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 static int update_block_group(struct btrfs_trans_handle
*trans
,
37 struct btrfs_root
*root
,
38 u64 bytenr
, u64 num_bytes
, int alloc
);
39 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
40 struct btrfs_root
*root
,
41 u64 bytenr
, u64 num_bytes
, u64 parent
,
42 u64 root_objectid
, u64 owner_objectid
,
43 u64 owner_offset
, int refs_to_drop
,
44 struct btrfs_delayed_extent_op
*extra_op
);
45 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
46 struct extent_buffer
*leaf
,
47 struct btrfs_extent_item
*ei
);
48 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
49 struct btrfs_root
*root
,
50 u64 parent
, u64 root_objectid
,
51 u64 flags
, u64 owner
, u64 offset
,
52 struct btrfs_key
*ins
, int ref_mod
);
53 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
54 struct btrfs_root
*root
,
55 u64 parent
, u64 root_objectid
,
56 u64 flags
, struct btrfs_disk_key
*key
,
57 int level
, struct btrfs_key
*ins
);
58 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
59 struct btrfs_root
*extent_root
, u64 alloc_bytes
,
60 u64 flags
, int force
);
61 static int find_next_key(struct btrfs_path
*path
, int level
,
62 struct btrfs_key
*key
);
63 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
64 int dump_block_groups
);
67 block_group_cache_done(struct btrfs_block_group_cache
*cache
)
70 return cache
->cached
== BTRFS_CACHE_FINISHED
;
73 static int block_group_bits(struct btrfs_block_group_cache
*cache
, u64 bits
)
75 return (cache
->flags
& bits
) == bits
;
78 void btrfs_get_block_group(struct btrfs_block_group_cache
*cache
)
80 atomic_inc(&cache
->count
);
83 void btrfs_put_block_group(struct btrfs_block_group_cache
*cache
)
85 if (atomic_dec_and_test(&cache
->count
)) {
86 WARN_ON(cache
->pinned
> 0);
87 WARN_ON(cache
->reserved
> 0);
88 WARN_ON(cache
->reserved_pinned
> 0);
94 * this adds the block group to the fs_info rb tree for the block group
97 static int btrfs_add_block_group_cache(struct btrfs_fs_info
*info
,
98 struct btrfs_block_group_cache
*block_group
)
101 struct rb_node
*parent
= NULL
;
102 struct btrfs_block_group_cache
*cache
;
104 spin_lock(&info
->block_group_cache_lock
);
105 p
= &info
->block_group_cache_tree
.rb_node
;
109 cache
= rb_entry(parent
, struct btrfs_block_group_cache
,
111 if (block_group
->key
.objectid
< cache
->key
.objectid
) {
113 } else if (block_group
->key
.objectid
> cache
->key
.objectid
) {
116 spin_unlock(&info
->block_group_cache_lock
);
121 rb_link_node(&block_group
->cache_node
, parent
, p
);
122 rb_insert_color(&block_group
->cache_node
,
123 &info
->block_group_cache_tree
);
124 spin_unlock(&info
->block_group_cache_lock
);
130 * This will return the block group at or after bytenr if contains is 0, else
131 * it will return the block group that contains the bytenr
133 static struct btrfs_block_group_cache
*
134 block_group_cache_tree_search(struct btrfs_fs_info
*info
, u64 bytenr
,
137 struct btrfs_block_group_cache
*cache
, *ret
= NULL
;
141 spin_lock(&info
->block_group_cache_lock
);
142 n
= info
->block_group_cache_tree
.rb_node
;
145 cache
= rb_entry(n
, struct btrfs_block_group_cache
,
147 end
= cache
->key
.objectid
+ cache
->key
.offset
- 1;
148 start
= cache
->key
.objectid
;
150 if (bytenr
< start
) {
151 if (!contains
&& (!ret
|| start
< ret
->key
.objectid
))
154 } else if (bytenr
> start
) {
155 if (contains
&& bytenr
<= end
) {
166 btrfs_get_block_group(ret
);
167 spin_unlock(&info
->block_group_cache_lock
);
172 static int add_excluded_extent(struct btrfs_root
*root
,
173 u64 start
, u64 num_bytes
)
175 u64 end
= start
+ num_bytes
- 1;
176 set_extent_bits(&root
->fs_info
->freed_extents
[0],
177 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
178 set_extent_bits(&root
->fs_info
->freed_extents
[1],
179 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
183 static void free_excluded_extents(struct btrfs_root
*root
,
184 struct btrfs_block_group_cache
*cache
)
188 start
= cache
->key
.objectid
;
189 end
= start
+ cache
->key
.offset
- 1;
191 clear_extent_bits(&root
->fs_info
->freed_extents
[0],
192 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
193 clear_extent_bits(&root
->fs_info
->freed_extents
[1],
194 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
197 static int exclude_super_stripes(struct btrfs_root
*root
,
198 struct btrfs_block_group_cache
*cache
)
205 if (cache
->key
.objectid
< BTRFS_SUPER_INFO_OFFSET
) {
206 stripe_len
= BTRFS_SUPER_INFO_OFFSET
- cache
->key
.objectid
;
207 cache
->bytes_super
+= stripe_len
;
208 ret
= add_excluded_extent(root
, cache
->key
.objectid
,
213 for (i
= 0; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
214 bytenr
= btrfs_sb_offset(i
);
215 ret
= btrfs_rmap_block(&root
->fs_info
->mapping_tree
,
216 cache
->key
.objectid
, bytenr
,
217 0, &logical
, &nr
, &stripe_len
);
221 cache
->bytes_super
+= stripe_len
;
222 ret
= add_excluded_extent(root
, logical
[nr
],
232 static struct btrfs_caching_control
*
233 get_caching_control(struct btrfs_block_group_cache
*cache
)
235 struct btrfs_caching_control
*ctl
;
237 spin_lock(&cache
->lock
);
238 if (cache
->cached
!= BTRFS_CACHE_STARTED
) {
239 spin_unlock(&cache
->lock
);
243 /* We're loading it the fast way, so we don't have a caching_ctl. */
244 if (!cache
->caching_ctl
) {
245 spin_unlock(&cache
->lock
);
249 ctl
= cache
->caching_ctl
;
250 atomic_inc(&ctl
->count
);
251 spin_unlock(&cache
->lock
);
255 static void put_caching_control(struct btrfs_caching_control
*ctl
)
257 if (atomic_dec_and_test(&ctl
->count
))
262 * this is only called by cache_block_group, since we could have freed extents
263 * we need to check the pinned_extents for any extents that can't be used yet
264 * since their free space will be released as soon as the transaction commits.
266 static u64
add_new_free_space(struct btrfs_block_group_cache
*block_group
,
267 struct btrfs_fs_info
*info
, u64 start
, u64 end
)
269 u64 extent_start
, extent_end
, size
, total_added
= 0;
272 while (start
< end
) {
273 ret
= find_first_extent_bit(info
->pinned_extents
, start
,
274 &extent_start
, &extent_end
,
275 EXTENT_DIRTY
| EXTENT_UPTODATE
);
279 if (extent_start
<= start
) {
280 start
= extent_end
+ 1;
281 } else if (extent_start
> start
&& extent_start
< end
) {
282 size
= extent_start
- start
;
284 ret
= btrfs_add_free_space(block_group
, start
,
287 start
= extent_end
+ 1;
296 ret
= btrfs_add_free_space(block_group
, start
, size
);
303 static int caching_kthread(void *data
)
305 struct btrfs_block_group_cache
*block_group
= data
;
306 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
307 struct btrfs_caching_control
*caching_ctl
= block_group
->caching_ctl
;
308 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
309 struct btrfs_path
*path
;
310 struct extent_buffer
*leaf
;
311 struct btrfs_key key
;
317 path
= btrfs_alloc_path();
321 last
= max_t(u64
, block_group
->key
.objectid
, BTRFS_SUPER_INFO_OFFSET
);
324 * We don't want to deadlock with somebody trying to allocate a new
325 * extent for the extent root while also trying to search the extent
326 * root to add free space. So we skip locking and search the commit
327 * root, since its read-only
329 path
->skip_locking
= 1;
330 path
->search_commit_root
= 1;
335 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
337 mutex_lock(&caching_ctl
->mutex
);
338 /* need to make sure the commit_root doesn't disappear */
339 down_read(&fs_info
->extent_commit_sem
);
341 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
345 leaf
= path
->nodes
[0];
346 nritems
= btrfs_header_nritems(leaf
);
350 if (fs_info
->closing
> 1) {
355 if (path
->slots
[0] < nritems
) {
356 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
358 ret
= find_next_key(path
, 0, &key
);
362 caching_ctl
->progress
= last
;
363 btrfs_release_path(extent_root
, path
);
364 up_read(&fs_info
->extent_commit_sem
);
365 mutex_unlock(&caching_ctl
->mutex
);
366 if (btrfs_transaction_in_commit(fs_info
))
373 if (key
.objectid
< block_group
->key
.objectid
) {
378 if (key
.objectid
>= block_group
->key
.objectid
+
379 block_group
->key
.offset
)
382 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
) {
383 total_found
+= add_new_free_space(block_group
,
386 last
= key
.objectid
+ key
.offset
;
388 if (total_found
> (1024 * 1024 * 2)) {
390 wake_up(&caching_ctl
->wait
);
397 total_found
+= add_new_free_space(block_group
, fs_info
, last
,
398 block_group
->key
.objectid
+
399 block_group
->key
.offset
);
400 caching_ctl
->progress
= (u64
)-1;
402 spin_lock(&block_group
->lock
);
403 block_group
->caching_ctl
= NULL
;
404 block_group
->cached
= BTRFS_CACHE_FINISHED
;
405 spin_unlock(&block_group
->lock
);
408 btrfs_free_path(path
);
409 up_read(&fs_info
->extent_commit_sem
);
411 free_excluded_extents(extent_root
, block_group
);
413 mutex_unlock(&caching_ctl
->mutex
);
414 wake_up(&caching_ctl
->wait
);
416 put_caching_control(caching_ctl
);
417 atomic_dec(&block_group
->space_info
->caching_threads
);
418 btrfs_put_block_group(block_group
);
423 static int cache_block_group(struct btrfs_block_group_cache
*cache
,
424 struct btrfs_trans_handle
*trans
,
425 struct btrfs_root
*root
,
428 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
429 struct btrfs_caching_control
*caching_ctl
;
430 struct task_struct
*tsk
;
434 if (cache
->cached
!= BTRFS_CACHE_NO
)
438 * We can't do the read from on-disk cache during a commit since we need
439 * to have the normal tree locking. Also if we are currently trying to
440 * allocate blocks for the tree root we can't do the fast caching since
441 * we likely hold important locks.
443 if (trans
&& (!trans
->transaction
->in_commit
) &&
444 (root
&& root
!= root
->fs_info
->tree_root
)) {
445 spin_lock(&cache
->lock
);
446 if (cache
->cached
!= BTRFS_CACHE_NO
) {
447 spin_unlock(&cache
->lock
);
450 cache
->cached
= BTRFS_CACHE_STARTED
;
451 spin_unlock(&cache
->lock
);
453 ret
= load_free_space_cache(fs_info
, cache
);
455 spin_lock(&cache
->lock
);
457 cache
->cached
= BTRFS_CACHE_FINISHED
;
458 cache
->last_byte_to_unpin
= (u64
)-1;
460 cache
->cached
= BTRFS_CACHE_NO
;
462 spin_unlock(&cache
->lock
);
464 free_excluded_extents(fs_info
->extent_root
, cache
);
472 caching_ctl
= kzalloc(sizeof(*caching_ctl
), GFP_NOFS
);
473 BUG_ON(!caching_ctl
);
475 INIT_LIST_HEAD(&caching_ctl
->list
);
476 mutex_init(&caching_ctl
->mutex
);
477 init_waitqueue_head(&caching_ctl
->wait
);
478 caching_ctl
->block_group
= cache
;
479 caching_ctl
->progress
= cache
->key
.objectid
;
480 /* one for caching kthread, one for caching block group list */
481 atomic_set(&caching_ctl
->count
, 2);
483 spin_lock(&cache
->lock
);
484 if (cache
->cached
!= BTRFS_CACHE_NO
) {
485 spin_unlock(&cache
->lock
);
489 cache
->caching_ctl
= caching_ctl
;
490 cache
->cached
= BTRFS_CACHE_STARTED
;
491 spin_unlock(&cache
->lock
);
493 down_write(&fs_info
->extent_commit_sem
);
494 list_add_tail(&caching_ctl
->list
, &fs_info
->caching_block_groups
);
495 up_write(&fs_info
->extent_commit_sem
);
497 atomic_inc(&cache
->space_info
->caching_threads
);
498 btrfs_get_block_group(cache
);
500 tsk
= kthread_run(caching_kthread
, cache
, "btrfs-cache-%llu\n",
501 cache
->key
.objectid
);
504 printk(KERN_ERR
"error running thread %d\n", ret
);
512 * return the block group that starts at or after bytenr
514 static struct btrfs_block_group_cache
*
515 btrfs_lookup_first_block_group(struct btrfs_fs_info
*info
, u64 bytenr
)
517 struct btrfs_block_group_cache
*cache
;
519 cache
= block_group_cache_tree_search(info
, bytenr
, 0);
525 * return the block group that contains the given bytenr
527 struct btrfs_block_group_cache
*btrfs_lookup_block_group(
528 struct btrfs_fs_info
*info
,
531 struct btrfs_block_group_cache
*cache
;
533 cache
= block_group_cache_tree_search(info
, bytenr
, 1);
538 static struct btrfs_space_info
*__find_space_info(struct btrfs_fs_info
*info
,
541 struct list_head
*head
= &info
->space_info
;
542 struct btrfs_space_info
*found
;
544 flags
&= BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_SYSTEM
|
545 BTRFS_BLOCK_GROUP_METADATA
;
548 list_for_each_entry_rcu(found
, head
, list
) {
549 if (found
->flags
& flags
) {
559 * after adding space to the filesystem, we need to clear the full flags
560 * on all the space infos.
562 void btrfs_clear_space_info_full(struct btrfs_fs_info
*info
)
564 struct list_head
*head
= &info
->space_info
;
565 struct btrfs_space_info
*found
;
568 list_for_each_entry_rcu(found
, head
, list
)
573 static u64
div_factor(u64 num
, int factor
)
582 static u64
div_factor_fine(u64 num
, int factor
)
591 u64
btrfs_find_block_group(struct btrfs_root
*root
,
592 u64 search_start
, u64 search_hint
, int owner
)
594 struct btrfs_block_group_cache
*cache
;
596 u64 last
= max(search_hint
, search_start
);
603 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
607 spin_lock(&cache
->lock
);
608 last
= cache
->key
.objectid
+ cache
->key
.offset
;
609 used
= btrfs_block_group_used(&cache
->item
);
611 if ((full_search
|| !cache
->ro
) &&
612 block_group_bits(cache
, BTRFS_BLOCK_GROUP_METADATA
)) {
613 if (used
+ cache
->pinned
+ cache
->reserved
<
614 div_factor(cache
->key
.offset
, factor
)) {
615 group_start
= cache
->key
.objectid
;
616 spin_unlock(&cache
->lock
);
617 btrfs_put_block_group(cache
);
621 spin_unlock(&cache
->lock
);
622 btrfs_put_block_group(cache
);
630 if (!full_search
&& factor
< 10) {
640 /* simple helper to search for an existing extent at a given offset */
641 int btrfs_lookup_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
644 struct btrfs_key key
;
645 struct btrfs_path
*path
;
647 path
= btrfs_alloc_path();
649 key
.objectid
= start
;
651 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
652 ret
= btrfs_search_slot(NULL
, root
->fs_info
->extent_root
, &key
, path
,
654 btrfs_free_path(path
);
659 * helper function to lookup reference count and flags of extent.
661 * the head node for delayed ref is used to store the sum of all the
662 * reference count modifications queued up in the rbtree. the head
663 * node may also store the extent flags to set. This way you can check
664 * to see what the reference count and extent flags would be if all of
665 * the delayed refs are not processed.
667 int btrfs_lookup_extent_info(struct btrfs_trans_handle
*trans
,
668 struct btrfs_root
*root
, u64 bytenr
,
669 u64 num_bytes
, u64
*refs
, u64
*flags
)
671 struct btrfs_delayed_ref_head
*head
;
672 struct btrfs_delayed_ref_root
*delayed_refs
;
673 struct btrfs_path
*path
;
674 struct btrfs_extent_item
*ei
;
675 struct extent_buffer
*leaf
;
676 struct btrfs_key key
;
682 path
= btrfs_alloc_path();
686 key
.objectid
= bytenr
;
687 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
688 key
.offset
= num_bytes
;
690 path
->skip_locking
= 1;
691 path
->search_commit_root
= 1;
694 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
,
700 leaf
= path
->nodes
[0];
701 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
702 if (item_size
>= sizeof(*ei
)) {
703 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
704 struct btrfs_extent_item
);
705 num_refs
= btrfs_extent_refs(leaf
, ei
);
706 extent_flags
= btrfs_extent_flags(leaf
, ei
);
708 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
709 struct btrfs_extent_item_v0
*ei0
;
710 BUG_ON(item_size
!= sizeof(*ei0
));
711 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
712 struct btrfs_extent_item_v0
);
713 num_refs
= btrfs_extent_refs_v0(leaf
, ei0
);
714 /* FIXME: this isn't correct for data */
715 extent_flags
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
720 BUG_ON(num_refs
== 0);
730 delayed_refs
= &trans
->transaction
->delayed_refs
;
731 spin_lock(&delayed_refs
->lock
);
732 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
734 if (!mutex_trylock(&head
->mutex
)) {
735 atomic_inc(&head
->node
.refs
);
736 spin_unlock(&delayed_refs
->lock
);
738 btrfs_release_path(root
->fs_info
->extent_root
, path
);
740 mutex_lock(&head
->mutex
);
741 mutex_unlock(&head
->mutex
);
742 btrfs_put_delayed_ref(&head
->node
);
745 if (head
->extent_op
&& head
->extent_op
->update_flags
)
746 extent_flags
|= head
->extent_op
->flags_to_set
;
748 BUG_ON(num_refs
== 0);
750 num_refs
+= head
->node
.ref_mod
;
751 mutex_unlock(&head
->mutex
);
753 spin_unlock(&delayed_refs
->lock
);
755 WARN_ON(num_refs
== 0);
759 *flags
= extent_flags
;
761 btrfs_free_path(path
);
766 * Back reference rules. Back refs have three main goals:
768 * 1) differentiate between all holders of references to an extent so that
769 * when a reference is dropped we can make sure it was a valid reference
770 * before freeing the extent.
772 * 2) Provide enough information to quickly find the holders of an extent
773 * if we notice a given block is corrupted or bad.
775 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
776 * maintenance. This is actually the same as #2, but with a slightly
777 * different use case.
779 * There are two kinds of back refs. The implicit back refs is optimized
780 * for pointers in non-shared tree blocks. For a given pointer in a block,
781 * back refs of this kind provide information about the block's owner tree
782 * and the pointer's key. These information allow us to find the block by
783 * b-tree searching. The full back refs is for pointers in tree blocks not
784 * referenced by their owner trees. The location of tree block is recorded
785 * in the back refs. Actually the full back refs is generic, and can be
786 * used in all cases the implicit back refs is used. The major shortcoming
787 * of the full back refs is its overhead. Every time a tree block gets
788 * COWed, we have to update back refs entry for all pointers in it.
790 * For a newly allocated tree block, we use implicit back refs for
791 * pointers in it. This means most tree related operations only involve
792 * implicit back refs. For a tree block created in old transaction, the
793 * only way to drop a reference to it is COW it. So we can detect the
794 * event that tree block loses its owner tree's reference and do the
795 * back refs conversion.
797 * When a tree block is COW'd through a tree, there are four cases:
799 * The reference count of the block is one and the tree is the block's
800 * owner tree. Nothing to do in this case.
802 * The reference count of the block is one and the tree is not the
803 * block's owner tree. In this case, full back refs is used for pointers
804 * in the block. Remove these full back refs, add implicit back refs for
805 * every pointers in the new block.
807 * The reference count of the block is greater than one and the tree is
808 * the block's owner tree. In this case, implicit back refs is used for
809 * pointers in the block. Add full back refs for every pointers in the
810 * block, increase lower level extents' reference counts. The original
811 * implicit back refs are entailed to the new block.
813 * The reference count of the block is greater than one and the tree is
814 * not the block's owner tree. Add implicit back refs for every pointer in
815 * the new block, increase lower level extents' reference count.
817 * Back Reference Key composing:
819 * The key objectid corresponds to the first byte in the extent,
820 * The key type is used to differentiate between types of back refs.
821 * There are different meanings of the key offset for different types
824 * File extents can be referenced by:
826 * - multiple snapshots, subvolumes, or different generations in one subvol
827 * - different files inside a single subvolume
828 * - different offsets inside a file (bookend extents in file.c)
830 * The extent ref structure for the implicit back refs has fields for:
832 * - Objectid of the subvolume root
833 * - objectid of the file holding the reference
834 * - original offset in the file
835 * - how many bookend extents
837 * The key offset for the implicit back refs is hash of the first
840 * The extent ref structure for the full back refs has field for:
842 * - number of pointers in the tree leaf
844 * The key offset for the implicit back refs is the first byte of
847 * When a file extent is allocated, The implicit back refs is used.
848 * the fields are filled in:
850 * (root_key.objectid, inode objectid, offset in file, 1)
852 * When a file extent is removed file truncation, we find the
853 * corresponding implicit back refs and check the following fields:
855 * (btrfs_header_owner(leaf), inode objectid, offset in file)
857 * Btree extents can be referenced by:
859 * - Different subvolumes
861 * Both the implicit back refs and the full back refs for tree blocks
862 * only consist of key. The key offset for the implicit back refs is
863 * objectid of block's owner tree. The key offset for the full back refs
864 * is the first byte of parent block.
866 * When implicit back refs is used, information about the lowest key and
867 * level of the tree block are required. These information are stored in
868 * tree block info structure.
871 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
872 static int convert_extent_item_v0(struct btrfs_trans_handle
*trans
,
873 struct btrfs_root
*root
,
874 struct btrfs_path
*path
,
875 u64 owner
, u32 extra_size
)
877 struct btrfs_extent_item
*item
;
878 struct btrfs_extent_item_v0
*ei0
;
879 struct btrfs_extent_ref_v0
*ref0
;
880 struct btrfs_tree_block_info
*bi
;
881 struct extent_buffer
*leaf
;
882 struct btrfs_key key
;
883 struct btrfs_key found_key
;
884 u32 new_size
= sizeof(*item
);
888 leaf
= path
->nodes
[0];
889 BUG_ON(btrfs_item_size_nr(leaf
, path
->slots
[0]) != sizeof(*ei0
));
891 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
892 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
893 struct btrfs_extent_item_v0
);
894 refs
= btrfs_extent_refs_v0(leaf
, ei0
);
896 if (owner
== (u64
)-1) {
898 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
899 ret
= btrfs_next_leaf(root
, path
);
903 leaf
= path
->nodes
[0];
905 btrfs_item_key_to_cpu(leaf
, &found_key
,
907 BUG_ON(key
.objectid
!= found_key
.objectid
);
908 if (found_key
.type
!= BTRFS_EXTENT_REF_V0_KEY
) {
912 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
913 struct btrfs_extent_ref_v0
);
914 owner
= btrfs_ref_objectid_v0(leaf
, ref0
);
918 btrfs_release_path(root
, path
);
920 if (owner
< BTRFS_FIRST_FREE_OBJECTID
)
921 new_size
+= sizeof(*bi
);
923 new_size
-= sizeof(*ei0
);
924 ret
= btrfs_search_slot(trans
, root
, &key
, path
,
925 new_size
+ extra_size
, 1);
930 ret
= btrfs_extend_item(trans
, root
, path
, new_size
);
933 leaf
= path
->nodes
[0];
934 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
935 btrfs_set_extent_refs(leaf
, item
, refs
);
936 /* FIXME: get real generation */
937 btrfs_set_extent_generation(leaf
, item
, 0);
938 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
939 btrfs_set_extent_flags(leaf
, item
,
940 BTRFS_EXTENT_FLAG_TREE_BLOCK
|
941 BTRFS_BLOCK_FLAG_FULL_BACKREF
);
942 bi
= (struct btrfs_tree_block_info
*)(item
+ 1);
943 /* FIXME: get first key of the block */
944 memset_extent_buffer(leaf
, 0, (unsigned long)bi
, sizeof(*bi
));
945 btrfs_set_tree_block_level(leaf
, bi
, (int)owner
);
947 btrfs_set_extent_flags(leaf
, item
, BTRFS_EXTENT_FLAG_DATA
);
949 btrfs_mark_buffer_dirty(leaf
);
954 static u64
hash_extent_data_ref(u64 root_objectid
, u64 owner
, u64 offset
)
956 u32 high_crc
= ~(u32
)0;
957 u32 low_crc
= ~(u32
)0;
960 lenum
= cpu_to_le64(root_objectid
);
961 high_crc
= crc32c(high_crc
, &lenum
, sizeof(lenum
));
962 lenum
= cpu_to_le64(owner
);
963 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
964 lenum
= cpu_to_le64(offset
);
965 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
967 return ((u64
)high_crc
<< 31) ^ (u64
)low_crc
;
970 static u64
hash_extent_data_ref_item(struct extent_buffer
*leaf
,
971 struct btrfs_extent_data_ref
*ref
)
973 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf
, ref
),
974 btrfs_extent_data_ref_objectid(leaf
, ref
),
975 btrfs_extent_data_ref_offset(leaf
, ref
));
978 static int match_extent_data_ref(struct extent_buffer
*leaf
,
979 struct btrfs_extent_data_ref
*ref
,
980 u64 root_objectid
, u64 owner
, u64 offset
)
982 if (btrfs_extent_data_ref_root(leaf
, ref
) != root_objectid
||
983 btrfs_extent_data_ref_objectid(leaf
, ref
) != owner
||
984 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
989 static noinline
int lookup_extent_data_ref(struct btrfs_trans_handle
*trans
,
990 struct btrfs_root
*root
,
991 struct btrfs_path
*path
,
992 u64 bytenr
, u64 parent
,
994 u64 owner
, u64 offset
)
996 struct btrfs_key key
;
997 struct btrfs_extent_data_ref
*ref
;
998 struct extent_buffer
*leaf
;
1004 key
.objectid
= bytenr
;
1006 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1007 key
.offset
= parent
;
1009 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1010 key
.offset
= hash_extent_data_ref(root_objectid
,
1015 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1024 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1025 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1026 btrfs_release_path(root
, path
);
1027 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1038 leaf
= path
->nodes
[0];
1039 nritems
= btrfs_header_nritems(leaf
);
1041 if (path
->slots
[0] >= nritems
) {
1042 ret
= btrfs_next_leaf(root
, path
);
1048 leaf
= path
->nodes
[0];
1049 nritems
= btrfs_header_nritems(leaf
);
1053 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1054 if (key
.objectid
!= bytenr
||
1055 key
.type
!= BTRFS_EXTENT_DATA_REF_KEY
)
1058 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1059 struct btrfs_extent_data_ref
);
1061 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1064 btrfs_release_path(root
, path
);
1076 static noinline
int insert_extent_data_ref(struct btrfs_trans_handle
*trans
,
1077 struct btrfs_root
*root
,
1078 struct btrfs_path
*path
,
1079 u64 bytenr
, u64 parent
,
1080 u64 root_objectid
, u64 owner
,
1081 u64 offset
, int refs_to_add
)
1083 struct btrfs_key key
;
1084 struct extent_buffer
*leaf
;
1089 key
.objectid
= bytenr
;
1091 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1092 key
.offset
= parent
;
1093 size
= sizeof(struct btrfs_shared_data_ref
);
1095 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1096 key
.offset
= hash_extent_data_ref(root_objectid
,
1098 size
= sizeof(struct btrfs_extent_data_ref
);
1101 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, size
);
1102 if (ret
&& ret
!= -EEXIST
)
1105 leaf
= path
->nodes
[0];
1107 struct btrfs_shared_data_ref
*ref
;
1108 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1109 struct btrfs_shared_data_ref
);
1111 btrfs_set_shared_data_ref_count(leaf
, ref
, refs_to_add
);
1113 num_refs
= btrfs_shared_data_ref_count(leaf
, ref
);
1114 num_refs
+= refs_to_add
;
1115 btrfs_set_shared_data_ref_count(leaf
, ref
, num_refs
);
1118 struct btrfs_extent_data_ref
*ref
;
1119 while (ret
== -EEXIST
) {
1120 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1121 struct btrfs_extent_data_ref
);
1122 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1125 btrfs_release_path(root
, path
);
1127 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1129 if (ret
&& ret
!= -EEXIST
)
1132 leaf
= path
->nodes
[0];
1134 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1135 struct btrfs_extent_data_ref
);
1137 btrfs_set_extent_data_ref_root(leaf
, ref
,
1139 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
1140 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
1141 btrfs_set_extent_data_ref_count(leaf
, ref
, refs_to_add
);
1143 num_refs
= btrfs_extent_data_ref_count(leaf
, ref
);
1144 num_refs
+= refs_to_add
;
1145 btrfs_set_extent_data_ref_count(leaf
, ref
, num_refs
);
1148 btrfs_mark_buffer_dirty(leaf
);
1151 btrfs_release_path(root
, path
);
1155 static noinline
int remove_extent_data_ref(struct btrfs_trans_handle
*trans
,
1156 struct btrfs_root
*root
,
1157 struct btrfs_path
*path
,
1160 struct btrfs_key key
;
1161 struct btrfs_extent_data_ref
*ref1
= NULL
;
1162 struct btrfs_shared_data_ref
*ref2
= NULL
;
1163 struct extent_buffer
*leaf
;
1167 leaf
= path
->nodes
[0];
1168 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1170 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1171 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1172 struct btrfs_extent_data_ref
);
1173 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1174 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1175 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1176 struct btrfs_shared_data_ref
);
1177 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1178 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1179 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1180 struct btrfs_extent_ref_v0
*ref0
;
1181 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1182 struct btrfs_extent_ref_v0
);
1183 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1189 BUG_ON(num_refs
< refs_to_drop
);
1190 num_refs
-= refs_to_drop
;
1192 if (num_refs
== 0) {
1193 ret
= btrfs_del_item(trans
, root
, path
);
1195 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
)
1196 btrfs_set_extent_data_ref_count(leaf
, ref1
, num_refs
);
1197 else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
)
1198 btrfs_set_shared_data_ref_count(leaf
, ref2
, num_refs
);
1199 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1201 struct btrfs_extent_ref_v0
*ref0
;
1202 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1203 struct btrfs_extent_ref_v0
);
1204 btrfs_set_ref_count_v0(leaf
, ref0
, num_refs
);
1207 btrfs_mark_buffer_dirty(leaf
);
1212 static noinline u32
extent_data_ref_count(struct btrfs_root
*root
,
1213 struct btrfs_path
*path
,
1214 struct btrfs_extent_inline_ref
*iref
)
1216 struct btrfs_key key
;
1217 struct extent_buffer
*leaf
;
1218 struct btrfs_extent_data_ref
*ref1
;
1219 struct btrfs_shared_data_ref
*ref2
;
1222 leaf
= path
->nodes
[0];
1223 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1225 if (btrfs_extent_inline_ref_type(leaf
, iref
) ==
1226 BTRFS_EXTENT_DATA_REF_KEY
) {
1227 ref1
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1228 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1230 ref2
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1231 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1233 } else if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1234 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1235 struct btrfs_extent_data_ref
);
1236 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1237 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1238 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1239 struct btrfs_shared_data_ref
);
1240 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1241 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1242 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1243 struct btrfs_extent_ref_v0
*ref0
;
1244 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1245 struct btrfs_extent_ref_v0
);
1246 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1254 static noinline
int lookup_tree_block_ref(struct btrfs_trans_handle
*trans
,
1255 struct btrfs_root
*root
,
1256 struct btrfs_path
*path
,
1257 u64 bytenr
, u64 parent
,
1260 struct btrfs_key key
;
1263 key
.objectid
= bytenr
;
1265 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1266 key
.offset
= parent
;
1268 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1269 key
.offset
= root_objectid
;
1272 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1275 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1276 if (ret
== -ENOENT
&& parent
) {
1277 btrfs_release_path(root
, path
);
1278 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1279 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1287 static noinline
int insert_tree_block_ref(struct btrfs_trans_handle
*trans
,
1288 struct btrfs_root
*root
,
1289 struct btrfs_path
*path
,
1290 u64 bytenr
, u64 parent
,
1293 struct btrfs_key key
;
1296 key
.objectid
= bytenr
;
1298 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1299 key
.offset
= parent
;
1301 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1302 key
.offset
= root_objectid
;
1305 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
1306 btrfs_release_path(root
, path
);
1310 static inline int extent_ref_type(u64 parent
, u64 owner
)
1313 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1315 type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1317 type
= BTRFS_TREE_BLOCK_REF_KEY
;
1320 type
= BTRFS_SHARED_DATA_REF_KEY
;
1322 type
= BTRFS_EXTENT_DATA_REF_KEY
;
1327 static int find_next_key(struct btrfs_path
*path
, int level
,
1328 struct btrfs_key
*key
)
1331 for (; level
< BTRFS_MAX_LEVEL
; level
++) {
1332 if (!path
->nodes
[level
])
1334 if (path
->slots
[level
] + 1 >=
1335 btrfs_header_nritems(path
->nodes
[level
]))
1338 btrfs_item_key_to_cpu(path
->nodes
[level
], key
,
1339 path
->slots
[level
] + 1);
1341 btrfs_node_key_to_cpu(path
->nodes
[level
], key
,
1342 path
->slots
[level
] + 1);
1349 * look for inline back ref. if back ref is found, *ref_ret is set
1350 * to the address of inline back ref, and 0 is returned.
1352 * if back ref isn't found, *ref_ret is set to the address where it
1353 * should be inserted, and -ENOENT is returned.
1355 * if insert is true and there are too many inline back refs, the path
1356 * points to the extent item, and -EAGAIN is returned.
1358 * NOTE: inline back refs are ordered in the same way that back ref
1359 * items in the tree are ordered.
1361 static noinline_for_stack
1362 int lookup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1363 struct btrfs_root
*root
,
1364 struct btrfs_path
*path
,
1365 struct btrfs_extent_inline_ref
**ref_ret
,
1366 u64 bytenr
, u64 num_bytes
,
1367 u64 parent
, u64 root_objectid
,
1368 u64 owner
, u64 offset
, int insert
)
1370 struct btrfs_key key
;
1371 struct extent_buffer
*leaf
;
1372 struct btrfs_extent_item
*ei
;
1373 struct btrfs_extent_inline_ref
*iref
;
1384 key
.objectid
= bytenr
;
1385 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1386 key
.offset
= num_bytes
;
1388 want
= extent_ref_type(parent
, owner
);
1390 extra_size
= btrfs_extent_inline_ref_size(want
);
1391 path
->keep_locks
= 1;
1394 ret
= btrfs_search_slot(trans
, root
, &key
, path
, extra_size
, 1);
1401 leaf
= path
->nodes
[0];
1402 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1403 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1404 if (item_size
< sizeof(*ei
)) {
1409 ret
= convert_extent_item_v0(trans
, root
, path
, owner
,
1415 leaf
= path
->nodes
[0];
1416 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1419 BUG_ON(item_size
< sizeof(*ei
));
1421 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1422 flags
= btrfs_extent_flags(leaf
, ei
);
1424 ptr
= (unsigned long)(ei
+ 1);
1425 end
= (unsigned long)ei
+ item_size
;
1427 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
) {
1428 ptr
+= sizeof(struct btrfs_tree_block_info
);
1431 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_DATA
));
1440 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1441 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1445 ptr
+= btrfs_extent_inline_ref_size(type
);
1449 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1450 struct btrfs_extent_data_ref
*dref
;
1451 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1452 if (match_extent_data_ref(leaf
, dref
, root_objectid
,
1457 if (hash_extent_data_ref_item(leaf
, dref
) <
1458 hash_extent_data_ref(root_objectid
, owner
, offset
))
1462 ref_offset
= btrfs_extent_inline_ref_offset(leaf
, iref
);
1464 if (parent
== ref_offset
) {
1468 if (ref_offset
< parent
)
1471 if (root_objectid
== ref_offset
) {
1475 if (ref_offset
< root_objectid
)
1479 ptr
+= btrfs_extent_inline_ref_size(type
);
1481 if (err
== -ENOENT
&& insert
) {
1482 if (item_size
+ extra_size
>=
1483 BTRFS_MAX_EXTENT_ITEM_SIZE(root
)) {
1488 * To add new inline back ref, we have to make sure
1489 * there is no corresponding back ref item.
1490 * For simplicity, we just do not add new inline back
1491 * ref if there is any kind of item for this block
1493 if (find_next_key(path
, 0, &key
) == 0 &&
1494 key
.objectid
== bytenr
&&
1495 key
.type
< BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1500 *ref_ret
= (struct btrfs_extent_inline_ref
*)ptr
;
1503 path
->keep_locks
= 0;
1504 btrfs_unlock_up_safe(path
, 1);
1510 * helper to add new inline back ref
1512 static noinline_for_stack
1513 int setup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1514 struct btrfs_root
*root
,
1515 struct btrfs_path
*path
,
1516 struct btrfs_extent_inline_ref
*iref
,
1517 u64 parent
, u64 root_objectid
,
1518 u64 owner
, u64 offset
, int refs_to_add
,
1519 struct btrfs_delayed_extent_op
*extent_op
)
1521 struct extent_buffer
*leaf
;
1522 struct btrfs_extent_item
*ei
;
1525 unsigned long item_offset
;
1531 leaf
= path
->nodes
[0];
1532 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1533 item_offset
= (unsigned long)iref
- (unsigned long)ei
;
1535 type
= extent_ref_type(parent
, owner
);
1536 size
= btrfs_extent_inline_ref_size(type
);
1538 ret
= btrfs_extend_item(trans
, root
, path
, size
);
1541 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1542 refs
= btrfs_extent_refs(leaf
, ei
);
1543 refs
+= refs_to_add
;
1544 btrfs_set_extent_refs(leaf
, ei
, refs
);
1546 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1548 ptr
= (unsigned long)ei
+ item_offset
;
1549 end
= (unsigned long)ei
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
1550 if (ptr
< end
- size
)
1551 memmove_extent_buffer(leaf
, ptr
+ size
, ptr
,
1554 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1555 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
1556 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1557 struct btrfs_extent_data_ref
*dref
;
1558 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1559 btrfs_set_extent_data_ref_root(leaf
, dref
, root_objectid
);
1560 btrfs_set_extent_data_ref_objectid(leaf
, dref
, owner
);
1561 btrfs_set_extent_data_ref_offset(leaf
, dref
, offset
);
1562 btrfs_set_extent_data_ref_count(leaf
, dref
, refs_to_add
);
1563 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1564 struct btrfs_shared_data_ref
*sref
;
1565 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1566 btrfs_set_shared_data_ref_count(leaf
, sref
, refs_to_add
);
1567 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1568 } else if (type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
1569 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1571 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
1573 btrfs_mark_buffer_dirty(leaf
);
1577 static int lookup_extent_backref(struct btrfs_trans_handle
*trans
,
1578 struct btrfs_root
*root
,
1579 struct btrfs_path
*path
,
1580 struct btrfs_extent_inline_ref
**ref_ret
,
1581 u64 bytenr
, u64 num_bytes
, u64 parent
,
1582 u64 root_objectid
, u64 owner
, u64 offset
)
1586 ret
= lookup_inline_extent_backref(trans
, root
, path
, ref_ret
,
1587 bytenr
, num_bytes
, parent
,
1588 root_objectid
, owner
, offset
, 0);
1592 btrfs_release_path(root
, path
);
1595 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1596 ret
= lookup_tree_block_ref(trans
, root
, path
, bytenr
, parent
,
1599 ret
= lookup_extent_data_ref(trans
, root
, path
, bytenr
, parent
,
1600 root_objectid
, owner
, offset
);
1606 * helper to update/remove inline back ref
1608 static noinline_for_stack
1609 int update_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1610 struct btrfs_root
*root
,
1611 struct btrfs_path
*path
,
1612 struct btrfs_extent_inline_ref
*iref
,
1614 struct btrfs_delayed_extent_op
*extent_op
)
1616 struct extent_buffer
*leaf
;
1617 struct btrfs_extent_item
*ei
;
1618 struct btrfs_extent_data_ref
*dref
= NULL
;
1619 struct btrfs_shared_data_ref
*sref
= NULL
;
1628 leaf
= path
->nodes
[0];
1629 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1630 refs
= btrfs_extent_refs(leaf
, ei
);
1631 WARN_ON(refs_to_mod
< 0 && refs
+ refs_to_mod
<= 0);
1632 refs
+= refs_to_mod
;
1633 btrfs_set_extent_refs(leaf
, ei
, refs
);
1635 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1637 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1639 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1640 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1641 refs
= btrfs_extent_data_ref_count(leaf
, dref
);
1642 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1643 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1644 refs
= btrfs_shared_data_ref_count(leaf
, sref
);
1647 BUG_ON(refs_to_mod
!= -1);
1650 BUG_ON(refs_to_mod
< 0 && refs
< -refs_to_mod
);
1651 refs
+= refs_to_mod
;
1654 if (type
== BTRFS_EXTENT_DATA_REF_KEY
)
1655 btrfs_set_extent_data_ref_count(leaf
, dref
, refs
);
1657 btrfs_set_shared_data_ref_count(leaf
, sref
, refs
);
1659 size
= btrfs_extent_inline_ref_size(type
);
1660 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1661 ptr
= (unsigned long)iref
;
1662 end
= (unsigned long)ei
+ item_size
;
1663 if (ptr
+ size
< end
)
1664 memmove_extent_buffer(leaf
, ptr
, ptr
+ size
,
1667 ret
= btrfs_truncate_item(trans
, root
, path
, item_size
, 1);
1670 btrfs_mark_buffer_dirty(leaf
);
1674 static noinline_for_stack
1675 int insert_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1676 struct btrfs_root
*root
,
1677 struct btrfs_path
*path
,
1678 u64 bytenr
, u64 num_bytes
, u64 parent
,
1679 u64 root_objectid
, u64 owner
,
1680 u64 offset
, int refs_to_add
,
1681 struct btrfs_delayed_extent_op
*extent_op
)
1683 struct btrfs_extent_inline_ref
*iref
;
1686 ret
= lookup_inline_extent_backref(trans
, root
, path
, &iref
,
1687 bytenr
, num_bytes
, parent
,
1688 root_objectid
, owner
, offset
, 1);
1690 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
);
1691 ret
= update_inline_extent_backref(trans
, root
, path
, iref
,
1692 refs_to_add
, extent_op
);
1693 } else if (ret
== -ENOENT
) {
1694 ret
= setup_inline_extent_backref(trans
, root
, path
, iref
,
1695 parent
, root_objectid
,
1696 owner
, offset
, refs_to_add
,
1702 static int insert_extent_backref(struct btrfs_trans_handle
*trans
,
1703 struct btrfs_root
*root
,
1704 struct btrfs_path
*path
,
1705 u64 bytenr
, u64 parent
, u64 root_objectid
,
1706 u64 owner
, u64 offset
, int refs_to_add
)
1709 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1710 BUG_ON(refs_to_add
!= 1);
1711 ret
= insert_tree_block_ref(trans
, root
, path
, bytenr
,
1712 parent
, root_objectid
);
1714 ret
= insert_extent_data_ref(trans
, root
, path
, bytenr
,
1715 parent
, root_objectid
,
1716 owner
, offset
, refs_to_add
);
1721 static int remove_extent_backref(struct btrfs_trans_handle
*trans
,
1722 struct btrfs_root
*root
,
1723 struct btrfs_path
*path
,
1724 struct btrfs_extent_inline_ref
*iref
,
1725 int refs_to_drop
, int is_data
)
1729 BUG_ON(!is_data
&& refs_to_drop
!= 1);
1731 ret
= update_inline_extent_backref(trans
, root
, path
, iref
,
1732 -refs_to_drop
, NULL
);
1733 } else if (is_data
) {
1734 ret
= remove_extent_data_ref(trans
, root
, path
, refs_to_drop
);
1736 ret
= btrfs_del_item(trans
, root
, path
);
1741 static int btrfs_issue_discard(struct block_device
*bdev
,
1744 return blkdev_issue_discard(bdev
, start
>> 9, len
>> 9, GFP_NOFS
, 0);
1747 static int btrfs_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
1748 u64 num_bytes
, u64
*actual_bytes
)
1751 u64 discarded_bytes
= 0;
1752 struct btrfs_multi_bio
*multi
= NULL
;
1755 /* Tell the block device(s) that the sectors can be discarded */
1756 ret
= btrfs_map_block(&root
->fs_info
->mapping_tree
, REQ_DISCARD
,
1757 bytenr
, &num_bytes
, &multi
, 0);
1759 struct btrfs_bio_stripe
*stripe
= multi
->stripes
;
1763 for (i
= 0; i
< multi
->num_stripes
; i
++, stripe
++) {
1764 ret
= btrfs_issue_discard(stripe
->dev
->bdev
,
1768 discarded_bytes
+= stripe
->length
;
1769 else if (ret
!= -EOPNOTSUPP
)
1774 if (discarded_bytes
&& ret
== -EOPNOTSUPP
)
1778 *actual_bytes
= discarded_bytes
;
1784 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1785 struct btrfs_root
*root
,
1786 u64 bytenr
, u64 num_bytes
, u64 parent
,
1787 u64 root_objectid
, u64 owner
, u64 offset
)
1790 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
&&
1791 root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
1793 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1794 ret
= btrfs_add_delayed_tree_ref(trans
, bytenr
, num_bytes
,
1795 parent
, root_objectid
, (int)owner
,
1796 BTRFS_ADD_DELAYED_REF
, NULL
);
1798 ret
= btrfs_add_delayed_data_ref(trans
, bytenr
, num_bytes
,
1799 parent
, root_objectid
, owner
, offset
,
1800 BTRFS_ADD_DELAYED_REF
, NULL
);
1805 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1806 struct btrfs_root
*root
,
1807 u64 bytenr
, u64 num_bytes
,
1808 u64 parent
, u64 root_objectid
,
1809 u64 owner
, u64 offset
, int refs_to_add
,
1810 struct btrfs_delayed_extent_op
*extent_op
)
1812 struct btrfs_path
*path
;
1813 struct extent_buffer
*leaf
;
1814 struct btrfs_extent_item
*item
;
1819 path
= btrfs_alloc_path();
1824 path
->leave_spinning
= 1;
1825 /* this will setup the path even if it fails to insert the back ref */
1826 ret
= insert_inline_extent_backref(trans
, root
->fs_info
->extent_root
,
1827 path
, bytenr
, num_bytes
, parent
,
1828 root_objectid
, owner
, offset
,
1829 refs_to_add
, extent_op
);
1833 if (ret
!= -EAGAIN
) {
1838 leaf
= path
->nodes
[0];
1839 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1840 refs
= btrfs_extent_refs(leaf
, item
);
1841 btrfs_set_extent_refs(leaf
, item
, refs
+ refs_to_add
);
1843 __run_delayed_extent_op(extent_op
, leaf
, item
);
1845 btrfs_mark_buffer_dirty(leaf
);
1846 btrfs_release_path(root
->fs_info
->extent_root
, path
);
1849 path
->leave_spinning
= 1;
1851 /* now insert the actual backref */
1852 ret
= insert_extent_backref(trans
, root
->fs_info
->extent_root
,
1853 path
, bytenr
, parent
, root_objectid
,
1854 owner
, offset
, refs_to_add
);
1857 btrfs_free_path(path
);
1861 static int run_delayed_data_ref(struct btrfs_trans_handle
*trans
,
1862 struct btrfs_root
*root
,
1863 struct btrfs_delayed_ref_node
*node
,
1864 struct btrfs_delayed_extent_op
*extent_op
,
1865 int insert_reserved
)
1868 struct btrfs_delayed_data_ref
*ref
;
1869 struct btrfs_key ins
;
1874 ins
.objectid
= node
->bytenr
;
1875 ins
.offset
= node
->num_bytes
;
1876 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1878 ref
= btrfs_delayed_node_to_data_ref(node
);
1879 if (node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
1880 parent
= ref
->parent
;
1882 ref_root
= ref
->root
;
1884 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
1886 BUG_ON(extent_op
->update_key
);
1887 flags
|= extent_op
->flags_to_set
;
1889 ret
= alloc_reserved_file_extent(trans
, root
,
1890 parent
, ref_root
, flags
,
1891 ref
->objectid
, ref
->offset
,
1892 &ins
, node
->ref_mod
);
1893 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
1894 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
1895 node
->num_bytes
, parent
,
1896 ref_root
, ref
->objectid
,
1897 ref
->offset
, node
->ref_mod
,
1899 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
1900 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
1901 node
->num_bytes
, parent
,
1902 ref_root
, ref
->objectid
,
1903 ref
->offset
, node
->ref_mod
,
1911 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
1912 struct extent_buffer
*leaf
,
1913 struct btrfs_extent_item
*ei
)
1915 u64 flags
= btrfs_extent_flags(leaf
, ei
);
1916 if (extent_op
->update_flags
) {
1917 flags
|= extent_op
->flags_to_set
;
1918 btrfs_set_extent_flags(leaf
, ei
, flags
);
1921 if (extent_op
->update_key
) {
1922 struct btrfs_tree_block_info
*bi
;
1923 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
));
1924 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
1925 btrfs_set_tree_block_key(leaf
, bi
, &extent_op
->key
);
1929 static int run_delayed_extent_op(struct btrfs_trans_handle
*trans
,
1930 struct btrfs_root
*root
,
1931 struct btrfs_delayed_ref_node
*node
,
1932 struct btrfs_delayed_extent_op
*extent_op
)
1934 struct btrfs_key key
;
1935 struct btrfs_path
*path
;
1936 struct btrfs_extent_item
*ei
;
1937 struct extent_buffer
*leaf
;
1942 path
= btrfs_alloc_path();
1946 key
.objectid
= node
->bytenr
;
1947 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1948 key
.offset
= node
->num_bytes
;
1951 path
->leave_spinning
= 1;
1952 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
,
1963 leaf
= path
->nodes
[0];
1964 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1965 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1966 if (item_size
< sizeof(*ei
)) {
1967 ret
= convert_extent_item_v0(trans
, root
->fs_info
->extent_root
,
1973 leaf
= path
->nodes
[0];
1974 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1977 BUG_ON(item_size
< sizeof(*ei
));
1978 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1979 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1981 btrfs_mark_buffer_dirty(leaf
);
1983 btrfs_free_path(path
);
1987 static int run_delayed_tree_ref(struct btrfs_trans_handle
*trans
,
1988 struct btrfs_root
*root
,
1989 struct btrfs_delayed_ref_node
*node
,
1990 struct btrfs_delayed_extent_op
*extent_op
,
1991 int insert_reserved
)
1994 struct btrfs_delayed_tree_ref
*ref
;
1995 struct btrfs_key ins
;
1999 ins
.objectid
= node
->bytenr
;
2000 ins
.offset
= node
->num_bytes
;
2001 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2003 ref
= btrfs_delayed_node_to_tree_ref(node
);
2004 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2005 parent
= ref
->parent
;
2007 ref_root
= ref
->root
;
2009 BUG_ON(node
->ref_mod
!= 1);
2010 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2011 BUG_ON(!extent_op
|| !extent_op
->update_flags
||
2012 !extent_op
->update_key
);
2013 ret
= alloc_reserved_tree_block(trans
, root
,
2015 extent_op
->flags_to_set
,
2018 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2019 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
2020 node
->num_bytes
, parent
, ref_root
,
2021 ref
->level
, 0, 1, extent_op
);
2022 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2023 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
2024 node
->num_bytes
, parent
, ref_root
,
2025 ref
->level
, 0, 1, extent_op
);
2032 /* helper function to actually process a single delayed ref entry */
2033 static int run_one_delayed_ref(struct btrfs_trans_handle
*trans
,
2034 struct btrfs_root
*root
,
2035 struct btrfs_delayed_ref_node
*node
,
2036 struct btrfs_delayed_extent_op
*extent_op
,
2037 int insert_reserved
)
2040 if (btrfs_delayed_ref_is_head(node
)) {
2041 struct btrfs_delayed_ref_head
*head
;
2043 * we've hit the end of the chain and we were supposed
2044 * to insert this extent into the tree. But, it got
2045 * deleted before we ever needed to insert it, so all
2046 * we have to do is clean up the accounting
2049 head
= btrfs_delayed_node_to_head(node
);
2050 if (insert_reserved
) {
2051 btrfs_pin_extent(root
, node
->bytenr
,
2052 node
->num_bytes
, 1);
2053 if (head
->is_data
) {
2054 ret
= btrfs_del_csums(trans
, root
,
2060 mutex_unlock(&head
->mutex
);
2064 if (node
->type
== BTRFS_TREE_BLOCK_REF_KEY
||
2065 node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2066 ret
= run_delayed_tree_ref(trans
, root
, node
, extent_op
,
2068 else if (node
->type
== BTRFS_EXTENT_DATA_REF_KEY
||
2069 node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2070 ret
= run_delayed_data_ref(trans
, root
, node
, extent_op
,
2077 static noinline
struct btrfs_delayed_ref_node
*
2078 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
2080 struct rb_node
*node
;
2081 struct btrfs_delayed_ref_node
*ref
;
2082 int action
= BTRFS_ADD_DELAYED_REF
;
2085 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2086 * this prevents ref count from going down to zero when
2087 * there still are pending delayed ref.
2089 node
= rb_prev(&head
->node
.rb_node
);
2093 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
2095 if (ref
->bytenr
!= head
->node
.bytenr
)
2097 if (ref
->action
== action
)
2099 node
= rb_prev(node
);
2101 if (action
== BTRFS_ADD_DELAYED_REF
) {
2102 action
= BTRFS_DROP_DELAYED_REF
;
2108 static noinline
int run_clustered_refs(struct btrfs_trans_handle
*trans
,
2109 struct btrfs_root
*root
,
2110 struct list_head
*cluster
)
2112 struct btrfs_delayed_ref_root
*delayed_refs
;
2113 struct btrfs_delayed_ref_node
*ref
;
2114 struct btrfs_delayed_ref_head
*locked_ref
= NULL
;
2115 struct btrfs_delayed_extent_op
*extent_op
;
2118 int must_insert_reserved
= 0;
2120 delayed_refs
= &trans
->transaction
->delayed_refs
;
2123 /* pick a new head ref from the cluster list */
2124 if (list_empty(cluster
))
2127 locked_ref
= list_entry(cluster
->next
,
2128 struct btrfs_delayed_ref_head
, cluster
);
2130 /* grab the lock that says we are going to process
2131 * all the refs for this head */
2132 ret
= btrfs_delayed_ref_lock(trans
, locked_ref
);
2135 * we may have dropped the spin lock to get the head
2136 * mutex lock, and that might have given someone else
2137 * time to free the head. If that's true, it has been
2138 * removed from our list and we can move on.
2140 if (ret
== -EAGAIN
) {
2148 * record the must insert reserved flag before we
2149 * drop the spin lock.
2151 must_insert_reserved
= locked_ref
->must_insert_reserved
;
2152 locked_ref
->must_insert_reserved
= 0;
2154 extent_op
= locked_ref
->extent_op
;
2155 locked_ref
->extent_op
= NULL
;
2158 * locked_ref is the head node, so we have to go one
2159 * node back for any delayed ref updates
2161 ref
= select_delayed_ref(locked_ref
);
2163 /* All delayed refs have been processed, Go ahead
2164 * and send the head node to run_one_delayed_ref,
2165 * so that any accounting fixes can happen
2167 ref
= &locked_ref
->node
;
2169 if (extent_op
&& must_insert_reserved
) {
2175 spin_unlock(&delayed_refs
->lock
);
2177 ret
= run_delayed_extent_op(trans
, root
,
2183 spin_lock(&delayed_refs
->lock
);
2187 list_del_init(&locked_ref
->cluster
);
2192 rb_erase(&ref
->rb_node
, &delayed_refs
->root
);
2193 delayed_refs
->num_entries
--;
2195 spin_unlock(&delayed_refs
->lock
);
2197 ret
= run_one_delayed_ref(trans
, root
, ref
, extent_op
,
2198 must_insert_reserved
);
2201 btrfs_put_delayed_ref(ref
);
2206 spin_lock(&delayed_refs
->lock
);
2212 * this starts processing the delayed reference count updates and
2213 * extent insertions we have queued up so far. count can be
2214 * 0, which means to process everything in the tree at the start
2215 * of the run (but not newly added entries), or it can be some target
2216 * number you'd like to process.
2218 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2219 struct btrfs_root
*root
, unsigned long count
)
2221 struct rb_node
*node
;
2222 struct btrfs_delayed_ref_root
*delayed_refs
;
2223 struct btrfs_delayed_ref_node
*ref
;
2224 struct list_head cluster
;
2226 int run_all
= count
== (unsigned long)-1;
2229 if (root
== root
->fs_info
->extent_root
)
2230 root
= root
->fs_info
->tree_root
;
2232 delayed_refs
= &trans
->transaction
->delayed_refs
;
2233 INIT_LIST_HEAD(&cluster
);
2235 spin_lock(&delayed_refs
->lock
);
2237 count
= delayed_refs
->num_entries
* 2;
2241 if (!(run_all
|| run_most
) &&
2242 delayed_refs
->num_heads_ready
< 64)
2246 * go find something we can process in the rbtree. We start at
2247 * the beginning of the tree, and then build a cluster
2248 * of refs to process starting at the first one we are able to
2251 ret
= btrfs_find_ref_cluster(trans
, &cluster
,
2252 delayed_refs
->run_delayed_start
);
2256 ret
= run_clustered_refs(trans
, root
, &cluster
);
2259 count
-= min_t(unsigned long, ret
, count
);
2266 node
= rb_first(&delayed_refs
->root
);
2269 count
= (unsigned long)-1;
2272 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
2274 if (btrfs_delayed_ref_is_head(ref
)) {
2275 struct btrfs_delayed_ref_head
*head
;
2277 head
= btrfs_delayed_node_to_head(ref
);
2278 atomic_inc(&ref
->refs
);
2280 spin_unlock(&delayed_refs
->lock
);
2281 mutex_lock(&head
->mutex
);
2282 mutex_unlock(&head
->mutex
);
2284 btrfs_put_delayed_ref(ref
);
2288 node
= rb_next(node
);
2290 spin_unlock(&delayed_refs
->lock
);
2291 schedule_timeout(1);
2295 spin_unlock(&delayed_refs
->lock
);
2299 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
2300 struct btrfs_root
*root
,
2301 u64 bytenr
, u64 num_bytes
, u64 flags
,
2304 struct btrfs_delayed_extent_op
*extent_op
;
2307 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
2311 extent_op
->flags_to_set
= flags
;
2312 extent_op
->update_flags
= 1;
2313 extent_op
->update_key
= 0;
2314 extent_op
->is_data
= is_data
? 1 : 0;
2316 ret
= btrfs_add_delayed_extent_op(trans
, bytenr
, num_bytes
, extent_op
);
2322 static noinline
int check_delayed_ref(struct btrfs_trans_handle
*trans
,
2323 struct btrfs_root
*root
,
2324 struct btrfs_path
*path
,
2325 u64 objectid
, u64 offset
, u64 bytenr
)
2327 struct btrfs_delayed_ref_head
*head
;
2328 struct btrfs_delayed_ref_node
*ref
;
2329 struct btrfs_delayed_data_ref
*data_ref
;
2330 struct btrfs_delayed_ref_root
*delayed_refs
;
2331 struct rb_node
*node
;
2335 delayed_refs
= &trans
->transaction
->delayed_refs
;
2336 spin_lock(&delayed_refs
->lock
);
2337 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
2341 if (!mutex_trylock(&head
->mutex
)) {
2342 atomic_inc(&head
->node
.refs
);
2343 spin_unlock(&delayed_refs
->lock
);
2345 btrfs_release_path(root
->fs_info
->extent_root
, path
);
2347 mutex_lock(&head
->mutex
);
2348 mutex_unlock(&head
->mutex
);
2349 btrfs_put_delayed_ref(&head
->node
);
2353 node
= rb_prev(&head
->node
.rb_node
);
2357 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2359 if (ref
->bytenr
!= bytenr
)
2363 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
)
2366 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
2368 node
= rb_prev(node
);
2370 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2371 if (ref
->bytenr
== bytenr
)
2375 if (data_ref
->root
!= root
->root_key
.objectid
||
2376 data_ref
->objectid
!= objectid
|| data_ref
->offset
!= offset
)
2381 mutex_unlock(&head
->mutex
);
2383 spin_unlock(&delayed_refs
->lock
);
2387 static noinline
int check_committed_ref(struct btrfs_trans_handle
*trans
,
2388 struct btrfs_root
*root
,
2389 struct btrfs_path
*path
,
2390 u64 objectid
, u64 offset
, u64 bytenr
)
2392 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2393 struct extent_buffer
*leaf
;
2394 struct btrfs_extent_data_ref
*ref
;
2395 struct btrfs_extent_inline_ref
*iref
;
2396 struct btrfs_extent_item
*ei
;
2397 struct btrfs_key key
;
2401 key
.objectid
= bytenr
;
2402 key
.offset
= (u64
)-1;
2403 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2405 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
2411 if (path
->slots
[0] == 0)
2415 leaf
= path
->nodes
[0];
2416 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2418 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
2422 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2423 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2424 if (item_size
< sizeof(*ei
)) {
2425 WARN_ON(item_size
!= sizeof(struct btrfs_extent_item_v0
));
2429 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2431 if (item_size
!= sizeof(*ei
) +
2432 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
2435 if (btrfs_extent_generation(leaf
, ei
) <=
2436 btrfs_root_last_snapshot(&root
->root_item
))
2439 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
2440 if (btrfs_extent_inline_ref_type(leaf
, iref
) !=
2441 BTRFS_EXTENT_DATA_REF_KEY
)
2444 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
2445 if (btrfs_extent_refs(leaf
, ei
) !=
2446 btrfs_extent_data_ref_count(leaf
, ref
) ||
2447 btrfs_extent_data_ref_root(leaf
, ref
) !=
2448 root
->root_key
.objectid
||
2449 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
2450 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
2458 int btrfs_cross_ref_exist(struct btrfs_trans_handle
*trans
,
2459 struct btrfs_root
*root
,
2460 u64 objectid
, u64 offset
, u64 bytenr
)
2462 struct btrfs_path
*path
;
2466 path
= btrfs_alloc_path();
2471 ret
= check_committed_ref(trans
, root
, path
, objectid
,
2473 if (ret
&& ret
!= -ENOENT
)
2476 ret2
= check_delayed_ref(trans
, root
, path
, objectid
,
2478 } while (ret2
== -EAGAIN
);
2480 if (ret2
&& ret2
!= -ENOENT
) {
2485 if (ret
!= -ENOENT
|| ret2
!= -ENOENT
)
2488 btrfs_free_path(path
);
2489 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
2495 int btrfs_cache_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2496 struct extent_buffer
*buf
, u32 nr_extents
)
2498 struct btrfs_key key
;
2499 struct btrfs_file_extent_item
*fi
;
2507 if (!root
->ref_cows
)
2510 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
2512 root_gen
= root
->root_key
.offset
;
2515 root_gen
= trans
->transid
- 1;
2518 level
= btrfs_header_level(buf
);
2519 nritems
= btrfs_header_nritems(buf
);
2522 struct btrfs_leaf_ref
*ref
;
2523 struct btrfs_extent_info
*info
;
2525 ref
= btrfs_alloc_leaf_ref(root
, nr_extents
);
2531 ref
->root_gen
= root_gen
;
2532 ref
->bytenr
= buf
->start
;
2533 ref
->owner
= btrfs_header_owner(buf
);
2534 ref
->generation
= btrfs_header_generation(buf
);
2535 ref
->nritems
= nr_extents
;
2536 info
= ref
->extents
;
2538 for (i
= 0; nr_extents
> 0 && i
< nritems
; i
++) {
2540 btrfs_item_key_to_cpu(buf
, &key
, i
);
2541 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
2543 fi
= btrfs_item_ptr(buf
, i
,
2544 struct btrfs_file_extent_item
);
2545 if (btrfs_file_extent_type(buf
, fi
) ==
2546 BTRFS_FILE_EXTENT_INLINE
)
2548 disk_bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
2549 if (disk_bytenr
== 0)
2552 info
->bytenr
= disk_bytenr
;
2554 btrfs_file_extent_disk_num_bytes(buf
, fi
);
2555 info
->objectid
= key
.objectid
;
2556 info
->offset
= key
.offset
;
2560 ret
= btrfs_add_leaf_ref(root
, ref
, shared
);
2561 if (ret
== -EEXIST
&& shared
) {
2562 struct btrfs_leaf_ref
*old
;
2563 old
= btrfs_lookup_leaf_ref(root
, ref
->bytenr
);
2565 btrfs_remove_leaf_ref(root
, old
);
2566 btrfs_free_leaf_ref(root
, old
);
2567 ret
= btrfs_add_leaf_ref(root
, ref
, shared
);
2570 btrfs_free_leaf_ref(root
, ref
);
2576 /* when a block goes through cow, we update the reference counts of
2577 * everything that block points to. The internal pointers of the block
2578 * can be in just about any order, and it is likely to have clusters of
2579 * things that are close together and clusters of things that are not.
2581 * To help reduce the seeks that come with updating all of these reference
2582 * counts, sort them by byte number before actual updates are done.
2584 * struct refsort is used to match byte number to slot in the btree block.
2585 * we sort based on the byte number and then use the slot to actually
2588 * struct refsort is smaller than strcut btrfs_item and smaller than
2589 * struct btrfs_key_ptr. Since we're currently limited to the page size
2590 * for a btree block, there's no way for a kmalloc of refsorts for a
2591 * single node to be bigger than a page.
2599 * for passing into sort()
2601 static int refsort_cmp(const void *a_void
, const void *b_void
)
2603 const struct refsort
*a
= a_void
;
2604 const struct refsort
*b
= b_void
;
2606 if (a
->bytenr
< b
->bytenr
)
2608 if (a
->bytenr
> b
->bytenr
)
2614 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
2615 struct btrfs_root
*root
,
2616 struct extent_buffer
*buf
,
2617 int full_backref
, int inc
)
2624 struct btrfs_key key
;
2625 struct btrfs_file_extent_item
*fi
;
2629 int (*process_func
)(struct btrfs_trans_handle
*, struct btrfs_root
*,
2630 u64
, u64
, u64
, u64
, u64
, u64
);
2632 ref_root
= btrfs_header_owner(buf
);
2633 nritems
= btrfs_header_nritems(buf
);
2634 level
= btrfs_header_level(buf
);
2636 if (!root
->ref_cows
&& level
== 0)
2640 process_func
= btrfs_inc_extent_ref
;
2642 process_func
= btrfs_free_extent
;
2645 parent
= buf
->start
;
2649 for (i
= 0; i
< nritems
; i
++) {
2651 btrfs_item_key_to_cpu(buf
, &key
, i
);
2652 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
2654 fi
= btrfs_item_ptr(buf
, i
,
2655 struct btrfs_file_extent_item
);
2656 if (btrfs_file_extent_type(buf
, fi
) ==
2657 BTRFS_FILE_EXTENT_INLINE
)
2659 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
2663 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
2664 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
2665 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
2666 parent
, ref_root
, key
.objectid
,
2671 bytenr
= btrfs_node_blockptr(buf
, i
);
2672 num_bytes
= btrfs_level_size(root
, level
- 1);
2673 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
2674 parent
, ref_root
, level
- 1, 0);
2685 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2686 struct extent_buffer
*buf
, int full_backref
)
2688 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1);
2691 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2692 struct extent_buffer
*buf
, int full_backref
)
2694 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0);
2697 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
2698 struct btrfs_root
*root
,
2699 struct btrfs_path
*path
,
2700 struct btrfs_block_group_cache
*cache
)
2703 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2705 struct extent_buffer
*leaf
;
2707 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
2712 leaf
= path
->nodes
[0];
2713 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
2714 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
2715 btrfs_mark_buffer_dirty(leaf
);
2716 btrfs_release_path(extent_root
, path
);
2724 static struct btrfs_block_group_cache
*
2725 next_block_group(struct btrfs_root
*root
,
2726 struct btrfs_block_group_cache
*cache
)
2728 struct rb_node
*node
;
2729 spin_lock(&root
->fs_info
->block_group_cache_lock
);
2730 node
= rb_next(&cache
->cache_node
);
2731 btrfs_put_block_group(cache
);
2733 cache
= rb_entry(node
, struct btrfs_block_group_cache
,
2735 btrfs_get_block_group(cache
);
2738 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
2742 static int cache_save_setup(struct btrfs_block_group_cache
*block_group
,
2743 struct btrfs_trans_handle
*trans
,
2744 struct btrfs_path
*path
)
2746 struct btrfs_root
*root
= block_group
->fs_info
->tree_root
;
2747 struct inode
*inode
= NULL
;
2749 int dcs
= BTRFS_DC_ERROR
;
2755 * If this block group is smaller than 100 megs don't bother caching the
2758 if (block_group
->key
.offset
< (100 * 1024 * 1024)) {
2759 spin_lock(&block_group
->lock
);
2760 block_group
->disk_cache_state
= BTRFS_DC_WRITTEN
;
2761 spin_unlock(&block_group
->lock
);
2766 inode
= lookup_free_space_inode(root
, block_group
, path
);
2767 if (IS_ERR(inode
) && PTR_ERR(inode
) != -ENOENT
) {
2768 ret
= PTR_ERR(inode
);
2769 btrfs_release_path(root
, path
);
2773 if (IS_ERR(inode
)) {
2777 if (block_group
->ro
)
2780 ret
= create_free_space_inode(root
, trans
, block_group
, path
);
2787 * We want to set the generation to 0, that way if anything goes wrong
2788 * from here on out we know not to trust this cache when we load up next
2791 BTRFS_I(inode
)->generation
= 0;
2792 ret
= btrfs_update_inode(trans
, root
, inode
);
2795 if (i_size_read(inode
) > 0) {
2796 ret
= btrfs_truncate_free_space_cache(root
, trans
, path
,
2802 spin_lock(&block_group
->lock
);
2803 if (block_group
->cached
!= BTRFS_CACHE_FINISHED
) {
2804 /* We're not cached, don't bother trying to write stuff out */
2805 dcs
= BTRFS_DC_WRITTEN
;
2806 spin_unlock(&block_group
->lock
);
2809 spin_unlock(&block_group
->lock
);
2811 num_pages
= (int)div64_u64(block_group
->key
.offset
, 1024 * 1024 * 1024);
2816 * Just to make absolutely sure we have enough space, we're going to
2817 * preallocate 12 pages worth of space for each block group. In
2818 * practice we ought to use at most 8, but we need extra space so we can
2819 * add our header and have a terminator between the extents and the
2823 num_pages
*= PAGE_CACHE_SIZE
;
2825 ret
= btrfs_check_data_free_space(inode
, num_pages
);
2829 ret
= btrfs_prealloc_file_range_trans(inode
, trans
, 0, 0, num_pages
,
2830 num_pages
, num_pages
,
2833 dcs
= BTRFS_DC_SETUP
;
2834 btrfs_free_reserved_data_space(inode
, num_pages
);
2838 btrfs_release_path(root
, path
);
2840 spin_lock(&block_group
->lock
);
2841 block_group
->disk_cache_state
= dcs
;
2842 spin_unlock(&block_group
->lock
);
2847 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
2848 struct btrfs_root
*root
)
2850 struct btrfs_block_group_cache
*cache
;
2852 struct btrfs_path
*path
;
2855 path
= btrfs_alloc_path();
2861 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
2863 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
)
2865 cache
= next_block_group(root
, cache
);
2873 err
= cache_save_setup(cache
, trans
, path
);
2874 last
= cache
->key
.objectid
+ cache
->key
.offset
;
2875 btrfs_put_block_group(cache
);
2880 err
= btrfs_run_delayed_refs(trans
, root
,
2885 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
2887 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
) {
2888 btrfs_put_block_group(cache
);
2894 cache
= next_block_group(root
, cache
);
2903 if (cache
->disk_cache_state
== BTRFS_DC_SETUP
)
2904 cache
->disk_cache_state
= BTRFS_DC_NEED_WRITE
;
2906 last
= cache
->key
.objectid
+ cache
->key
.offset
;
2908 err
= write_one_cache_group(trans
, root
, path
, cache
);
2910 btrfs_put_block_group(cache
);
2915 * I don't think this is needed since we're just marking our
2916 * preallocated extent as written, but just in case it can't
2920 err
= btrfs_run_delayed_refs(trans
, root
,
2925 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
2928 * Really this shouldn't happen, but it could if we
2929 * couldn't write the entire preallocated extent and
2930 * splitting the extent resulted in a new block.
2933 btrfs_put_block_group(cache
);
2936 if (cache
->disk_cache_state
== BTRFS_DC_NEED_WRITE
)
2938 cache
= next_block_group(root
, cache
);
2947 btrfs_write_out_cache(root
, trans
, cache
, path
);
2950 * If we didn't have an error then the cache state is still
2951 * NEED_WRITE, so we can set it to WRITTEN.
2953 if (cache
->disk_cache_state
== BTRFS_DC_NEED_WRITE
)
2954 cache
->disk_cache_state
= BTRFS_DC_WRITTEN
;
2955 last
= cache
->key
.objectid
+ cache
->key
.offset
;
2956 btrfs_put_block_group(cache
);
2959 btrfs_free_path(path
);
2963 int btrfs_extent_readonly(struct btrfs_root
*root
, u64 bytenr
)
2965 struct btrfs_block_group_cache
*block_group
;
2968 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
2969 if (!block_group
|| block_group
->ro
)
2972 btrfs_put_block_group(block_group
);
2976 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
2977 u64 total_bytes
, u64 bytes_used
,
2978 struct btrfs_space_info
**space_info
)
2980 struct btrfs_space_info
*found
;
2984 if (flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
2985 BTRFS_BLOCK_GROUP_RAID10
))
2990 found
= __find_space_info(info
, flags
);
2992 spin_lock(&found
->lock
);
2993 found
->total_bytes
+= total_bytes
;
2994 found
->disk_total
+= total_bytes
* factor
;
2995 found
->bytes_used
+= bytes_used
;
2996 found
->disk_used
+= bytes_used
* factor
;
2998 spin_unlock(&found
->lock
);
2999 *space_info
= found
;
3002 found
= kzalloc(sizeof(*found
), GFP_NOFS
);
3006 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
3007 INIT_LIST_HEAD(&found
->block_groups
[i
]);
3008 init_rwsem(&found
->groups_sem
);
3009 spin_lock_init(&found
->lock
);
3010 found
->flags
= flags
& (BTRFS_BLOCK_GROUP_DATA
|
3011 BTRFS_BLOCK_GROUP_SYSTEM
|
3012 BTRFS_BLOCK_GROUP_METADATA
);
3013 found
->total_bytes
= total_bytes
;
3014 found
->disk_total
= total_bytes
* factor
;
3015 found
->bytes_used
= bytes_used
;
3016 found
->disk_used
= bytes_used
* factor
;
3017 found
->bytes_pinned
= 0;
3018 found
->bytes_reserved
= 0;
3019 found
->bytes_readonly
= 0;
3020 found
->bytes_may_use
= 0;
3022 found
->force_alloc
= 0;
3023 *space_info
= found
;
3024 list_add_rcu(&found
->list
, &info
->space_info
);
3025 atomic_set(&found
->caching_threads
, 0);
3029 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
3031 u64 extra_flags
= flags
& (BTRFS_BLOCK_GROUP_RAID0
|
3032 BTRFS_BLOCK_GROUP_RAID1
|
3033 BTRFS_BLOCK_GROUP_RAID10
|
3034 BTRFS_BLOCK_GROUP_DUP
);
3036 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3037 fs_info
->avail_data_alloc_bits
|= extra_flags
;
3038 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3039 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
3040 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3041 fs_info
->avail_system_alloc_bits
|= extra_flags
;
3045 u64
btrfs_reduce_alloc_profile(struct btrfs_root
*root
, u64 flags
)
3048 * we add in the count of missing devices because we want
3049 * to make sure that any RAID levels on a degraded FS
3050 * continue to be honored.
3052 u64 num_devices
= root
->fs_info
->fs_devices
->rw_devices
+
3053 root
->fs_info
->fs_devices
->missing_devices
;
3055 if (num_devices
== 1)
3056 flags
&= ~(BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID0
);
3057 if (num_devices
< 4)
3058 flags
&= ~BTRFS_BLOCK_GROUP_RAID10
;
3060 if ((flags
& BTRFS_BLOCK_GROUP_DUP
) &&
3061 (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
3062 BTRFS_BLOCK_GROUP_RAID10
))) {
3063 flags
&= ~BTRFS_BLOCK_GROUP_DUP
;
3066 if ((flags
& BTRFS_BLOCK_GROUP_RAID1
) &&
3067 (flags
& BTRFS_BLOCK_GROUP_RAID10
)) {
3068 flags
&= ~BTRFS_BLOCK_GROUP_RAID1
;
3071 if ((flags
& BTRFS_BLOCK_GROUP_RAID0
) &&
3072 ((flags
& BTRFS_BLOCK_GROUP_RAID1
) |
3073 (flags
& BTRFS_BLOCK_GROUP_RAID10
) |
3074 (flags
& BTRFS_BLOCK_GROUP_DUP
)))
3075 flags
&= ~BTRFS_BLOCK_GROUP_RAID0
;
3079 static u64
get_alloc_profile(struct btrfs_root
*root
, u64 flags
)
3081 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3082 flags
|= root
->fs_info
->avail_data_alloc_bits
&
3083 root
->fs_info
->data_alloc_profile
;
3084 else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3085 flags
|= root
->fs_info
->avail_system_alloc_bits
&
3086 root
->fs_info
->system_alloc_profile
;
3087 else if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3088 flags
|= root
->fs_info
->avail_metadata_alloc_bits
&
3089 root
->fs_info
->metadata_alloc_profile
;
3090 return btrfs_reduce_alloc_profile(root
, flags
);
3093 u64
btrfs_get_alloc_profile(struct btrfs_root
*root
, int data
)
3098 flags
= BTRFS_BLOCK_GROUP_DATA
;
3099 else if (root
== root
->fs_info
->chunk_root
)
3100 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
3102 flags
= BTRFS_BLOCK_GROUP_METADATA
;
3104 return get_alloc_profile(root
, flags
);
3107 void btrfs_set_inode_space_info(struct btrfs_root
*root
, struct inode
*inode
)
3109 BTRFS_I(inode
)->space_info
= __find_space_info(root
->fs_info
,
3110 BTRFS_BLOCK_GROUP_DATA
);
3114 * This will check the space that the inode allocates from to make sure we have
3115 * enough space for bytes.
3117 int btrfs_check_data_free_space(struct inode
*inode
, u64 bytes
)
3119 struct btrfs_space_info
*data_sinfo
;
3120 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3122 int ret
= 0, committed
= 0, alloc_chunk
= 1;
3124 /* make sure bytes are sectorsize aligned */
3125 bytes
= (bytes
+ root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
3127 if (root
== root
->fs_info
->tree_root
) {
3132 data_sinfo
= BTRFS_I(inode
)->space_info
;
3137 /* make sure we have enough space to handle the data first */
3138 spin_lock(&data_sinfo
->lock
);
3139 used
= data_sinfo
->bytes_used
+ data_sinfo
->bytes_reserved
+
3140 data_sinfo
->bytes_pinned
+ data_sinfo
->bytes_readonly
+
3141 data_sinfo
->bytes_may_use
;
3143 if (used
+ bytes
> data_sinfo
->total_bytes
) {
3144 struct btrfs_trans_handle
*trans
;
3147 * if we don't have enough free bytes in this space then we need
3148 * to alloc a new chunk.
3150 if (!data_sinfo
->full
&& alloc_chunk
) {
3153 data_sinfo
->force_alloc
= 1;
3154 spin_unlock(&data_sinfo
->lock
);
3156 alloc_target
= btrfs_get_alloc_profile(root
, 1);
3157 trans
= btrfs_join_transaction(root
, 1);
3159 return PTR_ERR(trans
);
3161 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
3162 bytes
+ 2 * 1024 * 1024,
3164 btrfs_end_transaction(trans
, root
);
3173 btrfs_set_inode_space_info(root
, inode
);
3174 data_sinfo
= BTRFS_I(inode
)->space_info
;
3178 spin_unlock(&data_sinfo
->lock
);
3180 /* commit the current transaction and try again */
3182 if (!committed
&& !root
->fs_info
->open_ioctl_trans
) {
3184 trans
= btrfs_join_transaction(root
, 1);
3186 return PTR_ERR(trans
);
3187 ret
= btrfs_commit_transaction(trans
, root
);
3193 #if 0 /* I hope we never need this code again, just in case */
3194 printk(KERN_ERR
"no space left, need %llu, %llu bytes_used, "
3195 "%llu bytes_reserved, " "%llu bytes_pinned, "
3196 "%llu bytes_readonly, %llu may use %llu total\n",
3197 (unsigned long long)bytes
,
3198 (unsigned long long)data_sinfo
->bytes_used
,
3199 (unsigned long long)data_sinfo
->bytes_reserved
,
3200 (unsigned long long)data_sinfo
->bytes_pinned
,
3201 (unsigned long long)data_sinfo
->bytes_readonly
,
3202 (unsigned long long)data_sinfo
->bytes_may_use
,
3203 (unsigned long long)data_sinfo
->total_bytes
);
3207 data_sinfo
->bytes_may_use
+= bytes
;
3208 BTRFS_I(inode
)->reserved_bytes
+= bytes
;
3209 spin_unlock(&data_sinfo
->lock
);
3215 * called when we are clearing an delalloc extent from the
3216 * inode's io_tree or there was an error for whatever reason
3217 * after calling btrfs_check_data_free_space
3219 void btrfs_free_reserved_data_space(struct inode
*inode
, u64 bytes
)
3221 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3222 struct btrfs_space_info
*data_sinfo
;
3224 /* make sure bytes are sectorsize aligned */
3225 bytes
= (bytes
+ root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
3227 data_sinfo
= BTRFS_I(inode
)->space_info
;
3228 spin_lock(&data_sinfo
->lock
);
3229 data_sinfo
->bytes_may_use
-= bytes
;
3230 BTRFS_I(inode
)->reserved_bytes
-= bytes
;
3231 spin_unlock(&data_sinfo
->lock
);
3234 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
3236 struct list_head
*head
= &info
->space_info
;
3237 struct btrfs_space_info
*found
;
3240 list_for_each_entry_rcu(found
, head
, list
) {
3241 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
3242 found
->force_alloc
= 1;
3247 static int should_alloc_chunk(struct btrfs_root
*root
,
3248 struct btrfs_space_info
*sinfo
, u64 alloc_bytes
)
3250 u64 num_bytes
= sinfo
->total_bytes
- sinfo
->bytes_readonly
;
3253 if (sinfo
->bytes_used
+ sinfo
->bytes_reserved
+
3254 alloc_bytes
+ 256 * 1024 * 1024 < num_bytes
)
3257 if (sinfo
->bytes_used
+ sinfo
->bytes_reserved
+
3258 alloc_bytes
< div_factor(num_bytes
, 8))
3261 thresh
= btrfs_super_total_bytes(&root
->fs_info
->super_copy
);
3262 thresh
= max_t(u64
, 256 * 1024 * 1024, div_factor_fine(thresh
, 5));
3264 if (num_bytes
> thresh
&& sinfo
->bytes_used
< div_factor(num_bytes
, 3))
3270 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
3271 struct btrfs_root
*extent_root
, u64 alloc_bytes
,
3272 u64 flags
, int force
)
3274 struct btrfs_space_info
*space_info
;
3275 struct btrfs_fs_info
*fs_info
= extent_root
->fs_info
;
3278 mutex_lock(&fs_info
->chunk_mutex
);
3280 flags
= btrfs_reduce_alloc_profile(extent_root
, flags
);
3282 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
3284 ret
= update_space_info(extent_root
->fs_info
, flags
,
3288 BUG_ON(!space_info
);
3290 spin_lock(&space_info
->lock
);
3291 if (space_info
->force_alloc
)
3293 if (space_info
->full
) {
3294 spin_unlock(&space_info
->lock
);
3298 if (!force
&& !should_alloc_chunk(extent_root
, space_info
,
3300 spin_unlock(&space_info
->lock
);
3303 spin_unlock(&space_info
->lock
);
3306 * If we have mixed data/metadata chunks we want to make sure we keep
3307 * allocating mixed chunks instead of individual chunks.
3309 if (btrfs_mixed_space_info(space_info
))
3310 flags
|= (BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
);
3313 * if we're doing a data chunk, go ahead and make sure that
3314 * we keep a reasonable number of metadata chunks allocated in the
3317 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
3318 fs_info
->data_chunk_allocations
++;
3319 if (!(fs_info
->data_chunk_allocations
%
3320 fs_info
->metadata_ratio
))
3321 force_metadata_allocation(fs_info
);
3324 ret
= btrfs_alloc_chunk(trans
, extent_root
, flags
);
3325 spin_lock(&space_info
->lock
);
3327 space_info
->full
= 1;
3330 space_info
->force_alloc
= 0;
3331 spin_unlock(&space_info
->lock
);
3333 mutex_unlock(&extent_root
->fs_info
->chunk_mutex
);
3338 * shrink metadata reservation for delalloc
3340 static int shrink_delalloc(struct btrfs_trans_handle
*trans
,
3341 struct btrfs_root
*root
, u64 to_reclaim
, int sync
)
3343 struct btrfs_block_rsv
*block_rsv
;
3344 struct btrfs_space_info
*space_info
;
3349 int nr_pages
= (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT
;
3351 unsigned long progress
;
3353 block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
3354 space_info
= block_rsv
->space_info
;
3357 reserved
= space_info
->bytes_reserved
;
3358 progress
= space_info
->reservation_progress
;
3363 max_reclaim
= min(reserved
, to_reclaim
);
3365 while (loops
< 1024) {
3366 /* have the flusher threads jump in and do some IO */
3368 nr_pages
= min_t(unsigned long, nr_pages
,
3369 root
->fs_info
->delalloc_bytes
>> PAGE_CACHE_SHIFT
);
3370 writeback_inodes_sb_nr_if_idle(root
->fs_info
->sb
, nr_pages
);
3372 spin_lock(&space_info
->lock
);
3373 if (reserved
> space_info
->bytes_reserved
)
3374 reclaimed
+= reserved
- space_info
->bytes_reserved
;
3375 reserved
= space_info
->bytes_reserved
;
3376 spin_unlock(&space_info
->lock
);
3380 if (reserved
== 0 || reclaimed
>= max_reclaim
)
3383 if (trans
&& trans
->transaction
->blocked
)
3386 time_left
= schedule_timeout_interruptible(1);
3388 /* We were interrupted, exit */
3392 /* we've kicked the IO a few times, if anything has been freed,
3393 * exit. There is no sense in looping here for a long time
3394 * when we really need to commit the transaction, or there are
3395 * just too many writers without enough free space
3400 if (progress
!= space_info
->reservation_progress
)
3405 return reclaimed
>= to_reclaim
;
3409 * Retries tells us how many times we've called reserve_metadata_bytes. The
3410 * idea is if this is the first call (retries == 0) then we will add to our
3411 * reserved count if we can't make the allocation in order to hold our place
3412 * while we go and try and free up space. That way for retries > 1 we don't try
3413 * and add space, we just check to see if the amount of unused space is >= the
3414 * total space, meaning that our reservation is valid.
3416 * However if we don't intend to retry this reservation, pass -1 as retries so
3417 * that it short circuits this logic.
3419 static int reserve_metadata_bytes(struct btrfs_trans_handle
*trans
,
3420 struct btrfs_root
*root
,
3421 struct btrfs_block_rsv
*block_rsv
,
3422 u64 orig_bytes
, int flush
)
3424 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
3426 u64 num_bytes
= orig_bytes
;
3429 bool reserved
= false;
3430 bool committed
= false;
3437 spin_lock(&space_info
->lock
);
3438 unused
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
3439 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
3440 space_info
->bytes_may_use
;
3443 * The idea here is that we've not already over-reserved the block group
3444 * then we can go ahead and save our reservation first and then start
3445 * flushing if we need to. Otherwise if we've already overcommitted
3446 * lets start flushing stuff first and then come back and try to make
3449 if (unused
<= space_info
->total_bytes
) {
3450 unused
= space_info
->total_bytes
- unused
;
3451 if (unused
>= num_bytes
) {
3453 space_info
->bytes_reserved
+= orig_bytes
;
3457 * Ok set num_bytes to orig_bytes since we aren't
3458 * overocmmitted, this way we only try and reclaim what
3461 num_bytes
= orig_bytes
;
3465 * Ok we're over committed, set num_bytes to the overcommitted
3466 * amount plus the amount of bytes that we need for this
3469 num_bytes
= unused
- space_info
->total_bytes
+
3470 (orig_bytes
* (retries
+ 1));
3474 * Couldn't make our reservation, save our place so while we're trying
3475 * to reclaim space we can actually use it instead of somebody else
3476 * stealing it from us.
3478 if (ret
&& !reserved
) {
3479 space_info
->bytes_reserved
+= orig_bytes
;
3483 spin_unlock(&space_info
->lock
);
3492 * We do synchronous shrinking since we don't actually unreserve
3493 * metadata until after the IO is completed.
3495 ret
= shrink_delalloc(trans
, root
, num_bytes
, 1);
3502 * So if we were overcommitted it's possible that somebody else flushed
3503 * out enough space and we simply didn't have enough space to reclaim,
3504 * so go back around and try again.
3511 spin_lock(&space_info
->lock
);
3513 * Not enough space to be reclaimed, don't bother committing the
3516 if (space_info
->bytes_pinned
< orig_bytes
)
3518 spin_unlock(&space_info
->lock
);
3523 if (trans
|| committed
)
3527 trans
= btrfs_join_transaction(root
, 1);
3530 ret
= btrfs_commit_transaction(trans
, root
);
3539 spin_lock(&space_info
->lock
);
3540 space_info
->bytes_reserved
-= orig_bytes
;
3541 spin_unlock(&space_info
->lock
);
3547 static struct btrfs_block_rsv
*get_block_rsv(struct btrfs_trans_handle
*trans
,
3548 struct btrfs_root
*root
)
3550 struct btrfs_block_rsv
*block_rsv
;
3552 block_rsv
= trans
->block_rsv
;
3554 block_rsv
= root
->block_rsv
;
3557 block_rsv
= &root
->fs_info
->empty_block_rsv
;
3562 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
3566 spin_lock(&block_rsv
->lock
);
3567 if (block_rsv
->reserved
>= num_bytes
) {
3568 block_rsv
->reserved
-= num_bytes
;
3569 if (block_rsv
->reserved
< block_rsv
->size
)
3570 block_rsv
->full
= 0;
3573 spin_unlock(&block_rsv
->lock
);
3577 static void block_rsv_add_bytes(struct btrfs_block_rsv
*block_rsv
,
3578 u64 num_bytes
, int update_size
)
3580 spin_lock(&block_rsv
->lock
);
3581 block_rsv
->reserved
+= num_bytes
;
3583 block_rsv
->size
+= num_bytes
;
3584 else if (block_rsv
->reserved
>= block_rsv
->size
)
3585 block_rsv
->full
= 1;
3586 spin_unlock(&block_rsv
->lock
);
3589 void block_rsv_release_bytes(struct btrfs_block_rsv
*block_rsv
,
3590 struct btrfs_block_rsv
*dest
, u64 num_bytes
)
3592 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
3594 spin_lock(&block_rsv
->lock
);
3595 if (num_bytes
== (u64
)-1)
3596 num_bytes
= block_rsv
->size
;
3597 block_rsv
->size
-= num_bytes
;
3598 if (block_rsv
->reserved
>= block_rsv
->size
) {
3599 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
3600 block_rsv
->reserved
= block_rsv
->size
;
3601 block_rsv
->full
= 1;
3605 spin_unlock(&block_rsv
->lock
);
3607 if (num_bytes
> 0) {
3609 spin_lock(&dest
->lock
);
3613 bytes_to_add
= dest
->size
- dest
->reserved
;
3614 bytes_to_add
= min(num_bytes
, bytes_to_add
);
3615 dest
->reserved
+= bytes_to_add
;
3616 if (dest
->reserved
>= dest
->size
)
3618 num_bytes
-= bytes_to_add
;
3620 spin_unlock(&dest
->lock
);
3623 spin_lock(&space_info
->lock
);
3624 space_info
->bytes_reserved
-= num_bytes
;
3625 space_info
->reservation_progress
++;
3626 spin_unlock(&space_info
->lock
);
3631 static int block_rsv_migrate_bytes(struct btrfs_block_rsv
*src
,
3632 struct btrfs_block_rsv
*dst
, u64 num_bytes
)
3636 ret
= block_rsv_use_bytes(src
, num_bytes
);
3640 block_rsv_add_bytes(dst
, num_bytes
, 1);
3644 void btrfs_init_block_rsv(struct btrfs_block_rsv
*rsv
)
3646 memset(rsv
, 0, sizeof(*rsv
));
3647 spin_lock_init(&rsv
->lock
);
3648 atomic_set(&rsv
->usage
, 1);
3650 INIT_LIST_HEAD(&rsv
->list
);
3653 struct btrfs_block_rsv
*btrfs_alloc_block_rsv(struct btrfs_root
*root
)
3655 struct btrfs_block_rsv
*block_rsv
;
3656 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3658 block_rsv
= kmalloc(sizeof(*block_rsv
), GFP_NOFS
);
3662 btrfs_init_block_rsv(block_rsv
);
3663 block_rsv
->space_info
= __find_space_info(fs_info
,
3664 BTRFS_BLOCK_GROUP_METADATA
);
3668 void btrfs_free_block_rsv(struct btrfs_root
*root
,
3669 struct btrfs_block_rsv
*rsv
)
3671 if (rsv
&& atomic_dec_and_test(&rsv
->usage
)) {
3672 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
3679 * make the block_rsv struct be able to capture freed space.
3680 * the captured space will re-add to the the block_rsv struct
3681 * after transaction commit
3683 void btrfs_add_durable_block_rsv(struct btrfs_fs_info
*fs_info
,
3684 struct btrfs_block_rsv
*block_rsv
)
3686 block_rsv
->durable
= 1;
3687 mutex_lock(&fs_info
->durable_block_rsv_mutex
);
3688 list_add_tail(&block_rsv
->list
, &fs_info
->durable_block_rsv_list
);
3689 mutex_unlock(&fs_info
->durable_block_rsv_mutex
);
3692 int btrfs_block_rsv_add(struct btrfs_trans_handle
*trans
,
3693 struct btrfs_root
*root
,
3694 struct btrfs_block_rsv
*block_rsv
,
3702 ret
= reserve_metadata_bytes(trans
, root
, block_rsv
, num_bytes
, 1);
3704 block_rsv_add_bytes(block_rsv
, num_bytes
, 1);
3711 int btrfs_block_rsv_check(struct btrfs_trans_handle
*trans
,
3712 struct btrfs_root
*root
,
3713 struct btrfs_block_rsv
*block_rsv
,
3714 u64 min_reserved
, int min_factor
)
3717 int commit_trans
= 0;
3723 spin_lock(&block_rsv
->lock
);
3725 num_bytes
= div_factor(block_rsv
->size
, min_factor
);
3726 if (min_reserved
> num_bytes
)
3727 num_bytes
= min_reserved
;
3729 if (block_rsv
->reserved
>= num_bytes
) {
3732 num_bytes
-= block_rsv
->reserved
;
3733 if (block_rsv
->durable
&&
3734 block_rsv
->freed
[0] + block_rsv
->freed
[1] >= num_bytes
)
3737 spin_unlock(&block_rsv
->lock
);
3741 if (block_rsv
->refill_used
) {
3742 ret
= reserve_metadata_bytes(trans
, root
, block_rsv
,
3745 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
3754 trans
= btrfs_join_transaction(root
, 1);
3755 BUG_ON(IS_ERR(trans
));
3756 ret
= btrfs_commit_transaction(trans
, root
);
3763 int btrfs_block_rsv_migrate(struct btrfs_block_rsv
*src_rsv
,
3764 struct btrfs_block_rsv
*dst_rsv
,
3767 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
3770 void btrfs_block_rsv_release(struct btrfs_root
*root
,
3771 struct btrfs_block_rsv
*block_rsv
,
3774 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
3775 if (global_rsv
->full
|| global_rsv
== block_rsv
||
3776 block_rsv
->space_info
!= global_rsv
->space_info
)
3778 block_rsv_release_bytes(block_rsv
, global_rsv
, num_bytes
);
3782 * helper to calculate size of global block reservation.
3783 * the desired value is sum of space used by extent tree,
3784 * checksum tree and root tree
3786 static u64
calc_global_metadata_size(struct btrfs_fs_info
*fs_info
)
3788 struct btrfs_space_info
*sinfo
;
3792 int csum_size
= btrfs_super_csum_size(&fs_info
->super_copy
);
3795 * per tree used space accounting can be inaccuracy, so we
3798 spin_lock(&fs_info
->extent_root
->accounting_lock
);
3799 num_bytes
= btrfs_root_used(&fs_info
->extent_root
->root_item
);
3800 spin_unlock(&fs_info
->extent_root
->accounting_lock
);
3802 spin_lock(&fs_info
->csum_root
->accounting_lock
);
3803 num_bytes
+= btrfs_root_used(&fs_info
->csum_root
->root_item
);
3804 spin_unlock(&fs_info
->csum_root
->accounting_lock
);
3806 spin_lock(&fs_info
->tree_root
->accounting_lock
);
3807 num_bytes
+= btrfs_root_used(&fs_info
->tree_root
->root_item
);
3808 spin_unlock(&fs_info
->tree_root
->accounting_lock
);
3810 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_DATA
);
3811 spin_lock(&sinfo
->lock
);
3812 data_used
= sinfo
->bytes_used
;
3813 spin_unlock(&sinfo
->lock
);
3815 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
3816 spin_lock(&sinfo
->lock
);
3817 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_DATA
)
3819 meta_used
= sinfo
->bytes_used
;
3820 spin_unlock(&sinfo
->lock
);
3822 num_bytes
= (data_used
>> fs_info
->sb
->s_blocksize_bits
) *
3824 num_bytes
+= div64_u64(data_used
+ meta_used
, 50);
3826 if (num_bytes
* 3 > meta_used
)
3827 num_bytes
= div64_u64(meta_used
, 3);
3829 return ALIGN(num_bytes
, fs_info
->extent_root
->leafsize
<< 10);
3832 static void update_global_block_rsv(struct btrfs_fs_info
*fs_info
)
3834 struct btrfs_block_rsv
*block_rsv
= &fs_info
->global_block_rsv
;
3835 struct btrfs_space_info
*sinfo
= block_rsv
->space_info
;
3838 num_bytes
= calc_global_metadata_size(fs_info
);
3840 spin_lock(&block_rsv
->lock
);
3841 spin_lock(&sinfo
->lock
);
3843 block_rsv
->size
= num_bytes
;
3845 num_bytes
= sinfo
->bytes_used
+ sinfo
->bytes_pinned
+
3846 sinfo
->bytes_reserved
+ sinfo
->bytes_readonly
+
3847 sinfo
->bytes_may_use
;
3849 if (sinfo
->total_bytes
> num_bytes
) {
3850 num_bytes
= sinfo
->total_bytes
- num_bytes
;
3851 block_rsv
->reserved
+= num_bytes
;
3852 sinfo
->bytes_reserved
+= num_bytes
;
3855 if (block_rsv
->reserved
>= block_rsv
->size
) {
3856 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
3857 sinfo
->bytes_reserved
-= num_bytes
;
3858 sinfo
->reservation_progress
++;
3859 block_rsv
->reserved
= block_rsv
->size
;
3860 block_rsv
->full
= 1;
3863 printk(KERN_INFO
"global block rsv size %llu reserved %llu\n",
3864 block_rsv
->size
, block_rsv
->reserved
);
3866 spin_unlock(&sinfo
->lock
);
3867 spin_unlock(&block_rsv
->lock
);
3870 static void init_global_block_rsv(struct btrfs_fs_info
*fs_info
)
3872 struct btrfs_space_info
*space_info
;
3874 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
3875 fs_info
->chunk_block_rsv
.space_info
= space_info
;
3876 fs_info
->chunk_block_rsv
.priority
= 10;
3878 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
3879 fs_info
->global_block_rsv
.space_info
= space_info
;
3880 fs_info
->global_block_rsv
.priority
= 10;
3881 fs_info
->global_block_rsv
.refill_used
= 1;
3882 fs_info
->delalloc_block_rsv
.space_info
= space_info
;
3883 fs_info
->trans_block_rsv
.space_info
= space_info
;
3884 fs_info
->empty_block_rsv
.space_info
= space_info
;
3885 fs_info
->empty_block_rsv
.priority
= 10;
3887 fs_info
->extent_root
->block_rsv
= &fs_info
->global_block_rsv
;
3888 fs_info
->csum_root
->block_rsv
= &fs_info
->global_block_rsv
;
3889 fs_info
->dev_root
->block_rsv
= &fs_info
->global_block_rsv
;
3890 fs_info
->tree_root
->block_rsv
= &fs_info
->global_block_rsv
;
3891 fs_info
->chunk_root
->block_rsv
= &fs_info
->chunk_block_rsv
;
3893 btrfs_add_durable_block_rsv(fs_info
, &fs_info
->global_block_rsv
);
3895 btrfs_add_durable_block_rsv(fs_info
, &fs_info
->delalloc_block_rsv
);
3897 update_global_block_rsv(fs_info
);
3900 static void release_global_block_rsv(struct btrfs_fs_info
*fs_info
)
3902 block_rsv_release_bytes(&fs_info
->global_block_rsv
, NULL
, (u64
)-1);
3903 WARN_ON(fs_info
->delalloc_block_rsv
.size
> 0);
3904 WARN_ON(fs_info
->delalloc_block_rsv
.reserved
> 0);
3905 WARN_ON(fs_info
->trans_block_rsv
.size
> 0);
3906 WARN_ON(fs_info
->trans_block_rsv
.reserved
> 0);
3907 WARN_ON(fs_info
->chunk_block_rsv
.size
> 0);
3908 WARN_ON(fs_info
->chunk_block_rsv
.reserved
> 0);
3911 static u64
calc_trans_metadata_size(struct btrfs_root
*root
, int num_items
)
3913 return (root
->leafsize
+ root
->nodesize
* (BTRFS_MAX_LEVEL
- 1)) *
3917 int btrfs_trans_reserve_metadata(struct btrfs_trans_handle
*trans
,
3918 struct btrfs_root
*root
,
3924 if (num_items
== 0 || root
->fs_info
->chunk_root
== root
)
3927 num_bytes
= calc_trans_metadata_size(root
, num_items
);
3928 ret
= btrfs_block_rsv_add(trans
, root
, &root
->fs_info
->trans_block_rsv
,
3931 trans
->bytes_reserved
+= num_bytes
;
3932 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
3937 void btrfs_trans_release_metadata(struct btrfs_trans_handle
*trans
,
3938 struct btrfs_root
*root
)
3940 if (!trans
->bytes_reserved
)
3943 BUG_ON(trans
->block_rsv
!= &root
->fs_info
->trans_block_rsv
);
3944 btrfs_block_rsv_release(root
, trans
->block_rsv
,
3945 trans
->bytes_reserved
);
3946 trans
->bytes_reserved
= 0;
3949 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle
*trans
,
3950 struct inode
*inode
)
3952 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3953 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
3954 struct btrfs_block_rsv
*dst_rsv
= root
->orphan_block_rsv
;
3957 * one for deleting orphan item, one for updating inode and
3958 * two for calling btrfs_truncate_inode_items.
3960 * btrfs_truncate_inode_items is a delete operation, it frees
3961 * more space than it uses in most cases. So two units of
3962 * metadata space should be enough for calling it many times.
3963 * If all of the metadata space is used, we can commit
3964 * transaction and use space it freed.
3966 u64 num_bytes
= calc_trans_metadata_size(root
, 4);
3967 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
3970 void btrfs_orphan_release_metadata(struct inode
*inode
)
3972 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3973 u64 num_bytes
= calc_trans_metadata_size(root
, 4);
3974 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
, num_bytes
);
3977 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle
*trans
,
3978 struct btrfs_pending_snapshot
*pending
)
3980 struct btrfs_root
*root
= pending
->root
;
3981 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
3982 struct btrfs_block_rsv
*dst_rsv
= &pending
->block_rsv
;
3984 * two for root back/forward refs, two for directory entries
3985 * and one for root of the snapshot.
3987 u64 num_bytes
= calc_trans_metadata_size(root
, 5);
3988 dst_rsv
->space_info
= src_rsv
->space_info
;
3989 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
3992 static u64
calc_csum_metadata_size(struct inode
*inode
, u64 num_bytes
)
3994 return num_bytes
>>= 3;
3997 int btrfs_delalloc_reserve_metadata(struct inode
*inode
, u64 num_bytes
)
3999 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4000 struct btrfs_block_rsv
*block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
4003 int reserved_extents
;
4006 if (btrfs_transaction_in_commit(root
->fs_info
))
4007 schedule_timeout(1);
4009 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
4011 nr_extents
= atomic_read(&BTRFS_I(inode
)->outstanding_extents
) + 1;
4012 reserved_extents
= atomic_read(&BTRFS_I(inode
)->reserved_extents
);
4014 if (nr_extents
> reserved_extents
) {
4015 nr_extents
-= reserved_extents
;
4016 to_reserve
= calc_trans_metadata_size(root
, nr_extents
);
4022 to_reserve
+= calc_csum_metadata_size(inode
, num_bytes
);
4023 ret
= reserve_metadata_bytes(NULL
, root
, block_rsv
, to_reserve
, 1);
4027 atomic_add(nr_extents
, &BTRFS_I(inode
)->reserved_extents
);
4028 atomic_inc(&BTRFS_I(inode
)->outstanding_extents
);
4030 block_rsv_add_bytes(block_rsv
, to_reserve
, 1);
4032 if (block_rsv
->size
> 512 * 1024 * 1024)
4033 shrink_delalloc(NULL
, root
, to_reserve
, 0);
4038 void btrfs_delalloc_release_metadata(struct inode
*inode
, u64 num_bytes
)
4040 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4043 int reserved_extents
;
4045 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
4046 atomic_dec(&BTRFS_I(inode
)->outstanding_extents
);
4047 WARN_ON(atomic_read(&BTRFS_I(inode
)->outstanding_extents
) < 0);
4049 reserved_extents
= atomic_read(&BTRFS_I(inode
)->reserved_extents
);
4053 nr_extents
= atomic_read(&BTRFS_I(inode
)->outstanding_extents
);
4054 if (nr_extents
>= reserved_extents
) {
4058 old
= reserved_extents
;
4059 nr_extents
= reserved_extents
- nr_extents
;
4060 new = reserved_extents
- nr_extents
;
4061 old
= atomic_cmpxchg(&BTRFS_I(inode
)->reserved_extents
,
4062 reserved_extents
, new);
4063 if (likely(old
== reserved_extents
))
4065 reserved_extents
= old
;
4068 to_free
= calc_csum_metadata_size(inode
, num_bytes
);
4070 to_free
+= calc_trans_metadata_size(root
, nr_extents
);
4072 btrfs_block_rsv_release(root
, &root
->fs_info
->delalloc_block_rsv
,
4076 int btrfs_delalloc_reserve_space(struct inode
*inode
, u64 num_bytes
)
4080 ret
= btrfs_check_data_free_space(inode
, num_bytes
);
4084 ret
= btrfs_delalloc_reserve_metadata(inode
, num_bytes
);
4086 btrfs_free_reserved_data_space(inode
, num_bytes
);
4093 void btrfs_delalloc_release_space(struct inode
*inode
, u64 num_bytes
)
4095 btrfs_delalloc_release_metadata(inode
, num_bytes
);
4096 btrfs_free_reserved_data_space(inode
, num_bytes
);
4099 static int update_block_group(struct btrfs_trans_handle
*trans
,
4100 struct btrfs_root
*root
,
4101 u64 bytenr
, u64 num_bytes
, int alloc
)
4103 struct btrfs_block_group_cache
*cache
= NULL
;
4104 struct btrfs_fs_info
*info
= root
->fs_info
;
4105 u64 total
= num_bytes
;
4110 /* block accounting for super block */
4111 spin_lock(&info
->delalloc_lock
);
4112 old_val
= btrfs_super_bytes_used(&info
->super_copy
);
4114 old_val
+= num_bytes
;
4116 old_val
-= num_bytes
;
4117 btrfs_set_super_bytes_used(&info
->super_copy
, old_val
);
4118 spin_unlock(&info
->delalloc_lock
);
4121 cache
= btrfs_lookup_block_group(info
, bytenr
);
4124 if (cache
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
4125 BTRFS_BLOCK_GROUP_RAID1
|
4126 BTRFS_BLOCK_GROUP_RAID10
))
4131 * If this block group has free space cache written out, we
4132 * need to make sure to load it if we are removing space. This
4133 * is because we need the unpinning stage to actually add the
4134 * space back to the block group, otherwise we will leak space.
4136 if (!alloc
&& cache
->cached
== BTRFS_CACHE_NO
)
4137 cache_block_group(cache
, trans
, NULL
, 1);
4139 byte_in_group
= bytenr
- cache
->key
.objectid
;
4140 WARN_ON(byte_in_group
> cache
->key
.offset
);
4142 spin_lock(&cache
->space_info
->lock
);
4143 spin_lock(&cache
->lock
);
4145 if (btrfs_super_cache_generation(&info
->super_copy
) != 0 &&
4146 cache
->disk_cache_state
< BTRFS_DC_CLEAR
)
4147 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
4150 old_val
= btrfs_block_group_used(&cache
->item
);
4151 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
4153 old_val
+= num_bytes
;
4154 btrfs_set_block_group_used(&cache
->item
, old_val
);
4155 cache
->reserved
-= num_bytes
;
4156 cache
->space_info
->bytes_reserved
-= num_bytes
;
4157 cache
->space_info
->reservation_progress
++;
4158 cache
->space_info
->bytes_used
+= num_bytes
;
4159 cache
->space_info
->disk_used
+= num_bytes
* factor
;
4160 spin_unlock(&cache
->lock
);
4161 spin_unlock(&cache
->space_info
->lock
);
4163 old_val
-= num_bytes
;
4164 btrfs_set_block_group_used(&cache
->item
, old_val
);
4165 cache
->pinned
+= num_bytes
;
4166 cache
->space_info
->bytes_pinned
+= num_bytes
;
4167 cache
->space_info
->bytes_used
-= num_bytes
;
4168 cache
->space_info
->disk_used
-= num_bytes
* factor
;
4169 spin_unlock(&cache
->lock
);
4170 spin_unlock(&cache
->space_info
->lock
);
4172 set_extent_dirty(info
->pinned_extents
,
4173 bytenr
, bytenr
+ num_bytes
- 1,
4174 GFP_NOFS
| __GFP_NOFAIL
);
4176 btrfs_put_block_group(cache
);
4178 bytenr
+= num_bytes
;
4183 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
4185 struct btrfs_block_group_cache
*cache
;
4188 cache
= btrfs_lookup_first_block_group(root
->fs_info
, search_start
);
4192 bytenr
= cache
->key
.objectid
;
4193 btrfs_put_block_group(cache
);
4198 static int pin_down_extent(struct btrfs_root
*root
,
4199 struct btrfs_block_group_cache
*cache
,
4200 u64 bytenr
, u64 num_bytes
, int reserved
)
4202 spin_lock(&cache
->space_info
->lock
);
4203 spin_lock(&cache
->lock
);
4204 cache
->pinned
+= num_bytes
;
4205 cache
->space_info
->bytes_pinned
+= num_bytes
;
4207 cache
->reserved
-= num_bytes
;
4208 cache
->space_info
->bytes_reserved
-= num_bytes
;
4209 cache
->space_info
->reservation_progress
++;
4211 spin_unlock(&cache
->lock
);
4212 spin_unlock(&cache
->space_info
->lock
);
4214 set_extent_dirty(root
->fs_info
->pinned_extents
, bytenr
,
4215 bytenr
+ num_bytes
- 1, GFP_NOFS
| __GFP_NOFAIL
);
4220 * this function must be called within transaction
4222 int btrfs_pin_extent(struct btrfs_root
*root
,
4223 u64 bytenr
, u64 num_bytes
, int reserved
)
4225 struct btrfs_block_group_cache
*cache
;
4227 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
4230 pin_down_extent(root
, cache
, bytenr
, num_bytes
, reserved
);
4232 btrfs_put_block_group(cache
);
4237 * update size of reserved extents. this function may return -EAGAIN
4238 * if 'reserve' is true or 'sinfo' is false.
4240 int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
4241 u64 num_bytes
, int reserve
, int sinfo
)
4245 struct btrfs_space_info
*space_info
= cache
->space_info
;
4246 spin_lock(&space_info
->lock
);
4247 spin_lock(&cache
->lock
);
4252 cache
->reserved
+= num_bytes
;
4253 space_info
->bytes_reserved
+= num_bytes
;
4257 space_info
->bytes_readonly
+= num_bytes
;
4258 cache
->reserved
-= num_bytes
;
4259 space_info
->bytes_reserved
-= num_bytes
;
4260 space_info
->reservation_progress
++;
4262 spin_unlock(&cache
->lock
);
4263 spin_unlock(&space_info
->lock
);
4265 spin_lock(&cache
->lock
);
4270 cache
->reserved
+= num_bytes
;
4272 cache
->reserved
-= num_bytes
;
4274 spin_unlock(&cache
->lock
);
4279 int btrfs_prepare_extent_commit(struct btrfs_trans_handle
*trans
,
4280 struct btrfs_root
*root
)
4282 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4283 struct btrfs_caching_control
*next
;
4284 struct btrfs_caching_control
*caching_ctl
;
4285 struct btrfs_block_group_cache
*cache
;
4287 down_write(&fs_info
->extent_commit_sem
);
4289 list_for_each_entry_safe(caching_ctl
, next
,
4290 &fs_info
->caching_block_groups
, list
) {
4291 cache
= caching_ctl
->block_group
;
4292 if (block_group_cache_done(cache
)) {
4293 cache
->last_byte_to_unpin
= (u64
)-1;
4294 list_del_init(&caching_ctl
->list
);
4295 put_caching_control(caching_ctl
);
4297 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
4301 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
4302 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
4304 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
4306 up_write(&fs_info
->extent_commit_sem
);
4308 update_global_block_rsv(fs_info
);
4312 static int unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
4314 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4315 struct btrfs_block_group_cache
*cache
= NULL
;
4318 while (start
<= end
) {
4320 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
4322 btrfs_put_block_group(cache
);
4323 cache
= btrfs_lookup_block_group(fs_info
, start
);
4327 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
4328 len
= min(len
, end
+ 1 - start
);
4330 if (start
< cache
->last_byte_to_unpin
) {
4331 len
= min(len
, cache
->last_byte_to_unpin
- start
);
4332 btrfs_add_free_space(cache
, start
, len
);
4337 spin_lock(&cache
->space_info
->lock
);
4338 spin_lock(&cache
->lock
);
4339 cache
->pinned
-= len
;
4340 cache
->space_info
->bytes_pinned
-= len
;
4342 cache
->space_info
->bytes_readonly
+= len
;
4343 } else if (cache
->reserved_pinned
> 0) {
4344 len
= min(len
, cache
->reserved_pinned
);
4345 cache
->reserved_pinned
-= len
;
4346 cache
->space_info
->bytes_reserved
+= len
;
4348 spin_unlock(&cache
->lock
);
4349 spin_unlock(&cache
->space_info
->lock
);
4353 btrfs_put_block_group(cache
);
4357 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
4358 struct btrfs_root
*root
)
4360 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4361 struct extent_io_tree
*unpin
;
4362 struct btrfs_block_rsv
*block_rsv
;
4363 struct btrfs_block_rsv
*next_rsv
;
4369 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
4370 unpin
= &fs_info
->freed_extents
[1];
4372 unpin
= &fs_info
->freed_extents
[0];
4375 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
4380 if (btrfs_test_opt(root
, DISCARD
))
4381 ret
= btrfs_discard_extent(root
, start
,
4382 end
+ 1 - start
, NULL
);
4384 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
4385 unpin_extent_range(root
, start
, end
);
4389 mutex_lock(&fs_info
->durable_block_rsv_mutex
);
4390 list_for_each_entry_safe(block_rsv
, next_rsv
,
4391 &fs_info
->durable_block_rsv_list
, list
) {
4393 idx
= trans
->transid
& 0x1;
4394 if (block_rsv
->freed
[idx
] > 0) {
4395 block_rsv_add_bytes(block_rsv
,
4396 block_rsv
->freed
[idx
], 0);
4397 block_rsv
->freed
[idx
] = 0;
4399 if (atomic_read(&block_rsv
->usage
) == 0) {
4400 btrfs_block_rsv_release(root
, block_rsv
, (u64
)-1);
4402 if (block_rsv
->freed
[0] == 0 &&
4403 block_rsv
->freed
[1] == 0) {
4404 list_del_init(&block_rsv
->list
);
4408 btrfs_block_rsv_release(root
, block_rsv
, 0);
4411 mutex_unlock(&fs_info
->durable_block_rsv_mutex
);
4416 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
4417 struct btrfs_root
*root
,
4418 u64 bytenr
, u64 num_bytes
, u64 parent
,
4419 u64 root_objectid
, u64 owner_objectid
,
4420 u64 owner_offset
, int refs_to_drop
,
4421 struct btrfs_delayed_extent_op
*extent_op
)
4423 struct btrfs_key key
;
4424 struct btrfs_path
*path
;
4425 struct btrfs_fs_info
*info
= root
->fs_info
;
4426 struct btrfs_root
*extent_root
= info
->extent_root
;
4427 struct extent_buffer
*leaf
;
4428 struct btrfs_extent_item
*ei
;
4429 struct btrfs_extent_inline_ref
*iref
;
4432 int extent_slot
= 0;
4433 int found_extent
= 0;
4438 path
= btrfs_alloc_path();
4443 path
->leave_spinning
= 1;
4445 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
4446 BUG_ON(!is_data
&& refs_to_drop
!= 1);
4448 ret
= lookup_extent_backref(trans
, extent_root
, path
, &iref
,
4449 bytenr
, num_bytes
, parent
,
4450 root_objectid
, owner_objectid
,
4453 extent_slot
= path
->slots
[0];
4454 while (extent_slot
>= 0) {
4455 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
4457 if (key
.objectid
!= bytenr
)
4459 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
4460 key
.offset
== num_bytes
) {
4464 if (path
->slots
[0] - extent_slot
> 5)
4468 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4469 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
4470 if (found_extent
&& item_size
< sizeof(*ei
))
4473 if (!found_extent
) {
4475 ret
= remove_extent_backref(trans
, extent_root
, path
,
4479 btrfs_release_path(extent_root
, path
);
4480 path
->leave_spinning
= 1;
4482 key
.objectid
= bytenr
;
4483 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
4484 key
.offset
= num_bytes
;
4486 ret
= btrfs_search_slot(trans
, extent_root
,
4489 printk(KERN_ERR
"umm, got %d back from search"
4490 ", was looking for %llu\n", ret
,
4491 (unsigned long long)bytenr
);
4492 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
4495 extent_slot
= path
->slots
[0];
4498 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
4500 printk(KERN_ERR
"btrfs unable to find ref byte nr %llu "
4501 "parent %llu root %llu owner %llu offset %llu\n",
4502 (unsigned long long)bytenr
,
4503 (unsigned long long)parent
,
4504 (unsigned long long)root_objectid
,
4505 (unsigned long long)owner_objectid
,
4506 (unsigned long long)owner_offset
);
4509 leaf
= path
->nodes
[0];
4510 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
4511 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4512 if (item_size
< sizeof(*ei
)) {
4513 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
4514 ret
= convert_extent_item_v0(trans
, extent_root
, path
,
4518 btrfs_release_path(extent_root
, path
);
4519 path
->leave_spinning
= 1;
4521 key
.objectid
= bytenr
;
4522 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
4523 key
.offset
= num_bytes
;
4525 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
4528 printk(KERN_ERR
"umm, got %d back from search"
4529 ", was looking for %llu\n", ret
,
4530 (unsigned long long)bytenr
);
4531 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
4534 extent_slot
= path
->slots
[0];
4535 leaf
= path
->nodes
[0];
4536 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
4539 BUG_ON(item_size
< sizeof(*ei
));
4540 ei
= btrfs_item_ptr(leaf
, extent_slot
,
4541 struct btrfs_extent_item
);
4542 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
) {
4543 struct btrfs_tree_block_info
*bi
;
4544 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
4545 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
4546 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
4549 refs
= btrfs_extent_refs(leaf
, ei
);
4550 BUG_ON(refs
< refs_to_drop
);
4551 refs
-= refs_to_drop
;
4555 __run_delayed_extent_op(extent_op
, leaf
, ei
);
4557 * In the case of inline back ref, reference count will
4558 * be updated by remove_extent_backref
4561 BUG_ON(!found_extent
);
4563 btrfs_set_extent_refs(leaf
, ei
, refs
);
4564 btrfs_mark_buffer_dirty(leaf
);
4567 ret
= remove_extent_backref(trans
, extent_root
, path
,
4574 BUG_ON(is_data
&& refs_to_drop
!=
4575 extent_data_ref_count(root
, path
, iref
));
4577 BUG_ON(path
->slots
[0] != extent_slot
);
4579 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
4580 path
->slots
[0] = extent_slot
;
4585 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
4588 btrfs_release_path(extent_root
, path
);
4591 ret
= btrfs_del_csums(trans
, root
, bytenr
, num_bytes
);
4594 invalidate_mapping_pages(info
->btree_inode
->i_mapping
,
4595 bytenr
>> PAGE_CACHE_SHIFT
,
4596 (bytenr
+ num_bytes
- 1) >> PAGE_CACHE_SHIFT
);
4599 ret
= update_block_group(trans
, root
, bytenr
, num_bytes
, 0);
4602 btrfs_free_path(path
);
4607 * when we free an block, it is possible (and likely) that we free the last
4608 * delayed ref for that extent as well. This searches the delayed ref tree for
4609 * a given extent, and if there are no other delayed refs to be processed, it
4610 * removes it from the tree.
4612 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
4613 struct btrfs_root
*root
, u64 bytenr
)
4615 struct btrfs_delayed_ref_head
*head
;
4616 struct btrfs_delayed_ref_root
*delayed_refs
;
4617 struct btrfs_delayed_ref_node
*ref
;
4618 struct rb_node
*node
;
4621 delayed_refs
= &trans
->transaction
->delayed_refs
;
4622 spin_lock(&delayed_refs
->lock
);
4623 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
4627 node
= rb_prev(&head
->node
.rb_node
);
4631 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
4633 /* there are still entries for this ref, we can't drop it */
4634 if (ref
->bytenr
== bytenr
)
4637 if (head
->extent_op
) {
4638 if (!head
->must_insert_reserved
)
4640 kfree(head
->extent_op
);
4641 head
->extent_op
= NULL
;
4645 * waiting for the lock here would deadlock. If someone else has it
4646 * locked they are already in the process of dropping it anyway
4648 if (!mutex_trylock(&head
->mutex
))
4652 * at this point we have a head with no other entries. Go
4653 * ahead and process it.
4655 head
->node
.in_tree
= 0;
4656 rb_erase(&head
->node
.rb_node
, &delayed_refs
->root
);
4658 delayed_refs
->num_entries
--;
4661 * we don't take a ref on the node because we're removing it from the
4662 * tree, so we just steal the ref the tree was holding.
4664 delayed_refs
->num_heads
--;
4665 if (list_empty(&head
->cluster
))
4666 delayed_refs
->num_heads_ready
--;
4668 list_del_init(&head
->cluster
);
4669 spin_unlock(&delayed_refs
->lock
);
4671 BUG_ON(head
->extent_op
);
4672 if (head
->must_insert_reserved
)
4675 mutex_unlock(&head
->mutex
);
4676 btrfs_put_delayed_ref(&head
->node
);
4679 spin_unlock(&delayed_refs
->lock
);
4683 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
4684 struct btrfs_root
*root
,
4685 struct extent_buffer
*buf
,
4686 u64 parent
, int last_ref
)
4688 struct btrfs_block_rsv
*block_rsv
;
4689 struct btrfs_block_group_cache
*cache
= NULL
;
4692 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
4693 ret
= btrfs_add_delayed_tree_ref(trans
, buf
->start
, buf
->len
,
4694 parent
, root
->root_key
.objectid
,
4695 btrfs_header_level(buf
),
4696 BTRFS_DROP_DELAYED_REF
, NULL
);
4703 block_rsv
= get_block_rsv(trans
, root
);
4704 cache
= btrfs_lookup_block_group(root
->fs_info
, buf
->start
);
4705 if (block_rsv
->space_info
!= cache
->space_info
)
4708 if (btrfs_header_generation(buf
) == trans
->transid
) {
4709 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
4710 ret
= check_ref_cleanup(trans
, root
, buf
->start
);
4715 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
4716 pin_down_extent(root
, cache
, buf
->start
, buf
->len
, 1);
4720 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
4722 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
4723 ret
= btrfs_update_reserved_bytes(cache
, buf
->len
, 0, 0);
4724 if (ret
== -EAGAIN
) {
4725 /* block group became read-only */
4726 btrfs_update_reserved_bytes(cache
, buf
->len
, 0, 1);
4731 spin_lock(&block_rsv
->lock
);
4732 if (block_rsv
->reserved
< block_rsv
->size
) {
4733 block_rsv
->reserved
+= buf
->len
;
4736 spin_unlock(&block_rsv
->lock
);
4739 spin_lock(&cache
->space_info
->lock
);
4740 cache
->space_info
->bytes_reserved
-= buf
->len
;
4741 cache
->space_info
->reservation_progress
++;
4742 spin_unlock(&cache
->space_info
->lock
);
4747 if (block_rsv
->durable
&& !cache
->ro
) {
4749 spin_lock(&cache
->lock
);
4751 cache
->reserved_pinned
+= buf
->len
;
4754 spin_unlock(&cache
->lock
);
4757 spin_lock(&block_rsv
->lock
);
4758 block_rsv
->freed
[trans
->transid
& 0x1] += buf
->len
;
4759 spin_unlock(&block_rsv
->lock
);
4764 * Deleting the buffer, clear the corrupt flag since it doesn't matter
4767 clear_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
);
4768 btrfs_put_block_group(cache
);
4771 int btrfs_free_extent(struct btrfs_trans_handle
*trans
,
4772 struct btrfs_root
*root
,
4773 u64 bytenr
, u64 num_bytes
, u64 parent
,
4774 u64 root_objectid
, u64 owner
, u64 offset
)
4779 * tree log blocks never actually go into the extent allocation
4780 * tree, just update pinning info and exit early.
4782 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
4783 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
4784 /* unlocks the pinned mutex */
4785 btrfs_pin_extent(root
, bytenr
, num_bytes
, 1);
4787 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
4788 ret
= btrfs_add_delayed_tree_ref(trans
, bytenr
, num_bytes
,
4789 parent
, root_objectid
, (int)owner
,
4790 BTRFS_DROP_DELAYED_REF
, NULL
);
4793 ret
= btrfs_add_delayed_data_ref(trans
, bytenr
, num_bytes
,
4794 parent
, root_objectid
, owner
,
4795 offset
, BTRFS_DROP_DELAYED_REF
, NULL
);
4801 static u64
stripe_align(struct btrfs_root
*root
, u64 val
)
4803 u64 mask
= ((u64
)root
->stripesize
- 1);
4804 u64 ret
= (val
+ mask
) & ~mask
;
4809 * when we wait for progress in the block group caching, its because
4810 * our allocation attempt failed at least once. So, we must sleep
4811 * and let some progress happen before we try again.
4813 * This function will sleep at least once waiting for new free space to
4814 * show up, and then it will check the block group free space numbers
4815 * for our min num_bytes. Another option is to have it go ahead
4816 * and look in the rbtree for a free extent of a given size, but this
4820 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
4823 struct btrfs_caching_control
*caching_ctl
;
4826 caching_ctl
= get_caching_control(cache
);
4830 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
4831 (cache
->free_space
>= num_bytes
));
4833 put_caching_control(caching_ctl
);
4838 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
4840 struct btrfs_caching_control
*caching_ctl
;
4843 caching_ctl
= get_caching_control(cache
);
4847 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
4849 put_caching_control(caching_ctl
);
4853 static int get_block_group_index(struct btrfs_block_group_cache
*cache
)
4856 if (cache
->flags
& BTRFS_BLOCK_GROUP_RAID10
)
4858 else if (cache
->flags
& BTRFS_BLOCK_GROUP_RAID1
)
4860 else if (cache
->flags
& BTRFS_BLOCK_GROUP_DUP
)
4862 else if (cache
->flags
& BTRFS_BLOCK_GROUP_RAID0
)
4869 enum btrfs_loop_type
{
4870 LOOP_FIND_IDEAL
= 0,
4871 LOOP_CACHING_NOWAIT
= 1,
4872 LOOP_CACHING_WAIT
= 2,
4873 LOOP_ALLOC_CHUNK
= 3,
4874 LOOP_NO_EMPTY_SIZE
= 4,
4878 * walks the btree of allocated extents and find a hole of a given size.
4879 * The key ins is changed to record the hole:
4880 * ins->objectid == block start
4881 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4882 * ins->offset == number of blocks
4883 * Any available blocks before search_start are skipped.
4885 static noinline
int find_free_extent(struct btrfs_trans_handle
*trans
,
4886 struct btrfs_root
*orig_root
,
4887 u64 num_bytes
, u64 empty_size
,
4888 u64 search_start
, u64 search_end
,
4889 u64 hint_byte
, struct btrfs_key
*ins
,
4893 struct btrfs_root
*root
= orig_root
->fs_info
->extent_root
;
4894 struct btrfs_free_cluster
*last_ptr
= NULL
;
4895 struct btrfs_block_group_cache
*block_group
= NULL
;
4896 int empty_cluster
= 2 * 1024 * 1024;
4897 int allowed_chunk_alloc
= 0;
4898 int done_chunk_alloc
= 0;
4899 struct btrfs_space_info
*space_info
;
4900 int last_ptr_loop
= 0;
4903 bool found_uncached_bg
= false;
4904 bool failed_cluster_refill
= false;
4905 bool failed_alloc
= false;
4906 bool use_cluster
= true;
4907 u64 ideal_cache_percent
= 0;
4908 u64 ideal_cache_offset
= 0;
4910 WARN_ON(num_bytes
< root
->sectorsize
);
4911 btrfs_set_key_type(ins
, BTRFS_EXTENT_ITEM_KEY
);
4915 space_info
= __find_space_info(root
->fs_info
, data
);
4917 printk(KERN_ERR
"No space info for %d\n", data
);
4922 * If the space info is for both data and metadata it means we have a
4923 * small filesystem and we can't use the clustering stuff.
4925 if (btrfs_mixed_space_info(space_info
))
4926 use_cluster
= false;
4928 if (orig_root
->ref_cows
|| empty_size
)
4929 allowed_chunk_alloc
= 1;
4931 if (data
& BTRFS_BLOCK_GROUP_METADATA
&& use_cluster
) {
4932 last_ptr
= &root
->fs_info
->meta_alloc_cluster
;
4933 if (!btrfs_test_opt(root
, SSD
))
4934 empty_cluster
= 64 * 1024;
4937 if ((data
& BTRFS_BLOCK_GROUP_DATA
) && use_cluster
&&
4938 btrfs_test_opt(root
, SSD
)) {
4939 last_ptr
= &root
->fs_info
->data_alloc_cluster
;
4943 spin_lock(&last_ptr
->lock
);
4944 if (last_ptr
->block_group
)
4945 hint_byte
= last_ptr
->window_start
;
4946 spin_unlock(&last_ptr
->lock
);
4949 search_start
= max(search_start
, first_logical_byte(root
, 0));
4950 search_start
= max(search_start
, hint_byte
);
4955 if (search_start
== hint_byte
) {
4957 block_group
= btrfs_lookup_block_group(root
->fs_info
,
4960 * we don't want to use the block group if it doesn't match our
4961 * allocation bits, or if its not cached.
4963 * However if we are re-searching with an ideal block group
4964 * picked out then we don't care that the block group is cached.
4966 if (block_group
&& block_group_bits(block_group
, data
) &&
4967 (block_group
->cached
!= BTRFS_CACHE_NO
||
4968 search_start
== ideal_cache_offset
)) {
4969 down_read(&space_info
->groups_sem
);
4970 if (list_empty(&block_group
->list
) ||
4973 * someone is removing this block group,
4974 * we can't jump into the have_block_group
4975 * target because our list pointers are not
4978 btrfs_put_block_group(block_group
);
4979 up_read(&space_info
->groups_sem
);
4981 index
= get_block_group_index(block_group
);
4982 goto have_block_group
;
4984 } else if (block_group
) {
4985 btrfs_put_block_group(block_group
);
4989 down_read(&space_info
->groups_sem
);
4990 list_for_each_entry(block_group
, &space_info
->block_groups
[index
],
4995 btrfs_get_block_group(block_group
);
4996 search_start
= block_group
->key
.objectid
;
4999 * this can happen if we end up cycling through all the
5000 * raid types, but we want to make sure we only allocate
5001 * for the proper type.
5003 if (!block_group_bits(block_group
, data
)) {
5004 u64 extra
= BTRFS_BLOCK_GROUP_DUP
|
5005 BTRFS_BLOCK_GROUP_RAID1
|
5006 BTRFS_BLOCK_GROUP_RAID10
;
5009 * if they asked for extra copies and this block group
5010 * doesn't provide them, bail. This does allow us to
5011 * fill raid0 from raid1.
5013 if ((data
& extra
) && !(block_group
->flags
& extra
))
5018 if (unlikely(block_group
->cached
== BTRFS_CACHE_NO
)) {
5021 ret
= cache_block_group(block_group
, trans
,
5023 if (block_group
->cached
== BTRFS_CACHE_FINISHED
)
5024 goto have_block_group
;
5026 free_percent
= btrfs_block_group_used(&block_group
->item
);
5027 free_percent
*= 100;
5028 free_percent
= div64_u64(free_percent
,
5029 block_group
->key
.offset
);
5030 free_percent
= 100 - free_percent
;
5031 if (free_percent
> ideal_cache_percent
&&
5032 likely(!block_group
->ro
)) {
5033 ideal_cache_offset
= block_group
->key
.objectid
;
5034 ideal_cache_percent
= free_percent
;
5038 * We only want to start kthread caching if we are at
5039 * the point where we will wait for caching to make
5040 * progress, or if our ideal search is over and we've
5041 * found somebody to start caching.
5043 if (loop
> LOOP_CACHING_NOWAIT
||
5044 (loop
> LOOP_FIND_IDEAL
&&
5045 atomic_read(&space_info
->caching_threads
) < 2)) {
5046 ret
= cache_block_group(block_group
, trans
,
5050 found_uncached_bg
= true;
5053 * If loop is set for cached only, try the next block
5056 if (loop
== LOOP_FIND_IDEAL
)
5060 cached
= block_group_cache_done(block_group
);
5061 if (unlikely(!cached
))
5062 found_uncached_bg
= true;
5064 if (unlikely(block_group
->ro
))
5068 * Ok we want to try and use the cluster allocator, so lets look
5069 * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
5070 * have tried the cluster allocator plenty of times at this
5071 * point and not have found anything, so we are likely way too
5072 * fragmented for the clustering stuff to find anything, so lets
5073 * just skip it and let the allocator find whatever block it can
5076 if (last_ptr
&& loop
< LOOP_NO_EMPTY_SIZE
) {
5078 * the refill lock keeps out other
5079 * people trying to start a new cluster
5081 spin_lock(&last_ptr
->refill_lock
);
5082 if (last_ptr
->block_group
&&
5083 (last_ptr
->block_group
->ro
||
5084 !block_group_bits(last_ptr
->block_group
, data
))) {
5086 goto refill_cluster
;
5089 offset
= btrfs_alloc_from_cluster(block_group
, last_ptr
,
5090 num_bytes
, search_start
);
5092 /* we have a block, we're done */
5093 spin_unlock(&last_ptr
->refill_lock
);
5097 spin_lock(&last_ptr
->lock
);
5099 * whoops, this cluster doesn't actually point to
5100 * this block group. Get a ref on the block
5101 * group is does point to and try again
5103 if (!last_ptr_loop
&& last_ptr
->block_group
&&
5104 last_ptr
->block_group
!= block_group
) {
5106 btrfs_put_block_group(block_group
);
5107 block_group
= last_ptr
->block_group
;
5108 btrfs_get_block_group(block_group
);
5109 spin_unlock(&last_ptr
->lock
);
5110 spin_unlock(&last_ptr
->refill_lock
);
5113 search_start
= block_group
->key
.objectid
;
5115 * we know this block group is properly
5116 * in the list because
5117 * btrfs_remove_block_group, drops the
5118 * cluster before it removes the block
5119 * group from the list
5121 goto have_block_group
;
5123 spin_unlock(&last_ptr
->lock
);
5126 * this cluster didn't work out, free it and
5129 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
5133 /* allocate a cluster in this block group */
5134 ret
= btrfs_find_space_cluster(trans
, root
,
5135 block_group
, last_ptr
,
5137 empty_cluster
+ empty_size
);
5140 * now pull our allocation out of this
5143 offset
= btrfs_alloc_from_cluster(block_group
,
5144 last_ptr
, num_bytes
,
5147 /* we found one, proceed */
5148 spin_unlock(&last_ptr
->refill_lock
);
5151 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
5152 && !failed_cluster_refill
) {
5153 spin_unlock(&last_ptr
->refill_lock
);
5155 failed_cluster_refill
= true;
5156 wait_block_group_cache_progress(block_group
,
5157 num_bytes
+ empty_cluster
+ empty_size
);
5158 goto have_block_group
;
5162 * at this point we either didn't find a cluster
5163 * or we weren't able to allocate a block from our
5164 * cluster. Free the cluster we've been trying
5165 * to use, and go to the next block group
5167 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
5168 spin_unlock(&last_ptr
->refill_lock
);
5172 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
5173 num_bytes
, empty_size
);
5175 * If we didn't find a chunk, and we haven't failed on this
5176 * block group before, and this block group is in the middle of
5177 * caching and we are ok with waiting, then go ahead and wait
5178 * for progress to be made, and set failed_alloc to true.
5180 * If failed_alloc is true then we've already waited on this
5181 * block group once and should move on to the next block group.
5183 if (!offset
&& !failed_alloc
&& !cached
&&
5184 loop
> LOOP_CACHING_NOWAIT
) {
5185 wait_block_group_cache_progress(block_group
,
5186 num_bytes
+ empty_size
);
5187 failed_alloc
= true;
5188 goto have_block_group
;
5189 } else if (!offset
) {
5193 search_start
= stripe_align(root
, offset
);
5194 /* move on to the next group */
5195 if (search_start
+ num_bytes
>= search_end
) {
5196 btrfs_add_free_space(block_group
, offset
, num_bytes
);
5200 /* move on to the next group */
5201 if (search_start
+ num_bytes
>
5202 block_group
->key
.objectid
+ block_group
->key
.offset
) {
5203 btrfs_add_free_space(block_group
, offset
, num_bytes
);
5207 ins
->objectid
= search_start
;
5208 ins
->offset
= num_bytes
;
5210 if (offset
< search_start
)
5211 btrfs_add_free_space(block_group
, offset
,
5212 search_start
- offset
);
5213 BUG_ON(offset
> search_start
);
5215 ret
= btrfs_update_reserved_bytes(block_group
, num_bytes
, 1,
5216 (data
& BTRFS_BLOCK_GROUP_DATA
));
5217 if (ret
== -EAGAIN
) {
5218 btrfs_add_free_space(block_group
, offset
, num_bytes
);
5222 /* we are all good, lets return */
5223 ins
->objectid
= search_start
;
5224 ins
->offset
= num_bytes
;
5226 if (offset
< search_start
)
5227 btrfs_add_free_space(block_group
, offset
,
5228 search_start
- offset
);
5229 BUG_ON(offset
> search_start
);
5232 failed_cluster_refill
= false;
5233 failed_alloc
= false;
5234 BUG_ON(index
!= get_block_group_index(block_group
));
5235 btrfs_put_block_group(block_group
);
5237 up_read(&space_info
->groups_sem
);
5239 if (!ins
->objectid
&& ++index
< BTRFS_NR_RAID_TYPES
)
5242 /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5243 * for them to make caching progress. Also
5244 * determine the best possible bg to cache
5245 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5246 * caching kthreads as we move along
5247 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5248 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5249 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5252 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
&&
5253 (found_uncached_bg
|| empty_size
|| empty_cluster
||
5254 allowed_chunk_alloc
)) {
5256 if (loop
== LOOP_FIND_IDEAL
&& found_uncached_bg
) {
5257 found_uncached_bg
= false;
5259 if (!ideal_cache_percent
&&
5260 atomic_read(&space_info
->caching_threads
))
5264 * 1 of the following 2 things have happened so far
5266 * 1) We found an ideal block group for caching that
5267 * is mostly full and will cache quickly, so we might
5268 * as well wait for it.
5270 * 2) We searched for cached only and we didn't find
5271 * anything, and we didn't start any caching kthreads
5272 * either, so chances are we will loop through and
5273 * start a couple caching kthreads, and then come back
5274 * around and just wait for them. This will be slower
5275 * because we will have 2 caching kthreads reading at
5276 * the same time when we could have just started one
5277 * and waited for it to get far enough to give us an
5278 * allocation, so go ahead and go to the wait caching
5281 loop
= LOOP_CACHING_WAIT
;
5282 search_start
= ideal_cache_offset
;
5283 ideal_cache_percent
= 0;
5285 } else if (loop
== LOOP_FIND_IDEAL
) {
5287 * Didn't find a uncached bg, wait on anything we find
5290 loop
= LOOP_CACHING_WAIT
;
5294 if (loop
< LOOP_CACHING_WAIT
) {
5299 if (loop
== LOOP_ALLOC_CHUNK
) {
5304 if (allowed_chunk_alloc
) {
5305 ret
= do_chunk_alloc(trans
, root
, num_bytes
+
5306 2 * 1024 * 1024, data
, 1);
5307 allowed_chunk_alloc
= 0;
5308 done_chunk_alloc
= 1;
5309 } else if (!done_chunk_alloc
) {
5310 space_info
->force_alloc
= 1;
5313 if (loop
< LOOP_NO_EMPTY_SIZE
) {
5318 } else if (!ins
->objectid
) {
5322 /* we found what we needed */
5323 if (ins
->objectid
) {
5324 if (!(data
& BTRFS_BLOCK_GROUP_DATA
))
5325 trans
->block_group
= block_group
->key
.objectid
;
5327 btrfs_put_block_group(block_group
);
5334 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
5335 int dump_block_groups
)
5337 struct btrfs_block_group_cache
*cache
;
5340 spin_lock(&info
->lock
);
5341 printk(KERN_INFO
"space_info has %llu free, is %sfull\n",
5342 (unsigned long long)(info
->total_bytes
- info
->bytes_used
-
5343 info
->bytes_pinned
- info
->bytes_reserved
-
5344 info
->bytes_readonly
),
5345 (info
->full
) ? "" : "not ");
5346 printk(KERN_INFO
"space_info total=%llu, used=%llu, pinned=%llu, "
5347 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5348 (unsigned long long)info
->total_bytes
,
5349 (unsigned long long)info
->bytes_used
,
5350 (unsigned long long)info
->bytes_pinned
,
5351 (unsigned long long)info
->bytes_reserved
,
5352 (unsigned long long)info
->bytes_may_use
,
5353 (unsigned long long)info
->bytes_readonly
);
5354 spin_unlock(&info
->lock
);
5356 if (!dump_block_groups
)
5359 down_read(&info
->groups_sem
);
5361 list_for_each_entry(cache
, &info
->block_groups
[index
], list
) {
5362 spin_lock(&cache
->lock
);
5363 printk(KERN_INFO
"block group %llu has %llu bytes, %llu used "
5364 "%llu pinned %llu reserved\n",
5365 (unsigned long long)cache
->key
.objectid
,
5366 (unsigned long long)cache
->key
.offset
,
5367 (unsigned long long)btrfs_block_group_used(&cache
->item
),
5368 (unsigned long long)cache
->pinned
,
5369 (unsigned long long)cache
->reserved
);
5370 btrfs_dump_free_space(cache
, bytes
);
5371 spin_unlock(&cache
->lock
);
5373 if (++index
< BTRFS_NR_RAID_TYPES
)
5375 up_read(&info
->groups_sem
);
5378 int btrfs_reserve_extent(struct btrfs_trans_handle
*trans
,
5379 struct btrfs_root
*root
,
5380 u64 num_bytes
, u64 min_alloc_size
,
5381 u64 empty_size
, u64 hint_byte
,
5382 u64 search_end
, struct btrfs_key
*ins
,
5386 u64 search_start
= 0;
5388 data
= btrfs_get_alloc_profile(root
, data
);
5391 * the only place that sets empty_size is btrfs_realloc_node, which
5392 * is not called recursively on allocations
5394 if (empty_size
|| root
->ref_cows
)
5395 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
5396 num_bytes
+ 2 * 1024 * 1024, data
, 0);
5398 WARN_ON(num_bytes
< root
->sectorsize
);
5399 ret
= find_free_extent(trans
, root
, num_bytes
, empty_size
,
5400 search_start
, search_end
, hint_byte
,
5403 if (ret
== -ENOSPC
&& num_bytes
> min_alloc_size
) {
5404 num_bytes
= num_bytes
>> 1;
5405 num_bytes
= num_bytes
& ~(root
->sectorsize
- 1);
5406 num_bytes
= max(num_bytes
, min_alloc_size
);
5407 do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
5408 num_bytes
, data
, 1);
5411 if (ret
== -ENOSPC
&& btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
5412 struct btrfs_space_info
*sinfo
;
5414 sinfo
= __find_space_info(root
->fs_info
, data
);
5415 printk(KERN_ERR
"btrfs allocation failed flags %llu, "
5416 "wanted %llu\n", (unsigned long long)data
,
5417 (unsigned long long)num_bytes
);
5418 dump_space_info(sinfo
, num_bytes
, 1);
5421 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, ins
->offset
);
5426 int btrfs_free_reserved_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
5428 struct btrfs_block_group_cache
*cache
;
5431 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
5433 printk(KERN_ERR
"Unable to find block group for %llu\n",
5434 (unsigned long long)start
);
5438 if (btrfs_test_opt(root
, DISCARD
))
5439 ret
= btrfs_discard_extent(root
, start
, len
, NULL
);
5441 btrfs_add_free_space(cache
, start
, len
);
5442 btrfs_update_reserved_bytes(cache
, len
, 0, 1);
5443 btrfs_put_block_group(cache
);
5445 trace_btrfs_reserved_extent_free(root
, start
, len
);
5450 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
5451 struct btrfs_root
*root
,
5452 u64 parent
, u64 root_objectid
,
5453 u64 flags
, u64 owner
, u64 offset
,
5454 struct btrfs_key
*ins
, int ref_mod
)
5457 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5458 struct btrfs_extent_item
*extent_item
;
5459 struct btrfs_extent_inline_ref
*iref
;
5460 struct btrfs_path
*path
;
5461 struct extent_buffer
*leaf
;
5466 type
= BTRFS_SHARED_DATA_REF_KEY
;
5468 type
= BTRFS_EXTENT_DATA_REF_KEY
;
5470 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
5472 path
= btrfs_alloc_path();
5476 path
->leave_spinning
= 1;
5477 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
5481 leaf
= path
->nodes
[0];
5482 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
5483 struct btrfs_extent_item
);
5484 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
5485 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
5486 btrfs_set_extent_flags(leaf
, extent_item
,
5487 flags
| BTRFS_EXTENT_FLAG_DATA
);
5489 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
5490 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
5492 struct btrfs_shared_data_ref
*ref
;
5493 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
5494 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
5495 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
5497 struct btrfs_extent_data_ref
*ref
;
5498 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
5499 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
5500 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
5501 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
5502 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
5505 btrfs_mark_buffer_dirty(path
->nodes
[0]);
5506 btrfs_free_path(path
);
5508 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1);
5510 printk(KERN_ERR
"btrfs update block group failed for %llu "
5511 "%llu\n", (unsigned long long)ins
->objectid
,
5512 (unsigned long long)ins
->offset
);
5518 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
5519 struct btrfs_root
*root
,
5520 u64 parent
, u64 root_objectid
,
5521 u64 flags
, struct btrfs_disk_key
*key
,
5522 int level
, struct btrfs_key
*ins
)
5525 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5526 struct btrfs_extent_item
*extent_item
;
5527 struct btrfs_tree_block_info
*block_info
;
5528 struct btrfs_extent_inline_ref
*iref
;
5529 struct btrfs_path
*path
;
5530 struct extent_buffer
*leaf
;
5531 u32 size
= sizeof(*extent_item
) + sizeof(*block_info
) + sizeof(*iref
);
5533 path
= btrfs_alloc_path();
5536 path
->leave_spinning
= 1;
5537 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
5541 leaf
= path
->nodes
[0];
5542 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
5543 struct btrfs_extent_item
);
5544 btrfs_set_extent_refs(leaf
, extent_item
, 1);
5545 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
5546 btrfs_set_extent_flags(leaf
, extent_item
,
5547 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
5548 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
5550 btrfs_set_tree_block_key(leaf
, block_info
, key
);
5551 btrfs_set_tree_block_level(leaf
, block_info
, level
);
5553 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
5555 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
5556 btrfs_set_extent_inline_ref_type(leaf
, iref
,
5557 BTRFS_SHARED_BLOCK_REF_KEY
);
5558 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
5560 btrfs_set_extent_inline_ref_type(leaf
, iref
,
5561 BTRFS_TREE_BLOCK_REF_KEY
);
5562 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
5565 btrfs_mark_buffer_dirty(leaf
);
5566 btrfs_free_path(path
);
5568 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1);
5570 printk(KERN_ERR
"btrfs update block group failed for %llu "
5571 "%llu\n", (unsigned long long)ins
->objectid
,
5572 (unsigned long long)ins
->offset
);
5578 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
5579 struct btrfs_root
*root
,
5580 u64 root_objectid
, u64 owner
,
5581 u64 offset
, struct btrfs_key
*ins
)
5585 BUG_ON(root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
5587 ret
= btrfs_add_delayed_data_ref(trans
, ins
->objectid
, ins
->offset
,
5588 0, root_objectid
, owner
, offset
,
5589 BTRFS_ADD_DELAYED_EXTENT
, NULL
);
5594 * this is used by the tree logging recovery code. It records that
5595 * an extent has been allocated and makes sure to clear the free
5596 * space cache bits as well
5598 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
5599 struct btrfs_root
*root
,
5600 u64 root_objectid
, u64 owner
, u64 offset
,
5601 struct btrfs_key
*ins
)
5604 struct btrfs_block_group_cache
*block_group
;
5605 struct btrfs_caching_control
*caching_ctl
;
5606 u64 start
= ins
->objectid
;
5607 u64 num_bytes
= ins
->offset
;
5609 block_group
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
5610 cache_block_group(block_group
, trans
, NULL
, 0);
5611 caching_ctl
= get_caching_control(block_group
);
5614 BUG_ON(!block_group_cache_done(block_group
));
5615 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
5618 mutex_lock(&caching_ctl
->mutex
);
5620 if (start
>= caching_ctl
->progress
) {
5621 ret
= add_excluded_extent(root
, start
, num_bytes
);
5623 } else if (start
+ num_bytes
<= caching_ctl
->progress
) {
5624 ret
= btrfs_remove_free_space(block_group
,
5628 num_bytes
= caching_ctl
->progress
- start
;
5629 ret
= btrfs_remove_free_space(block_group
,
5633 start
= caching_ctl
->progress
;
5634 num_bytes
= ins
->objectid
+ ins
->offset
-
5635 caching_ctl
->progress
;
5636 ret
= add_excluded_extent(root
, start
, num_bytes
);
5640 mutex_unlock(&caching_ctl
->mutex
);
5641 put_caching_control(caching_ctl
);
5644 ret
= btrfs_update_reserved_bytes(block_group
, ins
->offset
, 1, 1);
5646 btrfs_put_block_group(block_group
);
5647 ret
= alloc_reserved_file_extent(trans
, root
, 0, root_objectid
,
5648 0, owner
, offset
, ins
, 1);
5652 struct extent_buffer
*btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
,
5653 struct btrfs_root
*root
,
5654 u64 bytenr
, u32 blocksize
,
5657 struct extent_buffer
*buf
;
5659 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
5661 return ERR_PTR(-ENOMEM
);
5662 btrfs_set_header_generation(buf
, trans
->transid
);
5663 btrfs_set_buffer_lockdep_class(buf
, level
);
5664 btrfs_tree_lock(buf
);
5665 clean_tree_block(trans
, root
, buf
);
5667 btrfs_set_lock_blocking(buf
);
5668 btrfs_set_buffer_uptodate(buf
);
5670 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
5672 * we allow two log transactions at a time, use different
5673 * EXENT bit to differentiate dirty pages.
5675 if (root
->log_transid
% 2 == 0)
5676 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
5677 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
5679 set_extent_new(&root
->dirty_log_pages
, buf
->start
,
5680 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
5682 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
5683 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
5685 trans
->blocks_used
++;
5686 /* this returns a buffer locked for blocking */
5690 static struct btrfs_block_rsv
*
5691 use_block_rsv(struct btrfs_trans_handle
*trans
,
5692 struct btrfs_root
*root
, u32 blocksize
)
5694 struct btrfs_block_rsv
*block_rsv
;
5695 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
5698 block_rsv
= get_block_rsv(trans
, root
);
5700 if (block_rsv
->size
== 0) {
5701 ret
= reserve_metadata_bytes(trans
, root
, block_rsv
,
5704 * If we couldn't reserve metadata bytes try and use some from
5705 * the global reserve.
5707 if (ret
&& block_rsv
!= global_rsv
) {
5708 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
5711 return ERR_PTR(ret
);
5713 return ERR_PTR(ret
);
5718 ret
= block_rsv_use_bytes(block_rsv
, blocksize
);
5723 ret
= reserve_metadata_bytes(trans
, root
, block_rsv
, blocksize
,
5726 spin_lock(&block_rsv
->lock
);
5727 block_rsv
->size
+= blocksize
;
5728 spin_unlock(&block_rsv
->lock
);
5730 } else if (ret
&& block_rsv
!= global_rsv
) {
5731 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
5737 return ERR_PTR(-ENOSPC
);
5740 static void unuse_block_rsv(struct btrfs_block_rsv
*block_rsv
, u32 blocksize
)
5742 block_rsv_add_bytes(block_rsv
, blocksize
, 0);
5743 block_rsv_release_bytes(block_rsv
, NULL
, 0);
5747 * finds a free extent and does all the dirty work required for allocation
5748 * returns the key for the extent through ins, and a tree buffer for
5749 * the first block of the extent through buf.
5751 * returns the tree buffer or NULL.
5753 struct extent_buffer
*btrfs_alloc_free_block(struct btrfs_trans_handle
*trans
,
5754 struct btrfs_root
*root
, u32 blocksize
,
5755 u64 parent
, u64 root_objectid
,
5756 struct btrfs_disk_key
*key
, int level
,
5757 u64 hint
, u64 empty_size
)
5759 struct btrfs_key ins
;
5760 struct btrfs_block_rsv
*block_rsv
;
5761 struct extent_buffer
*buf
;
5766 block_rsv
= use_block_rsv(trans
, root
, blocksize
);
5767 if (IS_ERR(block_rsv
))
5768 return ERR_CAST(block_rsv
);
5770 ret
= btrfs_reserve_extent(trans
, root
, blocksize
, blocksize
,
5771 empty_size
, hint
, (u64
)-1, &ins
, 0);
5773 unuse_block_rsv(block_rsv
, blocksize
);
5774 return ERR_PTR(ret
);
5777 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
,
5779 BUG_ON(IS_ERR(buf
));
5781 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
5783 parent
= ins
.objectid
;
5784 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
5788 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
5789 struct btrfs_delayed_extent_op
*extent_op
;
5790 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
5793 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
5795 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
5796 extent_op
->flags_to_set
= flags
;
5797 extent_op
->update_key
= 1;
5798 extent_op
->update_flags
= 1;
5799 extent_op
->is_data
= 0;
5801 ret
= btrfs_add_delayed_tree_ref(trans
, ins
.objectid
,
5802 ins
.offset
, parent
, root_objectid
,
5803 level
, BTRFS_ADD_DELAYED_EXTENT
,
5810 struct walk_control
{
5811 u64 refs
[BTRFS_MAX_LEVEL
];
5812 u64 flags
[BTRFS_MAX_LEVEL
];
5813 struct btrfs_key update_progress
;
5823 #define DROP_REFERENCE 1
5824 #define UPDATE_BACKREF 2
5826 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
5827 struct btrfs_root
*root
,
5828 struct walk_control
*wc
,
5829 struct btrfs_path
*path
)
5837 struct btrfs_key key
;
5838 struct extent_buffer
*eb
;
5843 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
5844 wc
->reada_count
= wc
->reada_count
* 2 / 3;
5845 wc
->reada_count
= max(wc
->reada_count
, 2);
5847 wc
->reada_count
= wc
->reada_count
* 3 / 2;
5848 wc
->reada_count
= min_t(int, wc
->reada_count
,
5849 BTRFS_NODEPTRS_PER_BLOCK(root
));
5852 eb
= path
->nodes
[wc
->level
];
5853 nritems
= btrfs_header_nritems(eb
);
5854 blocksize
= btrfs_level_size(root
, wc
->level
- 1);
5856 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
5857 if (nread
>= wc
->reada_count
)
5861 bytenr
= btrfs_node_blockptr(eb
, slot
);
5862 generation
= btrfs_node_ptr_generation(eb
, slot
);
5864 if (slot
== path
->slots
[wc
->level
])
5867 if (wc
->stage
== UPDATE_BACKREF
&&
5868 generation
<= root
->root_key
.offset
)
5871 /* We don't lock the tree block, it's OK to be racy here */
5872 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, blocksize
,
5877 if (wc
->stage
== DROP_REFERENCE
) {
5881 if (wc
->level
== 1 &&
5882 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
5884 if (!wc
->update_ref
||
5885 generation
<= root
->root_key
.offset
)
5887 btrfs_node_key_to_cpu(eb
, &key
, slot
);
5888 ret
= btrfs_comp_cpu_keys(&key
,
5889 &wc
->update_progress
);
5893 if (wc
->level
== 1 &&
5894 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
5898 ret
= readahead_tree_block(root
, bytenr
, blocksize
,
5904 wc
->reada_slot
= slot
;
5908 * hepler to process tree block while walking down the tree.
5910 * when wc->stage == UPDATE_BACKREF, this function updates
5911 * back refs for pointers in the block.
5913 * NOTE: return value 1 means we should stop walking down.
5915 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
5916 struct btrfs_root
*root
,
5917 struct btrfs_path
*path
,
5918 struct walk_control
*wc
, int lookup_info
)
5920 int level
= wc
->level
;
5921 struct extent_buffer
*eb
= path
->nodes
[level
];
5922 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
5925 if (wc
->stage
== UPDATE_BACKREF
&&
5926 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
5930 * when reference count of tree block is 1, it won't increase
5931 * again. once full backref flag is set, we never clear it.
5934 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
5935 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
5936 BUG_ON(!path
->locks
[level
]);
5937 ret
= btrfs_lookup_extent_info(trans
, root
,
5942 BUG_ON(wc
->refs
[level
] == 0);
5945 if (wc
->stage
== DROP_REFERENCE
) {
5946 if (wc
->refs
[level
] > 1)
5949 if (path
->locks
[level
] && !wc
->keep_locks
) {
5950 btrfs_tree_unlock(eb
);
5951 path
->locks
[level
] = 0;
5956 /* wc->stage == UPDATE_BACKREF */
5957 if (!(wc
->flags
[level
] & flag
)) {
5958 BUG_ON(!path
->locks
[level
]);
5959 ret
= btrfs_inc_ref(trans
, root
, eb
, 1);
5961 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
5963 ret
= btrfs_set_disk_extent_flags(trans
, root
, eb
->start
,
5966 wc
->flags
[level
] |= flag
;
5970 * the block is shared by multiple trees, so it's not good to
5971 * keep the tree lock
5973 if (path
->locks
[level
] && level
> 0) {
5974 btrfs_tree_unlock(eb
);
5975 path
->locks
[level
] = 0;
5981 * hepler to process tree block pointer.
5983 * when wc->stage == DROP_REFERENCE, this function checks
5984 * reference count of the block pointed to. if the block
5985 * is shared and we need update back refs for the subtree
5986 * rooted at the block, this function changes wc->stage to
5987 * UPDATE_BACKREF. if the block is shared and there is no
5988 * need to update back, this function drops the reference
5991 * NOTE: return value 1 means we should stop walking down.
5993 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
5994 struct btrfs_root
*root
,
5995 struct btrfs_path
*path
,
5996 struct walk_control
*wc
, int *lookup_info
)
6002 struct btrfs_key key
;
6003 struct extent_buffer
*next
;
6004 int level
= wc
->level
;
6008 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
6009 path
->slots
[level
]);
6011 * if the lower level block was created before the snapshot
6012 * was created, we know there is no need to update back refs
6015 if (wc
->stage
== UPDATE_BACKREF
&&
6016 generation
<= root
->root_key
.offset
) {
6021 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
6022 blocksize
= btrfs_level_size(root
, level
- 1);
6024 next
= btrfs_find_tree_block(root
, bytenr
, blocksize
);
6026 next
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
6031 btrfs_tree_lock(next
);
6032 btrfs_set_lock_blocking(next
);
6034 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, blocksize
,
6035 &wc
->refs
[level
- 1],
6036 &wc
->flags
[level
- 1]);
6038 BUG_ON(wc
->refs
[level
- 1] == 0);
6041 if (wc
->stage
== DROP_REFERENCE
) {
6042 if (wc
->refs
[level
- 1] > 1) {
6044 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
6047 if (!wc
->update_ref
||
6048 generation
<= root
->root_key
.offset
)
6051 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
6052 path
->slots
[level
]);
6053 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
6057 wc
->stage
= UPDATE_BACKREF
;
6058 wc
->shared_level
= level
- 1;
6062 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
6066 if (!btrfs_buffer_uptodate(next
, generation
)) {
6067 btrfs_tree_unlock(next
);
6068 free_extent_buffer(next
);
6074 if (reada
&& level
== 1)
6075 reada_walk_down(trans
, root
, wc
, path
);
6076 next
= read_tree_block(root
, bytenr
, blocksize
, generation
);
6079 btrfs_tree_lock(next
);
6080 btrfs_set_lock_blocking(next
);
6084 BUG_ON(level
!= btrfs_header_level(next
));
6085 path
->nodes
[level
] = next
;
6086 path
->slots
[level
] = 0;
6087 path
->locks
[level
] = 1;
6093 wc
->refs
[level
- 1] = 0;
6094 wc
->flags
[level
- 1] = 0;
6095 if (wc
->stage
== DROP_REFERENCE
) {
6096 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
6097 parent
= path
->nodes
[level
]->start
;
6099 BUG_ON(root
->root_key
.objectid
!=
6100 btrfs_header_owner(path
->nodes
[level
]));
6104 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
, parent
,
6105 root
->root_key
.objectid
, level
- 1, 0);
6108 btrfs_tree_unlock(next
);
6109 free_extent_buffer(next
);
6115 * hepler to process tree block while walking up the tree.
6117 * when wc->stage == DROP_REFERENCE, this function drops
6118 * reference count on the block.
6120 * when wc->stage == UPDATE_BACKREF, this function changes
6121 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6122 * to UPDATE_BACKREF previously while processing the block.
6124 * NOTE: return value 1 means we should stop walking up.
6126 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
6127 struct btrfs_root
*root
,
6128 struct btrfs_path
*path
,
6129 struct walk_control
*wc
)
6132 int level
= wc
->level
;
6133 struct extent_buffer
*eb
= path
->nodes
[level
];
6136 if (wc
->stage
== UPDATE_BACKREF
) {
6137 BUG_ON(wc
->shared_level
< level
);
6138 if (level
< wc
->shared_level
)
6141 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
6145 wc
->stage
= DROP_REFERENCE
;
6146 wc
->shared_level
= -1;
6147 path
->slots
[level
] = 0;
6150 * check reference count again if the block isn't locked.
6151 * we should start walking down the tree again if reference
6154 if (!path
->locks
[level
]) {
6156 btrfs_tree_lock(eb
);
6157 btrfs_set_lock_blocking(eb
);
6158 path
->locks
[level
] = 1;
6160 ret
= btrfs_lookup_extent_info(trans
, root
,
6165 BUG_ON(wc
->refs
[level
] == 0);
6166 if (wc
->refs
[level
] == 1) {
6167 btrfs_tree_unlock(eb
);
6168 path
->locks
[level
] = 0;
6174 /* wc->stage == DROP_REFERENCE */
6175 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
6177 if (wc
->refs
[level
] == 1) {
6179 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
6180 ret
= btrfs_dec_ref(trans
, root
, eb
, 1);
6182 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
6185 /* make block locked assertion in clean_tree_block happy */
6186 if (!path
->locks
[level
] &&
6187 btrfs_header_generation(eb
) == trans
->transid
) {
6188 btrfs_tree_lock(eb
);
6189 btrfs_set_lock_blocking(eb
);
6190 path
->locks
[level
] = 1;
6192 clean_tree_block(trans
, root
, eb
);
6195 if (eb
== root
->node
) {
6196 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
6199 BUG_ON(root
->root_key
.objectid
!=
6200 btrfs_header_owner(eb
));
6202 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
6203 parent
= path
->nodes
[level
+ 1]->start
;
6205 BUG_ON(root
->root_key
.objectid
!=
6206 btrfs_header_owner(path
->nodes
[level
+ 1]));
6209 btrfs_free_tree_block(trans
, root
, eb
, parent
, wc
->refs
[level
] == 1);
6211 wc
->refs
[level
] = 0;
6212 wc
->flags
[level
] = 0;
6216 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
6217 struct btrfs_root
*root
,
6218 struct btrfs_path
*path
,
6219 struct walk_control
*wc
)
6221 int level
= wc
->level
;
6222 int lookup_info
= 1;
6225 while (level
>= 0) {
6226 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
6233 if (path
->slots
[level
] >=
6234 btrfs_header_nritems(path
->nodes
[level
]))
6237 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
6239 path
->slots
[level
]++;
6248 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
6249 struct btrfs_root
*root
,
6250 struct btrfs_path
*path
,
6251 struct walk_control
*wc
, int max_level
)
6253 int level
= wc
->level
;
6256 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
6257 while (level
< max_level
&& path
->nodes
[level
]) {
6259 if (path
->slots
[level
] + 1 <
6260 btrfs_header_nritems(path
->nodes
[level
])) {
6261 path
->slots
[level
]++;
6264 ret
= walk_up_proc(trans
, root
, path
, wc
);
6268 if (path
->locks
[level
]) {
6269 btrfs_tree_unlock(path
->nodes
[level
]);
6270 path
->locks
[level
] = 0;
6272 free_extent_buffer(path
->nodes
[level
]);
6273 path
->nodes
[level
] = NULL
;
6281 * drop a subvolume tree.
6283 * this function traverses the tree freeing any blocks that only
6284 * referenced by the tree.
6286 * when a shared tree block is found. this function decreases its
6287 * reference count by one. if update_ref is true, this function
6288 * also make sure backrefs for the shared block and all lower level
6289 * blocks are properly updated.
6291 int btrfs_drop_snapshot(struct btrfs_root
*root
,
6292 struct btrfs_block_rsv
*block_rsv
, int update_ref
)
6294 struct btrfs_path
*path
;
6295 struct btrfs_trans_handle
*trans
;
6296 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
6297 struct btrfs_root_item
*root_item
= &root
->root_item
;
6298 struct walk_control
*wc
;
6299 struct btrfs_key key
;
6304 path
= btrfs_alloc_path();
6307 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
6310 trans
= btrfs_start_transaction(tree_root
, 0);
6311 BUG_ON(IS_ERR(trans
));
6314 trans
->block_rsv
= block_rsv
;
6316 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
6317 level
= btrfs_header_level(root
->node
);
6318 path
->nodes
[level
] = btrfs_lock_root_node(root
);
6319 btrfs_set_lock_blocking(path
->nodes
[level
]);
6320 path
->slots
[level
] = 0;
6321 path
->locks
[level
] = 1;
6322 memset(&wc
->update_progress
, 0,
6323 sizeof(wc
->update_progress
));
6325 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
6326 memcpy(&wc
->update_progress
, &key
,
6327 sizeof(wc
->update_progress
));
6329 level
= root_item
->drop_level
;
6331 path
->lowest_level
= level
;
6332 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6333 path
->lowest_level
= 0;
6341 * unlock our path, this is safe because only this
6342 * function is allowed to delete this snapshot
6344 btrfs_unlock_up_safe(path
, 0);
6346 level
= btrfs_header_level(root
->node
);
6348 btrfs_tree_lock(path
->nodes
[level
]);
6349 btrfs_set_lock_blocking(path
->nodes
[level
]);
6351 ret
= btrfs_lookup_extent_info(trans
, root
,
6352 path
->nodes
[level
]->start
,
6353 path
->nodes
[level
]->len
,
6357 BUG_ON(wc
->refs
[level
] == 0);
6359 if (level
== root_item
->drop_level
)
6362 btrfs_tree_unlock(path
->nodes
[level
]);
6363 WARN_ON(wc
->refs
[level
] != 1);
6369 wc
->shared_level
= -1;
6370 wc
->stage
= DROP_REFERENCE
;
6371 wc
->update_ref
= update_ref
;
6373 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
6376 ret
= walk_down_tree(trans
, root
, path
, wc
);
6382 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
6389 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
6393 if (wc
->stage
== DROP_REFERENCE
) {
6395 btrfs_node_key(path
->nodes
[level
],
6396 &root_item
->drop_progress
,
6397 path
->slots
[level
]);
6398 root_item
->drop_level
= level
;
6401 BUG_ON(wc
->level
== 0);
6402 if (btrfs_should_end_transaction(trans
, tree_root
)) {
6403 ret
= btrfs_update_root(trans
, tree_root
,
6408 btrfs_end_transaction_throttle(trans
, tree_root
);
6409 trans
= btrfs_start_transaction(tree_root
, 0);
6410 BUG_ON(IS_ERR(trans
));
6412 trans
->block_rsv
= block_rsv
;
6415 btrfs_release_path(root
, path
);
6418 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
6421 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
6422 ret
= btrfs_find_last_root(tree_root
, root
->root_key
.objectid
,
6426 /* if we fail to delete the orphan item this time
6427 * around, it'll get picked up the next time.
6429 * The most common failure here is just -ENOENT.
6431 btrfs_del_orphan_item(trans
, tree_root
,
6432 root
->root_key
.objectid
);
6436 if (root
->in_radix
) {
6437 btrfs_free_fs_root(tree_root
->fs_info
, root
);
6439 free_extent_buffer(root
->node
);
6440 free_extent_buffer(root
->commit_root
);
6444 btrfs_end_transaction_throttle(trans
, tree_root
);
6446 btrfs_free_path(path
);
6451 * drop subtree rooted at tree block 'node'.
6453 * NOTE: this function will unlock and release tree block 'node'
6455 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
6456 struct btrfs_root
*root
,
6457 struct extent_buffer
*node
,
6458 struct extent_buffer
*parent
)
6460 struct btrfs_path
*path
;
6461 struct walk_control
*wc
;
6467 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
6469 path
= btrfs_alloc_path();
6473 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
6475 btrfs_free_path(path
);
6479 btrfs_assert_tree_locked(parent
);
6480 parent_level
= btrfs_header_level(parent
);
6481 extent_buffer_get(parent
);
6482 path
->nodes
[parent_level
] = parent
;
6483 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
6485 btrfs_assert_tree_locked(node
);
6486 level
= btrfs_header_level(node
);
6487 path
->nodes
[level
] = node
;
6488 path
->slots
[level
] = 0;
6489 path
->locks
[level
] = 1;
6491 wc
->refs
[parent_level
] = 1;
6492 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
6494 wc
->shared_level
= -1;
6495 wc
->stage
= DROP_REFERENCE
;
6498 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
6501 wret
= walk_down_tree(trans
, root
, path
, wc
);
6507 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
6515 btrfs_free_path(path
);
6520 static unsigned long calc_ra(unsigned long start
, unsigned long last
,
6523 return min(last
, start
+ nr
- 1);
6526 static noinline
int relocate_inode_pages(struct inode
*inode
, u64 start
,
6531 unsigned long first_index
;
6532 unsigned long last_index
;
6535 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
6536 struct file_ra_state
*ra
;
6537 struct btrfs_ordered_extent
*ordered
;
6538 unsigned int total_read
= 0;
6539 unsigned int total_dirty
= 0;
6542 ra
= kzalloc(sizeof(*ra
), GFP_NOFS
);
6546 mutex_lock(&inode
->i_mutex
);
6547 first_index
= start
>> PAGE_CACHE_SHIFT
;
6548 last_index
= (start
+ len
- 1) >> PAGE_CACHE_SHIFT
;
6550 /* make sure the dirty trick played by the caller work */
6551 ret
= invalidate_inode_pages2_range(inode
->i_mapping
,
6552 first_index
, last_index
);
6556 file_ra_state_init(ra
, inode
->i_mapping
);
6558 for (i
= first_index
; i
<= last_index
; i
++) {
6559 if (total_read
% ra
->ra_pages
== 0) {
6560 btrfs_force_ra(inode
->i_mapping
, ra
, NULL
, i
,
6561 calc_ra(i
, last_index
, ra
->ra_pages
));
6565 if (((u64
)i
<< PAGE_CACHE_SHIFT
) > i_size_read(inode
))
6567 page
= grab_cache_page(inode
->i_mapping
, i
);
6572 if (!PageUptodate(page
)) {
6573 btrfs_readpage(NULL
, page
);
6575 if (!PageUptodate(page
)) {
6577 page_cache_release(page
);
6582 wait_on_page_writeback(page
);
6584 page_start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
6585 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
6586 lock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
6588 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
6590 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
6592 page_cache_release(page
);
6593 btrfs_start_ordered_extent(inode
, ordered
, 1);
6594 btrfs_put_ordered_extent(ordered
);
6597 set_page_extent_mapped(page
);
6599 if (i
== first_index
)
6600 set_extent_bits(io_tree
, page_start
, page_end
,
6601 EXTENT_BOUNDARY
, GFP_NOFS
);
6602 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
6604 set_page_dirty(page
);
6607 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
6609 page_cache_release(page
);
6614 mutex_unlock(&inode
->i_mutex
);
6615 balance_dirty_pages_ratelimited_nr(inode
->i_mapping
, total_dirty
);
6619 static noinline
int relocate_data_extent(struct inode
*reloc_inode
,
6620 struct btrfs_key
*extent_key
,
6623 struct btrfs_root
*root
= BTRFS_I(reloc_inode
)->root
;
6624 struct extent_map_tree
*em_tree
= &BTRFS_I(reloc_inode
)->extent_tree
;
6625 struct extent_map
*em
;
6626 u64 start
= extent_key
->objectid
- offset
;
6627 u64 end
= start
+ extent_key
->offset
- 1;
6629 em
= alloc_extent_map(GFP_NOFS
);
6633 em
->len
= extent_key
->offset
;
6634 em
->block_len
= extent_key
->offset
;
6635 em
->block_start
= extent_key
->objectid
;
6636 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
6637 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
6639 /* setup extent map to cheat btrfs_readpage */
6640 lock_extent(&BTRFS_I(reloc_inode
)->io_tree
, start
, end
, GFP_NOFS
);
6643 write_lock(&em_tree
->lock
);
6644 ret
= add_extent_mapping(em_tree
, em
);
6645 write_unlock(&em_tree
->lock
);
6646 if (ret
!= -EEXIST
) {
6647 free_extent_map(em
);
6650 btrfs_drop_extent_cache(reloc_inode
, start
, end
, 0);
6652 unlock_extent(&BTRFS_I(reloc_inode
)->io_tree
, start
, end
, GFP_NOFS
);
6654 return relocate_inode_pages(reloc_inode
, start
, extent_key
->offset
);
6657 struct btrfs_ref_path
{
6659 u64 nodes
[BTRFS_MAX_LEVEL
];
6661 u64 root_generation
;
6668 struct btrfs_key node_keys
[BTRFS_MAX_LEVEL
];
6669 u64 new_nodes
[BTRFS_MAX_LEVEL
];
6672 struct disk_extent
{
6683 static int is_cowonly_root(u64 root_objectid
)
6685 if (root_objectid
== BTRFS_ROOT_TREE_OBJECTID
||
6686 root_objectid
== BTRFS_EXTENT_TREE_OBJECTID
||
6687 root_objectid
== BTRFS_CHUNK_TREE_OBJECTID
||
6688 root_objectid
== BTRFS_DEV_TREE_OBJECTID
||
6689 root_objectid
== BTRFS_TREE_LOG_OBJECTID
||
6690 root_objectid
== BTRFS_CSUM_TREE_OBJECTID
)
6695 static noinline
int __next_ref_path(struct btrfs_trans_handle
*trans
,
6696 struct btrfs_root
*extent_root
,
6697 struct btrfs_ref_path
*ref_path
,
6700 struct extent_buffer
*leaf
;
6701 struct btrfs_path
*path
;
6702 struct btrfs_extent_ref
*ref
;
6703 struct btrfs_key key
;
6704 struct btrfs_key found_key
;
6710 path
= btrfs_alloc_path();
6715 ref_path
->lowest_level
= -1;
6716 ref_path
->current_level
= -1;
6717 ref_path
->shared_level
= -1;
6721 level
= ref_path
->current_level
- 1;
6722 while (level
>= -1) {
6724 if (level
< ref_path
->lowest_level
)
6728 bytenr
= ref_path
->nodes
[level
];
6730 bytenr
= ref_path
->extent_start
;
6731 BUG_ON(bytenr
== 0);
6733 parent
= ref_path
->nodes
[level
+ 1];
6734 ref_path
->nodes
[level
+ 1] = 0;
6735 ref_path
->current_level
= level
;
6736 BUG_ON(parent
== 0);
6738 key
.objectid
= bytenr
;
6739 key
.offset
= parent
+ 1;
6740 key
.type
= BTRFS_EXTENT_REF_KEY
;
6742 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
, 0, 0);
6747 leaf
= path
->nodes
[0];
6748 nritems
= btrfs_header_nritems(leaf
);
6749 if (path
->slots
[0] >= nritems
) {
6750 ret
= btrfs_next_leaf(extent_root
, path
);
6755 leaf
= path
->nodes
[0];
6758 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6759 if (found_key
.objectid
== bytenr
&&
6760 found_key
.type
== BTRFS_EXTENT_REF_KEY
) {
6761 if (level
< ref_path
->shared_level
)
6762 ref_path
->shared_level
= level
;
6767 btrfs_release_path(extent_root
, path
);
6770 /* reached lowest level */
6774 level
= ref_path
->current_level
;
6775 while (level
< BTRFS_MAX_LEVEL
- 1) {
6779 bytenr
= ref_path
->nodes
[level
];
6781 bytenr
= ref_path
->extent_start
;
6783 BUG_ON(bytenr
== 0);
6785 key
.objectid
= bytenr
;
6787 key
.type
= BTRFS_EXTENT_REF_KEY
;
6789 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
, 0, 0);
6793 leaf
= path
->nodes
[0];
6794 nritems
= btrfs_header_nritems(leaf
);
6795 if (path
->slots
[0] >= nritems
) {
6796 ret
= btrfs_next_leaf(extent_root
, path
);
6800 /* the extent was freed by someone */
6801 if (ref_path
->lowest_level
== level
)
6803 btrfs_release_path(extent_root
, path
);
6806 leaf
= path
->nodes
[0];
6809 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6810 if (found_key
.objectid
!= bytenr
||
6811 found_key
.type
!= BTRFS_EXTENT_REF_KEY
) {
6812 /* the extent was freed by someone */
6813 if (ref_path
->lowest_level
== level
) {
6817 btrfs_release_path(extent_root
, path
);
6821 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
6822 struct btrfs_extent_ref
);
6823 ref_objectid
= btrfs_ref_objectid(leaf
, ref
);
6824 if (ref_objectid
< BTRFS_FIRST_FREE_OBJECTID
) {
6826 level
= (int)ref_objectid
;
6827 BUG_ON(level
>= BTRFS_MAX_LEVEL
);
6828 ref_path
->lowest_level
= level
;
6829 ref_path
->current_level
= level
;
6830 ref_path
->nodes
[level
] = bytenr
;
6832 WARN_ON(ref_objectid
!= level
);
6835 WARN_ON(level
!= -1);
6839 if (ref_path
->lowest_level
== level
) {
6840 ref_path
->owner_objectid
= ref_objectid
;
6841 ref_path
->num_refs
= btrfs_ref_num_refs(leaf
, ref
);
6845 * the block is tree root or the block isn't in reference
6848 if (found_key
.objectid
== found_key
.offset
||
6849 is_cowonly_root(btrfs_ref_root(leaf
, ref
))) {
6850 ref_path
->root_objectid
= btrfs_ref_root(leaf
, ref
);
6851 ref_path
->root_generation
=
6852 btrfs_ref_generation(leaf
, ref
);
6854 /* special reference from the tree log */
6855 ref_path
->nodes
[0] = found_key
.offset
;
6856 ref_path
->current_level
= 0;
6863 BUG_ON(ref_path
->nodes
[level
] != 0);
6864 ref_path
->nodes
[level
] = found_key
.offset
;
6865 ref_path
->current_level
= level
;
6868 * the reference was created in the running transaction,
6869 * no need to continue walking up.
6871 if (btrfs_ref_generation(leaf
, ref
) == trans
->transid
) {
6872 ref_path
->root_objectid
= btrfs_ref_root(leaf
, ref
);
6873 ref_path
->root_generation
=
6874 btrfs_ref_generation(leaf
, ref
);
6879 btrfs_release_path(extent_root
, path
);
6882 /* reached max tree level, but no tree root found. */
6885 btrfs_free_path(path
);
6889 static int btrfs_first_ref_path(struct btrfs_trans_handle
*trans
,
6890 struct btrfs_root
*extent_root
,
6891 struct btrfs_ref_path
*ref_path
,
6894 memset(ref_path
, 0, sizeof(*ref_path
));
6895 ref_path
->extent_start
= extent_start
;
6897 return __next_ref_path(trans
, extent_root
, ref_path
, 1);
6900 static int btrfs_next_ref_path(struct btrfs_trans_handle
*trans
,
6901 struct btrfs_root
*extent_root
,
6902 struct btrfs_ref_path
*ref_path
)
6904 return __next_ref_path(trans
, extent_root
, ref_path
, 0);
6907 static noinline
int get_new_locations(struct inode
*reloc_inode
,
6908 struct btrfs_key
*extent_key
,
6909 u64 offset
, int no_fragment
,
6910 struct disk_extent
**extents
,
6913 struct btrfs_root
*root
= BTRFS_I(reloc_inode
)->root
;
6914 struct btrfs_path
*path
;
6915 struct btrfs_file_extent_item
*fi
;
6916 struct extent_buffer
*leaf
;
6917 struct disk_extent
*exts
= *extents
;
6918 struct btrfs_key found_key
;
6923 int max
= *nr_extents
;
6926 WARN_ON(!no_fragment
&& *extents
);
6929 exts
= kmalloc(sizeof(*exts
) * max
, GFP_NOFS
);
6934 path
= btrfs_alloc_path();
6936 if (exts
!= *extents
)
6941 cur_pos
= extent_key
->objectid
- offset
;
6942 last_byte
= extent_key
->objectid
+ extent_key
->offset
;
6943 ret
= btrfs_lookup_file_extent(NULL
, root
, path
, reloc_inode
->i_ino
,
6953 leaf
= path
->nodes
[0];
6954 nritems
= btrfs_header_nritems(leaf
);
6955 if (path
->slots
[0] >= nritems
) {
6956 ret
= btrfs_next_leaf(root
, path
);
6961 leaf
= path
->nodes
[0];
6964 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6965 if (found_key
.offset
!= cur_pos
||
6966 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
||
6967 found_key
.objectid
!= reloc_inode
->i_ino
)
6970 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
6971 struct btrfs_file_extent_item
);
6972 if (btrfs_file_extent_type(leaf
, fi
) !=
6973 BTRFS_FILE_EXTENT_REG
||
6974 btrfs_file_extent_disk_bytenr(leaf
, fi
) == 0)
6978 struct disk_extent
*old
= exts
;
6980 exts
= kzalloc(sizeof(*exts
) * max
, GFP_NOFS
);
6985 memcpy(exts
, old
, sizeof(*exts
) * nr
);
6986 if (old
!= *extents
)
6990 exts
[nr
].disk_bytenr
=
6991 btrfs_file_extent_disk_bytenr(leaf
, fi
);
6992 exts
[nr
].disk_num_bytes
=
6993 btrfs_file_extent_disk_num_bytes(leaf
, fi
);
6994 exts
[nr
].offset
= btrfs_file_extent_offset(leaf
, fi
);
6995 exts
[nr
].num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
6996 exts
[nr
].ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
6997 exts
[nr
].compression
= btrfs_file_extent_compression(leaf
, fi
);
6998 exts
[nr
].encryption
= btrfs_file_extent_encryption(leaf
, fi
);
6999 exts
[nr
].other_encoding
= btrfs_file_extent_other_encoding(leaf
,
7001 BUG_ON(exts
[nr
].offset
> 0);
7002 BUG_ON(exts
[nr
].compression
|| exts
[nr
].encryption
);
7003 BUG_ON(exts
[nr
].num_bytes
!= exts
[nr
].disk_num_bytes
);
7005 cur_pos
+= exts
[nr
].num_bytes
;
7008 if (cur_pos
+ offset
>= last_byte
)
7018 BUG_ON(cur_pos
+ offset
> last_byte
);
7019 if (cur_pos
+ offset
< last_byte
) {
7025 btrfs_free_path(path
);
7027 if (exts
!= *extents
)
7036 static noinline
int replace_one_extent(struct btrfs_trans_handle
*trans
,
7037 struct btrfs_root
*root
,
7038 struct btrfs_path
*path
,
7039 struct btrfs_key
*extent_key
,
7040 struct btrfs_key
*leaf_key
,
7041 struct btrfs_ref_path
*ref_path
,
7042 struct disk_extent
*new_extents
,
7045 struct extent_buffer
*leaf
;
7046 struct btrfs_file_extent_item
*fi
;
7047 struct inode
*inode
= NULL
;
7048 struct btrfs_key key
;
7053 u64 search_end
= (u64
)-1;
7056 int extent_locked
= 0;
7060 memcpy(&key
, leaf_key
, sizeof(key
));
7061 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
) {
7062 if (key
.objectid
< ref_path
->owner_objectid
||
7063 (key
.objectid
== ref_path
->owner_objectid
&&
7064 key
.type
< BTRFS_EXTENT_DATA_KEY
)) {
7065 key
.objectid
= ref_path
->owner_objectid
;
7066 key
.type
= BTRFS_EXTENT_DATA_KEY
;
7072 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
7076 leaf
= path
->nodes
[0];
7077 nritems
= btrfs_header_nritems(leaf
);
7079 if (extent_locked
&& ret
> 0) {
7081 * the file extent item was modified by someone
7082 * before the extent got locked.
7084 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
7085 lock_end
, GFP_NOFS
);
7089 if (path
->slots
[0] >= nritems
) {
7090 if (++nr_scaned
> 2)
7093 BUG_ON(extent_locked
);
7094 ret
= btrfs_next_leaf(root
, path
);
7099 leaf
= path
->nodes
[0];
7100 nritems
= btrfs_header_nritems(leaf
);
7103 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
7105 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
) {
7106 if ((key
.objectid
> ref_path
->owner_objectid
) ||
7107 (key
.objectid
== ref_path
->owner_objectid
&&
7108 key
.type
> BTRFS_EXTENT_DATA_KEY
) ||
7109 key
.offset
>= search_end
)
7113 if (inode
&& key
.objectid
!= inode
->i_ino
) {
7114 BUG_ON(extent_locked
);
7115 btrfs_release_path(root
, path
);
7116 mutex_unlock(&inode
->i_mutex
);
7122 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
7127 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
7128 struct btrfs_file_extent_item
);
7129 extent_type
= btrfs_file_extent_type(leaf
, fi
);
7130 if ((extent_type
!= BTRFS_FILE_EXTENT_REG
&&
7131 extent_type
!= BTRFS_FILE_EXTENT_PREALLOC
) ||
7132 (btrfs_file_extent_disk_bytenr(leaf
, fi
) !=
7133 extent_key
->objectid
)) {
7139 num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
7140 ext_offset
= btrfs_file_extent_offset(leaf
, fi
);
7142 if (search_end
== (u64
)-1) {
7143 search_end
= key
.offset
- ext_offset
+
7144 btrfs_file_extent_ram_bytes(leaf
, fi
);
7147 if (!extent_locked
) {
7148 lock_start
= key
.offset
;
7149 lock_end
= lock_start
+ num_bytes
- 1;
7151 if (lock_start
> key
.offset
||
7152 lock_end
+ 1 < key
.offset
+ num_bytes
) {
7153 unlock_extent(&BTRFS_I(inode
)->io_tree
,
7154 lock_start
, lock_end
, GFP_NOFS
);
7160 btrfs_release_path(root
, path
);
7162 inode
= btrfs_iget_locked(root
->fs_info
->sb
,
7163 key
.objectid
, root
);
7164 if (inode
->i_state
& I_NEW
) {
7165 BTRFS_I(inode
)->root
= root
;
7166 BTRFS_I(inode
)->location
.objectid
=
7168 BTRFS_I(inode
)->location
.type
=
7169 BTRFS_INODE_ITEM_KEY
;
7170 BTRFS_I(inode
)->location
.offset
= 0;
7171 btrfs_read_locked_inode(inode
);
7172 unlock_new_inode(inode
);
7175 * some code call btrfs_commit_transaction while
7176 * holding the i_mutex, so we can't use mutex_lock
7179 if (is_bad_inode(inode
) ||
7180 !mutex_trylock(&inode
->i_mutex
)) {
7183 key
.offset
= (u64
)-1;
7188 if (!extent_locked
) {
7189 struct btrfs_ordered_extent
*ordered
;
7191 btrfs_release_path(root
, path
);
7193 lock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
7194 lock_end
, GFP_NOFS
);
7195 ordered
= btrfs_lookup_first_ordered_extent(inode
,
7198 ordered
->file_offset
<= lock_end
&&
7199 ordered
->file_offset
+ ordered
->len
> lock_start
) {
7200 unlock_extent(&BTRFS_I(inode
)->io_tree
,
7201 lock_start
, lock_end
, GFP_NOFS
);
7202 btrfs_start_ordered_extent(inode
, ordered
, 1);
7203 btrfs_put_ordered_extent(ordered
);
7204 key
.offset
+= num_bytes
;
7208 btrfs_put_ordered_extent(ordered
);
7214 if (nr_extents
== 1) {
7215 /* update extent pointer in place */
7216 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
7217 new_extents
[0].disk_bytenr
);
7218 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
7219 new_extents
[0].disk_num_bytes
);
7220 btrfs_mark_buffer_dirty(leaf
);
7222 btrfs_drop_extent_cache(inode
, key
.offset
,
7223 key
.offset
+ num_bytes
- 1, 0);
7225 ret
= btrfs_inc_extent_ref(trans
, root
,
7226 new_extents
[0].disk_bytenr
,
7227 new_extents
[0].disk_num_bytes
,
7229 root
->root_key
.objectid
,
7234 ret
= btrfs_free_extent(trans
, root
,
7235 extent_key
->objectid
,
7238 btrfs_header_owner(leaf
),
7239 btrfs_header_generation(leaf
),
7243 btrfs_release_path(root
, path
);
7244 key
.offset
+= num_bytes
;
7252 * drop old extent pointer at first, then insert the
7253 * new pointers one bye one
7255 btrfs_release_path(root
, path
);
7256 ret
= btrfs_drop_extents(trans
, root
, inode
, key
.offset
,
7257 key
.offset
+ num_bytes
,
7258 key
.offset
, &alloc_hint
);
7261 for (i
= 0; i
< nr_extents
; i
++) {
7262 if (ext_offset
>= new_extents
[i
].num_bytes
) {
7263 ext_offset
-= new_extents
[i
].num_bytes
;
7266 extent_len
= min(new_extents
[i
].num_bytes
-
7267 ext_offset
, num_bytes
);
7269 ret
= btrfs_insert_empty_item(trans
, root
,
7274 leaf
= path
->nodes
[0];
7275 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
7276 struct btrfs_file_extent_item
);
7277 btrfs_set_file_extent_generation(leaf
, fi
,
7279 btrfs_set_file_extent_type(leaf
, fi
,
7280 BTRFS_FILE_EXTENT_REG
);
7281 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
7282 new_extents
[i
].disk_bytenr
);
7283 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
7284 new_extents
[i
].disk_num_bytes
);
7285 btrfs_set_file_extent_ram_bytes(leaf
, fi
,
7286 new_extents
[i
].ram_bytes
);
7288 btrfs_set_file_extent_compression(leaf
, fi
,
7289 new_extents
[i
].compression
);
7290 btrfs_set_file_extent_encryption(leaf
, fi
,
7291 new_extents
[i
].encryption
);
7292 btrfs_set_file_extent_other_encoding(leaf
, fi
,
7293 new_extents
[i
].other_encoding
);
7295 btrfs_set_file_extent_num_bytes(leaf
, fi
,
7297 ext_offset
+= new_extents
[i
].offset
;
7298 btrfs_set_file_extent_offset(leaf
, fi
,
7300 btrfs_mark_buffer_dirty(leaf
);
7302 btrfs_drop_extent_cache(inode
, key
.offset
,
7303 key
.offset
+ extent_len
- 1, 0);
7305 ret
= btrfs_inc_extent_ref(trans
, root
,
7306 new_extents
[i
].disk_bytenr
,
7307 new_extents
[i
].disk_num_bytes
,
7309 root
->root_key
.objectid
,
7310 trans
->transid
, key
.objectid
);
7312 btrfs_release_path(root
, path
);
7314 inode_add_bytes(inode
, extent_len
);
7317 num_bytes
-= extent_len
;
7318 key
.offset
+= extent_len
;
7323 BUG_ON(i
>= nr_extents
);
7327 if (extent_locked
) {
7328 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
7329 lock_end
, GFP_NOFS
);
7333 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
&&
7334 key
.offset
>= search_end
)
7341 btrfs_release_path(root
, path
);
7343 mutex_unlock(&inode
->i_mutex
);
7344 if (extent_locked
) {
7345 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
7346 lock_end
, GFP_NOFS
);
7353 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle
*trans
,
7354 struct btrfs_root
*root
,
7355 struct extent_buffer
*buf
, u64 orig_start
)
7360 BUG_ON(btrfs_header_generation(buf
) != trans
->transid
);
7361 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
7363 level
= btrfs_header_level(buf
);
7365 struct btrfs_leaf_ref
*ref
;
7366 struct btrfs_leaf_ref
*orig_ref
;
7368 orig_ref
= btrfs_lookup_leaf_ref(root
, orig_start
);
7372 ref
= btrfs_alloc_leaf_ref(root
, orig_ref
->nritems
);
7374 btrfs_free_leaf_ref(root
, orig_ref
);
7378 ref
->nritems
= orig_ref
->nritems
;
7379 memcpy(ref
->extents
, orig_ref
->extents
,
7380 sizeof(ref
->extents
[0]) * ref
->nritems
);
7382 btrfs_free_leaf_ref(root
, orig_ref
);
7384 ref
->root_gen
= trans
->transid
;
7385 ref
->bytenr
= buf
->start
;
7386 ref
->owner
= btrfs_header_owner(buf
);
7387 ref
->generation
= btrfs_header_generation(buf
);
7389 ret
= btrfs_add_leaf_ref(root
, ref
, 0);
7391 btrfs_free_leaf_ref(root
, ref
);
7396 static noinline
int invalidate_extent_cache(struct btrfs_root
*root
,
7397 struct extent_buffer
*leaf
,
7398 struct btrfs_block_group_cache
*group
,
7399 struct btrfs_root
*target_root
)
7401 struct btrfs_key key
;
7402 struct inode
*inode
= NULL
;
7403 struct btrfs_file_extent_item
*fi
;
7404 struct extent_state
*cached_state
= NULL
;
7406 u64 skip_objectid
= 0;
7410 nritems
= btrfs_header_nritems(leaf
);
7411 for (i
= 0; i
< nritems
; i
++) {
7412 btrfs_item_key_to_cpu(leaf
, &key
, i
);
7413 if (key
.objectid
== skip_objectid
||
7414 key
.type
!= BTRFS_EXTENT_DATA_KEY
)
7416 fi
= btrfs_item_ptr(leaf
, i
, struct btrfs_file_extent_item
);
7417 if (btrfs_file_extent_type(leaf
, fi
) ==
7418 BTRFS_FILE_EXTENT_INLINE
)
7420 if (btrfs_file_extent_disk_bytenr(leaf
, fi
) == 0)
7422 if (!inode
|| inode
->i_ino
!= key
.objectid
) {
7424 inode
= btrfs_ilookup(target_root
->fs_info
->sb
,
7425 key
.objectid
, target_root
, 1);
7428 skip_objectid
= key
.objectid
;
7431 num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
7433 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, key
.offset
,
7434 key
.offset
+ num_bytes
- 1, 0, &cached_state
,
7436 btrfs_drop_extent_cache(inode
, key
.offset
,
7437 key
.offset
+ num_bytes
- 1, 1);
7438 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, key
.offset
,
7439 key
.offset
+ num_bytes
- 1, &cached_state
,
7447 static noinline
int replace_extents_in_leaf(struct btrfs_trans_handle
*trans
,
7448 struct btrfs_root
*root
,
7449 struct extent_buffer
*leaf
,
7450 struct btrfs_block_group_cache
*group
,
7451 struct inode
*reloc_inode
)
7453 struct btrfs_key key
;
7454 struct btrfs_key extent_key
;
7455 struct btrfs_file_extent_item
*fi
;
7456 struct btrfs_leaf_ref
*ref
;
7457 struct disk_extent
*new_extent
;
7466 new_extent
= kmalloc(sizeof(*new_extent
), GFP_NOFS
);
7470 ref
= btrfs_lookup_leaf_ref(root
, leaf
->start
);
7474 nritems
= btrfs_header_nritems(leaf
);
7475 for (i
= 0; i
< nritems
; i
++) {
7476 btrfs_item_key_to_cpu(leaf
, &key
, i
);
7477 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
7479 fi
= btrfs_item_ptr(leaf
, i
, struct btrfs_file_extent_item
);
7480 if (btrfs_file_extent_type(leaf
, fi
) ==
7481 BTRFS_FILE_EXTENT_INLINE
)
7483 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
7484 num_bytes
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
7489 if (bytenr
>= group
->key
.objectid
+ group
->key
.offset
||
7490 bytenr
+ num_bytes
<= group
->key
.objectid
)
7493 extent_key
.objectid
= bytenr
;
7494 extent_key
.offset
= num_bytes
;
7495 extent_key
.type
= BTRFS_EXTENT_ITEM_KEY
;
7497 ret
= get_new_locations(reloc_inode
, &extent_key
,
7498 group
->key
.objectid
, 1,
7499 &new_extent
, &nr_extent
);
7504 BUG_ON(ref
->extents
[ext_index
].bytenr
!= bytenr
);
7505 BUG_ON(ref
->extents
[ext_index
].num_bytes
!= num_bytes
);
7506 ref
->extents
[ext_index
].bytenr
= new_extent
->disk_bytenr
;
7507 ref
->extents
[ext_index
].num_bytes
= new_extent
->disk_num_bytes
;
7509 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
7510 new_extent
->disk_bytenr
);
7511 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
7512 new_extent
->disk_num_bytes
);
7513 btrfs_mark_buffer_dirty(leaf
);
7515 ret
= btrfs_inc_extent_ref(trans
, root
,
7516 new_extent
->disk_bytenr
,
7517 new_extent
->disk_num_bytes
,
7519 root
->root_key
.objectid
,
7520 trans
->transid
, key
.objectid
);
7523 ret
= btrfs_free_extent(trans
, root
,
7524 bytenr
, num_bytes
, leaf
->start
,
7525 btrfs_header_owner(leaf
),
7526 btrfs_header_generation(leaf
),
7532 BUG_ON(ext_index
+ 1 != ref
->nritems
);
7533 btrfs_free_leaf_ref(root
, ref
);
7537 int btrfs_free_reloc_root(struct btrfs_trans_handle
*trans
,
7538 struct btrfs_root
*root
)
7540 struct btrfs_root
*reloc_root
;
7543 if (root
->reloc_root
) {
7544 reloc_root
= root
->reloc_root
;
7545 root
->reloc_root
= NULL
;
7546 list_add(&reloc_root
->dead_list
,
7547 &root
->fs_info
->dead_reloc_roots
);
7549 btrfs_set_root_bytenr(&reloc_root
->root_item
,
7550 reloc_root
->node
->start
);
7551 btrfs_set_root_level(&root
->root_item
,
7552 btrfs_header_level(reloc_root
->node
));
7553 memset(&reloc_root
->root_item
.drop_progress
, 0,
7554 sizeof(struct btrfs_disk_key
));
7555 reloc_root
->root_item
.drop_level
= 0;
7557 ret
= btrfs_update_root(trans
, root
->fs_info
->tree_root
,
7558 &reloc_root
->root_key
,
7559 &reloc_root
->root_item
);
7565 int btrfs_drop_dead_reloc_roots(struct btrfs_root
*root
)
7567 struct btrfs_trans_handle
*trans
;
7568 struct btrfs_root
*reloc_root
;
7569 struct btrfs_root
*prev_root
= NULL
;
7570 struct list_head dead_roots
;
7574 INIT_LIST_HEAD(&dead_roots
);
7575 list_splice_init(&root
->fs_info
->dead_reloc_roots
, &dead_roots
);
7577 while (!list_empty(&dead_roots
)) {
7578 reloc_root
= list_entry(dead_roots
.prev
,
7579 struct btrfs_root
, dead_list
);
7580 list_del_init(&reloc_root
->dead_list
);
7582 BUG_ON(reloc_root
->commit_root
!= NULL
);
7584 trans
= btrfs_join_transaction(root
, 1);
7585 BUG_ON(IS_ERR(trans
));
7587 mutex_lock(&root
->fs_info
->drop_mutex
);
7588 ret
= btrfs_drop_snapshot(trans
, reloc_root
);
7591 mutex_unlock(&root
->fs_info
->drop_mutex
);
7593 nr
= trans
->blocks_used
;
7594 ret
= btrfs_end_transaction(trans
, root
);
7596 btrfs_btree_balance_dirty(root
, nr
);
7599 free_extent_buffer(reloc_root
->node
);
7601 ret
= btrfs_del_root(trans
, root
->fs_info
->tree_root
,
7602 &reloc_root
->root_key
);
7604 mutex_unlock(&root
->fs_info
->drop_mutex
);
7606 nr
= trans
->blocks_used
;
7607 ret
= btrfs_end_transaction(trans
, root
);
7609 btrfs_btree_balance_dirty(root
, nr
);
7612 prev_root
= reloc_root
;
7615 btrfs_remove_leaf_refs(prev_root
, (u64
)-1, 0);
7621 int btrfs_add_dead_reloc_root(struct btrfs_root
*root
)
7623 list_add(&root
->dead_list
, &root
->fs_info
->dead_reloc_roots
);
7627 int btrfs_cleanup_reloc_trees(struct btrfs_root
*root
)
7629 struct btrfs_root
*reloc_root
;
7630 struct btrfs_trans_handle
*trans
;
7631 struct btrfs_key location
;
7635 mutex_lock(&root
->fs_info
->tree_reloc_mutex
);
7636 ret
= btrfs_find_dead_roots(root
, BTRFS_TREE_RELOC_OBJECTID
, NULL
);
7638 found
= !list_empty(&root
->fs_info
->dead_reloc_roots
);
7639 mutex_unlock(&root
->fs_info
->tree_reloc_mutex
);
7642 trans
= btrfs_start_transaction(root
, 1);
7643 BUG_ON(IS_ERR(trans
));
7644 ret
= btrfs_commit_transaction(trans
, root
);
7648 location
.objectid
= BTRFS_DATA_RELOC_TREE_OBJECTID
;
7649 location
.offset
= (u64
)-1;
7650 location
.type
= BTRFS_ROOT_ITEM_KEY
;
7652 reloc_root
= btrfs_read_fs_root_no_name(root
->fs_info
, &location
);
7653 BUG_ON(!reloc_root
);
7654 ret
= btrfs_orphan_cleanup(reloc_root
);
7659 static noinline
int init_reloc_tree(struct btrfs_trans_handle
*trans
,
7660 struct btrfs_root
*root
)
7662 struct btrfs_root
*reloc_root
;
7663 struct extent_buffer
*eb
;
7664 struct btrfs_root_item
*root_item
;
7665 struct btrfs_key root_key
;
7668 BUG_ON(!root
->ref_cows
);
7669 if (root
->reloc_root
)
7672 root_item
= kmalloc(sizeof(*root_item
), GFP_NOFS
);
7676 ret
= btrfs_copy_root(trans
, root
, root
->commit_root
,
7677 &eb
, BTRFS_TREE_RELOC_OBJECTID
);
7680 root_key
.objectid
= BTRFS_TREE_RELOC_OBJECTID
;
7681 root_key
.offset
= root
->root_key
.objectid
;
7682 root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
7684 memcpy(root_item
, &root
->root_item
, sizeof(root_item
));
7685 btrfs_set_root_refs(root_item
, 0);
7686 btrfs_set_root_bytenr(root_item
, eb
->start
);
7687 btrfs_set_root_level(root_item
, btrfs_header_level(eb
));
7688 btrfs_set_root_generation(root_item
, trans
->transid
);
7690 btrfs_tree_unlock(eb
);
7691 free_extent_buffer(eb
);
7693 ret
= btrfs_insert_root(trans
, root
->fs_info
->tree_root
,
7694 &root_key
, root_item
);
7698 reloc_root
= btrfs_read_fs_root_no_radix(root
->fs_info
->tree_root
,
7700 BUG_ON(IS_ERR(reloc_root
));
7701 reloc_root
->last_trans
= trans
->transid
;
7702 reloc_root
->commit_root
= NULL
;
7703 reloc_root
->ref_tree
= &root
->fs_info
->reloc_ref_tree
;
7705 root
->reloc_root
= reloc_root
;
7710 * Core function of space balance.
7712 * The idea is using reloc trees to relocate tree blocks in reference
7713 * counted roots. There is one reloc tree for each subvol, and all
7714 * reloc trees share same root key objectid. Reloc trees are snapshots
7715 * of the latest committed roots of subvols (root->commit_root).
7717 * To relocate a tree block referenced by a subvol, there are two steps.
7718 * COW the block through subvol's reloc tree, then update block pointer
7719 * in the subvol to point to the new block. Since all reloc trees share
7720 * same root key objectid, doing special handing for tree blocks owned
7721 * by them is easy. Once a tree block has been COWed in one reloc tree,
7722 * we can use the resulting new block directly when the same block is
7723 * required to COW again through other reloc trees. By this way, relocated
7724 * tree blocks are shared between reloc trees, so they are also shared
7727 static noinline
int relocate_one_path(struct btrfs_trans_handle
*trans
,
7728 struct btrfs_root
*root
,
7729 struct btrfs_path
*path
,
7730 struct btrfs_key
*first_key
,
7731 struct btrfs_ref_path
*ref_path
,
7732 struct btrfs_block_group_cache
*group
,
7733 struct inode
*reloc_inode
)
7735 struct btrfs_root
*reloc_root
;
7736 struct extent_buffer
*eb
= NULL
;
7737 struct btrfs_key
*keys
;
7741 int lowest_level
= 0;
7744 if (ref_path
->owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
)
7745 lowest_level
= ref_path
->owner_objectid
;
7747 if (!root
->ref_cows
) {
7748 path
->lowest_level
= lowest_level
;
7749 ret
= btrfs_search_slot(trans
, root
, first_key
, path
, 0, 1);
7751 path
->lowest_level
= 0;
7752 btrfs_release_path(root
, path
);
7756 mutex_lock(&root
->fs_info
->tree_reloc_mutex
);
7757 ret
= init_reloc_tree(trans
, root
);
7759 reloc_root
= root
->reloc_root
;
7761 shared_level
= ref_path
->shared_level
;
7762 ref_path
->shared_level
= BTRFS_MAX_LEVEL
- 1;
7764 keys
= ref_path
->node_keys
;
7765 nodes
= ref_path
->new_nodes
;
7766 memset(&keys
[shared_level
+ 1], 0,
7767 sizeof(*keys
) * (BTRFS_MAX_LEVEL
- shared_level
- 1));
7768 memset(&nodes
[shared_level
+ 1], 0,
7769 sizeof(*nodes
) * (BTRFS_MAX_LEVEL
- shared_level
- 1));
7771 if (nodes
[lowest_level
] == 0) {
7772 path
->lowest_level
= lowest_level
;
7773 ret
= btrfs_search_slot(trans
, reloc_root
, first_key
, path
,
7776 for (level
= lowest_level
; level
< BTRFS_MAX_LEVEL
; level
++) {
7777 eb
= path
->nodes
[level
];
7778 if (!eb
|| eb
== reloc_root
->node
)
7780 nodes
[level
] = eb
->start
;
7782 btrfs_item_key_to_cpu(eb
, &keys
[level
], 0);
7784 btrfs_node_key_to_cpu(eb
, &keys
[level
], 0);
7787 ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
7788 eb
= path
->nodes
[0];
7789 ret
= replace_extents_in_leaf(trans
, reloc_root
, eb
,
7790 group
, reloc_inode
);
7793 btrfs_release_path(reloc_root
, path
);
7795 ret
= btrfs_merge_path(trans
, reloc_root
, keys
, nodes
,
7801 * replace tree blocks in the fs tree with tree blocks in
7804 ret
= btrfs_merge_path(trans
, root
, keys
, nodes
, lowest_level
);
7807 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
7808 ret
= btrfs_search_slot(trans
, reloc_root
, first_key
, path
,
7811 extent_buffer_get(path
->nodes
[0]);
7812 eb
= path
->nodes
[0];
7813 btrfs_release_path(reloc_root
, path
);
7814 ret
= invalidate_extent_cache(reloc_root
, eb
, group
, root
);
7816 free_extent_buffer(eb
);
7819 mutex_unlock(&root
->fs_info
->tree_reloc_mutex
);
7820 path
->lowest_level
= 0;
7824 static noinline
int relocate_tree_block(struct btrfs_trans_handle
*trans
,
7825 struct btrfs_root
*root
,
7826 struct btrfs_path
*path
,
7827 struct btrfs_key
*first_key
,
7828 struct btrfs_ref_path
*ref_path
)
7832 ret
= relocate_one_path(trans
, root
, path
, first_key
,
7833 ref_path
, NULL
, NULL
);
7839 static noinline
int del_extent_zero(struct btrfs_trans_handle
*trans
,
7840 struct btrfs_root
*extent_root
,
7841 struct btrfs_path
*path
,
7842 struct btrfs_key
*extent_key
)
7846 ret
= btrfs_search_slot(trans
, extent_root
, extent_key
, path
, -1, 1);
7849 ret
= btrfs_del_item(trans
, extent_root
, path
);
7851 btrfs_release_path(extent_root
, path
);
7855 static noinline
struct btrfs_root
*read_ref_root(struct btrfs_fs_info
*fs_info
,
7856 struct btrfs_ref_path
*ref_path
)
7858 struct btrfs_key root_key
;
7860 root_key
.objectid
= ref_path
->root_objectid
;
7861 root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
7862 if (is_cowonly_root(ref_path
->root_objectid
))
7863 root_key
.offset
= 0;
7865 root_key
.offset
= (u64
)-1;
7867 return btrfs_read_fs_root_no_name(fs_info
, &root_key
);
7870 static noinline
int relocate_one_extent(struct btrfs_root
*extent_root
,
7871 struct btrfs_path
*path
,
7872 struct btrfs_key
*extent_key
,
7873 struct btrfs_block_group_cache
*group
,
7874 struct inode
*reloc_inode
, int pass
)
7876 struct btrfs_trans_handle
*trans
;
7877 struct btrfs_root
*found_root
;
7878 struct btrfs_ref_path
*ref_path
= NULL
;
7879 struct disk_extent
*new_extents
= NULL
;
7884 struct btrfs_key first_key
;
7888 trans
= btrfs_start_transaction(extent_root
, 1);
7889 BUG_ON(IS_ERR(trans
));
7891 if (extent_key
->objectid
== 0) {
7892 ret
= del_extent_zero(trans
, extent_root
, path
, extent_key
);
7896 ref_path
= kmalloc(sizeof(*ref_path
), GFP_NOFS
);
7902 for (loops
= 0; ; loops
++) {
7904 ret
= btrfs_first_ref_path(trans
, extent_root
, ref_path
,
7905 extent_key
->objectid
);
7907 ret
= btrfs_next_ref_path(trans
, extent_root
, ref_path
);
7914 if (ref_path
->root_objectid
== BTRFS_TREE_LOG_OBJECTID
||
7915 ref_path
->root_objectid
== BTRFS_TREE_RELOC_OBJECTID
)
7918 found_root
= read_ref_root(extent_root
->fs_info
, ref_path
);
7919 BUG_ON(!found_root
);
7921 * for reference counted tree, only process reference paths
7922 * rooted at the latest committed root.
7924 if (found_root
->ref_cows
&&
7925 ref_path
->root_generation
!= found_root
->root_key
.offset
)
7928 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
7931 * copy data extents to new locations
7933 u64 group_start
= group
->key
.objectid
;
7934 ret
= relocate_data_extent(reloc_inode
,
7943 level
= ref_path
->owner_objectid
;
7946 if (prev_block
!= ref_path
->nodes
[level
]) {
7947 struct extent_buffer
*eb
;
7948 u64 block_start
= ref_path
->nodes
[level
];
7949 u64 block_size
= btrfs_level_size(found_root
, level
);
7951 eb
= read_tree_block(found_root
, block_start
,
7957 btrfs_tree_lock(eb
);
7958 BUG_ON(level
!= btrfs_header_level(eb
));
7961 btrfs_item_key_to_cpu(eb
, &first_key
, 0);
7963 btrfs_node_key_to_cpu(eb
, &first_key
, 0);
7965 btrfs_tree_unlock(eb
);
7966 free_extent_buffer(eb
);
7967 prev_block
= block_start
;
7970 mutex_lock(&extent_root
->fs_info
->trans_mutex
);
7971 btrfs_record_root_in_trans(found_root
);
7972 mutex_unlock(&extent_root
->fs_info
->trans_mutex
);
7973 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
7975 * try to update data extent references while
7976 * keeping metadata shared between snapshots.
7979 ret
= relocate_one_path(trans
, found_root
,
7980 path
, &first_key
, ref_path
,
7981 group
, reloc_inode
);
7987 * use fallback method to process the remaining
7991 u64 group_start
= group
->key
.objectid
;
7992 new_extents
= kmalloc(sizeof(*new_extents
),
7995 ret
= get_new_locations(reloc_inode
,
8003 ret
= replace_one_extent(trans
, found_root
,
8005 &first_key
, ref_path
,
8006 new_extents
, nr_extents
);
8008 ret
= relocate_tree_block(trans
, found_root
, path
,
8009 &first_key
, ref_path
);
8016 btrfs_end_transaction(trans
, extent_root
);
8023 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
8026 u64 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
8027 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
8030 * we add in the count of missing devices because we want
8031 * to make sure that any RAID levels on a degraded FS
8032 * continue to be honored.
8034 num_devices
= root
->fs_info
->fs_devices
->rw_devices
+
8035 root
->fs_info
->fs_devices
->missing_devices
;
8037 if (num_devices
== 1) {
8038 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
8039 stripped
= flags
& ~stripped
;
8041 /* turn raid0 into single device chunks */
8042 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
8045 /* turn mirroring into duplication */
8046 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
8047 BTRFS_BLOCK_GROUP_RAID10
))
8048 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
8051 /* they already had raid on here, just return */
8052 if (flags
& stripped
)
8055 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
8056 stripped
= flags
& ~stripped
;
8058 /* switch duplicated blocks with raid1 */
8059 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
8060 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
8062 /* turn single device chunks into raid0 */
8063 return stripped
| BTRFS_BLOCK_GROUP_RAID0
;
8068 static int set_block_group_ro(struct btrfs_block_group_cache
*cache
)
8070 struct btrfs_space_info
*sinfo
= cache
->space_info
;
8077 spin_lock(&sinfo
->lock
);
8078 spin_lock(&cache
->lock
);
8079 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
8080 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
8082 if (sinfo
->bytes_used
+ sinfo
->bytes_reserved
+ sinfo
->bytes_pinned
+
8083 sinfo
->bytes_may_use
+ sinfo
->bytes_readonly
+
8084 cache
->reserved_pinned
+ num_bytes
<= sinfo
->total_bytes
) {
8085 sinfo
->bytes_readonly
+= num_bytes
;
8086 sinfo
->bytes_reserved
+= cache
->reserved_pinned
;
8087 cache
->reserved_pinned
= 0;
8092 spin_unlock(&cache
->lock
);
8093 spin_unlock(&sinfo
->lock
);
8097 int btrfs_set_block_group_ro(struct btrfs_root
*root
,
8098 struct btrfs_block_group_cache
*cache
)
8101 struct btrfs_trans_handle
*trans
;
8107 trans
= btrfs_join_transaction(root
, 1);
8108 BUG_ON(IS_ERR(trans
));
8110 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
8111 if (alloc_flags
!= cache
->flags
)
8112 do_chunk_alloc(trans
, root
, 2 * 1024 * 1024, alloc_flags
, 1);
8114 ret
= set_block_group_ro(cache
);
8117 alloc_flags
= get_alloc_profile(root
, cache
->space_info
->flags
);
8118 ret
= do_chunk_alloc(trans
, root
, 2 * 1024 * 1024, alloc_flags
, 1);
8121 ret
= set_block_group_ro(cache
);
8123 btrfs_end_transaction(trans
, root
);
8127 int btrfs_force_chunk_alloc(struct btrfs_trans_handle
*trans
,
8128 struct btrfs_root
*root
, u64 type
)
8130 u64 alloc_flags
= get_alloc_profile(root
, type
);
8131 return do_chunk_alloc(trans
, root
, 2 * 1024 * 1024, alloc_flags
, 1);
8135 * helper to account the unused space of all the readonly block group in the
8136 * list. takes mirrors into account.
8138 static u64
__btrfs_get_ro_block_group_free_space(struct list_head
*groups_list
)
8140 struct btrfs_block_group_cache
*block_group
;
8144 list_for_each_entry(block_group
, groups_list
, list
) {
8145 spin_lock(&block_group
->lock
);
8147 if (!block_group
->ro
) {
8148 spin_unlock(&block_group
->lock
);
8152 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_RAID1
|
8153 BTRFS_BLOCK_GROUP_RAID10
|
8154 BTRFS_BLOCK_GROUP_DUP
))
8159 free_bytes
+= (block_group
->key
.offset
-
8160 btrfs_block_group_used(&block_group
->item
)) *
8163 spin_unlock(&block_group
->lock
);
8170 * helper to account the unused space of all the readonly block group in the
8171 * space_info. takes mirrors into account.
8173 u64
btrfs_account_ro_block_groups_free_space(struct btrfs_space_info
*sinfo
)
8178 spin_lock(&sinfo
->lock
);
8180 for(i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
8181 if (!list_empty(&sinfo
->block_groups
[i
]))
8182 free_bytes
+= __btrfs_get_ro_block_group_free_space(
8183 &sinfo
->block_groups
[i
]);
8185 spin_unlock(&sinfo
->lock
);
8190 int btrfs_set_block_group_rw(struct btrfs_root
*root
,
8191 struct btrfs_block_group_cache
*cache
)
8193 struct btrfs_space_info
*sinfo
= cache
->space_info
;
8198 spin_lock(&sinfo
->lock
);
8199 spin_lock(&cache
->lock
);
8200 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
8201 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
8202 sinfo
->bytes_readonly
-= num_bytes
;
8204 spin_unlock(&cache
->lock
);
8205 spin_unlock(&sinfo
->lock
);
8210 * checks to see if its even possible to relocate this block group.
8212 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8213 * ok to go ahead and try.
8215 int btrfs_can_relocate(struct btrfs_root
*root
, u64 bytenr
)
8217 struct btrfs_block_group_cache
*block_group
;
8218 struct btrfs_space_info
*space_info
;
8219 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
8220 struct btrfs_device
*device
;
8224 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
8226 /* odd, couldn't find the block group, leave it alone */
8230 /* no bytes used, we're good */
8231 if (!btrfs_block_group_used(&block_group
->item
))
8234 space_info
= block_group
->space_info
;
8235 spin_lock(&space_info
->lock
);
8237 full
= space_info
->full
;
8240 * if this is the last block group we have in this space, we can't
8241 * relocate it unless we're able to allocate a new chunk below.
8243 * Otherwise, we need to make sure we have room in the space to handle
8244 * all of the extents from this block group. If we can, we're good
8246 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
8247 (space_info
->bytes_used
+ space_info
->bytes_reserved
+
8248 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
8249 btrfs_block_group_used(&block_group
->item
) <
8250 space_info
->total_bytes
)) {
8251 spin_unlock(&space_info
->lock
);
8254 spin_unlock(&space_info
->lock
);
8257 * ok we don't have enough space, but maybe we have free space on our
8258 * devices to allocate new chunks for relocation, so loop through our
8259 * alloc devices and guess if we have enough space. However, if we
8260 * were marked as full, then we know there aren't enough chunks, and we
8267 mutex_lock(&root
->fs_info
->chunk_mutex
);
8268 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
8269 u64 min_free
= btrfs_block_group_used(&block_group
->item
);
8273 * check to make sure we can actually find a chunk with enough
8274 * space to fit our block group in.
8276 if (device
->total_bytes
> device
->bytes_used
+ min_free
) {
8277 ret
= find_free_dev_extent(NULL
, device
, min_free
,
8284 mutex_unlock(&root
->fs_info
->chunk_mutex
);
8286 btrfs_put_block_group(block_group
);
8290 static int find_first_block_group(struct btrfs_root
*root
,
8291 struct btrfs_path
*path
, struct btrfs_key
*key
)
8294 struct btrfs_key found_key
;
8295 struct extent_buffer
*leaf
;
8298 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
8303 slot
= path
->slots
[0];
8304 leaf
= path
->nodes
[0];
8305 if (slot
>= btrfs_header_nritems(leaf
)) {
8306 ret
= btrfs_next_leaf(root
, path
);
8313 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
8315 if (found_key
.objectid
>= key
->objectid
&&
8316 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
8326 void btrfs_put_block_group_cache(struct btrfs_fs_info
*info
)
8328 struct btrfs_block_group_cache
*block_group
;
8332 struct inode
*inode
;
8334 block_group
= btrfs_lookup_first_block_group(info
, last
);
8335 while (block_group
) {
8336 spin_lock(&block_group
->lock
);
8337 if (block_group
->iref
)
8339 spin_unlock(&block_group
->lock
);
8340 block_group
= next_block_group(info
->tree_root
,
8350 inode
= block_group
->inode
;
8351 block_group
->iref
= 0;
8352 block_group
->inode
= NULL
;
8353 spin_unlock(&block_group
->lock
);
8355 last
= block_group
->key
.objectid
+ block_group
->key
.offset
;
8356 btrfs_put_block_group(block_group
);
8360 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
8362 struct btrfs_block_group_cache
*block_group
;
8363 struct btrfs_space_info
*space_info
;
8364 struct btrfs_caching_control
*caching_ctl
;
8367 down_write(&info
->extent_commit_sem
);
8368 while (!list_empty(&info
->caching_block_groups
)) {
8369 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
8370 struct btrfs_caching_control
, list
);
8371 list_del(&caching_ctl
->list
);
8372 put_caching_control(caching_ctl
);
8374 up_write(&info
->extent_commit_sem
);
8376 spin_lock(&info
->block_group_cache_lock
);
8377 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
8378 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
8380 rb_erase(&block_group
->cache_node
,
8381 &info
->block_group_cache_tree
);
8382 spin_unlock(&info
->block_group_cache_lock
);
8384 down_write(&block_group
->space_info
->groups_sem
);
8385 list_del(&block_group
->list
);
8386 up_write(&block_group
->space_info
->groups_sem
);
8388 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
8389 wait_block_group_cache_done(block_group
);
8392 * We haven't cached this block group, which means we could
8393 * possibly have excluded extents on this block group.
8395 if (block_group
->cached
== BTRFS_CACHE_NO
)
8396 free_excluded_extents(info
->extent_root
, block_group
);
8398 btrfs_remove_free_space_cache(block_group
);
8399 btrfs_put_block_group(block_group
);
8401 spin_lock(&info
->block_group_cache_lock
);
8403 spin_unlock(&info
->block_group_cache_lock
);
8405 /* now that all the block groups are freed, go through and
8406 * free all the space_info structs. This is only called during
8407 * the final stages of unmount, and so we know nobody is
8408 * using them. We call synchronize_rcu() once before we start,
8409 * just to be on the safe side.
8413 release_global_block_rsv(info
);
8415 while(!list_empty(&info
->space_info
)) {
8416 space_info
= list_entry(info
->space_info
.next
,
8417 struct btrfs_space_info
,
8419 if (space_info
->bytes_pinned
> 0 ||
8420 space_info
->bytes_reserved
> 0) {
8422 dump_space_info(space_info
, 0, 0);
8424 list_del(&space_info
->list
);
8430 static void __link_block_group(struct btrfs_space_info
*space_info
,
8431 struct btrfs_block_group_cache
*cache
)
8433 int index
= get_block_group_index(cache
);
8435 down_write(&space_info
->groups_sem
);
8436 list_add_tail(&cache
->list
, &space_info
->block_groups
[index
]);
8437 up_write(&space_info
->groups_sem
);
8440 int btrfs_read_block_groups(struct btrfs_root
*root
)
8442 struct btrfs_path
*path
;
8444 struct btrfs_block_group_cache
*cache
;
8445 struct btrfs_fs_info
*info
= root
->fs_info
;
8446 struct btrfs_space_info
*space_info
;
8447 struct btrfs_key key
;
8448 struct btrfs_key found_key
;
8449 struct extent_buffer
*leaf
;
8453 root
= info
->extent_root
;
8456 btrfs_set_key_type(&key
, BTRFS_BLOCK_GROUP_ITEM_KEY
);
8457 path
= btrfs_alloc_path();
8461 cache_gen
= btrfs_super_cache_generation(&root
->fs_info
->super_copy
);
8462 if (cache_gen
!= 0 &&
8463 btrfs_super_generation(&root
->fs_info
->super_copy
) != cache_gen
)
8465 if (btrfs_test_opt(root
, CLEAR_CACHE
))
8467 if (!btrfs_test_opt(root
, SPACE_CACHE
) && cache_gen
)
8468 printk(KERN_INFO
"btrfs: disk space caching is enabled\n");
8471 ret
= find_first_block_group(root
, path
, &key
);
8476 leaf
= path
->nodes
[0];
8477 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
8478 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
8484 atomic_set(&cache
->count
, 1);
8485 spin_lock_init(&cache
->lock
);
8486 spin_lock_init(&cache
->tree_lock
);
8487 cache
->fs_info
= info
;
8488 INIT_LIST_HEAD(&cache
->list
);
8489 INIT_LIST_HEAD(&cache
->cluster_list
);
8492 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
8495 * we only want to have 32k of ram per block group for keeping
8496 * track of free space, and if we pass 1/2 of that we want to
8497 * start converting things over to using bitmaps
8499 cache
->extents_thresh
= ((1024 * 32) / 2) /
8500 sizeof(struct btrfs_free_space
);
8502 read_extent_buffer(leaf
, &cache
->item
,
8503 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
8504 sizeof(cache
->item
));
8505 memcpy(&cache
->key
, &found_key
, sizeof(found_key
));
8507 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
8508 btrfs_release_path(root
, path
);
8509 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
8510 cache
->sectorsize
= root
->sectorsize
;
8513 * We need to exclude the super stripes now so that the space
8514 * info has super bytes accounted for, otherwise we'll think
8515 * we have more space than we actually do.
8517 exclude_super_stripes(root
, cache
);
8520 * check for two cases, either we are full, and therefore
8521 * don't need to bother with the caching work since we won't
8522 * find any space, or we are empty, and we can just add all
8523 * the space in and be done with it. This saves us _alot_ of
8524 * time, particularly in the full case.
8526 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
8527 cache
->last_byte_to_unpin
= (u64
)-1;
8528 cache
->cached
= BTRFS_CACHE_FINISHED
;
8529 free_excluded_extents(root
, cache
);
8530 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
8531 cache
->last_byte_to_unpin
= (u64
)-1;
8532 cache
->cached
= BTRFS_CACHE_FINISHED
;
8533 add_new_free_space(cache
, root
->fs_info
,
8535 found_key
.objectid
+
8537 free_excluded_extents(root
, cache
);
8540 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
8541 btrfs_block_group_used(&cache
->item
),
8544 cache
->space_info
= space_info
;
8545 spin_lock(&cache
->space_info
->lock
);
8546 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
8547 spin_unlock(&cache
->space_info
->lock
);
8549 __link_block_group(space_info
, cache
);
8551 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
8554 set_avail_alloc_bits(root
->fs_info
, cache
->flags
);
8555 if (btrfs_chunk_readonly(root
, cache
->key
.objectid
))
8556 set_block_group_ro(cache
);
8559 list_for_each_entry_rcu(space_info
, &root
->fs_info
->space_info
, list
) {
8560 if (!(get_alloc_profile(root
, space_info
->flags
) &
8561 (BTRFS_BLOCK_GROUP_RAID10
|
8562 BTRFS_BLOCK_GROUP_RAID1
|
8563 BTRFS_BLOCK_GROUP_DUP
)))
8566 * avoid allocating from un-mirrored block group if there are
8567 * mirrored block groups.
8569 list_for_each_entry(cache
, &space_info
->block_groups
[3], list
)
8570 set_block_group_ro(cache
);
8571 list_for_each_entry(cache
, &space_info
->block_groups
[4], list
)
8572 set_block_group_ro(cache
);
8575 init_global_block_rsv(info
);
8578 btrfs_free_path(path
);
8582 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
8583 struct btrfs_root
*root
, u64 bytes_used
,
8584 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
8588 struct btrfs_root
*extent_root
;
8589 struct btrfs_block_group_cache
*cache
;
8591 extent_root
= root
->fs_info
->extent_root
;
8593 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
8595 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
8599 cache
->key
.objectid
= chunk_offset
;
8600 cache
->key
.offset
= size
;
8601 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
8602 cache
->sectorsize
= root
->sectorsize
;
8603 cache
->fs_info
= root
->fs_info
;
8606 * we only want to have 32k of ram per block group for keeping track
8607 * of free space, and if we pass 1/2 of that we want to start
8608 * converting things over to using bitmaps
8610 cache
->extents_thresh
= ((1024 * 32) / 2) /
8611 sizeof(struct btrfs_free_space
);
8612 atomic_set(&cache
->count
, 1);
8613 spin_lock_init(&cache
->lock
);
8614 spin_lock_init(&cache
->tree_lock
);
8615 INIT_LIST_HEAD(&cache
->list
);
8616 INIT_LIST_HEAD(&cache
->cluster_list
);
8618 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
8619 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
8620 cache
->flags
= type
;
8621 btrfs_set_block_group_flags(&cache
->item
, type
);
8623 cache
->last_byte_to_unpin
= (u64
)-1;
8624 cache
->cached
= BTRFS_CACHE_FINISHED
;
8625 exclude_super_stripes(root
, cache
);
8627 add_new_free_space(cache
, root
->fs_info
, chunk_offset
,
8628 chunk_offset
+ size
);
8630 free_excluded_extents(root
, cache
);
8632 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
8633 &cache
->space_info
);
8636 spin_lock(&cache
->space_info
->lock
);
8637 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
8638 spin_unlock(&cache
->space_info
->lock
);
8640 __link_block_group(cache
->space_info
, cache
);
8642 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
8645 ret
= btrfs_insert_item(trans
, extent_root
, &cache
->key
, &cache
->item
,
8646 sizeof(cache
->item
));
8649 set_avail_alloc_bits(extent_root
->fs_info
, type
);
8654 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
8655 struct btrfs_root
*root
, u64 group_start
)
8657 struct btrfs_path
*path
;
8658 struct btrfs_block_group_cache
*block_group
;
8659 struct btrfs_free_cluster
*cluster
;
8660 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
8661 struct btrfs_key key
;
8662 struct inode
*inode
;
8666 root
= root
->fs_info
->extent_root
;
8668 block_group
= btrfs_lookup_block_group(root
->fs_info
, group_start
);
8669 BUG_ON(!block_group
);
8670 BUG_ON(!block_group
->ro
);
8673 * Free the reserved super bytes from this block group before
8676 free_excluded_extents(root
, block_group
);
8678 memcpy(&key
, &block_group
->key
, sizeof(key
));
8679 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
8680 BTRFS_BLOCK_GROUP_RAID1
|
8681 BTRFS_BLOCK_GROUP_RAID10
))
8686 /* make sure this block group isn't part of an allocation cluster */
8687 cluster
= &root
->fs_info
->data_alloc_cluster
;
8688 spin_lock(&cluster
->refill_lock
);
8689 btrfs_return_cluster_to_free_space(block_group
, cluster
);
8690 spin_unlock(&cluster
->refill_lock
);
8693 * make sure this block group isn't part of a metadata
8694 * allocation cluster
8696 cluster
= &root
->fs_info
->meta_alloc_cluster
;
8697 spin_lock(&cluster
->refill_lock
);
8698 btrfs_return_cluster_to_free_space(block_group
, cluster
);
8699 spin_unlock(&cluster
->refill_lock
);
8701 path
= btrfs_alloc_path();
8704 inode
= lookup_free_space_inode(root
, block_group
, path
);
8705 if (!IS_ERR(inode
)) {
8706 btrfs_orphan_add(trans
, inode
);
8708 /* One for the block groups ref */
8709 spin_lock(&block_group
->lock
);
8710 if (block_group
->iref
) {
8711 block_group
->iref
= 0;
8712 block_group
->inode
= NULL
;
8713 spin_unlock(&block_group
->lock
);
8716 spin_unlock(&block_group
->lock
);
8718 /* One for our lookup ref */
8722 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
8723 key
.offset
= block_group
->key
.objectid
;
8726 ret
= btrfs_search_slot(trans
, tree_root
, &key
, path
, -1, 1);
8730 btrfs_release_path(tree_root
, path
);
8732 ret
= btrfs_del_item(trans
, tree_root
, path
);
8735 btrfs_release_path(tree_root
, path
);
8738 spin_lock(&root
->fs_info
->block_group_cache_lock
);
8739 rb_erase(&block_group
->cache_node
,
8740 &root
->fs_info
->block_group_cache_tree
);
8741 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
8743 down_write(&block_group
->space_info
->groups_sem
);
8745 * we must use list_del_init so people can check to see if they
8746 * are still on the list after taking the semaphore
8748 list_del_init(&block_group
->list
);
8749 up_write(&block_group
->space_info
->groups_sem
);
8751 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
8752 wait_block_group_cache_done(block_group
);
8754 btrfs_remove_free_space_cache(block_group
);
8756 spin_lock(&block_group
->space_info
->lock
);
8757 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
8758 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
8759 block_group
->space_info
->disk_total
-= block_group
->key
.offset
* factor
;
8760 spin_unlock(&block_group
->space_info
->lock
);
8762 memcpy(&key
, &block_group
->key
, sizeof(key
));
8764 btrfs_clear_space_info_full(root
->fs_info
);
8766 btrfs_put_block_group(block_group
);
8767 btrfs_put_block_group(block_group
);
8769 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
8775 ret
= btrfs_del_item(trans
, root
, path
);
8777 btrfs_free_path(path
);
8781 int btrfs_init_space_info(struct btrfs_fs_info
*fs_info
)
8783 struct btrfs_space_info
*space_info
;
8786 ret
= update_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
, 0, 0,
8791 ret
= update_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
, 0, 0,
8796 ret
= update_space_info(fs_info
, BTRFS_BLOCK_GROUP_DATA
, 0, 0,
8804 int btrfs_error_unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
8806 return unpin_extent_range(root
, start
, end
);
8809 int btrfs_error_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
8810 u64 num_bytes
, u64
*actual_bytes
)
8812 return btrfs_discard_extent(root
, bytenr
, num_bytes
, actual_bytes
);
8815 int btrfs_trim_fs(struct btrfs_root
*root
, struct fstrim_range
*range
)
8817 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8818 struct btrfs_block_group_cache
*cache
= NULL
;
8825 cache
= btrfs_lookup_block_group(fs_info
, range
->start
);
8828 if (cache
->key
.objectid
>= (range
->start
+ range
->len
)) {
8829 btrfs_put_block_group(cache
);
8833 start
= max(range
->start
, cache
->key
.objectid
);
8834 end
= min(range
->start
+ range
->len
,
8835 cache
->key
.objectid
+ cache
->key
.offset
);
8837 if (end
- start
>= range
->minlen
) {
8838 if (!block_group_cache_done(cache
)) {
8839 ret
= cache_block_group(cache
, NULL
, root
, 0);
8841 wait_block_group_cache_done(cache
);
8843 ret
= btrfs_trim_block_group(cache
,
8849 trimmed
+= group_trimmed
;
8851 btrfs_put_block_group(cache
);
8856 cache
= next_block_group(fs_info
->tree_root
, cache
);
8859 range
->len
= trimmed
;