_BLOCK_GROUP_TYPE_MASK
;
3080 found
->total_bytes
= total_bytes
;
3081 found
->disk_total
= total_bytes
* factor
;
3082 found
->bytes_used
= bytes_used
;
3083 found
->disk_used
= bytes_used
* factor
;
3084 found
->bytes_pinned
= 0;
3085 found
->bytes_reserved
= 0;
3086 found
->bytes_readonly
= 0;
3087 found
->bytes_may_use
= 0;
3089 found
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3090 found
->chunk_alloc
= 0;
3092 init_waitqueue_head(&found
->wait
);
3093 *space_info
= found
;
3094 list_add_rcu(&found
->list
, &info
->space_info
);
3098 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
3100 u64 extra_flags
= flags
& BTRFS_BLOCK_GROUP_PROFILE_MASK
;
3102 /* chunk -> extended profile */
3103 if (extra_flags
== 0)
3104 extra_flags
= BTRFS_AVAIL_ALLOC_BIT_SINGLE
;
3106 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3107 fs_info
->avail_data_alloc_bits
|= extra_flags
;
3108 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3109 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
3110 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3111 fs_info
->avail_system_alloc_bits
|= extra_flags
;
3115 * @flags: available profiles in extended format (see ctree.h)
3117 * Returns reduced profile in chunk format. If profile changing is in
3118 * progress (either running or paused) picks the target profile (if it's
3119 * already available), otherwise falls back to plain reducing.
3121 u64
btrfs_reduce_alloc_profile(struct btrfs_root
*root
, u64 flags
)
3124 * we add in the count of missing devices because we want
3125 * to make sure that any RAID levels on a degraded FS
3126 * continue to be honored.
3128 u64 num_devices
= root
->fs_info
->fs_devices
->rw_devices
+
3129 root
->fs_info
->fs_devices
->missing_devices
;
3131 /* pick restriper's target profile if it's available */
3132 spin_lock(&root
->fs_info
->balance_lock
);
3133 if (root
->fs_info
->balance_ctl
) {
3134 struct btrfs_balance_control
*bctl
= root
->fs_info
->balance_ctl
;
3137 if ((flags
& BTRFS_BLOCK_GROUP_DATA
) &&
3138 (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3139 (flags
& bctl
->data
.target
)) {
3140 tgt
= BTRFS_BLOCK_GROUP_DATA
| bctl
->data
.target
;
3141 } else if ((flags
& BTRFS_BLOCK_GROUP_SYSTEM
) &&
3142 (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3143 (flags
& bctl
->sys
.target
)) {
3144 tgt
= BTRFS_BLOCK_GROUP_SYSTEM
| bctl
->sys
.target
;
3145 } else if ((flags
& BTRFS_BLOCK_GROUP_METADATA
) &&
3146 (bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3147 (flags
& bctl
->meta
.target
)) {
3148 tgt
= BTRFS_BLOCK_GROUP_METADATA
| bctl
->meta
.target
;
3152 spin_unlock(&root
->fs_info
->balance_lock
);
3157 spin_unlock(&root
->fs_info
->balance_lock
);
3159 if (num_devices
== 1)
3160 flags
&= ~(BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID0
);
3161 if (num_devices
< 4)
3162 flags
&= ~BTRFS_BLOCK_GROUP_RAID10
;
3164 if ((flags
& BTRFS_BLOCK_GROUP_DUP
) &&
3165 (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
3166 BTRFS_BLOCK_GROUP_RAID10
))) {
3167 flags
&= ~BTRFS_BLOCK_GROUP_DUP
;
3170 if ((flags
& BTRFS_BLOCK_GROUP_RAID1
) &&
3171 (flags
& BTRFS_BLOCK_GROUP_RAID10
)) {
3172 flags
&= ~BTRFS_BLOCK_GROUP_RAID1
;
3175 if ((flags
& BTRFS_BLOCK_GROUP_RAID0
) &&
3176 ((flags
& BTRFS_BLOCK_GROUP_RAID1
) |
3177 (flags
& BTRFS_BLOCK_GROUP_RAID10
) |
3178 (flags
& BTRFS_BLOCK_GROUP_DUP
))) {
3179 flags
&= ~BTRFS_BLOCK_GROUP_RAID0
;
3183 /* extended -> chunk profile */
3184 flags
&= ~BTRFS_AVAIL_ALLOC_BIT_SINGLE
;
3188 static u64
get_alloc_profile(struct btrfs_root
*root
, u64 flags
)
3190 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3191 flags
|= root
->fs_info
->avail_data_alloc_bits
;
3192 else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3193 flags
|= root
->fs_info
->avail_system_alloc_bits
;
3194 else if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3195 flags
|= root
->fs_info
->avail_metadata_alloc_bits
;
3197 return btrfs_reduce_alloc_profile(root
, flags
);
3200 u64
btrfs_get_alloc_profile(struct btrfs_root
*root
, int data
)
3205 flags
= BTRFS_BLOCK_GROUP_DATA
;
3206 else if (root
== root
->fs_info
->chunk_root
)
3207 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
3209 flags
= BTRFS_BLOCK_GROUP_METADATA
;
3211 return get_alloc_profile(root
, flags
);
3214 void btrfs_set_inode_space_info(struct btrfs_root
*root
, struct inode
*inode
)
3216 BTRFS_I(inode
)->space_info
= __find_space_info(root
->fs_info
,
3217 BTRFS_BLOCK_GROUP_DATA
);
3221 * This will check the space that the inode allocates from to make sure we have
3222 * enough space for bytes.
3224 int btrfs_check_data_free_space(struct inode
*inode
, u64 bytes
)
3226 struct btrfs_space_info
*data_sinfo
;
3227 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3229 int ret
= 0, committed
= 0, alloc_chunk
= 1;
3231 /* make sure bytes are sectorsize aligned */
3232 bytes
= (bytes
+ root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
3234 if (root
== root
->fs_info
->tree_root
||
3235 BTRFS_I(inode
)->location
.objectid
== BTRFS_FREE_INO_OBJECTID
) {
3240 data_sinfo
= BTRFS_I(inode
)->space_info
;
3245 /* make sure we have enough space to handle the data first */
3246 spin_lock(&data_sinfo
->lock
);
3247 used
= data_sinfo
->bytes_used
+ data_sinfo
->bytes_reserved
+
3248 data_sinfo
->bytes_pinned
+ data_sinfo
->bytes_readonly
+
3249 data_sinfo
->bytes_may_use
;
3251 if (used
+ bytes
> data_sinfo
->total_bytes
) {
3252 struct btrfs_trans_handle
*trans
;
3255 * if we don't have enough free bytes in this space then we need
3256 * to alloc a new chunk.
3258 if (!data_sinfo
->full
&& alloc_chunk
) {
3261 data_sinfo
->force_alloc
= CHUNK_ALLOC_FORCE
;
3262 spin_unlock(&data_sinfo
->lock
);
3264 alloc_target
= btrfs_get_alloc_profile(root
, 1);
3265 trans
= btrfs_join_transaction(root
);
3267 return PTR_ERR(trans
);
3269 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
3270 bytes
+ 2 * 1024 * 1024,
3272 CHUNK_ALLOC_NO_FORCE
);
3273 btrfs_end_transaction(trans
, root
);
3282 btrfs_set_inode_space_info(root
, inode
);
3283 data_sinfo
= BTRFS_I(inode
)->space_info
;
3289 * If we have less pinned bytes than we want to allocate then
3290 * don't bother committing the transaction, it won't help us.
3292 if (data_sinfo
->bytes_pinned
< bytes
)
3294 spin_unlock(&data_sinfo
->lock
);
3296 /* commit the current transaction and try again */
3299 !atomic_read(&root
->fs_info
->open_ioctl_trans
)) {
3301 trans
= btrfs_join_transaction(root
);
3303 return PTR_ERR(trans
);
3304 ret
= btrfs_commit_transaction(trans
, root
);
3312 data_sinfo
->bytes_may_use
+= bytes
;
3313 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
3314 (u64
)data_sinfo
, bytes
, 1);
3315 spin_unlock(&data_sinfo
->lock
);
3321 * Called if we need to clear a data reservation for this inode.
3323 void btrfs_free_reserved_data_space(struct inode
*inode
, u64 bytes
)
3325 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3326 struct btrfs_space_info
*data_sinfo
;
3328 /* make sure bytes are sectorsize aligned */
3329 bytes
= (bytes
+ root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
3331 data_sinfo
= BTRFS_I(inode
)->space_info
;
3332 spin_lock(&data_sinfo
->lock
);
3333 data_sinfo
->bytes_may_use
-= bytes
;
3334 trace_btrfs_space_reservation(root
->fs_info
, "space_info",
3335 (u64
)data_sinfo
, bytes
, 0);
3336 spin_unlock(&data_sinfo
->lock
);
3339 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
3341 struct list_head
*head
= &info
->space_info
;
3342 struct btrfs_space_info
*found
;
3345 list_for_each_entry_rcu(found
, head
, list
) {
3346 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
3347 found
->force_alloc
= CHUNK_ALLOC_FORCE
;
3352 static int should_alloc_chunk(struct btrfs_root
*root
,
3353 struct btrfs_space_info
*sinfo
, u64 alloc_bytes
,
3356 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
3357 u64 num_bytes
= sinfo
->total_bytes
- sinfo
->bytes_readonly
;
3358 u64 num_allocated
= sinfo
->bytes_used
+ sinfo
->bytes_reserved
;
3361 if (force
== CHUNK_ALLOC_FORCE
)
3365 * We need to take into account the global rsv because for all intents
3366 * and purposes it's used space. Don't worry about locking the
3367 * global_rsv, it doesn't change except when the transaction commits.
3369 num_allocated
+= global_rsv
->size
;
3372 * in limited mode, we want to have some free space up to
3373 * about 1% of the FS size.
3375 if (force
== CHUNK_ALLOC_LIMITED
) {
3376 thresh
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
3377 thresh
= max_t(u64
, 64 * 1024 * 1024,
3378 div_factor_fine(thresh
, 1));
3380 if (num_bytes
- num_allocated
< thresh
)
3383 thresh
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
3385 /* 256MB or 2% of the FS */
3386 thresh
= max_t(u64
, 256 * 1024 * 1024, div_factor_fine(thresh
, 2));
3388 if (num_bytes
> thresh
&& sinfo
->bytes_used
< div_factor(num_bytes
, 8))
3393 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
3394 struct btrfs_root
*extent_root
, u64 alloc_bytes
,
3395 u64 flags
, int force
)
3397 struct btrfs_space_info
*space_info
;
3398 struct btrfs_fs_info
*fs_info
= extent_root
->fs_info
;
3399 int wait_for_alloc
= 0;
3402 BUG_ON(!profile_is_valid(flags
, 0));
3404 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
3406 ret
= update_space_info(extent_root
->fs_info
, flags
,
3410 BUG_ON(!space_info
);
3413 spin_lock(&space_info
->lock
);
3414 if (space_info
->force_alloc
)
3415 force
= space_info
->force_alloc
;
3416 if (space_info
->full
) {
3417 spin_unlock(&space_info
->lock
);
3421 if (!should_alloc_chunk(extent_root
, space_info
, alloc_bytes
, force
)) {
3422 spin_unlock(&space_info
->lock
);
3424 } else if (space_info
->chunk_alloc
) {
3427 space_info
->chunk_alloc
= 1;
3430 spin_unlock(&space_info
->lock
);
3432 mutex_lock(&fs_info
->chunk_mutex
);
3435 * The chunk_mutex is held throughout the entirety of a chunk
3436 * allocation, so once we've acquired the chunk_mutex we know that the
3437 * other guy is done and we need to recheck and see if we should
3440 if (wait_for_alloc
) {
3441 mutex_unlock(&fs_info
->chunk_mutex
);
3447 * If we have mixed data/metadata chunks we want to make sure we keep
3448 * allocating mixed chunks instead of individual chunks.
3450 if (btrfs_mixed_space_info(space_info
))
3451 flags
|= (BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
);
3454 * if we're doing a data chunk, go ahead and make sure that
3455 * we keep a reasonable number of metadata chunks allocated in the
3458 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
3459 fs_info
->data_chunk_allocations
++;
3460 if (!(fs_info
->data_chunk_allocations
%
3461 fs_info
->metadata_ratio
))
3462 force_metadata_allocation(fs_info
);
3465 ret
= btrfs_alloc_chunk(trans
, extent_root
, flags
);
3466 if (ret
< 0 && ret
!= -ENOSPC
)
3469 spin_lock(&space_info
->lock
);
3471 space_info
->full
= 1;
3475 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3476 space_info
->chunk_alloc
= 0;
3477 spin_unlock(&space_info
->lock
);
3479 mutex_unlock(&extent_root
->fs_info
->chunk_mutex
);
3484 * shrink metadata reservation for delalloc
3486 static int shrink_delalloc(struct btrfs_root
*root
, u64 to_reclaim
,
3489 struct btrfs_block_rsv
*block_rsv
;
3490 struct btrfs_space_info
*space_info
;
3491 struct btrfs_trans_handle
*trans
;
3496 unsigned long nr_pages
= (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT
;
3498 unsigned long progress
;
3500 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
3501 block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
3502 space_info
= block_rsv
->space_info
;
3505 reserved
= space_info
->bytes_may_use
;
3506 progress
= space_info
->reservation_progress
;
3512 if (root
->fs_info
->delalloc_bytes
== 0) {
3515 btrfs_wait_ordered_extents(root
, 0, 0);
3519 max_reclaim
= min(reserved
, to_reclaim
);
3520 nr_pages
= max_t(unsigned long, nr_pages
,
3521 max_reclaim
>> PAGE_CACHE_SHIFT
);
3522 while (loops
< 1024) {
3523 /* have the flusher threads jump in and do some IO */
3525 nr_pages
= min_t(unsigned long, nr_pages
,
3526 root
->fs_info
->delalloc_bytes
>> PAGE_CACHE_SHIFT
);
3527 writeback_inodes_sb_nr_if_idle(root
->fs_info
->sb
, nr_pages
,
3528 WB_REASON_FS_FREE_SPACE
);
3530 spin_lock(&space_info
->lock
);
3531 if (reserved
> space_info
->bytes_may_use
)
3532 reclaimed
+= reserved
- space_info
->bytes_may_use
;
3533 reserved
= space_info
->bytes_may_use
;
3534 spin_unlock(&space_info
->lock
);
3538 if (reserved
== 0 || reclaimed
>= max_reclaim
)
3541 if (trans
&& trans
->transaction
->blocked
)
3544 if (wait_ordered
&& !trans
) {
3545 btrfs_wait_ordered_extents(root
, 0, 0);
3547 time_left
= schedule_timeout_interruptible(1);
3549 /* We were interrupted, exit */
3554 /* we've kicked the IO a few times, if anything has been freed,
3555 * exit. There is no sense in looping here for a long time
3556 * when we really need to commit the transaction, or there are
3557 * just too many writers without enough free space
3562 if (progress
!= space_info
->reservation_progress
)
3568 return reclaimed
>= to_reclaim
;
3572 * maybe_commit_transaction - possibly commit the transaction if its ok to
3573 * @root - the root we're allocating for
3574 * @bytes - the number of bytes we want to reserve
3575 * @force - force the commit
3577 * This will check to make sure that committing the transaction will actually
3578 * get us somewhere and then commit the transaction if it does. Otherwise it
3579 * will return -ENOSPC.
3581 static int may_commit_transaction(struct btrfs_root
*root
,
3582 struct btrfs_space_info
*space_info
,
3583 u64 bytes
, int force
)
3585 struct btrfs_block_rsv
*delayed_rsv
= &root
->fs_info
->delayed_block_rsv
;
3586 struct btrfs_trans_handle
*trans
;
3588 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
3595 /* See if there is enough pinned space to make this reservation */
3596 spin_lock(&space_info
->lock
);
3597 if (space_info
->bytes_pinned
>= bytes
) {
3598 spin_unlock(&space_info
->lock
);
3601 spin_unlock(&space_info
->lock
);
3604 * See if there is some space in the delayed insertion reservation for
3607 if (space_info
!= delayed_rsv
->space_info
)
3610 spin_lock(&delayed_rsv
->lock
);
3611 if (delayed_rsv
->size
< bytes
) {
3612 spin_unlock(&delayed_rsv
->lock
);
3615 spin_unlock(&delayed_rsv
->lock
);
3618 trans
= btrfs_join_transaction(root
);
3622 return btrfs_commit_transaction(trans
, root
);
3626 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3627 * @root - the root we're allocating for
3628 * @block_rsv - the block_rsv we're allocating for
3629 * @orig_bytes - the number of bytes we want
3630 * @flush - wether or not we can flush to make our reservation
3632 * This will reserve orgi_bytes number of bytes from the space info associated
3633 * with the block_rsv. If there is not enough space it will make an attempt to
3634 * flush out space to make room. It will do this by flushing delalloc if
3635 * possible or committing the transaction. If flush is 0 then no attempts to
3636 * regain reservations will be made and this will fail if there is not enough
3639 static int reserve_metadata_bytes(struct btrfs_root
*root
,
3640 struct btrfs_block_rsv
*block_rsv
,
3641 u64 orig_bytes
, int flush
)
3643 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
3645 u64 num_bytes
= orig_bytes
;
3648 bool committed
= false;
3649 bool flushing
= false;
3650 bool wait_ordered
= false;
3654 spin_lock(&space_info
->lock
);
3656 * We only want to wait if somebody other than us is flushing and we are
3657 * actually alloed to flush.
3659 while (flush
&& !flushing
&& space_info
->flush
) {
3660 spin_unlock(&space_info
->lock
);
3662 * If we have a trans handle we can't wait because the flusher
3663 * may have to commit the transaction, which would mean we would
3664 * deadlock since we are waiting for the flusher to finish, but
3665 * hold the current transaction open.
3667 if (current
->journal_info
)
3669 ret
= wait_event_interruptible(space_info
->wait
,
3670 !space_info
->flush
);
3671 /* Must have been interrupted, return */
3675 spin_lock(&space_info
->lock
);
3679 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
3680 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
3681 space_info
->bytes_may_use
;
3684 * The idea here is that we've not already over-reserved the block group
3685 * then we can go ahead and save our reservation first and then start
3686 * flushing if we need to. Otherwise if we've already overcommitted
3687 * lets start flushing stuff first and then come back and try to make
3690 if (used
<= space_info
->total_bytes
) {
3691 if (used
+ orig_bytes
<= space_info
->total_bytes
) {
3692 space_info
->bytes_may_use
+= orig_bytes
;
3693 trace_btrfs_space_reservation(root
->fs_info
,
3700 * Ok set num_bytes to orig_bytes since we aren't
3701 * overocmmitted, this way we only try and reclaim what
3704 num_bytes
= orig_bytes
;
3708 * Ok we're over committed, set num_bytes to the overcommitted
3709 * amount plus the amount of bytes that we need for this
3712 wait_ordered
= true;
3713 num_bytes
= used
- space_info
->total_bytes
+
3714 (orig_bytes
* (retries
+ 1));
3718 u64 profile
= btrfs_get_alloc_profile(root
, 0);
3722 * If we have a lot of space that's pinned, don't bother doing
3723 * the overcommit dance yet and just commit the transaction.
3725 avail
= (space_info
->total_bytes
- space_info
->bytes_used
) * 8;
3727 if (space_info
->bytes_pinned
>= avail
&& flush
&& !committed
) {
3728 space_info
->flush
= 1;
3730 spin_unlock(&space_info
->lock
);
3731 ret
= may_commit_transaction(root
, space_info
,
3739 spin_lock(&root
->fs_info
->free_chunk_lock
);
3740 avail
= root
->fs_info
->free_chunk_space
;
3743 * If we have dup, raid1 or raid10 then only half of the free
3744 * space is actually useable.
3746 if (profile
& (BTRFS_BLOCK_GROUP_DUP
|
3747 BTRFS_BLOCK_GROUP_RAID1
|
3748 BTRFS_BLOCK_GROUP_RAID10
))
3752 * If we aren't flushing don't let us overcommit too much, say
3753 * 1/8th of the space. If we can flush, let it overcommit up to
3760 spin_unlock(&root
->fs_info
->free_chunk_lock
);
3762 if (used
+ num_bytes
< space_info
->total_bytes
+ avail
) {
3763 space_info
->bytes_may_use
+= orig_bytes
;
3764 trace_btrfs_space_reservation(root
->fs_info
,
3770 wait_ordered
= true;
3775 * Couldn't make our reservation, save our place so while we're trying
3776 * to reclaim space we can actually use it instead of somebody else
3777 * stealing it from us.
3781 space_info
->flush
= 1;
3784 spin_unlock(&space_info
->lock
);
3790 * We do synchronous shrinking since we don't actually unreserve
3791 * metadata until after the IO is completed.
3793 ret
= shrink_delalloc(root
, num_bytes
, wait_ordered
);
3800 * So if we were overcommitted it's possible that somebody else flushed
3801 * out enough space and we simply didn't have enough space to reclaim,
3802 * so go back around and try again.
3805 wait_ordered
= true;
3814 ret
= may_commit_transaction(root
, space_info
, orig_bytes
, 0);
3822 spin_lock(&space_info
->lock
);
3823 space_info
->flush
= 0;
3824 wake_up_all(&space_info
->wait
);
3825 spin_unlock(&space_info
->lock
);
3830 static struct btrfs_block_rsv
*get_block_rsv(struct btrfs_trans_handle
*trans
,
3831 struct btrfs_root
*root
)
3833 struct btrfs_block_rsv
*block_rsv
= NULL
;
3835 if (root
->ref_cows
|| root
== root
->fs_info
->csum_root
)
3836 block_rsv
= trans
->block_rsv
;
3839 block_rsv
= root
->block_rsv
;
3842 block_rsv
= &root
->fs_info
->empty_block_rsv
;
3847 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
3851 spin_lock(&block_rsv
->lock
);
3852 if (block_rsv
->reserved
>= num_bytes
) {
3853 block_rsv
->reserved
-= num_bytes
;
3854 if (block_rsv
->reserved
< block_rsv
->size
)
3855 block_rsv
->full
= 0;
3858 spin_unlock(&block_rsv
->lock
);
3862 static void block_rsv_add_bytes(struct btrfs_block_rsv
*block_rsv
,
3863 u64 num_bytes
, int update_size
)
3865 spin_lock(&block_rsv
->lock
);
3866 block_rsv
->reserved
+= num_bytes
;
3868 block_rsv
->size
+= num_bytes
;
3869 else if (block_rsv
->reserved
>= block_rsv
->size
)
3870 block_rsv
->full
= 1;
3871 spin_unlock(&block_rsv
->lock
);
3874 static void block_rsv_release_bytes(struct btrfs_fs_info
*fs_info
,
3875 struct btrfs_block_rsv
*block_rsv
,
3876 struct btrfs_block_rsv
*dest
, u64 num_bytes
)
3878 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
3880 spin_lock(&block_rsv
->lock
);
3881 if (num_bytes
== (u64
)-1)
3882 num_bytes
= block_rsv
->size
;
3883 block_rsv
->size
-= num_bytes
;
3884 if (block_rsv
->reserved
>= block_rsv
->size
) {
3885 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
3886 block_rsv
->reserved
= block_rsv
->size
;
3887 block_rsv
->full
= 1;
3891 spin_unlock(&block_rsv
->lock
);
3893 if (num_bytes
> 0) {
3895 spin_lock(&dest
->lock
);
3899 bytes_to_add
= dest
->size
- dest
->reserved
;
3900 bytes_to_add
= min(num_bytes
, bytes_to_add
);
3901 dest
->reserved
+= bytes_to_add
;
3902 if (dest
->reserved
>= dest
->size
)
3904 num_bytes
-= bytes_to_add
;
3906 spin_unlock(&dest
->lock
);
3909 spin_lock(&space_info
->lock
);
3910 space_info
->bytes_may_use
-= num_bytes
;
3911 trace_btrfs_space_reservation(fs_info
, "space_info",
3914 space_info
->reservation_progress
++;
3915 spin_unlock(&space_info
->lock
);
3920 static int block_rsv_migrate_bytes(struct btrfs_block_rsv
*src
,
3921 struct btrfs_block_rsv
*dst
, u64 num_bytes
)
3925 ret
= block_rsv_use_bytes(src
, num_bytes
);
3929 block_rsv_add_bytes(dst
, num_bytes
, 1);
3933 void btrfs_init_block_rsv(struct btrfs_block_rsv
*rsv
)
3935 memset(rsv
, 0, sizeof(*rsv
));
3936 spin_lock_init(&rsv
->lock
);
3939 struct btrfs_block_rsv
*btrfs_alloc_block_rsv(struct btrfs_root
*root
)
3941 struct btrfs_block_rsv
*block_rsv
;
3942 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3944 block_rsv
= kmalloc(sizeof(*block_rsv
), GFP_NOFS
);
3948 btrfs_init_block_rsv(block_rsv
);
3949 block_rsv
->space_info
= __find_space_info(fs_info
,
3950 BTRFS_BLOCK_GROUP_METADATA
);
3954 void btrfs_free_block_rsv(struct btrfs_root
*root
,
3955 struct btrfs_block_rsv
*rsv
)
3957 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
3961 static inline int __block_rsv_add(struct btrfs_root
*root
,
3962 struct btrfs_block_rsv
*block_rsv
,
3963 u64 num_bytes
, int flush
)
3970 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
3972 block_rsv_add_bytes(block_rsv
, num_bytes
, 1);
3979 int btrfs_block_rsv_add(struct btrfs_root
*root
,
3980 struct btrfs_block_rsv
*block_rsv
,
3983 return __block_rsv_add(root
, block_rsv
, num_bytes
, 1);
3986 int btrfs_block_rsv_add_noflush(struct btrfs_root
*root
,
3987 struct btrfs_block_rsv
*block_rsv
,
3990 return __block_rsv_add(root
, block_rsv
, num_bytes
, 0);
3993 int btrfs_block_rsv_check(struct btrfs_root
*root
,
3994 struct btrfs_block_rsv
*block_rsv
, int min_factor
)
4002 spin_lock(&block_rsv
->lock
);
4003 num_bytes
= div_factor(block_rsv
->size
, min_factor
);
4004 if (block_rsv
->reserved
>= num_bytes
)
4006 spin_unlock(&block_rsv
->lock
);
4011 static inline int __btrfs_block_rsv_refill(struct btrfs_root
*root
,
4012 struct btrfs_block_rsv
*block_rsv
,
4013 u64 min_reserved
, int flush
)
4021 spin_lock(&block_rsv
->lock
);
4022 num_bytes
= min_reserved
;
4023 if (block_rsv
->reserved
>= num_bytes
)
4026 num_bytes
-= block_rsv
->reserved
;
4027 spin_unlock(&block_rsv
->lock
);
4032 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
4034 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
4041 int btrfs_block_rsv_refill(struct btrfs_root
*root
,
4042 struct btrfs_block_rsv
*block_rsv
,
4045 return __btrfs_block_rsv_refill(root
, block_rsv
, min_reserved
, 1);
4048 int btrfs_block_rsv_refill_noflush(struct btrfs_root
*root
,
4049 struct btrfs_block_rsv
*block_rsv
,
4052 return __btrfs_block_rsv_refill(root
, block_rsv
, min_reserved
, 0);
4055 int btrfs_block_rsv_migrate(struct btrfs_block_rsv
*src_rsv
,
4056 struct btrfs_block_rsv
*dst_rsv
,
4059 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
4062 void btrfs_block_rsv_release(struct btrfs_root
*root
,
4063 struct btrfs_block_rsv
*block_rsv
,
4066 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
4067 if (global_rsv
->full
|| global_rsv
== block_rsv
||
4068 block_rsv
->space_info
!= global_rsv
->space_info
)
4070 block_rsv_release_bytes(root
->fs_info
, block_rsv
, global_rsv
,
4075 * helper to calculate size of global block reservation.
4076 * the desired value is sum of space used by extent tree,
4077 * checksum tree and root tree
4079 static u64
calc_global_metadata_size(struct btrfs_fs_info
*fs_info
)
4081 struct btrfs_space_info
*sinfo
;
4085 int csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
4087 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_DATA
);
4088 spin_lock(&sinfo
->lock
);
4089 data_used
= sinfo
->bytes_used
;
4090 spin_unlock(&sinfo
->lock
);
4092 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4093 spin_lock(&sinfo
->lock
);
4094 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_DATA
)
4096 meta_used
= sinfo
->bytes_used
;
4097 spin_unlock(&sinfo
->lock
);
4099 num_bytes
= (data_used
>> fs_info
->sb
->s_blocksize_bits
) *
4101 num_bytes
+= div64_u64(data_used
+ meta_used
, 50);
4103 if (num_bytes
* 3 > meta_used
)
4104 num_bytes
= div64_u64(meta_used
, 3);
4106 return ALIGN(num_bytes
, fs_info
->extent_root
->leafsize
<< 10);
4109 static void update_global_block_rsv(struct btrfs_fs_info
*fs_info
)
4111 struct btrfs_block_rsv
*block_rsv
= &fs_info
->global_block_rsv
;
4112 struct btrfs_space_info
*sinfo
= block_rsv
->space_info
;
4115 num_bytes
= calc_global_metadata_size(fs_info
);
4117 spin_lock(&block_rsv
->lock
);
4118 spin_lock(&sinfo
->lock
);
4120 block_rsv
->size
= num_bytes
;
4122 num_bytes
= sinfo
->bytes_used
+ sinfo
->bytes_pinned
+
4123 sinfo
->bytes_reserved
+ sinfo
->bytes_readonly
+
4124 sinfo
->bytes_may_use
;
4126 if (sinfo
->total_bytes
> num_bytes
) {
4127 num_bytes
= sinfo
->total_bytes
- num_bytes
;
4128 block_rsv
->reserved
+= num_bytes
;
4129 sinfo
->bytes_may_use
+= num_bytes
;
4130 trace_btrfs_space_reservation(fs_info
, "space_info",
4131 (u64
)sinfo
, num_bytes
, 1);
4134 if (block_rsv
->reserved
>= block_rsv
->size
) {
4135 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
4136 sinfo
->bytes_may_use
-= num_bytes
;
4137 trace_btrfs_space_reservation(fs_info
, "space_info",
4138 (u64
)sinfo
, num_bytes
, 0);
4139 sinfo
->reservation_progress
++;
4140 block_rsv
->reserved
= block_rsv
->size
;
4141 block_rsv
->full
= 1;
4144 spin_unlock(&sinfo
->lock
);
4145 spin_unlock(&block_rsv
->lock
);
4148 static void init_global_block_rsv(struct btrfs_fs_info
*fs_info
)
4150 struct btrfs_space_info
*space_info
;
4152 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
4153 fs_info
->chunk_block_rsv
.space_info
= space_info
;
4155 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4156 fs_info
->global_block_rsv
.space_info
= space_info
;
4157 fs_info
->delalloc_block_rsv
.space_info
= space_info
;
4158 fs_info
->trans_block_rsv
.space_info
= space_info
;
4159 fs_info
->empty_block_rsv
.space_info
= space_info
;
4160 fs_info
->delayed_block_rsv
.space_info
= space_info
;
4162 fs_info
->extent_root
->block_rsv
= &fs_info
->global_block_rsv
;
4163 fs_info
->csum_root
->block_rsv
= &fs_info
->global_block_rsv
;
4164 fs_info
->dev_root
->block_rsv
= &fs_info
->global_block_rsv
;
4165 fs_info
->tree_root
->block_rsv
= &fs_info
->global_block_rsv
;
4166 fs_info
->chunk_root
->block_rsv
= &fs_info
->chunk_block_rsv
;
4168 update_global_block_rsv(fs_info
);
4171 static void release_global_block_rsv(struct btrfs_fs_info
*fs_info
)
4173 block_rsv_release_bytes(fs_info
, &fs_info
->global_block_rsv
, NULL
,
4175 WARN_ON(fs_info
->delalloc_block_rsv
.size
> 0);
4176 WARN_ON(fs_info
->delalloc_block_rsv
.reserved
> 0);
4177 WARN_ON(fs_info
->trans_block_rsv
.size
> 0);
4178 WARN_ON(fs_info
->trans_block_rsv
.reserved
> 0);
4179 WARN_ON(fs_info
->chunk_block_rsv
.size
> 0);
4180 WARN_ON(fs_info
->chunk_block_rsv
.reserved
> 0);
4181 WARN_ON(fs_info
->delayed_block_rsv
.size
> 0);
4182 WARN_ON(fs_info
->delayed_block_rsv
.reserved
> 0);
4185 void btrfs_trans_release_metadata(struct btrfs_trans_handle
*trans
,
4186 struct btrfs_root
*root
)
4188 if (!trans
->bytes_reserved
)
4191 trace_btrfs_space_reservation(root
->fs_info
, "transaction", (u64
)trans
,
4192 trans
->bytes_reserved
, 0);
4193 btrfs_block_rsv_release(root
, trans
->block_rsv
, trans
->bytes_reserved
);
4194 trans
->bytes_reserved
= 0;
4197 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle
*trans
,
4198 struct inode
*inode
)
4200 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4201 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
4202 struct btrfs_block_rsv
*dst_rsv
= root
->orphan_block_rsv
;
4205 * We need to hold space in order to delete our orphan item once we've
4206 * added it, so this takes the reservation so we can release it later
4207 * when we are truly done with the orphan item.
4209 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
4210 trace_btrfs_space_reservation(root
->fs_info
, "orphan",
4211 btrfs_ino(inode
), num_bytes
, 1);
4212 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
4215 void btrfs_orphan_release_metadata(struct inode
*inode
)
4217 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4218 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
4219 trace_btrfs_space_reservation(root
->fs_info
, "orphan",
4220 btrfs_ino(inode
), num_bytes
, 0);
4221 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
, num_bytes
);
4224 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle
*trans
,
4225 struct btrfs_pending_snapshot
*pending
)
4227 struct btrfs_root
*root
= pending
->root
;
4228 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
4229 struct btrfs_block_rsv
*dst_rsv
= &pending
->block_rsv
;
4231 * two for root back/forward refs, two for directory entries
4232 * and one for root of the snapshot.
4234 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 5);
4235 dst_rsv
->space_info
= src_rsv
->space_info
;
4236 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
4240 * drop_outstanding_extent - drop an outstanding extent
4241 * @inode: the inode we're dropping the extent for
4243 * This is called when we are freeing up an outstanding extent, either called
4244 * after an error or after an extent is written. This will return the number of
4245 * reserved extents that need to be freed. This must be called with
4246 * BTRFS_I(inode)->lock held.
4248 static unsigned drop_outstanding_extent(struct inode
*inode
)
4250 unsigned drop_inode_space
= 0;
4251 unsigned dropped_extents
= 0;
4253 BUG_ON(!BTRFS_I(inode
)->outstanding_extents
);
4254 BTRFS_I(inode
)->outstanding_extents
--;
4256 if (BTRFS_I(inode
)->outstanding_extents
== 0 &&
4257 BTRFS_I(inode
)->delalloc_meta_reserved
) {
4258 drop_inode_space
= 1;
4259 BTRFS_I(inode
)->delalloc_meta_reserved
= 0;
4263 * If we have more or the same amount of outsanding extents than we have
4264 * reserved then we need to leave the reserved extents count alone.
4266 if (BTRFS_I(inode
)->outstanding_extents
>=
4267 BTRFS_I(inode
)->reserved_extents
)
4268 return drop_inode_space
;
4270 dropped_extents
= BTRFS_I(inode
)->reserved_extents
-
4271 BTRFS_I(inode
)->outstanding_extents
;
4272 BTRFS_I(inode
)->reserved_extents
-= dropped_extents
;
4273 return dropped_extents
+ drop_inode_space
;
4277 * calc_csum_metadata_size - return the amount of metada space that must be
4278 * reserved/free'd for the given bytes.
4279 * @inode: the inode we're manipulating
4280 * @num_bytes: the number of bytes in question
4281 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4283 * This adjusts the number of csum_bytes in the inode and then returns the
4284 * correct amount of metadata that must either be reserved or freed. We
4285 * calculate how many checksums we can fit into one leaf and then divide the
4286 * number of bytes that will need to be checksumed by this value to figure out
4287 * how many checksums will be required. If we are adding bytes then the number
4288 * may go up and we will return the number of additional bytes that must be
4289 * reserved. If it is going down we will return the number of bytes that must
4292 * This must be called with BTRFS_I(inode)->lock held.
4294 static u64
calc_csum_metadata_size(struct inode
*inode
, u64 num_bytes
,
4297 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4299 int num_csums_per_leaf
;
4303 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
&&
4304 BTRFS_I(inode
)->csum_bytes
== 0)
4307 old_csums
= (int)div64_u64(BTRFS_I(inode
)->csum_bytes
, root
->sectorsize
);
4309 BTRFS_I(inode
)->csum_bytes
+= num_bytes
;
4311 BTRFS_I(inode
)->csum_bytes
-= num_bytes
;
4312 csum_size
= BTRFS_LEAF_DATA_SIZE(root
) - sizeof(struct btrfs_item
);
4313 num_csums_per_leaf
= (int)div64_u64(csum_size
,
4314 sizeof(struct btrfs_csum_item
) +
4315 sizeof(struct btrfs_disk_key
));
4316 num_csums
= (int)div64_u64(BTRFS_I(inode
)->csum_bytes
, root
->sectorsize
);
4317 num_csums
= num_csums
+ num_csums_per_leaf
- 1;
4318 num_csums
= num_csums
/ num_csums_per_leaf
;
4320 old_csums
= old_csums
+ num_csums_per_leaf
- 1;
4321 old_csums
= old_csums
/ num_csums_per_leaf
;
4323 /* No change, no need to reserve more */
4324 if (old_csums
== num_csums
)
4328 return btrfs_calc_trans_metadata_size(root
,
4329 num_csums
- old_csums
);
4331 return btrfs_calc_trans_metadata_size(root
, old_csums
- num_csums
);
4334 int btrfs_delalloc_reserve_metadata(struct inode
*inode
, u64 num_bytes
)
4336 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4337 struct btrfs_block_rsv
*block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
4340 unsigned nr_extents
= 0;
4341 int extra_reserve
= 0;
4345 /* Need to be holding the i_mutex here if we aren't free space cache */
4346 if (btrfs_is_free_space_inode(root
, inode
))
4349 WARN_ON(!mutex_is_locked(&inode
->i_mutex
));
4351 if (flush
&& btrfs_transaction_in_commit(root
->fs_info
))
4352 schedule_timeout(1);
4354 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
4356 spin_lock(&BTRFS_I(inode
)->lock
);
4357 BTRFS_I(inode
)->outstanding_extents
++;
4359 if (BTRFS_I(inode
)->outstanding_extents
>
4360 BTRFS_I(inode
)->reserved_extents
)
4361 nr_extents
= BTRFS_I(inode
)->outstanding_extents
-
4362 BTRFS_I(inode
)->reserved_extents
;
4365 * Add an item to reserve for updating the inode when we complete the
4368 if (!BTRFS_I(inode
)->delalloc_meta_reserved
) {
4373 to_reserve
= btrfs_calc_trans_metadata_size(root
, nr_extents
);
4374 to_reserve
+= calc_csum_metadata_size(inode
, num_bytes
, 1);
4375 csum_bytes
= BTRFS_I(inode
)->csum_bytes
;
4376 spin_unlock(&BTRFS_I(inode
)->lock
);
4378 ret
= reserve_metadata_bytes(root
, block_rsv
, to_reserve
, flush
);
4383 spin_lock(&BTRFS_I(inode
)->lock
);
4384 dropped
= drop_outstanding_extent(inode
);
4386 * If the inodes csum_bytes is the same as the original
4387 * csum_bytes then we know we haven't raced with any free()ers
4388 * so we can just reduce our inodes csum bytes and carry on.
4389 * Otherwise we have to do the normal free thing to account for
4390 * the case that the free side didn't free up its reserve
4391 * because of this outstanding reservation.
4393 if (BTRFS_I(inode
)->csum_bytes
== csum_bytes
)
4394 calc_csum_metadata_size(inode
, num_bytes
, 0);
4396 to_free
= calc_csum_metadata_size(inode
, num_bytes
, 0);
4397 spin_unlock(&BTRFS_I(inode
)->lock
);
4399 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
4402 btrfs_block_rsv_release(root
, block_rsv
, to_free
);
4403 trace_btrfs_space_reservation(root
->fs_info
,
4411 spin_lock(&BTRFS_I(inode
)->lock
);
4412 if (extra_reserve
) {
4413 BTRFS_I(inode
)->delalloc_meta_reserved
= 1;
4416 BTRFS_I(inode
)->reserved_extents
+= nr_extents
;
4417 spin_unlock(&BTRFS_I(inode
)->lock
);
4420 trace_btrfs_space_reservation(root
->fs_info
,"delalloc",
4421 btrfs_ino(inode
), to_reserve
, 1);
4422 block_rsv_add_bytes(block_rsv
, to_reserve
, 1);
4428 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4429 * @inode: the inode to release the reservation for
4430 * @num_bytes: the number of bytes we're releasing
4432 * This will release the metadata reservation for an inode. This can be called
4433 * once we complete IO for a given set of bytes to release their metadata
4436 void btrfs_delalloc_release_metadata(struct inode
*inode
, u64 num_bytes
)
4438 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4442 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
4443 spin_lock(&BTRFS_I(inode
)->lock
);
4444 dropped
= drop_outstanding_extent(inode
);
4446 to_free
= calc_csum_metadata_size(inode
, num_bytes
, 0);
4447 spin_unlock(&BTRFS_I(inode
)->lock
);
4449 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
4451 trace_btrfs_space_reservation(root
->fs_info
, "delalloc",
4452 btrfs_ino(inode
), to_free
, 0);
4453 btrfs_block_rsv_release(root
, &root
->fs_info
->delalloc_block_rsv
,
4458 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4459 * @inode: inode we're writing to
4460 * @num_bytes: the number of bytes we want to allocate
4462 * This will do the following things
4464 * o reserve space in the data space info for num_bytes
4465 * o reserve space in the metadata space info based on number of outstanding
4466 * extents and how much csums will be needed
4467 * o add to the inodes ->delalloc_bytes
4468 * o add it to the fs_info's delalloc inodes list.
4470 * This will return 0 for success and -ENOSPC if there is no space left.
4472 int btrfs_delalloc_reserve_space(struct inode
*inode
, u64 num_bytes
)
4476 ret
= btrfs_check_data_free_space(inode
, num_bytes
);
4480 ret
= btrfs_delalloc_reserve_metadata(inode
, num_bytes
);
4482 btrfs_free_reserved_data_space(inode
, num_bytes
);
4490 * btrfs_delalloc_release_space - release data and metadata space for delalloc
4491 * @inode: inode we're releasing space for
4492 * @num_bytes: the number of bytes we want to free up
4494 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
4495 * called in the case that we don't need the metadata AND data reservations
4496 * anymore. So if there is an error or we insert an inline extent.
4498 * This function will release the metadata space that was not used and will
4499 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4500 * list if there are no delalloc bytes left.
4502 void btrfs_delalloc_release_space(struct inode
*inode
, u64 num_bytes
)
4504 btrfs_delalloc_release_metadata(inode
, num_bytes
);
4505 btrfs_free_reserved_data_space(inode
, num_bytes
);
4508 static int update_block_group(struct btrfs_trans_handle
*trans
,
4509 struct btrfs_root
*root
,
4510 u64 bytenr
, u64 num_bytes
, int alloc
)
4512 struct btrfs_block_group_cache
*cache
= NULL
;
4513 struct btrfs_fs_info
*info
= root
->fs_info
;
4514 u64 total
= num_bytes
;
4519 /* block accounting for super block */
4520 spin_lock(&info
->delalloc_lock
);
4521 old_val
= btrfs_super_bytes_used(info
->super_copy
);
4523 old_val
+= num_bytes
;
4525 old_val
-= num_bytes
;
4526 btrfs_set_super_bytes_used(info
->super_copy
, old_val
);
4527 spin_unlock(&info
->delalloc_lock
);
4530 cache
= btrfs_lookup_block_group(info
, bytenr
);
4533 if (cache
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
4534 BTRFS_BLOCK_GROUP_RAID1
|
4535 BTRFS_BLOCK_GROUP_RAID10
))
4540 * If this block group has free space cache written out, we
4541 * need to make sure to load it if we are removing space. This
4542 * is because we need the unpinning stage to actually add the
4543 * space back to the block group, otherwise we will leak space.
4545 if (!alloc
&& cache
->cached
== BTRFS_CACHE_NO
)
4546 cache_block_group(cache
, trans
, NULL
, 1);
4548 byte_in_group
= bytenr
- cache
->key
.objectid
;
4549 WARN_ON(byte_in_group
> cache
->key
.offset
);
4551 spin_lock(&cache
->space_info
->lock
);
4552 spin_lock(&cache
->lock
);
4554 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
4555 cache
->disk_cache_state
< BTRFS_DC_CLEAR
)
4556 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
4559 old_val
= btrfs_block_group_used(&cache
->item
);
4560 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
4562 old_val
+= num_bytes
;
4563 btrfs_set_block_group_used(&cache
->item
, old_val
);
4564 cache
->reserved
-= num_bytes
;
4565 cache
->space_info
->bytes_reserved
-= num_bytes
;
4566 cache
->space_info
->bytes_used
+= num_bytes
;
4567 cache
->space_info
->disk_used
+= num_bytes
* factor
;
4568 spin_unlock(&cache
->lock
);
4569 spin_unlock(&cache
->space_info
->lock
);
4571 old_val
-= num_bytes
;
4572 btrfs_set_block_group_used(&cache
->item
, old_val
);
4573 cache
->pinned
+= num_bytes
;
4574 cache
->space_info
->bytes_pinned
+= num_bytes
;
4575 cache
->space_info
->bytes_used
-= num_bytes
;
4576 cache
->space_info
->disk_used
-= num_bytes
* factor
;
4577 spin_unlock(&cache
->lock
);
4578 spin_unlock(&cache
->space_info
->lock
);
4580 set_extent_dirty(info
->pinned_extents
,
4581 bytenr
, bytenr
+ num_bytes
- 1,
4582 GFP_NOFS
| __GFP_NOFAIL
);
4584 btrfs_put_block_group(cache
);
4586 bytenr
+= num_bytes
;
4591 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
4593 struct btrfs_block_group_cache
*cache
;
4596 cache
= btrfs_lookup_first_block_group(root
->fs_info
, search_start
);
4600 bytenr
= cache
->key
.objectid
;
4601 btrfs_put_block_group(cache
);
4606 static int pin_down_extent(struct btrfs_root
*root
,
4607 struct btrfs_block_group_cache
*cache
,
4608 u64 bytenr
, u64 num_bytes
, int reserved
)
4610 spin_lock(&cache
->space_info
->lock
);
4611 spin_lock(&cache
->lock
);
4612 cache
->pinned
+= num_bytes
;
4613 cache
->space_info
->bytes_pinned
+= num_bytes
;
4615 cache
->reserved
-= num_bytes
;
4616 cache
->space_info
->bytes_reserved
-= num_bytes
;
4618 spin_unlock(&cache
->lock
);
4619 spin_unlock(&cache
->space_info
->lock
);
4621 set_extent_dirty(root
->fs_info
->pinned_extents
, bytenr
,
4622 bytenr
+ num_bytes
- 1, GFP_NOFS
| __GFP_NOFAIL
);
4627 * this function must be called within transaction
4629 int btrfs_pin_extent(struct btrfs_root
*root
,
4630 u64 bytenr
, u64 num_bytes
, int reserved
)
4632 struct btrfs_block_group_cache
*cache
;
4634 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
4637 pin_down_extent(root
, cache
, bytenr
, num_bytes
, reserved
);
4639 btrfs_put_block_group(cache
);
4644 * this function must be called within transaction
4646 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle
*trans
,
4647 struct btrfs_root
*root
,
4648 u64 bytenr
, u64 num_bytes
)
4650 struct btrfs_block_group_cache
*cache
;
4652 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
4656 * pull in the free space cache (if any) so that our pin
4657 * removes the free space from the cache. We have load_only set
4658 * to one because the slow code to read in the free extents does check
4659 * the pinned extents.
4661 cache_block_group(cache
, trans
, root
, 1);
4663 pin_down_extent(root
, cache
, bytenr
, num_bytes
, 0);
4665 /* remove us from the free space cache (if we're there at all) */
4666 btrfs_remove_free_space(cache
, bytenr
, num_bytes
);
4667 btrfs_put_block_group(cache
);
4672 * btrfs_update_reserved_bytes - update the block_group and space info counters
4673 * @cache: The cache we are manipulating
4674 * @num_bytes: The number of bytes in question
4675 * @reserve: One of the reservation enums
4677 * This is called by the allocator when it reserves space, or by somebody who is
4678 * freeing space that was never actually used on disk. For example if you
4679 * reserve some space for a new leaf in transaction A and before transaction A
4680 * commits you free that leaf, you call this with reserve set to 0 in order to
4681 * clear the reservation.
4683 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4684 * ENOSPC accounting. For data we handle the reservation through clearing the
4685 * delalloc bits in the io_tree. We have to do this since we could end up
4686 * allocating less disk space for the amount of data we have reserved in the
4687 * case of compression.
4689 * If this is a reservation and the block group has become read only we cannot
4690 * make the reservation and return -EAGAIN, otherwise this function always
4693 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
4694 u64 num_bytes
, int reserve
)
4696 struct btrfs_space_info
*space_info
= cache
->space_info
;
4698 spin_lock(&space_info
->lock
);
4699 spin_lock(&cache
->lock
);
4700 if (reserve
!= RESERVE_FREE
) {
4704 cache
->reserved
+= num_bytes
;
4705 space_info
->bytes_reserved
+= num_bytes
;
4706 if (reserve
== RESERVE_ALLOC
) {
4707 trace_btrfs_space_reservation(cache
->fs_info
,
4711 space_info
->bytes_may_use
-= num_bytes
;
4716 space_info
->bytes_readonly
+= num_bytes
;
4717 cache
->reserved
-= num_bytes
;
4718 space_info
->bytes_reserved
-= num_bytes
;
4719 space_info
->reservation_progress
++;
4721 spin_unlock(&cache
->lock
);
4722 spin_unlock(&space_info
->lock
);
4726 int btrfs_prepare_extent_commit(struct btrfs_trans_handle
*trans
,
4727 struct btrfs_root
*root
)
4729 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4730 struct btrfs_caching_control
*next
;
4731 struct btrfs_caching_control
*caching_ctl
;
4732 struct btrfs_block_group_cache
*cache
;
4734 down_write(&fs_info
->extent_commit_sem
);
4736 list_for_each_entry_safe(caching_ctl
, next
,
4737 &fs_info
->caching_block_groups
, list
) {
4738 cache
= caching_ctl
->block_group
;
4739 if (block_group_cache_done(cache
)) {
4740 cache
->last_byte_to_unpin
= (u64
)-1;
4741 list_del_init(&caching_ctl
->list
);
4742 put_caching_control(caching_ctl
);
4744 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
4748 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
4749 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
4751 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
4753 up_write(&fs_info
->extent_commit_sem
);
4755 update_global_block_rsv(fs_info
);
4759 static int unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
4761 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4762 struct btrfs_block_group_cache
*cache
= NULL
;
4765 while (start
<= end
) {
4767 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
4769 btrfs_put_block_group(cache
);
4770 cache
= btrfs_lookup_block_group(fs_info
, start
);
4774 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
4775 len
= min(len
, end
+ 1 - start
);
4777 if (start
< cache
->last_byte_to_unpin
) {
4778 len
= min(len
, cache
->last_byte_to_unpin
- start
);
4779 btrfs_add_free_space(cache
, start
, len
);
4784 spin_lock(&cache
->space_info
->lock
);
4785 spin_lock(&cache
->lock
);
4786 cache
->pinned
-= len
;
4787 cache
->space_info
->bytes_pinned
-= len
;
4789 cache
->space_info
->bytes_readonly
+= len
;
4790 spin_unlock(&cache
->lock
);
4791 spin_unlock(&cache
->space_info
->lock
);
4795 btrfs_put_block_group(cache
);
4799 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
4800 struct btrfs_root
*root
)
4802 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4803 struct extent_io_tree
*unpin
;
4808 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
4809 unpin
= &fs_info
->freed_extents
[1];
4811 unpin
= &fs_info
->freed_extents
[0];
4814 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
4819 if (btrfs_test_opt(root
, DISCARD
))
4820 ret
= btrfs_discard_extent(root
, start
,
4821 end
+ 1 - start
, NULL
);
4823 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
4824 unpin_extent_range(root
, start
, end
);
4831 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
4832 struct btrfs_root
*root
,
4833 u64 bytenr
, u64 num_bytes
, u64 parent
,
4834 u64 root_objectid
, u64 owner_objectid
,
4835 u64 owner_offset
, int refs_to_drop
,
4836 struct btrfs_delayed_extent_op
*extent_op
)
4838 struct btrfs_key key
;
4839 struct btrfs_path
*path
;
4840 struct btrfs_fs_info
*info
= root
->fs_info
;
4841 struct btrfs_root
*extent_root
= info
->extent_root
;
4842 struct extent_buffer
*leaf
;
4843 struct btrfs_extent_item
*ei
;
4844 struct btrfs_extent_inline_ref
*iref
;
4847 int extent_slot
= 0;
4848 int found_extent
= 0;
4853 path
= btrfs_alloc_path();
4858 path
->leave_spinning
= 1;
4860 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
4861 BUG_ON(!is_data
&& refs_to_drop
!= 1);
4863 ret
= lookup_extent_backref(trans
, extent_root
, path
, &iref
,
4864 bytenr
, num_bytes
, parent
,
4865 root_objectid
, owner_objectid
,
4868 extent_slot
= path
->slots
[0];
4869 while (extent_slot
>= 0) {
4870 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
4872 if (key
.objectid
!= bytenr
)
4874 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
4875 key
.offset
== num_bytes
) {
4879 if (path
->slots
[0] - extent_slot
> 5)
4883 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4884 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
4885 if (found_extent
&& item_size
< sizeof(*ei
))
4888 if (!found_extent
) {
4890 ret
= remove_extent_backref(trans
, extent_root
, path
,
4894 btrfs_release_path(path
);
4895 path
->leave_spinning
= 1;
4897 key
.objectid
= bytenr
;
4898 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
4899 key
.offset
= num_bytes
;
4901 ret
= btrfs_search_slot(trans
, extent_root
,
4904 printk(KERN_ERR
"umm, got %d back from search"
4905 ", was looking for %llu\n", ret
,
4906 (unsigned long long)bytenr
);
4908 btrfs_print_leaf(extent_root
,
4912 extent_slot
= path
->slots
[0];
4915 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
4917 printk(KERN_ERR
"btrfs unable to find ref byte nr %llu "
4918 "parent %llu root %llu owner %llu offset %llu\n",
4919 (unsigned long long)bytenr
,
4920 (unsigned long long)parent
,
4921 (unsigned long long)root_objectid
,
4922 (unsigned long long)owner_objectid
,
4923 (unsigned long long)owner_offset
);
4926 leaf
= path
->nodes
[0];
4927 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
4928 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4929 if (item_size
< sizeof(*ei
)) {
4930 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
4931 ret
= convert_extent_item_v0(trans
, extent_root
, path
,
4935 btrfs_release_path(path
);
4936 path
->leave_spinning
= 1;
4938 key
.objectid
= bytenr
;
4939 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
4940 key
.offset
= num_bytes
;
4942 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
4945 printk(KERN_ERR
"umm, got %d back from search"
4946 ", was looking for %llu\n", ret
,
4947 (unsigned long long)bytenr
);
4948 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
4951 extent_slot
= path
->slots
[0];
4952 leaf
= path
->nodes
[0];
4953 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
4956 BUG_ON(item_size
< sizeof(*ei
));
4957 ei
= btrfs_item_ptr(leaf
, extent_slot
,
4958 struct btrfs_extent_item
);
4959 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
) {
4960 struct btrfs_tree_block_info
*bi
;
4961 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
4962 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
4963 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
4966 refs
= btrfs_extent_refs(leaf
, ei
);
4967 BUG_ON(refs
< refs_to_drop
);
4968 refs
-= refs_to_drop
;
4972 __run_delayed_extent_op(extent_op
, leaf
, ei
);
4974 * In the case of inline back ref, reference count will
4975 * be updated by remove_extent_backref
4978 BUG_ON(!found_extent
);
4980 btrfs_set_extent_refs(leaf
, ei
, refs
);
4981 btrfs_mark_buffer_dirty(leaf
);
4984 ret
= remove_extent_backref(trans
, extent_root
, path
,
4991 BUG_ON(is_data
&& refs_to_drop
!=
4992 extent_data_ref_count(root
, path
, iref
));
4994 BUG_ON(path
->slots
[0] != extent_slot
);
4996 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
4997 path
->slots
[0] = extent_slot
;
5002 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
5005 btrfs_release_path(path
);
5008 ret
= btrfs_del_csums(trans
, root
, bytenr
, num_bytes
);
5011 invalidate_mapping_pages(info
->btree_inode
->i_mapping
,
5012 bytenr
>> PAGE_CACHE_SHIFT
,
5013 (bytenr
+ num_bytes
- 1) >> PAGE_CACHE_SHIFT
);
5016 ret
= update_block_group(trans
, root
, bytenr
, num_bytes
, 0);
5019 btrfs_free_path(path
);
5024 * when we free an block, it is possible (and likely) that we free the last
5025 * delayed ref for that extent as well. This searches the delayed ref tree for
5026 * a given extent, and if there are no other delayed refs to be processed, it
5027 * removes it from the tree.
5029 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
5030 struct btrfs_root
*root
, u64 bytenr
)
5032 struct btrfs_delayed_ref_head
*head
;
5033 struct btrfs_delayed_ref_root
*delayed_refs
;
5034 struct btrfs_delayed_ref_node
*ref
;
5035 struct rb_node
*node
;
5038 delayed_refs
= &trans
->transaction
->delayed_refs
;
5039 spin_lock(&delayed_refs
->lock
);
5040 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
5044 node
= rb_prev(&head
->node
.rb_node
);
5048 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
5050 /* there are still entries for this ref, we can't drop it */
5051 if (ref
->bytenr
== bytenr
)
5054 if (head
->extent_op
) {
5055 if (!head
->must_insert_reserved
)
5057 kfree(head
->extent_op
);
5058 head
->extent_op
= NULL
;
5062 * waiting for the lock here would deadlock. If someone else has it
5063 * locked they are already in the process of dropping it anyway
5065 if (!mutex_trylock(&head
->mutex
))
5069 * at this point we have a head with no other entries. Go
5070 * ahead and process it.
5072 head
->node
.in_tree
= 0;
5073 rb_erase(&head
->node
.rb_node
, &delayed_refs
->root
);
5075 delayed_refs
->num_entries
--;
5076 if (waitqueue_active(&delayed_refs
->seq_wait
))
5077 wake_up(&delayed_refs
->seq_wait
);
5080 * we don't take a ref on the node because we're removing it from the
5081 * tree, so we just steal the ref the tree was holding.
5083 delayed_refs
->num_heads
--;
5084 if (list_empty(&head
->cluster
))
5085 delayed_refs
->num_heads_ready
--;
5087 list_del_init(&head
->cluster
);
5088 spin_unlock(&delayed_refs
->lock
);
5090 BUG_ON(head
->extent_op
);
5091 if (head
->must_insert_reserved
)
5094 mutex_unlock(&head
->mutex
);
5095 btrfs_put_delayed_ref(&head
->node
);
5098 spin_unlock(&delayed_refs
->lock
);
5102 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
5103 struct btrfs_root
*root
,
5104 struct extent_buffer
*buf
,
5105 u64 parent
, int last_ref
, int for_cow
)
5107 struct btrfs_block_group_cache
*cache
= NULL
;
5110 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
5111 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
5112 buf
->start
, buf
->len
,
5113 parent
, root
->root_key
.objectid
,
5114 btrfs_header_level(buf
),
5115 BTRFS_DROP_DELAYED_REF
, NULL
, for_cow
);
5122 cache
= btrfs_lookup_block_group(root
->fs_info
, buf
->start
);
5124 if (btrfs_header_generation(buf
) == trans
->transid
) {
5125 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
5126 ret
= check_ref_cleanup(trans
, root
, buf
->start
);
5131 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
5132 pin_down_extent(root
, cache
, buf
->start
, buf
->len
, 1);
5136 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
5138 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
5139 btrfs_update_reserved_bytes(cache
, buf
->len
, RESERVE_FREE
);
5143 * Deleting the buffer, clear the corrupt flag since it doesn't matter
5146 clear_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
);
5147 btrfs_put_block_group(cache
);
5150 int btrfs_free_extent(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
5151 u64 bytenr
, u64 num_bytes
, u64 parent
, u64 root_objectid
,
5152 u64 owner
, u64 offset
, int for_cow
)
5155 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5158 * tree log blocks never actually go into the extent allocation
5159 * tree, just update pinning info and exit early.
5161 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
5162 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
5163 /* unlocks the pinned mutex */
5164 btrfs_pin_extent(root
, bytenr
, num_bytes
, 1);
5166 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
5167 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
5169 parent
, root_objectid
, (int)owner
,
5170 BTRFS_DROP_DELAYED_REF
, NULL
, for_cow
);
5173 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
5175 parent
, root_objectid
, owner
,
5176 offset
, BTRFS_DROP_DELAYED_REF
,
5183 static u64
stripe_align(struct btrfs_root
*root
, u64 val
)
5185 u64 mask
= ((u64
)root
->stripesize
- 1);
5186 u64 ret
= (val
+ mask
) & ~mask
;
5191 * when we wait for progress in the block group caching, its because
5192 * our allocation attempt failed at least once. So, we must sleep
5193 * and let some progress happen before we try again.
5195 * This function will sleep at least once waiting for new free space to
5196 * show up, and then it will check the block group free space numbers
5197 * for our min num_bytes. Another option is to have it go ahead
5198 * and look in the rbtree for a free extent of a given size, but this
5202 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
5205 struct btrfs_caching_control
*caching_ctl
;
5208 caching_ctl
= get_caching_control(cache
);
5212 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
5213 (cache
->free_space_ctl
->free_space
>= num_bytes
));
5215 put_caching_control(caching_ctl
);
5220 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
5222 struct btrfs_caching_control
*caching_ctl
;
5225 caching_ctl
= get_caching_control(cache
);
5229 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
5231 put_caching_control(caching_ctl
);
5235 static int get_block_group_index(struct btrfs_block_group_cache
*cache
)
5238 if (cache
->flags
& BTRFS_BLOCK_GROUP_RAID10
)
5240 else if (cache
->flags
& BTRFS_BLOCK_GROUP_RAID1
)
5242 else if (cache
->flags
& BTRFS_BLOCK_GROUP_DUP
)
5244 else if (cache
->flags
& BTRFS_BLOCK_GROUP_RAID0
)
5251 enum btrfs_loop_type
{
5252 LOOP_FIND_IDEAL
= 0,
5253 LOOP_CACHING_NOWAIT
= 1,
5254 LOOP_CACHING_WAIT
= 2,
5255 LOOP_ALLOC_CHUNK
= 3,
5256 LOOP_NO_EMPTY_SIZE
= 4,
5260 * walks the btree of allocated extents and find a hole of a given size.
5261 * The key ins is changed to record the hole:
5262 * ins->objectid == block start
5263 * ins->flags = BTRFS_EXTENT_ITEM_KEY
5264 * ins->offset == number of blocks
5265 * Any available blocks before search_start are skipped.
5267 static noinline
int find_free_extent(struct btrfs_trans_handle
*trans
,
5268 struct btrfs_root
*orig_root
,
5269 u64 num_bytes
, u64 empty_size
,
5270 u64 search_start
, u64 search_end
,
5271 u64 hint_byte
, struct btrfs_key
*ins
,
5275 struct btrfs_root
*root
= orig_root
->fs_info
->extent_root
;
5276 struct btrfs_free_cluster
*last_ptr
= NULL
;
5277 struct btrfs_block_group_cache
*block_group
= NULL
;
5278 struct btrfs_block_group_cache
*used_block_group
;
5279 int empty_cluster
= 2 * 1024 * 1024;
5280 int allowed_chunk_alloc
= 0;
5281 int done_chunk_alloc
= 0;
5282 struct btrfs_space_info
*space_info
;
5285 int alloc_type
= (data
& BTRFS_BLOCK_GROUP_DATA
) ?
5286 RESERVE_ALLOC_NO_ACCOUNT
: RESERVE_ALLOC
;
5287 bool found_uncached_bg
= false;
5288 bool failed_cluster_refill
= false;
5289 bool failed_alloc
= false;
5290 bool use_cluster
= true;
5291 bool have_caching_bg
= false;
5292 u64 ideal_cache_percent
= 0;
5293 u64 ideal_cache_offset
= 0;
5295 WARN_ON(num_bytes
< root
->sectorsize
);
5296 btrfs_set_key_type(ins
, BTRFS_EXTENT_ITEM_KEY
);
5300 trace_find_free_extent(orig_root
, num_bytes
, empty_size
, data
);
5302 space_info
= __find_space_info(root
->fs_info
, data
);
5304 printk(KERN_ERR
"No space info for %llu\n", data
);
5309 * If the space info is for both data and metadata it means we have a
5310 * small filesystem and we can't use the clustering stuff.
5312 if (btrfs_mixed_space_info(space_info
))
5313 use_cluster
= false;
5315 if (orig_root
->ref_cows
|| empty_size
)
5316 allowed_chunk_alloc
= 1;
5318 if (data
& BTRFS_BLOCK_GROUP_METADATA
&& use_cluster
) {
5319 last_ptr
= &root
->fs_info
->meta_alloc_cluster
;
5320 if (!btrfs_test_opt(root
, SSD
))
5321 empty_cluster
= 64 * 1024;
5324 if ((data
& BTRFS_BLOCK_GROUP_DATA
) && use_cluster
&&
5325 btrfs_test_opt(root
, SSD
)) {
5326 last_ptr
= &root
->fs_info
->data_alloc_cluster
;
5330 spin_lock(&last_ptr
->lock
);
5331 if (last_ptr
->block_group
)
5332 hint_byte
= last_ptr
->window_start
;
5333 spin_unlock(&last_ptr
->lock
);
5336 search_start
= max(search_start
, first_logical_byte(root
, 0));
5337 search_start
= max(search_start
, hint_byte
);
5342 if (search_start
== hint_byte
) {
5344 block_group
= btrfs_lookup_block_group(root
->fs_info
,
5346 used_block_group
= block_group
;
5348 * we don't want to use the block group if it doesn't match our
5349 * allocation bits, or if its not cached.
5351 * However if we are re-searching with an ideal block group
5352 * picked out then we don't care that the block group is cached.
5354 if (block_group
&& block_group_bits(block_group
, data
) &&
5355 (block_group
->cached
!= BTRFS_CACHE_NO
||
5356 search_start
== ideal_cache_offset
)) {
5357 down_read(&space_info
->groups_sem
);
5358 if (list_empty(&block_group
->list
) ||
5361 * someone is removing this block group,
5362 * we can't jump into the have_block_group
5363 * target because our list pointers are not
5366 btrfs_put_block_group(block_group
);
5367 up_read(&space_info
->groups_sem
);
5369 index
= get_block_group_index(block_group
);
5370 goto have_block_group
;
5372 } else if (block_group
) {
5373 btrfs_put_block_group(block_group
);
5377 have_caching_bg
= false;
5378 down_read(&space_info
->groups_sem
);
5379 list_for_each_entry(block_group
, &space_info
->block_groups
[index
],
5384 used_block_group
= block_group
;
5385 btrfs_get_block_group(block_group
);
5386 search_start
= block_group
->key
.objectid
;
5389 * this can happen if we end up cycling through all the
5390 * raid types, but we want to make sure we only allocate
5391 * for the proper type.
5393 if (!block_group_bits(block_group
, data
)) {
5394 u64 extra
= BTRFS_BLOCK_GROUP_DUP
|
5395 BTRFS_BLOCK_GROUP_RAID1
|
5396 BTRFS_BLOCK_GROUP_RAID10
;
5399 * if they asked for extra copies and this block group
5400 * doesn't provide them, bail. This does allow us to
5401 * fill raid0 from raid1.
5403 if ((data
& extra
) && !(block_group
->flags
& extra
))
5408 cached
= block_group_cache_done(block_group
);
5409 if (unlikely(!cached
)) {
5412 found_uncached_bg
= true;
5413 ret
= cache_block_group(block_group
, trans
,
5415 if (block_group
->cached
== BTRFS_CACHE_FINISHED
)
5418 free_percent
= btrfs_block_group_used(&block_group
->item
);
5419 free_percent
*= 100;
5420 free_percent
= div64_u64(free_percent
,
5421 block_group
->key
.offset
);
5422 free_percent
= 100 - free_percent
;
5423 if (free_percent
> ideal_cache_percent
&&
5424 likely(!block_group
->ro
)) {
5425 ideal_cache_offset
= block_group
->key
.objectid
;
5426 ideal_cache_percent
= free_percent
;
5430 * The caching workers are limited to 2 threads, so we
5431 * can queue as much work as we care to.
5433 if (loop
> LOOP_FIND_IDEAL
) {
5434 ret
= cache_block_group(block_group
, trans
,
5440 * If loop is set for cached only, try the next block
5443 if (loop
== LOOP_FIND_IDEAL
)
5448 if (unlikely(block_group
->ro
))
5452 * Ok we want to try and use the cluster allocator, so
5457 * the refill lock keeps out other
5458 * people trying to start a new cluster
5460 spin_lock(&last_ptr
->refill_lock
);
5461 used_block_group
= last_ptr
->block_group
;
5462 if (used_block_group
!= block_group
&&
5463 (!used_block_group
||
5464 used_block_group
->ro
||
5465 !block_group_bits(used_block_group
, data
))) {
5466 used_block_group
= block_group
;
5467 goto refill_cluster
;
5470 if (used_block_group
!= block_group
)
5471 btrfs_get_block_group(used_block_group
);
5473 offset
= btrfs_alloc_from_cluster(used_block_group
,
5474 last_ptr
, num_bytes
, used_block_group
->key
.objectid
);
5476 /* we have a block, we're done */
5477 spin_unlock(&last_ptr
->refill_lock
);
5478 trace_btrfs_reserve_extent_cluster(root
,
5479 block_group
, search_start
, num_bytes
);
5483 WARN_ON(last_ptr
->block_group
!= used_block_group
);
5484 if (used_block_group
!= block_group
) {
5485 btrfs_put_block_group(used_block_group
);
5486 used_block_group
= block_group
;
5489 BUG_ON(used_block_group
!= block_group
);
5490 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
5491 * set up a new clusters, so lets just skip it
5492 * and let the allocator find whatever block
5493 * it can find. If we reach this point, we
5494 * will have tried the cluster allocator
5495 * plenty of times and not have found
5496 * anything, so we are likely way too
5497 * fragmented for the clustering stuff to find
5500 * However, if the cluster is taken from the
5501 * current block group, release the cluster
5502 * first, so that we stand a better chance of
5503 * succeeding in the unclustered
5505 if (loop
>= LOOP_NO_EMPTY_SIZE
&&
5506 last_ptr
->block_group
!= block_group
) {
5507 spin_unlock(&last_ptr
->refill_lock
);
5508 goto unclustered_alloc
;
5512 * this cluster didn't work out, free it and
5515 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
5517 if (loop
>= LOOP_NO_EMPTY_SIZE
) {
5518 spin_unlock(&last_ptr
->refill_lock
);
5519 goto unclustered_alloc
;
5522 /* allocate a cluster in this block group */
5523 ret
= btrfs_find_space_cluster(trans
, root
,
5524 block_group
, last_ptr
,
5525 search_start
, num_bytes
,
5526 empty_cluster
+ empty_size
);
5529 * now pull our allocation out of this
5532 offset
= btrfs_alloc_from_cluster(block_group
,
5533 last_ptr
, num_bytes
,
5536 /* we found one, proceed */
5537 spin_unlock(&last_ptr
->refill_lock
);
5538 trace_btrfs_reserve_extent_cluster(root
,
5539 block_group
, search_start
,
5543 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
5544 && !failed_cluster_refill
) {
5545 spin_unlock(&last_ptr
->refill_lock
);
5547 failed_cluster_refill
= true;
5548 wait_block_group_cache_progress(block_group
,
5549 num_bytes
+ empty_cluster
+ empty_size
);
5550 goto have_block_group
;
5554 * at this point we either didn't find a cluster
5555 * or we weren't able to allocate a block from our
5556 * cluster. Free the cluster we've been trying
5557 * to use, and go to the next block group
5559 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
5560 spin_unlock(&last_ptr
->refill_lock
);
5565 spin_lock(&block_group
->free_space_ctl
->tree_lock
);
5567 block_group
->free_space_ctl
->free_space
<
5568 num_bytes
+ empty_cluster
+ empty_size
) {
5569 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
5572 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
5574 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
5575 num_bytes
, empty_size
);
5577 * If we didn't find a chunk, and we haven't failed on this
5578 * block group before, and this block group is in the middle of
5579 * caching and we are ok with waiting, then go ahead and wait
5580 * for progress to be made, and set failed_alloc to true.
5582 * If failed_alloc is true then we've already waited on this
5583 * block group once and should move on to the next block group.
5585 if (!offset
&& !failed_alloc
&& !cached
&&
5586 loop
> LOOP_CACHING_NOWAIT
) {
5587 wait_block_group_cache_progress(block_group
,
5588 num_bytes
+ empty_size
);
5589 failed_alloc
= true;
5590 goto have_block_group
;
5591 } else if (!offset
) {
5593 have_caching_bg
= true;
5597 search_start
= stripe_align(root
, offset
);
5598 /* move on to the next group */
5599 if (search_start
+ num_bytes
>= search_end
) {
5600 btrfs_add_free_space(used_block_group
, offset
, num_bytes
);
5604 /* move on to the next group */
5605 if (search_start
+ num_bytes
>
5606 used_block_group
->key
.objectid
+ used_block_group
->key
.offset
) {
5607 btrfs_add_free_space(used_block_group
, offset
, num_bytes
);
5611 if (offset
< search_start
)
5612 btrfs_add_free_space(used_block_group
, offset
,
5613 search_start
- offset
);
5614 BUG_ON(offset
> search_start
);
5616 ret
= btrfs_update_reserved_bytes(used_block_group
, num_bytes
,
5618 if (ret
== -EAGAIN
) {
5619 btrfs_add_free_space(used_block_group
, offset
, num_bytes
);
5623 /* we are all good, lets return */
5624 ins
->objectid
= search_start
;
5625 ins
->offset
= num_bytes
;
5627 trace_btrfs_reserve_extent(orig_root
, block_group
,
5628 search_start
, num_bytes
);
5629 if (offset
< search_start
)
5630 btrfs_add_free_space(used_block_group
, offset
,
5631 search_start
- offset
);
5632 BUG_ON(offset
> search_start
);
5633 if (used_block_group
!= block_group
)
5634 btrfs_put_block_group(used_block_group
);
5635 btrfs_put_block_group(block_group
);
5638 failed_cluster_refill
= false;
5639 failed_alloc
= false;
5640 BUG_ON(index
!= get_block_group_index(block_group
));
5641 if (used_block_group
!= block_group
)
5642 btrfs_put_block_group(used_block_group
);
5643 btrfs_put_block_group(block_group
);
5645 up_read(&space_info
->groups_sem
);
5647 if (!ins
->objectid
&& loop
>= LOOP_CACHING_WAIT
&& have_caching_bg
)
5650 if (!ins
->objectid
&& ++index
< BTRFS_NR_RAID_TYPES
)
5653 /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5654 * for them to make caching progress. Also
5655 * determine the best possible bg to cache
5656 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5657 * caching kthreads as we move along
5658 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5659 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5660 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5663 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
) {
5665 if (loop
== LOOP_FIND_IDEAL
&& found_uncached_bg
) {
5666 found_uncached_bg
= false;
5668 if (!ideal_cache_percent
)
5672 * 1 of the following 2 things have happened so far
5674 * 1) We found an ideal block group for caching that
5675 * is mostly full and will cache quickly, so we might
5676 * as well wait for it.
5678 * 2) We searched for cached only and we didn't find
5679 * anything, and we didn't start any caching kthreads
5680 * either, so chances are we will loop through and
5681 * start a couple caching kthreads, and then come back
5682 * around and just wait for them. This will be slower
5683 * because we will have 2 caching kthreads reading at
5684 * the same time when we could have just started one
5685 * and waited for it to get far enough to give us an
5686 * allocation, so go ahead and go to the wait caching
5689 loop
= LOOP_CACHING_WAIT
;
5690 search_start
= ideal_cache_offset
;
5691 ideal_cache_percent
= 0;
5693 } else if (loop
== LOOP_FIND_IDEAL
) {
5695 * Didn't find a uncached bg, wait on anything we find
5698 loop
= LOOP_CACHING_WAIT
;
5704 if (loop
== LOOP_ALLOC_CHUNK
) {
5705 if (allowed_chunk_alloc
) {
5706 ret
= do_chunk_alloc(trans
, root
, num_bytes
+
5707 2 * 1024 * 1024, data
,
5708 CHUNK_ALLOC_LIMITED
);
5709 allowed_chunk_alloc
= 0;
5711 done_chunk_alloc
= 1;
5712 } else if (!done_chunk_alloc
&&
5713 space_info
->force_alloc
==
5714 CHUNK_ALLOC_NO_FORCE
) {
5715 space_info
->force_alloc
= CHUNK_ALLOC_LIMITED
;
5719 * We didn't allocate a chunk, go ahead and drop the
5720 * empty size and loop again.
5722 if (!done_chunk_alloc
)
5723 loop
= LOOP_NO_EMPTY_SIZE
;
5726 if (loop
== LOOP_NO_EMPTY_SIZE
) {
5732 } else if (!ins
->objectid
) {
5734 } else if (ins
->objectid
) {
5741 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
5742 int dump_block_groups
)
5744 struct btrfs_block_group_cache
*cache
;
5747 spin_lock(&info
->lock
);
5748 printk(KERN_INFO
"space_info %llu has %llu free, is %sfull\n",
5749 (unsigned long long)info
->flags
,
5750 (unsigned long long)(info
->total_bytes
- info
->bytes_used
-
5751 info
->bytes_pinned
- info
->bytes_reserved
-
5752 info
->bytes_readonly
),
5753 (info
->full
) ? "" : "not ");
5754 printk(KERN_INFO
"space_info total=%llu, used=%llu, pinned=%llu, "
5755 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5756 (unsigned long long)info
->total_bytes
,
5757 (unsigned long long)info
->bytes_used
,
5758 (unsigned long long)info
->bytes_pinned
,
5759 (unsigned long long)info
->bytes_reserved
,
5760 (unsigned long long)info
->bytes_may_use
,
5761 (unsigned long long)info
->bytes_readonly
);
5762 spin_unlock(&info
->lock
);
5764 if (!dump_block_groups
)
5767 down_read(&info
->groups_sem
);
5769 list_for_each_entry(cache
, &info
->block_groups
[index
], list
) {
5770 spin_lock(&cache
->lock
);
5771 printk(KERN_INFO
"block group %llu has %llu bytes, %llu used "
5772 "%llu pinned %llu reserved\n",
5773 (unsigned long long)cache
->key
.objectid
,
5774 (unsigned long long)cache
->key
.offset
,
5775 (unsigned long long)btrfs_block_group_used(&cache
->item
),
5776 (unsigned long long)cache
->pinned
,
5777 (unsigned long long)cache
->reserved
);
5778 btrfs_dump_free_space(cache
, bytes
);
5779 spin_unlock(&cache
->lock
);
5781 if (++index
< BTRFS_NR_RAID_TYPES
)
5783 up_read(&info
->groups_sem
);
5786 int btrfs_reserve_extent(struct btrfs_trans_handle
*trans
,
5787 struct btrfs_root
*root
,
5788 u64 num_bytes
, u64 min_alloc_size
,
5789 u64 empty_size
, u64 hint_byte
,
5790 u64 search_end
, struct btrfs_key
*ins
,
5794 u64 search_start
= 0;
5796 data
= btrfs_get_alloc_profile(root
, data
);
5799 * the only place that sets empty_size is btrfs_realloc_node, which
5800 * is not called recursively on allocations
5802 if (empty_size
|| root
->ref_cows
)
5803 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
5804 num_bytes
+ 2 * 1024 * 1024, data
,
5805 CHUNK_ALLOC_NO_FORCE
);
5807 WARN_ON(num_bytes
< root
->sectorsize
);
5808 ret
= find_free_extent(trans
, root
, num_bytes
, empty_size
,
5809 search_start
, search_end
, hint_byte
,
5812 if (ret
== -ENOSPC
&& num_bytes
> min_alloc_size
) {
5813 num_bytes
= num_bytes
>> 1;
5814 num_bytes
= num_bytes
& ~(root
->sectorsize
- 1);
5815 num_bytes
= max(num_bytes
, min_alloc_size
);
5816 do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
5817 num_bytes
, data
, CHUNK_ALLOC_FORCE
);
5820 if (ret
== -ENOSPC
&& btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
5821 struct btrfs_space_info
*sinfo
;
5823 sinfo
= __find_space_info(root
->fs_info
, data
);
5824 printk(KERN_ERR
"btrfs allocation failed flags %llu, "
5825 "wanted %llu\n", (unsigned long long)data
,
5826 (unsigned long long)num_bytes
);
5827 dump_space_info(sinfo
, num_bytes
, 1);
5830 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, ins
->offset
);
5835 static int __btrfs_free_reserved_extent(struct btrfs_root
*root
,
5836 u64 start
, u64 len
, int pin
)
5838 struct btrfs_block_group_cache
*cache
;
5841 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
5843 printk(KERN_ERR
"Unable to find block group for %llu\n",
5844 (unsigned long long)start
);
5848 if (btrfs_test_opt(root
, DISCARD
))
5849 ret
= btrfs_discard_extent(root
, start
, len
, NULL
);
5852 pin_down_extent(root
, cache
, start
, len
, 1);
5854 btrfs_add_free_space(cache
, start
, len
);
5855 btrfs_update_reserved_bytes(cache
, len
, RESERVE_FREE
);
5857 btrfs_put_block_group(cache
);
5859 trace_btrfs_reserved_extent_free(root
, start
, len
);
5864 int btrfs_free_reserved_extent(struct btrfs_root
*root
,
5867 return __btrfs_free_reserved_extent(root
, start
, len
, 0);
5870 int btrfs_free_and_pin_reserved_extent(struct btrfs_root
*root
,
5873 return __btrfs_free_reserved_extent(root
, start
, len
, 1);
5876 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
5877 struct btrfs_root
*root
,
5878 u64 parent
, u64 root_objectid
,
5879 u64 flags
, u64 owner
, u64 offset
,
5880 struct btrfs_key
*ins
, int ref_mod
)
5883 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5884 struct btrfs_extent_item
*extent_item
;
5885 struct btrfs_extent_inline_ref
*iref
;
5886 struct btrfs_path
*path
;
5887 struct extent_buffer
*leaf
;
5892 type
= BTRFS_SHARED_DATA_REF_KEY
;
5894 type
= BTRFS_EXTENT_DATA_REF_KEY
;
5896 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
5898 path
= btrfs_alloc_path();
5902 path
->leave_spinning
= 1;
5903 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
5907 leaf
= path
->nodes
[0];
5908 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
5909 struct btrfs_extent_item
);
5910 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
5911 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
5912 btrfs_set_extent_flags(leaf
, extent_item
,
5913 flags
| BTRFS_EXTENT_FLAG_DATA
);
5915 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
5916 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
5918 struct btrfs_shared_data_ref
*ref
;
5919 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
5920 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
5921 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
5923 struct btrfs_extent_data_ref
*ref
;
5924 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
5925 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
5926 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
5927 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
5928 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
5931 btrfs_mark_buffer_dirty(path
->nodes
[0]);
5932 btrfs_free_path(path
);
5934 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1);
5936 printk(KERN_ERR
"btrfs update block group failed for %llu "
5937 "%llu\n", (unsigned long long)ins
->objectid
,
5938 (unsigned long long)ins
->offset
);
5944 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
5945 struct btrfs_root
*root
,
5946 u64 parent
, u64 root_objectid
,
5947 u64 flags
, struct btrfs_disk_key
*key
,
5948 int level
, struct btrfs_key
*ins
)
5951 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5952 struct btrfs_extent_item
*extent_item
;
5953 struct btrfs_tree_block_info
*block_info
;
5954 struct btrfs_extent_inline_ref
*iref
;
5955 struct btrfs_path
*path
;
5956 struct extent_buffer
*leaf
;
5957 u32 size
= sizeof(*extent_item
) + sizeof(*block_info
) + sizeof(*iref
);
5959 path
= btrfs_alloc_path();
5963 path
->leave_spinning
= 1;
5964 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
5968 leaf
= path
->nodes
[0];
5969 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
5970 struct btrfs_extent_item
);
5971 btrfs_set_extent_refs(leaf
, extent_item
, 1);
5972 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
5973 btrfs_set_extent_flags(leaf
, extent_item
,
5974 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
5975 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
5977 btrfs_set_tree_block_key(leaf
, block_info
, key
);
5978 btrfs_set_tree_block_level(leaf
, block_info
, level
);
5980 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
5982 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
5983 btrfs_set_extent_inline_ref_type(leaf
, iref
,
5984 BTRFS_SHARED_BLOCK_REF_KEY
);
5985 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
5987 btrfs_set_extent_inline_ref_type(leaf
, iref
,
5988 BTRFS_TREE_BLOCK_REF_KEY
);
5989 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
5992 btrfs_mark_buffer_dirty(leaf
);
5993 btrfs_free_path(path
);
5995 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1);
5997 printk(KERN_ERR
"btrfs update block group failed for %llu "
5998 "%llu\n", (unsigned long long)ins
->objectid
,
5999 (unsigned long long)ins
->offset
);
6005 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
6006 struct btrfs_root
*root
,
6007 u64 root_objectid
, u64 owner
,
6008 u64 offset
, struct btrfs_key
*ins
)
6012 BUG_ON(root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
6014 ret
= btrfs_add_delayed_data_ref(root
->fs_info
, trans
, ins
->objectid
,
6016 root_objectid
, owner
, offset
,
6017 BTRFS_ADD_DELAYED_EXTENT
, NULL
, 0);
6022 * this is used by the tree logging recovery code. It records that
6023 * an extent has been allocated and makes sure to clear the free
6024 * space cache bits as well
6026 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
6027 struct btrfs_root
*root
,
6028 u64 root_objectid
, u64 owner
, u64 offset
,
6029 struct btrfs_key
*ins
)
6032 struct btrfs_block_group_cache
*block_group
;
6033 struct btrfs_caching_control
*caching_ctl
;
6034 u64 start
= ins
->objectid
;
6035 u64 num_bytes
= ins
->offset
;
6037 block_group
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
6038 cache_block_group(block_group
, trans
, NULL
, 0);
6039 caching_ctl
= get_caching_control(block_group
);
6042 BUG_ON(!block_group_cache_done(block_group
));
6043 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
6046 mutex_lock(&caching_ctl
->mutex
);
6048 if (start
>= caching_ctl
->progress
) {
6049 ret
= add_excluded_extent(root
, start
, num_bytes
);
6051 } else if (start
+ num_bytes
<= caching_ctl
->progress
) {
6052 ret
= btrfs_remove_free_space(block_group
,
6056 num_bytes
= caching_ctl
->progress
- start
;
6057 ret
= btrfs_remove_free_space(block_group
,
6061 start
= caching_ctl
->progress
;
6062 num_bytes
= ins
->objectid
+ ins
->offset
-
6063 caching_ctl
->progress
;
6064 ret
= add_excluded_extent(root
, start
, num_bytes
);
6068 mutex_unlock(&caching_ctl
->mutex
);
6069 put_caching_control(caching_ctl
);
6072 ret
= btrfs_update_reserved_bytes(block_group
, ins
->offset
,
6073 RESERVE_ALLOC_NO_ACCOUNT
);
6075 btrfs_put_block_group(block_group
);
6076 ret
= alloc_reserved_file_extent(trans
, root
, 0, root_objectid
,
6077 0, owner
, offset
, ins
, 1);
6081 struct extent_buffer
*btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
,
6082 struct btrfs_root
*root
,
6083 u64 bytenr
, u32 blocksize
,
6086 struct extent_buffer
*buf
;
6088 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
6090 return ERR_PTR(-ENOMEM
);
6091 btrfs_set_header_generation(buf
, trans
->transid
);
6092 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, buf
, level
);
6093 btrfs_tree_lock(buf
);
6094 clean_tree_block(trans
, root
, buf
);
6096 btrfs_set_lock_blocking(buf
);
6097 btrfs_set_buffer_uptodate(buf
);
6099 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
6101 * we allow two log transactions at a time, use different
6102 * EXENT bit to differentiate dirty pages.
6104 if (root
->log_transid
% 2 == 0)
6105 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
6106 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
6108 set_extent_new(&root
->dirty_log_pages
, buf
->start
,
6109 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
6111 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
6112 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
6114 trans
->blocks_used
++;
6115 /* this returns a buffer locked for blocking */
6119 static struct btrfs_block_rsv
*
6120 use_block_rsv(struct btrfs_trans_handle
*trans
,
6121 struct btrfs_root
*root
, u32 blocksize
)
6123 struct btrfs_block_rsv
*block_rsv
;
6124 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
6127 block_rsv
= get_block_rsv(trans
, root
);
6129 if (block_rsv
->size
== 0) {
6130 ret
= reserve_metadata_bytes(root
, block_rsv
, blocksize
, 0);
6132 * If we couldn't reserve metadata bytes try and use some from
6133 * the global reserve.
6135 if (ret
&& block_rsv
!= global_rsv
) {
6136 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
6139 return ERR_PTR(ret
);
6141 return ERR_PTR(ret
);
6146 ret
= block_rsv_use_bytes(block_rsv
, blocksize
);
6150 static DEFINE_RATELIMIT_STATE(_rs
,
6151 DEFAULT_RATELIMIT_INTERVAL
,
6152 /*DEFAULT_RATELIMIT_BURST*/ 2);
6153 if (__ratelimit(&_rs
)) {
6154 printk(KERN_DEBUG
"btrfs: block rsv returned %d\n", ret
);
6157 ret
= reserve_metadata_bytes(root
, block_rsv
, blocksize
, 0);
6160 } else if (ret
&& block_rsv
!= global_rsv
) {
6161 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
6167 return ERR_PTR(-ENOSPC
);
6170 static void unuse_block_rsv(struct btrfs_fs_info
*fs_info
,
6171 struct btrfs_block_rsv
*block_rsv
, u32 blocksize
)
6173 block_rsv_add_bytes(block_rsv
, blocksize
, 0);
6174 block_rsv_release_bytes(fs_info
, block_rsv
, NULL
, 0);
6178 * finds a free extent and does all the dirty work required for allocation
6179 * returns the key for the extent through ins, and a tree buffer for
6180 * the first block of the extent through buf.
6182 * returns the tree buffer or NULL.
6184 struct extent_buffer
*btrfs_alloc_free_block(struct btrfs_trans_handle
*trans
,
6185 struct btrfs_root
*root
, u32 blocksize
,
6186 u64 parent
, u64 root_objectid
,
6187 struct btrfs_disk_key
*key
, int level
,
6188 u64 hint
, u64 empty_size
, int for_cow
)
6190 struct btrfs_key ins
;
6191 struct btrfs_block_rsv
*block_rsv
;
6192 struct extent_buffer
*buf
;
6197 block_rsv
= use_block_rsv(trans
, root
, blocksize
);
6198 if (IS_ERR(block_rsv
))
6199 return ERR_CAST(block_rsv
);
6201 ret
= btrfs_reserve_extent(trans
, root
, blocksize
, blocksize
,
6202 empty_size
, hint
, (u64
)-1, &ins
, 0);
6204 unuse_block_rsv(root
->fs_info
, block_rsv
, blocksize
);
6205 return ERR_PTR(ret
);
6208 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
,
6210 BUG_ON(IS_ERR(buf
));
6212 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
6214 parent
= ins
.objectid
;
6215 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
6219 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
6220 struct btrfs_delayed_extent_op
*extent_op
;
6221 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
6224 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
6226 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
6227 extent_op
->flags_to_set
= flags
;
6228 extent_op
->update_key
= 1;
6229 extent_op
->update_flags
= 1;
6230 extent_op
->is_data
= 0;
6232 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
6234 ins
.offset
, parent
, root_objectid
,
6235 level
, BTRFS_ADD_DELAYED_EXTENT
,
6236 extent_op
, for_cow
);
6242 struct walk_control
{
6243 u64 refs
[BTRFS_MAX_LEVEL
];
6244 u64 flags
[BTRFS_MAX_LEVEL
];
6245 struct btrfs_key update_progress
;
6256 #define DROP_REFERENCE 1
6257 #define UPDATE_BACKREF 2
6259 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
6260 struct btrfs_root
*root
,
6261 struct walk_control
*wc
,
6262 struct btrfs_path
*path
)
6270 struct btrfs_key key
;
6271 struct extent_buffer
*eb
;
6276 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
6277 wc
->reada_count
= wc
->reada_count
* 2 / 3;
6278 wc
->reada_count
= max(wc
->reada_count
, 2);
6280 wc
->reada_count
= wc
->reada_count
* 3 / 2;
6281 wc
->reada_count
= min_t(int, wc
->reada_count
,
6282 BTRFS_NODEPTRS_PER_BLOCK(root
));
6285 eb
= path
->nodes
[wc
->level
];
6286 nritems
= btrfs_header_nritems(eb
);
6287 blocksize
= btrfs_level_size(root
, wc
->level
- 1);
6289 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
6290 if (nread
>= wc
->reada_count
)
6294 bytenr
= btrfs_node_blockptr(eb
, slot
);
6295 generation
= btrfs_node_ptr_generation(eb
, slot
);
6297 if (slot
== path
->slots
[wc
->level
])
6300 if (wc
->stage
== UPDATE_BACKREF
&&
6301 generation
<= root
->root_key
.offset
)
6304 /* We don't lock the tree block, it's OK to be racy here */
6305 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, blocksize
,
6310 if (wc
->stage
== DROP_REFERENCE
) {
6314 if (wc
->level
== 1 &&
6315 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
6317 if (!wc
->update_ref
||
6318 generation
<= root
->root_key
.offset
)
6320 btrfs_node_key_to_cpu(eb
, &key
, slot
);
6321 ret
= btrfs_comp_cpu_keys(&key
,
6322 &wc
->update_progress
);
6326 if (wc
->level
== 1 &&
6327 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
6331 ret
= readahead_tree_block(root
, bytenr
, blocksize
,
6337 wc
->reada_slot
= slot
;
6341 * hepler to process tree block while walking down the tree.
6343 * when wc->stage == UPDATE_BACKREF, this function updates
6344 * back refs for pointers in the block.
6346 * NOTE: return value 1 means we should stop walking down.
6348 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
6349 struct btrfs_root
*root
,
6350 struct btrfs_path
*path
,
6351 struct walk_control
*wc
, int lookup_info
)
6353 int level
= wc
->level
;
6354 struct extent_buffer
*eb
= path
->nodes
[level
];
6355 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
6358 if (wc
->stage
== UPDATE_BACKREF
&&
6359 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
6363 * when reference count of tree block is 1, it won't increase
6364 * again. once full backref flag is set, we never clear it.
6367 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
6368 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
6369 BUG_ON(!path
->locks
[level
]);
6370 ret
= btrfs_lookup_extent_info(trans
, root
,
6375 BUG_ON(wc
->refs
[level
] == 0);
6378 if (wc
->stage
== DROP_REFERENCE
) {
6379 if (wc
->refs
[level
] > 1)
6382 if (path
->locks
[level
] && !wc
->keep_locks
) {
6383 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
6384 path
->locks
[level
] = 0;
6389 /* wc->stage == UPDATE_BACKREF */
6390 if (!(wc
->flags
[level
] & flag
)) {
6391 BUG_ON(!path
->locks
[level
]);
6392 ret
= btrfs_inc_ref(trans
, root
, eb
, 1, wc
->for_reloc
);
6394 ret
= btrfs_dec_ref(trans
, root
, eb
, 0, wc
->for_reloc
);
6396 ret
= btrfs_set_disk_extent_flags(trans
, root
, eb
->start
,
6399 wc
->flags
[level
] |= flag
;
6403 * the block is shared by multiple trees, so it's not good to
6404 * keep the tree lock
6406 if (path
->locks
[level
] && level
> 0) {
6407 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
6408 path
->locks
[level
] = 0;
6414 * hepler to process tree block pointer.
6416 * when wc->stage == DROP_REFERENCE, this function checks
6417 * reference count of the block pointed to. if the block
6418 * is shared and we need update back refs for the subtree
6419 * rooted at the block, this function changes wc->stage to
6420 * UPDATE_BACKREF. if the block is shared and there is no
6421 * need to update back, this function drops the reference
6424 * NOTE: return value 1 means we should stop walking down.
6426 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
6427 struct btrfs_root
*root
,
6428 struct btrfs_path
*path
,
6429 struct walk_control
*wc
, int *lookup_info
)
6435 struct btrfs_key key
;
6436 struct extent_buffer
*next
;
6437 int level
= wc
->level
;
6441 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
6442 path
->slots
[level
]);
6444 * if the lower level block was created before the snapshot
6445 * was created, we know there is no need to update back refs
6448 if (wc
->stage
== UPDATE_BACKREF
&&
6449 generation
<= root
->root_key
.offset
) {
6454 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
6455 blocksize
= btrfs_level_size(root
, level
- 1);
6457 next
= btrfs_find_tree_block(root
, bytenr
, blocksize
);
6459 next
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
6464 btrfs_tree_lock(next
);
6465 btrfs_set_lock_blocking(next
);
6467 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, blocksize
,
6468 &wc
->refs
[level
- 1],
6469 &wc
->flags
[level
- 1]);
6471 BUG_ON(wc
->refs
[level
- 1] == 0);
6474 if (wc
->stage
== DROP_REFERENCE
) {
6475 if (wc
->refs
[level
- 1] > 1) {
6477 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
6480 if (!wc
->update_ref
||
6481 generation
<= root
->root_key
.offset
)
6484 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
6485 path
->slots
[level
]);
6486 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
6490 wc
->stage
= UPDATE_BACKREF
;
6491 wc
->shared_level
= level
- 1;
6495 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
6499 if (!btrfs_buffer_uptodate(next
, generation
)) {
6500 btrfs_tree_unlock(next
);
6501 free_extent_buffer(next
);
6507 if (reada
&& level
== 1)
6508 reada_walk_down(trans
, root
, wc
, path
);
6509 next
= read_tree_block(root
, bytenr
, blocksize
, generation
);
6512 btrfs_tree_lock(next
);
6513 btrfs_set_lock_blocking(next
);
6517 BUG_ON(level
!= btrfs_header_level(next
));
6518 path
->nodes
[level
] = next
;
6519 path
->slots
[level
] = 0;
6520 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
6526 wc
->refs
[level
- 1] = 0;
6527 wc
->flags
[level
- 1] = 0;
6528 if (wc
->stage
== DROP_REFERENCE
) {
6529 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
6530 parent
= path
->nodes
[level
]->start
;
6532 BUG_ON(root
->root_key
.objectid
!=
6533 btrfs_header_owner(path
->nodes
[level
]));
6537 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
, parent
,
6538 root
->root_key
.objectid
, level
- 1, 0, 0);
6541 btrfs_tree_unlock(next
);
6542 free_extent_buffer(next
);
6548 * hepler to process tree block while walking up the tree.
6550 * when wc->stage == DROP_REFERENCE, this function drops
6551 * reference count on the block.
6553 * when wc->stage == UPDATE_BACKREF, this function changes
6554 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6555 * to UPDATE_BACKREF previously while processing the block.
6557 * NOTE: return value 1 means we should stop walking up.
6559 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
6560 struct btrfs_root
*root
,
6561 struct btrfs_path
*path
,
6562 struct walk_control
*wc
)
6565 int level
= wc
->level
;
6566 struct extent_buffer
*eb
= path
->nodes
[level
];
6569 if (wc
->stage
== UPDATE_BACKREF
) {
6570 BUG_ON(wc
->shared_level
< level
);
6571 if (level
< wc
->shared_level
)
6574 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
6578 wc
->stage
= DROP_REFERENCE
;
6579 wc
->shared_level
= -1;
6580 path
->slots
[level
] = 0;
6583 * check reference count again if the block isn't locked.
6584 * we should start walking down the tree again if reference
6587 if (!path
->locks
[level
]) {
6589 btrfs_tree_lock(eb
);
6590 btrfs_set_lock_blocking(eb
);
6591 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
6593 ret
= btrfs_lookup_extent_info(trans
, root
,
6598 BUG_ON(wc
->refs
[level
] == 0);
6599 if (wc
->refs
[level
] == 1) {
6600 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
6606 /* wc->stage == DROP_REFERENCE */
6607 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
6609 if (wc
->refs
[level
] == 1) {
6611 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
6612 ret
= btrfs_dec_ref(trans
, root
, eb
, 1,
6615 ret
= btrfs_dec_ref(trans
, root
, eb
, 0,
6619 /* make block locked assertion in clean_tree_block happy */
6620 if (!path
->locks
[level
] &&
6621 btrfs_header_generation(eb
) == trans
->transid
) {
6622 btrfs_tree_lock(eb
);
6623 btrfs_set_lock_blocking(eb
);
6624 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
6626 clean_tree_block(trans
, root
, eb
);
6629 if (eb
== root
->node
) {
6630 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
6633 BUG_ON(root
->root_key
.objectid
!=
6634 btrfs_header_owner(eb
));
6636 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
6637 parent
= path
->nodes
[level
+ 1]->start
;
6639 BUG_ON(root
->root_key
.objectid
!=
6640 btrfs_header_owner(path
->nodes
[level
+ 1]));
6643 btrfs_free_tree_block(trans
, root
, eb
, parent
, wc
->refs
[level
] == 1, 0);
6645 wc
->refs
[level
] = 0;
6646 wc
->flags
[level
] = 0;
6650 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
6651 struct btrfs_root
*root
,
6652 struct btrfs_path
*path
,
6653 struct walk_control
*wc
)
6655 int level
= wc
->level
;
6656 int lookup_info
= 1;
6659 while (level
>= 0) {
6660 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
6667 if (path
->slots
[level
] >=
6668 btrfs_header_nritems(path
->nodes
[level
]))
6671 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
6673 path
->slots
[level
]++;
6682 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
6683 struct btrfs_root
*root
,
6684 struct btrfs_path
*path
,
6685 struct walk_control
*wc
, int max_level
)
6687 int level
= wc
->level
;
6690 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
6691 while (level
< max_level
&& path
->nodes
[level
]) {
6693 if (path
->slots
[level
] + 1 <
6694 btrfs_header_nritems(path
->nodes
[level
])) {
6695 path
->slots
[level
]++;
6698 ret
= walk_up_proc(trans
, root
, path
, wc
);
6702 if (path
->locks
[level
]) {
6703 btrfs_tree_unlock_rw(path
->nodes
[level
],
6704 path
->locks
[level
]);
6705 path
->locks
[level
] = 0;
6707 free_extent_buffer(path
->nodes
[level
]);
6708 path
->nodes
[level
] = NULL
;
6716 * drop a subvolume tree.
6718 * this function traverses the tree freeing any blocks that only
6719 * referenced by the tree.
6721 * when a shared tree block is found. this function decreases its
6722 * reference count by one. if update_ref is true, this function
6723 * also make sure backrefs for the shared block and all lower level
6724 * blocks are properly updated.
6726 void btrfs_drop_snapshot(struct btrfs_root
*root
,
6727 struct btrfs_block_rsv
*block_rsv
, int update_ref
,
6730 struct btrfs_path
*path
;
6731 struct btrfs_trans_handle
*trans
;
6732 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
6733 struct btrfs_root_item
*root_item
= &root
->root_item
;
6734 struct walk_control
*wc
;
6735 struct btrfs_key key
;
6740 path
= btrfs_alloc_path();
6746 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
6748 btrfs_free_path(path
);
6753 trans
= btrfs_start_transaction(tree_root
, 0);
6754 BUG_ON(IS_ERR(trans
));
6757 trans
->block_rsv
= block_rsv
;
6759 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
6760 level
= btrfs_header_level(root
->node
);
6761 path
->nodes
[level
] = btrfs_lock_root_node(root
);
6762 btrfs_set_lock_blocking(path
->nodes
[level
]);
6763 path
->slots
[level
] = 0;
6764 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
6765 memset(&wc
->update_progress
, 0,
6766 sizeof(wc
->update_progress
));
6768 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
6769 memcpy(&wc
->update_progress
, &key
,
6770 sizeof(wc
->update_progress
));
6772 level
= root_item
->drop_level
;
6774 path
->lowest_level
= level
;
6775 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6776 path
->lowest_level
= 0;
6784 * unlock our path, this is safe because only this
6785 * function is allowed to delete this snapshot
6787 btrfs_unlock_up_safe(path
, 0);
6789 level
= btrfs_header_level(root
->node
);
6791 btrfs_tree_lock(path
->nodes
[level
]);
6792 btrfs_set_lock_blocking(path
->nodes
[level
]);
6794 ret
= btrfs_lookup_extent_info(trans
, root
,
6795 path
->nodes
[level
]->start
,
6796 path
->nodes
[level
]->len
,
6800 BUG_ON(wc
->refs
[level
] == 0);
6802 if (level
== root_item
->drop_level
)
6805 btrfs_tree_unlock(path
->nodes
[level
]);
6806 WARN_ON(wc
->refs
[level
] != 1);
6812 wc
->shared_level
= -1;
6813 wc
->stage
= DROP_REFERENCE
;
6814 wc
->update_ref
= update_ref
;
6816 wc
->for_reloc
= for_reloc
;
6817 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
6820 ret
= walk_down_tree(trans
, root
, path
, wc
);
6826 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
6833 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
6837 if (wc
->stage
== DROP_REFERENCE
) {
6839 btrfs_node_key(path
->nodes
[level
],
6840 &root_item
->drop_progress
,
6841 path
->slots
[level
]);
6842 root_item
->drop_level
= level
;
6845 BUG_ON(wc
->level
== 0);
6846 if (btrfs_should_end_transaction(trans
, tree_root
)) {
6847 ret
= btrfs_update_root(trans
, tree_root
,
6852 btrfs_end_transaction_throttle(trans
, tree_root
);
6853 trans
= btrfs_start_transaction(tree_root
, 0);
6854 BUG_ON(IS_ERR(trans
));
6856 trans
->block_rsv
= block_rsv
;
6859 btrfs_release_path(path
);
6862 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
6865 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
6866 ret
= btrfs_find_last_root(tree_root
, root
->root_key
.objectid
,
6870 /* if we fail to delete the orphan item this time
6871 * around, it'll get picked up the next time.
6873 * The most common failure here is just -ENOENT.
6875 btrfs_del_orphan_item(trans
, tree_root
,
6876 root
->root_key
.objectid
);
6880 if (root
->in_radix
) {
6881 btrfs_free_fs_root(tree_root
->fs_info
, root
);
6883 free_extent_buffer(root
->node
);
6884 free_extent_buffer(root
->commit_root
);
6888 btrfs_end_transaction_throttle(trans
, tree_root
);
6890 btrfs_free_path(path
);
6893 btrfs_std_error(root
->fs_info
, err
);
6898 * drop subtree rooted at tree block 'node'.
6900 * NOTE: this function will unlock and release tree block 'node'
6901 * only used by relocation code
6903 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
6904 struct btrfs_root
*root
,
6905 struct extent_buffer
*node
,
6906 struct extent_buffer
*parent
)
6908 struct btrfs_path
*path
;
6909 struct walk_control
*wc
;
6915 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
6917 path
= btrfs_alloc_path();
6921 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
6923 btrfs_free_path(path
);
6927 btrfs_assert_tree_locked(parent
);
6928 parent_level
= btrfs_header_level(parent
);
6929 extent_buffer_get(parent
);
6930 path
->nodes
[parent_level
] = parent
;
6931 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
6933 btrfs_assert_tree_locked(node
);
6934 level
= btrfs_header_level(node
);
6935 path
->nodes
[level
] = node
;
6936 path
->slots
[level
] = 0;
6937 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
6939 wc
->refs
[parent_level
] = 1;
6940 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
6942 wc
->shared_level
= -1;
6943 wc
->stage
= DROP_REFERENCE
;
6947 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
6950 wret
= walk_down_tree(trans
, root
, path
, wc
);
6956 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
6964 btrfs_free_path(path
);
6968 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
6971 u64 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
6972 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
6974 if (root
->fs_info
->balance_ctl
) {
6975 struct btrfs_balance_control
*bctl
= root
->fs_info
->balance_ctl
;
6978 /* pick restriper's target profile and return */
6979 if (flags
& BTRFS_BLOCK_GROUP_DATA
&&
6980 bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
6981 tgt
= BTRFS_BLOCK_GROUP_DATA
| bctl
->data
.target
;
6982 } else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
&&
6983 bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
6984 tgt
= BTRFS_BLOCK_GROUP_SYSTEM
| bctl
->sys
.target
;
6985 } else if (flags
& BTRFS_BLOCK_GROUP_METADATA
&&
6986 bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
6987 tgt
= BTRFS_BLOCK_GROUP_METADATA
| bctl
->meta
.target
;
6991 /* extended -> chunk profile */
6992 tgt
&= ~BTRFS_AVAIL_ALLOC_BIT_SINGLE
;
6998 * we add in the count of missing devices because we want
6999 * to make sure that any RAID levels on a degraded FS
7000 * continue to be honored.
7002 num_devices
= root
->fs_info
->fs_devices
->rw_devices
+
7003 root
->fs_info
->fs_devices
->missing_devices
;
7005 if (num_devices
== 1) {
7006 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
7007 stripped
= flags
& ~stripped
;
7009 /* turn raid0 into single device chunks */
7010 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
7013 /* turn mirroring into duplication */
7014 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
7015 BTRFS_BLOCK_GROUP_RAID10
))
7016 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
7019 /* they already had raid on here, just return */
7020 if (flags
& stripped
)
7023 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
7024 stripped
= flags
& ~stripped
;
7026 /* switch duplicated blocks with raid1 */
7027 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
7028 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
7030 /* turn single device chunks into raid0 */
7031 return stripped
| BTRFS_BLOCK_GROUP_RAID0
;
7036 static int set_block_group_ro(struct btrfs_block_group_cache
*cache
, int force
)
7038 struct btrfs_space_info
*sinfo
= cache
->space_info
;
7040 u64 min_allocable_bytes
;
7045 * We need some metadata space and system metadata space for
7046 * allocating chunks in some corner cases until we force to set
7047 * it to be readonly.
7050 (BTRFS_BLOCK_GROUP_SYSTEM
| BTRFS_BLOCK_GROUP_METADATA
)) &&
7052 min_allocable_bytes
= 1 * 1024 * 1024;
7054 min_allocable_bytes
= 0;
7056 spin_lock(&sinfo
->lock
);
7057 spin_lock(&cache
->lock
);
7064 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
7065 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
7067 if (sinfo
->bytes_used
+ sinfo
->bytes_reserved
+ sinfo
->bytes_pinned
+
7068 sinfo
->bytes_may_use
+ sinfo
->bytes_readonly
+ num_bytes
+
7069 min_allocable_bytes
<= sinfo
->total_bytes
) {
7070 sinfo
->bytes_readonly
+= num_bytes
;
7075 spin_unlock(&cache
->lock
);
7076 spin_unlock(&sinfo
->lock
);
7080 int btrfs_set_block_group_ro(struct btrfs_root
*root
,
7081 struct btrfs_block_group_cache
*cache
)
7084 struct btrfs_trans_handle
*trans
;
7090 trans
= btrfs_join_transaction(root
);
7091 BUG_ON(IS_ERR(trans
));
7093 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
7094 if (alloc_flags
!= cache
->flags
)
7095 do_chunk_alloc(trans
, root
, 2 * 1024 * 1024, alloc_flags
,
7098 ret
= set_block_group_ro(cache
, 0);
7101 alloc_flags
= get_alloc_profile(root
, cache
->space_info
->flags
);
7102 ret
= do_chunk_alloc(trans
, root
, 2 * 1024 * 1024, alloc_flags
,
7106 ret
= set_block_group_ro(cache
, 0);
7108 btrfs_end_transaction(trans
, root
);
7112 int btrfs_force_chunk_alloc(struct btrfs_trans_handle
*trans
,
7113 struct btrfs_root
*root
, u64 type
)
7115 u64 alloc_flags
= get_alloc_profile(root
, type
);
7116 return do_chunk_alloc(trans
, root
, 2 * 1024 * 1024, alloc_flags
,
7121 * helper to account the unused space of all the readonly block group in the
7122 * list. takes mirrors into account.
7124 static u64
__btrfs_get_ro_block_group_free_space(struct list_head
*groups_list
)
7126 struct btrfs_block_group_cache
*block_group
;
7130 list_for_each_entry(block_group
, groups_list
, list
) {
7131 spin_lock(&block_group
->lock
);
7133 if (!block_group
->ro
) {
7134 spin_unlock(&block_group
->lock
);
7138 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_RAID1
|
7139 BTRFS_BLOCK_GROUP_RAID10
|
7140 BTRFS_BLOCK_GROUP_DUP
))
7145 free_bytes
+= (block_group
->key
.offset
-
7146 btrfs_block_group_used(&block_group
->item
)) *
7149 spin_unlock(&block_group
->lock
);
7156 * helper to account the unused space of all the readonly block group in the
7157 * space_info. takes mirrors into account.
7159 u64
btrfs_account_ro_block_groups_free_space(struct btrfs_space_info
*sinfo
)
7164 spin_lock(&sinfo
->lock
);
7166 for(i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
7167 if (!list_empty(&sinfo
->block_groups
[i
]))
7168 free_bytes
+= __btrfs_get_ro_block_group_free_space(
7169 &sinfo
->block_groups
[i
]);
7171 spin_unlock(&sinfo
->lock
);
7176 int btrfs_set_block_group_rw(struct btrfs_root
*root
,
7177 struct btrfs_block_group_cache
*cache
)
7179 struct btrfs_space_info
*sinfo
= cache
->space_info
;
7184 spin_lock(&sinfo
->lock
);
7185 spin_lock(&cache
->lock
);
7186 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
7187 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
7188 sinfo
->bytes_readonly
-= num_bytes
;
7190 spin_unlock(&cache
->lock
);
7191 spin_unlock(&sinfo
->lock
);
7196 * checks to see if its even possible to relocate this block group.
7198 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7199 * ok to go ahead and try.
7201 int btrfs_can_relocate(struct btrfs_root
*root
, u64 bytenr
)
7203 struct btrfs_block_group_cache
*block_group
;
7204 struct btrfs_space_info
*space_info
;
7205 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
7206 struct btrfs_device
*device
;
7214 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
7216 /* odd, couldn't find the block group, leave it alone */
7220 min_free
= btrfs_block_group_used(&block_group
->item
);
7222 /* no bytes used, we're good */
7226 space_info
= block_group
->space_info
;
7227 spin_lock(&space_info
->lock
);
7229 full
= space_info
->full
;
7232 * if this is the last block group we have in this space, we can't
7233 * relocate it unless we're able to allocate a new chunk below.
7235 * Otherwise, we need to make sure we have room in the space to handle
7236 * all of the extents from this block group. If we can, we're good
7238 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
7239 (space_info
->bytes_used
+ space_info
->bytes_reserved
+
7240 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
7241 min_free
< space_info
->total_bytes
)) {
7242 spin_unlock(&space_info
->lock
);
7245 spin_unlock(&space_info
->lock
);
7248 * ok we don't have enough space, but maybe we have free space on our
7249 * devices to allocate new chunks for relocation, so loop through our
7250 * alloc devices and guess if we have enough space. However, if we
7251 * were marked as full, then we know there aren't enough chunks, and we
7266 index
= get_block_group_index(block_group
);
7271 } else if (index
== 1) {
7273 } else if (index
== 2) {
7276 } else if (index
== 3) {
7277 dev_min
= fs_devices
->rw_devices
;
7278 do_div(min_free
, dev_min
);
7281 mutex_lock(&root
->fs_info
->chunk_mutex
);
7282 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
7286 * check to make sure we can actually find a chunk with enough
7287 * space to fit our block group in.
7289 if (device
->total_bytes
> device
->bytes_used
+ min_free
) {
7290 ret
= find_free_dev_extent(device
, min_free
,
7295 if (dev_nr
>= dev_min
)
7301 mutex_unlock(&root
->fs_info
->chunk_mutex
);
7303 btrfs_put_block_group(block_group
);
7307 static int find_first_block_group(struct btrfs_root
*root
,
7308 struct btrfs_path
*path
, struct btrfs_key
*key
)
7311 struct btrfs_key found_key
;
7312 struct extent_buffer
*leaf
;
7315 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
7320 slot
= path
->slots
[0];
7321 leaf
= path
->nodes
[0];
7322 if (slot
>= btrfs_header_nritems(leaf
)) {
7323 ret
= btrfs_next_leaf(root
, path
);
7330 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
7332 if (found_key
.objectid
>= key
->objectid
&&
7333 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
7343 void btrfs_put_block_group_cache(struct btrfs_fs_info
*info
)
7345 struct btrfs_block_group_cache
*block_group
;
7349 struct inode
*inode
;
7351 block_group
= btrfs_lookup_first_block_group(info
, last
);
7352 while (block_group
) {
7353 spin_lock(&block_group
->lock
);
7354 if (block_group
->iref
)
7356 spin_unlock(&block_group
->lock
);
7357 block_group
= next_block_group(info
->tree_root
,
7367 inode
= block_group
->inode
;
7368 block_group
->iref
= 0;
7369 block_group
->inode
= NULL
;
7370 spin_unlock(&block_group
->lock
);
7372 last
= block_group
->key
.objectid
+ block_group
->key
.offset
;
7373 btrfs_put_block_group(block_group
);
7377 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
7379 struct btrfs_block_group_cache
*block_group
;
7380 struct btrfs_space_info
*space_info
;
7381 struct btrfs_caching_control
*caching_ctl
;
7384 down_write(&info
->extent_commit_sem
);
7385 while (!list_empty(&info
->caching_block_groups
)) {
7386 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
7387 struct btrfs_caching_control
, list
);
7388 list_del(&caching_ctl
->list
);
7389 put_caching_control(caching_ctl
);
7391 up_write(&info
->extent_commit_sem
);
7393 spin_lock(&info
->block_group_cache_lock
);
7394 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
7395 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
7397 rb_erase(&block_group
->cache_node
,
7398 &info
->block_group_cache_tree
);
7399 spin_unlock(&info
->block_group_cache_lock
);
7401 down_write(&block_group
->space_info
->groups_sem
);
7402 list_del(&block_group
->list
);
7403 up_write(&block_group
->space_info
->groups_sem
);
7405 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
7406 wait_block_group_cache_done(block_group
);
7409 * We haven't cached this block group, which means we could
7410 * possibly have excluded extents on this block group.
7412 if (block_group
->cached
== BTRFS_CACHE_NO
)
7413 free_excluded_extents(info
->extent_root
, block_group
);
7415 btrfs_remove_free_space_cache(block_group
);
7416 btrfs_put_block_group(block_group
);
7418 spin_lock(&info
->block_group_cache_lock
);
7420 spin_unlock(&info
->block_group_cache_lock
);
7422 /* now that all the block groups are freed, go through and
7423 * free all the space_info structs. This is only called during
7424 * the final stages of unmount, and so we know nobody is
7425 * using them. We call synchronize_rcu() once before we start,
7426 * just to be on the safe side.
7430 release_global_block_rsv(info
);
7432 while(!list_empty(&info
->space_info
)) {
7433 space_info
= list_entry(info
->space_info
.next
,
7434 struct btrfs_space_info
,
7436 if (space_info
->bytes_pinned
> 0 ||
7437 space_info
->bytes_reserved
> 0 ||
7438 space_info
->bytes_may_use
> 0) {
7440 dump_space_info(space_info
, 0, 0);
7442 list_del(&space_info
->list
);
7448 static void __link_block_group(struct btrfs_space_info
*space_info
,
7449 struct btrfs_block_group_cache
*cache
)
7451 int index
= get_block_group_index(cache
);
7453 down_write(&space_info
->groups_sem
);
7454 list_add_tail(&cache
->list
, &space_info
->block_groups
[index
]);
7455 up_write(&space_info
->groups_sem
);
7458 int btrfs_read_block_groups(struct btrfs_root
*root
)
7460 struct btrfs_path
*path
;
7462 struct btrfs_block_group_cache
*cache
;
7463 struct btrfs_fs_info
*info
= root
->fs_info
;
7464 struct btrfs_space_info
*space_info
;
7465 struct btrfs_key key
;
7466 struct btrfs_key found_key
;
7467 struct extent_buffer
*leaf
;
7471 root
= info
->extent_root
;
7474 btrfs_set_key_type(&key
, BTRFS_BLOCK_GROUP_ITEM_KEY
);
7475 path
= btrfs_alloc_path();
7480 cache_gen
= btrfs_super_cache_generation(root
->fs_info
->super_copy
);
7481 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
7482 btrfs_super_generation(root
->fs_info
->super_copy
) != cache_gen
)
7484 if (btrfs_test_opt(root
, CLEAR_CACHE
))
7488 ret
= find_first_block_group(root
, path
, &key
);
7493 leaf
= path
->nodes
[0];
7494 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
7495 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
7500 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
7502 if (!cache
->free_space_ctl
) {
7508 atomic_set(&cache
->count
, 1);
7509 spin_lock_init(&cache
->lock
);
7510 cache
->fs_info
= info
;
7511 INIT_LIST_HEAD(&cache
->list
);
7512 INIT_LIST_HEAD(&cache
->cluster_list
);
7515 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
7517 read_extent_buffer(leaf
, &cache
->item
,
7518 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
7519 sizeof(cache
->item
));
7520 memcpy(&cache
->key
, &found_key
, sizeof(found_key
));
7522 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
7523 btrfs_release_path(path
);
7524 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
7525 cache
->sectorsize
= root
->sectorsize
;
7527 btrfs_init_free_space_ctl(cache
);
7530 * We need to exclude the super stripes now so that the space
7531 * info has super bytes accounted for, otherwise we'll think
7532 * we have more space than we actually do.
7534 exclude_super_stripes(root
, cache
);
7537 * check for two cases, either we are full, and therefore
7538 * don't need to bother with the caching work since we won't
7539 * find any space, or we are empty, and we can just add all
7540 * the space in and be done with it. This saves us _alot_ of
7541 * time, particularly in the full case.
7543 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
7544 cache
->last_byte_to_unpin
= (u64
)-1;
7545 cache
->cached
= BTRFS_CACHE_FINISHED
;
7546 free_excluded_extents(root
, cache
);
7547 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
7548 cache
->last_byte_to_unpin
= (u64
)-1;
7549 cache
->cached
= BTRFS_CACHE_FINISHED
;
7550 add_new_free_space(cache
, root
->fs_info
,
7552 found_key
.objectid
+
7554 free_excluded_extents(root
, cache
);
7557 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
7558 btrfs_block_group_used(&cache
->item
),
7561 cache
->space_info
= space_info
;
7562 spin_lock(&cache
->space_info
->lock
);
7563 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
7564 spin_unlock(&cache
->space_info
->lock
);
7566 __link_block_group(space_info
, cache
);
7568 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
7571 set_avail_alloc_bits(root
->fs_info
, cache
->flags
);
7572 if (btrfs_chunk_readonly(root
, cache
->key
.objectid
))
7573 set_block_group_ro(cache
, 1);
7576 list_for_each_entry_rcu(space_info
, &root
->fs_info
->space_info
, list
) {
7577 if (!(get_alloc_profile(root
, space_info
->flags
) &
7578 (BTRFS_BLOCK_GROUP_RAID10
|
7579 BTRFS_BLOCK_GROUP_RAID1
|
7580 BTRFS_BLOCK_GROUP_DUP
)))
7583 * avoid allocating from un-mirrored block group if there are
7584 * mirrored block groups.
7586 list_for_each_entry(cache
, &space_info
->block_groups
[3], list
)
7587 set_block_group_ro(cache
, 1);
7588 list_for_each_entry(cache
, &space_info
->block_groups
[4], list
)
7589 set_block_group_ro(cache
, 1);
7592 init_global_block_rsv(info
);
7595 btrfs_free_path(path
);
7599 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
7600 struct btrfs_root
*root
, u64 bytes_used
,
7601 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
7605 struct btrfs_root
*extent_root
;
7606 struct btrfs_block_group_cache
*cache
;
7608 extent_root
= root
->fs_info
->extent_root
;
7610 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
7612 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
7615 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
7617 if (!cache
->free_space_ctl
) {
7622 cache
->key
.objectid
= chunk_offset
;
7623 cache
->key
.offset
= size
;
7624 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
7625 cache
->sectorsize
= root
->sectorsize
;
7626 cache
->fs_info
= root
->fs_info
;
7628 atomic_set(&cache
->count
, 1);
7629 spin_lock_init(&cache
->lock
);
7630 INIT_LIST_HEAD(&cache
->list
);
7631 INIT_LIST_HEAD(&cache
->cluster_list
);
7633 btrfs_init_free_space_ctl(cache
);
7635 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
7636 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
7637 cache
->flags
= type
;
7638 btrfs_set_block_group_flags(&cache
->item
, type
);
7640 cache
->last_byte_to_unpin
= (u64
)-1;
7641 cache
->cached
= BTRFS_CACHE_FINISHED
;
7642 exclude_super_stripes(root
, cache
);
7644 add_new_free_space(cache
, root
->fs_info
, chunk_offset
,
7645 chunk_offset
+ size
);
7647 free_excluded_extents(root
, cache
);
7649 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
7650 &cache
->space_info
);
7652 update_global_block_rsv(root
->fs_info
);
7654 spin_lock(&cache
->space_info
->lock
);
7655 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
7656 spin_unlock(&cache
->space_info
->lock
);
7658 __link_block_group(cache
->space_info
, cache
);
7660 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
7663 ret
= btrfs_insert_item(trans
, extent_root
, &cache
->key
, &cache
->item
,
7664 sizeof(cache
->item
));
7667 set_avail_alloc_bits(extent_root
->fs_info
, type
);
7672 static void clear_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
7674 u64 extra_flags
= flags
& BTRFS_BLOCK_GROUP_PROFILE_MASK
;
7676 /* chunk -> extended profile */
7677 if (extra_flags
== 0)
7678 extra_flags
= BTRFS_AVAIL_ALLOC_BIT_SINGLE
;
7680 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
7681 fs_info
->avail_data_alloc_bits
&= ~extra_flags
;
7682 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
7683 fs_info
->avail_metadata_alloc_bits
&= ~extra_flags
;
7684 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
7685 fs_info
->avail_system_alloc_bits
&= ~extra_flags
;
7688 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
7689 struct btrfs_root
*root
, u64 group_start
)
7691 struct btrfs_path
*path
;
7692 struct btrfs_block_group_cache
*block_group
;
7693 struct btrfs_free_cluster
*cluster
;
7694 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
7695 struct btrfs_key key
;
7696 struct inode
*inode
;
7701 root
= root
->fs_info
->extent_root
;
7703 block_group
= btrfs_lookup_block_group(root
->fs_info
, group_start
);
7704 BUG_ON(!block_group
);
7705 BUG_ON(!block_group
->ro
);
7708 * Free the reserved super bytes from this block group before
7711 free_excluded_extents(root
, block_group
);
7713 memcpy(&key
, &block_group
->key
, sizeof(key
));
7714 index
= get_block_group_index(block_group
);
7715 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
7716 BTRFS_BLOCK_GROUP_RAID1
|
7717 BTRFS_BLOCK_GROUP_RAID10
))
7722 /* make sure this block group isn't part of an allocation cluster */
7723 cluster
= &root
->fs_info
->data_alloc_cluster
;
7724 spin_lock(&cluster
->refill_lock
);
7725 btrfs_return_cluster_to_free_space(block_group
, cluster
);
7726 spin_unlock(&cluster
->refill_lock
);
7729 * make sure this block group isn't part of a metadata
7730 * allocation cluster
7732 cluster
= &root
->fs_info
->meta_alloc_cluster
;
7733 spin_lock(&cluster
->refill_lock
);
7734 btrfs_return_cluster_to_free_space(block_group
, cluster
);
7735 spin_unlock(&cluster
->refill_lock
);
7737 path
= btrfs_alloc_path();
7743 inode
= lookup_free_space_inode(tree_root
, block_group
, path
);
7744 if (!IS_ERR(inode
)) {
7745 ret
= btrfs_orphan_add(trans
, inode
);
7748 /* One for the block groups ref */
7749 spin_lock(&block_group
->lock
);
7750 if (block_group
->iref
) {
7751 block_group
->iref
= 0;
7752 block_group
->inode
= NULL
;
7753 spin_unlock(&block_group
->lock
);
7756 spin_unlock(&block_group
->lock
);
7758 /* One for our lookup ref */
7759 btrfs_add_delayed_iput(inode
);
7762 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
7763 key
.offset
= block_group
->key
.objectid
;
7766 ret
= btrfs_search_slot(trans
, tree_root
, &key
, path
, -1, 1);
7770 btrfs_release_path(path
);
7772 ret
= btrfs_del_item(trans
, tree_root
, path
);
7775 btrfs_release_path(path
);
7778 spin_lock(&root
->fs_info
->block_group_cache_lock
);
7779 rb_erase(&block_group
->cache_node
,
7780 &root
->fs_info
->block_group_cache_tree
);
7781 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
7783 down_write(&block_group
->space_info
->groups_sem
);
7785 * we must use list_del_init so people can check to see if they
7786 * are still on the list after taking the semaphore
7788 list_del_init(&block_group
->list
);
7789 if (list_empty(&block_group
->space_info
->block_groups
[index
]))
7790 clear_avail_alloc_bits(root
->fs_info
, block_group
->flags
);
7791 up_write(&block_group
->space_info
->groups_sem
);
7793 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
7794 wait_block_group_cache_done(block_group
);
7796 btrfs_remove_free_space_cache(block_group
);
7798 spin_lock(&block_group
->space_info
->lock
);
7799 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
7800 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
7801 block_group
->space_info
->disk_total
-= block_group
->key
.offset
* factor
;
7802 spin_unlock(&block_group
->space_info
->lock
);
7804 memcpy(&key
, &block_group
->key
, sizeof(key
));
7806 btrfs_clear_space_info_full(root
->fs_info
);
7808 btrfs_put_block_group(block_group
);
7809 btrfs_put_block_group(block_group
);
7811 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
7817 ret
= btrfs_del_item(trans
, root
, path
);
7819 btrfs_free_path(path
);
7823 int btrfs_init_space_info(struct btrfs_fs_info
*fs_info
)
7825 struct btrfs_space_info
*space_info
;
7826 struct btrfs_super_block
*disk_super
;
7832 disk_super
= fs_info
->super_copy
;
7833 if (!btrfs_super_root(disk_super
))
7836 features
= btrfs_super_incompat_flags(disk_super
);
7837 if (features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
7840 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
7841 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
7846 flags
= BTRFS_BLOCK_GROUP_METADATA
| BTRFS_BLOCK_GROUP_DATA
;
7847 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
7849 flags
= BTRFS_BLOCK_GROUP_METADATA
;
7850 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
7854 flags
= BTRFS_BLOCK_GROUP_DATA
;
7855 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
7861 int btrfs_error_unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
7863 return unpin_extent_range(root
, start
, end
);
7866 int btrfs_error_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
7867 u64 num_bytes
, u64
*actual_bytes
)
7869 return btrfs_discard_extent(root
, bytenr
, num_bytes
, actual_bytes
);
7872 int btrfs_trim_fs(struct btrfs_root
*root
, struct fstrim_range
*range
)
7874 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7875 struct btrfs_block_group_cache
*cache
= NULL
;
7882 cache
= btrfs_lookup_block_group(fs_info
, range
->start
);
7885 if (cache
->key
.objectid
>= (range
->start
+ range
->len
)) {
7886 btrfs_put_block_group(cache
);
7890 start
= max(range
->start
, cache
->key
.objectid
);
7891 end
= min(range
->start
+ range
->len
,
7892 cache
->key
.objectid
+ cache
->key
.offset
);
7894 if (end
- start
>= range
->minlen
) {
7895 if (!block_group_cache_done(cache
)) {
7896 ret
= cache_block_group(cache
, NULL
, root
, 0);
7898 wait_block_group_cache_done(cache
);
7900 ret
= btrfs_trim_block_group(cache
,
7906 trimmed
+= group_trimmed
;
7908 btrfs_put_block_group(cache
);
7913 cache
= next_block_group(fs_info
->tree_root
, cache
);
7916 range
->len
= trimmed
;