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_DATA
| BTRFS_BLOCK_GROUP_SYSTEM
|
622 BTRFS_BLOCK_GROUP_METADATA
;
625 list_for_each_entry_rcu(found
, head
, list
) {
626 if (found
->flags
& flags
) {
636 * after adding space to the filesystem, we need to clear the full flags
637 * on all the space infos.
639 void btrfs_clear_space_info_full(struct btrfs_fs_info
*info
)
641 struct list_head
*head
= &info
->space_info
;
642 struct btrfs_space_info
*found
;
645 list_for_each_entry_rcu(found
, head
, list
)
650 static u64
div_factor(u64 num
, int factor
)
659 static u64
div_factor_fine(u64 num
, int factor
)
668 u64
btrfs_find_block_group(struct btrfs_root
*root
,
669 u64 search_start
, u64 search_hint
, int owner
)
671 struct btrfs_block_group_cache
*cache
;
673 u64 last
= max(search_hint
, search_start
);
680 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
684 spin_lock(&cache
->lock
);
685 last
= cache
->key
.objectid
+ cache
->key
.offset
;
686 used
= btrfs_block_group_used(&cache
->item
);
688 if ((full_search
|| !cache
->ro
) &&
689 block_group_bits(cache
, BTRFS_BLOCK_GROUP_METADATA
)) {
690 if (used
+ cache
->pinned
+ cache
->reserved
<
691 div_factor(cache
->key
.offset
, factor
)) {
692 group_start
= cache
->key
.objectid
;
693 spin_unlock(&cache
->lock
);
694 btrfs_put_block_group(cache
);
698 spin_unlock(&cache
->lock
);
699 btrfs_put_block_group(cache
);
707 if (!full_search
&& factor
< 10) {
717 /* simple helper to search for an existing extent at a given offset */
718 int btrfs_lookup_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
721 struct btrfs_key key
;
722 struct btrfs_path
*path
;
724 path
= btrfs_alloc_path();
728 key
.objectid
= start
;
730 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
731 ret
= btrfs_search_slot(NULL
, root
->fs_info
->extent_root
, &key
, path
,
733 btrfs_free_path(path
);
738 * helper function to lookup reference count and flags of extent.
740 * the head node for delayed ref is used to store the sum of all the
741 * reference count modifications queued up in the rbtree. the head
742 * node may also store the extent flags to set. This way you can check
743 * to see what the reference count and extent flags would be if all of
744 * the delayed refs are not processed.
746 int btrfs_lookup_extent_info(struct btrfs_trans_handle
*trans
,
747 struct btrfs_root
*root
, u64 bytenr
,
748 u64 num_bytes
, u64
*refs
, u64
*flags
)
750 struct btrfs_delayed_ref_head
*head
;
751 struct btrfs_delayed_ref_root
*delayed_refs
;
752 struct btrfs_path
*path
;
753 struct btrfs_extent_item
*ei
;
754 struct extent_buffer
*leaf
;
755 struct btrfs_key key
;
761 path
= btrfs_alloc_path();
765 key
.objectid
= bytenr
;
766 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
767 key
.offset
= num_bytes
;
769 path
->skip_locking
= 1;
770 path
->search_commit_root
= 1;
773 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
,
779 leaf
= path
->nodes
[0];
780 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
781 if (item_size
>= sizeof(*ei
)) {
782 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
783 struct btrfs_extent_item
);
784 num_refs
= btrfs_extent_refs(leaf
, ei
);
785 extent_flags
= btrfs_extent_flags(leaf
, ei
);
787 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
788 struct btrfs_extent_item_v0
*ei0
;
789 BUG_ON(item_size
!= sizeof(*ei0
));
790 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
791 struct btrfs_extent_item_v0
);
792 num_refs
= btrfs_extent_refs_v0(leaf
, ei0
);
793 /* FIXME: this isn't correct for data */
794 extent_flags
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
799 BUG_ON(num_refs
== 0);
809 delayed_refs
= &trans
->transaction
->delayed_refs
;
810 spin_lock(&delayed_refs
->lock
);
811 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
813 if (!mutex_trylock(&head
->mutex
)) {
814 atomic_inc(&head
->node
.refs
);
815 spin_unlock(&delayed_refs
->lock
);
817 btrfs_release_path(path
);
820 * Mutex was contended, block until it's released and try
823 mutex_lock(&head
->mutex
);
824 mutex_unlock(&head
->mutex
);
825 btrfs_put_delayed_ref(&head
->node
);
828 if (head
->extent_op
&& head
->extent_op
->update_flags
)
829 extent_flags
|= head
->extent_op
->flags_to_set
;
831 BUG_ON(num_refs
== 0);
833 num_refs
+= head
->node
.ref_mod
;
834 mutex_unlock(&head
->mutex
);
836 spin_unlock(&delayed_refs
->lock
);
838 WARN_ON(num_refs
== 0);
842 *flags
= extent_flags
;
844 btrfs_free_path(path
);
849 * Back reference rules. Back refs have three main goals:
851 * 1) differentiate between all holders of references to an extent so that
852 * when a reference is dropped we can make sure it was a valid reference
853 * before freeing the extent.
855 * 2) Provide enough information to quickly find the holders of an extent
856 * if we notice a given block is corrupted or bad.
858 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
859 * maintenance. This is actually the same as #2, but with a slightly
860 * different use case.
862 * There are two kinds of back refs. The implicit back refs is optimized
863 * for pointers in non-shared tree blocks. For a given pointer in a block,
864 * back refs of this kind provide information about the block's owner tree
865 * and the pointer's key. These information allow us to find the block by
866 * b-tree searching. The full back refs is for pointers in tree blocks not
867 * referenced by their owner trees. The location of tree block is recorded
868 * in the back refs. Actually the full back refs is generic, and can be
869 * used in all cases the implicit back refs is used. The major shortcoming
870 * of the full back refs is its overhead. Every time a tree block gets
871 * COWed, we have to update back refs entry for all pointers in it.
873 * For a newly allocated tree block, we use implicit back refs for
874 * pointers in it. This means most tree related operations only involve
875 * implicit back refs. For a tree block created in old transaction, the
876 * only way to drop a reference to it is COW it. So we can detect the
877 * event that tree block loses its owner tree's reference and do the
878 * back refs conversion.
880 * When a tree block is COW'd through a tree, there are four cases:
882 * The reference count of the block is one and the tree is the block's
883 * owner tree. Nothing to do in this case.
885 * The reference count of the block is one and the tree is not the
886 * block's owner tree. In this case, full back refs is used for pointers
887 * in the block. Remove these full back refs, add implicit back refs for
888 * every pointers in the new block.
890 * The reference count of the block is greater than one and the tree is
891 * the block's owner tree. In this case, implicit back refs is used for
892 * pointers in the block. Add full back refs for every pointers in the
893 * block, increase lower level extents' reference counts. The original
894 * implicit back refs are entailed to the new block.
896 * The reference count of the block is greater than one and the tree is
897 * not the block's owner tree. Add implicit back refs for every pointer in
898 * the new block, increase lower level extents' reference count.
900 * Back Reference Key composing:
902 * The key objectid corresponds to the first byte in the extent,
903 * The key type is used to differentiate between types of back refs.
904 * There are different meanings of the key offset for different types
907 * File extents can be referenced by:
909 * - multiple snapshots, subvolumes, or different generations in one subvol
910 * - different files inside a single subvolume
911 * - different offsets inside a file (bookend extents in file.c)
913 * The extent ref structure for the implicit back refs has fields for:
915 * - Objectid of the subvolume root
916 * - objectid of the file holding the reference
917 * - original offset in the file
918 * - how many bookend extents
920 * The key offset for the implicit back refs is hash of the first
923 * The extent ref structure for the full back refs has field for:
925 * - number of pointers in the tree leaf
927 * The key offset for the implicit back refs is the first byte of
930 * When a file extent is allocated, The implicit back refs is used.
931 * the fields are filled in:
933 * (root_key.objectid, inode objectid, offset in file, 1)
935 * When a file extent is removed file truncation, we find the
936 * corresponding implicit back refs and check the following fields:
938 * (btrfs_header_owner(leaf), inode objectid, offset in file)
940 * Btree extents can be referenced by:
942 * - Different subvolumes
944 * Both the implicit back refs and the full back refs for tree blocks
945 * only consist of key. The key offset for the implicit back refs is
946 * objectid of block's owner tree. The key offset for the full back refs
947 * is the first byte of parent block.
949 * When implicit back refs is used, information about the lowest key and
950 * level of the tree block are required. These information are stored in
951 * tree block info structure.
954 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
955 static int convert_extent_item_v0(struct btrfs_trans_handle
*trans
,
956 struct btrfs_root
*root
,
957 struct btrfs_path
*path
,
958 u64 owner
, u32 extra_size
)
960 struct btrfs_extent_item
*item
;
961 struct btrfs_extent_item_v0
*ei0
;
962 struct btrfs_extent_ref_v0
*ref0
;
963 struct btrfs_tree_block_info
*bi
;
964 struct extent_buffer
*leaf
;
965 struct btrfs_key key
;
966 struct btrfs_key found_key
;
967 u32 new_size
= sizeof(*item
);
971 leaf
= path
->nodes
[0];
972 BUG_ON(btrfs_item_size_nr(leaf
, path
->slots
[0]) != sizeof(*ei0
));
974 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
975 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
976 struct btrfs_extent_item_v0
);
977 refs
= btrfs_extent_refs_v0(leaf
, ei0
);
979 if (owner
== (u64
)-1) {
981 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
982 ret
= btrfs_next_leaf(root
, path
);
986 leaf
= path
->nodes
[0];
988 btrfs_item_key_to_cpu(leaf
, &found_key
,
990 BUG_ON(key
.objectid
!= found_key
.objectid
);
991 if (found_key
.type
!= BTRFS_EXTENT_REF_V0_KEY
) {
995 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
996 struct btrfs_extent_ref_v0
);
997 owner
= btrfs_ref_objectid_v0(leaf
, ref0
);
1001 btrfs_release_path(path
);
1003 if (owner
< BTRFS_FIRST_FREE_OBJECTID
)
1004 new_size
+= sizeof(*bi
);
1006 new_size
-= sizeof(*ei0
);
1007 ret
= btrfs_search_slot(trans
, root
, &key
, path
,
1008 new_size
+ extra_size
, 1);
1013 ret
= btrfs_extend_item(trans
, root
, path
, new_size
);
1015 leaf
= path
->nodes
[0];
1016 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1017 btrfs_set_extent_refs(leaf
, item
, refs
);
1018 /* FIXME: get real generation */
1019 btrfs_set_extent_generation(leaf
, item
, 0);
1020 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1021 btrfs_set_extent_flags(leaf
, item
,
1022 BTRFS_EXTENT_FLAG_TREE_BLOCK
|
1023 BTRFS_BLOCK_FLAG_FULL_BACKREF
);
1024 bi
= (struct btrfs_tree_block_info
*)(item
+ 1);
1025 /* FIXME: get first key of the block */
1026 memset_extent_buffer(leaf
, 0, (unsigned long)bi
, sizeof(*bi
));
1027 btrfs_set_tree_block_level(leaf
, bi
, (int)owner
);
1029 btrfs_set_extent_flags(leaf
, item
, BTRFS_EXTENT_FLAG_DATA
);
1031 btrfs_mark_buffer_dirty(leaf
);
1036 static u64
hash_extent_data_ref(u64 root_objectid
, u64 owner
, u64 offset
)
1038 u32 high_crc
= ~(u32
)0;
1039 u32 low_crc
= ~(u32
)0;
1042 lenum
= cpu_to_le64(root_objectid
);
1043 high_crc
= crc32c(high_crc
, &lenum
, sizeof(lenum
));
1044 lenum
= cpu_to_le64(owner
);
1045 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
1046 lenum
= cpu_to_le64(offset
);
1047 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
1049 return ((u64
)high_crc
<< 31) ^ (u64
)low_crc
;
1052 static u64
hash_extent_data_ref_item(struct extent_buffer
*leaf
,
1053 struct btrfs_extent_data_ref
*ref
)
1055 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf
, ref
),
1056 btrfs_extent_data_ref_objectid(leaf
, ref
),
1057 btrfs_extent_data_ref_offset(leaf
, ref
));
1060 static int match_extent_data_ref(struct extent_buffer
*leaf
,
1061 struct btrfs_extent_data_ref
*ref
,
1062 u64 root_objectid
, u64 owner
, u64 offset
)
1064 if (btrfs_extent_data_ref_root(leaf
, ref
) != root_objectid
||
1065 btrfs_extent_data_ref_objectid(leaf
, ref
) != owner
||
1066 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
1071 static noinline
int lookup_extent_data_ref(struct btrfs_trans_handle
*trans
,
1072 struct btrfs_root
*root
,
1073 struct btrfs_path
*path
,
1074 u64 bytenr
, u64 parent
,
1076 u64 owner
, u64 offset
)
1078 struct btrfs_key key
;
1079 struct btrfs_extent_data_ref
*ref
;
1080 struct extent_buffer
*leaf
;
1086 key
.objectid
= bytenr
;
1088 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1089 key
.offset
= parent
;
1091 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1092 key
.offset
= hash_extent_data_ref(root_objectid
,
1097 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1106 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1107 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1108 btrfs_release_path(path
);
1109 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1120 leaf
= path
->nodes
[0];
1121 nritems
= btrfs_header_nritems(leaf
);
1123 if (path
->slots
[0] >= nritems
) {
1124 ret
= btrfs_next_leaf(root
, path
);
1130 leaf
= path
->nodes
[0];
1131 nritems
= btrfs_header_nritems(leaf
);
1135 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1136 if (key
.objectid
!= bytenr
||
1137 key
.type
!= BTRFS_EXTENT_DATA_REF_KEY
)
1140 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1141 struct btrfs_extent_data_ref
);
1143 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1146 btrfs_release_path(path
);
1158 static noinline
int insert_extent_data_ref(struct btrfs_trans_handle
*trans
,
1159 struct btrfs_root
*root
,
1160 struct btrfs_path
*path
,
1161 u64 bytenr
, u64 parent
,
1162 u64 root_objectid
, u64 owner
,
1163 u64 offset
, int refs_to_add
)
1165 struct btrfs_key key
;
1166 struct extent_buffer
*leaf
;
1171 key
.objectid
= bytenr
;
1173 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1174 key
.offset
= parent
;
1175 size
= sizeof(struct btrfs_shared_data_ref
);
1177 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1178 key
.offset
= hash_extent_data_ref(root_objectid
,
1180 size
= sizeof(struct btrfs_extent_data_ref
);
1183 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, size
);
1184 if (ret
&& ret
!= -EEXIST
)
1187 leaf
= path
->nodes
[0];
1189 struct btrfs_shared_data_ref
*ref
;
1190 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1191 struct btrfs_shared_data_ref
);
1193 btrfs_set_shared_data_ref_count(leaf
, ref
, refs_to_add
);
1195 num_refs
= btrfs_shared_data_ref_count(leaf
, ref
);
1196 num_refs
+= refs_to_add
;
1197 btrfs_set_shared_data_ref_count(leaf
, ref
, num_refs
);
1200 struct btrfs_extent_data_ref
*ref
;
1201 while (ret
== -EEXIST
) {
1202 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1203 struct btrfs_extent_data_ref
);
1204 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1207 btrfs_release_path(path
);
1209 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1211 if (ret
&& ret
!= -EEXIST
)
1214 leaf
= path
->nodes
[0];
1216 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1217 struct btrfs_extent_data_ref
);
1219 btrfs_set_extent_data_ref_root(leaf
, ref
,
1221 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
1222 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
1223 btrfs_set_extent_data_ref_count(leaf
, ref
, refs_to_add
);
1225 num_refs
= btrfs_extent_data_ref_count(leaf
, ref
);
1226 num_refs
+= refs_to_add
;
1227 btrfs_set_extent_data_ref_count(leaf
, ref
, num_refs
);
1230 btrfs_mark_buffer_dirty(leaf
);
1233 btrfs_release_path(path
);
1237 static noinline
int remove_extent_data_ref(struct btrfs_trans_handle
*trans
,
1238 struct btrfs_root
*root
,
1239 struct btrfs_path
*path
,
1242 struct btrfs_key key
;
1243 struct btrfs_extent_data_ref
*ref1
= NULL
;
1244 struct btrfs_shared_data_ref
*ref2
= NULL
;
1245 struct extent_buffer
*leaf
;
1249 leaf
= path
->nodes
[0];
1250 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1252 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1253 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1254 struct btrfs_extent_data_ref
);
1255 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1256 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1257 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1258 struct btrfs_shared_data_ref
);
1259 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1260 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1261 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1262 struct btrfs_extent_ref_v0
*ref0
;
1263 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1264 struct btrfs_extent_ref_v0
);
1265 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1271 BUG_ON(num_refs
< refs_to_drop
);
1272 num_refs
-= refs_to_drop
;
1274 if (num_refs
== 0) {
1275 ret
= btrfs_del_item(trans
, root
, path
);
1277 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
)
1278 btrfs_set_extent_data_ref_count(leaf
, ref1
, num_refs
);
1279 else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
)
1280 btrfs_set_shared_data_ref_count(leaf
, ref2
, num_refs
);
1281 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1283 struct btrfs_extent_ref_v0
*ref0
;
1284 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1285 struct btrfs_extent_ref_v0
);
1286 btrfs_set_ref_count_v0(leaf
, ref0
, num_refs
);
1289 btrfs_mark_buffer_dirty(leaf
);
1294 static noinline u32
extent_data_ref_count(struct btrfs_root
*root
,
1295 struct btrfs_path
*path
,
1296 struct btrfs_extent_inline_ref
*iref
)
1298 struct btrfs_key key
;
1299 struct extent_buffer
*leaf
;
1300 struct btrfs_extent_data_ref
*ref1
;
1301 struct btrfs_shared_data_ref
*ref2
;
1304 leaf
= path
->nodes
[0];
1305 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1307 if (btrfs_extent_inline_ref_type(leaf
, iref
) ==
1308 BTRFS_EXTENT_DATA_REF_KEY
) {
1309 ref1
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1310 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1312 ref2
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1313 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1315 } else if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1316 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1317 struct btrfs_extent_data_ref
);
1318 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1319 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1320 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1321 struct btrfs_shared_data_ref
);
1322 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1323 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1324 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1325 struct btrfs_extent_ref_v0
*ref0
;
1326 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1327 struct btrfs_extent_ref_v0
);
1328 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1336 static noinline
int lookup_tree_block_ref(struct btrfs_trans_handle
*trans
,
1337 struct btrfs_root
*root
,
1338 struct btrfs_path
*path
,
1339 u64 bytenr
, u64 parent
,
1342 struct btrfs_key key
;
1345 key
.objectid
= bytenr
;
1347 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1348 key
.offset
= parent
;
1350 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1351 key
.offset
= root_objectid
;
1354 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1357 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1358 if (ret
== -ENOENT
&& parent
) {
1359 btrfs_release_path(path
);
1360 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1361 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1369 static noinline
int insert_tree_block_ref(struct btrfs_trans_handle
*trans
,
1370 struct btrfs_root
*root
,
1371 struct btrfs_path
*path
,
1372 u64 bytenr
, u64 parent
,
1375 struct btrfs_key key
;
1378 key
.objectid
= bytenr
;
1380 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1381 key
.offset
= parent
;
1383 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1384 key
.offset
= root_objectid
;
1387 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
1388 btrfs_release_path(path
);
1392 static inline int extent_ref_type(u64 parent
, u64 owner
)
1395 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1397 type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1399 type
= BTRFS_TREE_BLOCK_REF_KEY
;
1402 type
= BTRFS_SHARED_DATA_REF_KEY
;
1404 type
= BTRFS_EXTENT_DATA_REF_KEY
;
1409 static int find_next_key(struct btrfs_path
*path
, int level
,
1410 struct btrfs_key
*key
)
1413 for (; level
< BTRFS_MAX_LEVEL
; level
++) {
1414 if (!path
->nodes
[level
])
1416 if (path
->slots
[level
] + 1 >=
1417 btrfs_header_nritems(path
->nodes
[level
]))
1420 btrfs_item_key_to_cpu(path
->nodes
[level
], key
,
1421 path
->slots
[level
] + 1);
1423 btrfs_node_key_to_cpu(path
->nodes
[level
], key
,
1424 path
->slots
[level
] + 1);
1431 * look for inline back ref. if back ref is found, *ref_ret is set
1432 * to the address of inline back ref, and 0 is returned.
1434 * if back ref isn't found, *ref_ret is set to the address where it
1435 * should be inserted, and -ENOENT is returned.
1437 * if insert is true and there are too many inline back refs, the path
1438 * points to the extent item, and -EAGAIN is returned.
1440 * NOTE: inline back refs are ordered in the same way that back ref
1441 * items in the tree are ordered.
1443 static noinline_for_stack
1444 int lookup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1445 struct btrfs_root
*root
,
1446 struct btrfs_path
*path
,
1447 struct btrfs_extent_inline_ref
**ref_ret
,
1448 u64 bytenr
, u64 num_bytes
,
1449 u64 parent
, u64 root_objectid
,
1450 u64 owner
, u64 offset
, int insert
)
1452 struct btrfs_key key
;
1453 struct extent_buffer
*leaf
;
1454 struct btrfs_extent_item
*ei
;
1455 struct btrfs_extent_inline_ref
*iref
;
1466 key
.objectid
= bytenr
;
1467 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1468 key
.offset
= num_bytes
;
1470 want
= extent_ref_type(parent
, owner
);
1472 extra_size
= btrfs_extent_inline_ref_size(want
);
1473 path
->keep_locks
= 1;
1476 ret
= btrfs_search_slot(trans
, root
, &key
, path
, extra_size
, 1);
1483 leaf
= path
->nodes
[0];
1484 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1485 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1486 if (item_size
< sizeof(*ei
)) {
1491 ret
= convert_extent_item_v0(trans
, root
, path
, owner
,
1497 leaf
= path
->nodes
[0];
1498 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1501 BUG_ON(item_size
< sizeof(*ei
));
1503 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1504 flags
= btrfs_extent_flags(leaf
, ei
);
1506 ptr
= (unsigned long)(ei
+ 1);
1507 end
= (unsigned long)ei
+ item_size
;
1509 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
) {
1510 ptr
+= sizeof(struct btrfs_tree_block_info
);
1513 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_DATA
));
1522 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1523 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1527 ptr
+= btrfs_extent_inline_ref_size(type
);
1531 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1532 struct btrfs_extent_data_ref
*dref
;
1533 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1534 if (match_extent_data_ref(leaf
, dref
, root_objectid
,
1539 if (hash_extent_data_ref_item(leaf
, dref
) <
1540 hash_extent_data_ref(root_objectid
, owner
, offset
))
1544 ref_offset
= btrfs_extent_inline_ref_offset(leaf
, iref
);
1546 if (parent
== ref_offset
) {
1550 if (ref_offset
< parent
)
1553 if (root_objectid
== ref_offset
) {
1557 if (ref_offset
< root_objectid
)
1561 ptr
+= btrfs_extent_inline_ref_size(type
);
1563 if (err
== -ENOENT
&& insert
) {
1564 if (item_size
+ extra_size
>=
1565 BTRFS_MAX_EXTENT_ITEM_SIZE(root
)) {
1570 * To add new inline back ref, we have to make sure
1571 * there is no corresponding back ref item.
1572 * For simplicity, we just do not add new inline back
1573 * ref if there is any kind of item for this block
1575 if (find_next_key(path
, 0, &key
) == 0 &&
1576 key
.objectid
== bytenr
&&
1577 key
.type
< BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1582 *ref_ret
= (struct btrfs_extent_inline_ref
*)ptr
;
1585 path
->keep_locks
= 0;
1586 btrfs_unlock_up_safe(path
, 1);
1592 * helper to add new inline back ref
1594 static noinline_for_stack
1595 int setup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1596 struct btrfs_root
*root
,
1597 struct btrfs_path
*path
,
1598 struct btrfs_extent_inline_ref
*iref
,
1599 u64 parent
, u64 root_objectid
,
1600 u64 owner
, u64 offset
, int refs_to_add
,
1601 struct btrfs_delayed_extent_op
*extent_op
)
1603 struct extent_buffer
*leaf
;
1604 struct btrfs_extent_item
*ei
;
1607 unsigned long item_offset
;
1613 leaf
= path
->nodes
[0];
1614 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1615 item_offset
= (unsigned long)iref
- (unsigned long)ei
;
1617 type
= extent_ref_type(parent
, owner
);
1618 size
= btrfs_extent_inline_ref_size(type
);
1620 ret
= btrfs_extend_item(trans
, root
, path
, size
);
1622 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1623 refs
= btrfs_extent_refs(leaf
, ei
);
1624 refs
+= refs_to_add
;
1625 btrfs_set_extent_refs(leaf
, ei
, refs
);
1627 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1629 ptr
= (unsigned long)ei
+ item_offset
;
1630 end
= (unsigned long)ei
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
1631 if (ptr
< end
- size
)
1632 memmove_extent_buffer(leaf
, ptr
+ size
, ptr
,
1635 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1636 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
1637 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1638 struct btrfs_extent_data_ref
*dref
;
1639 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1640 btrfs_set_extent_data_ref_root(leaf
, dref
, root_objectid
);
1641 btrfs_set_extent_data_ref_objectid(leaf
, dref
, owner
);
1642 btrfs_set_extent_data_ref_offset(leaf
, dref
, offset
);
1643 btrfs_set_extent_data_ref_count(leaf
, dref
, refs_to_add
);
1644 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1645 struct btrfs_shared_data_ref
*sref
;
1646 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1647 btrfs_set_shared_data_ref_count(leaf
, sref
, refs_to_add
);
1648 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1649 } else if (type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
1650 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1652 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
1654 btrfs_mark_buffer_dirty(leaf
);
1658 static int lookup_extent_backref(struct btrfs_trans_handle
*trans
,
1659 struct btrfs_root
*root
,
1660 struct btrfs_path
*path
,
1661 struct btrfs_extent_inline_ref
**ref_ret
,
1662 u64 bytenr
, u64 num_bytes
, u64 parent
,
1663 u64 root_objectid
, u64 owner
, u64 offset
)
1667 ret
= lookup_inline_extent_backref(trans
, root
, path
, ref_ret
,
1668 bytenr
, num_bytes
, parent
,
1669 root_objectid
, owner
, offset
, 0);
1673 btrfs_release_path(path
);
1676 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1677 ret
= lookup_tree_block_ref(trans
, root
, path
, bytenr
, parent
,
1680 ret
= lookup_extent_data_ref(trans
, root
, path
, bytenr
, parent
,
1681 root_objectid
, owner
, offset
);
1687 * helper to update/remove inline back ref
1689 static noinline_for_stack
1690 int update_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1691 struct btrfs_root
*root
,
1692 struct btrfs_path
*path
,
1693 struct btrfs_extent_inline_ref
*iref
,
1695 struct btrfs_delayed_extent_op
*extent_op
)
1697 struct extent_buffer
*leaf
;
1698 struct btrfs_extent_item
*ei
;
1699 struct btrfs_extent_data_ref
*dref
= NULL
;
1700 struct btrfs_shared_data_ref
*sref
= NULL
;
1709 leaf
= path
->nodes
[0];
1710 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1711 refs
= btrfs_extent_refs(leaf
, ei
);
1712 WARN_ON(refs_to_mod
< 0 && refs
+ refs_to_mod
<= 0);
1713 refs
+= refs_to_mod
;
1714 btrfs_set_extent_refs(leaf
, ei
, refs
);
1716 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1718 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1720 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1721 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1722 refs
= btrfs_extent_data_ref_count(leaf
, dref
);
1723 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1724 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1725 refs
= btrfs_shared_data_ref_count(leaf
, sref
);
1728 BUG_ON(refs_to_mod
!= -1);
1731 BUG_ON(refs_to_mod
< 0 && refs
< -refs_to_mod
);
1732 refs
+= refs_to_mod
;
1735 if (type
== BTRFS_EXTENT_DATA_REF_KEY
)
1736 btrfs_set_extent_data_ref_count(leaf
, dref
, refs
);
1738 btrfs_set_shared_data_ref_count(leaf
, sref
, refs
);
1740 size
= btrfs_extent_inline_ref_size(type
);
1741 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1742 ptr
= (unsigned long)iref
;
1743 end
= (unsigned long)ei
+ item_size
;
1744 if (ptr
+ size
< end
)
1745 memmove_extent_buffer(leaf
, ptr
, ptr
+ size
,
1748 ret
= btrfs_truncate_item(trans
, root
, path
, item_size
, 1);
1750 btrfs_mark_buffer_dirty(leaf
);
1754 static noinline_for_stack
1755 int insert_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1756 struct btrfs_root
*root
,
1757 struct btrfs_path
*path
,
1758 u64 bytenr
, u64 num_bytes
, u64 parent
,
1759 u64 root_objectid
, u64 owner
,
1760 u64 offset
, int refs_to_add
,
1761 struct btrfs_delayed_extent_op
*extent_op
)
1763 struct btrfs_extent_inline_ref
*iref
;
1766 ret
= lookup_inline_extent_backref(trans
, root
, path
, &iref
,
1767 bytenr
, num_bytes
, parent
,
1768 root_objectid
, owner
, offset
, 1);
1770 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
);
1771 ret
= update_inline_extent_backref(trans
, root
, path
, iref
,
1772 refs_to_add
, extent_op
);
1773 } else if (ret
== -ENOENT
) {
1774 ret
= setup_inline_extent_backref(trans
, root
, path
, iref
,
1775 parent
, root_objectid
,
1776 owner
, offset
, refs_to_add
,
1782 static int insert_extent_backref(struct btrfs_trans_handle
*trans
,
1783 struct btrfs_root
*root
,
1784 struct btrfs_path
*path
,
1785 u64 bytenr
, u64 parent
, u64 root_objectid
,
1786 u64 owner
, u64 offset
, int refs_to_add
)
1789 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1790 BUG_ON(refs_to_add
!= 1);
1791 ret
= insert_tree_block_ref(trans
, root
, path
, bytenr
,
1792 parent
, root_objectid
);
1794 ret
= insert_extent_data_ref(trans
, root
, path
, bytenr
,
1795 parent
, root_objectid
,
1796 owner
, offset
, refs_to_add
);
1801 static int remove_extent_backref(struct btrfs_trans_handle
*trans
,
1802 struct btrfs_root
*root
,
1803 struct btrfs_path
*path
,
1804 struct btrfs_extent_inline_ref
*iref
,
1805 int refs_to_drop
, int is_data
)
1809 BUG_ON(!is_data
&& refs_to_drop
!= 1);
1811 ret
= update_inline_extent_backref(trans
, root
, path
, iref
,
1812 -refs_to_drop
, NULL
);
1813 } else if (is_data
) {
1814 ret
= remove_extent_data_ref(trans
, root
, path
, refs_to_drop
);
1816 ret
= btrfs_del_item(trans
, root
, path
);
1821 static int btrfs_issue_discard(struct block_device
*bdev
,
1824 return blkdev_issue_discard(bdev
, start
>> 9, len
>> 9, GFP_NOFS
, 0);
1827 static int btrfs_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
1828 u64 num_bytes
, u64
*actual_bytes
)
1831 u64 discarded_bytes
= 0;
1832 struct btrfs_bio
*bbio
= NULL
;
1835 /* Tell the block device(s) that the sectors can be discarded */
1836 ret
= btrfs_map_block(&root
->fs_info
->mapping_tree
, REQ_DISCARD
,
1837 bytenr
, &num_bytes
, &bbio
, 0);
1839 struct btrfs_bio_stripe
*stripe
= bbio
->stripes
;
1843 for (i
= 0; i
< bbio
->num_stripes
; i
++, stripe
++) {
1844 if (!stripe
->dev
->can_discard
)
1847 ret
= btrfs_issue_discard(stripe
->dev
->bdev
,
1851 discarded_bytes
+= stripe
->length
;
1852 else if (ret
!= -EOPNOTSUPP
)
1856 * Just in case we get back EOPNOTSUPP for some reason,
1857 * just ignore the return value so we don't screw up
1858 * people calling discard_extent.
1866 *actual_bytes
= discarded_bytes
;
1872 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1873 struct btrfs_root
*root
,
1874 u64 bytenr
, u64 num_bytes
, u64 parent
,
1875 u64 root_objectid
, u64 owner
, u64 offset
, int for_cow
)
1878 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1880 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
&&
1881 root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
1883 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1884 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
1886 parent
, root_objectid
, (int)owner
,
1887 BTRFS_ADD_DELAYED_REF
, NULL
, for_cow
);
1889 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
1891 parent
, root_objectid
, owner
, offset
,
1892 BTRFS_ADD_DELAYED_REF
, NULL
, for_cow
);
1897 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1898 struct btrfs_root
*root
,
1899 u64 bytenr
, u64 num_bytes
,
1900 u64 parent
, u64 root_objectid
,
1901 u64 owner
, u64 offset
, int refs_to_add
,
1902 struct btrfs_delayed_extent_op
*extent_op
)
1904 struct btrfs_path
*path
;
1905 struct extent_buffer
*leaf
;
1906 struct btrfs_extent_item
*item
;
1911 path
= btrfs_alloc_path();
1916 path
->leave_spinning
= 1;
1917 /* this will setup the path even if it fails to insert the back ref */
1918 ret
= insert_inline_extent_backref(trans
, root
->fs_info
->extent_root
,
1919 path
, bytenr
, num_bytes
, parent
,
1920 root_objectid
, owner
, offset
,
1921 refs_to_add
, extent_op
);
1925 if (ret
!= -EAGAIN
) {
1930 leaf
= path
->nodes
[0];
1931 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1932 refs
= btrfs_extent_refs(leaf
, item
);
1933 btrfs_set_extent_refs(leaf
, item
, refs
+ refs_to_add
);
1935 __run_delayed_extent_op(extent_op
, leaf
, item
);
1937 btrfs_mark_buffer_dirty(leaf
);
1938 btrfs_release_path(path
);
1941 path
->leave_spinning
= 1;
1943 /* now insert the actual backref */
1944 ret
= insert_extent_backref(trans
, root
->fs_info
->extent_root
,
1945 path
, bytenr
, parent
, root_objectid
,
1946 owner
, offset
, refs_to_add
);
1949 btrfs_free_path(path
);
1953 static int run_delayed_data_ref(struct btrfs_trans_handle
*trans
,
1954 struct btrfs_root
*root
,
1955 struct btrfs_delayed_ref_node
*node
,
1956 struct btrfs_delayed_extent_op
*extent_op
,
1957 int insert_reserved
)
1960 struct btrfs_delayed_data_ref
*ref
;
1961 struct btrfs_key ins
;
1966 ins
.objectid
= node
->bytenr
;
1967 ins
.offset
= node
->num_bytes
;
1968 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1970 ref
= btrfs_delayed_node_to_data_ref(node
);
1971 if (node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
1972 parent
= ref
->parent
;
1974 ref_root
= ref
->root
;
1976 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
1978 BUG_ON(extent_op
->update_key
);
1979 flags
|= extent_op
->flags_to_set
;
1981 ret
= alloc_reserved_file_extent(trans
, root
,
1982 parent
, ref_root
, flags
,
1983 ref
->objectid
, ref
->offset
,
1984 &ins
, node
->ref_mod
);
1985 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
1986 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
1987 node
->num_bytes
, parent
,
1988 ref_root
, ref
->objectid
,
1989 ref
->offset
, node
->ref_mod
,
1991 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
1992 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
1993 node
->num_bytes
, parent
,
1994 ref_root
, ref
->objectid
,
1995 ref
->offset
, node
->ref_mod
,
2003 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
2004 struct extent_buffer
*leaf
,
2005 struct btrfs_extent_item
*ei
)
2007 u64 flags
= btrfs_extent_flags(leaf
, ei
);
2008 if (extent_op
->update_flags
) {
2009 flags
|= extent_op
->flags_to_set
;
2010 btrfs_set_extent_flags(leaf
, ei
, flags
);
2013 if (extent_op
->update_key
) {
2014 struct btrfs_tree_block_info
*bi
;
2015 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
));
2016 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
2017 btrfs_set_tree_block_key(leaf
, bi
, &extent_op
->key
);
2021 static int run_delayed_extent_op(struct btrfs_trans_handle
*trans
,
2022 struct btrfs_root
*root
,
2023 struct btrfs_delayed_ref_node
*node
,
2024 struct btrfs_delayed_extent_op
*extent_op
)
2026 struct btrfs_key key
;
2027 struct btrfs_path
*path
;
2028 struct btrfs_extent_item
*ei
;
2029 struct extent_buffer
*leaf
;
2034 path
= btrfs_alloc_path();
2038 key
.objectid
= node
->bytenr
;
2039 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2040 key
.offset
= node
->num_bytes
;
2043 path
->leave_spinning
= 1;
2044 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
,
2055 leaf
= path
->nodes
[0];
2056 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2057 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2058 if (item_size
< sizeof(*ei
)) {
2059 ret
= convert_extent_item_v0(trans
, root
->fs_info
->extent_root
,
2065 leaf
= path
->nodes
[0];
2066 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2069 BUG_ON(item_size
< sizeof(*ei
));
2070 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2071 __run_delayed_extent_op(extent_op
, leaf
, ei
);
2073 btrfs_mark_buffer_dirty(leaf
);
2075 btrfs_free_path(path
);
2079 static int run_delayed_tree_ref(struct btrfs_trans_handle
*trans
,
2080 struct btrfs_root
*root
,
2081 struct btrfs_delayed_ref_node
*node
,
2082 struct btrfs_delayed_extent_op
*extent_op
,
2083 int insert_reserved
)
2086 struct btrfs_delayed_tree_ref
*ref
;
2087 struct btrfs_key ins
;
2091 ins
.objectid
= node
->bytenr
;
2092 ins
.offset
= node
->num_bytes
;
2093 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2095 ref
= btrfs_delayed_node_to_tree_ref(node
);
2096 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2097 parent
= ref
->parent
;
2099 ref_root
= ref
->root
;
2101 BUG_ON(node
->ref_mod
!= 1);
2102 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2103 BUG_ON(!extent_op
|| !extent_op
->update_flags
||
2104 !extent_op
->update_key
);
2105 ret
= alloc_reserved_tree_block(trans
, root
,
2107 extent_op
->flags_to_set
,
2110 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2111 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
2112 node
->num_bytes
, parent
, ref_root
,
2113 ref
->level
, 0, 1, extent_op
);
2114 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2115 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
2116 node
->num_bytes
, parent
, ref_root
,
2117 ref
->level
, 0, 1, extent_op
);
2124 /* helper function to actually process a single delayed ref entry */
2125 static int run_one_delayed_ref(struct btrfs_trans_handle
*trans
,
2126 struct btrfs_root
*root
,
2127 struct btrfs_delayed_ref_node
*node
,
2128 struct btrfs_delayed_extent_op
*extent_op
,
2129 int insert_reserved
)
2132 if (btrfs_delayed_ref_is_head(node
)) {
2133 struct btrfs_delayed_ref_head
*head
;
2135 * we've hit the end of the chain and we were supposed
2136 * to insert this extent into the tree. But, it got
2137 * deleted before we ever needed to insert it, so all
2138 * we have to do is clean up the accounting
2141 head
= btrfs_delayed_node_to_head(node
);
2142 if (insert_reserved
) {
2143 btrfs_pin_extent(root
, node
->bytenr
,
2144 node
->num_bytes
, 1);
2145 if (head
->is_data
) {
2146 ret
= btrfs_del_csums(trans
, root
,
2152 mutex_unlock(&head
->mutex
);
2156 if (node
->type
== BTRFS_TREE_BLOCK_REF_KEY
||
2157 node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2158 ret
= run_delayed_tree_ref(trans
, root
, node
, extent_op
,
2160 else if (node
->type
== BTRFS_EXTENT_DATA_REF_KEY
||
2161 node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2162 ret
= run_delayed_data_ref(trans
, root
, node
, extent_op
,
2169 static noinline
struct btrfs_delayed_ref_node
*
2170 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
2172 struct rb_node
*node
;
2173 struct btrfs_delayed_ref_node
*ref
;
2174 int action
= BTRFS_ADD_DELAYED_REF
;
2177 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2178 * this prevents ref count from going down to zero when
2179 * there still are pending delayed ref.
2181 node
= rb_prev(&head
->node
.rb_node
);
2185 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
2187 if (ref
->bytenr
!= head
->node
.bytenr
)
2189 if (ref
->action
== action
)
2191 node
= rb_prev(node
);
2193 if (action
== BTRFS_ADD_DELAYED_REF
) {
2194 action
= BTRFS_DROP_DELAYED_REF
;
2200 static noinline
int run_clustered_refs(struct btrfs_trans_handle
*trans
,
2201 struct btrfs_root
*root
,
2202 struct list_head
*cluster
)
2204 struct btrfs_delayed_ref_root
*delayed_refs
;
2205 struct btrfs_delayed_ref_node
*ref
;
2206 struct btrfs_delayed_ref_head
*locked_ref
= NULL
;
2207 struct btrfs_delayed_extent_op
*extent_op
;
2210 int must_insert_reserved
= 0;
2212 delayed_refs
= &trans
->transaction
->delayed_refs
;
2215 /* pick a new head ref from the cluster list */
2216 if (list_empty(cluster
))
2219 locked_ref
= list_entry(cluster
->next
,
2220 struct btrfs_delayed_ref_head
, cluster
);
2222 /* grab the lock that says we are going to process
2223 * all the refs for this head */
2224 ret
= btrfs_delayed_ref_lock(trans
, locked_ref
);
2227 * we may have dropped the spin lock to get the head
2228 * mutex lock, and that might have given someone else
2229 * time to free the head. If that's true, it has been
2230 * removed from our list and we can move on.
2232 if (ret
== -EAGAIN
) {
2240 * record the must insert reserved flag before we
2241 * drop the spin lock.
2243 must_insert_reserved
= locked_ref
->must_insert_reserved
;
2244 locked_ref
->must_insert_reserved
= 0;
2246 extent_op
= locked_ref
->extent_op
;
2247 locked_ref
->extent_op
= NULL
;
2250 * locked_ref is the head node, so we have to go one
2251 * node back for any delayed ref updates
2253 ref
= select_delayed_ref(locked_ref
);
2255 /* All delayed refs have been processed, Go ahead
2256 * and send the head node to run_one_delayed_ref,
2257 * so that any accounting fixes can happen
2259 ref
= &locked_ref
->node
;
2261 if (extent_op
&& must_insert_reserved
) {
2267 spin_unlock(&delayed_refs
->lock
);
2269 ret
= run_delayed_extent_op(trans
, root
,
2275 spin_lock(&delayed_refs
->lock
);
2279 list_del_init(&locked_ref
->cluster
);
2284 rb_erase(&ref
->rb_node
, &delayed_refs
->root
);
2285 delayed_refs
->num_entries
--;
2287 spin_unlock(&delayed_refs
->lock
);
2289 ret
= run_one_delayed_ref(trans
, root
, ref
, extent_op
,
2290 must_insert_reserved
);
2293 btrfs_put_delayed_ref(ref
);
2298 spin_lock(&delayed_refs
->lock
);
2304 * this starts processing the delayed reference count updates and
2305 * extent insertions we have queued up so far. count can be
2306 * 0, which means to process everything in the tree at the start
2307 * of the run (but not newly added entries), or it can be some target
2308 * number you'd like to process.
2310 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2311 struct btrfs_root
*root
, unsigned long count
)
2313 struct rb_node
*node
;
2314 struct btrfs_delayed_ref_root
*delayed_refs
;
2315 struct btrfs_delayed_ref_node
*ref
;
2316 struct list_head cluster
;
2318 int run_all
= count
== (unsigned long)-1;
2321 if (root
== root
->fs_info
->extent_root
)
2322 root
= root
->fs_info
->tree_root
;
2324 delayed_refs
= &trans
->transaction
->delayed_refs
;
2325 INIT_LIST_HEAD(&cluster
);
2327 spin_lock(&delayed_refs
->lock
);
2329 count
= delayed_refs
->num_entries
* 2;
2333 if (!(run_all
|| run_most
) &&
2334 delayed_refs
->num_heads_ready
< 64)
2338 * go find something we can process in the rbtree. We start at
2339 * the beginning of the tree, and then build a cluster
2340 * of refs to process starting at the first one we are able to
2343 ret
= btrfs_find_ref_cluster(trans
, &cluster
,
2344 delayed_refs
->run_delayed_start
);
2348 ret
= run_clustered_refs(trans
, root
, &cluster
);
2351 count
-= min_t(unsigned long, ret
, count
);
2358 node
= rb_first(&delayed_refs
->root
);
2361 count
= (unsigned long)-1;
2364 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
2366 if (btrfs_delayed_ref_is_head(ref
)) {
2367 struct btrfs_delayed_ref_head
*head
;
2369 head
= btrfs_delayed_node_to_head(ref
);
2370 atomic_inc(&ref
->refs
);
2372 spin_unlock(&delayed_refs
->lock
);
2374 * Mutex was contended, block until it's
2375 * released and try again
2377 mutex_lock(&head
->mutex
);
2378 mutex_unlock(&head
->mutex
);
2380 btrfs_put_delayed_ref(ref
);
2384 node
= rb_next(node
);
2386 spin_unlock(&delayed_refs
->lock
);
2387 schedule_timeout(1);
2391 spin_unlock(&delayed_refs
->lock
);
2395 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
2396 struct btrfs_root
*root
,
2397 u64 bytenr
, u64 num_bytes
, u64 flags
,
2400 struct btrfs_delayed_extent_op
*extent_op
;
2403 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
2407 extent_op
->flags_to_set
= flags
;
2408 extent_op
->update_flags
= 1;
2409 extent_op
->update_key
= 0;
2410 extent_op
->is_data
= is_data
? 1 : 0;
2412 ret
= btrfs_add_delayed_extent_op(root
->fs_info
, trans
, bytenr
,
2413 num_bytes
, extent_op
);
2419 static noinline
int check_delayed_ref(struct btrfs_trans_handle
*trans
,
2420 struct btrfs_root
*root
,
2421 struct btrfs_path
*path
,
2422 u64 objectid
, u64 offset
, u64 bytenr
)
2424 struct btrfs_delayed_ref_head
*head
;
2425 struct btrfs_delayed_ref_node
*ref
;
2426 struct btrfs_delayed_data_ref
*data_ref
;
2427 struct btrfs_delayed_ref_root
*delayed_refs
;
2428 struct rb_node
*node
;
2432 delayed_refs
= &trans
->transaction
->delayed_refs
;
2433 spin_lock(&delayed_refs
->lock
);
2434 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
2438 if (!mutex_trylock(&head
->mutex
)) {
2439 atomic_inc(&head
->node
.refs
);
2440 spin_unlock(&delayed_refs
->lock
);
2442 btrfs_release_path(path
);
2445 * Mutex was contended, block until it's released and let
2448 mutex_lock(&head
->mutex
);
2449 mutex_unlock(&head
->mutex
);
2450 btrfs_put_delayed_ref(&head
->node
);
2454 node
= rb_prev(&head
->node
.rb_node
);
2458 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2460 if (ref
->bytenr
!= bytenr
)
2464 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
)
2467 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
2469 node
= rb_prev(node
);
2471 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2472 if (ref
->bytenr
== bytenr
)
2476 if (data_ref
->root
!= root
->root_key
.objectid
||
2477 data_ref
->objectid
!= objectid
|| data_ref
->offset
!= offset
)
2482 mutex_unlock(&head
->mutex
);
2484 spin_unlock(&delayed_refs
->lock
);
2488 static noinline
int check_committed_ref(struct btrfs_trans_handle
*trans
,
2489 struct btrfs_root
*root
,
2490 struct btrfs_path
*path
,
2491 u64 objectid
, u64 offset
, u64 bytenr
)
2493 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2494 struct extent_buffer
*leaf
;
2495 struct btrfs_extent_data_ref
*ref
;
2496 struct btrfs_extent_inline_ref
*iref
;
2497 struct btrfs_extent_item
*ei
;
2498 struct btrfs_key key
;
2502 key
.objectid
= bytenr
;
2503 key
.offset
= (u64
)-1;
2504 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2506 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
2512 if (path
->slots
[0] == 0)
2516 leaf
= path
->nodes
[0];
2517 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2519 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
2523 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2524 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2525 if (item_size
< sizeof(*ei
)) {
2526 WARN_ON(item_size
!= sizeof(struct btrfs_extent_item_v0
));
2530 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2532 if (item_size
!= sizeof(*ei
) +
2533 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
2536 if (btrfs_extent_generation(leaf
, ei
) <=
2537 btrfs_root_last_snapshot(&root
->root_item
))
2540 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
2541 if (btrfs_extent_inline_ref_type(leaf
, iref
) !=
2542 BTRFS_EXTENT_DATA_REF_KEY
)
2545 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
2546 if (btrfs_extent_refs(leaf
, ei
) !=
2547 btrfs_extent_data_ref_count(leaf
, ref
) ||
2548 btrfs_extent_data_ref_root(leaf
, ref
) !=
2549 root
->root_key
.objectid
||
2550 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
2551 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
2559 int btrfs_cross_ref_exist(struct btrfs_trans_handle
*trans
,
2560 struct btrfs_root
*root
,
2561 u64 objectid
, u64 offset
, u64 bytenr
)
2563 struct btrfs_path
*path
;
2567 path
= btrfs_alloc_path();
2572 ret
= check_committed_ref(trans
, root
, path
, objectid
,
2574 if (ret
&& ret
!= -ENOENT
)
2577 ret2
= check_delayed_ref(trans
, root
, path
, objectid
,
2579 } while (ret2
== -EAGAIN
);
2581 if (ret2
&& ret2
!= -ENOENT
) {
2586 if (ret
!= -ENOENT
|| ret2
!= -ENOENT
)
2589 btrfs_free_path(path
);
2590 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
2595 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
2596 struct btrfs_root
*root
,
2597 struct extent_buffer
*buf
,
2598 int full_backref
, int inc
, int for_cow
)
2605 struct btrfs_key key
;
2606 struct btrfs_file_extent_item
*fi
;
2610 int (*process_func
)(struct btrfs_trans_handle
*, struct btrfs_root
*,
2611 u64
, u64
, u64
, u64
, u64
, u64
, int);
2613 ref_root
= btrfs_header_owner(buf
);
2614 nritems
= btrfs_header_nritems(buf
);
2615 level
= btrfs_header_level(buf
);
2617 if (!root
->ref_cows
&& level
== 0)
2621 process_func
= btrfs_inc_extent_ref
;
2623 process_func
= btrfs_free_extent
;
2626 parent
= buf
->start
;
2630 for (i
= 0; i
< nritems
; i
++) {
2632 btrfs_item_key_to_cpu(buf
, &key
, i
);
2633 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
2635 fi
= btrfs_item_ptr(buf
, i
,
2636 struct btrfs_file_extent_item
);
2637 if (btrfs_file_extent_type(buf
, fi
) ==
2638 BTRFS_FILE_EXTENT_INLINE
)
2640 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
2644 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
2645 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
2646 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
2647 parent
, ref_root
, key
.objectid
,
2648 key
.offset
, for_cow
);
2652 bytenr
= btrfs_node_blockptr(buf
, i
);
2653 num_bytes
= btrfs_level_size(root
, level
- 1);
2654 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
2655 parent
, ref_root
, level
- 1, 0,
2667 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2668 struct extent_buffer
*buf
, int full_backref
, int for_cow
)
2670 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1, for_cow
);
2673 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2674 struct extent_buffer
*buf
, int full_backref
, int for_cow
)
2676 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0, for_cow
);
2679 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
2680 struct btrfs_root
*root
,
2681 struct btrfs_path
*path
,
2682 struct btrfs_block_group_cache
*cache
)
2685 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2687 struct extent_buffer
*leaf
;
2689 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
2694 leaf
= path
->nodes
[0];
2695 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
2696 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
2697 btrfs_mark_buffer_dirty(leaf
);
2698 btrfs_release_path(path
);
2706 static struct btrfs_block_group_cache
*
2707 next_block_group(struct btrfs_root
*root
,
2708 struct btrfs_block_group_cache
*cache
)
2710 struct rb_node
*node
;
2711 spin_lock(&root
->fs_info
->block_group_cache_lock
);
2712 node
= rb_next(&cache
->cache_node
);
2713 btrfs_put_block_group(cache
);
2715 cache
= rb_entry(node
, struct btrfs_block_group_cache
,
2717 btrfs_get_block_group(cache
);
2720 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
2724 static int cache_save_setup(struct btrfs_block_group_cache
*block_group
,
2725 struct btrfs_trans_handle
*trans
,
2726 struct btrfs_path
*path
)
2728 struct btrfs_root
*root
= block_group
->fs_info
->tree_root
;
2729 struct inode
*inode
= NULL
;
2731 int dcs
= BTRFS_DC_ERROR
;
2737 * If this block group is smaller than 100 megs don't bother caching the
2740 if (block_group
->key
.offset
< (100 * 1024 * 1024)) {
2741 spin_lock(&block_group
->lock
);
2742 block_group
->disk_cache_state
= BTRFS_DC_WRITTEN
;
2743 spin_unlock(&block_group
->lock
);
2748 inode
= lookup_free_space_inode(root
, block_group
, path
);
2749 if (IS_ERR(inode
) && PTR_ERR(inode
) != -ENOENT
) {
2750 ret
= PTR_ERR(inode
);
2751 btrfs_release_path(path
);
2755 if (IS_ERR(inode
)) {
2759 if (block_group
->ro
)
2762 ret
= create_free_space_inode(root
, trans
, block_group
, path
);
2768 /* We've already setup this transaction, go ahead and exit */
2769 if (block_group
->cache_generation
== trans
->transid
&&
2770 i_size_read(inode
)) {
2771 dcs
= BTRFS_DC_SETUP
;
2776 * We want to set the generation to 0, that way if anything goes wrong
2777 * from here on out we know not to trust this cache when we load up next
2780 BTRFS_I(inode
)->generation
= 0;
2781 ret
= btrfs_update_inode(trans
, root
, inode
);
2784 if (i_size_read(inode
) > 0) {
2785 ret
= btrfs_truncate_free_space_cache(root
, trans
, path
,
2791 spin_lock(&block_group
->lock
);
2792 if (block_group
->cached
!= BTRFS_CACHE_FINISHED
) {
2793 /* We're not cached, don't bother trying to write stuff out */
2794 dcs
= BTRFS_DC_WRITTEN
;
2795 spin_unlock(&block_group
->lock
);
2798 spin_unlock(&block_group
->lock
);
2800 num_pages
= (int)div64_u64(block_group
->key
.offset
, 1024 * 1024 * 1024);
2805 * Just to make absolutely sure we have enough space, we're going to
2806 * preallocate 12 pages worth of space for each block group. In
2807 * practice we ought to use at most 8, but we need extra space so we can
2808 * add our header and have a terminator between the extents and the
2812 num_pages
*= PAGE_CACHE_SIZE
;
2814 ret
= btrfs_check_data_free_space(inode
, num_pages
);
2818 ret
= btrfs_prealloc_file_range_trans(inode
, trans
, 0, 0, num_pages
,
2819 num_pages
, num_pages
,
2822 dcs
= BTRFS_DC_SETUP
;
2823 btrfs_free_reserved_data_space(inode
, num_pages
);
2828 btrfs_release_path(path
);
2830 spin_lock(&block_group
->lock
);
2832 block_group
->cache_generation
= trans
->transid
;
2833 block_group
->disk_cache_state
= dcs
;
2834 spin_unlock(&block_group
->lock
);
2839 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
2840 struct btrfs_root
*root
)
2842 struct btrfs_block_group_cache
*cache
;
2844 struct btrfs_path
*path
;
2847 path
= btrfs_alloc_path();
2853 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
2855 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
)
2857 cache
= next_block_group(root
, cache
);
2865 err
= cache_save_setup(cache
, trans
, path
);
2866 last
= cache
->key
.objectid
+ cache
->key
.offset
;
2867 btrfs_put_block_group(cache
);
2872 err
= btrfs_run_delayed_refs(trans
, root
,
2877 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
2879 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
) {
2880 btrfs_put_block_group(cache
);
2886 cache
= next_block_group(root
, cache
);
2895 if (cache
->disk_cache_state
== BTRFS_DC_SETUP
)
2896 cache
->disk_cache_state
= BTRFS_DC_NEED_WRITE
;
2898 last
= cache
->key
.objectid
+ cache
->key
.offset
;
2900 err
= write_one_cache_group(trans
, root
, path
, cache
);
2902 btrfs_put_block_group(cache
);
2907 * I don't think this is needed since we're just marking our
2908 * preallocated extent as written, but just in case it can't
2912 err
= btrfs_run_delayed_refs(trans
, root
,
2917 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
2920 * Really this shouldn't happen, but it could if we
2921 * couldn't write the entire preallocated extent and
2922 * splitting the extent resulted in a new block.
2925 btrfs_put_block_group(cache
);
2928 if (cache
->disk_cache_state
== BTRFS_DC_NEED_WRITE
)
2930 cache
= next_block_group(root
, cache
);
2939 btrfs_write_out_cache(root
, trans
, cache
, path
);
2942 * If we didn't have an error then the cache state is still
2943 * NEED_WRITE, so we can set it to WRITTEN.
2945 if (cache
->disk_cache_state
== BTRFS_DC_NEED_WRITE
)
2946 cache
->disk_cache_state
= BTRFS_DC_WRITTEN
;
2947 last
= cache
->key
.objectid
+ cache
->key
.offset
;
2948 btrfs_put_block_group(cache
);
2951 btrfs_free_path(path
);
2955 int btrfs_extent_readonly(struct btrfs_root
*root
, u64 bytenr
)
2957 struct btrfs_block_group_cache
*block_group
;
2960 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
2961 if (!block_group
|| block_group
->ro
)
2964 btrfs_put_block_group(block_group
);
2968 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
2969 u64 total_bytes
, u64 bytes_used
,
2970 struct btrfs_space_info
**space_info
)
2972 struct btrfs_space_info
*found
;
2976 if (flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
2977 BTRFS_BLOCK_GROUP_RAID10
))
2982 found
= __find_space_info(info
, flags
);
2984 spin_lock(&found
->lock
);
2985 found
->total_bytes
+= total_bytes
;
2986 found
->disk_total
+= total_bytes
* factor
;
2987 found
->bytes_used
+= bytes_used
;
2988 found
->disk_used
+= bytes_used
* factor
;
2990 spin_unlock(&found
->lock
);
2991 *space_info
= found
;
2994 found
= kzalloc(sizeof(*found
), GFP_NOFS
);
2998 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
2999 INIT_LIST_HEAD(&found
->block_groups
[i
]);
3000 init_rwsem(&found
->groups_sem
);
3001 spin_lock_init(&found
->lock
);
3002 found
->flags
= flags
& (BTRFS_BLOCK_GROUP_DATA
|
3003 BTRFS_BLOCK_GROUP_SYSTEM
|
3004 BTRFS_BLOCK_GROUP_METADATA
);
3005 found
->total_bytes
= total_bytes
;
3006 found
->disk_total
= total_bytes
* factor
;
3007 found
->bytes_used
= bytes_used
;
3008 found
->disk_used
= bytes_used
* factor
;
3009 found
->bytes_pinned
= 0;
3010 found
->bytes_reserved
= 0;
3011 found
->bytes_readonly
= 0;
3012 found
->bytes_may_use
= 0;
3014 found
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3015 found
->chunk_alloc
= 0;
3017 init_waitqueue_head(&found
->wait
);
3018 *space_info
= found
;
3019 list_add_rcu(&found
->list
, &info
->space_info
);
3023 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
3025 u64 extra_flags
= flags
& (BTRFS_BLOCK_GROUP_RAID0
|
3026 BTRFS_BLOCK_GROUP_RAID1
|
3027 BTRFS_BLOCK_GROUP_RAID10
|
3028 BTRFS_BLOCK_GROUP_DUP
);
3030 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3031 fs_info
->avail_data_alloc_bits
|= extra_flags
;
3032 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3033 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
3034 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3035 fs_info
->avail_system_alloc_bits
|= extra_flags
;
3039 u64
btrfs_reduce_alloc_profile(struct btrfs_root
*root
, u64 flags
)
3042 * we add in the count of missing devices because we want
3043 * to make sure that any RAID levels on a degraded FS
3044 * continue to be honored.
3046 u64 num_devices
= root
->fs_info
->fs_devices
->rw_devices
+
3047 root
->fs_info
->fs_devices
->missing_devices
;
3049 if (num_devices
== 1)
3050 flags
&= ~(BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID0
);
3051 if (num_devices
< 4)
3052 flags
&= ~BTRFS_BLOCK_GROUP_RAID10
;
3054 if ((flags
& BTRFS_BLOCK_GROUP_DUP
) &&
3055 (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
3056 BTRFS_BLOCK_GROUP_RAID10
))) {
3057 flags
&= ~BTRFS_BLOCK_GROUP_DUP
;
3060 if ((flags
& BTRFS_BLOCK_GROUP_RAID1
) &&
3061 (flags
& BTRFS_BLOCK_GROUP_RAID10
)) {
3062 flags
&= ~BTRFS_BLOCK_GROUP_RAID1
;
3065 if ((flags
& BTRFS_BLOCK_GROUP_RAID0
) &&
3066 ((flags
& BTRFS_BLOCK_GROUP_RAID1
) |
3067 (flags
& BTRFS_BLOCK_GROUP_RAID10
) |
3068 (flags
& BTRFS_BLOCK_GROUP_DUP
)))
3069 flags
&= ~BTRFS_BLOCK_GROUP_RAID0
;
3073 static u64
get_alloc_profile(struct btrfs_root
*root
, u64 flags
)
3075 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3076 flags
|= root
->fs_info
->avail_data_alloc_bits
&
3077 root
->fs_info
->data_alloc_profile
;
3078 else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3079 flags
|= root
->fs_info
->avail_system_alloc_bits
&
3080 root
->fs_info
->system_alloc_profile
;
3081 else if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3082 flags
|= root
->fs_info
->avail_metadata_alloc_bits
&
3083 root
->fs_info
->metadata_alloc_profile
;
3084 return btrfs_reduce_alloc_profile(root
, flags
);
3087 u64
btrfs_get_alloc_profile(struct btrfs_root
*root
, int data
)
3092 flags
= BTRFS_BLOCK_GROUP_DATA
;
3093 else if (root
== root
->fs_info
->chunk_root
)
3094 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
3096 flags
= BTRFS_BLOCK_GROUP_METADATA
;
3098 return get_alloc_profile(root
, flags
);
3101 void btrfs_set_inode_space_info(struct btrfs_root
*root
, struct inode
*inode
)
3103 BTRFS_I(inode
)->space_info
= __find_space_info(root
->fs_info
,
3104 BTRFS_BLOCK_GROUP_DATA
);
3108 * This will check the space that the inode allocates from to make sure we have
3109 * enough space for bytes.
3111 int btrfs_check_data_free_space(struct inode
*inode
, u64 bytes
)
3113 struct btrfs_space_info
*data_sinfo
;
3114 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3116 int ret
= 0, committed
= 0, alloc_chunk
= 1;
3118 /* make sure bytes are sectorsize aligned */
3119 bytes
= (bytes
+ root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
3121 if (root
== root
->fs_info
->tree_root
||
3122 BTRFS_I(inode
)->location
.objectid
== BTRFS_FREE_INO_OBJECTID
) {
3127 data_sinfo
= BTRFS_I(inode
)->space_info
;
3132 /* make sure we have enough space to handle the data first */
3133 spin_lock(&data_sinfo
->lock
);
3134 used
= data_sinfo
->bytes_used
+ data_sinfo
->bytes_reserved
+
3135 data_sinfo
->bytes_pinned
+ data_sinfo
->bytes_readonly
+
3136 data_sinfo
->bytes_may_use
;
3138 if (used
+ bytes
> data_sinfo
->total_bytes
) {
3139 struct btrfs_trans_handle
*trans
;
3142 * if we don't have enough free bytes in this space then we need
3143 * to alloc a new chunk.
3145 if (!data_sinfo
->full
&& alloc_chunk
) {
3148 data_sinfo
->force_alloc
= CHUNK_ALLOC_FORCE
;
3149 spin_unlock(&data_sinfo
->lock
);
3151 alloc_target
= btrfs_get_alloc_profile(root
, 1);
3152 trans
= btrfs_join_transaction(root
);
3154 return PTR_ERR(trans
);
3156 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
3157 bytes
+ 2 * 1024 * 1024,
3159 CHUNK_ALLOC_NO_FORCE
);
3160 btrfs_end_transaction(trans
, root
);
3169 btrfs_set_inode_space_info(root
, inode
);
3170 data_sinfo
= BTRFS_I(inode
)->space_info
;
3176 * If we have less pinned bytes than we want to allocate then
3177 * don't bother committing the transaction, it won't help us.
3179 if (data_sinfo
->bytes_pinned
< bytes
)
3181 spin_unlock(&data_sinfo
->lock
);
3183 /* commit the current transaction and try again */
3186 !atomic_read(&root
->fs_info
->open_ioctl_trans
)) {
3188 trans
= btrfs_join_transaction(root
);
3190 return PTR_ERR(trans
);
3191 ret
= btrfs_commit_transaction(trans
, root
);
3199 data_sinfo
->bytes_may_use
+= bytes
;
3200 spin_unlock(&data_sinfo
->lock
);
3206 * Called if we need to clear a data reservation for this inode.
3208 void btrfs_free_reserved_data_space(struct inode
*inode
, u64 bytes
)
3210 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3211 struct btrfs_space_info
*data_sinfo
;
3213 /* make sure bytes are sectorsize aligned */
3214 bytes
= (bytes
+ root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
3216 data_sinfo
= BTRFS_I(inode
)->space_info
;
3217 spin_lock(&data_sinfo
->lock
);
3218 data_sinfo
->bytes_may_use
-= bytes
;
3219 spin_unlock(&data_sinfo
->lock
);
3222 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
3224 struct list_head
*head
= &info
->space_info
;
3225 struct btrfs_space_info
*found
;
3228 list_for_each_entry_rcu(found
, head
, list
) {
3229 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
3230 found
->force_alloc
= CHUNK_ALLOC_FORCE
;
3235 static int should_alloc_chunk(struct btrfs_root
*root
,
3236 struct btrfs_space_info
*sinfo
, u64 alloc_bytes
,
3239 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
3240 u64 num_bytes
= sinfo
->total_bytes
- sinfo
->bytes_readonly
;
3241 u64 num_allocated
= sinfo
->bytes_used
+ sinfo
->bytes_reserved
;
3244 if (force
== CHUNK_ALLOC_FORCE
)
3248 * We need to take into account the global rsv because for all intents
3249 * and purposes it's used space. Don't worry about locking the
3250 * global_rsv, it doesn't change except when the transaction commits.
3252 num_allocated
+= global_rsv
->size
;
3255 * in limited mode, we want to have some free space up to
3256 * about 1% of the FS size.
3258 if (force
== CHUNK_ALLOC_LIMITED
) {
3259 thresh
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
3260 thresh
= max_t(u64
, 64 * 1024 * 1024,
3261 div_factor_fine(thresh
, 1));
3263 if (num_bytes
- num_allocated
< thresh
)
3268 * we have two similar checks here, one based on percentage
3269 * and once based on a hard number of 256MB. The idea
3270 * is that if we have a good amount of free
3271 * room, don't allocate a chunk. A good mount is
3272 * less than 80% utilized of the chunks we have allocated,
3273 * or more than 256MB free
3275 if (num_allocated
+ alloc_bytes
+ 256 * 1024 * 1024 < num_bytes
)
3278 if (num_allocated
+ alloc_bytes
< div_factor(num_bytes
, 8))
3281 thresh
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
3283 /* 256MB or 5% of the FS */
3284 thresh
= max_t(u64
, 256 * 1024 * 1024, div_factor_fine(thresh
, 5));
3286 if (num_bytes
> thresh
&& sinfo
->bytes_used
< div_factor(num_bytes
, 3))
3291 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
3292 struct btrfs_root
*extent_root
, u64 alloc_bytes
,
3293 u64 flags
, int force
)
3295 struct btrfs_space_info
*space_info
;
3296 struct btrfs_fs_info
*fs_info
= extent_root
->fs_info
;
3297 int wait_for_alloc
= 0;
3300 flags
= btrfs_reduce_alloc_profile(extent_root
, flags
);
3302 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
3304 ret
= update_space_info(extent_root
->fs_info
, flags
,
3308 BUG_ON(!space_info
);
3311 spin_lock(&space_info
->lock
);
3312 if (space_info
->force_alloc
)
3313 force
= space_info
->force_alloc
;
3314 if (space_info
->full
) {
3315 spin_unlock(&space_info
->lock
);
3319 if (!should_alloc_chunk(extent_root
, space_info
, alloc_bytes
, force
)) {
3320 spin_unlock(&space_info
->lock
);
3322 } else if (space_info
->chunk_alloc
) {
3325 space_info
->chunk_alloc
= 1;
3328 spin_unlock(&space_info
->lock
);
3330 mutex_lock(&fs_info
->chunk_mutex
);
3333 * The chunk_mutex is held throughout the entirety of a chunk
3334 * allocation, so once we've acquired the chunk_mutex we know that the
3335 * other guy is done and we need to recheck and see if we should
3338 if (wait_for_alloc
) {
3339 mutex_unlock(&fs_info
->chunk_mutex
);
3345 * If we have mixed data/metadata chunks we want to make sure we keep
3346 * allocating mixed chunks instead of individual chunks.
3348 if (btrfs_mixed_space_info(space_info
))
3349 flags
|= (BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
);
3352 * if we're doing a data chunk, go ahead and make sure that
3353 * we keep a reasonable number of metadata chunks allocated in the
3356 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
3357 fs_info
->data_chunk_allocations
++;
3358 if (!(fs_info
->data_chunk_allocations
%
3359 fs_info
->metadata_ratio
))
3360 force_metadata_allocation(fs_info
);
3363 ret
= btrfs_alloc_chunk(trans
, extent_root
, flags
);
3364 if (ret
< 0 && ret
!= -ENOSPC
)
3367 spin_lock(&space_info
->lock
);
3369 space_info
->full
= 1;
3373 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3374 space_info
->chunk_alloc
= 0;
3375 spin_unlock(&space_info
->lock
);
3377 mutex_unlock(&extent_root
->fs_info
->chunk_mutex
);
3382 * shrink metadata reservation for delalloc
3384 static int shrink_delalloc(struct btrfs_root
*root
, u64 to_reclaim
,
3387 struct btrfs_block_rsv
*block_rsv
;
3388 struct btrfs_space_info
*space_info
;
3389 struct btrfs_trans_handle
*trans
;
3394 unsigned long nr_pages
= (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT
;
3396 unsigned long progress
;
3398 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
3399 block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
3400 space_info
= block_rsv
->space_info
;
3403 reserved
= space_info
->bytes_may_use
;
3404 progress
= space_info
->reservation_progress
;
3410 if (root
->fs_info
->delalloc_bytes
== 0) {
3413 btrfs_wait_ordered_extents(root
, 0, 0);
3417 max_reclaim
= min(reserved
, to_reclaim
);
3418 nr_pages
= max_t(unsigned long, nr_pages
,
3419 max_reclaim
>> PAGE_CACHE_SHIFT
);
3420 while (loops
< 1024) {
3421 /* have the flusher threads jump in and do some IO */
3423 nr_pages
= min_t(unsigned long, nr_pages
,
3424 root
->fs_info
->delalloc_bytes
>> PAGE_CACHE_SHIFT
);
3425 writeback_inodes_sb_nr_if_idle(root
->fs_info
->sb
, nr_pages
);
3427 spin_lock(&space_info
->lock
);
3428 if (reserved
> space_info
->bytes_may_use
)
3429 reclaimed
+= reserved
- space_info
->bytes_may_use
;
3430 reserved
= space_info
->bytes_may_use
;
3431 spin_unlock(&space_info
->lock
);
3435 if (reserved
== 0 || reclaimed
>= max_reclaim
)
3438 if (trans
&& trans
->transaction
->blocked
)
3441 if (wait_ordered
&& !trans
) {
3442 btrfs_wait_ordered_extents(root
, 0, 0);
3444 time_left
= schedule_timeout_interruptible(1);
3446 /* We were interrupted, exit */
3451 /* we've kicked the IO a few times, if anything has been freed,
3452 * exit. There is no sense in looping here for a long time
3453 * when we really need to commit the transaction, or there are
3454 * just too many writers without enough free space
3459 if (progress
!= space_info
->reservation_progress
)
3465 return reclaimed
>= to_reclaim
;
3469 * maybe_commit_transaction - possibly commit the transaction if its ok to
3470 * @root - the root we're allocating for
3471 * @bytes - the number of bytes we want to reserve
3472 * @force - force the commit
3474 * This will check to make sure that committing the transaction will actually
3475 * get us somewhere and then commit the transaction if it does. Otherwise it
3476 * will return -ENOSPC.
3478 static int may_commit_transaction(struct btrfs_root
*root
,
3479 struct btrfs_space_info
*space_info
,
3480 u64 bytes
, int force
)
3482 struct btrfs_block_rsv
*delayed_rsv
= &root
->fs_info
->delayed_block_rsv
;
3483 struct btrfs_trans_handle
*trans
;
3485 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
3492 /* See if there is enough pinned space to make this reservation */
3493 spin_lock(&space_info
->lock
);
3494 if (space_info
->bytes_pinned
>= bytes
) {
3495 spin_unlock(&space_info
->lock
);
3498 spin_unlock(&space_info
->lock
);
3501 * See if there is some space in the delayed insertion reservation for
3504 if (space_info
!= delayed_rsv
->space_info
)
3507 spin_lock(&delayed_rsv
->lock
);
3508 if (delayed_rsv
->size
< bytes
) {
3509 spin_unlock(&delayed_rsv
->lock
);
3512 spin_unlock(&delayed_rsv
->lock
);
3515 trans
= btrfs_join_transaction(root
);
3519 return btrfs_commit_transaction(trans
, root
);
3523 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3524 * @root - the root we're allocating for
3525 * @block_rsv - the block_rsv we're allocating for
3526 * @orig_bytes - the number of bytes we want
3527 * @flush - wether or not we can flush to make our reservation
3529 * This will reserve orgi_bytes number of bytes from the space info associated
3530 * with the block_rsv. If there is not enough space it will make an attempt to
3531 * flush out space to make room. It will do this by flushing delalloc if
3532 * possible or committing the transaction. If flush is 0 then no attempts to
3533 * regain reservations will be made and this will fail if there is not enough
3536 static int reserve_metadata_bytes(struct btrfs_root
*root
,
3537 struct btrfs_block_rsv
*block_rsv
,
3538 u64 orig_bytes
, int flush
)
3540 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
3542 u64 num_bytes
= orig_bytes
;
3545 bool committed
= false;
3546 bool flushing
= false;
3547 bool wait_ordered
= false;
3551 spin_lock(&space_info
->lock
);
3553 * We only want to wait if somebody other than us is flushing and we are
3554 * actually alloed to flush.
3556 while (flush
&& !flushing
&& space_info
->flush
) {
3557 spin_unlock(&space_info
->lock
);
3559 * If we have a trans handle we can't wait because the flusher
3560 * may have to commit the transaction, which would mean we would
3561 * deadlock since we are waiting for the flusher to finish, but
3562 * hold the current transaction open.
3564 if (current
->journal_info
)
3566 ret
= wait_event_interruptible(space_info
->wait
,
3567 !space_info
->flush
);
3568 /* Must have been interrupted, return */
3572 spin_lock(&space_info
->lock
);
3576 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
3577 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
3578 space_info
->bytes_may_use
;
3581 * The idea here is that we've not already over-reserved the block group
3582 * then we can go ahead and save our reservation first and then start
3583 * flushing if we need to. Otherwise if we've already overcommitted
3584 * lets start flushing stuff first and then come back and try to make
3587 if (used
<= space_info
->total_bytes
) {
3588 if (used
+ orig_bytes
<= space_info
->total_bytes
) {
3589 space_info
->bytes_may_use
+= orig_bytes
;
3593 * Ok set num_bytes to orig_bytes since we aren't
3594 * overocmmitted, this way we only try and reclaim what
3597 num_bytes
= orig_bytes
;
3601 * Ok we're over committed, set num_bytes to the overcommitted
3602 * amount plus the amount of bytes that we need for this
3605 wait_ordered
= true;
3606 num_bytes
= used
- space_info
->total_bytes
+
3607 (orig_bytes
* (retries
+ 1));
3611 u64 profile
= btrfs_get_alloc_profile(root
, 0);
3615 * If we have a lot of space that's pinned, don't bother doing
3616 * the overcommit dance yet and just commit the transaction.
3618 avail
= (space_info
->total_bytes
- space_info
->bytes_used
) * 8;
3620 if (space_info
->bytes_pinned
>= avail
&& flush
&& !committed
) {
3621 space_info
->flush
= 1;
3623 spin_unlock(&space_info
->lock
);
3624 ret
= may_commit_transaction(root
, space_info
,
3632 spin_lock(&root
->fs_info
->free_chunk_lock
);
3633 avail
= root
->fs_info
->free_chunk_space
;
3636 * If we have dup, raid1 or raid10 then only half of the free
3637 * space is actually useable.
3639 if (profile
& (BTRFS_BLOCK_GROUP_DUP
|
3640 BTRFS_BLOCK_GROUP_RAID1
|
3641 BTRFS_BLOCK_GROUP_RAID10
))
3645 * If we aren't flushing don't let us overcommit too much, say
3646 * 1/8th of the space. If we can flush, let it overcommit up to
3653 spin_unlock(&root
->fs_info
->free_chunk_lock
);
3655 if (used
+ num_bytes
< space_info
->total_bytes
+ avail
) {
3656 space_info
->bytes_may_use
+= orig_bytes
;
3659 wait_ordered
= true;
3664 * Couldn't make our reservation, save our place so while we're trying
3665 * to reclaim space we can actually use it instead of somebody else
3666 * stealing it from us.
3670 space_info
->flush
= 1;
3673 spin_unlock(&space_info
->lock
);
3679 * We do synchronous shrinking since we don't actually unreserve
3680 * metadata until after the IO is completed.
3682 ret
= shrink_delalloc(root
, num_bytes
, wait_ordered
);
3689 * So if we were overcommitted it's possible that somebody else flushed
3690 * out enough space and we simply didn't have enough space to reclaim,
3691 * so go back around and try again.
3694 wait_ordered
= true;
3703 ret
= may_commit_transaction(root
, space_info
, orig_bytes
, 0);
3711 spin_lock(&space_info
->lock
);
3712 space_info
->flush
= 0;
3713 wake_up_all(&space_info
->wait
);
3714 spin_unlock(&space_info
->lock
);
3719 static struct btrfs_block_rsv
*get_block_rsv(struct btrfs_trans_handle
*trans
,
3720 struct btrfs_root
*root
)
3722 struct btrfs_block_rsv
*block_rsv
= NULL
;
3724 if (root
->ref_cows
|| root
== root
->fs_info
->csum_root
)
3725 block_rsv
= trans
->block_rsv
;
3728 block_rsv
= root
->block_rsv
;
3731 block_rsv
= &root
->fs_info
->empty_block_rsv
;
3736 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
3740 spin_lock(&block_rsv
->lock
);
3741 if (block_rsv
->reserved
>= num_bytes
) {
3742 block_rsv
->reserved
-= num_bytes
;
3743 if (block_rsv
->reserved
< block_rsv
->size
)
3744 block_rsv
->full
= 0;
3747 spin_unlock(&block_rsv
->lock
);
3751 static void block_rsv_add_bytes(struct btrfs_block_rsv
*block_rsv
,
3752 u64 num_bytes
, int update_size
)
3754 spin_lock(&block_rsv
->lock
);
3755 block_rsv
->reserved
+= num_bytes
;
3757 block_rsv
->size
+= num_bytes
;
3758 else if (block_rsv
->reserved
>= block_rsv
->size
)
3759 block_rsv
->full
= 1;
3760 spin_unlock(&block_rsv
->lock
);
3763 static void block_rsv_release_bytes(struct btrfs_block_rsv
*block_rsv
,
3764 struct btrfs_block_rsv
*dest
, u64 num_bytes
)
3766 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
3768 spin_lock(&block_rsv
->lock
);
3769 if (num_bytes
== (u64
)-1)
3770 num_bytes
= block_rsv
->size
;
3771 block_rsv
->size
-= num_bytes
;
3772 if (block_rsv
->reserved
>= block_rsv
->size
) {
3773 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
3774 block_rsv
->reserved
= block_rsv
->size
;
3775 block_rsv
->full
= 1;
3779 spin_unlock(&block_rsv
->lock
);
3781 if (num_bytes
> 0) {
3783 spin_lock(&dest
->lock
);
3787 bytes_to_add
= dest
->size
- dest
->reserved
;
3788 bytes_to_add
= min(num_bytes
, bytes_to_add
);
3789 dest
->reserved
+= bytes_to_add
;
3790 if (dest
->reserved
>= dest
->size
)
3792 num_bytes
-= bytes_to_add
;
3794 spin_unlock(&dest
->lock
);
3797 spin_lock(&space_info
->lock
);
3798 space_info
->bytes_may_use
-= num_bytes
;
3799 space_info
->reservation_progress
++;
3800 spin_unlock(&space_info
->lock
);
3805 static int block_rsv_migrate_bytes(struct btrfs_block_rsv
*src
,
3806 struct btrfs_block_rsv
*dst
, u64 num_bytes
)
3810 ret
= block_rsv_use_bytes(src
, num_bytes
);
3814 block_rsv_add_bytes(dst
, num_bytes
, 1);
3818 void btrfs_init_block_rsv(struct btrfs_block_rsv
*rsv
)
3820 memset(rsv
, 0, sizeof(*rsv
));
3821 spin_lock_init(&rsv
->lock
);
3824 struct btrfs_block_rsv
*btrfs_alloc_block_rsv(struct btrfs_root
*root
)
3826 struct btrfs_block_rsv
*block_rsv
;
3827 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3829 block_rsv
= kmalloc(sizeof(*block_rsv
), GFP_NOFS
);
3833 btrfs_init_block_rsv(block_rsv
);
3834 block_rsv
->space_info
= __find_space_info(fs_info
,
3835 BTRFS_BLOCK_GROUP_METADATA
);
3839 void btrfs_free_block_rsv(struct btrfs_root
*root
,
3840 struct btrfs_block_rsv
*rsv
)
3842 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
3846 static inline int __block_rsv_add(struct btrfs_root
*root
,
3847 struct btrfs_block_rsv
*block_rsv
,
3848 u64 num_bytes
, int flush
)
3855 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
3857 block_rsv_add_bytes(block_rsv
, num_bytes
, 1);
3864 int btrfs_block_rsv_add(struct btrfs_root
*root
,
3865 struct btrfs_block_rsv
*block_rsv
,
3868 return __block_rsv_add(root
, block_rsv
, num_bytes
, 1);
3871 int btrfs_block_rsv_add_noflush(struct btrfs_root
*root
,
3872 struct btrfs_block_rsv
*block_rsv
,
3875 return __block_rsv_add(root
, block_rsv
, num_bytes
, 0);
3878 int btrfs_block_rsv_check(struct btrfs_root
*root
,
3879 struct btrfs_block_rsv
*block_rsv
, int min_factor
)
3887 spin_lock(&block_rsv
->lock
);
3888 num_bytes
= div_factor(block_rsv
->size
, min_factor
);
3889 if (block_rsv
->reserved
>= num_bytes
)
3891 spin_unlock(&block_rsv
->lock
);
3896 static inline int __btrfs_block_rsv_refill(struct btrfs_root
*root
,
3897 struct btrfs_block_rsv
*block_rsv
,
3898 u64 min_reserved
, int flush
)
3906 spin_lock(&block_rsv
->lock
);
3907 num_bytes
= min_reserved
;
3908 if (block_rsv
->reserved
>= num_bytes
)
3911 num_bytes
-= block_rsv
->reserved
;
3912 spin_unlock(&block_rsv
->lock
);
3917 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
);
3919 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
3926 int btrfs_block_rsv_refill(struct btrfs_root
*root
,
3927 struct btrfs_block_rsv
*block_rsv
,
3930 return __btrfs_block_rsv_refill(root
, block_rsv
, min_reserved
, 1);
3933 int btrfs_block_rsv_refill_noflush(struct btrfs_root
*root
,
3934 struct btrfs_block_rsv
*block_rsv
,
3937 return __btrfs_block_rsv_refill(root
, block_rsv
, min_reserved
, 0);
3940 int btrfs_block_rsv_migrate(struct btrfs_block_rsv
*src_rsv
,
3941 struct btrfs_block_rsv
*dst_rsv
,
3944 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
3947 void btrfs_block_rsv_release(struct btrfs_root
*root
,
3948 struct btrfs_block_rsv
*block_rsv
,
3951 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
3952 if (global_rsv
->full
|| global_rsv
== block_rsv
||
3953 block_rsv
->space_info
!= global_rsv
->space_info
)
3955 block_rsv_release_bytes(block_rsv
, global_rsv
, num_bytes
);
3959 * helper to calculate size of global block reservation.
3960 * the desired value is sum of space used by extent tree,
3961 * checksum tree and root tree
3963 static u64
calc_global_metadata_size(struct btrfs_fs_info
*fs_info
)
3965 struct btrfs_space_info
*sinfo
;
3969 int csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
3971 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_DATA
);
3972 spin_lock(&sinfo
->lock
);
3973 data_used
= sinfo
->bytes_used
;
3974 spin_unlock(&sinfo
->lock
);
3976 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
3977 spin_lock(&sinfo
->lock
);
3978 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_DATA
)
3980 meta_used
= sinfo
->bytes_used
;
3981 spin_unlock(&sinfo
->lock
);
3983 num_bytes
= (data_used
>> fs_info
->sb
->s_blocksize_bits
) *
3985 num_bytes
+= div64_u64(data_used
+ meta_used
, 50);
3987 if (num_bytes
* 3 > meta_used
)
3988 num_bytes
= div64_u64(meta_used
, 3);
3990 return ALIGN(num_bytes
, fs_info
->extent_root
->leafsize
<< 10);
3993 static void update_global_block_rsv(struct btrfs_fs_info
*fs_info
)
3995 struct btrfs_block_rsv
*block_rsv
= &fs_info
->global_block_rsv
;
3996 struct btrfs_space_info
*sinfo
= block_rsv
->space_info
;
3999 num_bytes
= calc_global_metadata_size(fs_info
);
4001 spin_lock(&block_rsv
->lock
);
4002 spin_lock(&sinfo
->lock
);
4004 block_rsv
->size
= num_bytes
;
4006 num_bytes
= sinfo
->bytes_used
+ sinfo
->bytes_pinned
+
4007 sinfo
->bytes_reserved
+ sinfo
->bytes_readonly
+
4008 sinfo
->bytes_may_use
;
4010 if (sinfo
->total_bytes
> num_bytes
) {
4011 num_bytes
= sinfo
->total_bytes
- num_bytes
;
4012 block_rsv
->reserved
+= num_bytes
;
4013 sinfo
->bytes_may_use
+= num_bytes
;
4016 if (block_rsv
->reserved
>= block_rsv
->size
) {
4017 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
4018 sinfo
->bytes_may_use
-= num_bytes
;
4019 sinfo
->reservation_progress
++;
4020 block_rsv
->reserved
= block_rsv
->size
;
4021 block_rsv
->full
= 1;
4024 spin_unlock(&sinfo
->lock
);
4025 spin_unlock(&block_rsv
->lock
);
4028 static void init_global_block_rsv(struct btrfs_fs_info
*fs_info
)
4030 struct btrfs_space_info
*space_info
;
4032 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
4033 fs_info
->chunk_block_rsv
.space_info
= space_info
;
4035 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
4036 fs_info
->global_block_rsv
.space_info
= space_info
;
4037 fs_info
->delalloc_block_rsv
.space_info
= space_info
;
4038 fs_info
->trans_block_rsv
.space_info
= space_info
;
4039 fs_info
->empty_block_rsv
.space_info
= space_info
;
4040 fs_info
->delayed_block_rsv
.space_info
= space_info
;
4042 fs_info
->extent_root
->block_rsv
= &fs_info
->global_block_rsv
;
4043 fs_info
->csum_root
->block_rsv
= &fs_info
->global_block_rsv
;
4044 fs_info
->dev_root
->block_rsv
= &fs_info
->global_block_rsv
;
4045 fs_info
->tree_root
->block_rsv
= &fs_info
->global_block_rsv
;
4046 fs_info
->chunk_root
->block_rsv
= &fs_info
->chunk_block_rsv
;
4048 update_global_block_rsv(fs_info
);
4051 static void release_global_block_rsv(struct btrfs_fs_info
*fs_info
)
4053 block_rsv_release_bytes(&fs_info
->global_block_rsv
, NULL
, (u64
)-1);
4054 WARN_ON(fs_info
->delalloc_block_rsv
.size
> 0);
4055 WARN_ON(fs_info
->delalloc_block_rsv
.reserved
> 0);
4056 WARN_ON(fs_info
->trans_block_rsv
.size
> 0);
4057 WARN_ON(fs_info
->trans_block_rsv
.reserved
> 0);
4058 WARN_ON(fs_info
->chunk_block_rsv
.size
> 0);
4059 WARN_ON(fs_info
->chunk_block_rsv
.reserved
> 0);
4060 WARN_ON(fs_info
->delayed_block_rsv
.size
> 0);
4061 WARN_ON(fs_info
->delayed_block_rsv
.reserved
> 0);
4064 void btrfs_trans_release_metadata(struct btrfs_trans_handle
*trans
,
4065 struct btrfs_root
*root
)
4067 if (!trans
->bytes_reserved
)
4070 btrfs_block_rsv_release(root
, trans
->block_rsv
, trans
->bytes_reserved
);
4071 trans
->bytes_reserved
= 0;
4074 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle
*trans
,
4075 struct inode
*inode
)
4077 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4078 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
4079 struct btrfs_block_rsv
*dst_rsv
= root
->orphan_block_rsv
;
4082 * We need to hold space in order to delete our orphan item once we've
4083 * added it, so this takes the reservation so we can release it later
4084 * when we are truly done with the orphan item.
4086 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
4087 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
4090 void btrfs_orphan_release_metadata(struct inode
*inode
)
4092 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4093 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
4094 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
, num_bytes
);
4097 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle
*trans
,
4098 struct btrfs_pending_snapshot
*pending
)
4100 struct btrfs_root
*root
= pending
->root
;
4101 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
4102 struct btrfs_block_rsv
*dst_rsv
= &pending
->block_rsv
;
4104 * two for root back/forward refs, two for directory entries
4105 * and one for root of the snapshot.
4107 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 5);
4108 dst_rsv
->space_info
= src_rsv
->space_info
;
4109 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
4113 * drop_outstanding_extent - drop an outstanding extent
4114 * @inode: the inode we're dropping the extent for
4116 * This is called when we are freeing up an outstanding extent, either called
4117 * after an error or after an extent is written. This will return the number of
4118 * reserved extents that need to be freed. This must be called with
4119 * BTRFS_I(inode)->lock held.
4121 static unsigned drop_outstanding_extent(struct inode
*inode
)
4123 unsigned drop_inode_space
= 0;
4124 unsigned dropped_extents
= 0;
4126 BUG_ON(!BTRFS_I(inode
)->outstanding_extents
);
4127 BTRFS_I(inode
)->outstanding_extents
--;
4129 if (BTRFS_I(inode
)->outstanding_extents
== 0 &&
4130 BTRFS_I(inode
)->delalloc_meta_reserved
) {
4131 drop_inode_space
= 1;
4132 BTRFS_I(inode
)->delalloc_meta_reserved
= 0;
4136 * If we have more or the same amount of outsanding extents than we have
4137 * reserved then we need to leave the reserved extents count alone.
4139 if (BTRFS_I(inode
)->outstanding_extents
>=
4140 BTRFS_I(inode
)->reserved_extents
)
4141 return drop_inode_space
;
4143 dropped_extents
= BTRFS_I(inode
)->reserved_extents
-
4144 BTRFS_I(inode
)->outstanding_extents
;
4145 BTRFS_I(inode
)->reserved_extents
-= dropped_extents
;
4146 return dropped_extents
+ drop_inode_space
;
4150 * calc_csum_metadata_size - return the amount of metada space that must be
4151 * reserved/free'd for the given bytes.
4152 * @inode: the inode we're manipulating
4153 * @num_bytes: the number of bytes in question
4154 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4156 * This adjusts the number of csum_bytes in the inode and then returns the
4157 * correct amount of metadata that must either be reserved or freed. We
4158 * calculate how many checksums we can fit into one leaf and then divide the
4159 * number of bytes that will need to be checksumed by this value to figure out
4160 * how many checksums will be required. If we are adding bytes then the number
4161 * may go up and we will return the number of additional bytes that must be
4162 * reserved. If it is going down we will return the number of bytes that must
4165 * This must be called with BTRFS_I(inode)->lock held.
4167 static u64
calc_csum_metadata_size(struct inode
*inode
, u64 num_bytes
,
4170 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4172 int num_csums_per_leaf
;
4176 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
&&
4177 BTRFS_I(inode
)->csum_bytes
== 0)
4180 old_csums
= (int)div64_u64(BTRFS_I(inode
)->csum_bytes
, root
->sectorsize
);
4182 BTRFS_I(inode
)->csum_bytes
+= num_bytes
;
4184 BTRFS_I(inode
)->csum_bytes
-= num_bytes
;
4185 csum_size
= BTRFS_LEAF_DATA_SIZE(root
) - sizeof(struct btrfs_item
);
4186 num_csums_per_leaf
= (int)div64_u64(csum_size
,
4187 sizeof(struct btrfs_csum_item
) +
4188 sizeof(struct btrfs_disk_key
));
4189 num_csums
= (int)div64_u64(BTRFS_I(inode
)->csum_bytes
, root
->sectorsize
);
4190 num_csums
= num_csums
+ num_csums_per_leaf
- 1;
4191 num_csums
= num_csums
/ num_csums_per_leaf
;
4193 old_csums
= old_csums
+ num_csums_per_leaf
- 1;
4194 old_csums
= old_csums
/ num_csums_per_leaf
;
4196 /* No change, no need to reserve more */
4197 if (old_csums
== num_csums
)
4201 return btrfs_calc_trans_metadata_size(root
,
4202 num_csums
- old_csums
);
4204 return btrfs_calc_trans_metadata_size(root
, old_csums
- num_csums
);
4207 int btrfs_delalloc_reserve_metadata(struct inode
*inode
, u64 num_bytes
)
4209 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4210 struct btrfs_block_rsv
*block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
4212 unsigned nr_extents
= 0;
4216 if (btrfs_is_free_space_inode(root
, inode
))
4219 if (flush
&& btrfs_transaction_in_commit(root
->fs_info
))
4220 schedule_timeout(1);
4222 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
4224 spin_lock(&BTRFS_I(inode
)->lock
);
4225 BTRFS_I(inode
)->outstanding_extents
++;
4227 if (BTRFS_I(inode
)->outstanding_extents
>
4228 BTRFS_I(inode
)->reserved_extents
) {
4229 nr_extents
= BTRFS_I(inode
)->outstanding_extents
-
4230 BTRFS_I(inode
)->reserved_extents
;
4231 BTRFS_I(inode
)->reserved_extents
+= nr_extents
;
4235 * Add an item to reserve for updating the inode when we complete the
4238 if (!BTRFS_I(inode
)->delalloc_meta_reserved
) {
4240 BTRFS_I(inode
)->delalloc_meta_reserved
= 1;
4243 to_reserve
= btrfs_calc_trans_metadata_size(root
, nr_extents
);
4244 to_reserve
+= calc_csum_metadata_size(inode
, num_bytes
, 1);
4245 spin_unlock(&BTRFS_I(inode
)->lock
);
4247 ret
= reserve_metadata_bytes(root
, block_rsv
, to_reserve
, flush
);
4252 spin_lock(&BTRFS_I(inode
)->lock
);
4253 dropped
= drop_outstanding_extent(inode
);
4254 to_free
= calc_csum_metadata_size(inode
, num_bytes
, 0);
4255 spin_unlock(&BTRFS_I(inode
)->lock
);
4256 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
4259 * Somebody could have come in and twiddled with the
4260 * reservation, so if we have to free more than we would have
4261 * reserved from this reservation go ahead and release those
4264 to_free
-= to_reserve
;
4266 btrfs_block_rsv_release(root
, block_rsv
, to_free
);
4270 block_rsv_add_bytes(block_rsv
, to_reserve
, 1);
4276 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4277 * @inode: the inode to release the reservation for
4278 * @num_bytes: the number of bytes we're releasing
4280 * This will release the metadata reservation for an inode. This can be called
4281 * once we complete IO for a given set of bytes to release their metadata
4284 void btrfs_delalloc_release_metadata(struct inode
*inode
, u64 num_bytes
)
4286 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4290 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
4291 spin_lock(&BTRFS_I(inode
)->lock
);
4292 dropped
= drop_outstanding_extent(inode
);
4294 to_free
= calc_csum_metadata_size(inode
, num_bytes
, 0);
4295 spin_unlock(&BTRFS_I(inode
)->lock
);
4297 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
4299 btrfs_block_rsv_release(root
, &root
->fs_info
->delalloc_block_rsv
,
4304 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4305 * @inode: inode we're writing to
4306 * @num_bytes: the number of bytes we want to allocate
4308 * This will do the following things
4310 * o reserve space in the data space info for num_bytes
4311 * o reserve space in the metadata space info based on number of outstanding
4312 * extents and how much csums will be needed
4313 * o add to the inodes ->delalloc_bytes
4314 * o add it to the fs_info's delalloc inodes list.
4316 * This will return 0 for success and -ENOSPC if there is no space left.
4318 int btrfs_delalloc_reserve_space(struct inode
*inode
, u64 num_bytes
)
4322 ret
= btrfs_check_data_free_space(inode
, num_bytes
);
4326 ret
= btrfs_delalloc_reserve_metadata(inode
, num_bytes
);
4328 btrfs_free_reserved_data_space(inode
, num_bytes
);
4336 * btrfs_delalloc_release_space - release data and metadata space for delalloc
4337 * @inode: inode we're releasing space for
4338 * @num_bytes: the number of bytes we want to free up
4340 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
4341 * called in the case that we don't need the metadata AND data reservations
4342 * anymore. So if there is an error or we insert an inline extent.
4344 * This function will release the metadata space that was not used and will
4345 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4346 * list if there are no delalloc bytes left.
4348 void btrfs_delalloc_release_space(struct inode
*inode
, u64 num_bytes
)
4350 btrfs_delalloc_release_metadata(inode
, num_bytes
);
4351 btrfs_free_reserved_data_space(inode
, num_bytes
);
4354 static int update_block_group(struct btrfs_trans_handle
*trans
,
4355 struct btrfs_root
*root
,
4356 u64 bytenr
, u64 num_bytes
, int alloc
)
4358 struct btrfs_block_group_cache
*cache
= NULL
;
4359 struct btrfs_fs_info
*info
= root
->fs_info
;
4360 u64 total
= num_bytes
;
4365 /* block accounting for super block */
4366 spin_lock(&info
->delalloc_lock
);
4367 old_val
= btrfs_super_bytes_used(info
->super_copy
);
4369 old_val
+= num_bytes
;
4371 old_val
-= num_bytes
;
4372 btrfs_set_super_bytes_used(info
->super_copy
, old_val
);
4373 spin_unlock(&info
->delalloc_lock
);
4376 cache
= btrfs_lookup_block_group(info
, bytenr
);
4379 if (cache
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
4380 BTRFS_BLOCK_GROUP_RAID1
|
4381 BTRFS_BLOCK_GROUP_RAID10
))
4386 * If this block group has free space cache written out, we
4387 * need to make sure to load it if we are removing space. This
4388 * is because we need the unpinning stage to actually add the
4389 * space back to the block group, otherwise we will leak space.
4391 if (!alloc
&& cache
->cached
== BTRFS_CACHE_NO
)
4392 cache_block_group(cache
, trans
, NULL
, 1);
4394 byte_in_group
= bytenr
- cache
->key
.objectid
;
4395 WARN_ON(byte_in_group
> cache
->key
.offset
);
4397 spin_lock(&cache
->space_info
->lock
);
4398 spin_lock(&cache
->lock
);
4400 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
4401 cache
->disk_cache_state
< BTRFS_DC_CLEAR
)
4402 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
4405 old_val
= btrfs_block_group_used(&cache
->item
);
4406 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
4408 old_val
+= num_bytes
;
4409 btrfs_set_block_group_used(&cache
->item
, old_val
);
4410 cache
->reserved
-= num_bytes
;
4411 cache
->space_info
->bytes_reserved
-= num_bytes
;
4412 cache
->space_info
->bytes_used
+= num_bytes
;
4413 cache
->space_info
->disk_used
+= num_bytes
* factor
;
4414 spin_unlock(&cache
->lock
);
4415 spin_unlock(&cache
->space_info
->lock
);
4417 old_val
-= num_bytes
;
4418 btrfs_set_block_group_used(&cache
->item
, old_val
);
4419 cache
->pinned
+= num_bytes
;
4420 cache
->space_info
->bytes_pinned
+= num_bytes
;
4421 cache
->space_info
->bytes_used
-= num_bytes
;
4422 cache
->space_info
->disk_used
-= num_bytes
* factor
;
4423 spin_unlock(&cache
->lock
);
4424 spin_unlock(&cache
->space_info
->lock
);
4426 set_extent_dirty(info
->pinned_extents
,
4427 bytenr
, bytenr
+ num_bytes
- 1,
4428 GFP_NOFS
| __GFP_NOFAIL
);
4430 btrfs_put_block_group(cache
);
4432 bytenr
+= num_bytes
;
4437 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
4439 struct btrfs_block_group_cache
*cache
;
4442 cache
= btrfs_lookup_first_block_group(root
->fs_info
, search_start
);
4446 bytenr
= cache
->key
.objectid
;
4447 btrfs_put_block_group(cache
);
4452 static int pin_down_extent(struct btrfs_root
*root
,
4453 struct btrfs_block_group_cache
*cache
,
4454 u64 bytenr
, u64 num_bytes
, int reserved
)
4456 spin_lock(&cache
->space_info
->lock
);
4457 spin_lock(&cache
->lock
);
4458 cache
->pinned
+= num_bytes
;
4459 cache
->space_info
->bytes_pinned
+= num_bytes
;
4461 cache
->reserved
-= num_bytes
;
4462 cache
->space_info
->bytes_reserved
-= num_bytes
;
4464 spin_unlock(&cache
->lock
);
4465 spin_unlock(&cache
->space_info
->lock
);
4467 set_extent_dirty(root
->fs_info
->pinned_extents
, bytenr
,
4468 bytenr
+ num_bytes
- 1, GFP_NOFS
| __GFP_NOFAIL
);
4473 * this function must be called within transaction
4475 int btrfs_pin_extent(struct btrfs_root
*root
,
4476 u64 bytenr
, u64 num_bytes
, int reserved
)
4478 struct btrfs_block_group_cache
*cache
;
4480 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
4483 pin_down_extent(root
, cache
, bytenr
, num_bytes
, reserved
);
4485 btrfs_put_block_group(cache
);
4490 * this function must be called within transaction
4492 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle
*trans
,
4493 struct btrfs_root
*root
,
4494 u64 bytenr
, u64 num_bytes
)
4496 struct btrfs_block_group_cache
*cache
;
4498 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
4502 * pull in the free space cache (if any) so that our pin
4503 * removes the free space from the cache. We have load_only set
4504 * to one because the slow code to read in the free extents does check
4505 * the pinned extents.
4507 cache_block_group(cache
, trans
, root
, 1);
4509 pin_down_extent(root
, cache
, bytenr
, num_bytes
, 0);
4511 /* remove us from the free space cache (if we're there at all) */
4512 btrfs_remove_free_space(cache
, bytenr
, num_bytes
);
4513 btrfs_put_block_group(cache
);
4518 * btrfs_update_reserved_bytes - update the block_group and space info counters
4519 * @cache: The cache we are manipulating
4520 * @num_bytes: The number of bytes in question
4521 * @reserve: One of the reservation enums
4523 * This is called by the allocator when it reserves space, or by somebody who is
4524 * freeing space that was never actually used on disk. For example if you
4525 * reserve some space for a new leaf in transaction A and before transaction A
4526 * commits you free that leaf, you call this with reserve set to 0 in order to
4527 * clear the reservation.
4529 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4530 * ENOSPC accounting. For data we handle the reservation through clearing the
4531 * delalloc bits in the io_tree. We have to do this since we could end up
4532 * allocating less disk space for the amount of data we have reserved in the
4533 * case of compression.
4535 * If this is a reservation and the block group has become read only we cannot
4536 * make the reservation and return -EAGAIN, otherwise this function always
4539 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
4540 u64 num_bytes
, int reserve
)
4542 struct btrfs_space_info
*space_info
= cache
->space_info
;
4544 spin_lock(&space_info
->lock
);
4545 spin_lock(&cache
->lock
);
4546 if (reserve
!= RESERVE_FREE
) {
4550 cache
->reserved
+= num_bytes
;
4551 space_info
->bytes_reserved
+= num_bytes
;
4552 if (reserve
== RESERVE_ALLOC
) {
4553 BUG_ON(space_info
->bytes_may_use
< num_bytes
);
4554 space_info
->bytes_may_use
-= num_bytes
;
4559 space_info
->bytes_readonly
+= num_bytes
;
4560 cache
->reserved
-= num_bytes
;
4561 space_info
->bytes_reserved
-= num_bytes
;
4562 space_info
->reservation_progress
++;
4564 spin_unlock(&cache
->lock
);
4565 spin_unlock(&space_info
->lock
);
4569 int btrfs_prepare_extent_commit(struct btrfs_trans_handle
*trans
,
4570 struct btrfs_root
*root
)
4572 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4573 struct btrfs_caching_control
*next
;
4574 struct btrfs_caching_control
*caching_ctl
;
4575 struct btrfs_block_group_cache
*cache
;
4577 down_write(&fs_info
->extent_commit_sem
);
4579 list_for_each_entry_safe(caching_ctl
, next
,
4580 &fs_info
->caching_block_groups
, list
) {
4581 cache
= caching_ctl
->block_group
;
4582 if (block_group_cache_done(cache
)) {
4583 cache
->last_byte_to_unpin
= (u64
)-1;
4584 list_del_init(&caching_ctl
->list
);
4585 put_caching_control(caching_ctl
);
4587 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
4591 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
4592 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
4594 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
4596 up_write(&fs_info
->extent_commit_sem
);
4598 update_global_block_rsv(fs_info
);
4602 static int unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
4604 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4605 struct btrfs_block_group_cache
*cache
= NULL
;
4608 while (start
<= end
) {
4610 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
4612 btrfs_put_block_group(cache
);
4613 cache
= btrfs_lookup_block_group(fs_info
, start
);
4617 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
4618 len
= min(len
, end
+ 1 - start
);
4620 if (start
< cache
->last_byte_to_unpin
) {
4621 len
= min(len
, cache
->last_byte_to_unpin
- start
);
4622 btrfs_add_free_space(cache
, start
, len
);
4627 spin_lock(&cache
->space_info
->lock
);
4628 spin_lock(&cache
->lock
);
4629 cache
->pinned
-= len
;
4630 cache
->space_info
->bytes_pinned
-= len
;
4632 cache
->space_info
->bytes_readonly
+= len
;
4633 spin_unlock(&cache
->lock
);
4634 spin_unlock(&cache
->space_info
->lock
);
4638 btrfs_put_block_group(cache
);
4642 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
4643 struct btrfs_root
*root
)
4645 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4646 struct extent_io_tree
*unpin
;
4651 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
4652 unpin
= &fs_info
->freed_extents
[1];
4654 unpin
= &fs_info
->freed_extents
[0];
4657 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
4662 if (btrfs_test_opt(root
, DISCARD
))
4663 ret
= btrfs_discard_extent(root
, start
,
4664 end
+ 1 - start
, NULL
);
4666 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
4667 unpin_extent_range(root
, start
, end
);
4674 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
4675 struct btrfs_root
*root
,
4676 u64 bytenr
, u64 num_bytes
, u64 parent
,
4677 u64 root_objectid
, u64 owner_objectid
,
4678 u64 owner_offset
, int refs_to_drop
,
4679 struct btrfs_delayed_extent_op
*extent_op
)
4681 struct btrfs_key key
;
4682 struct btrfs_path
*path
;
4683 struct btrfs_fs_info
*info
= root
->fs_info
;
4684 struct btrfs_root
*extent_root
= info
->extent_root
;
4685 struct extent_buffer
*leaf
;
4686 struct btrfs_extent_item
*ei
;
4687 struct btrfs_extent_inline_ref
*iref
;
4690 int extent_slot
= 0;
4691 int found_extent
= 0;
4696 path
= btrfs_alloc_path();
4701 path
->leave_spinning
= 1;
4703 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
4704 BUG_ON(!is_data
&& refs_to_drop
!= 1);
4706 ret
= lookup_extent_backref(trans
, extent_root
, path
, &iref
,
4707 bytenr
, num_bytes
, parent
,
4708 root_objectid
, owner_objectid
,
4711 extent_slot
= path
->slots
[0];
4712 while (extent_slot
>= 0) {
4713 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
4715 if (key
.objectid
!= bytenr
)
4717 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
4718 key
.offset
== num_bytes
) {
4722 if (path
->slots
[0] - extent_slot
> 5)
4726 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4727 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
4728 if (found_extent
&& item_size
< sizeof(*ei
))
4731 if (!found_extent
) {
4733 ret
= remove_extent_backref(trans
, extent_root
, path
,
4737 btrfs_release_path(path
);
4738 path
->leave_spinning
= 1;
4740 key
.objectid
= bytenr
;
4741 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
4742 key
.offset
= num_bytes
;
4744 ret
= btrfs_search_slot(trans
, extent_root
,
4747 printk(KERN_ERR
"umm, got %d back from search"
4748 ", was looking for %llu\n", ret
,
4749 (unsigned long long)bytenr
);
4751 btrfs_print_leaf(extent_root
,
4755 extent_slot
= path
->slots
[0];
4758 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
4760 printk(KERN_ERR
"btrfs unable to find ref byte nr %llu "
4761 "parent %llu root %llu owner %llu offset %llu\n",
4762 (unsigned long long)bytenr
,
4763 (unsigned long long)parent
,
4764 (unsigned long long)root_objectid
,
4765 (unsigned long long)owner_objectid
,
4766 (unsigned long long)owner_offset
);
4769 leaf
= path
->nodes
[0];
4770 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
4771 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4772 if (item_size
< sizeof(*ei
)) {
4773 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
4774 ret
= convert_extent_item_v0(trans
, extent_root
, path
,
4778 btrfs_release_path(path
);
4779 path
->leave_spinning
= 1;
4781 key
.objectid
= bytenr
;
4782 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
4783 key
.offset
= num_bytes
;
4785 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
4788 printk(KERN_ERR
"umm, got %d back from search"
4789 ", was looking for %llu\n", ret
,
4790 (unsigned long long)bytenr
);
4791 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
4794 extent_slot
= path
->slots
[0];
4795 leaf
= path
->nodes
[0];
4796 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
4799 BUG_ON(item_size
< sizeof(*ei
));
4800 ei
= btrfs_item_ptr(leaf
, extent_slot
,
4801 struct btrfs_extent_item
);
4802 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
) {
4803 struct btrfs_tree_block_info
*bi
;
4804 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
4805 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
4806 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
4809 refs
= btrfs_extent_refs(leaf
, ei
);
4810 BUG_ON(refs
< refs_to_drop
);
4811 refs
-= refs_to_drop
;
4815 __run_delayed_extent_op(extent_op
, leaf
, ei
);
4817 * In the case of inline back ref, reference count will
4818 * be updated by remove_extent_backref
4821 BUG_ON(!found_extent
);
4823 btrfs_set_extent_refs(leaf
, ei
, refs
);
4824 btrfs_mark_buffer_dirty(leaf
);
4827 ret
= remove_extent_backref(trans
, extent_root
, path
,
4834 BUG_ON(is_data
&& refs_to_drop
!=
4835 extent_data_ref_count(root
, path
, iref
));
4837 BUG_ON(path
->slots
[0] != extent_slot
);
4839 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
4840 path
->slots
[0] = extent_slot
;
4845 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
4848 btrfs_release_path(path
);
4851 ret
= btrfs_del_csums(trans
, root
, bytenr
, num_bytes
);
4854 invalidate_mapping_pages(info
->btree_inode
->i_mapping
,
4855 bytenr
>> PAGE_CACHE_SHIFT
,
4856 (bytenr
+ num_bytes
- 1) >> PAGE_CACHE_SHIFT
);
4859 ret
= update_block_group(trans
, root
, bytenr
, num_bytes
, 0);
4862 btrfs_free_path(path
);
4867 * when we free an block, it is possible (and likely) that we free the last
4868 * delayed ref for that extent as well. This searches the delayed ref tree for
4869 * a given extent, and if there are no other delayed refs to be processed, it
4870 * removes it from the tree.
4872 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
4873 struct btrfs_root
*root
, u64 bytenr
)
4875 struct btrfs_delayed_ref_head
*head
;
4876 struct btrfs_delayed_ref_root
*delayed_refs
;
4877 struct btrfs_delayed_ref_node
*ref
;
4878 struct rb_node
*node
;
4881 delayed_refs
= &trans
->transaction
->delayed_refs
;
4882 spin_lock(&delayed_refs
->lock
);
4883 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
4887 node
= rb_prev(&head
->node
.rb_node
);
4891 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
4893 /* there are still entries for this ref, we can't drop it */
4894 if (ref
->bytenr
== bytenr
)
4897 if (head
->extent_op
) {
4898 if (!head
->must_insert_reserved
)
4900 kfree(head
->extent_op
);
4901 head
->extent_op
= NULL
;
4905 * waiting for the lock here would deadlock. If someone else has it
4906 * locked they are already in the process of dropping it anyway
4908 if (!mutex_trylock(&head
->mutex
))
4912 * at this point we have a head with no other entries. Go
4913 * ahead and process it.
4915 head
->node
.in_tree
= 0;
4916 rb_erase(&head
->node
.rb_node
, &delayed_refs
->root
);
4918 delayed_refs
->num_entries
--;
4921 * we don't take a ref on the node because we're removing it from the
4922 * tree, so we just steal the ref the tree was holding.
4924 delayed_refs
->num_heads
--;
4925 if (list_empty(&head
->cluster
))
4926 delayed_refs
->num_heads_ready
--;
4928 list_del_init(&head
->cluster
);
4929 spin_unlock(&delayed_refs
->lock
);
4931 BUG_ON(head
->extent_op
);
4932 if (head
->must_insert_reserved
)
4935 mutex_unlock(&head
->mutex
);
4936 btrfs_put_delayed_ref(&head
->node
);
4939 spin_unlock(&delayed_refs
->lock
);
4943 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
4944 struct btrfs_root
*root
,
4945 struct extent_buffer
*buf
,
4946 u64 parent
, int last_ref
, int for_cow
)
4948 struct btrfs_block_group_cache
*cache
= NULL
;
4951 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
4952 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
4953 buf
->start
, buf
->len
,
4954 parent
, root
->root_key
.objectid
,
4955 btrfs_header_level(buf
),
4956 BTRFS_DROP_DELAYED_REF
, NULL
, for_cow
);
4963 cache
= btrfs_lookup_block_group(root
->fs_info
, buf
->start
);
4965 if (btrfs_header_generation(buf
) == trans
->transid
) {
4966 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
4967 ret
= check_ref_cleanup(trans
, root
, buf
->start
);
4972 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
4973 pin_down_extent(root
, cache
, buf
->start
, buf
->len
, 1);
4977 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
4979 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
4980 btrfs_update_reserved_bytes(cache
, buf
->len
, RESERVE_FREE
);
4984 * Deleting the buffer, clear the corrupt flag since it doesn't matter
4987 clear_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
);
4988 btrfs_put_block_group(cache
);
4991 int btrfs_free_extent(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
4992 u64 bytenr
, u64 num_bytes
, u64 parent
, u64 root_objectid
,
4993 u64 owner
, u64 offset
, int for_cow
)
4996 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4999 * tree log blocks never actually go into the extent allocation
5000 * tree, just update pinning info and exit early.
5002 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
5003 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
5004 /* unlocks the pinned mutex */
5005 btrfs_pin_extent(root
, bytenr
, num_bytes
, 1);
5007 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
5008 ret
= btrfs_add_delayed_tree_ref(fs_info
, trans
, bytenr
,
5010 parent
, root_objectid
, (int)owner
,
5011 BTRFS_DROP_DELAYED_REF
, NULL
, for_cow
);
5014 ret
= btrfs_add_delayed_data_ref(fs_info
, trans
, bytenr
,
5016 parent
, root_objectid
, owner
,
5017 offset
, BTRFS_DROP_DELAYED_REF
,
5024 static u64
stripe_align(struct btrfs_root
*root
, u64 val
)
5026 u64 mask
= ((u64
)root
->stripesize
- 1);
5027 u64 ret
= (val
+ mask
) & ~mask
;
5032 * when we wait for progress in the block group caching, its because
5033 * our allocation attempt failed at least once. So, we must sleep
5034 * and let some progress happen before we try again.
5036 * This function will sleep at least once waiting for new free space to
5037 * show up, and then it will check the block group free space numbers
5038 * for our min num_bytes. Another option is to have it go ahead
5039 * and look in the rbtree for a free extent of a given size, but this
5043 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
5046 struct btrfs_caching_control
*caching_ctl
;
5049 caching_ctl
= get_caching_control(cache
);
5053 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
5054 (cache
->free_space_ctl
->free_space
>= num_bytes
));
5056 put_caching_control(caching_ctl
);
5061 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
5063 struct btrfs_caching_control
*caching_ctl
;
5066 caching_ctl
= get_caching_control(cache
);
5070 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
5072 put_caching_control(caching_ctl
);
5076 static int get_block_group_index(struct btrfs_block_group_cache
*cache
)
5079 if (cache
->flags
& BTRFS_BLOCK_GROUP_RAID10
)
5081 else if (cache
->flags
& BTRFS_BLOCK_GROUP_RAID1
)
5083 else if (cache
->flags
& BTRFS_BLOCK_GROUP_DUP
)
5085 else if (cache
->flags
& BTRFS_BLOCK_GROUP_RAID0
)
5092 enum btrfs_loop_type
{
5093 LOOP_FIND_IDEAL
= 0,
5094 LOOP_CACHING_NOWAIT
= 1,
5095 LOOP_CACHING_WAIT
= 2,
5096 LOOP_ALLOC_CHUNK
= 3,
5097 LOOP_NO_EMPTY_SIZE
= 4,
5101 * walks the btree of allocated extents and find a hole of a given size.
5102 * The key ins is changed to record the hole:
5103 * ins->objectid == block start
5104 * ins->flags = BTRFS_EXTENT_ITEM_KEY
5105 * ins->offset == number of blocks
5106 * Any available blocks before search_start are skipped.
5108 static noinline
int find_free_extent(struct btrfs_trans_handle
*trans
,
5109 struct btrfs_root
*orig_root
,
5110 u64 num_bytes
, u64 empty_size
,
5111 u64 search_start
, u64 search_end
,
5112 u64 hint_byte
, struct btrfs_key
*ins
,
5116 struct btrfs_root
*root
= orig_root
->fs_info
->extent_root
;
5117 struct btrfs_free_cluster
*last_ptr
= NULL
;
5118 struct btrfs_block_group_cache
*block_group
= NULL
;
5119 int empty_cluster
= 2 * 1024 * 1024;
5120 int allowed_chunk_alloc
= 0;
5121 int done_chunk_alloc
= 0;
5122 struct btrfs_space_info
*space_info
;
5123 int last_ptr_loop
= 0;
5126 int alloc_type
= (data
& BTRFS_BLOCK_GROUP_DATA
) ?
5127 RESERVE_ALLOC_NO_ACCOUNT
: RESERVE_ALLOC
;
5128 bool found_uncached_bg
= false;
5129 bool failed_cluster_refill
= false;
5130 bool failed_alloc
= false;
5131 bool use_cluster
= true;
5132 bool have_caching_bg
= false;
5133 u64 ideal_cache_percent
= 0;
5134 u64 ideal_cache_offset
= 0;
5136 WARN_ON(num_bytes
< root
->sectorsize
);
5137 btrfs_set_key_type(ins
, BTRFS_EXTENT_ITEM_KEY
);
5141 space_info
= __find_space_info(root
->fs_info
, data
);
5143 printk(KERN_ERR
"No space info for %llu\n", data
);
5148 * If the space info is for both data and metadata it means we have a
5149 * small filesystem and we can't use the clustering stuff.
5151 if (btrfs_mixed_space_info(space_info
))
5152 use_cluster
= false;
5154 if (orig_root
->ref_cows
|| empty_size
)
5155 allowed_chunk_alloc
= 1;
5157 if (data
& BTRFS_BLOCK_GROUP_METADATA
&& use_cluster
) {
5158 last_ptr
= &root
->fs_info
->meta_alloc_cluster
;
5159 if (!btrfs_test_opt(root
, SSD
))
5160 empty_cluster
= 64 * 1024;
5163 if ((data
& BTRFS_BLOCK_GROUP_DATA
) && use_cluster
&&
5164 btrfs_test_opt(root
, SSD
)) {
5165 last_ptr
= &root
->fs_info
->data_alloc_cluster
;
5169 spin_lock(&last_ptr
->lock
);
5170 if (last_ptr
->block_group
)
5171 hint_byte
= last_ptr
->window_start
;
5172 spin_unlock(&last_ptr
->lock
);
5175 search_start
= max(search_start
, first_logical_byte(root
, 0));
5176 search_start
= max(search_start
, hint_byte
);
5181 if (search_start
== hint_byte
) {
5183 block_group
= btrfs_lookup_block_group(root
->fs_info
,
5186 * we don't want to use the block group if it doesn't match our
5187 * allocation bits, or if its not cached.
5189 * However if we are re-searching with an ideal block group
5190 * picked out then we don't care that the block group is cached.
5192 if (block_group
&& block_group_bits(block_group
, data
) &&
5193 (block_group
->cached
!= BTRFS_CACHE_NO
||
5194 search_start
== ideal_cache_offset
)) {
5195 down_read(&space_info
->groups_sem
);
5196 if (list_empty(&block_group
->list
) ||
5199 * someone is removing this block group,
5200 * we can't jump into the have_block_group
5201 * target because our list pointers are not
5204 btrfs_put_block_group(block_group
);
5205 up_read(&space_info
->groups_sem
);
5207 index
= get_block_group_index(block_group
);
5208 goto have_block_group
;
5210 } else if (block_group
) {
5211 btrfs_put_block_group(block_group
);
5215 have_caching_bg
= false;
5216 down_read(&space_info
->groups_sem
);
5217 list_for_each_entry(block_group
, &space_info
->block_groups
[index
],
5222 btrfs_get_block_group(block_group
);
5223 search_start
= block_group
->key
.objectid
;
5226 * this can happen if we end up cycling through all the
5227 * raid types, but we want to make sure we only allocate
5228 * for the proper type.
5230 if (!block_group_bits(block_group
, data
)) {
5231 u64 extra
= BTRFS_BLOCK_GROUP_DUP
|
5232 BTRFS_BLOCK_GROUP_RAID1
|
5233 BTRFS_BLOCK_GROUP_RAID10
;
5236 * if they asked for extra copies and this block group
5237 * doesn't provide them, bail. This does allow us to
5238 * fill raid0 from raid1.
5240 if ((data
& extra
) && !(block_group
->flags
& extra
))
5245 cached
= block_group_cache_done(block_group
);
5246 if (unlikely(!cached
)) {
5249 found_uncached_bg
= true;
5250 ret
= cache_block_group(block_group
, trans
,
5252 if (block_group
->cached
== BTRFS_CACHE_FINISHED
)
5255 free_percent
= btrfs_block_group_used(&block_group
->item
);
5256 free_percent
*= 100;
5257 free_percent
= div64_u64(free_percent
,
5258 block_group
->key
.offset
);
5259 free_percent
= 100 - free_percent
;
5260 if (free_percent
> ideal_cache_percent
&&
5261 likely(!block_group
->ro
)) {
5262 ideal_cache_offset
= block_group
->key
.objectid
;
5263 ideal_cache_percent
= free_percent
;
5267 * The caching workers are limited to 2 threads, so we
5268 * can queue as much work as we care to.
5270 if (loop
> LOOP_FIND_IDEAL
) {
5271 ret
= cache_block_group(block_group
, trans
,
5277 * If loop is set for cached only, try the next block
5280 if (loop
== LOOP_FIND_IDEAL
)
5285 if (unlikely(block_group
->ro
))
5288 spin_lock(&block_group
->free_space_ctl
->tree_lock
);
5290 block_group
->free_space_ctl
->free_space
<
5291 num_bytes
+ empty_cluster
+ empty_size
) {
5292 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
5295 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
5298 * Ok we want to try and use the cluster allocator, so lets look
5299 * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
5300 * have tried the cluster allocator plenty of times at this
5301 * point and not have found anything, so we are likely way too
5302 * fragmented for the clustering stuff to find anything, so lets
5303 * just skip it and let the allocator find whatever block it can
5306 if (last_ptr
&& loop
< LOOP_NO_EMPTY_SIZE
) {
5308 * the refill lock keeps out other
5309 * people trying to start a new cluster
5311 spin_lock(&last_ptr
->refill_lock
);
5312 if (!last_ptr
->block_group
||
5313 last_ptr
->block_group
->ro
||
5314 !block_group_bits(last_ptr
->block_group
, data
))
5315 goto refill_cluster
;
5317 offset
= btrfs_alloc_from_cluster(block_group
, last_ptr
,
5318 num_bytes
, search_start
);
5320 /* we have a block, we're done */
5321 spin_unlock(&last_ptr
->refill_lock
);
5325 spin_lock(&last_ptr
->lock
);
5327 * whoops, this cluster doesn't actually point to
5328 * this block group. Get a ref on the block
5329 * group is does point to and try again
5331 if (!last_ptr_loop
&& last_ptr
->block_group
&&
5332 last_ptr
->block_group
!= block_group
&&
5334 get_block_group_index(last_ptr
->block_group
)) {
5336 btrfs_put_block_group(block_group
);
5337 block_group
= last_ptr
->block_group
;
5338 btrfs_get_block_group(block_group
);
5339 spin_unlock(&last_ptr
->lock
);
5340 spin_unlock(&last_ptr
->refill_lock
);
5343 search_start
= block_group
->key
.objectid
;
5345 * we know this block group is properly
5346 * in the list because
5347 * btrfs_remove_block_group, drops the
5348 * cluster before it removes the block
5349 * group from the list
5351 goto have_block_group
;
5353 spin_unlock(&last_ptr
->lock
);
5356 * this cluster didn't work out, free it and
5359 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
5363 /* allocate a cluster in this block group */
5364 ret
= btrfs_find_space_cluster(trans
, root
,
5365 block_group
, last_ptr
,
5366 search_start
, num_bytes
,
5367 empty_cluster
+ empty_size
);
5370 * now pull our allocation out of this
5373 offset
= btrfs_alloc_from_cluster(block_group
,
5374 last_ptr
, num_bytes
,
5377 /* we found one, proceed */
5378 spin_unlock(&last_ptr
->refill_lock
);
5381 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
5382 && !failed_cluster_refill
) {
5383 spin_unlock(&last_ptr
->refill_lock
);
5385 failed_cluster_refill
= true;
5386 wait_block_group_cache_progress(block_group
,
5387 num_bytes
+ empty_cluster
+ empty_size
);
5388 goto have_block_group
;
5392 * at this point we either didn't find a cluster
5393 * or we weren't able to allocate a block from our
5394 * cluster. Free the cluster we've been trying
5395 * to use, and go to the next block group
5397 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
5398 spin_unlock(&last_ptr
->refill_lock
);
5402 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
5403 num_bytes
, empty_size
);
5405 * If we didn't find a chunk, and we haven't failed on this
5406 * block group before, and this block group is in the middle of
5407 * caching and we are ok with waiting, then go ahead and wait
5408 * for progress to be made, and set failed_alloc to true.
5410 * If failed_alloc is true then we've already waited on this
5411 * block group once and should move on to the next block group.
5413 if (!offset
&& !failed_alloc
&& !cached
&&
5414 loop
> LOOP_CACHING_NOWAIT
) {
5415 wait_block_group_cache_progress(block_group
,
5416 num_bytes
+ empty_size
);
5417 failed_alloc
= true;
5418 goto have_block_group
;
5419 } else if (!offset
) {
5421 have_caching_bg
= true;
5425 search_start
= stripe_align(root
, offset
);
5426 /* move on to the next group */
5427 if (search_start
+ num_bytes
>= search_end
) {
5428 btrfs_add_free_space(block_group
, offset
, num_bytes
);
5432 /* move on to the next group */
5433 if (search_start
+ num_bytes
>
5434 block_group
->key
.objectid
+ block_group
->key
.offset
) {
5435 btrfs_add_free_space(block_group
, offset
, num_bytes
);
5439 ins
->objectid
= search_start
;
5440 ins
->offset
= num_bytes
;
5442 if (offset
< search_start
)
5443 btrfs_add_free_space(block_group
, offset
,
5444 search_start
- offset
);
5445 BUG_ON(offset
> search_start
);
5447 ret
= btrfs_update_reserved_bytes(block_group
, num_bytes
,
5449 if (ret
== -EAGAIN
) {
5450 btrfs_add_free_space(block_group
, offset
, num_bytes
);
5454 /* we are all good, lets return */
5455 ins
->objectid
= search_start
;
5456 ins
->offset
= num_bytes
;
5458 if (offset
< search_start
)
5459 btrfs_add_free_space(block_group
, offset
,
5460 search_start
- offset
);
5461 BUG_ON(offset
> search_start
);
5462 btrfs_put_block_group(block_group
);
5465 failed_cluster_refill
= false;
5466 failed_alloc
= false;
5467 BUG_ON(index
!= get_block_group_index(block_group
));
5468 btrfs_put_block_group(block_group
);
5470 up_read(&space_info
->groups_sem
);
5472 if (!ins
->objectid
&& loop
>= LOOP_CACHING_WAIT
&& have_caching_bg
)
5475 if (!ins
->objectid
&& ++index
< BTRFS_NR_RAID_TYPES
)
5478 /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5479 * for them to make caching progress. Also
5480 * determine the best possible bg to cache
5481 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5482 * caching kthreads as we move along
5483 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5484 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5485 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5488 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
) {
5490 if (loop
== LOOP_FIND_IDEAL
&& found_uncached_bg
) {
5491 found_uncached_bg
= false;
5493 if (!ideal_cache_percent
)
5497 * 1 of the following 2 things have happened so far
5499 * 1) We found an ideal block group for caching that
5500 * is mostly full and will cache quickly, so we might
5501 * as well wait for it.
5503 * 2) We searched for cached only and we didn't find
5504 * anything, and we didn't start any caching kthreads
5505 * either, so chances are we will loop through and
5506 * start a couple caching kthreads, and then come back
5507 * around and just wait for them. This will be slower
5508 * because we will have 2 caching kthreads reading at
5509 * the same time when we could have just started one
5510 * and waited for it to get far enough to give us an
5511 * allocation, so go ahead and go to the wait caching
5514 loop
= LOOP_CACHING_WAIT
;
5515 search_start
= ideal_cache_offset
;
5516 ideal_cache_percent
= 0;
5518 } else if (loop
== LOOP_FIND_IDEAL
) {
5520 * Didn't find a uncached bg, wait on anything we find
5523 loop
= LOOP_CACHING_WAIT
;
5529 if (loop
== LOOP_ALLOC_CHUNK
) {
5530 if (allowed_chunk_alloc
) {
5531 ret
= do_chunk_alloc(trans
, root
, num_bytes
+
5532 2 * 1024 * 1024, data
,
5533 CHUNK_ALLOC_LIMITED
);
5534 allowed_chunk_alloc
= 0;
5536 done_chunk_alloc
= 1;
5537 } else if (!done_chunk_alloc
&&
5538 space_info
->force_alloc
==
5539 CHUNK_ALLOC_NO_FORCE
) {
5540 space_info
->force_alloc
= CHUNK_ALLOC_LIMITED
;
5544 * We didn't allocate a chunk, go ahead and drop the
5545 * empty size and loop again.
5547 if (!done_chunk_alloc
)
5548 loop
= LOOP_NO_EMPTY_SIZE
;
5551 if (loop
== LOOP_NO_EMPTY_SIZE
) {
5557 } else if (!ins
->objectid
) {
5559 } else if (ins
->objectid
) {
5566 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
5567 int dump_block_groups
)
5569 struct btrfs_block_group_cache
*cache
;
5572 spin_lock(&info
->lock
);
5573 printk(KERN_INFO
"space_info %llu has %llu free, is %sfull\n",
5574 (unsigned long long)info
->flags
,
5575 (unsigned long long)(info
->total_bytes
- info
->bytes_used
-
5576 info
->bytes_pinned
- info
->bytes_reserved
-
5577 info
->bytes_readonly
),
5578 (info
->full
) ? "" : "not ");
5579 printk(KERN_INFO
"space_info total=%llu, used=%llu, pinned=%llu, "
5580 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5581 (unsigned long long)info
->total_bytes
,
5582 (unsigned long long)info
->bytes_used
,
5583 (unsigned long long)info
->bytes_pinned
,
5584 (unsigned long long)info
->bytes_reserved
,
5585 (unsigned long long)info
->bytes_may_use
,
5586 (unsigned long long)info
->bytes_readonly
);
5587 spin_unlock(&info
->lock
);
5589 if (!dump_block_groups
)
5592 down_read(&info
->groups_sem
);
5594 list_for_each_entry(cache
, &info
->block_groups
[index
], list
) {
5595 spin_lock(&cache
->lock
);
5596 printk(KERN_INFO
"block group %llu has %llu bytes, %llu used "
5597 "%llu pinned %llu reserved\n",
5598 (unsigned long long)cache
->key
.objectid
,
5599 (unsigned long long)cache
->key
.offset
,
5600 (unsigned long long)btrfs_block_group_used(&cache
->item
),
5601 (unsigned long long)cache
->pinned
,
5602 (unsigned long long)cache
->reserved
);
5603 btrfs_dump_free_space(cache
, bytes
);
5604 spin_unlock(&cache
->lock
);
5606 if (++index
< BTRFS_NR_RAID_TYPES
)
5608 up_read(&info
->groups_sem
);
5611 int btrfs_reserve_extent(struct btrfs_trans_handle
*trans
,
5612 struct btrfs_root
*root
,
5613 u64 num_bytes
, u64 min_alloc_size
,
5614 u64 empty_size
, u64 hint_byte
,
5615 u64 search_end
, struct btrfs_key
*ins
,
5619 u64 search_start
= 0;
5621 data
= btrfs_get_alloc_profile(root
, data
);
5624 * the only place that sets empty_size is btrfs_realloc_node, which
5625 * is not called recursively on allocations
5627 if (empty_size
|| root
->ref_cows
)
5628 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
5629 num_bytes
+ 2 * 1024 * 1024, data
,
5630 CHUNK_ALLOC_NO_FORCE
);
5632 WARN_ON(num_bytes
< root
->sectorsize
);
5633 ret
= find_free_extent(trans
, root
, num_bytes
, empty_size
,
5634 search_start
, search_end
, hint_byte
,
5637 if (ret
== -ENOSPC
&& num_bytes
> min_alloc_size
) {
5638 num_bytes
= num_bytes
>> 1;
5639 num_bytes
= num_bytes
& ~(root
->sectorsize
- 1);
5640 num_bytes
= max(num_bytes
, min_alloc_size
);
5641 do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
5642 num_bytes
, data
, CHUNK_ALLOC_FORCE
);
5645 if (ret
== -ENOSPC
&& btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
5646 struct btrfs_space_info
*sinfo
;
5648 sinfo
= __find_space_info(root
->fs_info
, data
);
5649 printk(KERN_ERR
"btrfs allocation failed flags %llu, "
5650 "wanted %llu\n", (unsigned long long)data
,
5651 (unsigned long long)num_bytes
);
5652 dump_space_info(sinfo
, num_bytes
, 1);
5655 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, ins
->offset
);
5660 static int __btrfs_free_reserved_extent(struct btrfs_root
*root
,
5661 u64 start
, u64 len
, int pin
)
5663 struct btrfs_block_group_cache
*cache
;
5666 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
5668 printk(KERN_ERR
"Unable to find block group for %llu\n",
5669 (unsigned long long)start
);
5673 if (btrfs_test_opt(root
, DISCARD
))
5674 ret
= btrfs_discard_extent(root
, start
, len
, NULL
);
5677 pin_down_extent(root
, cache
, start
, len
, 1);
5679 btrfs_add_free_space(cache
, start
, len
);
5680 btrfs_update_reserved_bytes(cache
, len
, RESERVE_FREE
);
5682 btrfs_put_block_group(cache
);
5684 trace_btrfs_reserved_extent_free(root
, start
, len
);
5689 int btrfs_free_reserved_extent(struct btrfs_root
*root
,
5692 return __btrfs_free_reserved_extent(root
, start
, len
, 0);
5695 int btrfs_free_and_pin_reserved_extent(struct btrfs_root
*root
,
5698 return __btrfs_free_reserved_extent(root
, start
, len
, 1);
5701 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
5702 struct btrfs_root
*root
,
5703 u64 parent
, u64 root_objectid
,
5704 u64 flags
, u64 owner
, u64 offset
,
5705 struct btrfs_key
*ins
, int ref_mod
)
5708 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5709 struct btrfs_extent_item
*extent_item
;
5710 struct btrfs_extent_inline_ref
*iref
;
5711 struct btrfs_path
*path
;
5712 struct extent_buffer
*leaf
;
5717 type
= BTRFS_SHARED_DATA_REF_KEY
;
5719 type
= BTRFS_EXTENT_DATA_REF_KEY
;
5721 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
5723 path
= btrfs_alloc_path();
5727 path
->leave_spinning
= 1;
5728 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
5732 leaf
= path
->nodes
[0];
5733 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
5734 struct btrfs_extent_item
);
5735 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
5736 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
5737 btrfs_set_extent_flags(leaf
, extent_item
,
5738 flags
| BTRFS_EXTENT_FLAG_DATA
);
5740 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
5741 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
5743 struct btrfs_shared_data_ref
*ref
;
5744 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
5745 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
5746 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
5748 struct btrfs_extent_data_ref
*ref
;
5749 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
5750 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
5751 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
5752 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
5753 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
5756 btrfs_mark_buffer_dirty(path
->nodes
[0]);
5757 btrfs_free_path(path
);
5759 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1);
5761 printk(KERN_ERR
"btrfs update block group failed for %llu "
5762 "%llu\n", (unsigned long long)ins
->objectid
,
5763 (unsigned long long)ins
->offset
);
5769 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
5770 struct btrfs_root
*root
,
5771 u64 parent
, u64 root_objectid
,
5772 u64 flags
, struct btrfs_disk_key
*key
,
5773 int level
, struct btrfs_key
*ins
)
5776 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5777 struct btrfs_extent_item
*extent_item
;
5778 struct btrfs_tree_block_info
*block_info
;
5779 struct btrfs_extent_inline_ref
*iref
;
5780 struct btrfs_path
*path
;
5781 struct extent_buffer
*leaf
;
5782 u32 size
= sizeof(*extent_item
) + sizeof(*block_info
) + sizeof(*iref
);
5784 path
= btrfs_alloc_path();
5788 path
->leave_spinning
= 1;
5789 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
5793 leaf
= path
->nodes
[0];
5794 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
5795 struct btrfs_extent_item
);
5796 btrfs_set_extent_refs(leaf
, extent_item
, 1);
5797 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
5798 btrfs_set_extent_flags(leaf
, extent_item
,
5799 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
5800 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
5802 btrfs_set_tree_block_key(leaf
, block_info
, key
);
5803 btrfs_set_tree_block_level(leaf
, block_info
, level
);
5805 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
5807 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
5808 btrfs_set_extent_inline_ref_type(leaf
, iref
,
5809 BTRFS_SHARED_BLOCK_REF_KEY
);
5810 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
5812 btrfs_set_extent_inline_ref_type(leaf
, iref
,
5813 BTRFS_TREE_BLOCK_REF_KEY
);
5814 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
5817 btrfs_mark_buffer_dirty(leaf
);
5818 btrfs_free_path(path
);
5820 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1);
5822 printk(KERN_ERR
"btrfs update block group failed for %llu "
5823 "%llu\n", (unsigned long long)ins
->objectid
,
5824 (unsigned long long)ins
->offset
);
5830 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
5831 struct btrfs_root
*root
,
5832 u64 root_objectid
, u64 owner
,
5833 u64 offset
, struct btrfs_key
*ins
)
5837 BUG_ON(root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
5839 ret
= btrfs_add_delayed_data_ref(root
->fs_info
, trans
, ins
->objectid
,
5841 root_objectid
, owner
, offset
,
5842 BTRFS_ADD_DELAYED_EXTENT
, NULL
, 0);
5847 * this is used by the tree logging recovery code. It records that
5848 * an extent has been allocated and makes sure to clear the free
5849 * space cache bits as well
5851 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
5852 struct btrfs_root
*root
,
5853 u64 root_objectid
, u64 owner
, u64 offset
,
5854 struct btrfs_key
*ins
)
5857 struct btrfs_block_group_cache
*block_group
;
5858 struct btrfs_caching_control
*caching_ctl
;
5859 u64 start
= ins
->objectid
;
5860 u64 num_bytes
= ins
->offset
;
5862 block_group
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
5863 cache_block_group(block_group
, trans
, NULL
, 0);
5864 caching_ctl
= get_caching_control(block_group
);
5867 BUG_ON(!block_group_cache_done(block_group
));
5868 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
5871 mutex_lock(&caching_ctl
->mutex
);
5873 if (start
>= caching_ctl
->progress
) {
5874 ret
= add_excluded_extent(root
, start
, num_bytes
);
5876 } else if (start
+ num_bytes
<= caching_ctl
->progress
) {
5877 ret
= btrfs_remove_free_space(block_group
,
5881 num_bytes
= caching_ctl
->progress
- start
;
5882 ret
= btrfs_remove_free_space(block_group
,
5886 start
= caching_ctl
->progress
;
5887 num_bytes
= ins
->objectid
+ ins
->offset
-
5888 caching_ctl
->progress
;
5889 ret
= add_excluded_extent(root
, start
, num_bytes
);
5893 mutex_unlock(&caching_ctl
->mutex
);
5894 put_caching_control(caching_ctl
);
5897 ret
= btrfs_update_reserved_bytes(block_group
, ins
->offset
,
5898 RESERVE_ALLOC_NO_ACCOUNT
);
5900 btrfs_put_block_group(block_group
);
5901 ret
= alloc_reserved_file_extent(trans
, root
, 0, root_objectid
,
5902 0, owner
, offset
, ins
, 1);
5906 struct extent_buffer
*btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
,
5907 struct btrfs_root
*root
,
5908 u64 bytenr
, u32 blocksize
,
5911 struct extent_buffer
*buf
;
5913 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
5915 return ERR_PTR(-ENOMEM
);
5916 btrfs_set_header_generation(buf
, trans
->transid
);
5917 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, buf
, level
);
5918 btrfs_tree_lock(buf
);
5919 clean_tree_block(trans
, root
, buf
);
5921 btrfs_set_lock_blocking(buf
);
5922 btrfs_set_buffer_uptodate(buf
);
5924 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
5926 * we allow two log transactions at a time, use different
5927 * EXENT bit to differentiate dirty pages.
5929 if (root
->log_transid
% 2 == 0)
5930 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
5931 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
5933 set_extent_new(&root
->dirty_log_pages
, buf
->start
,
5934 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
5936 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
5937 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
5939 trans
->blocks_used
++;
5940 /* this returns a buffer locked for blocking */
5944 static struct btrfs_block_rsv
*
5945 use_block_rsv(struct btrfs_trans_handle
*trans
,
5946 struct btrfs_root
*root
, u32 blocksize
)
5948 struct btrfs_block_rsv
*block_rsv
;
5949 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
5952 block_rsv
= get_block_rsv(trans
, root
);
5954 if (block_rsv
->size
== 0) {
5955 ret
= reserve_metadata_bytes(root
, block_rsv
, blocksize
, 0);
5957 * If we couldn't reserve metadata bytes try and use some from
5958 * the global reserve.
5960 if (ret
&& block_rsv
!= global_rsv
) {
5961 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
5964 return ERR_PTR(ret
);
5966 return ERR_PTR(ret
);
5971 ret
= block_rsv_use_bytes(block_rsv
, blocksize
);
5975 static DEFINE_RATELIMIT_STATE(_rs
,
5976 DEFAULT_RATELIMIT_INTERVAL
,
5977 /*DEFAULT_RATELIMIT_BURST*/ 2);
5978 if (__ratelimit(&_rs
)) {
5979 printk(KERN_DEBUG
"btrfs: block rsv returned %d\n", ret
);
5982 ret
= reserve_metadata_bytes(root
, block_rsv
, blocksize
, 0);
5985 } else if (ret
&& block_rsv
!= global_rsv
) {
5986 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
5992 return ERR_PTR(-ENOSPC
);
5995 static void unuse_block_rsv(struct btrfs_block_rsv
*block_rsv
, u32 blocksize
)
5997 block_rsv_add_bytes(block_rsv
, blocksize
, 0);
5998 block_rsv_release_bytes(block_rsv
, NULL
, 0);
6002 * finds a free extent and does all the dirty work required for allocation
6003 * returns the key for the extent through ins, and a tree buffer for
6004 * the first block of the extent through buf.
6006 * returns the tree buffer or NULL.
6008 struct extent_buffer
*btrfs_alloc_free_block(struct btrfs_trans_handle
*trans
,
6009 struct btrfs_root
*root
, u32 blocksize
,
6010 u64 parent
, u64 root_objectid
,
6011 struct btrfs_disk_key
*key
, int level
,
6012 u64 hint
, u64 empty_size
, int for_cow
)
6014 struct btrfs_key ins
;
6015 struct btrfs_block_rsv
*block_rsv
;
6016 struct extent_buffer
*buf
;
6021 block_rsv
= use_block_rsv(trans
, root
, blocksize
);
6022 if (IS_ERR(block_rsv
))
6023 return ERR_CAST(block_rsv
);
6025 ret
= btrfs_reserve_extent(trans
, root
, blocksize
, blocksize
,
6026 empty_size
, hint
, (u64
)-1, &ins
, 0);
6028 unuse_block_rsv(block_rsv
, blocksize
);
6029 return ERR_PTR(ret
);
6032 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
,
6034 BUG_ON(IS_ERR(buf
));
6036 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
6038 parent
= ins
.objectid
;
6039 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
6043 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
6044 struct btrfs_delayed_extent_op
*extent_op
;
6045 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
6048 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
6050 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
6051 extent_op
->flags_to_set
= flags
;
6052 extent_op
->update_key
= 1;
6053 extent_op
->update_flags
= 1;
6054 extent_op
->is_data
= 0;
6056 ret
= btrfs_add_delayed_tree_ref(root
->fs_info
, trans
,
6058 ins
.offset
, parent
, root_objectid
,
6059 level
, BTRFS_ADD_DELAYED_EXTENT
,
6060 extent_op
, for_cow
);
6066 struct walk_control
{
6067 u64 refs
[BTRFS_MAX_LEVEL
];
6068 u64 flags
[BTRFS_MAX_LEVEL
];
6069 struct btrfs_key update_progress
;
6080 #define DROP_REFERENCE 1
6081 #define UPDATE_BACKREF 2
6083 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
6084 struct btrfs_root
*root
,
6085 struct walk_control
*wc
,
6086 struct btrfs_path
*path
)
6094 struct btrfs_key key
;
6095 struct extent_buffer
*eb
;
6100 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
6101 wc
->reada_count
= wc
->reada_count
* 2 / 3;
6102 wc
->reada_count
= max(wc
->reada_count
, 2);
6104 wc
->reada_count
= wc
->reada_count
* 3 / 2;
6105 wc
->reada_count
= min_t(int, wc
->reada_count
,
6106 BTRFS_NODEPTRS_PER_BLOCK(root
));
6109 eb
= path
->nodes
[wc
->level
];
6110 nritems
= btrfs_header_nritems(eb
);
6111 blocksize
= btrfs_level_size(root
, wc
->level
- 1);
6113 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
6114 if (nread
>= wc
->reada_count
)
6118 bytenr
= btrfs_node_blockptr(eb
, slot
);
6119 generation
= btrfs_node_ptr_generation(eb
, slot
);
6121 if (slot
== path
->slots
[wc
->level
])
6124 if (wc
->stage
== UPDATE_BACKREF
&&
6125 generation
<= root
->root_key
.offset
)
6128 /* We don't lock the tree block, it's OK to be racy here */
6129 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, blocksize
,
6134 if (wc
->stage
== DROP_REFERENCE
) {
6138 if (wc
->level
== 1 &&
6139 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
6141 if (!wc
->update_ref
||
6142 generation
<= root
->root_key
.offset
)
6144 btrfs_node_key_to_cpu(eb
, &key
, slot
);
6145 ret
= btrfs_comp_cpu_keys(&key
,
6146 &wc
->update_progress
);
6150 if (wc
->level
== 1 &&
6151 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
6155 ret
= readahead_tree_block(root
, bytenr
, blocksize
,
6161 wc
->reada_slot
= slot
;
6165 * hepler to process tree block while walking down the tree.
6167 * when wc->stage == UPDATE_BACKREF, this function updates
6168 * back refs for pointers in the block.
6170 * NOTE: return value 1 means we should stop walking down.
6172 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
6173 struct btrfs_root
*root
,
6174 struct btrfs_path
*path
,
6175 struct walk_control
*wc
, int lookup_info
)
6177 int level
= wc
->level
;
6178 struct extent_buffer
*eb
= path
->nodes
[level
];
6179 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
6182 if (wc
->stage
== UPDATE_BACKREF
&&
6183 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
6187 * when reference count of tree block is 1, it won't increase
6188 * again. once full backref flag is set, we never clear it.
6191 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
6192 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
6193 BUG_ON(!path
->locks
[level
]);
6194 ret
= btrfs_lookup_extent_info(trans
, root
,
6199 BUG_ON(wc
->refs
[level
] == 0);
6202 if (wc
->stage
== DROP_REFERENCE
) {
6203 if (wc
->refs
[level
] > 1)
6206 if (path
->locks
[level
] && !wc
->keep_locks
) {
6207 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
6208 path
->locks
[level
] = 0;
6213 /* wc->stage == UPDATE_BACKREF */
6214 if (!(wc
->flags
[level
] & flag
)) {
6215 BUG_ON(!path
->locks
[level
]);
6216 ret
= btrfs_inc_ref(trans
, root
, eb
, 1, wc
->for_reloc
);
6218 ret
= btrfs_dec_ref(trans
, root
, eb
, 0, wc
->for_reloc
);
6220 ret
= btrfs_set_disk_extent_flags(trans
, root
, eb
->start
,
6223 wc
->flags
[level
] |= flag
;
6227 * the block is shared by multiple trees, so it's not good to
6228 * keep the tree lock
6230 if (path
->locks
[level
] && level
> 0) {
6231 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
6232 path
->locks
[level
] = 0;
6238 * hepler to process tree block pointer.
6240 * when wc->stage == DROP_REFERENCE, this function checks
6241 * reference count of the block pointed to. if the block
6242 * is shared and we need update back refs for the subtree
6243 * rooted at the block, this function changes wc->stage to
6244 * UPDATE_BACKREF. if the block is shared and there is no
6245 * need to update back, this function drops the reference
6248 * NOTE: return value 1 means we should stop walking down.
6250 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
6251 struct btrfs_root
*root
,
6252 struct btrfs_path
*path
,
6253 struct walk_control
*wc
, int *lookup_info
)
6259 struct btrfs_key key
;
6260 struct extent_buffer
*next
;
6261 int level
= wc
->level
;
6265 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
6266 path
->slots
[level
]);
6268 * if the lower level block was created before the snapshot
6269 * was created, we know there is no need to update back refs
6272 if (wc
->stage
== UPDATE_BACKREF
&&
6273 generation
<= root
->root_key
.offset
) {
6278 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
6279 blocksize
= btrfs_level_size(root
, level
- 1);
6281 next
= btrfs_find_tree_block(root
, bytenr
, blocksize
);
6283 next
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
6288 btrfs_tree_lock(next
);
6289 btrfs_set_lock_blocking(next
);
6291 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, blocksize
,
6292 &wc
->refs
[level
- 1],
6293 &wc
->flags
[level
- 1]);
6295 BUG_ON(wc
->refs
[level
- 1] == 0);
6298 if (wc
->stage
== DROP_REFERENCE
) {
6299 if (wc
->refs
[level
- 1] > 1) {
6301 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
6304 if (!wc
->update_ref
||
6305 generation
<= root
->root_key
.offset
)
6308 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
6309 path
->slots
[level
]);
6310 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
6314 wc
->stage
= UPDATE_BACKREF
;
6315 wc
->shared_level
= level
- 1;
6319 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
6323 if (!btrfs_buffer_uptodate(next
, generation
)) {
6324 btrfs_tree_unlock(next
);
6325 free_extent_buffer(next
);
6331 if (reada
&& level
== 1)
6332 reada_walk_down(trans
, root
, wc
, path
);
6333 next
= read_tree_block(root
, bytenr
, blocksize
, generation
);
6336 btrfs_tree_lock(next
);
6337 btrfs_set_lock_blocking(next
);
6341 BUG_ON(level
!= btrfs_header_level(next
));
6342 path
->nodes
[level
] = next
;
6343 path
->slots
[level
] = 0;
6344 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
6350 wc
->refs
[level
- 1] = 0;
6351 wc
->flags
[level
- 1] = 0;
6352 if (wc
->stage
== DROP_REFERENCE
) {
6353 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
6354 parent
= path
->nodes
[level
]->start
;
6356 BUG_ON(root
->root_key
.objectid
!=
6357 btrfs_header_owner(path
->nodes
[level
]));
6361 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
, parent
,
6362 root
->root_key
.objectid
, level
- 1, 0, 0);
6365 btrfs_tree_unlock(next
);
6366 free_extent_buffer(next
);
6372 * hepler to process tree block while walking up the tree.
6374 * when wc->stage == DROP_REFERENCE, this function drops
6375 * reference count on the block.
6377 * when wc->stage == UPDATE_BACKREF, this function changes
6378 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6379 * to UPDATE_BACKREF previously while processing the block.
6381 * NOTE: return value 1 means we should stop walking up.
6383 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
6384 struct btrfs_root
*root
,
6385 struct btrfs_path
*path
,
6386 struct walk_control
*wc
)
6389 int level
= wc
->level
;
6390 struct extent_buffer
*eb
= path
->nodes
[level
];
6393 if (wc
->stage
== UPDATE_BACKREF
) {
6394 BUG_ON(wc
->shared_level
< level
);
6395 if (level
< wc
->shared_level
)
6398 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
6402 wc
->stage
= DROP_REFERENCE
;
6403 wc
->shared_level
= -1;
6404 path
->slots
[level
] = 0;
6407 * check reference count again if the block isn't locked.
6408 * we should start walking down the tree again if reference
6411 if (!path
->locks
[level
]) {
6413 btrfs_tree_lock(eb
);
6414 btrfs_set_lock_blocking(eb
);
6415 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
6417 ret
= btrfs_lookup_extent_info(trans
, root
,
6422 BUG_ON(wc
->refs
[level
] == 0);
6423 if (wc
->refs
[level
] == 1) {
6424 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
6430 /* wc->stage == DROP_REFERENCE */
6431 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
6433 if (wc
->refs
[level
] == 1) {
6435 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
6436 ret
= btrfs_dec_ref(trans
, root
, eb
, 1,
6439 ret
= btrfs_dec_ref(trans
, root
, eb
, 0,
6443 /* make block locked assertion in clean_tree_block happy */
6444 if (!path
->locks
[level
] &&
6445 btrfs_header_generation(eb
) == trans
->transid
) {
6446 btrfs_tree_lock(eb
);
6447 btrfs_set_lock_blocking(eb
);
6448 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
6450 clean_tree_block(trans
, root
, eb
);
6453 if (eb
== root
->node
) {
6454 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
6457 BUG_ON(root
->root_key
.objectid
!=
6458 btrfs_header_owner(eb
));
6460 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
6461 parent
= path
->nodes
[level
+ 1]->start
;
6463 BUG_ON(root
->root_key
.objectid
!=
6464 btrfs_header_owner(path
->nodes
[level
+ 1]));
6467 btrfs_free_tree_block(trans
, root
, eb
, parent
, wc
->refs
[level
] == 1, 0);
6469 wc
->refs
[level
] = 0;
6470 wc
->flags
[level
] = 0;
6474 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
6475 struct btrfs_root
*root
,
6476 struct btrfs_path
*path
,
6477 struct walk_control
*wc
)
6479 int level
= wc
->level
;
6480 int lookup_info
= 1;
6483 while (level
>= 0) {
6484 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
6491 if (path
->slots
[level
] >=
6492 btrfs_header_nritems(path
->nodes
[level
]))
6495 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
6497 path
->slots
[level
]++;
6506 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
6507 struct btrfs_root
*root
,
6508 struct btrfs_path
*path
,
6509 struct walk_control
*wc
, int max_level
)
6511 int level
= wc
->level
;
6514 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
6515 while (level
< max_level
&& path
->nodes
[level
]) {
6517 if (path
->slots
[level
] + 1 <
6518 btrfs_header_nritems(path
->nodes
[level
])) {
6519 path
->slots
[level
]++;
6522 ret
= walk_up_proc(trans
, root
, path
, wc
);
6526 if (path
->locks
[level
]) {
6527 btrfs_tree_unlock_rw(path
->nodes
[level
],
6528 path
->locks
[level
]);
6529 path
->locks
[level
] = 0;
6531 free_extent_buffer(path
->nodes
[level
]);
6532 path
->nodes
[level
] = NULL
;
6540 * drop a subvolume tree.
6542 * this function traverses the tree freeing any blocks that only
6543 * referenced by the tree.
6545 * when a shared tree block is found. this function decreases its
6546 * reference count by one. if update_ref is true, this function
6547 * also make sure backrefs for the shared block and all lower level
6548 * blocks are properly updated.
6550 void btrfs_drop_snapshot(struct btrfs_root
*root
,
6551 struct btrfs_block_rsv
*block_rsv
, int update_ref
,
6554 struct btrfs_path
*path
;
6555 struct btrfs_trans_handle
*trans
;
6556 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
6557 struct btrfs_root_item
*root_item
= &root
->root_item
;
6558 struct walk_control
*wc
;
6559 struct btrfs_key key
;
6564 path
= btrfs_alloc_path();
6570 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
6572 btrfs_free_path(path
);
6577 trans
= btrfs_start_transaction(tree_root
, 0);
6578 BUG_ON(IS_ERR(trans
));
6581 trans
->block_rsv
= block_rsv
;
6583 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
6584 level
= btrfs_header_level(root
->node
);
6585 path
->nodes
[level
] = btrfs_lock_root_node(root
);
6586 btrfs_set_lock_blocking(path
->nodes
[level
]);
6587 path
->slots
[level
] = 0;
6588 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
6589 memset(&wc
->update_progress
, 0,
6590 sizeof(wc
->update_progress
));
6592 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
6593 memcpy(&wc
->update_progress
, &key
,
6594 sizeof(wc
->update_progress
));
6596 level
= root_item
->drop_level
;
6598 path
->lowest_level
= level
;
6599 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6600 path
->lowest_level
= 0;
6608 * unlock our path, this is safe because only this
6609 * function is allowed to delete this snapshot
6611 btrfs_unlock_up_safe(path
, 0);
6613 level
= btrfs_header_level(root
->node
);
6615 btrfs_tree_lock(path
->nodes
[level
]);
6616 btrfs_set_lock_blocking(path
->nodes
[level
]);
6618 ret
= btrfs_lookup_extent_info(trans
, root
,
6619 path
->nodes
[level
]->start
,
6620 path
->nodes
[level
]->len
,
6624 BUG_ON(wc
->refs
[level
] == 0);
6626 if (level
== root_item
->drop_level
)
6629 btrfs_tree_unlock(path
->nodes
[level
]);
6630 WARN_ON(wc
->refs
[level
] != 1);
6636 wc
->shared_level
= -1;
6637 wc
->stage
= DROP_REFERENCE
;
6638 wc
->update_ref
= update_ref
;
6640 wc
->for_reloc
= for_reloc
;
6641 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
6644 ret
= walk_down_tree(trans
, root
, path
, wc
);
6650 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
6657 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
6661 if (wc
->stage
== DROP_REFERENCE
) {
6663 btrfs_node_key(path
->nodes
[level
],
6664 &root_item
->drop_progress
,
6665 path
->slots
[level
]);
6666 root_item
->drop_level
= level
;
6669 BUG_ON(wc
->level
== 0);
6670 if (btrfs_should_end_transaction(trans
, tree_root
)) {
6671 ret
= btrfs_update_root(trans
, tree_root
,
6676 btrfs_end_transaction_throttle(trans
, tree_root
);
6677 trans
= btrfs_start_transaction(tree_root
, 0);
6678 BUG_ON(IS_ERR(trans
));
6680 trans
->block_rsv
= block_rsv
;
6683 btrfs_release_path(path
);
6686 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
6689 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
6690 ret
= btrfs_find_last_root(tree_root
, root
->root_key
.objectid
,
6694 /* if we fail to delete the orphan item this time
6695 * around, it'll get picked up the next time.
6697 * The most common failure here is just -ENOENT.
6699 btrfs_del_orphan_item(trans
, tree_root
,
6700 root
->root_key
.objectid
);
6704 if (root
->in_radix
) {
6705 btrfs_free_fs_root(tree_root
->fs_info
, root
);
6707 free_extent_buffer(root
->node
);
6708 free_extent_buffer(root
->commit_root
);
6712 btrfs_end_transaction_throttle(trans
, tree_root
);
6714 btrfs_free_path(path
);
6717 btrfs_std_error(root
->fs_info
, err
);
6722 * drop subtree rooted at tree block 'node'.
6724 * NOTE: this function will unlock and release tree block 'node'
6725 * only used by relocation code
6727 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
6728 struct btrfs_root
*root
,
6729 struct extent_buffer
*node
,
6730 struct extent_buffer
*parent
)
6732 struct btrfs_path
*path
;
6733 struct walk_control
*wc
;
6739 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
6741 path
= btrfs_alloc_path();
6745 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
6747 btrfs_free_path(path
);
6751 btrfs_assert_tree_locked(parent
);
6752 parent_level
= btrfs_header_level(parent
);
6753 extent_buffer_get(parent
);
6754 path
->nodes
[parent_level
] = parent
;
6755 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
6757 btrfs_assert_tree_locked(node
);
6758 level
= btrfs_header_level(node
);
6759 path
->nodes
[level
] = node
;
6760 path
->slots
[level
] = 0;
6761 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
6763 wc
->refs
[parent_level
] = 1;
6764 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
6766 wc
->shared_level
= -1;
6767 wc
->stage
= DROP_REFERENCE
;
6771 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
6774 wret
= walk_down_tree(trans
, root
, path
, wc
);
6780 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
6788 btrfs_free_path(path
);
6792 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
6795 u64 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
6796 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
6799 * we add in the count of missing devices because we want
6800 * to make sure that any RAID levels on a degraded FS
6801 * continue to be honored.
6803 num_devices
= root
->fs_info
->fs_devices
->rw_devices
+
6804 root
->fs_info
->fs_devices
->missing_devices
;
6806 if (num_devices
== 1) {
6807 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
6808 stripped
= flags
& ~stripped
;
6810 /* turn raid0 into single device chunks */
6811 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
6814 /* turn mirroring into duplication */
6815 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
6816 BTRFS_BLOCK_GROUP_RAID10
))
6817 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
6820 /* they already had raid on here, just return */
6821 if (flags
& stripped
)
6824 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
6825 stripped
= flags
& ~stripped
;
6827 /* switch duplicated blocks with raid1 */
6828 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
6829 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
6831 /* turn single device chunks into raid0 */
6832 return stripped
| BTRFS_BLOCK_GROUP_RAID0
;
6837 static int set_block_group_ro(struct btrfs_block_group_cache
*cache
, int force
)
6839 struct btrfs_space_info
*sinfo
= cache
->space_info
;
6841 u64 min_allocable_bytes
;
6846 * We need some metadata space and system metadata space for
6847 * allocating chunks in some corner cases until we force to set
6848 * it to be readonly.
6851 (BTRFS_BLOCK_GROUP_SYSTEM
| BTRFS_BLOCK_GROUP_METADATA
)) &&
6853 min_allocable_bytes
= 1 * 1024 * 1024;
6855 min_allocable_bytes
= 0;
6857 spin_lock(&sinfo
->lock
);
6858 spin_lock(&cache
->lock
);
6865 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
6866 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
6868 if (sinfo
->bytes_used
+ sinfo
->bytes_reserved
+ sinfo
->bytes_pinned
+
6869 sinfo
->bytes_may_use
+ sinfo
->bytes_readonly
+ num_bytes
+
6870 min_allocable_bytes
<= sinfo
->total_bytes
) {
6871 sinfo
->bytes_readonly
+= num_bytes
;
6876 spin_unlock(&cache
->lock
);
6877 spin_unlock(&sinfo
->lock
);
6881 int btrfs_set_block_group_ro(struct btrfs_root
*root
,
6882 struct btrfs_block_group_cache
*cache
)
6885 struct btrfs_trans_handle
*trans
;
6891 trans
= btrfs_join_transaction(root
);
6892 BUG_ON(IS_ERR(trans
));
6894 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
6895 if (alloc_flags
!= cache
->flags
)
6896 do_chunk_alloc(trans
, root
, 2 * 1024 * 1024, alloc_flags
,
6899 ret
= set_block_group_ro(cache
, 0);
6902 alloc_flags
= get_alloc_profile(root
, cache
->space_info
->flags
);
6903 ret
= do_chunk_alloc(trans
, root
, 2 * 1024 * 1024, alloc_flags
,
6907 ret
= set_block_group_ro(cache
, 0);
6909 btrfs_end_transaction(trans
, root
);
6913 int btrfs_force_chunk_alloc(struct btrfs_trans_handle
*trans
,
6914 struct btrfs_root
*root
, u64 type
)
6916 u64 alloc_flags
= get_alloc_profile(root
, type
);
6917 return do_chunk_alloc(trans
, root
, 2 * 1024 * 1024, alloc_flags
,
6922 * helper to account the unused space of all the readonly block group in the
6923 * list. takes mirrors into account.
6925 static u64
__btrfs_get_ro_block_group_free_space(struct list_head
*groups_list
)
6927 struct btrfs_block_group_cache
*block_group
;
6931 list_for_each_entry(block_group
, groups_list
, list
) {
6932 spin_lock(&block_group
->lock
);
6934 if (!block_group
->ro
) {
6935 spin_unlock(&block_group
->lock
);
6939 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_RAID1
|
6940 BTRFS_BLOCK_GROUP_RAID10
|
6941 BTRFS_BLOCK_GROUP_DUP
))
6946 free_bytes
+= (block_group
->key
.offset
-
6947 btrfs_block_group_used(&block_group
->item
)) *
6950 spin_unlock(&block_group
->lock
);
6957 * helper to account the unused space of all the readonly block group in the
6958 * space_info. takes mirrors into account.
6960 u64
btrfs_account_ro_block_groups_free_space(struct btrfs_space_info
*sinfo
)
6965 spin_lock(&sinfo
->lock
);
6967 for(i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
6968 if (!list_empty(&sinfo
->block_groups
[i
]))
6969 free_bytes
+= __btrfs_get_ro_block_group_free_space(
6970 &sinfo
->block_groups
[i
]);
6972 spin_unlock(&sinfo
->lock
);
6977 int btrfs_set_block_group_rw(struct btrfs_root
*root
,
6978 struct btrfs_block_group_cache
*cache
)
6980 struct btrfs_space_info
*sinfo
= cache
->space_info
;
6985 spin_lock(&sinfo
->lock
);
6986 spin_lock(&cache
->lock
);
6987 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
6988 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
6989 sinfo
->bytes_readonly
-= num_bytes
;
6991 spin_unlock(&cache
->lock
);
6992 spin_unlock(&sinfo
->lock
);
6997 * checks to see if its even possible to relocate this block group.
6999 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7000 * ok to go ahead and try.
7002 int btrfs_can_relocate(struct btrfs_root
*root
, u64 bytenr
)
7004 struct btrfs_block_group_cache
*block_group
;
7005 struct btrfs_space_info
*space_info
;
7006 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
7007 struct btrfs_device
*device
;
7015 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
7017 /* odd, couldn't find the block group, leave it alone */
7021 min_free
= btrfs_block_group_used(&block_group
->item
);
7023 /* no bytes used, we're good */
7027 space_info
= block_group
->space_info
;
7028 spin_lock(&space_info
->lock
);
7030 full
= space_info
->full
;
7033 * if this is the last block group we have in this space, we can't
7034 * relocate it unless we're able to allocate a new chunk below.
7036 * Otherwise, we need to make sure we have room in the space to handle
7037 * all of the extents from this block group. If we can, we're good
7039 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
7040 (space_info
->bytes_used
+ space_info
->bytes_reserved
+
7041 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
7042 min_free
< space_info
->total_bytes
)) {
7043 spin_unlock(&space_info
->lock
);
7046 spin_unlock(&space_info
->lock
);
7049 * ok we don't have enough space, but maybe we have free space on our
7050 * devices to allocate new chunks for relocation, so loop through our
7051 * alloc devices and guess if we have enough space. However, if we
7052 * were marked as full, then we know there aren't enough chunks, and we
7067 index
= get_block_group_index(block_group
);
7072 } else if (index
== 1) {
7074 } else if (index
== 2) {
7077 } else if (index
== 3) {
7078 dev_min
= fs_devices
->rw_devices
;
7079 do_div(min_free
, dev_min
);
7082 mutex_lock(&root
->fs_info
->chunk_mutex
);
7083 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
7087 * check to make sure we can actually find a chunk with enough
7088 * space to fit our block group in.
7090 if (device
->total_bytes
> device
->bytes_used
+ min_free
) {
7091 ret
= find_free_dev_extent(NULL
, device
, min_free
,
7096 if (dev_nr
>= dev_min
)
7102 mutex_unlock(&root
->fs_info
->chunk_mutex
);
7104 btrfs_put_block_group(block_group
);
7108 static int find_first_block_group(struct btrfs_root
*root
,
7109 struct btrfs_path
*path
, struct btrfs_key
*key
)
7112 struct btrfs_key found_key
;
7113 struct extent_buffer
*leaf
;
7116 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
7121 slot
= path
->slots
[0];
7122 leaf
= path
->nodes
[0];
7123 if (slot
>= btrfs_header_nritems(leaf
)) {
7124 ret
= btrfs_next_leaf(root
, path
);
7131 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
7133 if (found_key
.objectid
>= key
->objectid
&&
7134 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
7144 void btrfs_put_block_group_cache(struct btrfs_fs_info
*info
)
7146 struct btrfs_block_group_cache
*block_group
;
7150 struct inode
*inode
;
7152 block_group
= btrfs_lookup_first_block_group(info
, last
);
7153 while (block_group
) {
7154 spin_lock(&block_group
->lock
);
7155 if (block_group
->iref
)
7157 spin_unlock(&block_group
->lock
);
7158 block_group
= next_block_group(info
->tree_root
,
7168 inode
= block_group
->inode
;
7169 block_group
->iref
= 0;
7170 block_group
->inode
= NULL
;
7171 spin_unlock(&block_group
->lock
);
7173 last
= block_group
->key
.objectid
+ block_group
->key
.offset
;
7174 btrfs_put_block_group(block_group
);
7178 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
7180 struct btrfs_block_group_cache
*block_group
;
7181 struct btrfs_space_info
*space_info
;
7182 struct btrfs_caching_control
*caching_ctl
;
7185 down_write(&info
->extent_commit_sem
);
7186 while (!list_empty(&info
->caching_block_groups
)) {
7187 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
7188 struct btrfs_caching_control
, list
);
7189 list_del(&caching_ctl
->list
);
7190 put_caching_control(caching_ctl
);
7192 up_write(&info
->extent_commit_sem
);
7194 spin_lock(&info
->block_group_cache_lock
);
7195 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
7196 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
7198 rb_erase(&block_group
->cache_node
,
7199 &info
->block_group_cache_tree
);
7200 spin_unlock(&info
->block_group_cache_lock
);
7202 down_write(&block_group
->space_info
->groups_sem
);
7203 list_del(&block_group
->list
);
7204 up_write(&block_group
->space_info
->groups_sem
);
7206 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
7207 wait_block_group_cache_done(block_group
);
7210 * We haven't cached this block group, which means we could
7211 * possibly have excluded extents on this block group.
7213 if (block_group
->cached
== BTRFS_CACHE_NO
)
7214 free_excluded_extents(info
->extent_root
, block_group
);
7216 btrfs_remove_free_space_cache(block_group
);
7217 btrfs_put_block_group(block_group
);
7219 spin_lock(&info
->block_group_cache_lock
);
7221 spin_unlock(&info
->block_group_cache_lock
);
7223 /* now that all the block groups are freed, go through and
7224 * free all the space_info structs. This is only called during
7225 * the final stages of unmount, and so we know nobody is
7226 * using them. We call synchronize_rcu() once before we start,
7227 * just to be on the safe side.
7231 release_global_block_rsv(info
);
7233 while(!list_empty(&info
->space_info
)) {
7234 space_info
= list_entry(info
->space_info
.next
,
7235 struct btrfs_space_info
,
7237 if (space_info
->bytes_pinned
> 0 ||
7238 space_info
->bytes_reserved
> 0 ||
7239 space_info
->bytes_may_use
> 0) {
7241 dump_space_info(space_info
, 0, 0);
7243 list_del(&space_info
->list
);
7249 static void __link_block_group(struct btrfs_space_info
*space_info
,
7250 struct btrfs_block_group_cache
*cache
)
7252 int index
= get_block_group_index(cache
);
7254 down_write(&space_info
->groups_sem
);
7255 list_add_tail(&cache
->list
, &space_info
->block_groups
[index
]);
7256 up_write(&space_info
->groups_sem
);
7259 int btrfs_read_block_groups(struct btrfs_root
*root
)
7261 struct btrfs_path
*path
;
7263 struct btrfs_block_group_cache
*cache
;
7264 struct btrfs_fs_info
*info
= root
->fs_info
;
7265 struct btrfs_space_info
*space_info
;
7266 struct btrfs_key key
;
7267 struct btrfs_key found_key
;
7268 struct extent_buffer
*leaf
;
7272 root
= info
->extent_root
;
7275 btrfs_set_key_type(&key
, BTRFS_BLOCK_GROUP_ITEM_KEY
);
7276 path
= btrfs_alloc_path();
7281 cache_gen
= btrfs_super_cache_generation(root
->fs_info
->super_copy
);
7282 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
7283 btrfs_super_generation(root
->fs_info
->super_copy
) != cache_gen
)
7285 if (btrfs_test_opt(root
, CLEAR_CACHE
))
7289 ret
= find_first_block_group(root
, path
, &key
);
7294 leaf
= path
->nodes
[0];
7295 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
7296 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
7301 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
7303 if (!cache
->free_space_ctl
) {
7309 atomic_set(&cache
->count
, 1);
7310 spin_lock_init(&cache
->lock
);
7311 cache
->fs_info
= info
;
7312 INIT_LIST_HEAD(&cache
->list
);
7313 INIT_LIST_HEAD(&cache
->cluster_list
);
7316 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
7318 read_extent_buffer(leaf
, &cache
->item
,
7319 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
7320 sizeof(cache
->item
));
7321 memcpy(&cache
->key
, &found_key
, sizeof(found_key
));
7323 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
7324 btrfs_release_path(path
);
7325 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
7326 cache
->sectorsize
= root
->sectorsize
;
7328 btrfs_init_free_space_ctl(cache
);
7331 * We need to exclude the super stripes now so that the space
7332 * info has super bytes accounted for, otherwise we'll think
7333 * we have more space than we actually do.
7335 exclude_super_stripes(root
, cache
);
7338 * check for two cases, either we are full, and therefore
7339 * don't need to bother with the caching work since we won't
7340 * find any space, or we are empty, and we can just add all
7341 * the space in and be done with it. This saves us _alot_ of
7342 * time, particularly in the full case.
7344 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
7345 cache
->last_byte_to_unpin
= (u64
)-1;
7346 cache
->cached
= BTRFS_CACHE_FINISHED
;
7347 free_excluded_extents(root
, cache
);
7348 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
7349 cache
->last_byte_to_unpin
= (u64
)-1;
7350 cache
->cached
= BTRFS_CACHE_FINISHED
;
7351 add_new_free_space(cache
, root
->fs_info
,
7353 found_key
.objectid
+
7355 free_excluded_extents(root
, cache
);
7358 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
7359 btrfs_block_group_used(&cache
->item
),
7362 cache
->space_info
= space_info
;
7363 spin_lock(&cache
->space_info
->lock
);
7364 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
7365 spin_unlock(&cache
->space_info
->lock
);
7367 __link_block_group(space_info
, cache
);
7369 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
7372 set_avail_alloc_bits(root
->fs_info
, cache
->flags
);
7373 if (btrfs_chunk_readonly(root
, cache
->key
.objectid
))
7374 set_block_group_ro(cache
, 1);
7377 list_for_each_entry_rcu(space_info
, &root
->fs_info
->space_info
, list
) {
7378 if (!(get_alloc_profile(root
, space_info
->flags
) &
7379 (BTRFS_BLOCK_GROUP_RAID10
|
7380 BTRFS_BLOCK_GROUP_RAID1
|
7381 BTRFS_BLOCK_GROUP_DUP
)))
7384 * avoid allocating from un-mirrored block group if there are
7385 * mirrored block groups.
7387 list_for_each_entry(cache
, &space_info
->block_groups
[3], list
)
7388 set_block_group_ro(cache
, 1);
7389 list_for_each_entry(cache
, &space_info
->block_groups
[4], list
)
7390 set_block_group_ro(cache
, 1);
7393 init_global_block_rsv(info
);
7396 btrfs_free_path(path
);
7400 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
7401 struct btrfs_root
*root
, u64 bytes_used
,
7402 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
7406 struct btrfs_root
*extent_root
;
7407 struct btrfs_block_group_cache
*cache
;
7409 extent_root
= root
->fs_info
->extent_root
;
7411 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
7413 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
7416 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
7418 if (!cache
->free_space_ctl
) {
7423 cache
->key
.objectid
= chunk_offset
;
7424 cache
->key
.offset
= size
;
7425 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
7426 cache
->sectorsize
= root
->sectorsize
;
7427 cache
->fs_info
= root
->fs_info
;
7429 atomic_set(&cache
->count
, 1);
7430 spin_lock_init(&cache
->lock
);
7431 INIT_LIST_HEAD(&cache
->list
);
7432 INIT_LIST_HEAD(&cache
->cluster_list
);
7434 btrfs_init_free_space_ctl(cache
);
7436 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
7437 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
7438 cache
->flags
= type
;
7439 btrfs_set_block_group_flags(&cache
->item
, type
);
7441 cache
->last_byte_to_unpin
= (u64
)-1;
7442 cache
->cached
= BTRFS_CACHE_FINISHED
;
7443 exclude_super_stripes(root
, cache
);
7445 add_new_free_space(cache
, root
->fs_info
, chunk_offset
,
7446 chunk_offset
+ size
);
7448 free_excluded_extents(root
, cache
);
7450 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
7451 &cache
->space_info
);
7454 spin_lock(&cache
->space_info
->lock
);
7455 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
7456 spin_unlock(&cache
->space_info
->lock
);
7458 __link_block_group(cache
->space_info
, cache
);
7460 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
7463 ret
= btrfs_insert_item(trans
, extent_root
, &cache
->key
, &cache
->item
,
7464 sizeof(cache
->item
));
7467 set_avail_alloc_bits(extent_root
->fs_info
, type
);
7472 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
7473 struct btrfs_root
*root
, u64 group_start
)
7475 struct btrfs_path
*path
;
7476 struct btrfs_block_group_cache
*block_group
;
7477 struct btrfs_free_cluster
*cluster
;
7478 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
7479 struct btrfs_key key
;
7480 struct inode
*inode
;
7484 root
= root
->fs_info
->extent_root
;
7486 block_group
= btrfs_lookup_block_group(root
->fs_info
, group_start
);
7487 BUG_ON(!block_group
);
7488 BUG_ON(!block_group
->ro
);
7491 * Free the reserved super bytes from this block group before
7494 free_excluded_extents(root
, block_group
);
7496 memcpy(&key
, &block_group
->key
, sizeof(key
));
7497 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
7498 BTRFS_BLOCK_GROUP_RAID1
|
7499 BTRFS_BLOCK_GROUP_RAID10
))
7504 /* make sure this block group isn't part of an allocation cluster */
7505 cluster
= &root
->fs_info
->data_alloc_cluster
;
7506 spin_lock(&cluster
->refill_lock
);
7507 btrfs_return_cluster_to_free_space(block_group
, cluster
);
7508 spin_unlock(&cluster
->refill_lock
);
7511 * make sure this block group isn't part of a metadata
7512 * allocation cluster
7514 cluster
= &root
->fs_info
->meta_alloc_cluster
;
7515 spin_lock(&cluster
->refill_lock
);
7516 btrfs_return_cluster_to_free_space(block_group
, cluster
);
7517 spin_unlock(&cluster
->refill_lock
);
7519 path
= btrfs_alloc_path();
7525 inode
= lookup_free_space_inode(tree_root
, block_group
, path
);
7526 if (!IS_ERR(inode
)) {
7527 ret
= btrfs_orphan_add(trans
, inode
);
7530 /* One for the block groups ref */
7531 spin_lock(&block_group
->lock
);
7532 if (block_group
->iref
) {
7533 block_group
->iref
= 0;
7534 block_group
->inode
= NULL
;
7535 spin_unlock(&block_group
->lock
);
7538 spin_unlock(&block_group
->lock
);
7540 /* One for our lookup ref */
7541 btrfs_add_delayed_iput(inode
);
7544 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
7545 key
.offset
= block_group
->key
.objectid
;
7548 ret
= btrfs_search_slot(trans
, tree_root
, &key
, path
, -1, 1);
7552 btrfs_release_path(path
);
7554 ret
= btrfs_del_item(trans
, tree_root
, path
);
7557 btrfs_release_path(path
);
7560 spin_lock(&root
->fs_info
->block_group_cache_lock
);
7561 rb_erase(&block_group
->cache_node
,
7562 &root
->fs_info
->block_group_cache_tree
);
7563 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
7565 down_write(&block_group
->space_info
->groups_sem
);
7567 * we must use list_del_init so people can check to see if they
7568 * are still on the list after taking the semaphore
7570 list_del_init(&block_group
->list
);
7571 up_write(&block_group
->space_info
->groups_sem
);
7573 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
7574 wait_block_group_cache_done(block_group
);
7576 btrfs_remove_free_space_cache(block_group
);
7578 spin_lock(&block_group
->space_info
->lock
);
7579 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
7580 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
7581 block_group
->space_info
->disk_total
-= block_group
->key
.offset
* factor
;
7582 spin_unlock(&block_group
->space_info
->lock
);
7584 memcpy(&key
, &block_group
->key
, sizeof(key
));
7586 btrfs_clear_space_info_full(root
->fs_info
);
7588 btrfs_put_block_group(block_group
);
7589 btrfs_put_block_group(block_group
);
7591 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
7597 ret
= btrfs_del_item(trans
, root
, path
);
7599 btrfs_free_path(path
);
7603 int btrfs_init_space_info(struct btrfs_fs_info
*fs_info
)
7605 struct btrfs_space_info
*space_info
;
7606 struct btrfs_super_block
*disk_super
;
7612 disk_super
= fs_info
->super_copy
;
7613 if (!btrfs_super_root(disk_super
))
7616 features
= btrfs_super_incompat_flags(disk_super
);
7617 if (features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
7620 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
7621 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
7626 flags
= BTRFS_BLOCK_GROUP_METADATA
| BTRFS_BLOCK_GROUP_DATA
;
7627 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
7629 flags
= BTRFS_BLOCK_GROUP_METADATA
;
7630 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
7634 flags
= BTRFS_BLOCK_GROUP_DATA
;
7635 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
7641 int btrfs_error_unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
7643 return unpin_extent_range(root
, start
, end
);
7646 int btrfs_error_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
7647 u64 num_bytes
, u64
*actual_bytes
)
7649 return btrfs_discard_extent(root
, bytenr
, num_bytes
, actual_bytes
);
7652 int btrfs_trim_fs(struct btrfs_root
*root
, struct fstrim_range
*range
)
7654 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7655 struct btrfs_block_group_cache
*cache
= NULL
;
7662 cache
= btrfs_lookup_block_group(fs_info
, range
->start
);
7665 if (cache
->key
.objectid
>= (range
->start
+ range
->len
)) {
7666 btrfs_put_block_group(cache
);
7670 start
= max(range
->start
, cache
->key
.objectid
);
7671 end
= min(range
->start
+ range
->len
,
7672 cache
->key
.objectid
+ cache
->key
.offset
);
7674 if (end
- start
>= range
->minlen
) {
7675 if (!block_group_cache_done(cache
)) {
7676 ret
= cache_block_group(cache
, NULL
, root
, 0);
7678 wait_block_group_cache_done(cache
);
7680 ret
= btrfs_trim_block_group(cache
,
7686 trimmed
+= group_trimmed
;
7688 btrfs_put_block_group(cache
);
7693 cache
= next_block_group(fs_info
->tree_root
, cache
);
7696 range
->len
= trimmed
;