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>
30 #include "print-tree.h"
31 #include "transaction.h"
34 #include "free-space-cache.h"
36 /* control flags for do_chunk_alloc's force field
37 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
38 * if we really need one.
40 * CHUNK_ALLOC_FORCE means it must try to allocate one
42 * CHUNK_ALLOC_LIMITED means to only try and allocate one
43 * if we have very few chunks already allocated. This is
44 * used as part of the clustering code to help make sure
45 * we have a good pool of storage to cluster in, without
46 * filling the FS with empty chunks
50 CHUNK_ALLOC_NO_FORCE
= 0,
51 CHUNK_ALLOC_FORCE
= 1,
52 CHUNK_ALLOC_LIMITED
= 2,
55 static int update_block_group(struct btrfs_trans_handle
*trans
,
56 struct btrfs_root
*root
,
57 u64 bytenr
, u64 num_bytes
, int alloc
);
58 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
59 struct btrfs_root
*root
,
60 u64 bytenr
, u64 num_bytes
, u64 parent
,
61 u64 root_objectid
, u64 owner_objectid
,
62 u64 owner_offset
, int refs_to_drop
,
63 struct btrfs_delayed_extent_op
*extra_op
);
64 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
65 struct extent_buffer
*leaf
,
66 struct btrfs_extent_item
*ei
);
67 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
68 struct btrfs_root
*root
,
69 u64 parent
, u64 root_objectid
,
70 u64 flags
, u64 owner
, u64 offset
,
71 struct btrfs_key
*ins
, int ref_mod
);
72 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
73 struct btrfs_root
*root
,
74 u64 parent
, u64 root_objectid
,
75 u64 flags
, struct btrfs_disk_key
*key
,
76 int level
, struct btrfs_key
*ins
);
77 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
78 struct btrfs_root
*extent_root
, u64 alloc_bytes
,
79 u64 flags
, int force
);
80 static int find_next_key(struct btrfs_path
*path
, int level
,
81 struct btrfs_key
*key
);
82 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
83 int dump_block_groups
);
86 block_group_cache_done(struct btrfs_block_group_cache
*cache
)
89 return cache
->cached
== BTRFS_CACHE_FINISHED
;
92 static int block_group_bits(struct btrfs_block_group_cache
*cache
, u64 bits
)
94 return (cache
->flags
& bits
) == bits
;
97 static void btrfs_get_block_group(struct btrfs_block_group_cache
*cache
)
99 atomic_inc(&cache
->count
);
102 void btrfs_put_block_group(struct btrfs_block_group_cache
*cache
)
104 if (atomic_dec_and_test(&cache
->count
)) {
105 WARN_ON(cache
->pinned
> 0);
106 WARN_ON(cache
->reserved
> 0);
107 WARN_ON(cache
->reserved_pinned
> 0);
108 kfree(cache
->free_space_ctl
);
114 * this adds the block group to the fs_info rb tree for the block group
117 static int btrfs_add_block_group_cache(struct btrfs_fs_info
*info
,
118 struct btrfs_block_group_cache
*block_group
)
121 struct rb_node
*parent
= NULL
;
122 struct btrfs_block_group_cache
*cache
;
124 spin_lock(&info
->block_group_cache_lock
);
125 p
= &info
->block_group_cache_tree
.rb_node
;
129 cache
= rb_entry(parent
, struct btrfs_block_group_cache
,
131 if (block_group
->key
.objectid
< cache
->key
.objectid
) {
133 } else if (block_group
->key
.objectid
> cache
->key
.objectid
) {
136 spin_unlock(&info
->block_group_cache_lock
);
141 rb_link_node(&block_group
->cache_node
, parent
, p
);
142 rb_insert_color(&block_group
->cache_node
,
143 &info
->block_group_cache_tree
);
144 spin_unlock(&info
->block_group_cache_lock
);
150 * This will return the block group at or after bytenr if contains is 0, else
151 * it will return the block group that contains the bytenr
153 static struct btrfs_block_group_cache
*
154 block_group_cache_tree_search(struct btrfs_fs_info
*info
, u64 bytenr
,
157 struct btrfs_block_group_cache
*cache
, *ret
= NULL
;
161 spin_lock(&info
->block_group_cache_lock
);
162 n
= info
->block_group_cache_tree
.rb_node
;
165 cache
= rb_entry(n
, struct btrfs_block_group_cache
,
167 end
= cache
->key
.objectid
+ cache
->key
.offset
- 1;
168 start
= cache
->key
.objectid
;
170 if (bytenr
< start
) {
171 if (!contains
&& (!ret
|| start
< ret
->key
.objectid
))
174 } else if (bytenr
> start
) {
175 if (contains
&& bytenr
<= end
) {
186 btrfs_get_block_group(ret
);
187 spin_unlock(&info
->block_group_cache_lock
);
192 static int add_excluded_extent(struct btrfs_root
*root
,
193 u64 start
, u64 num_bytes
)
195 u64 end
= start
+ num_bytes
- 1;
196 set_extent_bits(&root
->fs_info
->freed_extents
[0],
197 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
198 set_extent_bits(&root
->fs_info
->freed_extents
[1],
199 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
203 static void free_excluded_extents(struct btrfs_root
*root
,
204 struct btrfs_block_group_cache
*cache
)
208 start
= cache
->key
.objectid
;
209 end
= start
+ cache
->key
.offset
- 1;
211 clear_extent_bits(&root
->fs_info
->freed_extents
[0],
212 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
213 clear_extent_bits(&root
->fs_info
->freed_extents
[1],
214 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
217 static int exclude_super_stripes(struct btrfs_root
*root
,
218 struct btrfs_block_group_cache
*cache
)
225 if (cache
->key
.objectid
< BTRFS_SUPER_INFO_OFFSET
) {
226 stripe_len
= BTRFS_SUPER_INFO_OFFSET
- cache
->key
.objectid
;
227 cache
->bytes_super
+= stripe_len
;
228 ret
= add_excluded_extent(root
, cache
->key
.objectid
,
233 for (i
= 0; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
234 bytenr
= btrfs_sb_offset(i
);
235 ret
= btrfs_rmap_block(&root
->fs_info
->mapping_tree
,
236 cache
->key
.objectid
, bytenr
,
237 0, &logical
, &nr
, &stripe_len
);
241 cache
->bytes_super
+= stripe_len
;
242 ret
= add_excluded_extent(root
, logical
[nr
],
252 static struct btrfs_caching_control
*
253 get_caching_control(struct btrfs_block_group_cache
*cache
)
255 struct btrfs_caching_control
*ctl
;
257 spin_lock(&cache
->lock
);
258 if (cache
->cached
!= BTRFS_CACHE_STARTED
) {
259 spin_unlock(&cache
->lock
);
263 /* We're loading it the fast way, so we don't have a caching_ctl. */
264 if (!cache
->caching_ctl
) {
265 spin_unlock(&cache
->lock
);
269 ctl
= cache
->caching_ctl
;
270 atomic_inc(&ctl
->count
);
271 spin_unlock(&cache
->lock
);
275 static void put_caching_control(struct btrfs_caching_control
*ctl
)
277 if (atomic_dec_and_test(&ctl
->count
))
282 * this is only called by cache_block_group, since we could have freed extents
283 * we need to check the pinned_extents for any extents that can't be used yet
284 * since their free space will be released as soon as the transaction commits.
286 static u64
add_new_free_space(struct btrfs_block_group_cache
*block_group
,
287 struct btrfs_fs_info
*info
, u64 start
, u64 end
)
289 u64 extent_start
, extent_end
, size
, total_added
= 0;
292 while (start
< end
) {
293 ret
= find_first_extent_bit(info
->pinned_extents
, start
,
294 &extent_start
, &extent_end
,
295 EXTENT_DIRTY
| EXTENT_UPTODATE
);
299 if (extent_start
<= start
) {
300 start
= extent_end
+ 1;
301 } else if (extent_start
> start
&& extent_start
< end
) {
302 size
= extent_start
- start
;
304 ret
= btrfs_add_free_space(block_group
, start
,
307 start
= extent_end
+ 1;
316 ret
= btrfs_add_free_space(block_group
, start
, size
);
323 static noinline
void caching_thread(struct btrfs_work
*work
)
325 struct btrfs_block_group_cache
*block_group
;
326 struct btrfs_fs_info
*fs_info
;
327 struct btrfs_caching_control
*caching_ctl
;
328 struct btrfs_root
*extent_root
;
329 struct btrfs_path
*path
;
330 struct extent_buffer
*leaf
;
331 struct btrfs_key key
;
337 caching_ctl
= container_of(work
, struct btrfs_caching_control
, work
);
338 block_group
= caching_ctl
->block_group
;
339 fs_info
= block_group
->fs_info
;
340 extent_root
= fs_info
->extent_root
;
342 path
= btrfs_alloc_path();
346 last
= max_t(u64
, block_group
->key
.objectid
, BTRFS_SUPER_INFO_OFFSET
);
349 * We don't want to deadlock with somebody trying to allocate a new
350 * extent for the extent root while also trying to search the extent
351 * root to add free space. So we skip locking and search the commit
352 * root, since its read-only
354 path
->skip_locking
= 1;
355 path
->search_commit_root
= 1;
360 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
362 mutex_lock(&caching_ctl
->mutex
);
363 /* need to make sure the commit_root doesn't disappear */
364 down_read(&fs_info
->extent_commit_sem
);
366 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
370 leaf
= path
->nodes
[0];
371 nritems
= btrfs_header_nritems(leaf
);
374 if (btrfs_fs_closing(fs_info
) > 1) {
379 if (path
->slots
[0] < nritems
) {
380 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
382 ret
= find_next_key(path
, 0, &key
);
386 if (need_resched() ||
387 btrfs_next_leaf(extent_root
, path
)) {
388 caching_ctl
->progress
= last
;
389 btrfs_release_path(path
);
390 up_read(&fs_info
->extent_commit_sem
);
391 mutex_unlock(&caching_ctl
->mutex
);
395 leaf
= path
->nodes
[0];
396 nritems
= btrfs_header_nritems(leaf
);
400 if (key
.objectid
< block_group
->key
.objectid
) {
405 if (key
.objectid
>= block_group
->key
.objectid
+
406 block_group
->key
.offset
)
409 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
) {
410 total_found
+= add_new_free_space(block_group
,
413 last
= key
.objectid
+ key
.offset
;
415 if (total_found
> (1024 * 1024 * 2)) {
417 wake_up(&caching_ctl
->wait
);
424 total_found
+= add_new_free_space(block_group
, fs_info
, last
,
425 block_group
->key
.objectid
+
426 block_group
->key
.offset
);
427 caching_ctl
->progress
= (u64
)-1;
429 spin_lock(&block_group
->lock
);
430 block_group
->caching_ctl
= NULL
;
431 block_group
->cached
= BTRFS_CACHE_FINISHED
;
432 spin_unlock(&block_group
->lock
);
435 btrfs_free_path(path
);
436 up_read(&fs_info
->extent_commit_sem
);
438 free_excluded_extents(extent_root
, block_group
);
440 mutex_unlock(&caching_ctl
->mutex
);
442 wake_up(&caching_ctl
->wait
);
444 put_caching_control(caching_ctl
);
445 btrfs_put_block_group(block_group
);
448 static int cache_block_group(struct btrfs_block_group_cache
*cache
,
449 struct btrfs_trans_handle
*trans
,
450 struct btrfs_root
*root
,
453 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
454 struct btrfs_caching_control
*caching_ctl
;
458 if (cache
->cached
!= BTRFS_CACHE_NO
)
462 * We can't do the read from on-disk cache during a commit since we need
463 * to have the normal tree locking. Also if we are currently trying to
464 * allocate blocks for the tree root we can't do the fast caching since
465 * we likely hold important locks.
467 if (trans
&& (!trans
->transaction
->in_commit
) &&
468 (root
&& root
!= root
->fs_info
->tree_root
)) {
469 spin_lock(&cache
->lock
);
470 if (cache
->cached
!= BTRFS_CACHE_NO
) {
471 spin_unlock(&cache
->lock
);
474 cache
->cached
= BTRFS_CACHE_STARTED
;
475 spin_unlock(&cache
->lock
);
477 ret
= load_free_space_cache(fs_info
, cache
);
479 spin_lock(&cache
->lock
);
481 cache
->cached
= BTRFS_CACHE_FINISHED
;
482 cache
->last_byte_to_unpin
= (u64
)-1;
484 cache
->cached
= BTRFS_CACHE_NO
;
486 spin_unlock(&cache
->lock
);
488 free_excluded_extents(fs_info
->extent_root
, cache
);
496 caching_ctl
= kzalloc(sizeof(*caching_ctl
), GFP_NOFS
);
497 BUG_ON(!caching_ctl
);
499 INIT_LIST_HEAD(&caching_ctl
->list
);
500 mutex_init(&caching_ctl
->mutex
);
501 init_waitqueue_head(&caching_ctl
->wait
);
502 caching_ctl
->block_group
= cache
;
503 caching_ctl
->progress
= cache
->key
.objectid
;
504 /* one for caching kthread, one for caching block group list */
505 atomic_set(&caching_ctl
->count
, 2);
506 caching_ctl
->work
.func
= caching_thread
;
508 spin_lock(&cache
->lock
);
509 if (cache
->cached
!= BTRFS_CACHE_NO
) {
510 spin_unlock(&cache
->lock
);
514 cache
->caching_ctl
= caching_ctl
;
515 cache
->cached
= BTRFS_CACHE_STARTED
;
516 spin_unlock(&cache
->lock
);
518 down_write(&fs_info
->extent_commit_sem
);
519 list_add_tail(&caching_ctl
->list
, &fs_info
->caching_block_groups
);
520 up_write(&fs_info
->extent_commit_sem
);
522 btrfs_get_block_group(cache
);
524 btrfs_queue_worker(&fs_info
->caching_workers
, &caching_ctl
->work
);
530 * return the block group that starts at or after bytenr
532 static struct btrfs_block_group_cache
*
533 btrfs_lookup_first_block_group(struct btrfs_fs_info
*info
, u64 bytenr
)
535 struct btrfs_block_group_cache
*cache
;
537 cache
= block_group_cache_tree_search(info
, bytenr
, 0);
543 * return the block group that contains the given bytenr
545 struct btrfs_block_group_cache
*btrfs_lookup_block_group(
546 struct btrfs_fs_info
*info
,
549 struct btrfs_block_group_cache
*cache
;
551 cache
= block_group_cache_tree_search(info
, bytenr
, 1);
556 static struct btrfs_space_info
*__find_space_info(struct btrfs_fs_info
*info
,
559 struct list_head
*head
= &info
->space_info
;
560 struct btrfs_space_info
*found
;
562 flags
&= BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_SYSTEM
|
563 BTRFS_BLOCK_GROUP_METADATA
;
566 list_for_each_entry_rcu(found
, head
, list
) {
567 if (found
->flags
& flags
) {
577 * after adding space to the filesystem, we need to clear the full flags
578 * on all the space infos.
580 void btrfs_clear_space_info_full(struct btrfs_fs_info
*info
)
582 struct list_head
*head
= &info
->space_info
;
583 struct btrfs_space_info
*found
;
586 list_for_each_entry_rcu(found
, head
, list
)
591 static u64
div_factor(u64 num
, int factor
)
600 static u64
div_factor_fine(u64 num
, int factor
)
609 u64
btrfs_find_block_group(struct btrfs_root
*root
,
610 u64 search_start
, u64 search_hint
, int owner
)
612 struct btrfs_block_group_cache
*cache
;
614 u64 last
= max(search_hint
, search_start
);
621 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
625 spin_lock(&cache
->lock
);
626 last
= cache
->key
.objectid
+ cache
->key
.offset
;
627 used
= btrfs_block_group_used(&cache
->item
);
629 if ((full_search
|| !cache
->ro
) &&
630 block_group_bits(cache
, BTRFS_BLOCK_GROUP_METADATA
)) {
631 if (used
+ cache
->pinned
+ cache
->reserved
<
632 div_factor(cache
->key
.offset
, factor
)) {
633 group_start
= cache
->key
.objectid
;
634 spin_unlock(&cache
->lock
);
635 btrfs_put_block_group(cache
);
639 spin_unlock(&cache
->lock
);
640 btrfs_put_block_group(cache
);
648 if (!full_search
&& factor
< 10) {
658 /* simple helper to search for an existing extent at a given offset */
659 int btrfs_lookup_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
662 struct btrfs_key key
;
663 struct btrfs_path
*path
;
665 path
= btrfs_alloc_path();
669 key
.objectid
= start
;
671 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
672 ret
= btrfs_search_slot(NULL
, root
->fs_info
->extent_root
, &key
, path
,
674 btrfs_free_path(path
);
679 * helper function to lookup reference count and flags of extent.
681 * the head node for delayed ref is used to store the sum of all the
682 * reference count modifications queued up in the rbtree. the head
683 * node may also store the extent flags to set. This way you can check
684 * to see what the reference count and extent flags would be if all of
685 * the delayed refs are not processed.
687 int btrfs_lookup_extent_info(struct btrfs_trans_handle
*trans
,
688 struct btrfs_root
*root
, u64 bytenr
,
689 u64 num_bytes
, u64
*refs
, u64
*flags
)
691 struct btrfs_delayed_ref_head
*head
;
692 struct btrfs_delayed_ref_root
*delayed_refs
;
693 struct btrfs_path
*path
;
694 struct btrfs_extent_item
*ei
;
695 struct extent_buffer
*leaf
;
696 struct btrfs_key key
;
702 path
= btrfs_alloc_path();
706 key
.objectid
= bytenr
;
707 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
708 key
.offset
= num_bytes
;
710 path
->skip_locking
= 1;
711 path
->search_commit_root
= 1;
714 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
,
720 leaf
= path
->nodes
[0];
721 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
722 if (item_size
>= sizeof(*ei
)) {
723 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
724 struct btrfs_extent_item
);
725 num_refs
= btrfs_extent_refs(leaf
, ei
);
726 extent_flags
= btrfs_extent_flags(leaf
, ei
);
728 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
729 struct btrfs_extent_item_v0
*ei0
;
730 BUG_ON(item_size
!= sizeof(*ei0
));
731 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
732 struct btrfs_extent_item_v0
);
733 num_refs
= btrfs_extent_refs_v0(leaf
, ei0
);
734 /* FIXME: this isn't correct for data */
735 extent_flags
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
740 BUG_ON(num_refs
== 0);
750 delayed_refs
= &trans
->transaction
->delayed_refs
;
751 spin_lock(&delayed_refs
->lock
);
752 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
754 if (!mutex_trylock(&head
->mutex
)) {
755 atomic_inc(&head
->node
.refs
);
756 spin_unlock(&delayed_refs
->lock
);
758 btrfs_release_path(path
);
761 * Mutex was contended, block until it's released and try
764 mutex_lock(&head
->mutex
);
765 mutex_unlock(&head
->mutex
);
766 btrfs_put_delayed_ref(&head
->node
);
769 if (head
->extent_op
&& head
->extent_op
->update_flags
)
770 extent_flags
|= head
->extent_op
->flags_to_set
;
772 BUG_ON(num_refs
== 0);
774 num_refs
+= head
->node
.ref_mod
;
775 mutex_unlock(&head
->mutex
);
777 spin_unlock(&delayed_refs
->lock
);
779 WARN_ON(num_refs
== 0);
783 *flags
= extent_flags
;
785 btrfs_free_path(path
);
790 * Back reference rules. Back refs have three main goals:
792 * 1) differentiate between all holders of references to an extent so that
793 * when a reference is dropped we can make sure it was a valid reference
794 * before freeing the extent.
796 * 2) Provide enough information to quickly find the holders of an extent
797 * if we notice a given block is corrupted or bad.
799 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
800 * maintenance. This is actually the same as #2, but with a slightly
801 * different use case.
803 * There are two kinds of back refs. The implicit back refs is optimized
804 * for pointers in non-shared tree blocks. For a given pointer in a block,
805 * back refs of this kind provide information about the block's owner tree
806 * and the pointer's key. These information allow us to find the block by
807 * b-tree searching. The full back refs is for pointers in tree blocks not
808 * referenced by their owner trees. The location of tree block is recorded
809 * in the back refs. Actually the full back refs is generic, and can be
810 * used in all cases the implicit back refs is used. The major shortcoming
811 * of the full back refs is its overhead. Every time a tree block gets
812 * COWed, we have to update back refs entry for all pointers in it.
814 * For a newly allocated tree block, we use implicit back refs for
815 * pointers in it. This means most tree related operations only involve
816 * implicit back refs. For a tree block created in old transaction, the
817 * only way to drop a reference to it is COW it. So we can detect the
818 * event that tree block loses its owner tree's reference and do the
819 * back refs conversion.
821 * When a tree block is COW'd through a tree, there are four cases:
823 * The reference count of the block is one and the tree is the block's
824 * owner tree. Nothing to do in this case.
826 * The reference count of the block is one and the tree is not the
827 * block's owner tree. In this case, full back refs is used for pointers
828 * in the block. Remove these full back refs, add implicit back refs for
829 * every pointers in the new block.
831 * The reference count of the block is greater than one and the tree is
832 * the block's owner tree. In this case, implicit back refs is used for
833 * pointers in the block. Add full back refs for every pointers in the
834 * block, increase lower level extents' reference counts. The original
835 * implicit back refs are entailed to the new block.
837 * The reference count of the block is greater than one and the tree is
838 * not the block's owner tree. Add implicit back refs for every pointer in
839 * the new block, increase lower level extents' reference count.
841 * Back Reference Key composing:
843 * The key objectid corresponds to the first byte in the extent,
844 * The key type is used to differentiate between types of back refs.
845 * There are different meanings of the key offset for different types
848 * File extents can be referenced by:
850 * - multiple snapshots, subvolumes, or different generations in one subvol
851 * - different files inside a single subvolume
852 * - different offsets inside a file (bookend extents in file.c)
854 * The extent ref structure for the implicit back refs has fields for:
856 * - Objectid of the subvolume root
857 * - objectid of the file holding the reference
858 * - original offset in the file
859 * - how many bookend extents
861 * The key offset for the implicit back refs is hash of the first
864 * The extent ref structure for the full back refs has field for:
866 * - number of pointers in the tree leaf
868 * The key offset for the implicit back refs is the first byte of
871 * When a file extent is allocated, The implicit back refs is used.
872 * the fields are filled in:
874 * (root_key.objectid, inode objectid, offset in file, 1)
876 * When a file extent is removed file truncation, we find the
877 * corresponding implicit back refs and check the following fields:
879 * (btrfs_header_owner(leaf), inode objectid, offset in file)
881 * Btree extents can be referenced by:
883 * - Different subvolumes
885 * Both the implicit back refs and the full back refs for tree blocks
886 * only consist of key. The key offset for the implicit back refs is
887 * objectid of block's owner tree. The key offset for the full back refs
888 * is the first byte of parent block.
890 * When implicit back refs is used, information about the lowest key and
891 * level of the tree block are required. These information are stored in
892 * tree block info structure.
895 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
896 static int convert_extent_item_v0(struct btrfs_trans_handle
*trans
,
897 struct btrfs_root
*root
,
898 struct btrfs_path
*path
,
899 u64 owner
, u32 extra_size
)
901 struct btrfs_extent_item
*item
;
902 struct btrfs_extent_item_v0
*ei0
;
903 struct btrfs_extent_ref_v0
*ref0
;
904 struct btrfs_tree_block_info
*bi
;
905 struct extent_buffer
*leaf
;
906 struct btrfs_key key
;
907 struct btrfs_key found_key
;
908 u32 new_size
= sizeof(*item
);
912 leaf
= path
->nodes
[0];
913 BUG_ON(btrfs_item_size_nr(leaf
, path
->slots
[0]) != sizeof(*ei0
));
915 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
916 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
917 struct btrfs_extent_item_v0
);
918 refs
= btrfs_extent_refs_v0(leaf
, ei0
);
920 if (owner
== (u64
)-1) {
922 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
923 ret
= btrfs_next_leaf(root
, path
);
927 leaf
= path
->nodes
[0];
929 btrfs_item_key_to_cpu(leaf
, &found_key
,
931 BUG_ON(key
.objectid
!= found_key
.objectid
);
932 if (found_key
.type
!= BTRFS_EXTENT_REF_V0_KEY
) {
936 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
937 struct btrfs_extent_ref_v0
);
938 owner
= btrfs_ref_objectid_v0(leaf
, ref0
);
942 btrfs_release_path(path
);
944 if (owner
< BTRFS_FIRST_FREE_OBJECTID
)
945 new_size
+= sizeof(*bi
);
947 new_size
-= sizeof(*ei0
);
948 ret
= btrfs_search_slot(trans
, root
, &key
, path
,
949 new_size
+ extra_size
, 1);
954 ret
= btrfs_extend_item(trans
, root
, path
, new_size
);
956 leaf
= path
->nodes
[0];
957 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
958 btrfs_set_extent_refs(leaf
, item
, refs
);
959 /* FIXME: get real generation */
960 btrfs_set_extent_generation(leaf
, item
, 0);
961 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
962 btrfs_set_extent_flags(leaf
, item
,
963 BTRFS_EXTENT_FLAG_TREE_BLOCK
|
964 BTRFS_BLOCK_FLAG_FULL_BACKREF
);
965 bi
= (struct btrfs_tree_block_info
*)(item
+ 1);
966 /* FIXME: get first key of the block */
967 memset_extent_buffer(leaf
, 0, (unsigned long)bi
, sizeof(*bi
));
968 btrfs_set_tree_block_level(leaf
, bi
, (int)owner
);
970 btrfs_set_extent_flags(leaf
, item
, BTRFS_EXTENT_FLAG_DATA
);
972 btrfs_mark_buffer_dirty(leaf
);
977 static u64
hash_extent_data_ref(u64 root_objectid
, u64 owner
, u64 offset
)
979 u32 high_crc
= ~(u32
)0;
980 u32 low_crc
= ~(u32
)0;
983 lenum
= cpu_to_le64(root_objectid
);
984 high_crc
= crc32c(high_crc
, &lenum
, sizeof(lenum
));
985 lenum
= cpu_to_le64(owner
);
986 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
987 lenum
= cpu_to_le64(offset
);
988 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
990 return ((u64
)high_crc
<< 31) ^ (u64
)low_crc
;
993 static u64
hash_extent_data_ref_item(struct extent_buffer
*leaf
,
994 struct btrfs_extent_data_ref
*ref
)
996 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf
, ref
),
997 btrfs_extent_data_ref_objectid(leaf
, ref
),
998 btrfs_extent_data_ref_offset(leaf
, ref
));
1001 static int match_extent_data_ref(struct extent_buffer
*leaf
,
1002 struct btrfs_extent_data_ref
*ref
,
1003 u64 root_objectid
, u64 owner
, u64 offset
)
1005 if (btrfs_extent_data_ref_root(leaf
, ref
) != root_objectid
||
1006 btrfs_extent_data_ref_objectid(leaf
, ref
) != owner
||
1007 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
1012 static noinline
int lookup_extent_data_ref(struct btrfs_trans_handle
*trans
,
1013 struct btrfs_root
*root
,
1014 struct btrfs_path
*path
,
1015 u64 bytenr
, u64 parent
,
1017 u64 owner
, u64 offset
)
1019 struct btrfs_key key
;
1020 struct btrfs_extent_data_ref
*ref
;
1021 struct extent_buffer
*leaf
;
1027 key
.objectid
= bytenr
;
1029 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1030 key
.offset
= parent
;
1032 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1033 key
.offset
= hash_extent_data_ref(root_objectid
,
1038 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1047 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1048 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1049 btrfs_release_path(path
);
1050 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1061 leaf
= path
->nodes
[0];
1062 nritems
= btrfs_header_nritems(leaf
);
1064 if (path
->slots
[0] >= nritems
) {
1065 ret
= btrfs_next_leaf(root
, path
);
1071 leaf
= path
->nodes
[0];
1072 nritems
= btrfs_header_nritems(leaf
);
1076 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1077 if (key
.objectid
!= bytenr
||
1078 key
.type
!= BTRFS_EXTENT_DATA_REF_KEY
)
1081 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1082 struct btrfs_extent_data_ref
);
1084 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1087 btrfs_release_path(path
);
1099 static noinline
int insert_extent_data_ref(struct btrfs_trans_handle
*trans
,
1100 struct btrfs_root
*root
,
1101 struct btrfs_path
*path
,
1102 u64 bytenr
, u64 parent
,
1103 u64 root_objectid
, u64 owner
,
1104 u64 offset
, int refs_to_add
)
1106 struct btrfs_key key
;
1107 struct extent_buffer
*leaf
;
1112 key
.objectid
= bytenr
;
1114 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1115 key
.offset
= parent
;
1116 size
= sizeof(struct btrfs_shared_data_ref
);
1118 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1119 key
.offset
= hash_extent_data_ref(root_objectid
,
1121 size
= sizeof(struct btrfs_extent_data_ref
);
1124 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, size
);
1125 if (ret
&& ret
!= -EEXIST
)
1128 leaf
= path
->nodes
[0];
1130 struct btrfs_shared_data_ref
*ref
;
1131 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1132 struct btrfs_shared_data_ref
);
1134 btrfs_set_shared_data_ref_count(leaf
, ref
, refs_to_add
);
1136 num_refs
= btrfs_shared_data_ref_count(leaf
, ref
);
1137 num_refs
+= refs_to_add
;
1138 btrfs_set_shared_data_ref_count(leaf
, ref
, num_refs
);
1141 struct btrfs_extent_data_ref
*ref
;
1142 while (ret
== -EEXIST
) {
1143 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1144 struct btrfs_extent_data_ref
);
1145 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1148 btrfs_release_path(path
);
1150 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1152 if (ret
&& ret
!= -EEXIST
)
1155 leaf
= path
->nodes
[0];
1157 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1158 struct btrfs_extent_data_ref
);
1160 btrfs_set_extent_data_ref_root(leaf
, ref
,
1162 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
1163 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
1164 btrfs_set_extent_data_ref_count(leaf
, ref
, refs_to_add
);
1166 num_refs
= btrfs_extent_data_ref_count(leaf
, ref
);
1167 num_refs
+= refs_to_add
;
1168 btrfs_set_extent_data_ref_count(leaf
, ref
, num_refs
);
1171 btrfs_mark_buffer_dirty(leaf
);
1174 btrfs_release_path(path
);
1178 static noinline
int remove_extent_data_ref(struct btrfs_trans_handle
*trans
,
1179 struct btrfs_root
*root
,
1180 struct btrfs_path
*path
,
1183 struct btrfs_key key
;
1184 struct btrfs_extent_data_ref
*ref1
= NULL
;
1185 struct btrfs_shared_data_ref
*ref2
= NULL
;
1186 struct extent_buffer
*leaf
;
1190 leaf
= path
->nodes
[0];
1191 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1193 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1194 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1195 struct btrfs_extent_data_ref
);
1196 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1197 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1198 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1199 struct btrfs_shared_data_ref
);
1200 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1201 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1202 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1203 struct btrfs_extent_ref_v0
*ref0
;
1204 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1205 struct btrfs_extent_ref_v0
);
1206 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1212 BUG_ON(num_refs
< refs_to_drop
);
1213 num_refs
-= refs_to_drop
;
1215 if (num_refs
== 0) {
1216 ret
= btrfs_del_item(trans
, root
, path
);
1218 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
)
1219 btrfs_set_extent_data_ref_count(leaf
, ref1
, num_refs
);
1220 else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
)
1221 btrfs_set_shared_data_ref_count(leaf
, ref2
, num_refs
);
1222 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1224 struct btrfs_extent_ref_v0
*ref0
;
1225 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1226 struct btrfs_extent_ref_v0
);
1227 btrfs_set_ref_count_v0(leaf
, ref0
, num_refs
);
1230 btrfs_mark_buffer_dirty(leaf
);
1235 static noinline u32
extent_data_ref_count(struct btrfs_root
*root
,
1236 struct btrfs_path
*path
,
1237 struct btrfs_extent_inline_ref
*iref
)
1239 struct btrfs_key key
;
1240 struct extent_buffer
*leaf
;
1241 struct btrfs_extent_data_ref
*ref1
;
1242 struct btrfs_shared_data_ref
*ref2
;
1245 leaf
= path
->nodes
[0];
1246 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1248 if (btrfs_extent_inline_ref_type(leaf
, iref
) ==
1249 BTRFS_EXTENT_DATA_REF_KEY
) {
1250 ref1
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1251 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1253 ref2
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1254 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1256 } else if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1257 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1258 struct btrfs_extent_data_ref
);
1259 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1260 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1261 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1262 struct btrfs_shared_data_ref
);
1263 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1264 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1265 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1266 struct btrfs_extent_ref_v0
*ref0
;
1267 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1268 struct btrfs_extent_ref_v0
);
1269 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1277 static noinline
int lookup_tree_block_ref(struct btrfs_trans_handle
*trans
,
1278 struct btrfs_root
*root
,
1279 struct btrfs_path
*path
,
1280 u64 bytenr
, u64 parent
,
1283 struct btrfs_key key
;
1286 key
.objectid
= bytenr
;
1288 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1289 key
.offset
= parent
;
1291 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1292 key
.offset
= root_objectid
;
1295 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1298 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1299 if (ret
== -ENOENT
&& parent
) {
1300 btrfs_release_path(path
);
1301 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1302 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1310 static noinline
int insert_tree_block_ref(struct btrfs_trans_handle
*trans
,
1311 struct btrfs_root
*root
,
1312 struct btrfs_path
*path
,
1313 u64 bytenr
, u64 parent
,
1316 struct btrfs_key key
;
1319 key
.objectid
= bytenr
;
1321 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1322 key
.offset
= parent
;
1324 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1325 key
.offset
= root_objectid
;
1328 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
1329 btrfs_release_path(path
);
1333 static inline int extent_ref_type(u64 parent
, u64 owner
)
1336 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1338 type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1340 type
= BTRFS_TREE_BLOCK_REF_KEY
;
1343 type
= BTRFS_SHARED_DATA_REF_KEY
;
1345 type
= BTRFS_EXTENT_DATA_REF_KEY
;
1350 static int find_next_key(struct btrfs_path
*path
, int level
,
1351 struct btrfs_key
*key
)
1354 for (; level
< BTRFS_MAX_LEVEL
; level
++) {
1355 if (!path
->nodes
[level
])
1357 if (path
->slots
[level
] + 1 >=
1358 btrfs_header_nritems(path
->nodes
[level
]))
1361 btrfs_item_key_to_cpu(path
->nodes
[level
], key
,
1362 path
->slots
[level
] + 1);
1364 btrfs_node_key_to_cpu(path
->nodes
[level
], key
,
1365 path
->slots
[level
] + 1);
1372 * look for inline back ref. if back ref is found, *ref_ret is set
1373 * to the address of inline back ref, and 0 is returned.
1375 * if back ref isn't found, *ref_ret is set to the address where it
1376 * should be inserted, and -ENOENT is returned.
1378 * if insert is true and there are too many inline back refs, the path
1379 * points to the extent item, and -EAGAIN is returned.
1381 * NOTE: inline back refs are ordered in the same way that back ref
1382 * items in the tree are ordered.
1384 static noinline_for_stack
1385 int lookup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1386 struct btrfs_root
*root
,
1387 struct btrfs_path
*path
,
1388 struct btrfs_extent_inline_ref
**ref_ret
,
1389 u64 bytenr
, u64 num_bytes
,
1390 u64 parent
, u64 root_objectid
,
1391 u64 owner
, u64 offset
, int insert
)
1393 struct btrfs_key key
;
1394 struct extent_buffer
*leaf
;
1395 struct btrfs_extent_item
*ei
;
1396 struct btrfs_extent_inline_ref
*iref
;
1407 key
.objectid
= bytenr
;
1408 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1409 key
.offset
= num_bytes
;
1411 want
= extent_ref_type(parent
, owner
);
1413 extra_size
= btrfs_extent_inline_ref_size(want
);
1414 path
->keep_locks
= 1;
1417 ret
= btrfs_search_slot(trans
, root
, &key
, path
, extra_size
, 1);
1424 leaf
= path
->nodes
[0];
1425 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1426 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1427 if (item_size
< sizeof(*ei
)) {
1432 ret
= convert_extent_item_v0(trans
, root
, path
, owner
,
1438 leaf
= path
->nodes
[0];
1439 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1442 BUG_ON(item_size
< sizeof(*ei
));
1444 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1445 flags
= btrfs_extent_flags(leaf
, ei
);
1447 ptr
= (unsigned long)(ei
+ 1);
1448 end
= (unsigned long)ei
+ item_size
;
1450 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
) {
1451 ptr
+= sizeof(struct btrfs_tree_block_info
);
1454 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_DATA
));
1463 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1464 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1468 ptr
+= btrfs_extent_inline_ref_size(type
);
1472 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1473 struct btrfs_extent_data_ref
*dref
;
1474 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1475 if (match_extent_data_ref(leaf
, dref
, root_objectid
,
1480 if (hash_extent_data_ref_item(leaf
, dref
) <
1481 hash_extent_data_ref(root_objectid
, owner
, offset
))
1485 ref_offset
= btrfs_extent_inline_ref_offset(leaf
, iref
);
1487 if (parent
== ref_offset
) {
1491 if (ref_offset
< parent
)
1494 if (root_objectid
== ref_offset
) {
1498 if (ref_offset
< root_objectid
)
1502 ptr
+= btrfs_extent_inline_ref_size(type
);
1504 if (err
== -ENOENT
&& insert
) {
1505 if (item_size
+ extra_size
>=
1506 BTRFS_MAX_EXTENT_ITEM_SIZE(root
)) {
1511 * To add new inline back ref, we have to make sure
1512 * there is no corresponding back ref item.
1513 * For simplicity, we just do not add new inline back
1514 * ref if there is any kind of item for this block
1516 if (find_next_key(path
, 0, &key
) == 0 &&
1517 key
.objectid
== bytenr
&&
1518 key
.type
< BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1523 *ref_ret
= (struct btrfs_extent_inline_ref
*)ptr
;
1526 path
->keep_locks
= 0;
1527 btrfs_unlock_up_safe(path
, 1);
1533 * helper to add new inline back ref
1535 static noinline_for_stack
1536 int setup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1537 struct btrfs_root
*root
,
1538 struct btrfs_path
*path
,
1539 struct btrfs_extent_inline_ref
*iref
,
1540 u64 parent
, u64 root_objectid
,
1541 u64 owner
, u64 offset
, int refs_to_add
,
1542 struct btrfs_delayed_extent_op
*extent_op
)
1544 struct extent_buffer
*leaf
;
1545 struct btrfs_extent_item
*ei
;
1548 unsigned long item_offset
;
1554 leaf
= path
->nodes
[0];
1555 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1556 item_offset
= (unsigned long)iref
- (unsigned long)ei
;
1558 type
= extent_ref_type(parent
, owner
);
1559 size
= btrfs_extent_inline_ref_size(type
);
1561 ret
= btrfs_extend_item(trans
, root
, path
, size
);
1563 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1564 refs
= btrfs_extent_refs(leaf
, ei
);
1565 refs
+= refs_to_add
;
1566 btrfs_set_extent_refs(leaf
, ei
, refs
);
1568 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1570 ptr
= (unsigned long)ei
+ item_offset
;
1571 end
= (unsigned long)ei
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
1572 if (ptr
< end
- size
)
1573 memmove_extent_buffer(leaf
, ptr
+ size
, ptr
,
1576 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1577 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
1578 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1579 struct btrfs_extent_data_ref
*dref
;
1580 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1581 btrfs_set_extent_data_ref_root(leaf
, dref
, root_objectid
);
1582 btrfs_set_extent_data_ref_objectid(leaf
, dref
, owner
);
1583 btrfs_set_extent_data_ref_offset(leaf
, dref
, offset
);
1584 btrfs_set_extent_data_ref_count(leaf
, dref
, refs_to_add
);
1585 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1586 struct btrfs_shared_data_ref
*sref
;
1587 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1588 btrfs_set_shared_data_ref_count(leaf
, sref
, refs_to_add
);
1589 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1590 } else if (type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
1591 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1593 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
1595 btrfs_mark_buffer_dirty(leaf
);
1599 static int lookup_extent_backref(struct btrfs_trans_handle
*trans
,
1600 struct btrfs_root
*root
,
1601 struct btrfs_path
*path
,
1602 struct btrfs_extent_inline_ref
**ref_ret
,
1603 u64 bytenr
, u64 num_bytes
, u64 parent
,
1604 u64 root_objectid
, u64 owner
, u64 offset
)
1608 ret
= lookup_inline_extent_backref(trans
, root
, path
, ref_ret
,
1609 bytenr
, num_bytes
, parent
,
1610 root_objectid
, owner
, offset
, 0);
1614 btrfs_release_path(path
);
1617 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1618 ret
= lookup_tree_block_ref(trans
, root
, path
, bytenr
, parent
,
1621 ret
= lookup_extent_data_ref(trans
, root
, path
, bytenr
, parent
,
1622 root_objectid
, owner
, offset
);
1628 * helper to update/remove inline back ref
1630 static noinline_for_stack
1631 int update_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1632 struct btrfs_root
*root
,
1633 struct btrfs_path
*path
,
1634 struct btrfs_extent_inline_ref
*iref
,
1636 struct btrfs_delayed_extent_op
*extent_op
)
1638 struct extent_buffer
*leaf
;
1639 struct btrfs_extent_item
*ei
;
1640 struct btrfs_extent_data_ref
*dref
= NULL
;
1641 struct btrfs_shared_data_ref
*sref
= NULL
;
1650 leaf
= path
->nodes
[0];
1651 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1652 refs
= btrfs_extent_refs(leaf
, ei
);
1653 WARN_ON(refs_to_mod
< 0 && refs
+ refs_to_mod
<= 0);
1654 refs
+= refs_to_mod
;
1655 btrfs_set_extent_refs(leaf
, ei
, refs
);
1657 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1659 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1661 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1662 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1663 refs
= btrfs_extent_data_ref_count(leaf
, dref
);
1664 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1665 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1666 refs
= btrfs_shared_data_ref_count(leaf
, sref
);
1669 BUG_ON(refs_to_mod
!= -1);
1672 BUG_ON(refs_to_mod
< 0 && refs
< -refs_to_mod
);
1673 refs
+= refs_to_mod
;
1676 if (type
== BTRFS_EXTENT_DATA_REF_KEY
)
1677 btrfs_set_extent_data_ref_count(leaf
, dref
, refs
);
1679 btrfs_set_shared_data_ref_count(leaf
, sref
, refs
);
1681 size
= btrfs_extent_inline_ref_size(type
);
1682 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1683 ptr
= (unsigned long)iref
;
1684 end
= (unsigned long)ei
+ item_size
;
1685 if (ptr
+ size
< end
)
1686 memmove_extent_buffer(leaf
, ptr
, ptr
+ size
,
1689 ret
= btrfs_truncate_item(trans
, root
, path
, item_size
, 1);
1691 btrfs_mark_buffer_dirty(leaf
);
1695 static noinline_for_stack
1696 int insert_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1697 struct btrfs_root
*root
,
1698 struct btrfs_path
*path
,
1699 u64 bytenr
, u64 num_bytes
, u64 parent
,
1700 u64 root_objectid
, u64 owner
,
1701 u64 offset
, int refs_to_add
,
1702 struct btrfs_delayed_extent_op
*extent_op
)
1704 struct btrfs_extent_inline_ref
*iref
;
1707 ret
= lookup_inline_extent_backref(trans
, root
, path
, &iref
,
1708 bytenr
, num_bytes
, parent
,
1709 root_objectid
, owner
, offset
, 1);
1711 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
);
1712 ret
= update_inline_extent_backref(trans
, root
, path
, iref
,
1713 refs_to_add
, extent_op
);
1714 } else if (ret
== -ENOENT
) {
1715 ret
= setup_inline_extent_backref(trans
, root
, path
, iref
,
1716 parent
, root_objectid
,
1717 owner
, offset
, refs_to_add
,
1723 static int insert_extent_backref(struct btrfs_trans_handle
*trans
,
1724 struct btrfs_root
*root
,
1725 struct btrfs_path
*path
,
1726 u64 bytenr
, u64 parent
, u64 root_objectid
,
1727 u64 owner
, u64 offset
, int refs_to_add
)
1730 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1731 BUG_ON(refs_to_add
!= 1);
1732 ret
= insert_tree_block_ref(trans
, root
, path
, bytenr
,
1733 parent
, root_objectid
);
1735 ret
= insert_extent_data_ref(trans
, root
, path
, bytenr
,
1736 parent
, root_objectid
,
1737 owner
, offset
, refs_to_add
);
1742 static int remove_extent_backref(struct btrfs_trans_handle
*trans
,
1743 struct btrfs_root
*root
,
1744 struct btrfs_path
*path
,
1745 struct btrfs_extent_inline_ref
*iref
,
1746 int refs_to_drop
, int is_data
)
1750 BUG_ON(!is_data
&& refs_to_drop
!= 1);
1752 ret
= update_inline_extent_backref(trans
, root
, path
, iref
,
1753 -refs_to_drop
, NULL
);
1754 } else if (is_data
) {
1755 ret
= remove_extent_data_ref(trans
, root
, path
, refs_to_drop
);
1757 ret
= btrfs_del_item(trans
, root
, path
);
1762 static int btrfs_issue_discard(struct block_device
*bdev
,
1765 return blkdev_issue_discard(bdev
, start
>> 9, len
>> 9, GFP_NOFS
, 0);
1768 static int btrfs_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
1769 u64 num_bytes
, u64
*actual_bytes
)
1772 u64 discarded_bytes
= 0;
1773 struct btrfs_multi_bio
*multi
= NULL
;
1776 /* Tell the block device(s) that the sectors can be discarded */
1777 ret
= btrfs_map_block(&root
->fs_info
->mapping_tree
, REQ_DISCARD
,
1778 bytenr
, &num_bytes
, &multi
, 0);
1780 struct btrfs_bio_stripe
*stripe
= multi
->stripes
;
1784 for (i
= 0; i
< multi
->num_stripes
; i
++, stripe
++) {
1785 ret
= btrfs_issue_discard(stripe
->dev
->bdev
,
1789 discarded_bytes
+= stripe
->length
;
1790 else if (ret
!= -EOPNOTSUPP
)
1795 if (discarded_bytes
&& ret
== -EOPNOTSUPP
)
1799 *actual_bytes
= discarded_bytes
;
1805 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1806 struct btrfs_root
*root
,
1807 u64 bytenr
, u64 num_bytes
, u64 parent
,
1808 u64 root_objectid
, u64 owner
, u64 offset
)
1811 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
&&
1812 root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
1814 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1815 ret
= btrfs_add_delayed_tree_ref(trans
, bytenr
, num_bytes
,
1816 parent
, root_objectid
, (int)owner
,
1817 BTRFS_ADD_DELAYED_REF
, NULL
);
1819 ret
= btrfs_add_delayed_data_ref(trans
, bytenr
, num_bytes
,
1820 parent
, root_objectid
, owner
, offset
,
1821 BTRFS_ADD_DELAYED_REF
, NULL
);
1826 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1827 struct btrfs_root
*root
,
1828 u64 bytenr
, u64 num_bytes
,
1829 u64 parent
, u64 root_objectid
,
1830 u64 owner
, u64 offset
, int refs_to_add
,
1831 struct btrfs_delayed_extent_op
*extent_op
)
1833 struct btrfs_path
*path
;
1834 struct extent_buffer
*leaf
;
1835 struct btrfs_extent_item
*item
;
1840 path
= btrfs_alloc_path();
1845 path
->leave_spinning
= 1;
1846 /* this will setup the path even if it fails to insert the back ref */
1847 ret
= insert_inline_extent_backref(trans
, root
->fs_info
->extent_root
,
1848 path
, bytenr
, num_bytes
, parent
,
1849 root_objectid
, owner
, offset
,
1850 refs_to_add
, extent_op
);
1854 if (ret
!= -EAGAIN
) {
1859 leaf
= path
->nodes
[0];
1860 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1861 refs
= btrfs_extent_refs(leaf
, item
);
1862 btrfs_set_extent_refs(leaf
, item
, refs
+ refs_to_add
);
1864 __run_delayed_extent_op(extent_op
, leaf
, item
);
1866 btrfs_mark_buffer_dirty(leaf
);
1867 btrfs_release_path(path
);
1870 path
->leave_spinning
= 1;
1872 /* now insert the actual backref */
1873 ret
= insert_extent_backref(trans
, root
->fs_info
->extent_root
,
1874 path
, bytenr
, parent
, root_objectid
,
1875 owner
, offset
, refs_to_add
);
1878 btrfs_free_path(path
);
1882 static int run_delayed_data_ref(struct btrfs_trans_handle
*trans
,
1883 struct btrfs_root
*root
,
1884 struct btrfs_delayed_ref_node
*node
,
1885 struct btrfs_delayed_extent_op
*extent_op
,
1886 int insert_reserved
)
1889 struct btrfs_delayed_data_ref
*ref
;
1890 struct btrfs_key ins
;
1895 ins
.objectid
= node
->bytenr
;
1896 ins
.offset
= node
->num_bytes
;
1897 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1899 ref
= btrfs_delayed_node_to_data_ref(node
);
1900 if (node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
1901 parent
= ref
->parent
;
1903 ref_root
= ref
->root
;
1905 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
1907 BUG_ON(extent_op
->update_key
);
1908 flags
|= extent_op
->flags_to_set
;
1910 ret
= alloc_reserved_file_extent(trans
, root
,
1911 parent
, ref_root
, flags
,
1912 ref
->objectid
, ref
->offset
,
1913 &ins
, node
->ref_mod
);
1914 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
1915 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
1916 node
->num_bytes
, parent
,
1917 ref_root
, ref
->objectid
,
1918 ref
->offset
, node
->ref_mod
,
1920 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
1921 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
1922 node
->num_bytes
, parent
,
1923 ref_root
, ref
->objectid
,
1924 ref
->offset
, node
->ref_mod
,
1932 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
1933 struct extent_buffer
*leaf
,
1934 struct btrfs_extent_item
*ei
)
1936 u64 flags
= btrfs_extent_flags(leaf
, ei
);
1937 if (extent_op
->update_flags
) {
1938 flags
|= extent_op
->flags_to_set
;
1939 btrfs_set_extent_flags(leaf
, ei
, flags
);
1942 if (extent_op
->update_key
) {
1943 struct btrfs_tree_block_info
*bi
;
1944 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
));
1945 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
1946 btrfs_set_tree_block_key(leaf
, bi
, &extent_op
->key
);
1950 static int run_delayed_extent_op(struct btrfs_trans_handle
*trans
,
1951 struct btrfs_root
*root
,
1952 struct btrfs_delayed_ref_node
*node
,
1953 struct btrfs_delayed_extent_op
*extent_op
)
1955 struct btrfs_key key
;
1956 struct btrfs_path
*path
;
1957 struct btrfs_extent_item
*ei
;
1958 struct extent_buffer
*leaf
;
1963 path
= btrfs_alloc_path();
1967 key
.objectid
= node
->bytenr
;
1968 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1969 key
.offset
= node
->num_bytes
;
1972 path
->leave_spinning
= 1;
1973 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
,
1984 leaf
= path
->nodes
[0];
1985 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1986 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1987 if (item_size
< sizeof(*ei
)) {
1988 ret
= convert_extent_item_v0(trans
, root
->fs_info
->extent_root
,
1994 leaf
= path
->nodes
[0];
1995 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1998 BUG_ON(item_size
< sizeof(*ei
));
1999 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2000 __run_delayed_extent_op(extent_op
, leaf
, ei
);
2002 btrfs_mark_buffer_dirty(leaf
);
2004 btrfs_free_path(path
);
2008 static int run_delayed_tree_ref(struct btrfs_trans_handle
*trans
,
2009 struct btrfs_root
*root
,
2010 struct btrfs_delayed_ref_node
*node
,
2011 struct btrfs_delayed_extent_op
*extent_op
,
2012 int insert_reserved
)
2015 struct btrfs_delayed_tree_ref
*ref
;
2016 struct btrfs_key ins
;
2020 ins
.objectid
= node
->bytenr
;
2021 ins
.offset
= node
->num_bytes
;
2022 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2024 ref
= btrfs_delayed_node_to_tree_ref(node
);
2025 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2026 parent
= ref
->parent
;
2028 ref_root
= ref
->root
;
2030 BUG_ON(node
->ref_mod
!= 1);
2031 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2032 BUG_ON(!extent_op
|| !extent_op
->update_flags
||
2033 !extent_op
->update_key
);
2034 ret
= alloc_reserved_tree_block(trans
, root
,
2036 extent_op
->flags_to_set
,
2039 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2040 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
2041 node
->num_bytes
, parent
, ref_root
,
2042 ref
->level
, 0, 1, extent_op
);
2043 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2044 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
2045 node
->num_bytes
, parent
, ref_root
,
2046 ref
->level
, 0, 1, extent_op
);
2053 /* helper function to actually process a single delayed ref entry */
2054 static int run_one_delayed_ref(struct btrfs_trans_handle
*trans
,
2055 struct btrfs_root
*root
,
2056 struct btrfs_delayed_ref_node
*node
,
2057 struct btrfs_delayed_extent_op
*extent_op
,
2058 int insert_reserved
)
2061 if (btrfs_delayed_ref_is_head(node
)) {
2062 struct btrfs_delayed_ref_head
*head
;
2064 * we've hit the end of the chain and we were supposed
2065 * to insert this extent into the tree. But, it got
2066 * deleted before we ever needed to insert it, so all
2067 * we have to do is clean up the accounting
2070 head
= btrfs_delayed_node_to_head(node
);
2071 if (insert_reserved
) {
2072 btrfs_pin_extent(root
, node
->bytenr
,
2073 node
->num_bytes
, 1);
2074 if (head
->is_data
) {
2075 ret
= btrfs_del_csums(trans
, root
,
2081 mutex_unlock(&head
->mutex
);
2085 if (node
->type
== BTRFS_TREE_BLOCK_REF_KEY
||
2086 node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2087 ret
= run_delayed_tree_ref(trans
, root
, node
, extent_op
,
2089 else if (node
->type
== BTRFS_EXTENT_DATA_REF_KEY
||
2090 node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2091 ret
= run_delayed_data_ref(trans
, root
, node
, extent_op
,
2098 static noinline
struct btrfs_delayed_ref_node
*
2099 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
2101 struct rb_node
*node
;
2102 struct btrfs_delayed_ref_node
*ref
;
2103 int action
= BTRFS_ADD_DELAYED_REF
;
2106 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2107 * this prevents ref count from going down to zero when
2108 * there still are pending delayed ref.
2110 node
= rb_prev(&head
->node
.rb_node
);
2114 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
2116 if (ref
->bytenr
!= head
->node
.bytenr
)
2118 if (ref
->action
== action
)
2120 node
= rb_prev(node
);
2122 if (action
== BTRFS_ADD_DELAYED_REF
) {
2123 action
= BTRFS_DROP_DELAYED_REF
;
2129 static noinline
int run_clustered_refs(struct btrfs_trans_handle
*trans
,
2130 struct btrfs_root
*root
,
2131 struct list_head
*cluster
)
2133 struct btrfs_delayed_ref_root
*delayed_refs
;
2134 struct btrfs_delayed_ref_node
*ref
;
2135 struct btrfs_delayed_ref_head
*locked_ref
= NULL
;
2136 struct btrfs_delayed_extent_op
*extent_op
;
2139 int must_insert_reserved
= 0;
2141 delayed_refs
= &trans
->transaction
->delayed_refs
;
2144 /* pick a new head ref from the cluster list */
2145 if (list_empty(cluster
))
2148 locked_ref
= list_entry(cluster
->next
,
2149 struct btrfs_delayed_ref_head
, cluster
);
2151 /* grab the lock that says we are going to process
2152 * all the refs for this head */
2153 ret
= btrfs_delayed_ref_lock(trans
, locked_ref
);
2156 * we may have dropped the spin lock to get the head
2157 * mutex lock, and that might have given someone else
2158 * time to free the head. If that's true, it has been
2159 * removed from our list and we can move on.
2161 if (ret
== -EAGAIN
) {
2169 * record the must insert reserved flag before we
2170 * drop the spin lock.
2172 must_insert_reserved
= locked_ref
->must_insert_reserved
;
2173 locked_ref
->must_insert_reserved
= 0;
2175 extent_op
= locked_ref
->extent_op
;
2176 locked_ref
->extent_op
= NULL
;
2179 * locked_ref is the head node, so we have to go one
2180 * node back for any delayed ref updates
2182 ref
= select_delayed_ref(locked_ref
);
2184 /* All delayed refs have been processed, Go ahead
2185 * and send the head node to run_one_delayed_ref,
2186 * so that any accounting fixes can happen
2188 ref
= &locked_ref
->node
;
2190 if (extent_op
&& must_insert_reserved
) {
2196 spin_unlock(&delayed_refs
->lock
);
2198 ret
= run_delayed_extent_op(trans
, root
,
2204 spin_lock(&delayed_refs
->lock
);
2208 list_del_init(&locked_ref
->cluster
);
2213 rb_erase(&ref
->rb_node
, &delayed_refs
->root
);
2214 delayed_refs
->num_entries
--;
2216 spin_unlock(&delayed_refs
->lock
);
2218 ret
= run_one_delayed_ref(trans
, root
, ref
, extent_op
,
2219 must_insert_reserved
);
2222 btrfs_put_delayed_ref(ref
);
2227 spin_lock(&delayed_refs
->lock
);
2233 * this starts processing the delayed reference count updates and
2234 * extent insertions we have queued up so far. count can be
2235 * 0, which means to process everything in the tree at the start
2236 * of the run (but not newly added entries), or it can be some target
2237 * number you'd like to process.
2239 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2240 struct btrfs_root
*root
, unsigned long count
)
2242 struct rb_node
*node
;
2243 struct btrfs_delayed_ref_root
*delayed_refs
;
2244 struct btrfs_delayed_ref_node
*ref
;
2245 struct list_head cluster
;
2247 int run_all
= count
== (unsigned long)-1;
2250 if (root
== root
->fs_info
->extent_root
)
2251 root
= root
->fs_info
->tree_root
;
2253 delayed_refs
= &trans
->transaction
->delayed_refs
;
2254 INIT_LIST_HEAD(&cluster
);
2256 spin_lock(&delayed_refs
->lock
);
2258 count
= delayed_refs
->num_entries
* 2;
2262 if (!(run_all
|| run_most
) &&
2263 delayed_refs
->num_heads_ready
< 64)
2267 * go find something we can process in the rbtree. We start at
2268 * the beginning of the tree, and then build a cluster
2269 * of refs to process starting at the first one we are able to
2272 ret
= btrfs_find_ref_cluster(trans
, &cluster
,
2273 delayed_refs
->run_delayed_start
);
2277 ret
= run_clustered_refs(trans
, root
, &cluster
);
2280 count
-= min_t(unsigned long, ret
, count
);
2287 node
= rb_first(&delayed_refs
->root
);
2290 count
= (unsigned long)-1;
2293 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
2295 if (btrfs_delayed_ref_is_head(ref
)) {
2296 struct btrfs_delayed_ref_head
*head
;
2298 head
= btrfs_delayed_node_to_head(ref
);
2299 atomic_inc(&ref
->refs
);
2301 spin_unlock(&delayed_refs
->lock
);
2303 * Mutex was contended, block until it's
2304 * released and try again
2306 mutex_lock(&head
->mutex
);
2307 mutex_unlock(&head
->mutex
);
2309 btrfs_put_delayed_ref(ref
);
2313 node
= rb_next(node
);
2315 spin_unlock(&delayed_refs
->lock
);
2316 schedule_timeout(1);
2320 spin_unlock(&delayed_refs
->lock
);
2324 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
2325 struct btrfs_root
*root
,
2326 u64 bytenr
, u64 num_bytes
, u64 flags
,
2329 struct btrfs_delayed_extent_op
*extent_op
;
2332 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
2336 extent_op
->flags_to_set
= flags
;
2337 extent_op
->update_flags
= 1;
2338 extent_op
->update_key
= 0;
2339 extent_op
->is_data
= is_data
? 1 : 0;
2341 ret
= btrfs_add_delayed_extent_op(trans
, bytenr
, num_bytes
, extent_op
);
2347 static noinline
int check_delayed_ref(struct btrfs_trans_handle
*trans
,
2348 struct btrfs_root
*root
,
2349 struct btrfs_path
*path
,
2350 u64 objectid
, u64 offset
, u64 bytenr
)
2352 struct btrfs_delayed_ref_head
*head
;
2353 struct btrfs_delayed_ref_node
*ref
;
2354 struct btrfs_delayed_data_ref
*data_ref
;
2355 struct btrfs_delayed_ref_root
*delayed_refs
;
2356 struct rb_node
*node
;
2360 delayed_refs
= &trans
->transaction
->delayed_refs
;
2361 spin_lock(&delayed_refs
->lock
);
2362 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
2366 if (!mutex_trylock(&head
->mutex
)) {
2367 atomic_inc(&head
->node
.refs
);
2368 spin_unlock(&delayed_refs
->lock
);
2370 btrfs_release_path(path
);
2373 * Mutex was contended, block until it's released and let
2376 mutex_lock(&head
->mutex
);
2377 mutex_unlock(&head
->mutex
);
2378 btrfs_put_delayed_ref(&head
->node
);
2382 node
= rb_prev(&head
->node
.rb_node
);
2386 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2388 if (ref
->bytenr
!= bytenr
)
2392 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
)
2395 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
2397 node
= rb_prev(node
);
2399 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2400 if (ref
->bytenr
== bytenr
)
2404 if (data_ref
->root
!= root
->root_key
.objectid
||
2405 data_ref
->objectid
!= objectid
|| data_ref
->offset
!= offset
)
2410 mutex_unlock(&head
->mutex
);
2412 spin_unlock(&delayed_refs
->lock
);
2416 static noinline
int check_committed_ref(struct btrfs_trans_handle
*trans
,
2417 struct btrfs_root
*root
,
2418 struct btrfs_path
*path
,
2419 u64 objectid
, u64 offset
, u64 bytenr
)
2421 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2422 struct extent_buffer
*leaf
;
2423 struct btrfs_extent_data_ref
*ref
;
2424 struct btrfs_extent_inline_ref
*iref
;
2425 struct btrfs_extent_item
*ei
;
2426 struct btrfs_key key
;
2430 key
.objectid
= bytenr
;
2431 key
.offset
= (u64
)-1;
2432 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2434 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
2440 if (path
->slots
[0] == 0)
2444 leaf
= path
->nodes
[0];
2445 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2447 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
2451 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2452 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2453 if (item_size
< sizeof(*ei
)) {
2454 WARN_ON(item_size
!= sizeof(struct btrfs_extent_item_v0
));
2458 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2460 if (item_size
!= sizeof(*ei
) +
2461 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
2464 if (btrfs_extent_generation(leaf
, ei
) <=
2465 btrfs_root_last_snapshot(&root
->root_item
))
2468 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
2469 if (btrfs_extent_inline_ref_type(leaf
, iref
) !=
2470 BTRFS_EXTENT_DATA_REF_KEY
)
2473 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
2474 if (btrfs_extent_refs(leaf
, ei
) !=
2475 btrfs_extent_data_ref_count(leaf
, ref
) ||
2476 btrfs_extent_data_ref_root(leaf
, ref
) !=
2477 root
->root_key
.objectid
||
2478 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
2479 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
2487 int btrfs_cross_ref_exist(struct btrfs_trans_handle
*trans
,
2488 struct btrfs_root
*root
,
2489 u64 objectid
, u64 offset
, u64 bytenr
)
2491 struct btrfs_path
*path
;
2495 path
= btrfs_alloc_path();
2500 ret
= check_committed_ref(trans
, root
, path
, objectid
,
2502 if (ret
&& ret
!= -ENOENT
)
2505 ret2
= check_delayed_ref(trans
, root
, path
, objectid
,
2507 } while (ret2
== -EAGAIN
);
2509 if (ret2
&& ret2
!= -ENOENT
) {
2514 if (ret
!= -ENOENT
|| ret2
!= -ENOENT
)
2517 btrfs_free_path(path
);
2518 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
2523 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
2524 struct btrfs_root
*root
,
2525 struct extent_buffer
*buf
,
2526 int full_backref
, int inc
)
2533 struct btrfs_key key
;
2534 struct btrfs_file_extent_item
*fi
;
2538 int (*process_func
)(struct btrfs_trans_handle
*, struct btrfs_root
*,
2539 u64
, u64
, u64
, u64
, u64
, u64
);
2541 ref_root
= btrfs_header_owner(buf
);
2542 nritems
= btrfs_header_nritems(buf
);
2543 level
= btrfs_header_level(buf
);
2545 if (!root
->ref_cows
&& level
== 0)
2549 process_func
= btrfs_inc_extent_ref
;
2551 process_func
= btrfs_free_extent
;
2554 parent
= buf
->start
;
2558 for (i
= 0; i
< nritems
; i
++) {
2560 btrfs_item_key_to_cpu(buf
, &key
, i
);
2561 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
2563 fi
= btrfs_item_ptr(buf
, i
,
2564 struct btrfs_file_extent_item
);
2565 if (btrfs_file_extent_type(buf
, fi
) ==
2566 BTRFS_FILE_EXTENT_INLINE
)
2568 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
2572 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
2573 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
2574 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
2575 parent
, ref_root
, key
.objectid
,
2580 bytenr
= btrfs_node_blockptr(buf
, i
);
2581 num_bytes
= btrfs_level_size(root
, level
- 1);
2582 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
2583 parent
, ref_root
, level
- 1, 0);
2594 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2595 struct extent_buffer
*buf
, int full_backref
)
2597 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1);
2600 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2601 struct extent_buffer
*buf
, int full_backref
)
2603 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0);
2606 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
2607 struct btrfs_root
*root
,
2608 struct btrfs_path
*path
,
2609 struct btrfs_block_group_cache
*cache
)
2612 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2614 struct extent_buffer
*leaf
;
2616 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
2621 leaf
= path
->nodes
[0];
2622 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
2623 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
2624 btrfs_mark_buffer_dirty(leaf
);
2625 btrfs_release_path(path
);
2633 static struct btrfs_block_group_cache
*
2634 next_block_group(struct btrfs_root
*root
,
2635 struct btrfs_block_group_cache
*cache
)
2637 struct rb_node
*node
;
2638 spin_lock(&root
->fs_info
->block_group_cache_lock
);
2639 node
= rb_next(&cache
->cache_node
);
2640 btrfs_put_block_group(cache
);
2642 cache
= rb_entry(node
, struct btrfs_block_group_cache
,
2644 btrfs_get_block_group(cache
);
2647 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
2651 static int cache_save_setup(struct btrfs_block_group_cache
*block_group
,
2652 struct btrfs_trans_handle
*trans
,
2653 struct btrfs_path
*path
)
2655 struct btrfs_root
*root
= block_group
->fs_info
->tree_root
;
2656 struct inode
*inode
= NULL
;
2658 int dcs
= BTRFS_DC_ERROR
;
2664 * If this block group is smaller than 100 megs don't bother caching the
2667 if (block_group
->key
.offset
< (100 * 1024 * 1024)) {
2668 spin_lock(&block_group
->lock
);
2669 block_group
->disk_cache_state
= BTRFS_DC_WRITTEN
;
2670 spin_unlock(&block_group
->lock
);
2675 inode
= lookup_free_space_inode(root
, block_group
, path
);
2676 if (IS_ERR(inode
) && PTR_ERR(inode
) != -ENOENT
) {
2677 ret
= PTR_ERR(inode
);
2678 btrfs_release_path(path
);
2682 if (IS_ERR(inode
)) {
2686 if (block_group
->ro
)
2689 ret
= create_free_space_inode(root
, trans
, block_group
, path
);
2696 * We want to set the generation to 0, that way if anything goes wrong
2697 * from here on out we know not to trust this cache when we load up next
2700 BTRFS_I(inode
)->generation
= 0;
2701 ret
= btrfs_update_inode(trans
, root
, inode
);
2704 if (i_size_read(inode
) > 0) {
2705 ret
= btrfs_truncate_free_space_cache(root
, trans
, path
,
2711 spin_lock(&block_group
->lock
);
2712 if (block_group
->cached
!= BTRFS_CACHE_FINISHED
) {
2713 /* We're not cached, don't bother trying to write stuff out */
2714 dcs
= BTRFS_DC_WRITTEN
;
2715 spin_unlock(&block_group
->lock
);
2718 spin_unlock(&block_group
->lock
);
2720 num_pages
= (int)div64_u64(block_group
->key
.offset
, 1024 * 1024 * 1024);
2725 * Just to make absolutely sure we have enough space, we're going to
2726 * preallocate 12 pages worth of space for each block group. In
2727 * practice we ought to use at most 8, but we need extra space so we can
2728 * add our header and have a terminator between the extents and the
2732 num_pages
*= PAGE_CACHE_SIZE
;
2734 ret
= btrfs_check_data_free_space(inode
, num_pages
);
2738 ret
= btrfs_prealloc_file_range_trans(inode
, trans
, 0, 0, num_pages
,
2739 num_pages
, num_pages
,
2742 dcs
= BTRFS_DC_SETUP
;
2743 btrfs_free_reserved_data_space(inode
, num_pages
);
2747 btrfs_release_path(path
);
2749 spin_lock(&block_group
->lock
);
2750 block_group
->disk_cache_state
= dcs
;
2751 spin_unlock(&block_group
->lock
);
2756 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
2757 struct btrfs_root
*root
)
2759 struct btrfs_block_group_cache
*cache
;
2761 struct btrfs_path
*path
;
2764 path
= btrfs_alloc_path();
2770 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
2772 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
)
2774 cache
= next_block_group(root
, cache
);
2782 err
= cache_save_setup(cache
, trans
, path
);
2783 last
= cache
->key
.objectid
+ cache
->key
.offset
;
2784 btrfs_put_block_group(cache
);
2789 err
= btrfs_run_delayed_refs(trans
, root
,
2794 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
2796 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
) {
2797 btrfs_put_block_group(cache
);
2803 cache
= next_block_group(root
, cache
);
2812 if (cache
->disk_cache_state
== BTRFS_DC_SETUP
)
2813 cache
->disk_cache_state
= BTRFS_DC_NEED_WRITE
;
2815 last
= cache
->key
.objectid
+ cache
->key
.offset
;
2817 err
= write_one_cache_group(trans
, root
, path
, cache
);
2819 btrfs_put_block_group(cache
);
2824 * I don't think this is needed since we're just marking our
2825 * preallocated extent as written, but just in case it can't
2829 err
= btrfs_run_delayed_refs(trans
, root
,
2834 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
2837 * Really this shouldn't happen, but it could if we
2838 * couldn't write the entire preallocated extent and
2839 * splitting the extent resulted in a new block.
2842 btrfs_put_block_group(cache
);
2845 if (cache
->disk_cache_state
== BTRFS_DC_NEED_WRITE
)
2847 cache
= next_block_group(root
, cache
);
2856 btrfs_write_out_cache(root
, trans
, cache
, path
);
2859 * If we didn't have an error then the cache state is still
2860 * NEED_WRITE, so we can set it to WRITTEN.
2862 if (cache
->disk_cache_state
== BTRFS_DC_NEED_WRITE
)
2863 cache
->disk_cache_state
= BTRFS_DC_WRITTEN
;
2864 last
= cache
->key
.objectid
+ cache
->key
.offset
;
2865 btrfs_put_block_group(cache
);
2868 btrfs_free_path(path
);
2872 int btrfs_extent_readonly(struct btrfs_root
*root
, u64 bytenr
)
2874 struct btrfs_block_group_cache
*block_group
;
2877 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
2878 if (!block_group
|| block_group
->ro
)
2881 btrfs_put_block_group(block_group
);
2885 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
2886 u64 total_bytes
, u64 bytes_used
,
2887 struct btrfs_space_info
**space_info
)
2889 struct btrfs_space_info
*found
;
2893 if (flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
2894 BTRFS_BLOCK_GROUP_RAID10
))
2899 found
= __find_space_info(info
, flags
);
2901 spin_lock(&found
->lock
);
2902 found
->total_bytes
+= total_bytes
;
2903 found
->disk_total
+= total_bytes
* factor
;
2904 found
->bytes_used
+= bytes_used
;
2905 found
->disk_used
+= bytes_used
* factor
;
2907 spin_unlock(&found
->lock
);
2908 *space_info
= found
;
2911 found
= kzalloc(sizeof(*found
), GFP_NOFS
);
2915 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
2916 INIT_LIST_HEAD(&found
->block_groups
[i
]);
2917 init_rwsem(&found
->groups_sem
);
2918 spin_lock_init(&found
->lock
);
2919 found
->flags
= flags
& (BTRFS_BLOCK_GROUP_DATA
|
2920 BTRFS_BLOCK_GROUP_SYSTEM
|
2921 BTRFS_BLOCK_GROUP_METADATA
);
2922 found
->total_bytes
= total_bytes
;
2923 found
->disk_total
= total_bytes
* factor
;
2924 found
->bytes_used
= bytes_used
;
2925 found
->disk_used
= bytes_used
* factor
;
2926 found
->bytes_pinned
= 0;
2927 found
->bytes_reserved
= 0;
2928 found
->bytes_readonly
= 0;
2929 found
->bytes_may_use
= 0;
2931 found
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
2932 found
->chunk_alloc
= 0;
2934 init_waitqueue_head(&found
->wait
);
2935 *space_info
= found
;
2936 list_add_rcu(&found
->list
, &info
->space_info
);
2940 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
2942 u64 extra_flags
= flags
& (BTRFS_BLOCK_GROUP_RAID0
|
2943 BTRFS_BLOCK_GROUP_RAID1
|
2944 BTRFS_BLOCK_GROUP_RAID10
|
2945 BTRFS_BLOCK_GROUP_DUP
);
2947 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
2948 fs_info
->avail_data_alloc_bits
|= extra_flags
;
2949 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
2950 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
2951 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
2952 fs_info
->avail_system_alloc_bits
|= extra_flags
;
2956 u64
btrfs_reduce_alloc_profile(struct btrfs_root
*root
, u64 flags
)
2959 * we add in the count of missing devices because we want
2960 * to make sure that any RAID levels on a degraded FS
2961 * continue to be honored.
2963 u64 num_devices
= root
->fs_info
->fs_devices
->rw_devices
+
2964 root
->fs_info
->fs_devices
->missing_devices
;
2966 if (num_devices
== 1)
2967 flags
&= ~(BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID0
);
2968 if (num_devices
< 4)
2969 flags
&= ~BTRFS_BLOCK_GROUP_RAID10
;
2971 if ((flags
& BTRFS_BLOCK_GROUP_DUP
) &&
2972 (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
2973 BTRFS_BLOCK_GROUP_RAID10
))) {
2974 flags
&= ~BTRFS_BLOCK_GROUP_DUP
;
2977 if ((flags
& BTRFS_BLOCK_GROUP_RAID1
) &&
2978 (flags
& BTRFS_BLOCK_GROUP_RAID10
)) {
2979 flags
&= ~BTRFS_BLOCK_GROUP_RAID1
;
2982 if ((flags
& BTRFS_BLOCK_GROUP_RAID0
) &&
2983 ((flags
& BTRFS_BLOCK_GROUP_RAID1
) |
2984 (flags
& BTRFS_BLOCK_GROUP_RAID10
) |
2985 (flags
& BTRFS_BLOCK_GROUP_DUP
)))
2986 flags
&= ~BTRFS_BLOCK_GROUP_RAID0
;
2990 static u64
get_alloc_profile(struct btrfs_root
*root
, u64 flags
)
2992 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
2993 flags
|= root
->fs_info
->avail_data_alloc_bits
&
2994 root
->fs_info
->data_alloc_profile
;
2995 else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
2996 flags
|= root
->fs_info
->avail_system_alloc_bits
&
2997 root
->fs_info
->system_alloc_profile
;
2998 else if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
2999 flags
|= root
->fs_info
->avail_metadata_alloc_bits
&
3000 root
->fs_info
->metadata_alloc_profile
;
3001 return btrfs_reduce_alloc_profile(root
, flags
);
3004 u64
btrfs_get_alloc_profile(struct btrfs_root
*root
, int data
)
3009 flags
= BTRFS_BLOCK_GROUP_DATA
;
3010 else if (root
== root
->fs_info
->chunk_root
)
3011 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
3013 flags
= BTRFS_BLOCK_GROUP_METADATA
;
3015 return get_alloc_profile(root
, flags
);
3018 void btrfs_set_inode_space_info(struct btrfs_root
*root
, struct inode
*inode
)
3020 BTRFS_I(inode
)->space_info
= __find_space_info(root
->fs_info
,
3021 BTRFS_BLOCK_GROUP_DATA
);
3025 * This will check the space that the inode allocates from to make sure we have
3026 * enough space for bytes.
3028 int btrfs_check_data_free_space(struct inode
*inode
, u64 bytes
)
3030 struct btrfs_space_info
*data_sinfo
;
3031 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3033 int ret
= 0, committed
= 0, alloc_chunk
= 1;
3035 /* make sure bytes are sectorsize aligned */
3036 bytes
= (bytes
+ root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
3038 if (root
== root
->fs_info
->tree_root
||
3039 BTRFS_I(inode
)->location
.objectid
== BTRFS_FREE_INO_OBJECTID
) {
3044 data_sinfo
= BTRFS_I(inode
)->space_info
;
3049 /* make sure we have enough space to handle the data first */
3050 spin_lock(&data_sinfo
->lock
);
3051 used
= data_sinfo
->bytes_used
+ data_sinfo
->bytes_reserved
+
3052 data_sinfo
->bytes_pinned
+ data_sinfo
->bytes_readonly
+
3053 data_sinfo
->bytes_may_use
;
3055 if (used
+ bytes
> data_sinfo
->total_bytes
) {
3056 struct btrfs_trans_handle
*trans
;
3059 * if we don't have enough free bytes in this space then we need
3060 * to alloc a new chunk.
3062 if (!data_sinfo
->full
&& alloc_chunk
) {
3065 data_sinfo
->force_alloc
= CHUNK_ALLOC_FORCE
;
3066 spin_unlock(&data_sinfo
->lock
);
3068 alloc_target
= btrfs_get_alloc_profile(root
, 1);
3069 trans
= btrfs_join_transaction(root
);
3071 return PTR_ERR(trans
);
3073 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
3074 bytes
+ 2 * 1024 * 1024,
3076 CHUNK_ALLOC_NO_FORCE
);
3077 btrfs_end_transaction(trans
, root
);
3086 btrfs_set_inode_space_info(root
, inode
);
3087 data_sinfo
= BTRFS_I(inode
)->space_info
;
3093 * If we have less pinned bytes than we want to allocate then
3094 * don't bother committing the transaction, it won't help us.
3096 if (data_sinfo
->bytes_pinned
< bytes
)
3098 spin_unlock(&data_sinfo
->lock
);
3100 /* commit the current transaction and try again */
3103 !atomic_read(&root
->fs_info
->open_ioctl_trans
)) {
3105 trans
= btrfs_join_transaction(root
);
3107 return PTR_ERR(trans
);
3108 ret
= btrfs_commit_transaction(trans
, root
);
3116 data_sinfo
->bytes_may_use
+= bytes
;
3117 BTRFS_I(inode
)->reserved_bytes
+= bytes
;
3118 spin_unlock(&data_sinfo
->lock
);
3124 * called when we are clearing an delalloc extent from the
3125 * inode's io_tree or there was an error for whatever reason
3126 * after calling btrfs_check_data_free_space
3128 void btrfs_free_reserved_data_space(struct inode
*inode
, u64 bytes
)
3130 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3131 struct btrfs_space_info
*data_sinfo
;
3133 /* make sure bytes are sectorsize aligned */
3134 bytes
= (bytes
+ root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
3136 data_sinfo
= BTRFS_I(inode
)->space_info
;
3137 spin_lock(&data_sinfo
->lock
);
3138 data_sinfo
->bytes_may_use
-= bytes
;
3139 BTRFS_I(inode
)->reserved_bytes
-= bytes
;
3140 spin_unlock(&data_sinfo
->lock
);
3143 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
3145 struct list_head
*head
= &info
->space_info
;
3146 struct btrfs_space_info
*found
;
3149 list_for_each_entry_rcu(found
, head
, list
) {
3150 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
3151 found
->force_alloc
= CHUNK_ALLOC_FORCE
;
3156 static int should_alloc_chunk(struct btrfs_root
*root
,
3157 struct btrfs_space_info
*sinfo
, u64 alloc_bytes
,
3160 u64 num_bytes
= sinfo
->total_bytes
- sinfo
->bytes_readonly
;
3161 u64 num_allocated
= sinfo
->bytes_used
+ sinfo
->bytes_reserved
;
3164 if (force
== CHUNK_ALLOC_FORCE
)
3168 * in limited mode, we want to have some free space up to
3169 * about 1% of the FS size.
3171 if (force
== CHUNK_ALLOC_LIMITED
) {
3172 thresh
= btrfs_super_total_bytes(&root
->fs_info
->super_copy
);
3173 thresh
= max_t(u64
, 64 * 1024 * 1024,
3174 div_factor_fine(thresh
, 1));
3176 if (num_bytes
- num_allocated
< thresh
)
3181 * we have two similar checks here, one based on percentage
3182 * and once based on a hard number of 256MB. The idea
3183 * is that if we have a good amount of free
3184 * room, don't allocate a chunk. A good mount is
3185 * less than 80% utilized of the chunks we have allocated,
3186 * or more than 256MB free
3188 if (num_allocated
+ alloc_bytes
+ 256 * 1024 * 1024 < num_bytes
)
3191 if (num_allocated
+ alloc_bytes
< div_factor(num_bytes
, 8))
3194 thresh
= btrfs_super_total_bytes(&root
->fs_info
->super_copy
);
3196 /* 256MB or 5% of the FS */
3197 thresh
= max_t(u64
, 256 * 1024 * 1024, div_factor_fine(thresh
, 5));
3199 if (num_bytes
> thresh
&& sinfo
->bytes_used
< div_factor(num_bytes
, 3))
3204 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
3205 struct btrfs_root
*extent_root
, u64 alloc_bytes
,
3206 u64 flags
, int force
)
3208 struct btrfs_space_info
*space_info
;
3209 struct btrfs_fs_info
*fs_info
= extent_root
->fs_info
;
3210 int wait_for_alloc
= 0;
3213 flags
= btrfs_reduce_alloc_profile(extent_root
, flags
);
3215 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
3217 ret
= update_space_info(extent_root
->fs_info
, flags
,
3221 BUG_ON(!space_info
);
3224 spin_lock(&space_info
->lock
);
3225 if (space_info
->force_alloc
)
3226 force
= space_info
->force_alloc
;
3227 if (space_info
->full
) {
3228 spin_unlock(&space_info
->lock
);
3232 if (!should_alloc_chunk(extent_root
, space_info
, alloc_bytes
, force
)) {
3233 spin_unlock(&space_info
->lock
);
3235 } else if (space_info
->chunk_alloc
) {
3238 space_info
->chunk_alloc
= 1;
3241 spin_unlock(&space_info
->lock
);
3243 mutex_lock(&fs_info
->chunk_mutex
);
3246 * The chunk_mutex is held throughout the entirety of a chunk
3247 * allocation, so once we've acquired the chunk_mutex we know that the
3248 * other guy is done and we need to recheck and see if we should
3251 if (wait_for_alloc
) {
3252 mutex_unlock(&fs_info
->chunk_mutex
);
3258 * If we have mixed data/metadata chunks we want to make sure we keep
3259 * allocating mixed chunks instead of individual chunks.
3261 if (btrfs_mixed_space_info(space_info
))
3262 flags
|= (BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
);
3265 * if we're doing a data chunk, go ahead and make sure that
3266 * we keep a reasonable number of metadata chunks allocated in the
3269 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
3270 fs_info
->data_chunk_allocations
++;
3271 if (!(fs_info
->data_chunk_allocations
%
3272 fs_info
->metadata_ratio
))
3273 force_metadata_allocation(fs_info
);
3276 ret
= btrfs_alloc_chunk(trans
, extent_root
, flags
);
3277 if (ret
< 0 && ret
!= -ENOSPC
)
3280 spin_lock(&space_info
->lock
);
3282 space_info
->full
= 1;
3286 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3287 space_info
->chunk_alloc
= 0;
3288 spin_unlock(&space_info
->lock
);
3290 mutex_unlock(&extent_root
->fs_info
->chunk_mutex
);
3295 * shrink metadata reservation for delalloc
3297 static int shrink_delalloc(struct btrfs_trans_handle
*trans
,
3298 struct btrfs_root
*root
, u64 to_reclaim
, int sync
)
3300 struct btrfs_block_rsv
*block_rsv
;
3301 struct btrfs_space_info
*space_info
;
3306 int nr_pages
= (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT
;
3308 unsigned long progress
;
3310 block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
3311 space_info
= block_rsv
->space_info
;
3314 reserved
= space_info
->bytes_reserved
;
3315 progress
= space_info
->reservation_progress
;
3321 if (root
->fs_info
->delalloc_bytes
== 0) {
3324 btrfs_wait_ordered_extents(root
, 0, 0);
3328 max_reclaim
= min(reserved
, to_reclaim
);
3330 while (loops
< 1024) {
3331 /* have the flusher threads jump in and do some IO */
3333 nr_pages
= min_t(unsigned long, nr_pages
,
3334 root
->fs_info
->delalloc_bytes
>> PAGE_CACHE_SHIFT
);
3335 writeback_inodes_sb_nr_if_idle(root
->fs_info
->sb
, nr_pages
);
3337 spin_lock(&space_info
->lock
);
3338 if (reserved
> space_info
->bytes_reserved
)
3339 reclaimed
+= reserved
- space_info
->bytes_reserved
;
3340 reserved
= space_info
->bytes_reserved
;
3341 spin_unlock(&space_info
->lock
);
3345 if (reserved
== 0 || reclaimed
>= max_reclaim
)
3348 if (trans
&& trans
->transaction
->blocked
)
3351 time_left
= schedule_timeout_interruptible(1);
3353 /* We were interrupted, exit */
3357 /* we've kicked the IO a few times, if anything has been freed,
3358 * exit. There is no sense in looping here for a long time
3359 * when we really need to commit the transaction, or there are
3360 * just too many writers without enough free space
3365 if (progress
!= space_info
->reservation_progress
)
3370 if (reclaimed
>= to_reclaim
&& !trans
)
3371 btrfs_wait_ordered_extents(root
, 0, 0);
3372 return reclaimed
>= to_reclaim
;
3376 * Retries tells us how many times we've called reserve_metadata_bytes. The
3377 * idea is if this is the first call (retries == 0) then we will add to our
3378 * reserved count if we can't make the allocation in order to hold our place
3379 * while we go and try and free up space. That way for retries > 1 we don't try
3380 * and add space, we just check to see if the amount of unused space is >= the
3381 * total space, meaning that our reservation is valid.
3383 * However if we don't intend to retry this reservation, pass -1 as retries so
3384 * that it short circuits this logic.
3386 static int reserve_metadata_bytes(struct btrfs_trans_handle
*trans
,
3387 struct btrfs_root
*root
,
3388 struct btrfs_block_rsv
*block_rsv
,
3389 u64 orig_bytes
, int flush
)
3391 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
3393 u64 num_bytes
= orig_bytes
;
3396 bool committed
= false;
3397 bool flushing
= false;
3401 spin_lock(&space_info
->lock
);
3403 * We only want to wait if somebody other than us is flushing and we are
3404 * actually alloed to flush.
3406 while (flush
&& !flushing
&& space_info
->flush
) {
3407 spin_unlock(&space_info
->lock
);
3409 * If we have a trans handle we can't wait because the flusher
3410 * may have to commit the transaction, which would mean we would
3411 * deadlock since we are waiting for the flusher to finish, but
3412 * hold the current transaction open.
3416 ret
= wait_event_interruptible(space_info
->wait
,
3417 !space_info
->flush
);
3418 /* Must have been interrupted, return */
3422 spin_lock(&space_info
->lock
);
3426 unused
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
3427 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
3428 space_info
->bytes_may_use
;
3431 * The idea here is that we've not already over-reserved the block group
3432 * then we can go ahead and save our reservation first and then start
3433 * flushing if we need to. Otherwise if we've already overcommitted
3434 * lets start flushing stuff first and then come back and try to make
3437 if (unused
<= space_info
->total_bytes
) {
3438 unused
= space_info
->total_bytes
- unused
;
3439 if (unused
>= num_bytes
) {
3440 space_info
->bytes_reserved
+= orig_bytes
;
3444 * Ok set num_bytes to orig_bytes since we aren't
3445 * overocmmitted, this way we only try and reclaim what
3448 num_bytes
= orig_bytes
;
3452 * Ok we're over committed, set num_bytes to the overcommitted
3453 * amount plus the amount of bytes that we need for this
3456 num_bytes
= unused
- space_info
->total_bytes
+
3457 (orig_bytes
* (retries
+ 1));
3461 * Couldn't make our reservation, save our place so while we're trying
3462 * to reclaim space we can actually use it instead of somebody else
3463 * stealing it from us.
3467 space_info
->flush
= 1;
3470 spin_unlock(&space_info
->lock
);
3476 * We do synchronous shrinking since we don't actually unreserve
3477 * metadata until after the IO is completed.
3479 ret
= shrink_delalloc(trans
, root
, num_bytes
, 1);
3486 * So if we were overcommitted it's possible that somebody else flushed
3487 * out enough space and we simply didn't have enough space to reclaim,
3488 * so go back around and try again.
3496 * Not enough space to be reclaimed, don't bother committing the
3499 spin_lock(&space_info
->lock
);
3500 if (space_info
->bytes_pinned
< orig_bytes
)
3502 spin_unlock(&space_info
->lock
);
3514 trans
= btrfs_join_transaction(root
);
3517 ret
= btrfs_commit_transaction(trans
, root
);
3526 spin_lock(&space_info
->lock
);
3527 space_info
->flush
= 0;
3528 wake_up_all(&space_info
->wait
);
3529 spin_unlock(&space_info
->lock
);
3534 static struct btrfs_block_rsv
*get_block_rsv(struct btrfs_trans_handle
*trans
,
3535 struct btrfs_root
*root
)
3537 struct btrfs_block_rsv
*block_rsv
;
3539 block_rsv
= trans
->block_rsv
;
3541 block_rsv
= root
->block_rsv
;
3544 block_rsv
= &root
->fs_info
->empty_block_rsv
;
3549 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
3553 spin_lock(&block_rsv
->lock
);
3554 if (block_rsv
->reserved
>= num_bytes
) {
3555 block_rsv
->reserved
-= num_bytes
;
3556 if (block_rsv
->reserved
< block_rsv
->size
)
3557 block_rsv
->full
= 0;
3560 spin_unlock(&block_rsv
->lock
);
3564 static void block_rsv_add_bytes(struct btrfs_block_rsv
*block_rsv
,
3565 u64 num_bytes
, int update_size
)
3567 spin_lock(&block_rsv
->lock
);
3568 block_rsv
->reserved
+= num_bytes
;
3570 block_rsv
->size
+= num_bytes
;
3571 else if (block_rsv
->reserved
>= block_rsv
->size
)
3572 block_rsv
->full
= 1;
3573 spin_unlock(&block_rsv
->lock
);
3576 static void block_rsv_release_bytes(struct btrfs_block_rsv
*block_rsv
,
3577 struct btrfs_block_rsv
*dest
, u64 num_bytes
)
3579 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
3581 spin_lock(&block_rsv
->lock
);
3582 if (num_bytes
== (u64
)-1)
3583 num_bytes
= block_rsv
->size
;
3584 block_rsv
->size
-= num_bytes
;
3585 if (block_rsv
->reserved
>= block_rsv
->size
) {
3586 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
3587 block_rsv
->reserved
= block_rsv
->size
;
3588 block_rsv
->full
= 1;
3592 spin_unlock(&block_rsv
->lock
);
3594 if (num_bytes
> 0) {
3596 spin_lock(&dest
->lock
);
3600 bytes_to_add
= dest
->size
- dest
->reserved
;
3601 bytes_to_add
= min(num_bytes
, bytes_to_add
);
3602 dest
->reserved
+= bytes_to_add
;
3603 if (dest
->reserved
>= dest
->size
)
3605 num_bytes
-= bytes_to_add
;
3607 spin_unlock(&dest
->lock
);
3610 spin_lock(&space_info
->lock
);
3611 space_info
->bytes_reserved
-= num_bytes
;
3612 space_info
->reservation_progress
++;
3613 spin_unlock(&space_info
->lock
);
3618 static int block_rsv_migrate_bytes(struct btrfs_block_rsv
*src
,
3619 struct btrfs_block_rsv
*dst
, u64 num_bytes
)
3623 ret
= block_rsv_use_bytes(src
, num_bytes
);
3627 block_rsv_add_bytes(dst
, num_bytes
, 1);
3631 void btrfs_init_block_rsv(struct btrfs_block_rsv
*rsv
)
3633 memset(rsv
, 0, sizeof(*rsv
));
3634 spin_lock_init(&rsv
->lock
);
3635 atomic_set(&rsv
->usage
, 1);
3637 INIT_LIST_HEAD(&rsv
->list
);
3640 struct btrfs_block_rsv
*btrfs_alloc_block_rsv(struct btrfs_root
*root
)
3642 struct btrfs_block_rsv
*block_rsv
;
3643 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3645 block_rsv
= kmalloc(sizeof(*block_rsv
), GFP_NOFS
);
3649 btrfs_init_block_rsv(block_rsv
);
3650 block_rsv
->space_info
= __find_space_info(fs_info
,
3651 BTRFS_BLOCK_GROUP_METADATA
);
3655 void btrfs_free_block_rsv(struct btrfs_root
*root
,
3656 struct btrfs_block_rsv
*rsv
)
3658 if (rsv
&& atomic_dec_and_test(&rsv
->usage
)) {
3659 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
3666 * make the block_rsv struct be able to capture freed space.
3667 * the captured space will re-add to the the block_rsv struct
3668 * after transaction commit
3670 void btrfs_add_durable_block_rsv(struct btrfs_fs_info
*fs_info
,
3671 struct btrfs_block_rsv
*block_rsv
)
3673 block_rsv
->durable
= 1;
3674 mutex_lock(&fs_info
->durable_block_rsv_mutex
);
3675 list_add_tail(&block_rsv
->list
, &fs_info
->durable_block_rsv_list
);
3676 mutex_unlock(&fs_info
->durable_block_rsv_mutex
);
3679 int btrfs_block_rsv_add(struct btrfs_trans_handle
*trans
,
3680 struct btrfs_root
*root
,
3681 struct btrfs_block_rsv
*block_rsv
,
3689 ret
= reserve_metadata_bytes(trans
, root
, block_rsv
, num_bytes
, 1);
3691 block_rsv_add_bytes(block_rsv
, num_bytes
, 1);
3698 int btrfs_block_rsv_check(struct btrfs_trans_handle
*trans
,
3699 struct btrfs_root
*root
,
3700 struct btrfs_block_rsv
*block_rsv
,
3701 u64 min_reserved
, int min_factor
)
3704 int commit_trans
= 0;
3710 spin_lock(&block_rsv
->lock
);
3712 num_bytes
= div_factor(block_rsv
->size
, min_factor
);
3713 if (min_reserved
> num_bytes
)
3714 num_bytes
= min_reserved
;
3716 if (block_rsv
->reserved
>= num_bytes
) {
3719 num_bytes
-= block_rsv
->reserved
;
3720 if (block_rsv
->durable
&&
3721 block_rsv
->freed
[0] + block_rsv
->freed
[1] >= num_bytes
)
3724 spin_unlock(&block_rsv
->lock
);
3728 if (block_rsv
->refill_used
) {
3729 ret
= reserve_metadata_bytes(trans
, root
, block_rsv
,
3732 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
3740 trans
= btrfs_join_transaction(root
);
3741 BUG_ON(IS_ERR(trans
));
3742 ret
= btrfs_commit_transaction(trans
, root
);
3749 int btrfs_block_rsv_migrate(struct btrfs_block_rsv
*src_rsv
,
3750 struct btrfs_block_rsv
*dst_rsv
,
3753 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
3756 void btrfs_block_rsv_release(struct btrfs_root
*root
,
3757 struct btrfs_block_rsv
*block_rsv
,
3760 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
3761 if (global_rsv
->full
|| global_rsv
== block_rsv
||
3762 block_rsv
->space_info
!= global_rsv
->space_info
)
3764 block_rsv_release_bytes(block_rsv
, global_rsv
, num_bytes
);
3768 * helper to calculate size of global block reservation.
3769 * the desired value is sum of space used by extent tree,
3770 * checksum tree and root tree
3772 static u64
calc_global_metadata_size(struct btrfs_fs_info
*fs_info
)
3774 struct btrfs_space_info
*sinfo
;
3778 int csum_size
= btrfs_super_csum_size(&fs_info
->super_copy
);
3780 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_DATA
);
3781 spin_lock(&sinfo
->lock
);
3782 data_used
= sinfo
->bytes_used
;
3783 spin_unlock(&sinfo
->lock
);
3785 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
3786 spin_lock(&sinfo
->lock
);
3787 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_DATA
)
3789 meta_used
= sinfo
->bytes_used
;
3790 spin_unlock(&sinfo
->lock
);
3792 num_bytes
= (data_used
>> fs_info
->sb
->s_blocksize_bits
) *
3794 num_bytes
+= div64_u64(data_used
+ meta_used
, 50);
3796 if (num_bytes
* 3 > meta_used
)
3797 num_bytes
= div64_u64(meta_used
, 3);
3799 return ALIGN(num_bytes
, fs_info
->extent_root
->leafsize
<< 10);
3802 static void update_global_block_rsv(struct btrfs_fs_info
*fs_info
)
3804 struct btrfs_block_rsv
*block_rsv
= &fs_info
->global_block_rsv
;
3805 struct btrfs_space_info
*sinfo
= block_rsv
->space_info
;
3808 num_bytes
= calc_global_metadata_size(fs_info
);
3810 spin_lock(&block_rsv
->lock
);
3811 spin_lock(&sinfo
->lock
);
3813 block_rsv
->size
= num_bytes
;
3815 num_bytes
= sinfo
->bytes_used
+ sinfo
->bytes_pinned
+
3816 sinfo
->bytes_reserved
+ sinfo
->bytes_readonly
+
3817 sinfo
->bytes_may_use
;
3819 if (sinfo
->total_bytes
> num_bytes
) {
3820 num_bytes
= sinfo
->total_bytes
- num_bytes
;
3821 block_rsv
->reserved
+= num_bytes
;
3822 sinfo
->bytes_reserved
+= num_bytes
;
3825 if (block_rsv
->reserved
>= block_rsv
->size
) {
3826 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
3827 sinfo
->bytes_reserved
-= num_bytes
;
3828 sinfo
->reservation_progress
++;
3829 block_rsv
->reserved
= block_rsv
->size
;
3830 block_rsv
->full
= 1;
3833 spin_unlock(&sinfo
->lock
);
3834 spin_unlock(&block_rsv
->lock
);
3837 static void init_global_block_rsv(struct btrfs_fs_info
*fs_info
)
3839 struct btrfs_space_info
*space_info
;
3841 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
3842 fs_info
->chunk_block_rsv
.space_info
= space_info
;
3843 fs_info
->chunk_block_rsv
.priority
= 10;
3845 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
3846 fs_info
->global_block_rsv
.space_info
= space_info
;
3847 fs_info
->global_block_rsv
.priority
= 10;
3848 fs_info
->global_block_rsv
.refill_used
= 1;
3849 fs_info
->delalloc_block_rsv
.space_info
= space_info
;
3850 fs_info
->trans_block_rsv
.space_info
= space_info
;
3851 fs_info
->empty_block_rsv
.space_info
= space_info
;
3852 fs_info
->empty_block_rsv
.priority
= 10;
3854 fs_info
->extent_root
->block_rsv
= &fs_info
->global_block_rsv
;
3855 fs_info
->csum_root
->block_rsv
= &fs_info
->global_block_rsv
;
3856 fs_info
->dev_root
->block_rsv
= &fs_info
->global_block_rsv
;
3857 fs_info
->tree_root
->block_rsv
= &fs_info
->global_block_rsv
;
3858 fs_info
->chunk_root
->block_rsv
= &fs_info
->chunk_block_rsv
;
3860 btrfs_add_durable_block_rsv(fs_info
, &fs_info
->global_block_rsv
);
3862 btrfs_add_durable_block_rsv(fs_info
, &fs_info
->delalloc_block_rsv
);
3864 update_global_block_rsv(fs_info
);
3867 static void release_global_block_rsv(struct btrfs_fs_info
*fs_info
)
3869 block_rsv_release_bytes(&fs_info
->global_block_rsv
, NULL
, (u64
)-1);
3870 WARN_ON(fs_info
->delalloc_block_rsv
.size
> 0);
3871 WARN_ON(fs_info
->delalloc_block_rsv
.reserved
> 0);
3872 WARN_ON(fs_info
->trans_block_rsv
.size
> 0);
3873 WARN_ON(fs_info
->trans_block_rsv
.reserved
> 0);
3874 WARN_ON(fs_info
->chunk_block_rsv
.size
> 0);
3875 WARN_ON(fs_info
->chunk_block_rsv
.reserved
> 0);
3878 int btrfs_truncate_reserve_metadata(struct btrfs_trans_handle
*trans
,
3879 struct btrfs_root
*root
,
3880 struct btrfs_block_rsv
*rsv
)
3882 struct btrfs_block_rsv
*trans_rsv
= &root
->fs_info
->trans_block_rsv
;
3887 * Truncate should be freeing data, but give us 2 items just in case it
3888 * needs to use some space. We may want to be smarter about this in the
3891 num_bytes
= btrfs_calc_trans_metadata_size(root
, 2);
3893 /* We already have enough bytes, just return */
3894 if (rsv
->reserved
>= num_bytes
)
3897 num_bytes
-= rsv
->reserved
;
3900 * You should have reserved enough space before hand to do this, so this
3903 ret
= block_rsv_migrate_bytes(trans_rsv
, rsv
, num_bytes
);
3909 void btrfs_trans_release_metadata(struct btrfs_trans_handle
*trans
,
3910 struct btrfs_root
*root
)
3912 if (!trans
->bytes_reserved
)
3915 BUG_ON(trans
->block_rsv
!= &root
->fs_info
->trans_block_rsv
);
3916 btrfs_block_rsv_release(root
, trans
->block_rsv
,
3917 trans
->bytes_reserved
);
3918 trans
->bytes_reserved
= 0;
3921 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle
*trans
,
3922 struct inode
*inode
)
3924 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3925 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
3926 struct btrfs_block_rsv
*dst_rsv
= root
->orphan_block_rsv
;
3929 * We need to hold space in order to delete our orphan item once we've
3930 * added it, so this takes the reservation so we can release it later
3931 * when we are truly done with the orphan item.
3933 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
3934 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
3937 void btrfs_orphan_release_metadata(struct inode
*inode
)
3939 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3940 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
3941 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
, num_bytes
);
3944 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle
*trans
,
3945 struct btrfs_pending_snapshot
*pending
)
3947 struct btrfs_root
*root
= pending
->root
;
3948 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
3949 struct btrfs_block_rsv
*dst_rsv
= &pending
->block_rsv
;
3951 * two for root back/forward refs, two for directory entries
3952 * and one for root of the snapshot.
3954 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 5);
3955 dst_rsv
->space_info
= src_rsv
->space_info
;
3956 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
3959 static unsigned drop_outstanding_extent(struct inode
*inode
)
3961 unsigned dropped_extents
= 0;
3963 spin_lock(&BTRFS_I(inode
)->lock
);
3964 BUG_ON(!BTRFS_I(inode
)->outstanding_extents
);
3965 BTRFS_I(inode
)->outstanding_extents
--;
3968 * If we have more or the same amount of outsanding extents than we have
3969 * reserved then we need to leave the reserved extents count alone.
3971 if (BTRFS_I(inode
)->outstanding_extents
>=
3972 BTRFS_I(inode
)->reserved_extents
)
3975 dropped_extents
= BTRFS_I(inode
)->reserved_extents
-
3976 BTRFS_I(inode
)->outstanding_extents
;
3977 BTRFS_I(inode
)->reserved_extents
-= dropped_extents
;
3979 spin_unlock(&BTRFS_I(inode
)->lock
);
3980 return dropped_extents
;
3983 static u64
calc_csum_metadata_size(struct inode
*inode
, u64 num_bytes
)
3985 return num_bytes
>>= 3;
3988 int btrfs_delalloc_reserve_metadata(struct inode
*inode
, u64 num_bytes
)
3990 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3991 struct btrfs_block_rsv
*block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
3993 unsigned nr_extents
= 0;
3996 if (btrfs_transaction_in_commit(root
->fs_info
))
3997 schedule_timeout(1);
3999 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
4001 spin_lock(&BTRFS_I(inode
)->lock
);
4002 BTRFS_I(inode
)->outstanding_extents
++;
4004 if (BTRFS_I(inode
)->outstanding_extents
>
4005 BTRFS_I(inode
)->reserved_extents
) {
4006 nr_extents
= BTRFS_I(inode
)->outstanding_extents
-
4007 BTRFS_I(inode
)->reserved_extents
;
4008 BTRFS_I(inode
)->reserved_extents
+= nr_extents
;
4010 to_reserve
= btrfs_calc_trans_metadata_size(root
, nr_extents
);
4012 spin_unlock(&BTRFS_I(inode
)->lock
);
4014 to_reserve
+= calc_csum_metadata_size(inode
, num_bytes
);
4015 ret
= reserve_metadata_bytes(NULL
, root
, block_rsv
, to_reserve
, 1);
4019 * We don't need the return value since our reservation failed,
4020 * we just need to clean up our counter.
4022 dropped
= drop_outstanding_extent(inode
);
4023 WARN_ON(dropped
> 1);
4027 block_rsv_add_bytes(block_rsv
, to_reserve
, 1);
4032 void btrfs_delalloc_release_metadata(struct inode
*inode
, u64 num_bytes
)
4034 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4038 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
4039 dropped
= drop_outstanding_extent(inode
);
4041 to_free
= calc_csum_metadata_size(inode
, num_bytes
);
4043 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
4045 btrfs_block_rsv_release(root
, &root
->fs_info
->delalloc_block_rsv
,
4049 int btrfs_delalloc_reserve_space(struct inode
*inode
, u64 num_bytes
)
4053 ret
= btrfs_check_data_free_space(inode
, num_bytes
);
4057 ret
= btrfs_delalloc_reserve_metadata(inode
, num_bytes
);
4059 btrfs_free_reserved_data_space(inode
, num_bytes
);
4066 void btrfs_delalloc_release_space(struct inode
*inode
, u64 num_bytes
)
4068 btrfs_delalloc_release_metadata(inode
, num_bytes
);
4069 btrfs_free_reserved_data_space(inode
, num_bytes
);
4072 static int update_block_group(struct btrfs_trans_handle
*trans
,
4073 struct btrfs_root
*root
,
4074 u64 bytenr
, u64 num_bytes
, int alloc
)
4076 struct btrfs_block_group_cache
*cache
= NULL
;
4077 struct btrfs_fs_info
*info
= root
->fs_info
;
4078 u64 total
= num_bytes
;
4083 /* block accounting for super block */
4084 spin_lock(&info
->delalloc_lock
);
4085 old_val
= btrfs_super_bytes_used(&info
->super_copy
);
4087 old_val
+= num_bytes
;
4089 old_val
-= num_bytes
;
4090 btrfs_set_super_bytes_used(&info
->super_copy
, old_val
);
4091 spin_unlock(&info
->delalloc_lock
);
4094 cache
= btrfs_lookup_block_group(info
, bytenr
);
4097 if (cache
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
4098 BTRFS_BLOCK_GROUP_RAID1
|
4099 BTRFS_BLOCK_GROUP_RAID10
))
4104 * If this block group has free space cache written out, we
4105 * need to make sure to load it if we are removing space. This
4106 * is because we need the unpinning stage to actually add the
4107 * space back to the block group, otherwise we will leak space.
4109 if (!alloc
&& cache
->cached
== BTRFS_CACHE_NO
)
4110 cache_block_group(cache
, trans
, NULL
, 1);
4112 byte_in_group
= bytenr
- cache
->key
.objectid
;
4113 WARN_ON(byte_in_group
> cache
->key
.offset
);
4115 spin_lock(&cache
->space_info
->lock
);
4116 spin_lock(&cache
->lock
);
4118 if (btrfs_super_cache_generation(&info
->super_copy
) != 0 &&
4119 cache
->disk_cache_state
< BTRFS_DC_CLEAR
)
4120 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
4123 old_val
= btrfs_block_group_used(&cache
->item
);
4124 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
4126 old_val
+= num_bytes
;
4127 btrfs_set_block_group_used(&cache
->item
, old_val
);
4128 cache
->reserved
-= num_bytes
;
4129 cache
->space_info
->bytes_reserved
-= num_bytes
;
4130 cache
->space_info
->reservation_progress
++;
4131 cache
->space_info
->bytes_used
+= num_bytes
;
4132 cache
->space_info
->disk_used
+= num_bytes
* factor
;
4133 spin_unlock(&cache
->lock
);
4134 spin_unlock(&cache
->space_info
->lock
);
4136 old_val
-= num_bytes
;
4137 btrfs_set_block_group_used(&cache
->item
, old_val
);
4138 cache
->pinned
+= num_bytes
;
4139 cache
->space_info
->bytes_pinned
+= num_bytes
;
4140 cache
->space_info
->bytes_used
-= num_bytes
;
4141 cache
->space_info
->disk_used
-= num_bytes
* factor
;
4142 spin_unlock(&cache
->lock
);
4143 spin_unlock(&cache
->space_info
->lock
);
4145 set_extent_dirty(info
->pinned_extents
,
4146 bytenr
, bytenr
+ num_bytes
- 1,
4147 GFP_NOFS
| __GFP_NOFAIL
);
4149 btrfs_put_block_group(cache
);
4151 bytenr
+= num_bytes
;
4156 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
4158 struct btrfs_block_group_cache
*cache
;
4161 cache
= btrfs_lookup_first_block_group(root
->fs_info
, search_start
);
4165 bytenr
= cache
->key
.objectid
;
4166 btrfs_put_block_group(cache
);
4171 static int pin_down_extent(struct btrfs_root
*root
,
4172 struct btrfs_block_group_cache
*cache
,
4173 u64 bytenr
, u64 num_bytes
, int reserved
)
4175 spin_lock(&cache
->space_info
->lock
);
4176 spin_lock(&cache
->lock
);
4177 cache
->pinned
+= num_bytes
;
4178 cache
->space_info
->bytes_pinned
+= num_bytes
;
4180 cache
->reserved
-= num_bytes
;
4181 cache
->space_info
->bytes_reserved
-= num_bytes
;
4182 cache
->space_info
->reservation_progress
++;
4184 spin_unlock(&cache
->lock
);
4185 spin_unlock(&cache
->space_info
->lock
);
4187 set_extent_dirty(root
->fs_info
->pinned_extents
, bytenr
,
4188 bytenr
+ num_bytes
- 1, GFP_NOFS
| __GFP_NOFAIL
);
4193 * this function must be called within transaction
4195 int btrfs_pin_extent(struct btrfs_root
*root
,
4196 u64 bytenr
, u64 num_bytes
, int reserved
)
4198 struct btrfs_block_group_cache
*cache
;
4200 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
4203 pin_down_extent(root
, cache
, bytenr
, num_bytes
, reserved
);
4205 btrfs_put_block_group(cache
);
4210 * update size of reserved extents. this function may return -EAGAIN
4211 * if 'reserve' is true or 'sinfo' is false.
4213 int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
4214 u64 num_bytes
, int reserve
, int sinfo
)
4218 struct btrfs_space_info
*space_info
= cache
->space_info
;
4219 spin_lock(&space_info
->lock
);
4220 spin_lock(&cache
->lock
);
4225 cache
->reserved
+= num_bytes
;
4226 space_info
->bytes_reserved
+= num_bytes
;
4230 space_info
->bytes_readonly
+= num_bytes
;
4231 cache
->reserved
-= num_bytes
;
4232 space_info
->bytes_reserved
-= num_bytes
;
4233 space_info
->reservation_progress
++;
4235 spin_unlock(&cache
->lock
);
4236 spin_unlock(&space_info
->lock
);
4238 spin_lock(&cache
->lock
);
4243 cache
->reserved
+= num_bytes
;
4245 cache
->reserved
-= num_bytes
;
4247 spin_unlock(&cache
->lock
);
4252 int btrfs_prepare_extent_commit(struct btrfs_trans_handle
*trans
,
4253 struct btrfs_root
*root
)
4255 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4256 struct btrfs_caching_control
*next
;
4257 struct btrfs_caching_control
*caching_ctl
;
4258 struct btrfs_block_group_cache
*cache
;
4260 down_write(&fs_info
->extent_commit_sem
);
4262 list_for_each_entry_safe(caching_ctl
, next
,
4263 &fs_info
->caching_block_groups
, list
) {
4264 cache
= caching_ctl
->block_group
;
4265 if (block_group_cache_done(cache
)) {
4266 cache
->last_byte_to_unpin
= (u64
)-1;
4267 list_del_init(&caching_ctl
->list
);
4268 put_caching_control(caching_ctl
);
4270 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
4274 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
4275 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
4277 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
4279 up_write(&fs_info
->extent_commit_sem
);
4281 update_global_block_rsv(fs_info
);
4285 static int unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
4287 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4288 struct btrfs_block_group_cache
*cache
= NULL
;
4291 while (start
<= end
) {
4293 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
4295 btrfs_put_block_group(cache
);
4296 cache
= btrfs_lookup_block_group(fs_info
, start
);
4300 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
4301 len
= min(len
, end
+ 1 - start
);
4303 if (start
< cache
->last_byte_to_unpin
) {
4304 len
= min(len
, cache
->last_byte_to_unpin
- start
);
4305 btrfs_add_free_space(cache
, start
, len
);
4310 spin_lock(&cache
->space_info
->lock
);
4311 spin_lock(&cache
->lock
);
4312 cache
->pinned
-= len
;
4313 cache
->space_info
->bytes_pinned
-= len
;
4315 cache
->space_info
->bytes_readonly
+= len
;
4316 } else if (cache
->reserved_pinned
> 0) {
4317 len
= min(len
, cache
->reserved_pinned
);
4318 cache
->reserved_pinned
-= len
;
4319 cache
->space_info
->bytes_reserved
+= len
;
4321 spin_unlock(&cache
->lock
);
4322 spin_unlock(&cache
->space_info
->lock
);
4326 btrfs_put_block_group(cache
);
4330 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
4331 struct btrfs_root
*root
)
4333 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4334 struct extent_io_tree
*unpin
;
4335 struct btrfs_block_rsv
*block_rsv
;
4336 struct btrfs_block_rsv
*next_rsv
;
4342 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
4343 unpin
= &fs_info
->freed_extents
[1];
4345 unpin
= &fs_info
->freed_extents
[0];
4348 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
4353 if (btrfs_test_opt(root
, DISCARD
))
4354 ret
= btrfs_discard_extent(root
, start
,
4355 end
+ 1 - start
, NULL
);
4357 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
4358 unpin_extent_range(root
, start
, end
);
4362 mutex_lock(&fs_info
->durable_block_rsv_mutex
);
4363 list_for_each_entry_safe(block_rsv
, next_rsv
,
4364 &fs_info
->durable_block_rsv_list
, list
) {
4366 idx
= trans
->transid
& 0x1;
4367 if (block_rsv
->freed
[idx
] > 0) {
4368 block_rsv_add_bytes(block_rsv
,
4369 block_rsv
->freed
[idx
], 0);
4370 block_rsv
->freed
[idx
] = 0;
4372 if (atomic_read(&block_rsv
->usage
) == 0) {
4373 btrfs_block_rsv_release(root
, block_rsv
, (u64
)-1);
4375 if (block_rsv
->freed
[0] == 0 &&
4376 block_rsv
->freed
[1] == 0) {
4377 list_del_init(&block_rsv
->list
);
4381 btrfs_block_rsv_release(root
, block_rsv
, 0);
4384 mutex_unlock(&fs_info
->durable_block_rsv_mutex
);
4389 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
4390 struct btrfs_root
*root
,
4391 u64 bytenr
, u64 num_bytes
, u64 parent
,
4392 u64 root_objectid
, u64 owner_objectid
,
4393 u64 owner_offset
, int refs_to_drop
,
4394 struct btrfs_delayed_extent_op
*extent_op
)
4396 struct btrfs_key key
;
4397 struct btrfs_path
*path
;
4398 struct btrfs_fs_info
*info
= root
->fs_info
;
4399 struct btrfs_root
*extent_root
= info
->extent_root
;
4400 struct extent_buffer
*leaf
;
4401 struct btrfs_extent_item
*ei
;
4402 struct btrfs_extent_inline_ref
*iref
;
4405 int extent_slot
= 0;
4406 int found_extent
= 0;
4411 path
= btrfs_alloc_path();
4416 path
->leave_spinning
= 1;
4418 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
4419 BUG_ON(!is_data
&& refs_to_drop
!= 1);
4421 ret
= lookup_extent_backref(trans
, extent_root
, path
, &iref
,
4422 bytenr
, num_bytes
, parent
,
4423 root_objectid
, owner_objectid
,
4426 extent_slot
= path
->slots
[0];
4427 while (extent_slot
>= 0) {
4428 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
4430 if (key
.objectid
!= bytenr
)
4432 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
4433 key
.offset
== num_bytes
) {
4437 if (path
->slots
[0] - extent_slot
> 5)
4441 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4442 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
4443 if (found_extent
&& item_size
< sizeof(*ei
))
4446 if (!found_extent
) {
4448 ret
= remove_extent_backref(trans
, extent_root
, path
,
4452 btrfs_release_path(path
);
4453 path
->leave_spinning
= 1;
4455 key
.objectid
= bytenr
;
4456 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
4457 key
.offset
= num_bytes
;
4459 ret
= btrfs_search_slot(trans
, extent_root
,
4462 printk(KERN_ERR
"umm, got %d back from search"
4463 ", was looking for %llu\n", ret
,
4464 (unsigned long long)bytenr
);
4465 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
4468 extent_slot
= path
->slots
[0];
4471 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
4473 printk(KERN_ERR
"btrfs unable to find ref byte nr %llu "
4474 "parent %llu root %llu owner %llu offset %llu\n",
4475 (unsigned long long)bytenr
,
4476 (unsigned long long)parent
,
4477 (unsigned long long)root_objectid
,
4478 (unsigned long long)owner_objectid
,
4479 (unsigned long long)owner_offset
);
4482 leaf
= path
->nodes
[0];
4483 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
4484 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4485 if (item_size
< sizeof(*ei
)) {
4486 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
4487 ret
= convert_extent_item_v0(trans
, extent_root
, path
,
4491 btrfs_release_path(path
);
4492 path
->leave_spinning
= 1;
4494 key
.objectid
= bytenr
;
4495 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
4496 key
.offset
= num_bytes
;
4498 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
4501 printk(KERN_ERR
"umm, got %d back from search"
4502 ", was looking for %llu\n", ret
,
4503 (unsigned long long)bytenr
);
4504 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
4507 extent_slot
= path
->slots
[0];
4508 leaf
= path
->nodes
[0];
4509 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
4512 BUG_ON(item_size
< sizeof(*ei
));
4513 ei
= btrfs_item_ptr(leaf
, extent_slot
,
4514 struct btrfs_extent_item
);
4515 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
) {
4516 struct btrfs_tree_block_info
*bi
;
4517 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
4518 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
4519 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
4522 refs
= btrfs_extent_refs(leaf
, ei
);
4523 BUG_ON(refs
< refs_to_drop
);
4524 refs
-= refs_to_drop
;
4528 __run_delayed_extent_op(extent_op
, leaf
, ei
);
4530 * In the case of inline back ref, reference count will
4531 * be updated by remove_extent_backref
4534 BUG_ON(!found_extent
);
4536 btrfs_set_extent_refs(leaf
, ei
, refs
);
4537 btrfs_mark_buffer_dirty(leaf
);
4540 ret
= remove_extent_backref(trans
, extent_root
, path
,
4547 BUG_ON(is_data
&& refs_to_drop
!=
4548 extent_data_ref_count(root
, path
, iref
));
4550 BUG_ON(path
->slots
[0] != extent_slot
);
4552 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
4553 path
->slots
[0] = extent_slot
;
4558 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
4561 btrfs_release_path(path
);
4564 ret
= btrfs_del_csums(trans
, root
, bytenr
, num_bytes
);
4567 invalidate_mapping_pages(info
->btree_inode
->i_mapping
,
4568 bytenr
>> PAGE_CACHE_SHIFT
,
4569 (bytenr
+ num_bytes
- 1) >> PAGE_CACHE_SHIFT
);
4572 ret
= update_block_group(trans
, root
, bytenr
, num_bytes
, 0);
4575 btrfs_free_path(path
);
4580 * when we free an block, it is possible (and likely) that we free the last
4581 * delayed ref for that extent as well. This searches the delayed ref tree for
4582 * a given extent, and if there are no other delayed refs to be processed, it
4583 * removes it from the tree.
4585 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
4586 struct btrfs_root
*root
, u64 bytenr
)
4588 struct btrfs_delayed_ref_head
*head
;
4589 struct btrfs_delayed_ref_root
*delayed_refs
;
4590 struct btrfs_delayed_ref_node
*ref
;
4591 struct rb_node
*node
;
4594 delayed_refs
= &trans
->transaction
->delayed_refs
;
4595 spin_lock(&delayed_refs
->lock
);
4596 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
4600 node
= rb_prev(&head
->node
.rb_node
);
4604 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
4606 /* there are still entries for this ref, we can't drop it */
4607 if (ref
->bytenr
== bytenr
)
4610 if (head
->extent_op
) {
4611 if (!head
->must_insert_reserved
)
4613 kfree(head
->extent_op
);
4614 head
->extent_op
= NULL
;
4618 * waiting for the lock here would deadlock. If someone else has it
4619 * locked they are already in the process of dropping it anyway
4621 if (!mutex_trylock(&head
->mutex
))
4625 * at this point we have a head with no other entries. Go
4626 * ahead and process it.
4628 head
->node
.in_tree
= 0;
4629 rb_erase(&head
->node
.rb_node
, &delayed_refs
->root
);
4631 delayed_refs
->num_entries
--;
4634 * we don't take a ref on the node because we're removing it from the
4635 * tree, so we just steal the ref the tree was holding.
4637 delayed_refs
->num_heads
--;
4638 if (list_empty(&head
->cluster
))
4639 delayed_refs
->num_heads_ready
--;
4641 list_del_init(&head
->cluster
);
4642 spin_unlock(&delayed_refs
->lock
);
4644 BUG_ON(head
->extent_op
);
4645 if (head
->must_insert_reserved
)
4648 mutex_unlock(&head
->mutex
);
4649 btrfs_put_delayed_ref(&head
->node
);
4652 spin_unlock(&delayed_refs
->lock
);
4656 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
4657 struct btrfs_root
*root
,
4658 struct extent_buffer
*buf
,
4659 u64 parent
, int last_ref
)
4661 struct btrfs_block_rsv
*block_rsv
;
4662 struct btrfs_block_group_cache
*cache
= NULL
;
4665 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
4666 ret
= btrfs_add_delayed_tree_ref(trans
, buf
->start
, buf
->len
,
4667 parent
, root
->root_key
.objectid
,
4668 btrfs_header_level(buf
),
4669 BTRFS_DROP_DELAYED_REF
, NULL
);
4676 block_rsv
= get_block_rsv(trans
, root
);
4677 cache
= btrfs_lookup_block_group(root
->fs_info
, buf
->start
);
4678 if (block_rsv
->space_info
!= cache
->space_info
)
4681 if (btrfs_header_generation(buf
) == trans
->transid
) {
4682 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
4683 ret
= check_ref_cleanup(trans
, root
, buf
->start
);
4688 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
4689 pin_down_extent(root
, cache
, buf
->start
, buf
->len
, 1);
4693 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
4695 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
4696 ret
= btrfs_update_reserved_bytes(cache
, buf
->len
, 0, 0);
4697 if (ret
== -EAGAIN
) {
4698 /* block group became read-only */
4699 btrfs_update_reserved_bytes(cache
, buf
->len
, 0, 1);
4704 spin_lock(&block_rsv
->lock
);
4705 if (block_rsv
->reserved
< block_rsv
->size
) {
4706 block_rsv
->reserved
+= buf
->len
;
4709 spin_unlock(&block_rsv
->lock
);
4712 spin_lock(&cache
->space_info
->lock
);
4713 cache
->space_info
->bytes_reserved
-= buf
->len
;
4714 cache
->space_info
->reservation_progress
++;
4715 spin_unlock(&cache
->space_info
->lock
);
4720 if (block_rsv
->durable
&& !cache
->ro
) {
4722 spin_lock(&cache
->lock
);
4724 cache
->reserved_pinned
+= buf
->len
;
4727 spin_unlock(&cache
->lock
);
4730 spin_lock(&block_rsv
->lock
);
4731 block_rsv
->freed
[trans
->transid
& 0x1] += buf
->len
;
4732 spin_unlock(&block_rsv
->lock
);
4737 * Deleting the buffer, clear the corrupt flag since it doesn't matter
4740 clear_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
);
4741 btrfs_put_block_group(cache
);
4744 int btrfs_free_extent(struct btrfs_trans_handle
*trans
,
4745 struct btrfs_root
*root
,
4746 u64 bytenr
, u64 num_bytes
, u64 parent
,
4747 u64 root_objectid
, u64 owner
, u64 offset
)
4752 * tree log blocks never actually go into the extent allocation
4753 * tree, just update pinning info and exit early.
4755 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
4756 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
4757 /* unlocks the pinned mutex */
4758 btrfs_pin_extent(root
, bytenr
, num_bytes
, 1);
4760 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
4761 ret
= btrfs_add_delayed_tree_ref(trans
, bytenr
, num_bytes
,
4762 parent
, root_objectid
, (int)owner
,
4763 BTRFS_DROP_DELAYED_REF
, NULL
);
4766 ret
= btrfs_add_delayed_data_ref(trans
, bytenr
, num_bytes
,
4767 parent
, root_objectid
, owner
,
4768 offset
, BTRFS_DROP_DELAYED_REF
, NULL
);
4774 static u64
stripe_align(struct btrfs_root
*root
, u64 val
)
4776 u64 mask
= ((u64
)root
->stripesize
- 1);
4777 u64 ret
= (val
+ mask
) & ~mask
;
4782 * when we wait for progress in the block group caching, its because
4783 * our allocation attempt failed at least once. So, we must sleep
4784 * and let some progress happen before we try again.
4786 * This function will sleep at least once waiting for new free space to
4787 * show up, and then it will check the block group free space numbers
4788 * for our min num_bytes. Another option is to have it go ahead
4789 * and look in the rbtree for a free extent of a given size, but this
4793 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
4796 struct btrfs_caching_control
*caching_ctl
;
4799 caching_ctl
= get_caching_control(cache
);
4803 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
4804 (cache
->free_space_ctl
->free_space
>= num_bytes
));
4806 put_caching_control(caching_ctl
);
4811 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
4813 struct btrfs_caching_control
*caching_ctl
;
4816 caching_ctl
= get_caching_control(cache
);
4820 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
4822 put_caching_control(caching_ctl
);
4826 static int get_block_group_index(struct btrfs_block_group_cache
*cache
)
4829 if (cache
->flags
& BTRFS_BLOCK_GROUP_RAID10
)
4831 else if (cache
->flags
& BTRFS_BLOCK_GROUP_RAID1
)
4833 else if (cache
->flags
& BTRFS_BLOCK_GROUP_DUP
)
4835 else if (cache
->flags
& BTRFS_BLOCK_GROUP_RAID0
)
4842 enum btrfs_loop_type
{
4843 LOOP_FIND_IDEAL
= 0,
4844 LOOP_CACHING_NOWAIT
= 1,
4845 LOOP_CACHING_WAIT
= 2,
4846 LOOP_ALLOC_CHUNK
= 3,
4847 LOOP_NO_EMPTY_SIZE
= 4,
4851 * walks the btree of allocated extents and find a hole of a given size.
4852 * The key ins is changed to record the hole:
4853 * ins->objectid == block start
4854 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4855 * ins->offset == number of blocks
4856 * Any available blocks before search_start are skipped.
4858 static noinline
int find_free_extent(struct btrfs_trans_handle
*trans
,
4859 struct btrfs_root
*orig_root
,
4860 u64 num_bytes
, u64 empty_size
,
4861 u64 search_start
, u64 search_end
,
4862 u64 hint_byte
, struct btrfs_key
*ins
,
4866 struct btrfs_root
*root
= orig_root
->fs_info
->extent_root
;
4867 struct btrfs_free_cluster
*last_ptr
= NULL
;
4868 struct btrfs_block_group_cache
*block_group
= NULL
;
4869 int empty_cluster
= 2 * 1024 * 1024;
4870 int allowed_chunk_alloc
= 0;
4871 int done_chunk_alloc
= 0;
4872 struct btrfs_space_info
*space_info
;
4873 int last_ptr_loop
= 0;
4876 bool found_uncached_bg
= false;
4877 bool failed_cluster_refill
= false;
4878 bool failed_alloc
= false;
4879 bool use_cluster
= true;
4880 u64 ideal_cache_percent
= 0;
4881 u64 ideal_cache_offset
= 0;
4883 WARN_ON(num_bytes
< root
->sectorsize
);
4884 btrfs_set_key_type(ins
, BTRFS_EXTENT_ITEM_KEY
);
4888 space_info
= __find_space_info(root
->fs_info
, data
);
4890 printk(KERN_ERR
"No space info for %llu\n", data
);
4895 * If the space info is for both data and metadata it means we have a
4896 * small filesystem and we can't use the clustering stuff.
4898 if (btrfs_mixed_space_info(space_info
))
4899 use_cluster
= false;
4901 if (orig_root
->ref_cows
|| empty_size
)
4902 allowed_chunk_alloc
= 1;
4904 if (data
& BTRFS_BLOCK_GROUP_METADATA
&& use_cluster
) {
4905 last_ptr
= &root
->fs_info
->meta_alloc_cluster
;
4906 if (!btrfs_test_opt(root
, SSD
))
4907 empty_cluster
= 64 * 1024;
4910 if ((data
& BTRFS_BLOCK_GROUP_DATA
) && use_cluster
&&
4911 btrfs_test_opt(root
, SSD
)) {
4912 last_ptr
= &root
->fs_info
->data_alloc_cluster
;
4916 spin_lock(&last_ptr
->lock
);
4917 if (last_ptr
->block_group
)
4918 hint_byte
= last_ptr
->window_start
;
4919 spin_unlock(&last_ptr
->lock
);
4922 search_start
= max(search_start
, first_logical_byte(root
, 0));
4923 search_start
= max(search_start
, hint_byte
);
4928 if (search_start
== hint_byte
) {
4930 block_group
= btrfs_lookup_block_group(root
->fs_info
,
4933 * we don't want to use the block group if it doesn't match our
4934 * allocation bits, or if its not cached.
4936 * However if we are re-searching with an ideal block group
4937 * picked out then we don't care that the block group is cached.
4939 if (block_group
&& block_group_bits(block_group
, data
) &&
4940 (block_group
->cached
!= BTRFS_CACHE_NO
||
4941 search_start
== ideal_cache_offset
)) {
4942 down_read(&space_info
->groups_sem
);
4943 if (list_empty(&block_group
->list
) ||
4946 * someone is removing this block group,
4947 * we can't jump into the have_block_group
4948 * target because our list pointers are not
4951 btrfs_put_block_group(block_group
);
4952 up_read(&space_info
->groups_sem
);
4954 index
= get_block_group_index(block_group
);
4955 goto have_block_group
;
4957 } else if (block_group
) {
4958 btrfs_put_block_group(block_group
);
4962 down_read(&space_info
->groups_sem
);
4963 list_for_each_entry(block_group
, &space_info
->block_groups
[index
],
4968 btrfs_get_block_group(block_group
);
4969 search_start
= block_group
->key
.objectid
;
4972 * this can happen if we end up cycling through all the
4973 * raid types, but we want to make sure we only allocate
4974 * for the proper type.
4976 if (!block_group_bits(block_group
, data
)) {
4977 u64 extra
= BTRFS_BLOCK_GROUP_DUP
|
4978 BTRFS_BLOCK_GROUP_RAID1
|
4979 BTRFS_BLOCK_GROUP_RAID10
;
4982 * if they asked for extra copies and this block group
4983 * doesn't provide them, bail. This does allow us to
4984 * fill raid0 from raid1.
4986 if ((data
& extra
) && !(block_group
->flags
& extra
))
4991 if (unlikely(block_group
->cached
== BTRFS_CACHE_NO
)) {
4994 ret
= cache_block_group(block_group
, trans
,
4996 if (block_group
->cached
== BTRFS_CACHE_FINISHED
)
4997 goto have_block_group
;
4999 free_percent
= btrfs_block_group_used(&block_group
->item
);
5000 free_percent
*= 100;
5001 free_percent
= div64_u64(free_percent
,
5002 block_group
->key
.offset
);
5003 free_percent
= 100 - free_percent
;
5004 if (free_percent
> ideal_cache_percent
&&
5005 likely(!block_group
->ro
)) {
5006 ideal_cache_offset
= block_group
->key
.objectid
;
5007 ideal_cache_percent
= free_percent
;
5011 * The caching workers are limited to 2 threads, so we
5012 * can queue as much work as we care to.
5014 if (loop
> LOOP_FIND_IDEAL
) {
5015 ret
= cache_block_group(block_group
, trans
,
5019 found_uncached_bg
= true;
5022 * If loop is set for cached only, try the next block
5025 if (loop
== LOOP_FIND_IDEAL
)
5029 cached
= block_group_cache_done(block_group
);
5030 if (unlikely(!cached
))
5031 found_uncached_bg
= true;
5033 if (unlikely(block_group
->ro
))
5036 spin_lock(&block_group
->free_space_ctl
->tree_lock
);
5038 block_group
->free_space_ctl
->free_space
<
5039 num_bytes
+ empty_size
) {
5040 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
5043 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
5046 * Ok we want to try and use the cluster allocator, so lets look
5047 * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
5048 * have tried the cluster allocator plenty of times at this
5049 * point and not have found anything, so we are likely way too
5050 * fragmented for the clustering stuff to find anything, so lets
5051 * just skip it and let the allocator find whatever block it can
5054 if (last_ptr
&& loop
< LOOP_NO_EMPTY_SIZE
) {
5056 * the refill lock keeps out other
5057 * people trying to start a new cluster
5059 spin_lock(&last_ptr
->refill_lock
);
5060 if (last_ptr
->block_group
&&
5061 (last_ptr
->block_group
->ro
||
5062 !block_group_bits(last_ptr
->block_group
, data
))) {
5064 goto refill_cluster
;
5067 offset
= btrfs_alloc_from_cluster(block_group
, last_ptr
,
5068 num_bytes
, search_start
);
5070 /* we have a block, we're done */
5071 spin_unlock(&last_ptr
->refill_lock
);
5075 spin_lock(&last_ptr
->lock
);
5077 * whoops, this cluster doesn't actually point to
5078 * this block group. Get a ref on the block
5079 * group is does point to and try again
5081 if (!last_ptr_loop
&& last_ptr
->block_group
&&
5082 last_ptr
->block_group
!= block_group
) {
5084 btrfs_put_block_group(block_group
);
5085 block_group
= last_ptr
->block_group
;
5086 btrfs_get_block_group(block_group
);
5087 spin_unlock(&last_ptr
->lock
);
5088 spin_unlock(&last_ptr
->refill_lock
);
5091 search_start
= block_group
->key
.objectid
;
5093 * we know this block group is properly
5094 * in the list because
5095 * btrfs_remove_block_group, drops the
5096 * cluster before it removes the block
5097 * group from the list
5099 goto have_block_group
;
5101 spin_unlock(&last_ptr
->lock
);
5104 * this cluster didn't work out, free it and
5107 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
5111 /* allocate a cluster in this block group */
5112 ret
= btrfs_find_space_cluster(trans
, root
,
5113 block_group
, last_ptr
,
5115 empty_cluster
+ empty_size
);
5118 * now pull our allocation out of this
5121 offset
= btrfs_alloc_from_cluster(block_group
,
5122 last_ptr
, num_bytes
,
5125 /* we found one, proceed */
5126 spin_unlock(&last_ptr
->refill_lock
);
5129 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
5130 && !failed_cluster_refill
) {
5131 spin_unlock(&last_ptr
->refill_lock
);
5133 failed_cluster_refill
= true;
5134 wait_block_group_cache_progress(block_group
,
5135 num_bytes
+ empty_cluster
+ empty_size
);
5136 goto have_block_group
;
5140 * at this point we either didn't find a cluster
5141 * or we weren't able to allocate a block from our
5142 * cluster. Free the cluster we've been trying
5143 * to use, and go to the next block group
5145 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
5146 spin_unlock(&last_ptr
->refill_lock
);
5150 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
5151 num_bytes
, empty_size
);
5153 * If we didn't find a chunk, and we haven't failed on this
5154 * block group before, and this block group is in the middle of
5155 * caching and we are ok with waiting, then go ahead and wait
5156 * for progress to be made, and set failed_alloc to true.
5158 * If failed_alloc is true then we've already waited on this
5159 * block group once and should move on to the next block group.
5161 if (!offset
&& !failed_alloc
&& !cached
&&
5162 loop
> LOOP_CACHING_NOWAIT
) {
5163 wait_block_group_cache_progress(block_group
,
5164 num_bytes
+ empty_size
);
5165 failed_alloc
= true;
5166 goto have_block_group
;
5167 } else if (!offset
) {
5171 search_start
= stripe_align(root
, offset
);
5172 /* move on to the next group */
5173 if (search_start
+ num_bytes
>= search_end
) {
5174 btrfs_add_free_space(block_group
, offset
, num_bytes
);
5178 /* move on to the next group */
5179 if (search_start
+ num_bytes
>
5180 block_group
->key
.objectid
+ block_group
->key
.offset
) {
5181 btrfs_add_free_space(block_group
, offset
, num_bytes
);
5185 ins
->objectid
= search_start
;
5186 ins
->offset
= num_bytes
;
5188 if (offset
< search_start
)
5189 btrfs_add_free_space(block_group
, offset
,
5190 search_start
- offset
);
5191 BUG_ON(offset
> search_start
);
5193 ret
= btrfs_update_reserved_bytes(block_group
, num_bytes
, 1,
5194 (data
& BTRFS_BLOCK_GROUP_DATA
));
5195 if (ret
== -EAGAIN
) {
5196 btrfs_add_free_space(block_group
, offset
, num_bytes
);
5200 /* we are all good, lets return */
5201 ins
->objectid
= search_start
;
5202 ins
->offset
= num_bytes
;
5204 if (offset
< search_start
)
5205 btrfs_add_free_space(block_group
, offset
,
5206 search_start
- offset
);
5207 BUG_ON(offset
> search_start
);
5208 btrfs_put_block_group(block_group
);
5211 failed_cluster_refill
= false;
5212 failed_alloc
= false;
5213 BUG_ON(index
!= get_block_group_index(block_group
));
5214 btrfs_put_block_group(block_group
);
5216 up_read(&space_info
->groups_sem
);
5218 if (!ins
->objectid
&& ++index
< BTRFS_NR_RAID_TYPES
)
5221 /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5222 * for them to make caching progress. Also
5223 * determine the best possible bg to cache
5224 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5225 * caching kthreads as we move along
5226 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5227 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5228 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5231 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
) {
5233 if (loop
== LOOP_FIND_IDEAL
&& found_uncached_bg
) {
5234 found_uncached_bg
= false;
5236 if (!ideal_cache_percent
)
5240 * 1 of the following 2 things have happened so far
5242 * 1) We found an ideal block group for caching that
5243 * is mostly full and will cache quickly, so we might
5244 * as well wait for it.
5246 * 2) We searched for cached only and we didn't find
5247 * anything, and we didn't start any caching kthreads
5248 * either, so chances are we will loop through and
5249 * start a couple caching kthreads, and then come back
5250 * around and just wait for them. This will be slower
5251 * because we will have 2 caching kthreads reading at
5252 * the same time when we could have just started one
5253 * and waited for it to get far enough to give us an
5254 * allocation, so go ahead and go to the wait caching
5257 loop
= LOOP_CACHING_WAIT
;
5258 search_start
= ideal_cache_offset
;
5259 ideal_cache_percent
= 0;
5261 } else if (loop
== LOOP_FIND_IDEAL
) {
5263 * Didn't find a uncached bg, wait on anything we find
5266 loop
= LOOP_CACHING_WAIT
;
5272 if (loop
== LOOP_ALLOC_CHUNK
) {
5273 if (allowed_chunk_alloc
) {
5274 ret
= do_chunk_alloc(trans
, root
, num_bytes
+
5275 2 * 1024 * 1024, data
,
5276 CHUNK_ALLOC_LIMITED
);
5277 allowed_chunk_alloc
= 0;
5279 done_chunk_alloc
= 1;
5280 } else if (!done_chunk_alloc
&&
5281 space_info
->force_alloc
==
5282 CHUNK_ALLOC_NO_FORCE
) {
5283 space_info
->force_alloc
= CHUNK_ALLOC_LIMITED
;
5287 * We didn't allocate a chunk, go ahead and drop the
5288 * empty size and loop again.
5290 if (!done_chunk_alloc
)
5291 loop
= LOOP_NO_EMPTY_SIZE
;
5294 if (loop
== LOOP_NO_EMPTY_SIZE
) {
5300 } else if (!ins
->objectid
) {
5302 } else if (ins
->objectid
) {
5309 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
5310 int dump_block_groups
)
5312 struct btrfs_block_group_cache
*cache
;
5315 spin_lock(&info
->lock
);
5316 printk(KERN_INFO
"space_info has %llu free, is %sfull\n",
5317 (unsigned long long)(info
->total_bytes
- info
->bytes_used
-
5318 info
->bytes_pinned
- info
->bytes_reserved
-
5319 info
->bytes_readonly
),
5320 (info
->full
) ? "" : "not ");
5321 printk(KERN_INFO
"space_info total=%llu, used=%llu, pinned=%llu, "
5322 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5323 (unsigned long long)info
->total_bytes
,
5324 (unsigned long long)info
->bytes_used
,
5325 (unsigned long long)info
->bytes_pinned
,
5326 (unsigned long long)info
->bytes_reserved
,
5327 (unsigned long long)info
->bytes_may_use
,
5328 (unsigned long long)info
->bytes_readonly
);
5329 spin_unlock(&info
->lock
);
5331 if (!dump_block_groups
)
5334 down_read(&info
->groups_sem
);
5336 list_for_each_entry(cache
, &info
->block_groups
[index
], list
) {
5337 spin_lock(&cache
->lock
);
5338 printk(KERN_INFO
"block group %llu has %llu bytes, %llu used "
5339 "%llu pinned %llu reserved\n",
5340 (unsigned long long)cache
->key
.objectid
,
5341 (unsigned long long)cache
->key
.offset
,
5342 (unsigned long long)btrfs_block_group_used(&cache
->item
),
5343 (unsigned long long)cache
->pinned
,
5344 (unsigned long long)cache
->reserved
);
5345 btrfs_dump_free_space(cache
, bytes
);
5346 spin_unlock(&cache
->lock
);
5348 if (++index
< BTRFS_NR_RAID_TYPES
)
5350 up_read(&info
->groups_sem
);
5353 int btrfs_reserve_extent(struct btrfs_trans_handle
*trans
,
5354 struct btrfs_root
*root
,
5355 u64 num_bytes
, u64 min_alloc_size
,
5356 u64 empty_size
, u64 hint_byte
,
5357 u64 search_end
, struct btrfs_key
*ins
,
5361 u64 search_start
= 0;
5363 data
= btrfs_get_alloc_profile(root
, data
);
5366 * the only place that sets empty_size is btrfs_realloc_node, which
5367 * is not called recursively on allocations
5369 if (empty_size
|| root
->ref_cows
)
5370 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
5371 num_bytes
+ 2 * 1024 * 1024, data
,
5372 CHUNK_ALLOC_NO_FORCE
);
5374 WARN_ON(num_bytes
< root
->sectorsize
);
5375 ret
= find_free_extent(trans
, root
, num_bytes
, empty_size
,
5376 search_start
, search_end
, hint_byte
,
5379 if (ret
== -ENOSPC
&& num_bytes
> min_alloc_size
) {
5380 num_bytes
= num_bytes
>> 1;
5381 num_bytes
= num_bytes
& ~(root
->sectorsize
- 1);
5382 num_bytes
= max(num_bytes
, min_alloc_size
);
5383 do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
5384 num_bytes
, data
, CHUNK_ALLOC_FORCE
);
5387 if (ret
== -ENOSPC
&& btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
5388 struct btrfs_space_info
*sinfo
;
5390 sinfo
= __find_space_info(root
->fs_info
, data
);
5391 printk(KERN_ERR
"btrfs allocation failed flags %llu, "
5392 "wanted %llu\n", (unsigned long long)data
,
5393 (unsigned long long)num_bytes
);
5394 dump_space_info(sinfo
, num_bytes
, 1);
5397 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, ins
->offset
);
5402 int btrfs_free_reserved_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
5404 struct btrfs_block_group_cache
*cache
;
5407 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
5409 printk(KERN_ERR
"Unable to find block group for %llu\n",
5410 (unsigned long long)start
);
5414 if (btrfs_test_opt(root
, DISCARD
))
5415 ret
= btrfs_discard_extent(root
, start
, len
, NULL
);
5417 btrfs_add_free_space(cache
, start
, len
);
5418 btrfs_update_reserved_bytes(cache
, len
, 0, 1);
5419 btrfs_put_block_group(cache
);
5421 trace_btrfs_reserved_extent_free(root
, start
, len
);
5426 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
5427 struct btrfs_root
*root
,
5428 u64 parent
, u64 root_objectid
,
5429 u64 flags
, u64 owner
, u64 offset
,
5430 struct btrfs_key
*ins
, int ref_mod
)
5433 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5434 struct btrfs_extent_item
*extent_item
;
5435 struct btrfs_extent_inline_ref
*iref
;
5436 struct btrfs_path
*path
;
5437 struct extent_buffer
*leaf
;
5442 type
= BTRFS_SHARED_DATA_REF_KEY
;
5444 type
= BTRFS_EXTENT_DATA_REF_KEY
;
5446 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
5448 path
= btrfs_alloc_path();
5452 path
->leave_spinning
= 1;
5453 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
5457 leaf
= path
->nodes
[0];
5458 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
5459 struct btrfs_extent_item
);
5460 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
5461 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
5462 btrfs_set_extent_flags(leaf
, extent_item
,
5463 flags
| BTRFS_EXTENT_FLAG_DATA
);
5465 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
5466 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
5468 struct btrfs_shared_data_ref
*ref
;
5469 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
5470 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
5471 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
5473 struct btrfs_extent_data_ref
*ref
;
5474 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
5475 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
5476 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
5477 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
5478 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
5481 btrfs_mark_buffer_dirty(path
->nodes
[0]);
5482 btrfs_free_path(path
);
5484 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1);
5486 printk(KERN_ERR
"btrfs update block group failed for %llu "
5487 "%llu\n", (unsigned long long)ins
->objectid
,
5488 (unsigned long long)ins
->offset
);
5494 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
5495 struct btrfs_root
*root
,
5496 u64 parent
, u64 root_objectid
,
5497 u64 flags
, struct btrfs_disk_key
*key
,
5498 int level
, struct btrfs_key
*ins
)
5501 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5502 struct btrfs_extent_item
*extent_item
;
5503 struct btrfs_tree_block_info
*block_info
;
5504 struct btrfs_extent_inline_ref
*iref
;
5505 struct btrfs_path
*path
;
5506 struct extent_buffer
*leaf
;
5507 u32 size
= sizeof(*extent_item
) + sizeof(*block_info
) + sizeof(*iref
);
5509 path
= btrfs_alloc_path();
5513 path
->leave_spinning
= 1;
5514 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
5518 leaf
= path
->nodes
[0];
5519 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
5520 struct btrfs_extent_item
);
5521 btrfs_set_extent_refs(leaf
, extent_item
, 1);
5522 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
5523 btrfs_set_extent_flags(leaf
, extent_item
,
5524 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
5525 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
5527 btrfs_set_tree_block_key(leaf
, block_info
, key
);
5528 btrfs_set_tree_block_level(leaf
, block_info
, level
);
5530 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
5532 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
5533 btrfs_set_extent_inline_ref_type(leaf
, iref
,
5534 BTRFS_SHARED_BLOCK_REF_KEY
);
5535 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
5537 btrfs_set_extent_inline_ref_type(leaf
, iref
,
5538 BTRFS_TREE_BLOCK_REF_KEY
);
5539 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
5542 btrfs_mark_buffer_dirty(leaf
);
5543 btrfs_free_path(path
);
5545 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1);
5547 printk(KERN_ERR
"btrfs update block group failed for %llu "
5548 "%llu\n", (unsigned long long)ins
->objectid
,
5549 (unsigned long long)ins
->offset
);
5555 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
5556 struct btrfs_root
*root
,
5557 u64 root_objectid
, u64 owner
,
5558 u64 offset
, struct btrfs_key
*ins
)
5562 BUG_ON(root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
5564 ret
= btrfs_add_delayed_data_ref(trans
, ins
->objectid
, ins
->offset
,
5565 0, root_objectid
, owner
, offset
,
5566 BTRFS_ADD_DELAYED_EXTENT
, NULL
);
5571 * this is used by the tree logging recovery code. It records that
5572 * an extent has been allocated and makes sure to clear the free
5573 * space cache bits as well
5575 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
5576 struct btrfs_root
*root
,
5577 u64 root_objectid
, u64 owner
, u64 offset
,
5578 struct btrfs_key
*ins
)
5581 struct btrfs_block_group_cache
*block_group
;
5582 struct btrfs_caching_control
*caching_ctl
;
5583 u64 start
= ins
->objectid
;
5584 u64 num_bytes
= ins
->offset
;
5586 block_group
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
5587 cache_block_group(block_group
, trans
, NULL
, 0);
5588 caching_ctl
= get_caching_control(block_group
);
5591 BUG_ON(!block_group_cache_done(block_group
));
5592 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
5595 mutex_lock(&caching_ctl
->mutex
);
5597 if (start
>= caching_ctl
->progress
) {
5598 ret
= add_excluded_extent(root
, start
, num_bytes
);
5600 } else if (start
+ num_bytes
<= caching_ctl
->progress
) {
5601 ret
= btrfs_remove_free_space(block_group
,
5605 num_bytes
= caching_ctl
->progress
- start
;
5606 ret
= btrfs_remove_free_space(block_group
,
5610 start
= caching_ctl
->progress
;
5611 num_bytes
= ins
->objectid
+ ins
->offset
-
5612 caching_ctl
->progress
;
5613 ret
= add_excluded_extent(root
, start
, num_bytes
);
5617 mutex_unlock(&caching_ctl
->mutex
);
5618 put_caching_control(caching_ctl
);
5621 ret
= btrfs_update_reserved_bytes(block_group
, ins
->offset
, 1, 1);
5623 btrfs_put_block_group(block_group
);
5624 ret
= alloc_reserved_file_extent(trans
, root
, 0, root_objectid
,
5625 0, owner
, offset
, ins
, 1);
5629 struct extent_buffer
*btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
,
5630 struct btrfs_root
*root
,
5631 u64 bytenr
, u32 blocksize
,
5634 struct extent_buffer
*buf
;
5636 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
5638 return ERR_PTR(-ENOMEM
);
5639 btrfs_set_header_generation(buf
, trans
->transid
);
5640 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, buf
, level
);
5641 btrfs_tree_lock(buf
);
5642 clean_tree_block(trans
, root
, buf
);
5644 btrfs_set_lock_blocking(buf
);
5645 btrfs_set_buffer_uptodate(buf
);
5647 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
5649 * we allow two log transactions at a time, use different
5650 * EXENT bit to differentiate dirty pages.
5652 if (root
->log_transid
% 2 == 0)
5653 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
5654 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
5656 set_extent_new(&root
->dirty_log_pages
, buf
->start
,
5657 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
5659 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
5660 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
5662 trans
->blocks_used
++;
5663 /* this returns a buffer locked for blocking */
5667 static struct btrfs_block_rsv
*
5668 use_block_rsv(struct btrfs_trans_handle
*trans
,
5669 struct btrfs_root
*root
, u32 blocksize
)
5671 struct btrfs_block_rsv
*block_rsv
;
5672 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
5675 block_rsv
= get_block_rsv(trans
, root
);
5677 if (block_rsv
->size
== 0) {
5678 ret
= reserve_metadata_bytes(trans
, root
, block_rsv
,
5681 * If we couldn't reserve metadata bytes try and use some from
5682 * the global reserve.
5684 if (ret
&& block_rsv
!= global_rsv
) {
5685 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
5688 return ERR_PTR(ret
);
5690 return ERR_PTR(ret
);
5695 ret
= block_rsv_use_bytes(block_rsv
, blocksize
);
5700 ret
= reserve_metadata_bytes(trans
, root
, block_rsv
, blocksize
,
5703 spin_lock(&block_rsv
->lock
);
5704 block_rsv
->size
+= blocksize
;
5705 spin_unlock(&block_rsv
->lock
);
5707 } else if (ret
&& block_rsv
!= global_rsv
) {
5708 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
5714 return ERR_PTR(-ENOSPC
);
5717 static void unuse_block_rsv(struct btrfs_block_rsv
*block_rsv
, u32 blocksize
)
5719 block_rsv_add_bytes(block_rsv
, blocksize
, 0);
5720 block_rsv_release_bytes(block_rsv
, NULL
, 0);
5724 * finds a free extent and does all the dirty work required for allocation
5725 * returns the key for the extent through ins, and a tree buffer for
5726 * the first block of the extent through buf.
5728 * returns the tree buffer or NULL.
5730 struct extent_buffer
*btrfs_alloc_free_block(struct btrfs_trans_handle
*trans
,
5731 struct btrfs_root
*root
, u32 blocksize
,
5732 u64 parent
, u64 root_objectid
,
5733 struct btrfs_disk_key
*key
, int level
,
5734 u64 hint
, u64 empty_size
)
5736 struct btrfs_key ins
;
5737 struct btrfs_block_rsv
*block_rsv
;
5738 struct extent_buffer
*buf
;
5743 block_rsv
= use_block_rsv(trans
, root
, blocksize
);
5744 if (IS_ERR(block_rsv
))
5745 return ERR_CAST(block_rsv
);
5747 ret
= btrfs_reserve_extent(trans
, root
, blocksize
, blocksize
,
5748 empty_size
, hint
, (u64
)-1, &ins
, 0);
5750 unuse_block_rsv(block_rsv
, blocksize
);
5751 return ERR_PTR(ret
);
5754 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
,
5756 BUG_ON(IS_ERR(buf
));
5758 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
5760 parent
= ins
.objectid
;
5761 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
5765 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
5766 struct btrfs_delayed_extent_op
*extent_op
;
5767 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
5770 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
5772 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
5773 extent_op
->flags_to_set
= flags
;
5774 extent_op
->update_key
= 1;
5775 extent_op
->update_flags
= 1;
5776 extent_op
->is_data
= 0;
5778 ret
= btrfs_add_delayed_tree_ref(trans
, ins
.objectid
,
5779 ins
.offset
, parent
, root_objectid
,
5780 level
, BTRFS_ADD_DELAYED_EXTENT
,
5787 struct walk_control
{
5788 u64 refs
[BTRFS_MAX_LEVEL
];
5789 u64 flags
[BTRFS_MAX_LEVEL
];
5790 struct btrfs_key update_progress
;
5800 #define DROP_REFERENCE 1
5801 #define UPDATE_BACKREF 2
5803 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
5804 struct btrfs_root
*root
,
5805 struct walk_control
*wc
,
5806 struct btrfs_path
*path
)
5814 struct btrfs_key key
;
5815 struct extent_buffer
*eb
;
5820 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
5821 wc
->reada_count
= wc
->reada_count
* 2 / 3;
5822 wc
->reada_count
= max(wc
->reada_count
, 2);
5824 wc
->reada_count
= wc
->reada_count
* 3 / 2;
5825 wc
->reada_count
= min_t(int, wc
->reada_count
,
5826 BTRFS_NODEPTRS_PER_BLOCK(root
));
5829 eb
= path
->nodes
[wc
->level
];
5830 nritems
= btrfs_header_nritems(eb
);
5831 blocksize
= btrfs_level_size(root
, wc
->level
- 1);
5833 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
5834 if (nread
>= wc
->reada_count
)
5838 bytenr
= btrfs_node_blockptr(eb
, slot
);
5839 generation
= btrfs_node_ptr_generation(eb
, slot
);
5841 if (slot
== path
->slots
[wc
->level
])
5844 if (wc
->stage
== UPDATE_BACKREF
&&
5845 generation
<= root
->root_key
.offset
)
5848 /* We don't lock the tree block, it's OK to be racy here */
5849 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, blocksize
,
5854 if (wc
->stage
== DROP_REFERENCE
) {
5858 if (wc
->level
== 1 &&
5859 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
5861 if (!wc
->update_ref
||
5862 generation
<= root
->root_key
.offset
)
5864 btrfs_node_key_to_cpu(eb
, &key
, slot
);
5865 ret
= btrfs_comp_cpu_keys(&key
,
5866 &wc
->update_progress
);
5870 if (wc
->level
== 1 &&
5871 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
5875 ret
= readahead_tree_block(root
, bytenr
, blocksize
,
5881 wc
->reada_slot
= slot
;
5885 * hepler to process tree block while walking down the tree.
5887 * when wc->stage == UPDATE_BACKREF, this function updates
5888 * back refs for pointers in the block.
5890 * NOTE: return value 1 means we should stop walking down.
5892 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
5893 struct btrfs_root
*root
,
5894 struct btrfs_path
*path
,
5895 struct walk_control
*wc
, int lookup_info
)
5897 int level
= wc
->level
;
5898 struct extent_buffer
*eb
= path
->nodes
[level
];
5899 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
5902 if (wc
->stage
== UPDATE_BACKREF
&&
5903 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
5907 * when reference count of tree block is 1, it won't increase
5908 * again. once full backref flag is set, we never clear it.
5911 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
5912 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
5913 BUG_ON(!path
->locks
[level
]);
5914 ret
= btrfs_lookup_extent_info(trans
, root
,
5919 BUG_ON(wc
->refs
[level
] == 0);
5922 if (wc
->stage
== DROP_REFERENCE
) {
5923 if (wc
->refs
[level
] > 1)
5926 if (path
->locks
[level
] && !wc
->keep_locks
) {
5927 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
5928 path
->locks
[level
] = 0;
5933 /* wc->stage == UPDATE_BACKREF */
5934 if (!(wc
->flags
[level
] & flag
)) {
5935 BUG_ON(!path
->locks
[level
]);
5936 ret
= btrfs_inc_ref(trans
, root
, eb
, 1);
5938 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
5940 ret
= btrfs_set_disk_extent_flags(trans
, root
, eb
->start
,
5943 wc
->flags
[level
] |= flag
;
5947 * the block is shared by multiple trees, so it's not good to
5948 * keep the tree lock
5950 if (path
->locks
[level
] && level
> 0) {
5951 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
5952 path
->locks
[level
] = 0;
5958 * hepler to process tree block pointer.
5960 * when wc->stage == DROP_REFERENCE, this function checks
5961 * reference count of the block pointed to. if the block
5962 * is shared and we need update back refs for the subtree
5963 * rooted at the block, this function changes wc->stage to
5964 * UPDATE_BACKREF. if the block is shared and there is no
5965 * need to update back, this function drops the reference
5968 * NOTE: return value 1 means we should stop walking down.
5970 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
5971 struct btrfs_root
*root
,
5972 struct btrfs_path
*path
,
5973 struct walk_control
*wc
, int *lookup_info
)
5979 struct btrfs_key key
;
5980 struct extent_buffer
*next
;
5981 int level
= wc
->level
;
5985 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
5986 path
->slots
[level
]);
5988 * if the lower level block was created before the snapshot
5989 * was created, we know there is no need to update back refs
5992 if (wc
->stage
== UPDATE_BACKREF
&&
5993 generation
<= root
->root_key
.offset
) {
5998 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
5999 blocksize
= btrfs_level_size(root
, level
- 1);
6001 next
= btrfs_find_tree_block(root
, bytenr
, blocksize
);
6003 next
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
6008 btrfs_tree_lock(next
);
6009 btrfs_set_lock_blocking(next
);
6011 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, blocksize
,
6012 &wc
->refs
[level
- 1],
6013 &wc
->flags
[level
- 1]);
6015 BUG_ON(wc
->refs
[level
- 1] == 0);
6018 if (wc
->stage
== DROP_REFERENCE
) {
6019 if (wc
->refs
[level
- 1] > 1) {
6021 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
6024 if (!wc
->update_ref
||
6025 generation
<= root
->root_key
.offset
)
6028 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
6029 path
->slots
[level
]);
6030 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
6034 wc
->stage
= UPDATE_BACKREF
;
6035 wc
->shared_level
= level
- 1;
6039 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
6043 if (!btrfs_buffer_uptodate(next
, generation
)) {
6044 btrfs_tree_unlock(next
);
6045 free_extent_buffer(next
);
6051 if (reada
&& level
== 1)
6052 reada_walk_down(trans
, root
, wc
, path
);
6053 next
= read_tree_block(root
, bytenr
, blocksize
, generation
);
6056 btrfs_tree_lock(next
);
6057 btrfs_set_lock_blocking(next
);
6061 BUG_ON(level
!= btrfs_header_level(next
));
6062 path
->nodes
[level
] = next
;
6063 path
->slots
[level
] = 0;
6064 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
6070 wc
->refs
[level
- 1] = 0;
6071 wc
->flags
[level
- 1] = 0;
6072 if (wc
->stage
== DROP_REFERENCE
) {
6073 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
6074 parent
= path
->nodes
[level
]->start
;
6076 BUG_ON(root
->root_key
.objectid
!=
6077 btrfs_header_owner(path
->nodes
[level
]));
6081 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
, parent
,
6082 root
->root_key
.objectid
, level
- 1, 0);
6085 btrfs_tree_unlock(next
);
6086 free_extent_buffer(next
);
6092 * hepler to process tree block while walking up the tree.
6094 * when wc->stage == DROP_REFERENCE, this function drops
6095 * reference count on the block.
6097 * when wc->stage == UPDATE_BACKREF, this function changes
6098 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6099 * to UPDATE_BACKREF previously while processing the block.
6101 * NOTE: return value 1 means we should stop walking up.
6103 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
6104 struct btrfs_root
*root
,
6105 struct btrfs_path
*path
,
6106 struct walk_control
*wc
)
6109 int level
= wc
->level
;
6110 struct extent_buffer
*eb
= path
->nodes
[level
];
6113 if (wc
->stage
== UPDATE_BACKREF
) {
6114 BUG_ON(wc
->shared_level
< level
);
6115 if (level
< wc
->shared_level
)
6118 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
6122 wc
->stage
= DROP_REFERENCE
;
6123 wc
->shared_level
= -1;
6124 path
->slots
[level
] = 0;
6127 * check reference count again if the block isn't locked.
6128 * we should start walking down the tree again if reference
6131 if (!path
->locks
[level
]) {
6133 btrfs_tree_lock(eb
);
6134 btrfs_set_lock_blocking(eb
);
6135 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
6137 ret
= btrfs_lookup_extent_info(trans
, root
,
6142 BUG_ON(wc
->refs
[level
] == 0);
6143 if (wc
->refs
[level
] == 1) {
6144 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
6150 /* wc->stage == DROP_REFERENCE */
6151 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
6153 if (wc
->refs
[level
] == 1) {
6155 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
6156 ret
= btrfs_dec_ref(trans
, root
, eb
, 1);
6158 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
6161 /* make block locked assertion in clean_tree_block happy */
6162 if (!path
->locks
[level
] &&
6163 btrfs_header_generation(eb
) == trans
->transid
) {
6164 btrfs_tree_lock(eb
);
6165 btrfs_set_lock_blocking(eb
);
6166 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
6168 clean_tree_block(trans
, root
, eb
);
6171 if (eb
== root
->node
) {
6172 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
6175 BUG_ON(root
->root_key
.objectid
!=
6176 btrfs_header_owner(eb
));
6178 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
6179 parent
= path
->nodes
[level
+ 1]->start
;
6181 BUG_ON(root
->root_key
.objectid
!=
6182 btrfs_header_owner(path
->nodes
[level
+ 1]));
6185 btrfs_free_tree_block(trans
, root
, eb
, parent
, wc
->refs
[level
] == 1);
6187 wc
->refs
[level
] = 0;
6188 wc
->flags
[level
] = 0;
6192 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
6193 struct btrfs_root
*root
,
6194 struct btrfs_path
*path
,
6195 struct walk_control
*wc
)
6197 int level
= wc
->level
;
6198 int lookup_info
= 1;
6201 while (level
>= 0) {
6202 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
6209 if (path
->slots
[level
] >=
6210 btrfs_header_nritems(path
->nodes
[level
]))
6213 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
6215 path
->slots
[level
]++;
6224 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
6225 struct btrfs_root
*root
,
6226 struct btrfs_path
*path
,
6227 struct walk_control
*wc
, int max_level
)
6229 int level
= wc
->level
;
6232 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
6233 while (level
< max_level
&& path
->nodes
[level
]) {
6235 if (path
->slots
[level
] + 1 <
6236 btrfs_header_nritems(path
->nodes
[level
])) {
6237 path
->slots
[level
]++;
6240 ret
= walk_up_proc(trans
, root
, path
, wc
);
6244 if (path
->locks
[level
]) {
6245 btrfs_tree_unlock_rw(path
->nodes
[level
],
6246 path
->locks
[level
]);
6247 path
->locks
[level
] = 0;
6249 free_extent_buffer(path
->nodes
[level
]);
6250 path
->nodes
[level
] = NULL
;
6258 * drop a subvolume tree.
6260 * this function traverses the tree freeing any blocks that only
6261 * referenced by the tree.
6263 * when a shared tree block is found. this function decreases its
6264 * reference count by one. if update_ref is true, this function
6265 * also make sure backrefs for the shared block and all lower level
6266 * blocks are properly updated.
6268 int btrfs_drop_snapshot(struct btrfs_root
*root
,
6269 struct btrfs_block_rsv
*block_rsv
, int update_ref
)
6271 struct btrfs_path
*path
;
6272 struct btrfs_trans_handle
*trans
;
6273 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
6274 struct btrfs_root_item
*root_item
= &root
->root_item
;
6275 struct walk_control
*wc
;
6276 struct btrfs_key key
;
6281 path
= btrfs_alloc_path();
6285 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
6287 btrfs_free_path(path
);
6291 trans
= btrfs_start_transaction(tree_root
, 0);
6292 BUG_ON(IS_ERR(trans
));
6295 trans
->block_rsv
= block_rsv
;
6297 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
6298 level
= btrfs_header_level(root
->node
);
6299 path
->nodes
[level
] = btrfs_lock_root_node(root
);
6300 btrfs_set_lock_blocking(path
->nodes
[level
]);
6301 path
->slots
[level
] = 0;
6302 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
6303 memset(&wc
->update_progress
, 0,
6304 sizeof(wc
->update_progress
));
6306 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
6307 memcpy(&wc
->update_progress
, &key
,
6308 sizeof(wc
->update_progress
));
6310 level
= root_item
->drop_level
;
6312 path
->lowest_level
= level
;
6313 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6314 path
->lowest_level
= 0;
6322 * unlock our path, this is safe because only this
6323 * function is allowed to delete this snapshot
6325 btrfs_unlock_up_safe(path
, 0);
6327 level
= btrfs_header_level(root
->node
);
6329 btrfs_tree_lock(path
->nodes
[level
]);
6330 btrfs_set_lock_blocking(path
->nodes
[level
]);
6332 ret
= btrfs_lookup_extent_info(trans
, root
,
6333 path
->nodes
[level
]->start
,
6334 path
->nodes
[level
]->len
,
6338 BUG_ON(wc
->refs
[level
] == 0);
6340 if (level
== root_item
->drop_level
)
6343 btrfs_tree_unlock(path
->nodes
[level
]);
6344 WARN_ON(wc
->refs
[level
] != 1);
6350 wc
->shared_level
= -1;
6351 wc
->stage
= DROP_REFERENCE
;
6352 wc
->update_ref
= update_ref
;
6354 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
6357 ret
= walk_down_tree(trans
, root
, path
, wc
);
6363 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
6370 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
6374 if (wc
->stage
== DROP_REFERENCE
) {
6376 btrfs_node_key(path
->nodes
[level
],
6377 &root_item
->drop_progress
,
6378 path
->slots
[level
]);
6379 root_item
->drop_level
= level
;
6382 BUG_ON(wc
->level
== 0);
6383 if (btrfs_should_end_transaction(trans
, tree_root
)) {
6384 ret
= btrfs_update_root(trans
, tree_root
,
6389 btrfs_end_transaction_throttle(trans
, tree_root
);
6390 trans
= btrfs_start_transaction(tree_root
, 0);
6391 BUG_ON(IS_ERR(trans
));
6393 trans
->block_rsv
= block_rsv
;
6396 btrfs_release_path(path
);
6399 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
6402 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
6403 ret
= btrfs_find_last_root(tree_root
, root
->root_key
.objectid
,
6407 /* if we fail to delete the orphan item this time
6408 * around, it'll get picked up the next time.
6410 * The most common failure here is just -ENOENT.
6412 btrfs_del_orphan_item(trans
, tree_root
,
6413 root
->root_key
.objectid
);
6417 if (root
->in_radix
) {
6418 btrfs_free_fs_root(tree_root
->fs_info
, root
);
6420 free_extent_buffer(root
->node
);
6421 free_extent_buffer(root
->commit_root
);
6425 btrfs_end_transaction_throttle(trans
, tree_root
);
6427 btrfs_free_path(path
);
6432 * drop subtree rooted at tree block 'node'.
6434 * NOTE: this function will unlock and release tree block 'node'
6436 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
6437 struct btrfs_root
*root
,
6438 struct extent_buffer
*node
,
6439 struct extent_buffer
*parent
)
6441 struct btrfs_path
*path
;
6442 struct walk_control
*wc
;
6448 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
6450 path
= btrfs_alloc_path();
6454 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
6456 btrfs_free_path(path
);
6460 btrfs_assert_tree_locked(parent
);
6461 parent_level
= btrfs_header_level(parent
);
6462 extent_buffer_get(parent
);
6463 path
->nodes
[parent_level
] = parent
;
6464 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
6466 btrfs_assert_tree_locked(node
);
6467 level
= btrfs_header_level(node
);
6468 path
->nodes
[level
] = node
;
6469 path
->slots
[level
] = 0;
6470 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
6472 wc
->refs
[parent_level
] = 1;
6473 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
6475 wc
->shared_level
= -1;
6476 wc
->stage
= DROP_REFERENCE
;
6479 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
6482 wret
= walk_down_tree(trans
, root
, path
, wc
);
6488 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
6496 btrfs_free_path(path
);
6500 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
6503 u64 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
6504 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
6507 * we add in the count of missing devices because we want
6508 * to make sure that any RAID levels on a degraded FS
6509 * continue to be honored.
6511 num_devices
= root
->fs_info
->fs_devices
->rw_devices
+
6512 root
->fs_info
->fs_devices
->missing_devices
;
6514 if (num_devices
== 1) {
6515 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
6516 stripped
= flags
& ~stripped
;
6518 /* turn raid0 into single device chunks */
6519 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
6522 /* turn mirroring into duplication */
6523 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
6524 BTRFS_BLOCK_GROUP_RAID10
))
6525 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
6528 /* they already had raid on here, just return */
6529 if (flags
& stripped
)
6532 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
6533 stripped
= flags
& ~stripped
;
6535 /* switch duplicated blocks with raid1 */
6536 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
6537 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
6539 /* turn single device chunks into raid0 */
6540 return stripped
| BTRFS_BLOCK_GROUP_RAID0
;
6545 static int set_block_group_ro(struct btrfs_block_group_cache
*cache
, int force
)
6547 struct btrfs_space_info
*sinfo
= cache
->space_info
;
6549 u64 min_allocable_bytes
;
6556 * We need some metadata space and system metadata space for
6557 * allocating chunks in some corner cases until we force to set
6558 * it to be readonly.
6561 (BTRFS_BLOCK_GROUP_SYSTEM
| BTRFS_BLOCK_GROUP_METADATA
)) &&
6563 min_allocable_bytes
= 1 * 1024 * 1024;
6565 min_allocable_bytes
= 0;
6567 spin_lock(&sinfo
->lock
);
6568 spin_lock(&cache
->lock
);
6569 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
6570 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
6572 if (sinfo
->bytes_used
+ sinfo
->bytes_reserved
+ sinfo
->bytes_pinned
+
6573 sinfo
->bytes_may_use
+ sinfo
->bytes_readonly
+
6574 cache
->reserved_pinned
+ num_bytes
+ min_allocable_bytes
<=
6575 sinfo
->total_bytes
) {
6576 sinfo
->bytes_readonly
+= num_bytes
;
6577 sinfo
->bytes_reserved
+= cache
->reserved_pinned
;
6578 cache
->reserved_pinned
= 0;
6583 spin_unlock(&cache
->lock
);
6584 spin_unlock(&sinfo
->lock
);
6588 int btrfs_set_block_group_ro(struct btrfs_root
*root
,
6589 struct btrfs_block_group_cache
*cache
)
6592 struct btrfs_trans_handle
*trans
;
6598 trans
= btrfs_join_transaction(root
);
6599 BUG_ON(IS_ERR(trans
));
6601 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
6602 if (alloc_flags
!= cache
->flags
)
6603 do_chunk_alloc(trans
, root
, 2 * 1024 * 1024, alloc_flags
,
6606 ret
= set_block_group_ro(cache
, 0);
6609 alloc_flags
= get_alloc_profile(root
, cache
->space_info
->flags
);
6610 ret
= do_chunk_alloc(trans
, root
, 2 * 1024 * 1024, alloc_flags
,
6614 ret
= set_block_group_ro(cache
, 0);
6616 btrfs_end_transaction(trans
, root
);
6620 int btrfs_force_chunk_alloc(struct btrfs_trans_handle
*trans
,
6621 struct btrfs_root
*root
, u64 type
)
6623 u64 alloc_flags
= get_alloc_profile(root
, type
);
6624 return do_chunk_alloc(trans
, root
, 2 * 1024 * 1024, alloc_flags
,
6629 * helper to account the unused space of all the readonly block group in the
6630 * list. takes mirrors into account.
6632 static u64
__btrfs_get_ro_block_group_free_space(struct list_head
*groups_list
)
6634 struct btrfs_block_group_cache
*block_group
;
6638 list_for_each_entry(block_group
, groups_list
, list
) {
6639 spin_lock(&block_group
->lock
);
6641 if (!block_group
->ro
) {
6642 spin_unlock(&block_group
->lock
);
6646 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_RAID1
|
6647 BTRFS_BLOCK_GROUP_RAID10
|
6648 BTRFS_BLOCK_GROUP_DUP
))
6653 free_bytes
+= (block_group
->key
.offset
-
6654 btrfs_block_group_used(&block_group
->item
)) *
6657 spin_unlock(&block_group
->lock
);
6664 * helper to account the unused space of all the readonly block group in the
6665 * space_info. takes mirrors into account.
6667 u64
btrfs_account_ro_block_groups_free_space(struct btrfs_space_info
*sinfo
)
6672 spin_lock(&sinfo
->lock
);
6674 for(i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
6675 if (!list_empty(&sinfo
->block_groups
[i
]))
6676 free_bytes
+= __btrfs_get_ro_block_group_free_space(
6677 &sinfo
->block_groups
[i
]);
6679 spin_unlock(&sinfo
->lock
);
6684 int btrfs_set_block_group_rw(struct btrfs_root
*root
,
6685 struct btrfs_block_group_cache
*cache
)
6687 struct btrfs_space_info
*sinfo
= cache
->space_info
;
6692 spin_lock(&sinfo
->lock
);
6693 spin_lock(&cache
->lock
);
6694 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
6695 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
6696 sinfo
->bytes_readonly
-= num_bytes
;
6698 spin_unlock(&cache
->lock
);
6699 spin_unlock(&sinfo
->lock
);
6704 * checks to see if its even possible to relocate this block group.
6706 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
6707 * ok to go ahead and try.
6709 int btrfs_can_relocate(struct btrfs_root
*root
, u64 bytenr
)
6711 struct btrfs_block_group_cache
*block_group
;
6712 struct btrfs_space_info
*space_info
;
6713 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
6714 struct btrfs_device
*device
;
6718 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
6720 /* odd, couldn't find the block group, leave it alone */
6724 /* no bytes used, we're good */
6725 if (!btrfs_block_group_used(&block_group
->item
))
6728 space_info
= block_group
->space_info
;
6729 spin_lock(&space_info
->lock
);
6731 full
= space_info
->full
;
6734 * if this is the last block group we have in this space, we can't
6735 * relocate it unless we're able to allocate a new chunk below.
6737 * Otherwise, we need to make sure we have room in the space to handle
6738 * all of the extents from this block group. If we can, we're good
6740 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
6741 (space_info
->bytes_used
+ space_info
->bytes_reserved
+
6742 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
6743 btrfs_block_group_used(&block_group
->item
) <
6744 space_info
->total_bytes
)) {
6745 spin_unlock(&space_info
->lock
);
6748 spin_unlock(&space_info
->lock
);
6751 * ok we don't have enough space, but maybe we have free space on our
6752 * devices to allocate new chunks for relocation, so loop through our
6753 * alloc devices and guess if we have enough space. However, if we
6754 * were marked as full, then we know there aren't enough chunks, and we
6761 mutex_lock(&root
->fs_info
->chunk_mutex
);
6762 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
6763 u64 min_free
= btrfs_block_group_used(&block_group
->item
);
6767 * check to make sure we can actually find a chunk with enough
6768 * space to fit our block group in.
6770 if (device
->total_bytes
> device
->bytes_used
+ min_free
) {
6771 ret
= find_free_dev_extent(NULL
, device
, min_free
,
6778 mutex_unlock(&root
->fs_info
->chunk_mutex
);
6780 btrfs_put_block_group(block_group
);
6784 static int find_first_block_group(struct btrfs_root
*root
,
6785 struct btrfs_path
*path
, struct btrfs_key
*key
)
6788 struct btrfs_key found_key
;
6789 struct extent_buffer
*leaf
;
6792 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
6797 slot
= path
->slots
[0];
6798 leaf
= path
->nodes
[0];
6799 if (slot
>= btrfs_header_nritems(leaf
)) {
6800 ret
= btrfs_next_leaf(root
, path
);
6807 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
6809 if (found_key
.objectid
>= key
->objectid
&&
6810 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
6820 void btrfs_put_block_group_cache(struct btrfs_fs_info
*info
)
6822 struct btrfs_block_group_cache
*block_group
;
6826 struct inode
*inode
;
6828 block_group
= btrfs_lookup_first_block_group(info
, last
);
6829 while (block_group
) {
6830 spin_lock(&block_group
->lock
);
6831 if (block_group
->iref
)
6833 spin_unlock(&block_group
->lock
);
6834 block_group
= next_block_group(info
->tree_root
,
6844 inode
= block_group
->inode
;
6845 block_group
->iref
= 0;
6846 block_group
->inode
= NULL
;
6847 spin_unlock(&block_group
->lock
);
6849 last
= block_group
->key
.objectid
+ block_group
->key
.offset
;
6850 btrfs_put_block_group(block_group
);
6854 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
6856 struct btrfs_block_group_cache
*block_group
;
6857 struct btrfs_space_info
*space_info
;
6858 struct btrfs_caching_control
*caching_ctl
;
6861 down_write(&info
->extent_commit_sem
);
6862 while (!list_empty(&info
->caching_block_groups
)) {
6863 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
6864 struct btrfs_caching_control
, list
);
6865 list_del(&caching_ctl
->list
);
6866 put_caching_control(caching_ctl
);
6868 up_write(&info
->extent_commit_sem
);
6870 spin_lock(&info
->block_group_cache_lock
);
6871 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
6872 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
6874 rb_erase(&block_group
->cache_node
,
6875 &info
->block_group_cache_tree
);
6876 spin_unlock(&info
->block_group_cache_lock
);
6878 down_write(&block_group
->space_info
->groups_sem
);
6879 list_del(&block_group
->list
);
6880 up_write(&block_group
->space_info
->groups_sem
);
6882 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
6883 wait_block_group_cache_done(block_group
);
6886 * We haven't cached this block group, which means we could
6887 * possibly have excluded extents on this block group.
6889 if (block_group
->cached
== BTRFS_CACHE_NO
)
6890 free_excluded_extents(info
->extent_root
, block_group
);
6892 btrfs_remove_free_space_cache(block_group
);
6893 btrfs_put_block_group(block_group
);
6895 spin_lock(&info
->block_group_cache_lock
);
6897 spin_unlock(&info
->block_group_cache_lock
);
6899 /* now that all the block groups are freed, go through and
6900 * free all the space_info structs. This is only called during
6901 * the final stages of unmount, and so we know nobody is
6902 * using them. We call synchronize_rcu() once before we start,
6903 * just to be on the safe side.
6907 release_global_block_rsv(info
);
6909 while(!list_empty(&info
->space_info
)) {
6910 space_info
= list_entry(info
->space_info
.next
,
6911 struct btrfs_space_info
,
6913 if (space_info
->bytes_pinned
> 0 ||
6914 space_info
->bytes_reserved
> 0) {
6916 dump_space_info(space_info
, 0, 0);
6918 list_del(&space_info
->list
);
6924 static void __link_block_group(struct btrfs_space_info
*space_info
,
6925 struct btrfs_block_group_cache
*cache
)
6927 int index
= get_block_group_index(cache
);
6929 down_write(&space_info
->groups_sem
);
6930 list_add_tail(&cache
->list
, &space_info
->block_groups
[index
]);
6931 up_write(&space_info
->groups_sem
);
6934 int btrfs_read_block_groups(struct btrfs_root
*root
)
6936 struct btrfs_path
*path
;
6938 struct btrfs_block_group_cache
*cache
;
6939 struct btrfs_fs_info
*info
= root
->fs_info
;
6940 struct btrfs_space_info
*space_info
;
6941 struct btrfs_key key
;
6942 struct btrfs_key found_key
;
6943 struct extent_buffer
*leaf
;
6947 root
= info
->extent_root
;
6950 btrfs_set_key_type(&key
, BTRFS_BLOCK_GROUP_ITEM_KEY
);
6951 path
= btrfs_alloc_path();
6956 cache_gen
= btrfs_super_cache_generation(&root
->fs_info
->super_copy
);
6957 if (cache_gen
!= 0 &&
6958 btrfs_super_generation(&root
->fs_info
->super_copy
) != cache_gen
)
6960 if (btrfs_test_opt(root
, CLEAR_CACHE
))
6962 if (!btrfs_test_opt(root
, SPACE_CACHE
) && cache_gen
)
6963 printk(KERN_INFO
"btrfs: disk space caching is enabled\n");
6966 ret
= find_first_block_group(root
, path
, &key
);
6971 leaf
= path
->nodes
[0];
6972 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6973 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
6978 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
6980 if (!cache
->free_space_ctl
) {
6986 atomic_set(&cache
->count
, 1);
6987 spin_lock_init(&cache
->lock
);
6988 cache
->fs_info
= info
;
6989 INIT_LIST_HEAD(&cache
->list
);
6990 INIT_LIST_HEAD(&cache
->cluster_list
);
6993 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
6995 read_extent_buffer(leaf
, &cache
->item
,
6996 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
6997 sizeof(cache
->item
));
6998 memcpy(&cache
->key
, &found_key
, sizeof(found_key
));
7000 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
7001 btrfs_release_path(path
);
7002 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
7003 cache
->sectorsize
= root
->sectorsize
;
7005 btrfs_init_free_space_ctl(cache
);
7008 * We need to exclude the super stripes now so that the space
7009 * info has super bytes accounted for, otherwise we'll think
7010 * we have more space than we actually do.
7012 exclude_super_stripes(root
, cache
);
7015 * check for two cases, either we are full, and therefore
7016 * don't need to bother with the caching work since we won't
7017 * find any space, or we are empty, and we can just add all
7018 * the space in and be done with it. This saves us _alot_ of
7019 * time, particularly in the full case.
7021 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
7022 cache
->last_byte_to_unpin
= (u64
)-1;
7023 cache
->cached
= BTRFS_CACHE_FINISHED
;
7024 free_excluded_extents(root
, cache
);
7025 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
7026 cache
->last_byte_to_unpin
= (u64
)-1;
7027 cache
->cached
= BTRFS_CACHE_FINISHED
;
7028 add_new_free_space(cache
, root
->fs_info
,
7030 found_key
.objectid
+
7032 free_excluded_extents(root
, cache
);
7035 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
7036 btrfs_block_group_used(&cache
->item
),
7039 cache
->space_info
= space_info
;
7040 spin_lock(&cache
->space_info
->lock
);
7041 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
7042 spin_unlock(&cache
->space_info
->lock
);
7044 __link_block_group(space_info
, cache
);
7046 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
7049 set_avail_alloc_bits(root
->fs_info
, cache
->flags
);
7050 if (btrfs_chunk_readonly(root
, cache
->key
.objectid
))
7051 set_block_group_ro(cache
, 1);
7054 list_for_each_entry_rcu(space_info
, &root
->fs_info
->space_info
, list
) {
7055 if (!(get_alloc_profile(root
, space_info
->flags
) &
7056 (BTRFS_BLOCK_GROUP_RAID10
|
7057 BTRFS_BLOCK_GROUP_RAID1
|
7058 BTRFS_BLOCK_GROUP_DUP
)))
7061 * avoid allocating from un-mirrored block group if there are
7062 * mirrored block groups.
7064 list_for_each_entry(cache
, &space_info
->block_groups
[3], list
)
7065 set_block_group_ro(cache
, 1);
7066 list_for_each_entry(cache
, &space_info
->block_groups
[4], list
)
7067 set_block_group_ro(cache
, 1);
7070 init_global_block_rsv(info
);
7073 btrfs_free_path(path
);
7077 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
7078 struct btrfs_root
*root
, u64 bytes_used
,
7079 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
7083 struct btrfs_root
*extent_root
;
7084 struct btrfs_block_group_cache
*cache
;
7086 extent_root
= root
->fs_info
->extent_root
;
7088 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
7090 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
7093 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
7095 if (!cache
->free_space_ctl
) {
7100 cache
->key
.objectid
= chunk_offset
;
7101 cache
->key
.offset
= size
;
7102 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
7103 cache
->sectorsize
= root
->sectorsize
;
7104 cache
->fs_info
= root
->fs_info
;
7106 atomic_set(&cache
->count
, 1);
7107 spin_lock_init(&cache
->lock
);
7108 INIT_LIST_HEAD(&cache
->list
);
7109 INIT_LIST_HEAD(&cache
->cluster_list
);
7111 btrfs_init_free_space_ctl(cache
);
7113 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
7114 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
7115 cache
->flags
= type
;
7116 btrfs_set_block_group_flags(&cache
->item
, type
);
7118 cache
->last_byte_to_unpin
= (u64
)-1;
7119 cache
->cached
= BTRFS_CACHE_FINISHED
;
7120 exclude_super_stripes(root
, cache
);
7122 add_new_free_space(cache
, root
->fs_info
, chunk_offset
,
7123 chunk_offset
+ size
);
7125 free_excluded_extents(root
, cache
);
7127 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
7128 &cache
->space_info
);
7131 spin_lock(&cache
->space_info
->lock
);
7132 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
7133 spin_unlock(&cache
->space_info
->lock
);
7135 __link_block_group(cache
->space_info
, cache
);
7137 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
7140 ret
= btrfs_insert_item(trans
, extent_root
, &cache
->key
, &cache
->item
,
7141 sizeof(cache
->item
));
7144 set_avail_alloc_bits(extent_root
->fs_info
, type
);
7149 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
7150 struct btrfs_root
*root
, u64 group_start
)
7152 struct btrfs_path
*path
;
7153 struct btrfs_block_group_cache
*block_group
;
7154 struct btrfs_free_cluster
*cluster
;
7155 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
7156 struct btrfs_key key
;
7157 struct inode
*inode
;
7161 root
= root
->fs_info
->extent_root
;
7163 block_group
= btrfs_lookup_block_group(root
->fs_info
, group_start
);
7164 BUG_ON(!block_group
);
7165 BUG_ON(!block_group
->ro
);
7168 * Free the reserved super bytes from this block group before
7171 free_excluded_extents(root
, block_group
);
7173 memcpy(&key
, &block_group
->key
, sizeof(key
));
7174 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
7175 BTRFS_BLOCK_GROUP_RAID1
|
7176 BTRFS_BLOCK_GROUP_RAID10
))
7181 /* make sure this block group isn't part of an allocation cluster */
7182 cluster
= &root
->fs_info
->data_alloc_cluster
;
7183 spin_lock(&cluster
->refill_lock
);
7184 btrfs_return_cluster_to_free_space(block_group
, cluster
);
7185 spin_unlock(&cluster
->refill_lock
);
7188 * make sure this block group isn't part of a metadata
7189 * allocation cluster
7191 cluster
= &root
->fs_info
->meta_alloc_cluster
;
7192 spin_lock(&cluster
->refill_lock
);
7193 btrfs_return_cluster_to_free_space(block_group
, cluster
);
7194 spin_unlock(&cluster
->refill_lock
);
7196 path
= btrfs_alloc_path();
7202 inode
= lookup_free_space_inode(root
, block_group
, path
);
7203 if (!IS_ERR(inode
)) {
7204 ret
= btrfs_orphan_add(trans
, inode
);
7207 /* One for the block groups ref */
7208 spin_lock(&block_group
->lock
);
7209 if (block_group
->iref
) {
7210 block_group
->iref
= 0;
7211 block_group
->inode
= NULL
;
7212 spin_unlock(&block_group
->lock
);
7215 spin_unlock(&block_group
->lock
);
7217 /* One for our lookup ref */
7221 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
7222 key
.offset
= block_group
->key
.objectid
;
7225 ret
= btrfs_search_slot(trans
, tree_root
, &key
, path
, -1, 1);
7229 btrfs_release_path(path
);
7231 ret
= btrfs_del_item(trans
, tree_root
, path
);
7234 btrfs_release_path(path
);
7237 spin_lock(&root
->fs_info
->block_group_cache_lock
);
7238 rb_erase(&block_group
->cache_node
,
7239 &root
->fs_info
->block_group_cache_tree
);
7240 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
7242 down_write(&block_group
->space_info
->groups_sem
);
7244 * we must use list_del_init so people can check to see if they
7245 * are still on the list after taking the semaphore
7247 list_del_init(&block_group
->list
);
7248 up_write(&block_group
->space_info
->groups_sem
);
7250 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
7251 wait_block_group_cache_done(block_group
);
7253 btrfs_remove_free_space_cache(block_group
);
7255 spin_lock(&block_group
->space_info
->lock
);
7256 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
7257 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
7258 block_group
->space_info
->disk_total
-= block_group
->key
.offset
* factor
;
7259 spin_unlock(&block_group
->space_info
->lock
);
7261 memcpy(&key
, &block_group
->key
, sizeof(key
));
7263 btrfs_clear_space_info_full(root
->fs_info
);
7265 btrfs_put_block_group(block_group
);
7266 btrfs_put_block_group(block_group
);
7268 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
7274 ret
= btrfs_del_item(trans
, root
, path
);
7276 btrfs_free_path(path
);
7280 int btrfs_init_space_info(struct btrfs_fs_info
*fs_info
)
7282 struct btrfs_space_info
*space_info
;
7283 struct btrfs_super_block
*disk_super
;
7289 disk_super
= &fs_info
->super_copy
;
7290 if (!btrfs_super_root(disk_super
))
7293 features
= btrfs_super_incompat_flags(disk_super
);
7294 if (features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
7297 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
7298 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
7303 flags
= BTRFS_BLOCK_GROUP_METADATA
| BTRFS_BLOCK_GROUP_DATA
;
7304 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
7306 flags
= BTRFS_BLOCK_GROUP_METADATA
;
7307 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
7311 flags
= BTRFS_BLOCK_GROUP_DATA
;
7312 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
7318 int btrfs_error_unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
7320 return unpin_extent_range(root
, start
, end
);
7323 int btrfs_error_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
7324 u64 num_bytes
, u64
*actual_bytes
)
7326 return btrfs_discard_extent(root
, bytenr
, num_bytes
, actual_bytes
);
7329 int btrfs_trim_fs(struct btrfs_root
*root
, struct fstrim_range
*range
)
7331 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7332 struct btrfs_block_group_cache
*cache
= NULL
;
7339 cache
= btrfs_lookup_block_group(fs_info
, range
->start
);
7342 if (cache
->key
.objectid
>= (range
->start
+ range
->len
)) {
7343 btrfs_put_block_group(cache
);
7347 start
= max(range
->start
, cache
->key
.objectid
);
7348 end
= min(range
->start
+ range
->len
,
7349 cache
->key
.objectid
+ cache
->key
.offset
);
7351 if (end
- start
>= range
->minlen
) {
7352 if (!block_group_cache_done(cache
)) {
7353 ret
= cache_block_group(cache
, NULL
, root
, 0);
7355 wait_block_group_cache_done(cache
);
7357 ret
= btrfs_trim_block_group(cache
,
7363 trimmed
+= group_trimmed
;
7365 btrfs_put_block_group(cache
);
7370 cache
= next_block_group(fs_info
->tree_root
, cache
);
7373 range
->len
= trimmed
;