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
->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 void btrfs_issue_discard(struct block_device
*bdev
,
1744 blkdev_issue_discard(bdev
, start
>> 9, len
>> 9, GFP_NOFS
, 0);
1747 static int btrfs_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
1751 u64 map_length
= num_bytes
;
1752 struct btrfs_multi_bio
*multi
= NULL
;
1754 if (!btrfs_test_opt(root
, DISCARD
))
1757 /* Tell the block device(s) that the sectors can be discarded */
1758 ret
= btrfs_map_block(&root
->fs_info
->mapping_tree
, READ
,
1759 bytenr
, &map_length
, &multi
, 0);
1761 struct btrfs_bio_stripe
*stripe
= multi
->stripes
;
1764 if (map_length
> num_bytes
)
1765 map_length
= num_bytes
;
1767 for (i
= 0; i
< multi
->num_stripes
; i
++, stripe
++) {
1768 btrfs_issue_discard(stripe
->dev
->bdev
,
1778 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1779 struct btrfs_root
*root
,
1780 u64 bytenr
, u64 num_bytes
, u64 parent
,
1781 u64 root_objectid
, u64 owner
, u64 offset
)
1784 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
&&
1785 root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
1787 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1788 ret
= btrfs_add_delayed_tree_ref(trans
, bytenr
, num_bytes
,
1789 parent
, root_objectid
, (int)owner
,
1790 BTRFS_ADD_DELAYED_REF
, NULL
);
1792 ret
= btrfs_add_delayed_data_ref(trans
, bytenr
, num_bytes
,
1793 parent
, root_objectid
, owner
, offset
,
1794 BTRFS_ADD_DELAYED_REF
, NULL
);
1799 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1800 struct btrfs_root
*root
,
1801 u64 bytenr
, u64 num_bytes
,
1802 u64 parent
, u64 root_objectid
,
1803 u64 owner
, u64 offset
, int refs_to_add
,
1804 struct btrfs_delayed_extent_op
*extent_op
)
1806 struct btrfs_path
*path
;
1807 struct extent_buffer
*leaf
;
1808 struct btrfs_extent_item
*item
;
1813 path
= btrfs_alloc_path();
1818 path
->leave_spinning
= 1;
1819 /* this will setup the path even if it fails to insert the back ref */
1820 ret
= insert_inline_extent_backref(trans
, root
->fs_info
->extent_root
,
1821 path
, bytenr
, num_bytes
, parent
,
1822 root_objectid
, owner
, offset
,
1823 refs_to_add
, extent_op
);
1827 if (ret
!= -EAGAIN
) {
1832 leaf
= path
->nodes
[0];
1833 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1834 refs
= btrfs_extent_refs(leaf
, item
);
1835 btrfs_set_extent_refs(leaf
, item
, refs
+ refs_to_add
);
1837 __run_delayed_extent_op(extent_op
, leaf
, item
);
1839 btrfs_mark_buffer_dirty(leaf
);
1840 btrfs_release_path(root
->fs_info
->extent_root
, path
);
1843 path
->leave_spinning
= 1;
1845 /* now insert the actual backref */
1846 ret
= insert_extent_backref(trans
, root
->fs_info
->extent_root
,
1847 path
, bytenr
, parent
, root_objectid
,
1848 owner
, offset
, refs_to_add
);
1851 btrfs_free_path(path
);
1855 static int run_delayed_data_ref(struct btrfs_trans_handle
*trans
,
1856 struct btrfs_root
*root
,
1857 struct btrfs_delayed_ref_node
*node
,
1858 struct btrfs_delayed_extent_op
*extent_op
,
1859 int insert_reserved
)
1862 struct btrfs_delayed_data_ref
*ref
;
1863 struct btrfs_key ins
;
1868 ins
.objectid
= node
->bytenr
;
1869 ins
.offset
= node
->num_bytes
;
1870 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1872 ref
= btrfs_delayed_node_to_data_ref(node
);
1873 if (node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
1874 parent
= ref
->parent
;
1876 ref_root
= ref
->root
;
1878 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
1880 BUG_ON(extent_op
->update_key
);
1881 flags
|= extent_op
->flags_to_set
;
1883 ret
= alloc_reserved_file_extent(trans
, root
,
1884 parent
, ref_root
, flags
,
1885 ref
->objectid
, ref
->offset
,
1886 &ins
, node
->ref_mod
);
1887 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
1888 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
1889 node
->num_bytes
, parent
,
1890 ref_root
, ref
->objectid
,
1891 ref
->offset
, node
->ref_mod
,
1893 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
1894 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
1895 node
->num_bytes
, parent
,
1896 ref_root
, ref
->objectid
,
1897 ref
->offset
, node
->ref_mod
,
1905 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
1906 struct extent_buffer
*leaf
,
1907 struct btrfs_extent_item
*ei
)
1909 u64 flags
= btrfs_extent_flags(leaf
, ei
);
1910 if (extent_op
->update_flags
) {
1911 flags
|= extent_op
->flags_to_set
;
1912 btrfs_set_extent_flags(leaf
, ei
, flags
);
1915 if (extent_op
->update_key
) {
1916 struct btrfs_tree_block_info
*bi
;
1917 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
));
1918 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
1919 btrfs_set_tree_block_key(leaf
, bi
, &extent_op
->key
);
1923 static int run_delayed_extent_op(struct btrfs_trans_handle
*trans
,
1924 struct btrfs_root
*root
,
1925 struct btrfs_delayed_ref_node
*node
,
1926 struct btrfs_delayed_extent_op
*extent_op
)
1928 struct btrfs_key key
;
1929 struct btrfs_path
*path
;
1930 struct btrfs_extent_item
*ei
;
1931 struct extent_buffer
*leaf
;
1936 path
= btrfs_alloc_path();
1940 key
.objectid
= node
->bytenr
;
1941 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1942 key
.offset
= node
->num_bytes
;
1945 path
->leave_spinning
= 1;
1946 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
,
1957 leaf
= path
->nodes
[0];
1958 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1959 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1960 if (item_size
< sizeof(*ei
)) {
1961 ret
= convert_extent_item_v0(trans
, root
->fs_info
->extent_root
,
1967 leaf
= path
->nodes
[0];
1968 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1971 BUG_ON(item_size
< sizeof(*ei
));
1972 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1973 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1975 btrfs_mark_buffer_dirty(leaf
);
1977 btrfs_free_path(path
);
1981 static int run_delayed_tree_ref(struct btrfs_trans_handle
*trans
,
1982 struct btrfs_root
*root
,
1983 struct btrfs_delayed_ref_node
*node
,
1984 struct btrfs_delayed_extent_op
*extent_op
,
1985 int insert_reserved
)
1988 struct btrfs_delayed_tree_ref
*ref
;
1989 struct btrfs_key ins
;
1993 ins
.objectid
= node
->bytenr
;
1994 ins
.offset
= node
->num_bytes
;
1995 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1997 ref
= btrfs_delayed_node_to_tree_ref(node
);
1998 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
1999 parent
= ref
->parent
;
2001 ref_root
= ref
->root
;
2003 BUG_ON(node
->ref_mod
!= 1);
2004 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2005 BUG_ON(!extent_op
|| !extent_op
->update_flags
||
2006 !extent_op
->update_key
);
2007 ret
= alloc_reserved_tree_block(trans
, root
,
2009 extent_op
->flags_to_set
,
2012 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2013 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
2014 node
->num_bytes
, parent
, ref_root
,
2015 ref
->level
, 0, 1, extent_op
);
2016 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2017 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
2018 node
->num_bytes
, parent
, ref_root
,
2019 ref
->level
, 0, 1, extent_op
);
2026 /* helper function to actually process a single delayed ref entry */
2027 static int run_one_delayed_ref(struct btrfs_trans_handle
*trans
,
2028 struct btrfs_root
*root
,
2029 struct btrfs_delayed_ref_node
*node
,
2030 struct btrfs_delayed_extent_op
*extent_op
,
2031 int insert_reserved
)
2034 if (btrfs_delayed_ref_is_head(node
)) {
2035 struct btrfs_delayed_ref_head
*head
;
2037 * we've hit the end of the chain and we were supposed
2038 * to insert this extent into the tree. But, it got
2039 * deleted before we ever needed to insert it, so all
2040 * we have to do is clean up the accounting
2043 head
= btrfs_delayed_node_to_head(node
);
2044 if (insert_reserved
) {
2045 btrfs_pin_extent(root
, node
->bytenr
,
2046 node
->num_bytes
, 1);
2047 if (head
->is_data
) {
2048 ret
= btrfs_del_csums(trans
, root
,
2054 mutex_unlock(&head
->mutex
);
2058 if (node
->type
== BTRFS_TREE_BLOCK_REF_KEY
||
2059 node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2060 ret
= run_delayed_tree_ref(trans
, root
, node
, extent_op
,
2062 else if (node
->type
== BTRFS_EXTENT_DATA_REF_KEY
||
2063 node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2064 ret
= run_delayed_data_ref(trans
, root
, node
, extent_op
,
2071 static noinline
struct btrfs_delayed_ref_node
*
2072 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
2074 struct rb_node
*node
;
2075 struct btrfs_delayed_ref_node
*ref
;
2076 int action
= BTRFS_ADD_DELAYED_REF
;
2079 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2080 * this prevents ref count from going down to zero when
2081 * there still are pending delayed ref.
2083 node
= rb_prev(&head
->node
.rb_node
);
2087 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
2089 if (ref
->bytenr
!= head
->node
.bytenr
)
2091 if (ref
->action
== action
)
2093 node
= rb_prev(node
);
2095 if (action
== BTRFS_ADD_DELAYED_REF
) {
2096 action
= BTRFS_DROP_DELAYED_REF
;
2102 static noinline
int run_clustered_refs(struct btrfs_trans_handle
*trans
,
2103 struct btrfs_root
*root
,
2104 struct list_head
*cluster
)
2106 struct btrfs_delayed_ref_root
*delayed_refs
;
2107 struct btrfs_delayed_ref_node
*ref
;
2108 struct btrfs_delayed_ref_head
*locked_ref
= NULL
;
2109 struct btrfs_delayed_extent_op
*extent_op
;
2112 int must_insert_reserved
= 0;
2114 delayed_refs
= &trans
->transaction
->delayed_refs
;
2117 /* pick a new head ref from the cluster list */
2118 if (list_empty(cluster
))
2121 locked_ref
= list_entry(cluster
->next
,
2122 struct btrfs_delayed_ref_head
, cluster
);
2124 /* grab the lock that says we are going to process
2125 * all the refs for this head */
2126 ret
= btrfs_delayed_ref_lock(trans
, locked_ref
);
2129 * we may have dropped the spin lock to get the head
2130 * mutex lock, and that might have given someone else
2131 * time to free the head. If that's true, it has been
2132 * removed from our list and we can move on.
2134 if (ret
== -EAGAIN
) {
2142 * record the must insert reserved flag before we
2143 * drop the spin lock.
2145 must_insert_reserved
= locked_ref
->must_insert_reserved
;
2146 locked_ref
->must_insert_reserved
= 0;
2148 extent_op
= locked_ref
->extent_op
;
2149 locked_ref
->extent_op
= NULL
;
2152 * locked_ref is the head node, so we have to go one
2153 * node back for any delayed ref updates
2155 ref
= select_delayed_ref(locked_ref
);
2157 /* All delayed refs have been processed, Go ahead
2158 * and send the head node to run_one_delayed_ref,
2159 * so that any accounting fixes can happen
2161 ref
= &locked_ref
->node
;
2163 if (extent_op
&& must_insert_reserved
) {
2169 spin_unlock(&delayed_refs
->lock
);
2171 ret
= run_delayed_extent_op(trans
, root
,
2177 spin_lock(&delayed_refs
->lock
);
2181 list_del_init(&locked_ref
->cluster
);
2186 rb_erase(&ref
->rb_node
, &delayed_refs
->root
);
2187 delayed_refs
->num_entries
--;
2189 spin_unlock(&delayed_refs
->lock
);
2191 ret
= run_one_delayed_ref(trans
, root
, ref
, extent_op
,
2192 must_insert_reserved
);
2195 btrfs_put_delayed_ref(ref
);
2200 spin_lock(&delayed_refs
->lock
);
2206 * this starts processing the delayed reference count updates and
2207 * extent insertions we have queued up so far. count can be
2208 * 0, which means to process everything in the tree at the start
2209 * of the run (but not newly added entries), or it can be some target
2210 * number you'd like to process.
2212 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2213 struct btrfs_root
*root
, unsigned long count
)
2215 struct rb_node
*node
;
2216 struct btrfs_delayed_ref_root
*delayed_refs
;
2217 struct btrfs_delayed_ref_node
*ref
;
2218 struct list_head cluster
;
2220 int run_all
= count
== (unsigned long)-1;
2223 if (root
== root
->fs_info
->extent_root
)
2224 root
= root
->fs_info
->tree_root
;
2226 delayed_refs
= &trans
->transaction
->delayed_refs
;
2227 INIT_LIST_HEAD(&cluster
);
2229 spin_lock(&delayed_refs
->lock
);
2231 count
= delayed_refs
->num_entries
* 2;
2235 if (!(run_all
|| run_most
) &&
2236 delayed_refs
->num_heads_ready
< 64)
2240 * go find something we can process in the rbtree. We start at
2241 * the beginning of the tree, and then build a cluster
2242 * of refs to process starting at the first one we are able to
2245 ret
= btrfs_find_ref_cluster(trans
, &cluster
,
2246 delayed_refs
->run_delayed_start
);
2250 ret
= run_clustered_refs(trans
, root
, &cluster
);
2253 count
-= min_t(unsigned long, ret
, count
);
2260 node
= rb_first(&delayed_refs
->root
);
2263 count
= (unsigned long)-1;
2266 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
2268 if (btrfs_delayed_ref_is_head(ref
)) {
2269 struct btrfs_delayed_ref_head
*head
;
2271 head
= btrfs_delayed_node_to_head(ref
);
2272 atomic_inc(&ref
->refs
);
2274 spin_unlock(&delayed_refs
->lock
);
2275 mutex_lock(&head
->mutex
);
2276 mutex_unlock(&head
->mutex
);
2278 btrfs_put_delayed_ref(ref
);
2282 node
= rb_next(node
);
2284 spin_unlock(&delayed_refs
->lock
);
2285 schedule_timeout(1);
2289 spin_unlock(&delayed_refs
->lock
);
2293 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
2294 struct btrfs_root
*root
,
2295 u64 bytenr
, u64 num_bytes
, u64 flags
,
2298 struct btrfs_delayed_extent_op
*extent_op
;
2301 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
2305 extent_op
->flags_to_set
= flags
;
2306 extent_op
->update_flags
= 1;
2307 extent_op
->update_key
= 0;
2308 extent_op
->is_data
= is_data
? 1 : 0;
2310 ret
= btrfs_add_delayed_extent_op(trans
, bytenr
, num_bytes
, extent_op
);
2316 static noinline
int check_delayed_ref(struct btrfs_trans_handle
*trans
,
2317 struct btrfs_root
*root
,
2318 struct btrfs_path
*path
,
2319 u64 objectid
, u64 offset
, u64 bytenr
)
2321 struct btrfs_delayed_ref_head
*head
;
2322 struct btrfs_delayed_ref_node
*ref
;
2323 struct btrfs_delayed_data_ref
*data_ref
;
2324 struct btrfs_delayed_ref_root
*delayed_refs
;
2325 struct rb_node
*node
;
2329 delayed_refs
= &trans
->transaction
->delayed_refs
;
2330 spin_lock(&delayed_refs
->lock
);
2331 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
2335 if (!mutex_trylock(&head
->mutex
)) {
2336 atomic_inc(&head
->node
.refs
);
2337 spin_unlock(&delayed_refs
->lock
);
2339 btrfs_release_path(root
->fs_info
->extent_root
, path
);
2341 mutex_lock(&head
->mutex
);
2342 mutex_unlock(&head
->mutex
);
2343 btrfs_put_delayed_ref(&head
->node
);
2347 node
= rb_prev(&head
->node
.rb_node
);
2351 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2353 if (ref
->bytenr
!= bytenr
)
2357 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
)
2360 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
2362 node
= rb_prev(node
);
2364 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2365 if (ref
->bytenr
== bytenr
)
2369 if (data_ref
->root
!= root
->root_key
.objectid
||
2370 data_ref
->objectid
!= objectid
|| data_ref
->offset
!= offset
)
2375 mutex_unlock(&head
->mutex
);
2377 spin_unlock(&delayed_refs
->lock
);
2381 static noinline
int check_committed_ref(struct btrfs_trans_handle
*trans
,
2382 struct btrfs_root
*root
,
2383 struct btrfs_path
*path
,
2384 u64 objectid
, u64 offset
, u64 bytenr
)
2386 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2387 struct extent_buffer
*leaf
;
2388 struct btrfs_extent_data_ref
*ref
;
2389 struct btrfs_extent_inline_ref
*iref
;
2390 struct btrfs_extent_item
*ei
;
2391 struct btrfs_key key
;
2395 key
.objectid
= bytenr
;
2396 key
.offset
= (u64
)-1;
2397 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2399 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
2405 if (path
->slots
[0] == 0)
2409 leaf
= path
->nodes
[0];
2410 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2412 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
2416 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2417 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2418 if (item_size
< sizeof(*ei
)) {
2419 WARN_ON(item_size
!= sizeof(struct btrfs_extent_item_v0
));
2423 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2425 if (item_size
!= sizeof(*ei
) +
2426 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
2429 if (btrfs_extent_generation(leaf
, ei
) <=
2430 btrfs_root_last_snapshot(&root
->root_item
))
2433 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
2434 if (btrfs_extent_inline_ref_type(leaf
, iref
) !=
2435 BTRFS_EXTENT_DATA_REF_KEY
)
2438 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
2439 if (btrfs_extent_refs(leaf
, ei
) !=
2440 btrfs_extent_data_ref_count(leaf
, ref
) ||
2441 btrfs_extent_data_ref_root(leaf
, ref
) !=
2442 root
->root_key
.objectid
||
2443 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
2444 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
2452 int btrfs_cross_ref_exist(struct btrfs_trans_handle
*trans
,
2453 struct btrfs_root
*root
,
2454 u64 objectid
, u64 offset
, u64 bytenr
)
2456 struct btrfs_path
*path
;
2460 path
= btrfs_alloc_path();
2465 ret
= check_committed_ref(trans
, root
, path
, objectid
,
2467 if (ret
&& ret
!= -ENOENT
)
2470 ret2
= check_delayed_ref(trans
, root
, path
, objectid
,
2472 } while (ret2
== -EAGAIN
);
2474 if (ret2
&& ret2
!= -ENOENT
) {
2479 if (ret
!= -ENOENT
|| ret2
!= -ENOENT
)
2482 btrfs_free_path(path
);
2483 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
2489 int btrfs_cache_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2490 struct extent_buffer
*buf
, u32 nr_extents
)
2492 struct btrfs_key key
;
2493 struct btrfs_file_extent_item
*fi
;
2501 if (!root
->ref_cows
)
2504 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
2506 root_gen
= root
->root_key
.offset
;
2509 root_gen
= trans
->transid
- 1;
2512 level
= btrfs_header_level(buf
);
2513 nritems
= btrfs_header_nritems(buf
);
2516 struct btrfs_leaf_ref
*ref
;
2517 struct btrfs_extent_info
*info
;
2519 ref
= btrfs_alloc_leaf_ref(root
, nr_extents
);
2525 ref
->root_gen
= root_gen
;
2526 ref
->bytenr
= buf
->start
;
2527 ref
->owner
= btrfs_header_owner(buf
);
2528 ref
->generation
= btrfs_header_generation(buf
);
2529 ref
->nritems
= nr_extents
;
2530 info
= ref
->extents
;
2532 for (i
= 0; nr_extents
> 0 && i
< nritems
; i
++) {
2534 btrfs_item_key_to_cpu(buf
, &key
, i
);
2535 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
2537 fi
= btrfs_item_ptr(buf
, i
,
2538 struct btrfs_file_extent_item
);
2539 if (btrfs_file_extent_type(buf
, fi
) ==
2540 BTRFS_FILE_EXTENT_INLINE
)
2542 disk_bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
2543 if (disk_bytenr
== 0)
2546 info
->bytenr
= disk_bytenr
;
2548 btrfs_file_extent_disk_num_bytes(buf
, fi
);
2549 info
->objectid
= key
.objectid
;
2550 info
->offset
= key
.offset
;
2554 ret
= btrfs_add_leaf_ref(root
, ref
, shared
);
2555 if (ret
== -EEXIST
&& shared
) {
2556 struct btrfs_leaf_ref
*old
;
2557 old
= btrfs_lookup_leaf_ref(root
, ref
->bytenr
);
2559 btrfs_remove_leaf_ref(root
, old
);
2560 btrfs_free_leaf_ref(root
, old
);
2561 ret
= btrfs_add_leaf_ref(root
, ref
, shared
);
2564 btrfs_free_leaf_ref(root
, ref
);
2570 /* when a block goes through cow, we update the reference counts of
2571 * everything that block points to. The internal pointers of the block
2572 * can be in just about any order, and it is likely to have clusters of
2573 * things that are close together and clusters of things that are not.
2575 * To help reduce the seeks that come with updating all of these reference
2576 * counts, sort them by byte number before actual updates are done.
2578 * struct refsort is used to match byte number to slot in the btree block.
2579 * we sort based on the byte number and then use the slot to actually
2582 * struct refsort is smaller than strcut btrfs_item and smaller than
2583 * struct btrfs_key_ptr. Since we're currently limited to the page size
2584 * for a btree block, there's no way for a kmalloc of refsorts for a
2585 * single node to be bigger than a page.
2593 * for passing into sort()
2595 static int refsort_cmp(const void *a_void
, const void *b_void
)
2597 const struct refsort
*a
= a_void
;
2598 const struct refsort
*b
= b_void
;
2600 if (a
->bytenr
< b
->bytenr
)
2602 if (a
->bytenr
> b
->bytenr
)
2608 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
2609 struct btrfs_root
*root
,
2610 struct extent_buffer
*buf
,
2611 int full_backref
, int inc
)
2618 struct btrfs_key key
;
2619 struct btrfs_file_extent_item
*fi
;
2623 int (*process_func
)(struct btrfs_trans_handle
*, struct btrfs_root
*,
2624 u64
, u64
, u64
, u64
, u64
, u64
);
2626 ref_root
= btrfs_header_owner(buf
);
2627 nritems
= btrfs_header_nritems(buf
);
2628 level
= btrfs_header_level(buf
);
2630 if (!root
->ref_cows
&& level
== 0)
2634 process_func
= btrfs_inc_extent_ref
;
2636 process_func
= btrfs_free_extent
;
2639 parent
= buf
->start
;
2643 for (i
= 0; i
< nritems
; i
++) {
2645 btrfs_item_key_to_cpu(buf
, &key
, i
);
2646 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
2648 fi
= btrfs_item_ptr(buf
, i
,
2649 struct btrfs_file_extent_item
);
2650 if (btrfs_file_extent_type(buf
, fi
) ==
2651 BTRFS_FILE_EXTENT_INLINE
)
2653 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
2657 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
2658 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
2659 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
2660 parent
, ref_root
, key
.objectid
,
2665 bytenr
= btrfs_node_blockptr(buf
, i
);
2666 num_bytes
= btrfs_level_size(root
, level
- 1);
2667 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
2668 parent
, ref_root
, level
- 1, 0);
2679 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2680 struct extent_buffer
*buf
, int full_backref
)
2682 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1);
2685 int btrfs_dec_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
, 0);
2691 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
2692 struct btrfs_root
*root
,
2693 struct btrfs_path
*path
,
2694 struct btrfs_block_group_cache
*cache
)
2697 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2699 struct extent_buffer
*leaf
;
2701 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
2706 leaf
= path
->nodes
[0];
2707 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
2708 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
2709 btrfs_mark_buffer_dirty(leaf
);
2710 btrfs_release_path(extent_root
, path
);
2718 static struct btrfs_block_group_cache
*
2719 next_block_group(struct btrfs_root
*root
,
2720 struct btrfs_block_group_cache
*cache
)
2722 struct rb_node
*node
;
2723 spin_lock(&root
->fs_info
->block_group_cache_lock
);
2724 node
= rb_next(&cache
->cache_node
);
2725 btrfs_put_block_group(cache
);
2727 cache
= rb_entry(node
, struct btrfs_block_group_cache
,
2729 btrfs_get_block_group(cache
);
2732 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
2736 static int cache_save_setup(struct btrfs_block_group_cache
*block_group
,
2737 struct btrfs_trans_handle
*trans
,
2738 struct btrfs_path
*path
)
2740 struct btrfs_root
*root
= block_group
->fs_info
->tree_root
;
2741 struct inode
*inode
= NULL
;
2743 int dcs
= BTRFS_DC_ERROR
;
2749 * If this block group is smaller than 100 megs don't bother caching the
2752 if (block_group
->key
.offset
< (100 * 1024 * 1024)) {
2753 spin_lock(&block_group
->lock
);
2754 block_group
->disk_cache_state
= BTRFS_DC_WRITTEN
;
2755 spin_unlock(&block_group
->lock
);
2760 inode
= lookup_free_space_inode(root
, block_group
, path
);
2761 if (IS_ERR(inode
) && PTR_ERR(inode
) != -ENOENT
) {
2762 ret
= PTR_ERR(inode
);
2763 btrfs_release_path(root
, path
);
2767 if (IS_ERR(inode
)) {
2771 if (block_group
->ro
)
2774 ret
= create_free_space_inode(root
, trans
, block_group
, path
);
2781 * We want to set the generation to 0, that way if anything goes wrong
2782 * from here on out we know not to trust this cache when we load up next
2785 BTRFS_I(inode
)->generation
= 0;
2786 ret
= btrfs_update_inode(trans
, root
, inode
);
2789 if (i_size_read(inode
) > 0) {
2790 ret
= btrfs_truncate_free_space_cache(root
, trans
, path
,
2796 spin_lock(&block_group
->lock
);
2797 if (block_group
->cached
!= BTRFS_CACHE_FINISHED
) {
2798 /* We're not cached, don't bother trying to write stuff out */
2799 dcs
= BTRFS_DC_WRITTEN
;
2800 spin_unlock(&block_group
->lock
);
2803 spin_unlock(&block_group
->lock
);
2805 num_pages
= (int)div64_u64(block_group
->key
.offset
, 1024 * 1024 * 1024);
2810 * Just to make absolutely sure we have enough space, we're going to
2811 * preallocate 12 pages worth of space for each block group. In
2812 * practice we ought to use at most 8, but we need extra space so we can
2813 * add our header and have a terminator between the extents and the
2817 num_pages
*= PAGE_CACHE_SIZE
;
2819 ret
= btrfs_check_data_free_space(inode
, num_pages
);
2823 ret
= btrfs_prealloc_file_range_trans(inode
, trans
, 0, 0, num_pages
,
2824 num_pages
, num_pages
,
2827 dcs
= BTRFS_DC_SETUP
;
2828 btrfs_free_reserved_data_space(inode
, num_pages
);
2832 btrfs_release_path(root
, path
);
2834 spin_lock(&block_group
->lock
);
2835 block_group
->disk_cache_state
= dcs
;
2836 spin_unlock(&block_group
->lock
);
2841 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
2842 struct btrfs_root
*root
)
2844 struct btrfs_block_group_cache
*cache
;
2846 struct btrfs_path
*path
;
2849 path
= btrfs_alloc_path();
2855 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
2857 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
)
2859 cache
= next_block_group(root
, cache
);
2867 err
= cache_save_setup(cache
, trans
, path
);
2868 last
= cache
->key
.objectid
+ cache
->key
.offset
;
2869 btrfs_put_block_group(cache
);
2874 err
= btrfs_run_delayed_refs(trans
, root
,
2879 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
2881 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
) {
2882 btrfs_put_block_group(cache
);
2888 cache
= next_block_group(root
, cache
);
2897 if (cache
->disk_cache_state
== BTRFS_DC_SETUP
)
2898 cache
->disk_cache_state
= BTRFS_DC_NEED_WRITE
;
2900 last
= cache
->key
.objectid
+ cache
->key
.offset
;
2902 err
= write_one_cache_group(trans
, root
, path
, cache
);
2904 btrfs_put_block_group(cache
);
2909 * I don't think this is needed since we're just marking our
2910 * preallocated extent as written, but just in case it can't
2914 err
= btrfs_run_delayed_refs(trans
, root
,
2919 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
2922 * Really this shouldn't happen, but it could if we
2923 * couldn't write the entire preallocated extent and
2924 * splitting the extent resulted in a new block.
2927 btrfs_put_block_group(cache
);
2930 if (cache
->disk_cache_state
== BTRFS_DC_NEED_WRITE
)
2932 cache
= next_block_group(root
, cache
);
2941 btrfs_write_out_cache(root
, trans
, cache
, path
);
2944 * If we didn't have an error then the cache state is still
2945 * NEED_WRITE, so we can set it to WRITTEN.
2947 if (cache
->disk_cache_state
== BTRFS_DC_NEED_WRITE
)
2948 cache
->disk_cache_state
= BTRFS_DC_WRITTEN
;
2949 last
= cache
->key
.objectid
+ cache
->key
.offset
;
2950 btrfs_put_block_group(cache
);
2953 btrfs_free_path(path
);
2957 int btrfs_extent_readonly(struct btrfs_root
*root
, u64 bytenr
)
2959 struct btrfs_block_group_cache
*block_group
;
2962 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
2963 if (!block_group
|| block_group
->ro
)
2966 btrfs_put_block_group(block_group
);
2970 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
2971 u64 total_bytes
, u64 bytes_used
,
2972 struct btrfs_space_info
**space_info
)
2974 struct btrfs_space_info
*found
;
2978 if (flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
2979 BTRFS_BLOCK_GROUP_RAID10
))
2984 found
= __find_space_info(info
, flags
);
2986 spin_lock(&found
->lock
);
2987 found
->total_bytes
+= total_bytes
;
2988 found
->disk_total
+= total_bytes
* factor
;
2989 found
->bytes_used
+= bytes_used
;
2990 found
->disk_used
+= bytes_used
* factor
;
2992 spin_unlock(&found
->lock
);
2993 *space_info
= found
;
2996 found
= kzalloc(sizeof(*found
), GFP_NOFS
);
3000 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
3001 INIT_LIST_HEAD(&found
->block_groups
[i
]);
3002 init_rwsem(&found
->groups_sem
);
3003 spin_lock_init(&found
->lock
);
3004 found
->flags
= flags
& (BTRFS_BLOCK_GROUP_DATA
|
3005 BTRFS_BLOCK_GROUP_SYSTEM
|
3006 BTRFS_BLOCK_GROUP_METADATA
);
3007 found
->total_bytes
= total_bytes
;
3008 found
->disk_total
= total_bytes
* factor
;
3009 found
->bytes_used
= bytes_used
;
3010 found
->disk_used
= bytes_used
* factor
;
3011 found
->bytes_pinned
= 0;
3012 found
->bytes_reserved
= 0;
3013 found
->bytes_readonly
= 0;
3014 found
->bytes_may_use
= 0;
3016 found
->force_alloc
= 0;
3017 *space_info
= found
;
3018 list_add_rcu(&found
->list
, &info
->space_info
);
3019 atomic_set(&found
->caching_threads
, 0);
3023 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
3025 u64 extra_flags
= flags
& (BTRFS_BLOCK_GROUP_RAID0
|
3026 BTRFS_BLOCK_GROUP_RAID1
|
3027 BTRFS_BLOCK_GROUP_RAID10
|
3028 BTRFS_BLOCK_GROUP_DUP
);
3030 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3031 fs_info
->avail_data_alloc_bits
|= extra_flags
;
3032 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3033 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
3034 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3035 fs_info
->avail_system_alloc_bits
|= extra_flags
;
3039 u64
btrfs_reduce_alloc_profile(struct btrfs_root
*root
, u64 flags
)
3042 * we add in the count of missing devices because we want
3043 * to make sure that any RAID levels on a degraded FS
3044 * continue to be honored.
3046 u64 num_devices
= root
->fs_info
->fs_devices
->rw_devices
+
3047 root
->fs_info
->fs_devices
->missing_devices
;
3049 if (num_devices
== 1)
3050 flags
&= ~(BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID0
);
3051 if (num_devices
< 4)
3052 flags
&= ~BTRFS_BLOCK_GROUP_RAID10
;
3054 if ((flags
& BTRFS_BLOCK_GROUP_DUP
) &&
3055 (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
3056 BTRFS_BLOCK_GROUP_RAID10
))) {
3057 flags
&= ~BTRFS_BLOCK_GROUP_DUP
;
3060 if ((flags
& BTRFS_BLOCK_GROUP_RAID1
) &&
3061 (flags
& BTRFS_BLOCK_GROUP_RAID10
)) {
3062 flags
&= ~BTRFS_BLOCK_GROUP_RAID1
;
3065 if ((flags
& BTRFS_BLOCK_GROUP_RAID0
) &&
3066 ((flags
& BTRFS_BLOCK_GROUP_RAID1
) |
3067 (flags
& BTRFS_BLOCK_GROUP_RAID10
) |
3068 (flags
& BTRFS_BLOCK_GROUP_DUP
)))
3069 flags
&= ~BTRFS_BLOCK_GROUP_RAID0
;
3073 static u64
get_alloc_profile(struct btrfs_root
*root
, u64 flags
)
3075 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3076 flags
|= root
->fs_info
->avail_data_alloc_bits
&
3077 root
->fs_info
->data_alloc_profile
;
3078 else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3079 flags
|= root
->fs_info
->avail_system_alloc_bits
&
3080 root
->fs_info
->system_alloc_profile
;
3081 else if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3082 flags
|= root
->fs_info
->avail_metadata_alloc_bits
&
3083 root
->fs_info
->metadata_alloc_profile
;
3084 return btrfs_reduce_alloc_profile(root
, flags
);
3087 u64
btrfs_get_alloc_profile(struct btrfs_root
*root
, int data
)
3092 flags
= BTRFS_BLOCK_GROUP_DATA
;
3093 else if (root
== root
->fs_info
->chunk_root
)
3094 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
3096 flags
= BTRFS_BLOCK_GROUP_METADATA
;
3098 return get_alloc_profile(root
, flags
);
3101 void btrfs_set_inode_space_info(struct btrfs_root
*root
, struct inode
*inode
)
3103 BTRFS_I(inode
)->space_info
= __find_space_info(root
->fs_info
,
3104 BTRFS_BLOCK_GROUP_DATA
);
3108 * This will check the space that the inode allocates from to make sure we have
3109 * enough space for bytes.
3111 int btrfs_check_data_free_space(struct inode
*inode
, u64 bytes
)
3113 struct btrfs_space_info
*data_sinfo
;
3114 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3116 int ret
= 0, committed
= 0, alloc_chunk
= 1;
3118 /* make sure bytes are sectorsize aligned */
3119 bytes
= (bytes
+ root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
3121 if (root
== root
->fs_info
->tree_root
) {
3126 data_sinfo
= BTRFS_I(inode
)->space_info
;
3131 /* make sure we have enough space to handle the data first */
3132 spin_lock(&data_sinfo
->lock
);
3133 used
= data_sinfo
->bytes_used
+ data_sinfo
->bytes_reserved
+
3134 data_sinfo
->bytes_pinned
+ data_sinfo
->bytes_readonly
+
3135 data_sinfo
->bytes_may_use
;
3137 if (used
+ bytes
> data_sinfo
->total_bytes
) {
3138 struct btrfs_trans_handle
*trans
;
3141 * if we don't have enough free bytes in this space then we need
3142 * to alloc a new chunk.
3144 if (!data_sinfo
->full
&& alloc_chunk
) {
3147 data_sinfo
->force_alloc
= 1;
3148 spin_unlock(&data_sinfo
->lock
);
3150 alloc_target
= btrfs_get_alloc_profile(root
, 1);
3151 trans
= btrfs_join_transaction(root
, 1);
3153 return PTR_ERR(trans
);
3155 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
3156 bytes
+ 2 * 1024 * 1024,
3158 btrfs_end_transaction(trans
, root
);
3167 btrfs_set_inode_space_info(root
, inode
);
3168 data_sinfo
= BTRFS_I(inode
)->space_info
;
3172 spin_unlock(&data_sinfo
->lock
);
3174 /* commit the current transaction and try again */
3176 if (!committed
&& !root
->fs_info
->open_ioctl_trans
) {
3178 trans
= btrfs_join_transaction(root
, 1);
3180 return PTR_ERR(trans
);
3181 ret
= btrfs_commit_transaction(trans
, root
);
3187 #if 0 /* I hope we never need this code again, just in case */
3188 printk(KERN_ERR
"no space left, need %llu, %llu bytes_used, "
3189 "%llu bytes_reserved, " "%llu bytes_pinned, "
3190 "%llu bytes_readonly, %llu may use %llu total\n",
3191 (unsigned long long)bytes
,
3192 (unsigned long long)data_sinfo
->bytes_used
,
3193 (unsigned long long)data_sinfo
->bytes_reserved
,
3194 (unsigned long long)data_sinfo
->bytes_pinned
,
3195 (unsigned long long)data_sinfo
->bytes_readonly
,
3196 (unsigned long long)data_sinfo
->bytes_may_use
,
3197 (unsigned long long)data_sinfo
->total_bytes
);
3201 data_sinfo
->bytes_may_use
+= bytes
;
3202 BTRFS_I(inode
)->reserved_bytes
+= bytes
;
3203 spin_unlock(&data_sinfo
->lock
);
3209 * called when we are clearing an delalloc extent from the
3210 * inode's io_tree or there was an error for whatever reason
3211 * after calling btrfs_check_data_free_space
3213 void btrfs_free_reserved_data_space(struct inode
*inode
, u64 bytes
)
3215 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3216 struct btrfs_space_info
*data_sinfo
;
3218 /* make sure bytes are sectorsize aligned */
3219 bytes
= (bytes
+ root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
3221 data_sinfo
= BTRFS_I(inode
)->space_info
;
3222 spin_lock(&data_sinfo
->lock
);
3223 data_sinfo
->bytes_may_use
-= bytes
;
3224 BTRFS_I(inode
)->reserved_bytes
-= bytes
;
3225 spin_unlock(&data_sinfo
->lock
);
3228 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
3230 struct list_head
*head
= &info
->space_info
;
3231 struct btrfs_space_info
*found
;
3234 list_for_each_entry_rcu(found
, head
, list
) {
3235 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
3236 found
->force_alloc
= 1;
3241 static int should_alloc_chunk(struct btrfs_root
*root
,
3242 struct btrfs_space_info
*sinfo
, u64 alloc_bytes
)
3244 u64 num_bytes
= sinfo
->total_bytes
- sinfo
->bytes_readonly
;
3247 if (sinfo
->bytes_used
+ sinfo
->bytes_reserved
+
3248 alloc_bytes
+ 256 * 1024 * 1024 < num_bytes
)
3251 if (sinfo
->bytes_used
+ sinfo
->bytes_reserved
+
3252 alloc_bytes
< div_factor(num_bytes
, 8))
3255 thresh
= btrfs_super_total_bytes(&root
->fs_info
->super_copy
);
3256 thresh
= max_t(u64
, 256 * 1024 * 1024, div_factor_fine(thresh
, 5));
3258 if (num_bytes
> thresh
&& sinfo
->bytes_used
< div_factor(num_bytes
, 3))
3264 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
3265 struct btrfs_root
*extent_root
, u64 alloc_bytes
,
3266 u64 flags
, int force
)
3268 struct btrfs_space_info
*space_info
;
3269 struct btrfs_fs_info
*fs_info
= extent_root
->fs_info
;
3272 mutex_lock(&fs_info
->chunk_mutex
);
3274 flags
= btrfs_reduce_alloc_profile(extent_root
, flags
);
3276 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
3278 ret
= update_space_info(extent_root
->fs_info
, flags
,
3282 BUG_ON(!space_info
);
3284 spin_lock(&space_info
->lock
);
3285 if (space_info
->force_alloc
)
3287 if (space_info
->full
) {
3288 spin_unlock(&space_info
->lock
);
3292 if (!force
&& !should_alloc_chunk(extent_root
, space_info
,
3294 spin_unlock(&space_info
->lock
);
3297 spin_unlock(&space_info
->lock
);
3300 * If we have mixed data/metadata chunks we want to make sure we keep
3301 * allocating mixed chunks instead of individual chunks.
3303 if (btrfs_mixed_space_info(space_info
))
3304 flags
|= (BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
);
3307 * if we're doing a data chunk, go ahead and make sure that
3308 * we keep a reasonable number of metadata chunks allocated in the
3311 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
3312 fs_info
->data_chunk_allocations
++;
3313 if (!(fs_info
->data_chunk_allocations
%
3314 fs_info
->metadata_ratio
))
3315 force_metadata_allocation(fs_info
);
3318 ret
= btrfs_alloc_chunk(trans
, extent_root
, flags
);
3319 spin_lock(&space_info
->lock
);
3321 space_info
->full
= 1;
3324 space_info
->force_alloc
= 0;
3325 spin_unlock(&space_info
->lock
);
3327 mutex_unlock(&extent_root
->fs_info
->chunk_mutex
);
3332 * shrink metadata reservation for delalloc
3334 static int shrink_delalloc(struct btrfs_trans_handle
*trans
,
3335 struct btrfs_root
*root
, u64 to_reclaim
, int sync
)
3337 struct btrfs_block_rsv
*block_rsv
;
3338 struct btrfs_space_info
*space_info
;
3343 int nr_pages
= (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT
;
3345 unsigned long progress
;
3347 block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
3348 space_info
= block_rsv
->space_info
;
3351 reserved
= space_info
->bytes_reserved
;
3352 progress
= space_info
->reservation_progress
;
3357 max_reclaim
= min(reserved
, to_reclaim
);
3359 while (loops
< 1024) {
3360 /* have the flusher threads jump in and do some IO */
3362 nr_pages
= min_t(unsigned long, nr_pages
,
3363 root
->fs_info
->delalloc_bytes
>> PAGE_CACHE_SHIFT
);
3364 writeback_inodes_sb_nr_if_idle(root
->fs_info
->sb
, nr_pages
);
3366 spin_lock(&space_info
->lock
);
3367 if (reserved
> space_info
->bytes_reserved
)
3368 reclaimed
+= reserved
- space_info
->bytes_reserved
;
3369 reserved
= space_info
->bytes_reserved
;
3370 spin_unlock(&space_info
->lock
);
3374 if (reserved
== 0 || reclaimed
>= max_reclaim
)
3377 if (trans
&& trans
->transaction
->blocked
)
3380 time_left
= schedule_timeout_interruptible(1);
3382 /* We were interrupted, exit */
3386 /* we've kicked the IO a few times, if anything has been freed,
3387 * exit. There is no sense in looping here for a long time
3388 * when we really need to commit the transaction, or there are
3389 * just too many writers without enough free space
3394 if (progress
!= space_info
->reservation_progress
)
3399 return reclaimed
>= to_reclaim
;
3403 * Retries tells us how many times we've called reserve_metadata_bytes. The
3404 * idea is if this is the first call (retries == 0) then we will add to our
3405 * reserved count if we can't make the allocation in order to hold our place
3406 * while we go and try and free up space. That way for retries > 1 we don't try
3407 * and add space, we just check to see if the amount of unused space is >= the
3408 * total space, meaning that our reservation is valid.
3410 * However if we don't intend to retry this reservation, pass -1 as retries so
3411 * that it short circuits this logic.
3413 static int reserve_metadata_bytes(struct btrfs_trans_handle
*trans
,
3414 struct btrfs_root
*root
,
3415 struct btrfs_block_rsv
*block_rsv
,
3416 u64 orig_bytes
, int flush
)
3418 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
3420 u64 num_bytes
= orig_bytes
;
3423 bool reserved
= false;
3424 bool committed
= false;
3431 spin_lock(&space_info
->lock
);
3432 unused
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
3433 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
3434 space_info
->bytes_may_use
;
3437 * The idea here is that we've not already over-reserved the block group
3438 * then we can go ahead and save our reservation first and then start
3439 * flushing if we need to. Otherwise if we've already overcommitted
3440 * lets start flushing stuff first and then come back and try to make
3443 if (unused
<= space_info
->total_bytes
) {
3444 unused
= space_info
->total_bytes
- unused
;
3445 if (unused
>= num_bytes
) {
3447 space_info
->bytes_reserved
+= orig_bytes
;
3451 * Ok set num_bytes to orig_bytes since we aren't
3452 * overocmmitted, this way we only try and reclaim what
3455 num_bytes
= orig_bytes
;
3459 * Ok we're over committed, set num_bytes to the overcommitted
3460 * amount plus the amount of bytes that we need for this
3463 num_bytes
= unused
- space_info
->total_bytes
+
3464 (orig_bytes
* (retries
+ 1));
3468 * Couldn't make our reservation, save our place so while we're trying
3469 * to reclaim space we can actually use it instead of somebody else
3470 * stealing it from us.
3472 if (ret
&& !reserved
) {
3473 space_info
->bytes_reserved
+= orig_bytes
;
3477 spin_unlock(&space_info
->lock
);
3486 * We do synchronous shrinking since we don't actually unreserve
3487 * metadata until after the IO is completed.
3489 ret
= shrink_delalloc(trans
, root
, num_bytes
, 1);
3496 * So if we were overcommitted it's possible that somebody else flushed
3497 * out enough space and we simply didn't have enough space to reclaim,
3498 * so go back around and try again.
3505 spin_lock(&space_info
->lock
);
3507 * Not enough space to be reclaimed, don't bother committing the
3510 if (space_info
->bytes_pinned
< orig_bytes
)
3512 spin_unlock(&space_info
->lock
);
3517 if (trans
|| committed
)
3521 trans
= btrfs_join_transaction(root
, 1);
3524 ret
= btrfs_commit_transaction(trans
, root
);
3533 spin_lock(&space_info
->lock
);
3534 space_info
->bytes_reserved
-= orig_bytes
;
3535 spin_unlock(&space_info
->lock
);
3541 static struct btrfs_block_rsv
*get_block_rsv(struct btrfs_trans_handle
*trans
,
3542 struct btrfs_root
*root
)
3544 struct btrfs_block_rsv
*block_rsv
;
3546 block_rsv
= trans
->block_rsv
;
3548 block_rsv
= root
->block_rsv
;
3551 block_rsv
= &root
->fs_info
->empty_block_rsv
;
3556 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
3560 spin_lock(&block_rsv
->lock
);
3561 if (block_rsv
->reserved
>= num_bytes
) {
3562 block_rsv
->reserved
-= num_bytes
;
3563 if (block_rsv
->reserved
< block_rsv
->size
)
3564 block_rsv
->full
= 0;
3567 spin_unlock(&block_rsv
->lock
);
3571 static void block_rsv_add_bytes(struct btrfs_block_rsv
*block_rsv
,
3572 u64 num_bytes
, int update_size
)
3574 spin_lock(&block_rsv
->lock
);
3575 block_rsv
->reserved
+= num_bytes
;
3577 block_rsv
->size
+= num_bytes
;
3578 else if (block_rsv
->reserved
>= block_rsv
->size
)
3579 block_rsv
->full
= 1;
3580 spin_unlock(&block_rsv
->lock
);
3583 void block_rsv_release_bytes(struct btrfs_block_rsv
*block_rsv
,
3584 struct btrfs_block_rsv
*dest
, u64 num_bytes
)
3586 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
3588 spin_lock(&block_rsv
->lock
);
3589 if (num_bytes
== (u64
)-1)
3590 num_bytes
= block_rsv
->size
;
3591 block_rsv
->size
-= num_bytes
;
3592 if (block_rsv
->reserved
>= block_rsv
->size
) {
3593 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
3594 block_rsv
->reserved
= block_rsv
->size
;
3595 block_rsv
->full
= 1;
3599 spin_unlock(&block_rsv
->lock
);
3601 if (num_bytes
> 0) {
3603 spin_lock(&dest
->lock
);
3607 bytes_to_add
= dest
->size
- dest
->reserved
;
3608 bytes_to_add
= min(num_bytes
, bytes_to_add
);
3609 dest
->reserved
+= bytes_to_add
;
3610 if (dest
->reserved
>= dest
->size
)
3612 num_bytes
-= bytes_to_add
;
3614 spin_unlock(&dest
->lock
);
3617 spin_lock(&space_info
->lock
);
3618 space_info
->bytes_reserved
-= num_bytes
;
3619 space_info
->reservation_progress
++;
3620 spin_unlock(&space_info
->lock
);
3625 static int block_rsv_migrate_bytes(struct btrfs_block_rsv
*src
,
3626 struct btrfs_block_rsv
*dst
, u64 num_bytes
)
3630 ret
= block_rsv_use_bytes(src
, num_bytes
);
3634 block_rsv_add_bytes(dst
, num_bytes
, 1);
3638 void btrfs_init_block_rsv(struct btrfs_block_rsv
*rsv
)
3640 memset(rsv
, 0, sizeof(*rsv
));
3641 spin_lock_init(&rsv
->lock
);
3642 atomic_set(&rsv
->usage
, 1);
3644 INIT_LIST_HEAD(&rsv
->list
);
3647 struct btrfs_block_rsv
*btrfs_alloc_block_rsv(struct btrfs_root
*root
)
3649 struct btrfs_block_rsv
*block_rsv
;
3650 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3652 block_rsv
= kmalloc(sizeof(*block_rsv
), GFP_NOFS
);
3656 btrfs_init_block_rsv(block_rsv
);
3657 block_rsv
->space_info
= __find_space_info(fs_info
,
3658 BTRFS_BLOCK_GROUP_METADATA
);
3662 void btrfs_free_block_rsv(struct btrfs_root
*root
,
3663 struct btrfs_block_rsv
*rsv
)
3665 if (rsv
&& atomic_dec_and_test(&rsv
->usage
)) {
3666 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
3673 * make the block_rsv struct be able to capture freed space.
3674 * the captured space will re-add to the the block_rsv struct
3675 * after transaction commit
3677 void btrfs_add_durable_block_rsv(struct btrfs_fs_info
*fs_info
,
3678 struct btrfs_block_rsv
*block_rsv
)
3680 block_rsv
->durable
= 1;
3681 mutex_lock(&fs_info
->durable_block_rsv_mutex
);
3682 list_add_tail(&block_rsv
->list
, &fs_info
->durable_block_rsv_list
);
3683 mutex_unlock(&fs_info
->durable_block_rsv_mutex
);
3686 int btrfs_block_rsv_add(struct btrfs_trans_handle
*trans
,
3687 struct btrfs_root
*root
,
3688 struct btrfs_block_rsv
*block_rsv
,
3696 ret
= reserve_metadata_bytes(trans
, root
, block_rsv
, num_bytes
, 1);
3698 block_rsv_add_bytes(block_rsv
, num_bytes
, 1);
3705 int btrfs_block_rsv_check(struct btrfs_trans_handle
*trans
,
3706 struct btrfs_root
*root
,
3707 struct btrfs_block_rsv
*block_rsv
,
3708 u64 min_reserved
, int min_factor
)
3711 int commit_trans
= 0;
3717 spin_lock(&block_rsv
->lock
);
3719 num_bytes
= div_factor(block_rsv
->size
, min_factor
);
3720 if (min_reserved
> num_bytes
)
3721 num_bytes
= min_reserved
;
3723 if (block_rsv
->reserved
>= num_bytes
) {
3726 num_bytes
-= block_rsv
->reserved
;
3727 if (block_rsv
->durable
&&
3728 block_rsv
->freed
[0] + block_rsv
->freed
[1] >= num_bytes
)
3731 spin_unlock(&block_rsv
->lock
);
3735 if (block_rsv
->refill_used
) {
3736 ret
= reserve_metadata_bytes(trans
, root
, block_rsv
,
3739 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
3748 trans
= btrfs_join_transaction(root
, 1);
3749 BUG_ON(IS_ERR(trans
));
3750 ret
= btrfs_commit_transaction(trans
, root
);
3757 int btrfs_block_rsv_migrate(struct btrfs_block_rsv
*src_rsv
,
3758 struct btrfs_block_rsv
*dst_rsv
,
3761 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
3764 void btrfs_block_rsv_release(struct btrfs_root
*root
,
3765 struct btrfs_block_rsv
*block_rsv
,
3768 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
3769 if (global_rsv
->full
|| global_rsv
== block_rsv
||
3770 block_rsv
->space_info
!= global_rsv
->space_info
)
3772 block_rsv_release_bytes(block_rsv
, global_rsv
, num_bytes
);
3776 * helper to calculate size of global block reservation.
3777 * the desired value is sum of space used by extent tree,
3778 * checksum tree and root tree
3780 static u64
calc_global_metadata_size(struct btrfs_fs_info
*fs_info
)
3782 struct btrfs_space_info
*sinfo
;
3786 int csum_size
= btrfs_super_csum_size(&fs_info
->super_copy
);
3789 * per tree used space accounting can be inaccuracy, so we
3792 spin_lock(&fs_info
->extent_root
->accounting_lock
);
3793 num_bytes
= btrfs_root_used(&fs_info
->extent_root
->root_item
);
3794 spin_unlock(&fs_info
->extent_root
->accounting_lock
);
3796 spin_lock(&fs_info
->csum_root
->accounting_lock
);
3797 num_bytes
+= btrfs_root_used(&fs_info
->csum_root
->root_item
);
3798 spin_unlock(&fs_info
->csum_root
->accounting_lock
);
3800 spin_lock(&fs_info
->tree_root
->accounting_lock
);
3801 num_bytes
+= btrfs_root_used(&fs_info
->tree_root
->root_item
);
3802 spin_unlock(&fs_info
->tree_root
->accounting_lock
);
3804 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_DATA
);
3805 spin_lock(&sinfo
->lock
);
3806 data_used
= sinfo
->bytes_used
;
3807 spin_unlock(&sinfo
->lock
);
3809 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
3810 spin_lock(&sinfo
->lock
);
3811 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_DATA
)
3813 meta_used
= sinfo
->bytes_used
;
3814 spin_unlock(&sinfo
->lock
);
3816 num_bytes
= (data_used
>> fs_info
->sb
->s_blocksize_bits
) *
3818 num_bytes
+= div64_u64(data_used
+ meta_used
, 50);
3820 if (num_bytes
* 3 > meta_used
)
3821 num_bytes
= div64_u64(meta_used
, 3);
3823 return ALIGN(num_bytes
, fs_info
->extent_root
->leafsize
<< 10);
3826 static void update_global_block_rsv(struct btrfs_fs_info
*fs_info
)
3828 struct btrfs_block_rsv
*block_rsv
= &fs_info
->global_block_rsv
;
3829 struct btrfs_space_info
*sinfo
= block_rsv
->space_info
;
3832 num_bytes
= calc_global_metadata_size(fs_info
);
3834 spin_lock(&block_rsv
->lock
);
3835 spin_lock(&sinfo
->lock
);
3837 block_rsv
->size
= num_bytes
;
3839 num_bytes
= sinfo
->bytes_used
+ sinfo
->bytes_pinned
+
3840 sinfo
->bytes_reserved
+ sinfo
->bytes_readonly
+
3841 sinfo
->bytes_may_use
;
3843 if (sinfo
->total_bytes
> num_bytes
) {
3844 num_bytes
= sinfo
->total_bytes
- num_bytes
;
3845 block_rsv
->reserved
+= num_bytes
;
3846 sinfo
->bytes_reserved
+= num_bytes
;
3849 if (block_rsv
->reserved
>= block_rsv
->size
) {
3850 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
3851 sinfo
->bytes_reserved
-= num_bytes
;
3852 sinfo
->reservation_progress
++;
3853 block_rsv
->reserved
= block_rsv
->size
;
3854 block_rsv
->full
= 1;
3857 printk(KERN_INFO
"global block rsv size %llu reserved %llu\n",
3858 block_rsv
->size
, block_rsv
->reserved
);
3860 spin_unlock(&sinfo
->lock
);
3861 spin_unlock(&block_rsv
->lock
);
3864 static void init_global_block_rsv(struct btrfs_fs_info
*fs_info
)
3866 struct btrfs_space_info
*space_info
;
3868 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
3869 fs_info
->chunk_block_rsv
.space_info
= space_info
;
3870 fs_info
->chunk_block_rsv
.priority
= 10;
3872 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
3873 fs_info
->global_block_rsv
.space_info
= space_info
;
3874 fs_info
->global_block_rsv
.priority
= 10;
3875 fs_info
->global_block_rsv
.refill_used
= 1;
3876 fs_info
->delalloc_block_rsv
.space_info
= space_info
;
3877 fs_info
->trans_block_rsv
.space_info
= space_info
;
3878 fs_info
->empty_block_rsv
.space_info
= space_info
;
3879 fs_info
->empty_block_rsv
.priority
= 10;
3881 fs_info
->extent_root
->block_rsv
= &fs_info
->global_block_rsv
;
3882 fs_info
->csum_root
->block_rsv
= &fs_info
->global_block_rsv
;
3883 fs_info
->dev_root
->block_rsv
= &fs_info
->global_block_rsv
;
3884 fs_info
->tree_root
->block_rsv
= &fs_info
->global_block_rsv
;
3885 fs_info
->chunk_root
->block_rsv
= &fs_info
->chunk_block_rsv
;
3887 btrfs_add_durable_block_rsv(fs_info
, &fs_info
->global_block_rsv
);
3889 btrfs_add_durable_block_rsv(fs_info
, &fs_info
->delalloc_block_rsv
);
3891 update_global_block_rsv(fs_info
);
3894 static void release_global_block_rsv(struct btrfs_fs_info
*fs_info
)
3896 block_rsv_release_bytes(&fs_info
->global_block_rsv
, NULL
, (u64
)-1);
3897 WARN_ON(fs_info
->delalloc_block_rsv
.size
> 0);
3898 WARN_ON(fs_info
->delalloc_block_rsv
.reserved
> 0);
3899 WARN_ON(fs_info
->trans_block_rsv
.size
> 0);
3900 WARN_ON(fs_info
->trans_block_rsv
.reserved
> 0);
3901 WARN_ON(fs_info
->chunk_block_rsv
.size
> 0);
3902 WARN_ON(fs_info
->chunk_block_rsv
.reserved
> 0);
3905 static u64
calc_trans_metadata_size(struct btrfs_root
*root
, int num_items
)
3907 return (root
->leafsize
+ root
->nodesize
* (BTRFS_MAX_LEVEL
- 1)) *
3911 int btrfs_trans_reserve_metadata(struct btrfs_trans_handle
*trans
,
3912 struct btrfs_root
*root
,
3918 if (num_items
== 0 || root
->fs_info
->chunk_root
== root
)
3921 num_bytes
= calc_trans_metadata_size(root
, num_items
);
3922 ret
= btrfs_block_rsv_add(trans
, root
, &root
->fs_info
->trans_block_rsv
,
3925 trans
->bytes_reserved
+= num_bytes
;
3926 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
3931 void btrfs_trans_release_metadata(struct btrfs_trans_handle
*trans
,
3932 struct btrfs_root
*root
)
3934 if (!trans
->bytes_reserved
)
3937 BUG_ON(trans
->block_rsv
!= &root
->fs_info
->trans_block_rsv
);
3938 btrfs_block_rsv_release(root
, trans
->block_rsv
,
3939 trans
->bytes_reserved
);
3940 trans
->bytes_reserved
= 0;
3943 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle
*trans
,
3944 struct inode
*inode
)
3946 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3947 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
3948 struct btrfs_block_rsv
*dst_rsv
= root
->orphan_block_rsv
;
3951 * one for deleting orphan item, one for updating inode and
3952 * two for calling btrfs_truncate_inode_items.
3954 * btrfs_truncate_inode_items is a delete operation, it frees
3955 * more space than it uses in most cases. So two units of
3956 * metadata space should be enough for calling it many times.
3957 * If all of the metadata space is used, we can commit
3958 * transaction and use space it freed.
3960 u64 num_bytes
= calc_trans_metadata_size(root
, 4);
3961 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
3964 void btrfs_orphan_release_metadata(struct inode
*inode
)
3966 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3967 u64 num_bytes
= calc_trans_metadata_size(root
, 4);
3968 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
, num_bytes
);
3971 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle
*trans
,
3972 struct btrfs_pending_snapshot
*pending
)
3974 struct btrfs_root
*root
= pending
->root
;
3975 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
3976 struct btrfs_block_rsv
*dst_rsv
= &pending
->block_rsv
;
3978 * two for root back/forward refs, two for directory entries
3979 * and one for root of the snapshot.
3981 u64 num_bytes
= calc_trans_metadata_size(root
, 5);
3982 dst_rsv
->space_info
= src_rsv
->space_info
;
3983 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
3986 static u64
calc_csum_metadata_size(struct inode
*inode
, u64 num_bytes
)
3988 return num_bytes
>>= 3;
3991 int btrfs_delalloc_reserve_metadata(struct inode
*inode
, u64 num_bytes
)
3993 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3994 struct btrfs_block_rsv
*block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
3997 int reserved_extents
;
4000 if (btrfs_transaction_in_commit(root
->fs_info
))
4001 schedule_timeout(1);
4003 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
4005 nr_extents
= atomic_read(&BTRFS_I(inode
)->outstanding_extents
) + 1;
4006 reserved_extents
= atomic_read(&BTRFS_I(inode
)->reserved_extents
);
4008 if (nr_extents
> reserved_extents
) {
4009 nr_extents
-= reserved_extents
;
4010 to_reserve
= calc_trans_metadata_size(root
, nr_extents
);
4016 to_reserve
+= calc_csum_metadata_size(inode
, num_bytes
);
4017 ret
= reserve_metadata_bytes(NULL
, root
, block_rsv
, to_reserve
, 1);
4021 atomic_add(nr_extents
, &BTRFS_I(inode
)->reserved_extents
);
4022 atomic_inc(&BTRFS_I(inode
)->outstanding_extents
);
4024 block_rsv_add_bytes(block_rsv
, to_reserve
, 1);
4026 if (block_rsv
->size
> 512 * 1024 * 1024)
4027 shrink_delalloc(NULL
, root
, to_reserve
, 0);
4032 void btrfs_delalloc_release_metadata(struct inode
*inode
, u64 num_bytes
)
4034 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4037 int reserved_extents
;
4039 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
4040 atomic_dec(&BTRFS_I(inode
)->outstanding_extents
);
4041 WARN_ON(atomic_read(&BTRFS_I(inode
)->outstanding_extents
) < 0);
4043 reserved_extents
= atomic_read(&BTRFS_I(inode
)->reserved_extents
);
4047 nr_extents
= atomic_read(&BTRFS_I(inode
)->outstanding_extents
);
4048 if (nr_extents
>= reserved_extents
) {
4052 old
= reserved_extents
;
4053 nr_extents
= reserved_extents
- nr_extents
;
4054 new = reserved_extents
- nr_extents
;
4055 old
= atomic_cmpxchg(&BTRFS_I(inode
)->reserved_extents
,
4056 reserved_extents
, new);
4057 if (likely(old
== reserved_extents
))
4059 reserved_extents
= old
;
4062 to_free
= calc_csum_metadata_size(inode
, num_bytes
);
4064 to_free
+= calc_trans_metadata_size(root
, nr_extents
);
4066 btrfs_block_rsv_release(root
, &root
->fs_info
->delalloc_block_rsv
,
4070 int btrfs_delalloc_reserve_space(struct inode
*inode
, u64 num_bytes
)
4074 ret
= btrfs_check_data_free_space(inode
, num_bytes
);
4078 ret
= btrfs_delalloc_reserve_metadata(inode
, num_bytes
);
4080 btrfs_free_reserved_data_space(inode
, num_bytes
);
4087 void btrfs_delalloc_release_space(struct inode
*inode
, u64 num_bytes
)
4089 btrfs_delalloc_release_metadata(inode
, num_bytes
);
4090 btrfs_free_reserved_data_space(inode
, num_bytes
);
4093 static int update_block_group(struct btrfs_trans_handle
*trans
,
4094 struct btrfs_root
*root
,
4095 u64 bytenr
, u64 num_bytes
, int alloc
)
4097 struct btrfs_block_group_cache
*cache
= NULL
;
4098 struct btrfs_fs_info
*info
= root
->fs_info
;
4099 u64 total
= num_bytes
;
4104 /* block accounting for super block */
4105 spin_lock(&info
->delalloc_lock
);
4106 old_val
= btrfs_super_bytes_used(&info
->super_copy
);
4108 old_val
+= num_bytes
;
4110 old_val
-= num_bytes
;
4111 btrfs_set_super_bytes_used(&info
->super_copy
, old_val
);
4112 spin_unlock(&info
->delalloc_lock
);
4115 cache
= btrfs_lookup_block_group(info
, bytenr
);
4118 if (cache
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
4119 BTRFS_BLOCK_GROUP_RAID1
|
4120 BTRFS_BLOCK_GROUP_RAID10
))
4125 * If this block group has free space cache written out, we
4126 * need to make sure to load it if we are removing space. This
4127 * is because we need the unpinning stage to actually add the
4128 * space back to the block group, otherwise we will leak space.
4130 if (!alloc
&& cache
->cached
== BTRFS_CACHE_NO
)
4131 cache_block_group(cache
, trans
, NULL
, 1);
4133 byte_in_group
= bytenr
- cache
->key
.objectid
;
4134 WARN_ON(byte_in_group
> cache
->key
.offset
);
4136 spin_lock(&cache
->space_info
->lock
);
4137 spin_lock(&cache
->lock
);
4139 if (btrfs_super_cache_generation(&info
->super_copy
) != 0 &&
4140 cache
->disk_cache_state
< BTRFS_DC_CLEAR
)
4141 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
4144 old_val
= btrfs_block_group_used(&cache
->item
);
4145 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
4147 old_val
+= num_bytes
;
4148 btrfs_set_block_group_used(&cache
->item
, old_val
);
4149 cache
->reserved
-= num_bytes
;
4150 cache
->space_info
->bytes_reserved
-= num_bytes
;
4151 cache
->space_info
->reservation_progress
++;
4152 cache
->space_info
->bytes_used
+= num_bytes
;
4153 cache
->space_info
->disk_used
+= num_bytes
* factor
;
4154 spin_unlock(&cache
->lock
);
4155 spin_unlock(&cache
->space_info
->lock
);
4157 old_val
-= num_bytes
;
4158 btrfs_set_block_group_used(&cache
->item
, old_val
);
4159 cache
->pinned
+= num_bytes
;
4160 cache
->space_info
->bytes_pinned
+= num_bytes
;
4161 cache
->space_info
->bytes_used
-= num_bytes
;
4162 cache
->space_info
->disk_used
-= num_bytes
* factor
;
4163 spin_unlock(&cache
->lock
);
4164 spin_unlock(&cache
->space_info
->lock
);
4166 set_extent_dirty(info
->pinned_extents
,
4167 bytenr
, bytenr
+ num_bytes
- 1,
4168 GFP_NOFS
| __GFP_NOFAIL
);
4170 btrfs_put_block_group(cache
);
4172 bytenr
+= num_bytes
;
4177 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
4179 struct btrfs_block_group_cache
*cache
;
4182 cache
= btrfs_lookup_first_block_group(root
->fs_info
, search_start
);
4186 bytenr
= cache
->key
.objectid
;
4187 btrfs_put_block_group(cache
);
4192 static int pin_down_extent(struct btrfs_root
*root
,
4193 struct btrfs_block_group_cache
*cache
,
4194 u64 bytenr
, u64 num_bytes
, int reserved
)
4196 spin_lock(&cache
->space_info
->lock
);
4197 spin_lock(&cache
->lock
);
4198 cache
->pinned
+= num_bytes
;
4199 cache
->space_info
->bytes_pinned
+= num_bytes
;
4201 cache
->reserved
-= num_bytes
;
4202 cache
->space_info
->bytes_reserved
-= num_bytes
;
4203 cache
->space_info
->reservation_progress
++;
4205 spin_unlock(&cache
->lock
);
4206 spin_unlock(&cache
->space_info
->lock
);
4208 set_extent_dirty(root
->fs_info
->pinned_extents
, bytenr
,
4209 bytenr
+ num_bytes
- 1, GFP_NOFS
| __GFP_NOFAIL
);
4214 * this function must be called within transaction
4216 int btrfs_pin_extent(struct btrfs_root
*root
,
4217 u64 bytenr
, u64 num_bytes
, int reserved
)
4219 struct btrfs_block_group_cache
*cache
;
4221 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
4224 pin_down_extent(root
, cache
, bytenr
, num_bytes
, reserved
);
4226 btrfs_put_block_group(cache
);
4231 * update size of reserved extents. this function may return -EAGAIN
4232 * if 'reserve' is true or 'sinfo' is false.
4234 int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
4235 u64 num_bytes
, int reserve
, int sinfo
)
4239 struct btrfs_space_info
*space_info
= cache
->space_info
;
4240 spin_lock(&space_info
->lock
);
4241 spin_lock(&cache
->lock
);
4246 cache
->reserved
+= num_bytes
;
4247 space_info
->bytes_reserved
+= num_bytes
;
4251 space_info
->bytes_readonly
+= num_bytes
;
4252 cache
->reserved
-= num_bytes
;
4253 space_info
->bytes_reserved
-= num_bytes
;
4254 space_info
->reservation_progress
++;
4256 spin_unlock(&cache
->lock
);
4257 spin_unlock(&space_info
->lock
);
4259 spin_lock(&cache
->lock
);
4264 cache
->reserved
+= num_bytes
;
4266 cache
->reserved
-= num_bytes
;
4268 spin_unlock(&cache
->lock
);
4273 int btrfs_prepare_extent_commit(struct btrfs_trans_handle
*trans
,
4274 struct btrfs_root
*root
)
4276 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4277 struct btrfs_caching_control
*next
;
4278 struct btrfs_caching_control
*caching_ctl
;
4279 struct btrfs_block_group_cache
*cache
;
4281 down_write(&fs_info
->extent_commit_sem
);
4283 list_for_each_entry_safe(caching_ctl
, next
,
4284 &fs_info
->caching_block_groups
, list
) {
4285 cache
= caching_ctl
->block_group
;
4286 if (block_group_cache_done(cache
)) {
4287 cache
->last_byte_to_unpin
= (u64
)-1;
4288 list_del_init(&caching_ctl
->list
);
4289 put_caching_control(caching_ctl
);
4291 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
4295 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
4296 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
4298 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
4300 up_write(&fs_info
->extent_commit_sem
);
4302 update_global_block_rsv(fs_info
);
4306 static int unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
4308 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4309 struct btrfs_block_group_cache
*cache
= NULL
;
4312 while (start
<= end
) {
4314 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
4316 btrfs_put_block_group(cache
);
4317 cache
= btrfs_lookup_block_group(fs_info
, start
);
4321 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
4322 len
= min(len
, end
+ 1 - start
);
4324 if (start
< cache
->last_byte_to_unpin
) {
4325 len
= min(len
, cache
->last_byte_to_unpin
- start
);
4326 btrfs_add_free_space(cache
, start
, len
);
4331 spin_lock(&cache
->space_info
->lock
);
4332 spin_lock(&cache
->lock
);
4333 cache
->pinned
-= len
;
4334 cache
->space_info
->bytes_pinned
-= len
;
4336 cache
->space_info
->bytes_readonly
+= len
;
4337 } else if (cache
->reserved_pinned
> 0) {
4338 len
= min(len
, cache
->reserved_pinned
);
4339 cache
->reserved_pinned
-= len
;
4340 cache
->space_info
->bytes_reserved
+= len
;
4342 spin_unlock(&cache
->lock
);
4343 spin_unlock(&cache
->space_info
->lock
);
4347 btrfs_put_block_group(cache
);
4351 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
4352 struct btrfs_root
*root
)
4354 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4355 struct extent_io_tree
*unpin
;
4356 struct btrfs_block_rsv
*block_rsv
;
4357 struct btrfs_block_rsv
*next_rsv
;
4363 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
4364 unpin
= &fs_info
->freed_extents
[1];
4366 unpin
= &fs_info
->freed_extents
[0];
4369 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
4374 ret
= btrfs_discard_extent(root
, start
, end
+ 1 - start
);
4376 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
4377 unpin_extent_range(root
, start
, end
);
4381 mutex_lock(&fs_info
->durable_block_rsv_mutex
);
4382 list_for_each_entry_safe(block_rsv
, next_rsv
,
4383 &fs_info
->durable_block_rsv_list
, list
) {
4385 idx
= trans
->transid
& 0x1;
4386 if (block_rsv
->freed
[idx
] > 0) {
4387 block_rsv_add_bytes(block_rsv
,
4388 block_rsv
->freed
[idx
], 0);
4389 block_rsv
->freed
[idx
] = 0;
4391 if (atomic_read(&block_rsv
->usage
) == 0) {
4392 btrfs_block_rsv_release(root
, block_rsv
, (u64
)-1);
4394 if (block_rsv
->freed
[0] == 0 &&
4395 block_rsv
->freed
[1] == 0) {
4396 list_del_init(&block_rsv
->list
);
4400 btrfs_block_rsv_release(root
, block_rsv
, 0);
4403 mutex_unlock(&fs_info
->durable_block_rsv_mutex
);
4408 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
4409 struct btrfs_root
*root
,
4410 u64 bytenr
, u64 num_bytes
, u64 parent
,
4411 u64 root_objectid
, u64 owner_objectid
,
4412 u64 owner_offset
, int refs_to_drop
,
4413 struct btrfs_delayed_extent_op
*extent_op
)
4415 struct btrfs_key key
;
4416 struct btrfs_path
*path
;
4417 struct btrfs_fs_info
*info
= root
->fs_info
;
4418 struct btrfs_root
*extent_root
= info
->extent_root
;
4419 struct extent_buffer
*leaf
;
4420 struct btrfs_extent_item
*ei
;
4421 struct btrfs_extent_inline_ref
*iref
;
4424 int extent_slot
= 0;
4425 int found_extent
= 0;
4430 path
= btrfs_alloc_path();
4435 path
->leave_spinning
= 1;
4437 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
4438 BUG_ON(!is_data
&& refs_to_drop
!= 1);
4440 ret
= lookup_extent_backref(trans
, extent_root
, path
, &iref
,
4441 bytenr
, num_bytes
, parent
,
4442 root_objectid
, owner_objectid
,
4445 extent_slot
= path
->slots
[0];
4446 while (extent_slot
>= 0) {
4447 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
4449 if (key
.objectid
!= bytenr
)
4451 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
4452 key
.offset
== num_bytes
) {
4456 if (path
->slots
[0] - extent_slot
> 5)
4460 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4461 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
4462 if (found_extent
&& item_size
< sizeof(*ei
))
4465 if (!found_extent
) {
4467 ret
= remove_extent_backref(trans
, extent_root
, path
,
4471 btrfs_release_path(extent_root
, path
);
4472 path
->leave_spinning
= 1;
4474 key
.objectid
= bytenr
;
4475 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
4476 key
.offset
= num_bytes
;
4478 ret
= btrfs_search_slot(trans
, extent_root
,
4481 printk(KERN_ERR
"umm, got %d back from search"
4482 ", was looking for %llu\n", ret
,
4483 (unsigned long long)bytenr
);
4484 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
4487 extent_slot
= path
->slots
[0];
4490 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
4492 printk(KERN_ERR
"btrfs unable to find ref byte nr %llu "
4493 "parent %llu root %llu owner %llu offset %llu\n",
4494 (unsigned long long)bytenr
,
4495 (unsigned long long)parent
,
4496 (unsigned long long)root_objectid
,
4497 (unsigned long long)owner_objectid
,
4498 (unsigned long long)owner_offset
);
4501 leaf
= path
->nodes
[0];
4502 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
4503 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4504 if (item_size
< sizeof(*ei
)) {
4505 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
4506 ret
= convert_extent_item_v0(trans
, extent_root
, path
,
4510 btrfs_release_path(extent_root
, path
);
4511 path
->leave_spinning
= 1;
4513 key
.objectid
= bytenr
;
4514 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
4515 key
.offset
= num_bytes
;
4517 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
4520 printk(KERN_ERR
"umm, got %d back from search"
4521 ", was looking for %llu\n", ret
,
4522 (unsigned long long)bytenr
);
4523 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
4526 extent_slot
= path
->slots
[0];
4527 leaf
= path
->nodes
[0];
4528 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
4531 BUG_ON(item_size
< sizeof(*ei
));
4532 ei
= btrfs_item_ptr(leaf
, extent_slot
,
4533 struct btrfs_extent_item
);
4534 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
) {
4535 struct btrfs_tree_block_info
*bi
;
4536 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
4537 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
4538 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
4541 refs
= btrfs_extent_refs(leaf
, ei
);
4542 BUG_ON(refs
< refs_to_drop
);
4543 refs
-= refs_to_drop
;
4547 __run_delayed_extent_op(extent_op
, leaf
, ei
);
4549 * In the case of inline back ref, reference count will
4550 * be updated by remove_extent_backref
4553 BUG_ON(!found_extent
);
4555 btrfs_set_extent_refs(leaf
, ei
, refs
);
4556 btrfs_mark_buffer_dirty(leaf
);
4559 ret
= remove_extent_backref(trans
, extent_root
, path
,
4566 BUG_ON(is_data
&& refs_to_drop
!=
4567 extent_data_ref_count(root
, path
, iref
));
4569 BUG_ON(path
->slots
[0] != extent_slot
);
4571 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
4572 path
->slots
[0] = extent_slot
;
4577 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
4580 btrfs_release_path(extent_root
, path
);
4583 ret
= btrfs_del_csums(trans
, root
, bytenr
, num_bytes
);
4586 invalidate_mapping_pages(info
->btree_inode
->i_mapping
,
4587 bytenr
>> PAGE_CACHE_SHIFT
,
4588 (bytenr
+ num_bytes
- 1) >> PAGE_CACHE_SHIFT
);
4591 ret
= update_block_group(trans
, root
, bytenr
, num_bytes
, 0);
4594 btrfs_free_path(path
);
4599 * when we free an block, it is possible (and likely) that we free the last
4600 * delayed ref for that extent as well. This searches the delayed ref tree for
4601 * a given extent, and if there are no other delayed refs to be processed, it
4602 * removes it from the tree.
4604 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
4605 struct btrfs_root
*root
, u64 bytenr
)
4607 struct btrfs_delayed_ref_head
*head
;
4608 struct btrfs_delayed_ref_root
*delayed_refs
;
4609 struct btrfs_delayed_ref_node
*ref
;
4610 struct rb_node
*node
;
4613 delayed_refs
= &trans
->transaction
->delayed_refs
;
4614 spin_lock(&delayed_refs
->lock
);
4615 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
4619 node
= rb_prev(&head
->node
.rb_node
);
4623 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
4625 /* there are still entries for this ref, we can't drop it */
4626 if (ref
->bytenr
== bytenr
)
4629 if (head
->extent_op
) {
4630 if (!head
->must_insert_reserved
)
4632 kfree(head
->extent_op
);
4633 head
->extent_op
= NULL
;
4637 * waiting for the lock here would deadlock. If someone else has it
4638 * locked they are already in the process of dropping it anyway
4640 if (!mutex_trylock(&head
->mutex
))
4644 * at this point we have a head with no other entries. Go
4645 * ahead and process it.
4647 head
->node
.in_tree
= 0;
4648 rb_erase(&head
->node
.rb_node
, &delayed_refs
->root
);
4650 delayed_refs
->num_entries
--;
4653 * we don't take a ref on the node because we're removing it from the
4654 * tree, so we just steal the ref the tree was holding.
4656 delayed_refs
->num_heads
--;
4657 if (list_empty(&head
->cluster
))
4658 delayed_refs
->num_heads_ready
--;
4660 list_del_init(&head
->cluster
);
4661 spin_unlock(&delayed_refs
->lock
);
4663 BUG_ON(head
->extent_op
);
4664 if (head
->must_insert_reserved
)
4667 mutex_unlock(&head
->mutex
);
4668 btrfs_put_delayed_ref(&head
->node
);
4671 spin_unlock(&delayed_refs
->lock
);
4675 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
4676 struct btrfs_root
*root
,
4677 struct extent_buffer
*buf
,
4678 u64 parent
, int last_ref
)
4680 struct btrfs_block_rsv
*block_rsv
;
4681 struct btrfs_block_group_cache
*cache
= NULL
;
4684 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
4685 ret
= btrfs_add_delayed_tree_ref(trans
, buf
->start
, buf
->len
,
4686 parent
, root
->root_key
.objectid
,
4687 btrfs_header_level(buf
),
4688 BTRFS_DROP_DELAYED_REF
, NULL
);
4695 block_rsv
= get_block_rsv(trans
, root
);
4696 cache
= btrfs_lookup_block_group(root
->fs_info
, buf
->start
);
4697 if (block_rsv
->space_info
!= cache
->space_info
)
4700 if (btrfs_header_generation(buf
) == trans
->transid
) {
4701 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
4702 ret
= check_ref_cleanup(trans
, root
, buf
->start
);
4707 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
4708 pin_down_extent(root
, cache
, buf
->start
, buf
->len
, 1);
4712 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
4714 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
4715 ret
= btrfs_update_reserved_bytes(cache
, buf
->len
, 0, 0);
4716 if (ret
== -EAGAIN
) {
4717 /* block group became read-only */
4718 btrfs_update_reserved_bytes(cache
, buf
->len
, 0, 1);
4723 spin_lock(&block_rsv
->lock
);
4724 if (block_rsv
->reserved
< block_rsv
->size
) {
4725 block_rsv
->reserved
+= buf
->len
;
4728 spin_unlock(&block_rsv
->lock
);
4731 spin_lock(&cache
->space_info
->lock
);
4732 cache
->space_info
->bytes_reserved
-= buf
->len
;
4733 cache
->space_info
->reservation_progress
++;
4734 spin_unlock(&cache
->space_info
->lock
);
4739 if (block_rsv
->durable
&& !cache
->ro
) {
4741 spin_lock(&cache
->lock
);
4743 cache
->reserved_pinned
+= buf
->len
;
4746 spin_unlock(&cache
->lock
);
4749 spin_lock(&block_rsv
->lock
);
4750 block_rsv
->freed
[trans
->transid
& 0x1] += buf
->len
;
4751 spin_unlock(&block_rsv
->lock
);
4756 * Deleting the buffer, clear the corrupt flag since it doesn't matter
4759 clear_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
);
4760 btrfs_put_block_group(cache
);
4763 int btrfs_free_extent(struct btrfs_trans_handle
*trans
,
4764 struct btrfs_root
*root
,
4765 u64 bytenr
, u64 num_bytes
, u64 parent
,
4766 u64 root_objectid
, u64 owner
, u64 offset
)
4771 * tree log blocks never actually go into the extent allocation
4772 * tree, just update pinning info and exit early.
4774 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
4775 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
4776 /* unlocks the pinned mutex */
4777 btrfs_pin_extent(root
, bytenr
, num_bytes
, 1);
4779 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
4780 ret
= btrfs_add_delayed_tree_ref(trans
, bytenr
, num_bytes
,
4781 parent
, root_objectid
, (int)owner
,
4782 BTRFS_DROP_DELAYED_REF
, NULL
);
4785 ret
= btrfs_add_delayed_data_ref(trans
, bytenr
, num_bytes
,
4786 parent
, root_objectid
, owner
,
4787 offset
, BTRFS_DROP_DELAYED_REF
, NULL
);
4793 static u64
stripe_align(struct btrfs_root
*root
, u64 val
)
4795 u64 mask
= ((u64
)root
->stripesize
- 1);
4796 u64 ret
= (val
+ mask
) & ~mask
;
4801 * when we wait for progress in the block group caching, its because
4802 * our allocation attempt failed at least once. So, we must sleep
4803 * and let some progress happen before we try again.
4805 * This function will sleep at least once waiting for new free space to
4806 * show up, and then it will check the block group free space numbers
4807 * for our min num_bytes. Another option is to have it go ahead
4808 * and look in the rbtree for a free extent of a given size, but this
4812 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
4815 struct btrfs_caching_control
*caching_ctl
;
4818 caching_ctl
= get_caching_control(cache
);
4822 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
4823 (cache
->free_space
>= num_bytes
));
4825 put_caching_control(caching_ctl
);
4830 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
4832 struct btrfs_caching_control
*caching_ctl
;
4835 caching_ctl
= get_caching_control(cache
);
4839 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
4841 put_caching_control(caching_ctl
);
4845 static int get_block_group_index(struct btrfs_block_group_cache
*cache
)
4848 if (cache
->flags
& BTRFS_BLOCK_GROUP_RAID10
)
4850 else if (cache
->flags
& BTRFS_BLOCK_GROUP_RAID1
)
4852 else if (cache
->flags
& BTRFS_BLOCK_GROUP_DUP
)
4854 else if (cache
->flags
& BTRFS_BLOCK_GROUP_RAID0
)
4861 enum btrfs_loop_type
{
4862 LOOP_FIND_IDEAL
= 0,
4863 LOOP_CACHING_NOWAIT
= 1,
4864 LOOP_CACHING_WAIT
= 2,
4865 LOOP_ALLOC_CHUNK
= 3,
4866 LOOP_NO_EMPTY_SIZE
= 4,
4870 * walks the btree of allocated extents and find a hole of a given size.
4871 * The key ins is changed to record the hole:
4872 * ins->objectid == block start
4873 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4874 * ins->offset == number of blocks
4875 * Any available blocks before search_start are skipped.
4877 static noinline
int find_free_extent(struct btrfs_trans_handle
*trans
,
4878 struct btrfs_root
*orig_root
,
4879 u64 num_bytes
, u64 empty_size
,
4880 u64 search_start
, u64 search_end
,
4881 u64 hint_byte
, struct btrfs_key
*ins
,
4885 struct btrfs_root
*root
= orig_root
->fs_info
->extent_root
;
4886 struct btrfs_free_cluster
*last_ptr
= NULL
;
4887 struct btrfs_block_group_cache
*block_group
= NULL
;
4888 int empty_cluster
= 2 * 1024 * 1024;
4889 int allowed_chunk_alloc
= 0;
4890 int done_chunk_alloc
= 0;
4891 struct btrfs_space_info
*space_info
;
4892 int last_ptr_loop
= 0;
4895 bool found_uncached_bg
= false;
4896 bool failed_cluster_refill
= false;
4897 bool failed_alloc
= false;
4898 bool use_cluster
= true;
4899 u64 ideal_cache_percent
= 0;
4900 u64 ideal_cache_offset
= 0;
4902 WARN_ON(num_bytes
< root
->sectorsize
);
4903 btrfs_set_key_type(ins
, BTRFS_EXTENT_ITEM_KEY
);
4907 space_info
= __find_space_info(root
->fs_info
, data
);
4909 printk(KERN_ERR
"No space info for %d\n", data
);
4914 * If the space info is for both data and metadata it means we have a
4915 * small filesystem and we can't use the clustering stuff.
4917 if (btrfs_mixed_space_info(space_info
))
4918 use_cluster
= false;
4920 if (orig_root
->ref_cows
|| empty_size
)
4921 allowed_chunk_alloc
= 1;
4923 if (data
& BTRFS_BLOCK_GROUP_METADATA
&& use_cluster
) {
4924 last_ptr
= &root
->fs_info
->meta_alloc_cluster
;
4925 if (!btrfs_test_opt(root
, SSD
))
4926 empty_cluster
= 64 * 1024;
4929 if ((data
& BTRFS_BLOCK_GROUP_DATA
) && use_cluster
&&
4930 btrfs_test_opt(root
, SSD
)) {
4931 last_ptr
= &root
->fs_info
->data_alloc_cluster
;
4935 spin_lock(&last_ptr
->lock
);
4936 if (last_ptr
->block_group
)
4937 hint_byte
= last_ptr
->window_start
;
4938 spin_unlock(&last_ptr
->lock
);
4941 search_start
= max(search_start
, first_logical_byte(root
, 0));
4942 search_start
= max(search_start
, hint_byte
);
4947 if (search_start
== hint_byte
) {
4949 block_group
= btrfs_lookup_block_group(root
->fs_info
,
4952 * we don't want to use the block group if it doesn't match our
4953 * allocation bits, or if its not cached.
4955 * However if we are re-searching with an ideal block group
4956 * picked out then we don't care that the block group is cached.
4958 if (block_group
&& block_group_bits(block_group
, data
) &&
4959 (block_group
->cached
!= BTRFS_CACHE_NO
||
4960 search_start
== ideal_cache_offset
)) {
4961 down_read(&space_info
->groups_sem
);
4962 if (list_empty(&block_group
->list
) ||
4965 * someone is removing this block group,
4966 * we can't jump into the have_block_group
4967 * target because our list pointers are not
4970 btrfs_put_block_group(block_group
);
4971 up_read(&space_info
->groups_sem
);
4973 index
= get_block_group_index(block_group
);
4974 goto have_block_group
;
4976 } else if (block_group
) {
4977 btrfs_put_block_group(block_group
);
4981 down_read(&space_info
->groups_sem
);
4982 list_for_each_entry(block_group
, &space_info
->block_groups
[index
],
4987 btrfs_get_block_group(block_group
);
4988 search_start
= block_group
->key
.objectid
;
4991 * this can happen if we end up cycling through all the
4992 * raid types, but we want to make sure we only allocate
4993 * for the proper type.
4995 if (!block_group_bits(block_group
, data
)) {
4996 u64 extra
= BTRFS_BLOCK_GROUP_DUP
|
4997 BTRFS_BLOCK_GROUP_RAID1
|
4998 BTRFS_BLOCK_GROUP_RAID10
;
5001 * if they asked for extra copies and this block group
5002 * doesn't provide them, bail. This does allow us to
5003 * fill raid0 from raid1.
5005 if ((data
& extra
) && !(block_group
->flags
& extra
))
5010 if (unlikely(block_group
->cached
== BTRFS_CACHE_NO
)) {
5013 ret
= cache_block_group(block_group
, trans
,
5015 if (block_group
->cached
== BTRFS_CACHE_FINISHED
)
5016 goto have_block_group
;
5018 free_percent
= btrfs_block_group_used(&block_group
->item
);
5019 free_percent
*= 100;
5020 free_percent
= div64_u64(free_percent
,
5021 block_group
->key
.offset
);
5022 free_percent
= 100 - free_percent
;
5023 if (free_percent
> ideal_cache_percent
&&
5024 likely(!block_group
->ro
)) {
5025 ideal_cache_offset
= block_group
->key
.objectid
;
5026 ideal_cache_percent
= free_percent
;
5030 * We only want to start kthread caching if we are at
5031 * the point where we will wait for caching to make
5032 * progress, or if our ideal search is over and we've
5033 * found somebody to start caching.
5035 if (loop
> LOOP_CACHING_NOWAIT
||
5036 (loop
> LOOP_FIND_IDEAL
&&
5037 atomic_read(&space_info
->caching_threads
) < 2)) {
5038 ret
= cache_block_group(block_group
, trans
,
5042 found_uncached_bg
= true;
5045 * If loop is set for cached only, try the next block
5048 if (loop
== LOOP_FIND_IDEAL
)
5052 cached
= block_group_cache_done(block_group
);
5053 if (unlikely(!cached
))
5054 found_uncached_bg
= true;
5056 if (unlikely(block_group
->ro
))
5060 * Ok we want to try and use the cluster allocator, so lets look
5061 * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
5062 * have tried the cluster allocator plenty of times at this
5063 * point and not have found anything, so we are likely way too
5064 * fragmented for the clustering stuff to find anything, so lets
5065 * just skip it and let the allocator find whatever block it can
5068 if (last_ptr
&& loop
< LOOP_NO_EMPTY_SIZE
) {
5070 * the refill lock keeps out other
5071 * people trying to start a new cluster
5073 spin_lock(&last_ptr
->refill_lock
);
5074 if (last_ptr
->block_group
&&
5075 (last_ptr
->block_group
->ro
||
5076 !block_group_bits(last_ptr
->block_group
, data
))) {
5078 goto refill_cluster
;
5081 offset
= btrfs_alloc_from_cluster(block_group
, last_ptr
,
5082 num_bytes
, search_start
);
5084 /* we have a block, we're done */
5085 spin_unlock(&last_ptr
->refill_lock
);
5089 spin_lock(&last_ptr
->lock
);
5091 * whoops, this cluster doesn't actually point to
5092 * this block group. Get a ref on the block
5093 * group is does point to and try again
5095 if (!last_ptr_loop
&& last_ptr
->block_group
&&
5096 last_ptr
->block_group
!= block_group
) {
5098 btrfs_put_block_group(block_group
);
5099 block_group
= last_ptr
->block_group
;
5100 btrfs_get_block_group(block_group
);
5101 spin_unlock(&last_ptr
->lock
);
5102 spin_unlock(&last_ptr
->refill_lock
);
5105 search_start
= block_group
->key
.objectid
;
5107 * we know this block group is properly
5108 * in the list because
5109 * btrfs_remove_block_group, drops the
5110 * cluster before it removes the block
5111 * group from the list
5113 goto have_block_group
;
5115 spin_unlock(&last_ptr
->lock
);
5118 * this cluster didn't work out, free it and
5121 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
5125 /* allocate a cluster in this block group */
5126 ret
= btrfs_find_space_cluster(trans
, root
,
5127 block_group
, last_ptr
,
5129 empty_cluster
+ empty_size
);
5132 * now pull our allocation out of this
5135 offset
= btrfs_alloc_from_cluster(block_group
,
5136 last_ptr
, num_bytes
,
5139 /* we found one, proceed */
5140 spin_unlock(&last_ptr
->refill_lock
);
5143 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
5144 && !failed_cluster_refill
) {
5145 spin_unlock(&last_ptr
->refill_lock
);
5147 failed_cluster_refill
= true;
5148 wait_block_group_cache_progress(block_group
,
5149 num_bytes
+ empty_cluster
+ empty_size
);
5150 goto have_block_group
;
5154 * at this point we either didn't find a cluster
5155 * or we weren't able to allocate a block from our
5156 * cluster. Free the cluster we've been trying
5157 * to use, and go to the next block group
5159 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
5160 spin_unlock(&last_ptr
->refill_lock
);
5164 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
5165 num_bytes
, empty_size
);
5167 * If we didn't find a chunk, and we haven't failed on this
5168 * block group before, and this block group is in the middle of
5169 * caching and we are ok with waiting, then go ahead and wait
5170 * for progress to be made, and set failed_alloc to true.
5172 * If failed_alloc is true then we've already waited on this
5173 * block group once and should move on to the next block group.
5175 if (!offset
&& !failed_alloc
&& !cached
&&
5176 loop
> LOOP_CACHING_NOWAIT
) {
5177 wait_block_group_cache_progress(block_group
,
5178 num_bytes
+ empty_size
);
5179 failed_alloc
= true;
5180 goto have_block_group
;
5181 } else if (!offset
) {
5185 search_start
= stripe_align(root
, offset
);
5186 /* move on to the next group */
5187 if (search_start
+ num_bytes
>= search_end
) {
5188 btrfs_add_free_space(block_group
, offset
, num_bytes
);
5192 /* move on to the next group */
5193 if (search_start
+ num_bytes
>
5194 block_group
->key
.objectid
+ block_group
->key
.offset
) {
5195 btrfs_add_free_space(block_group
, offset
, num_bytes
);
5199 ins
->objectid
= search_start
;
5200 ins
->offset
= num_bytes
;
5202 if (offset
< search_start
)
5203 btrfs_add_free_space(block_group
, offset
,
5204 search_start
- offset
);
5205 BUG_ON(offset
> search_start
);
5207 ret
= btrfs_update_reserved_bytes(block_group
, num_bytes
, 1,
5208 (data
& BTRFS_BLOCK_GROUP_DATA
));
5209 if (ret
== -EAGAIN
) {
5210 btrfs_add_free_space(block_group
, offset
, num_bytes
);
5214 /* we are all good, lets return */
5215 ins
->objectid
= search_start
;
5216 ins
->offset
= num_bytes
;
5218 if (offset
< search_start
)
5219 btrfs_add_free_space(block_group
, offset
,
5220 search_start
- offset
);
5221 BUG_ON(offset
> search_start
);
5224 failed_cluster_refill
= false;
5225 failed_alloc
= false;
5226 BUG_ON(index
!= get_block_group_index(block_group
));
5227 btrfs_put_block_group(block_group
);
5229 up_read(&space_info
->groups_sem
);
5231 if (!ins
->objectid
&& ++index
< BTRFS_NR_RAID_TYPES
)
5234 /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5235 * for them to make caching progress. Also
5236 * determine the best possible bg to cache
5237 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5238 * caching kthreads as we move along
5239 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5240 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5241 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5244 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
&&
5245 (found_uncached_bg
|| empty_size
|| empty_cluster
||
5246 allowed_chunk_alloc
)) {
5248 if (loop
== LOOP_FIND_IDEAL
&& found_uncached_bg
) {
5249 found_uncached_bg
= false;
5251 if (!ideal_cache_percent
&&
5252 atomic_read(&space_info
->caching_threads
))
5256 * 1 of the following 2 things have happened so far
5258 * 1) We found an ideal block group for caching that
5259 * is mostly full and will cache quickly, so we might
5260 * as well wait for it.
5262 * 2) We searched for cached only and we didn't find
5263 * anything, and we didn't start any caching kthreads
5264 * either, so chances are we will loop through and
5265 * start a couple caching kthreads, and then come back
5266 * around and just wait for them. This will be slower
5267 * because we will have 2 caching kthreads reading at
5268 * the same time when we could have just started one
5269 * and waited for it to get far enough to give us an
5270 * allocation, so go ahead and go to the wait caching
5273 loop
= LOOP_CACHING_WAIT
;
5274 search_start
= ideal_cache_offset
;
5275 ideal_cache_percent
= 0;
5277 } else if (loop
== LOOP_FIND_IDEAL
) {
5279 * Didn't find a uncached bg, wait on anything we find
5282 loop
= LOOP_CACHING_WAIT
;
5286 if (loop
< LOOP_CACHING_WAIT
) {
5291 if (loop
== LOOP_ALLOC_CHUNK
) {
5296 if (allowed_chunk_alloc
) {
5297 ret
= do_chunk_alloc(trans
, root
, num_bytes
+
5298 2 * 1024 * 1024, data
, 1);
5299 allowed_chunk_alloc
= 0;
5300 done_chunk_alloc
= 1;
5301 } else if (!done_chunk_alloc
) {
5302 space_info
->force_alloc
= 1;
5305 if (loop
< LOOP_NO_EMPTY_SIZE
) {
5310 } else if (!ins
->objectid
) {
5314 /* we found what we needed */
5315 if (ins
->objectid
) {
5316 if (!(data
& BTRFS_BLOCK_GROUP_DATA
))
5317 trans
->block_group
= block_group
->key
.objectid
;
5319 btrfs_put_block_group(block_group
);
5326 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
5327 int dump_block_groups
)
5329 struct btrfs_block_group_cache
*cache
;
5332 spin_lock(&info
->lock
);
5333 printk(KERN_INFO
"space_info has %llu free, is %sfull\n",
5334 (unsigned long long)(info
->total_bytes
- info
->bytes_used
-
5335 info
->bytes_pinned
- info
->bytes_reserved
-
5336 info
->bytes_readonly
),
5337 (info
->full
) ? "" : "not ");
5338 printk(KERN_INFO
"space_info total=%llu, used=%llu, pinned=%llu, "
5339 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5340 (unsigned long long)info
->total_bytes
,
5341 (unsigned long long)info
->bytes_used
,
5342 (unsigned long long)info
->bytes_pinned
,
5343 (unsigned long long)info
->bytes_reserved
,
5344 (unsigned long long)info
->bytes_may_use
,
5345 (unsigned long long)info
->bytes_readonly
);
5346 spin_unlock(&info
->lock
);
5348 if (!dump_block_groups
)
5351 down_read(&info
->groups_sem
);
5353 list_for_each_entry(cache
, &info
->block_groups
[index
], list
) {
5354 spin_lock(&cache
->lock
);
5355 printk(KERN_INFO
"block group %llu has %llu bytes, %llu used "
5356 "%llu pinned %llu reserved\n",
5357 (unsigned long long)cache
->key
.objectid
,
5358 (unsigned long long)cache
->key
.offset
,
5359 (unsigned long long)btrfs_block_group_used(&cache
->item
),
5360 (unsigned long long)cache
->pinned
,
5361 (unsigned long long)cache
->reserved
);
5362 btrfs_dump_free_space(cache
, bytes
);
5363 spin_unlock(&cache
->lock
);
5365 if (++index
< BTRFS_NR_RAID_TYPES
)
5367 up_read(&info
->groups_sem
);
5370 int btrfs_reserve_extent(struct btrfs_trans_handle
*trans
,
5371 struct btrfs_root
*root
,
5372 u64 num_bytes
, u64 min_alloc_size
,
5373 u64 empty_size
, u64 hint_byte
,
5374 u64 search_end
, struct btrfs_key
*ins
,
5378 u64 search_start
= 0;
5380 data
= btrfs_get_alloc_profile(root
, data
);
5383 * the only place that sets empty_size is btrfs_realloc_node, which
5384 * is not called recursively on allocations
5386 if (empty_size
|| root
->ref_cows
)
5387 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
5388 num_bytes
+ 2 * 1024 * 1024, data
, 0);
5390 WARN_ON(num_bytes
< root
->sectorsize
);
5391 ret
= find_free_extent(trans
, root
, num_bytes
, empty_size
,
5392 search_start
, search_end
, hint_byte
,
5395 if (ret
== -ENOSPC
&& num_bytes
> min_alloc_size
) {
5396 num_bytes
= num_bytes
>> 1;
5397 num_bytes
= num_bytes
& ~(root
->sectorsize
- 1);
5398 num_bytes
= max(num_bytes
, min_alloc_size
);
5399 do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
5400 num_bytes
, data
, 1);
5403 if (ret
== -ENOSPC
&& btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
5404 struct btrfs_space_info
*sinfo
;
5406 sinfo
= __find_space_info(root
->fs_info
, data
);
5407 printk(KERN_ERR
"btrfs allocation failed flags %llu, "
5408 "wanted %llu\n", (unsigned long long)data
,
5409 (unsigned long long)num_bytes
);
5410 dump_space_info(sinfo
, num_bytes
, 1);
5413 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, ins
->offset
);
5418 int btrfs_free_reserved_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
5420 struct btrfs_block_group_cache
*cache
;
5423 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
5425 printk(KERN_ERR
"Unable to find block group for %llu\n",
5426 (unsigned long long)start
);
5430 ret
= btrfs_discard_extent(root
, start
, len
);
5432 btrfs_add_free_space(cache
, start
, len
);
5433 btrfs_update_reserved_bytes(cache
, len
, 0, 1);
5434 btrfs_put_block_group(cache
);
5436 trace_btrfs_reserved_extent_free(root
, start
, len
);
5441 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
5442 struct btrfs_root
*root
,
5443 u64 parent
, u64 root_objectid
,
5444 u64 flags
, u64 owner
, u64 offset
,
5445 struct btrfs_key
*ins
, int ref_mod
)
5448 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5449 struct btrfs_extent_item
*extent_item
;
5450 struct btrfs_extent_inline_ref
*iref
;
5451 struct btrfs_path
*path
;
5452 struct extent_buffer
*leaf
;
5457 type
= BTRFS_SHARED_DATA_REF_KEY
;
5459 type
= BTRFS_EXTENT_DATA_REF_KEY
;
5461 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
5463 path
= btrfs_alloc_path();
5467 path
->leave_spinning
= 1;
5468 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
5472 leaf
= path
->nodes
[0];
5473 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
5474 struct btrfs_extent_item
);
5475 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
5476 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
5477 btrfs_set_extent_flags(leaf
, extent_item
,
5478 flags
| BTRFS_EXTENT_FLAG_DATA
);
5480 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
5481 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
5483 struct btrfs_shared_data_ref
*ref
;
5484 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
5485 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
5486 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
5488 struct btrfs_extent_data_ref
*ref
;
5489 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
5490 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
5491 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
5492 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
5493 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
5496 btrfs_mark_buffer_dirty(path
->nodes
[0]);
5497 btrfs_free_path(path
);
5499 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1);
5501 printk(KERN_ERR
"btrfs update block group failed for %llu "
5502 "%llu\n", (unsigned long long)ins
->objectid
,
5503 (unsigned long long)ins
->offset
);
5509 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
5510 struct btrfs_root
*root
,
5511 u64 parent
, u64 root_objectid
,
5512 u64 flags
, struct btrfs_disk_key
*key
,
5513 int level
, struct btrfs_key
*ins
)
5516 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5517 struct btrfs_extent_item
*extent_item
;
5518 struct btrfs_tree_block_info
*block_info
;
5519 struct btrfs_extent_inline_ref
*iref
;
5520 struct btrfs_path
*path
;
5521 struct extent_buffer
*leaf
;
5522 u32 size
= sizeof(*extent_item
) + sizeof(*block_info
) + sizeof(*iref
);
5524 path
= btrfs_alloc_path();
5527 path
->leave_spinning
= 1;
5528 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
5532 leaf
= path
->nodes
[0];
5533 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
5534 struct btrfs_extent_item
);
5535 btrfs_set_extent_refs(leaf
, extent_item
, 1);
5536 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
5537 btrfs_set_extent_flags(leaf
, extent_item
,
5538 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
5539 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
5541 btrfs_set_tree_block_key(leaf
, block_info
, key
);
5542 btrfs_set_tree_block_level(leaf
, block_info
, level
);
5544 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
5546 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
5547 btrfs_set_extent_inline_ref_type(leaf
, iref
,
5548 BTRFS_SHARED_BLOCK_REF_KEY
);
5549 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
5551 btrfs_set_extent_inline_ref_type(leaf
, iref
,
5552 BTRFS_TREE_BLOCK_REF_KEY
);
5553 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
5556 btrfs_mark_buffer_dirty(leaf
);
5557 btrfs_free_path(path
);
5559 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1);
5561 printk(KERN_ERR
"btrfs update block group failed for %llu "
5562 "%llu\n", (unsigned long long)ins
->objectid
,
5563 (unsigned long long)ins
->offset
);
5569 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
5570 struct btrfs_root
*root
,
5571 u64 root_objectid
, u64 owner
,
5572 u64 offset
, struct btrfs_key
*ins
)
5576 BUG_ON(root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
5578 ret
= btrfs_add_delayed_data_ref(trans
, ins
->objectid
, ins
->offset
,
5579 0, root_objectid
, owner
, offset
,
5580 BTRFS_ADD_DELAYED_EXTENT
, NULL
);
5585 * this is used by the tree logging recovery code. It records that
5586 * an extent has been allocated and makes sure to clear the free
5587 * space cache bits as well
5589 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
5590 struct btrfs_root
*root
,
5591 u64 root_objectid
, u64 owner
, u64 offset
,
5592 struct btrfs_key
*ins
)
5595 struct btrfs_block_group_cache
*block_group
;
5596 struct btrfs_caching_control
*caching_ctl
;
5597 u64 start
= ins
->objectid
;
5598 u64 num_bytes
= ins
->offset
;
5600 block_group
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
5601 cache_block_group(block_group
, trans
, NULL
, 0);
5602 caching_ctl
= get_caching_control(block_group
);
5605 BUG_ON(!block_group_cache_done(block_group
));
5606 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
5609 mutex_lock(&caching_ctl
->mutex
);
5611 if (start
>= caching_ctl
->progress
) {
5612 ret
= add_excluded_extent(root
, start
, num_bytes
);
5614 } else if (start
+ num_bytes
<= caching_ctl
->progress
) {
5615 ret
= btrfs_remove_free_space(block_group
,
5619 num_bytes
= caching_ctl
->progress
- start
;
5620 ret
= btrfs_remove_free_space(block_group
,
5624 start
= caching_ctl
->progress
;
5625 num_bytes
= ins
->objectid
+ ins
->offset
-
5626 caching_ctl
->progress
;
5627 ret
= add_excluded_extent(root
, start
, num_bytes
);
5631 mutex_unlock(&caching_ctl
->mutex
);
5632 put_caching_control(caching_ctl
);
5635 ret
= btrfs_update_reserved_bytes(block_group
, ins
->offset
, 1, 1);
5637 btrfs_put_block_group(block_group
);
5638 ret
= alloc_reserved_file_extent(trans
, root
, 0, root_objectid
,
5639 0, owner
, offset
, ins
, 1);
5643 struct extent_buffer
*btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
,
5644 struct btrfs_root
*root
,
5645 u64 bytenr
, u32 blocksize
,
5648 struct extent_buffer
*buf
;
5650 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
5652 return ERR_PTR(-ENOMEM
);
5653 btrfs_set_header_generation(buf
, trans
->transid
);
5654 btrfs_set_buffer_lockdep_class(buf
, level
);
5655 btrfs_tree_lock(buf
);
5656 clean_tree_block(trans
, root
, buf
);
5658 btrfs_set_lock_blocking(buf
);
5659 btrfs_set_buffer_uptodate(buf
);
5661 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
5663 * we allow two log transactions at a time, use different
5664 * EXENT bit to differentiate dirty pages.
5666 if (root
->log_transid
% 2 == 0)
5667 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
5668 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
5670 set_extent_new(&root
->dirty_log_pages
, buf
->start
,
5671 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
5673 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
5674 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
5676 trans
->blocks_used
++;
5677 /* this returns a buffer locked for blocking */
5681 static struct btrfs_block_rsv
*
5682 use_block_rsv(struct btrfs_trans_handle
*trans
,
5683 struct btrfs_root
*root
, u32 blocksize
)
5685 struct btrfs_block_rsv
*block_rsv
;
5686 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
5689 block_rsv
= get_block_rsv(trans
, root
);
5691 if (block_rsv
->size
== 0) {
5692 ret
= reserve_metadata_bytes(trans
, root
, block_rsv
,
5695 * If we couldn't reserve metadata bytes try and use some from
5696 * the global reserve.
5698 if (ret
&& block_rsv
!= global_rsv
) {
5699 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
5702 return ERR_PTR(ret
);
5704 return ERR_PTR(ret
);
5709 ret
= block_rsv_use_bytes(block_rsv
, blocksize
);
5714 ret
= reserve_metadata_bytes(trans
, root
, block_rsv
, blocksize
,
5717 spin_lock(&block_rsv
->lock
);
5718 block_rsv
->size
+= blocksize
;
5719 spin_unlock(&block_rsv
->lock
);
5721 } else if (ret
&& block_rsv
!= global_rsv
) {
5722 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
5728 return ERR_PTR(-ENOSPC
);
5731 static void unuse_block_rsv(struct btrfs_block_rsv
*block_rsv
, u32 blocksize
)
5733 block_rsv_add_bytes(block_rsv
, blocksize
, 0);
5734 block_rsv_release_bytes(block_rsv
, NULL
, 0);
5738 * finds a free extent and does all the dirty work required for allocation
5739 * returns the key for the extent through ins, and a tree buffer for
5740 * the first block of the extent through buf.
5742 * returns the tree buffer or NULL.
5744 struct extent_buffer
*btrfs_alloc_free_block(struct btrfs_trans_handle
*trans
,
5745 struct btrfs_root
*root
, u32 blocksize
,
5746 u64 parent
, u64 root_objectid
,
5747 struct btrfs_disk_key
*key
, int level
,
5748 u64 hint
, u64 empty_size
)
5750 struct btrfs_key ins
;
5751 struct btrfs_block_rsv
*block_rsv
;
5752 struct extent_buffer
*buf
;
5757 block_rsv
= use_block_rsv(trans
, root
, blocksize
);
5758 if (IS_ERR(block_rsv
))
5759 return ERR_CAST(block_rsv
);
5761 ret
= btrfs_reserve_extent(trans
, root
, blocksize
, blocksize
,
5762 empty_size
, hint
, (u64
)-1, &ins
, 0);
5764 unuse_block_rsv(block_rsv
, blocksize
);
5765 return ERR_PTR(ret
);
5768 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
,
5770 BUG_ON(IS_ERR(buf
));
5772 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
5774 parent
= ins
.objectid
;
5775 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
5779 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
5780 struct btrfs_delayed_extent_op
*extent_op
;
5781 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
5784 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
5786 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
5787 extent_op
->flags_to_set
= flags
;
5788 extent_op
->update_key
= 1;
5789 extent_op
->update_flags
= 1;
5790 extent_op
->is_data
= 0;
5792 ret
= btrfs_add_delayed_tree_ref(trans
, ins
.objectid
,
5793 ins
.offset
, parent
, root_objectid
,
5794 level
, BTRFS_ADD_DELAYED_EXTENT
,
5801 struct walk_control
{
5802 u64 refs
[BTRFS_MAX_LEVEL
];
5803 u64 flags
[BTRFS_MAX_LEVEL
];
5804 struct btrfs_key update_progress
;
5814 #define DROP_REFERENCE 1
5815 #define UPDATE_BACKREF 2
5817 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
5818 struct btrfs_root
*root
,
5819 struct walk_control
*wc
,
5820 struct btrfs_path
*path
)
5828 struct btrfs_key key
;
5829 struct extent_buffer
*eb
;
5834 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
5835 wc
->reada_count
= wc
->reada_count
* 2 / 3;
5836 wc
->reada_count
= max(wc
->reada_count
, 2);
5838 wc
->reada_count
= wc
->reada_count
* 3 / 2;
5839 wc
->reada_count
= min_t(int, wc
->reada_count
,
5840 BTRFS_NODEPTRS_PER_BLOCK(root
));
5843 eb
= path
->nodes
[wc
->level
];
5844 nritems
= btrfs_header_nritems(eb
);
5845 blocksize
= btrfs_level_size(root
, wc
->level
- 1);
5847 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
5848 if (nread
>= wc
->reada_count
)
5852 bytenr
= btrfs_node_blockptr(eb
, slot
);
5853 generation
= btrfs_node_ptr_generation(eb
, slot
);
5855 if (slot
== path
->slots
[wc
->level
])
5858 if (wc
->stage
== UPDATE_BACKREF
&&
5859 generation
<= root
->root_key
.offset
)
5862 /* We don't lock the tree block, it's OK to be racy here */
5863 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, blocksize
,
5868 if (wc
->stage
== DROP_REFERENCE
) {
5872 if (wc
->level
== 1 &&
5873 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
5875 if (!wc
->update_ref
||
5876 generation
<= root
->root_key
.offset
)
5878 btrfs_node_key_to_cpu(eb
, &key
, slot
);
5879 ret
= btrfs_comp_cpu_keys(&key
,
5880 &wc
->update_progress
);
5884 if (wc
->level
== 1 &&
5885 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
5889 ret
= readahead_tree_block(root
, bytenr
, blocksize
,
5895 wc
->reada_slot
= slot
;
5899 * hepler to process tree block while walking down the tree.
5901 * when wc->stage == UPDATE_BACKREF, this function updates
5902 * back refs for pointers in the block.
5904 * NOTE: return value 1 means we should stop walking down.
5906 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
5907 struct btrfs_root
*root
,
5908 struct btrfs_path
*path
,
5909 struct walk_control
*wc
, int lookup_info
)
5911 int level
= wc
->level
;
5912 struct extent_buffer
*eb
= path
->nodes
[level
];
5913 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
5916 if (wc
->stage
== UPDATE_BACKREF
&&
5917 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
5921 * when reference count of tree block is 1, it won't increase
5922 * again. once full backref flag is set, we never clear it.
5925 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
5926 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
5927 BUG_ON(!path
->locks
[level
]);
5928 ret
= btrfs_lookup_extent_info(trans
, root
,
5933 BUG_ON(wc
->refs
[level
] == 0);
5936 if (wc
->stage
== DROP_REFERENCE
) {
5937 if (wc
->refs
[level
] > 1)
5940 if (path
->locks
[level
] && !wc
->keep_locks
) {
5941 btrfs_tree_unlock(eb
);
5942 path
->locks
[level
] = 0;
5947 /* wc->stage == UPDATE_BACKREF */
5948 if (!(wc
->flags
[level
] & flag
)) {
5949 BUG_ON(!path
->locks
[level
]);
5950 ret
= btrfs_inc_ref(trans
, root
, eb
, 1);
5952 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
5954 ret
= btrfs_set_disk_extent_flags(trans
, root
, eb
->start
,
5957 wc
->flags
[level
] |= flag
;
5961 * the block is shared by multiple trees, so it's not good to
5962 * keep the tree lock
5964 if (path
->locks
[level
] && level
> 0) {
5965 btrfs_tree_unlock(eb
);
5966 path
->locks
[level
] = 0;
5972 * hepler to process tree block pointer.
5974 * when wc->stage == DROP_REFERENCE, this function checks
5975 * reference count of the block pointed to. if the block
5976 * is shared and we need update back refs for the subtree
5977 * rooted at the block, this function changes wc->stage to
5978 * UPDATE_BACKREF. if the block is shared and there is no
5979 * need to update back, this function drops the reference
5982 * NOTE: return value 1 means we should stop walking down.
5984 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
5985 struct btrfs_root
*root
,
5986 struct btrfs_path
*path
,
5987 struct walk_control
*wc
, int *lookup_info
)
5993 struct btrfs_key key
;
5994 struct extent_buffer
*next
;
5995 int level
= wc
->level
;
5999 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
6000 path
->slots
[level
]);
6002 * if the lower level block was created before the snapshot
6003 * was created, we know there is no need to update back refs
6006 if (wc
->stage
== UPDATE_BACKREF
&&
6007 generation
<= root
->root_key
.offset
) {
6012 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
6013 blocksize
= btrfs_level_size(root
, level
- 1);
6015 next
= btrfs_find_tree_block(root
, bytenr
, blocksize
);
6017 next
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
6022 btrfs_tree_lock(next
);
6023 btrfs_set_lock_blocking(next
);
6025 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, blocksize
,
6026 &wc
->refs
[level
- 1],
6027 &wc
->flags
[level
- 1]);
6029 BUG_ON(wc
->refs
[level
- 1] == 0);
6032 if (wc
->stage
== DROP_REFERENCE
) {
6033 if (wc
->refs
[level
- 1] > 1) {
6035 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
6038 if (!wc
->update_ref
||
6039 generation
<= root
->root_key
.offset
)
6042 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
6043 path
->slots
[level
]);
6044 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
6048 wc
->stage
= UPDATE_BACKREF
;
6049 wc
->shared_level
= level
- 1;
6053 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
6057 if (!btrfs_buffer_uptodate(next
, generation
)) {
6058 btrfs_tree_unlock(next
);
6059 free_extent_buffer(next
);
6065 if (reada
&& level
== 1)
6066 reada_walk_down(trans
, root
, wc
, path
);
6067 next
= read_tree_block(root
, bytenr
, blocksize
, generation
);
6070 btrfs_tree_lock(next
);
6071 btrfs_set_lock_blocking(next
);
6075 BUG_ON(level
!= btrfs_header_level(next
));
6076 path
->nodes
[level
] = next
;
6077 path
->slots
[level
] = 0;
6078 path
->locks
[level
] = 1;
6084 wc
->refs
[level
- 1] = 0;
6085 wc
->flags
[level
- 1] = 0;
6086 if (wc
->stage
== DROP_REFERENCE
) {
6087 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
6088 parent
= path
->nodes
[level
]->start
;
6090 BUG_ON(root
->root_key
.objectid
!=
6091 btrfs_header_owner(path
->nodes
[level
]));
6095 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
, parent
,
6096 root
->root_key
.objectid
, level
- 1, 0);
6099 btrfs_tree_unlock(next
);
6100 free_extent_buffer(next
);
6106 * hepler to process tree block while walking up the tree.
6108 * when wc->stage == DROP_REFERENCE, this function drops
6109 * reference count on the block.
6111 * when wc->stage == UPDATE_BACKREF, this function changes
6112 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6113 * to UPDATE_BACKREF previously while processing the block.
6115 * NOTE: return value 1 means we should stop walking up.
6117 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
6118 struct btrfs_root
*root
,
6119 struct btrfs_path
*path
,
6120 struct walk_control
*wc
)
6123 int level
= wc
->level
;
6124 struct extent_buffer
*eb
= path
->nodes
[level
];
6127 if (wc
->stage
== UPDATE_BACKREF
) {
6128 BUG_ON(wc
->shared_level
< level
);
6129 if (level
< wc
->shared_level
)
6132 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
6136 wc
->stage
= DROP_REFERENCE
;
6137 wc
->shared_level
= -1;
6138 path
->slots
[level
] = 0;
6141 * check reference count again if the block isn't locked.
6142 * we should start walking down the tree again if reference
6145 if (!path
->locks
[level
]) {
6147 btrfs_tree_lock(eb
);
6148 btrfs_set_lock_blocking(eb
);
6149 path
->locks
[level
] = 1;
6151 ret
= btrfs_lookup_extent_info(trans
, root
,
6156 BUG_ON(wc
->refs
[level
] == 0);
6157 if (wc
->refs
[level
] == 1) {
6158 btrfs_tree_unlock(eb
);
6159 path
->locks
[level
] = 0;
6165 /* wc->stage == DROP_REFERENCE */
6166 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
6168 if (wc
->refs
[level
] == 1) {
6170 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
6171 ret
= btrfs_dec_ref(trans
, root
, eb
, 1);
6173 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
6176 /* make block locked assertion in clean_tree_block happy */
6177 if (!path
->locks
[level
] &&
6178 btrfs_header_generation(eb
) == trans
->transid
) {
6179 btrfs_tree_lock(eb
);
6180 btrfs_set_lock_blocking(eb
);
6181 path
->locks
[level
] = 1;
6183 clean_tree_block(trans
, root
, eb
);
6186 if (eb
== root
->node
) {
6187 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
6190 BUG_ON(root
->root_key
.objectid
!=
6191 btrfs_header_owner(eb
));
6193 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
6194 parent
= path
->nodes
[level
+ 1]->start
;
6196 BUG_ON(root
->root_key
.objectid
!=
6197 btrfs_header_owner(path
->nodes
[level
+ 1]));
6200 btrfs_free_tree_block(trans
, root
, eb
, parent
, wc
->refs
[level
] == 1);
6202 wc
->refs
[level
] = 0;
6203 wc
->flags
[level
] = 0;
6207 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
6208 struct btrfs_root
*root
,
6209 struct btrfs_path
*path
,
6210 struct walk_control
*wc
)
6212 int level
= wc
->level
;
6213 int lookup_info
= 1;
6216 while (level
>= 0) {
6217 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
6224 if (path
->slots
[level
] >=
6225 btrfs_header_nritems(path
->nodes
[level
]))
6228 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
6230 path
->slots
[level
]++;
6239 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
6240 struct btrfs_root
*root
,
6241 struct btrfs_path
*path
,
6242 struct walk_control
*wc
, int max_level
)
6244 int level
= wc
->level
;
6247 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
6248 while (level
< max_level
&& path
->nodes
[level
]) {
6250 if (path
->slots
[level
] + 1 <
6251 btrfs_header_nritems(path
->nodes
[level
])) {
6252 path
->slots
[level
]++;
6255 ret
= walk_up_proc(trans
, root
, path
, wc
);
6259 if (path
->locks
[level
]) {
6260 btrfs_tree_unlock(path
->nodes
[level
]);
6261 path
->locks
[level
] = 0;
6263 free_extent_buffer(path
->nodes
[level
]);
6264 path
->nodes
[level
] = NULL
;
6272 * drop a subvolume tree.
6274 * this function traverses the tree freeing any blocks that only
6275 * referenced by the tree.
6277 * when a shared tree block is found. this function decreases its
6278 * reference count by one. if update_ref is true, this function
6279 * also make sure backrefs for the shared block and all lower level
6280 * blocks are properly updated.
6282 int btrfs_drop_snapshot(struct btrfs_root
*root
,
6283 struct btrfs_block_rsv
*block_rsv
, int update_ref
)
6285 struct btrfs_path
*path
;
6286 struct btrfs_trans_handle
*trans
;
6287 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
6288 struct btrfs_root_item
*root_item
= &root
->root_item
;
6289 struct walk_control
*wc
;
6290 struct btrfs_key key
;
6295 path
= btrfs_alloc_path();
6298 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
6301 trans
= btrfs_start_transaction(tree_root
, 0);
6302 BUG_ON(IS_ERR(trans
));
6305 trans
->block_rsv
= block_rsv
;
6307 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
6308 level
= btrfs_header_level(root
->node
);
6309 path
->nodes
[level
] = btrfs_lock_root_node(root
);
6310 btrfs_set_lock_blocking(path
->nodes
[level
]);
6311 path
->slots
[level
] = 0;
6312 path
->locks
[level
] = 1;
6313 memset(&wc
->update_progress
, 0,
6314 sizeof(wc
->update_progress
));
6316 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
6317 memcpy(&wc
->update_progress
, &key
,
6318 sizeof(wc
->update_progress
));
6320 level
= root_item
->drop_level
;
6322 path
->lowest_level
= level
;
6323 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6324 path
->lowest_level
= 0;
6332 * unlock our path, this is safe because only this
6333 * function is allowed to delete this snapshot
6335 btrfs_unlock_up_safe(path
, 0);
6337 level
= btrfs_header_level(root
->node
);
6339 btrfs_tree_lock(path
->nodes
[level
]);
6340 btrfs_set_lock_blocking(path
->nodes
[level
]);
6342 ret
= btrfs_lookup_extent_info(trans
, root
,
6343 path
->nodes
[level
]->start
,
6344 path
->nodes
[level
]->len
,
6348 BUG_ON(wc
->refs
[level
] == 0);
6350 if (level
== root_item
->drop_level
)
6353 btrfs_tree_unlock(path
->nodes
[level
]);
6354 WARN_ON(wc
->refs
[level
] != 1);
6360 wc
->shared_level
= -1;
6361 wc
->stage
= DROP_REFERENCE
;
6362 wc
->update_ref
= update_ref
;
6364 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
6367 ret
= walk_down_tree(trans
, root
, path
, wc
);
6373 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
6380 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
6384 if (wc
->stage
== DROP_REFERENCE
) {
6386 btrfs_node_key(path
->nodes
[level
],
6387 &root_item
->drop_progress
,
6388 path
->slots
[level
]);
6389 root_item
->drop_level
= level
;
6392 BUG_ON(wc
->level
== 0);
6393 if (btrfs_should_end_transaction(trans
, tree_root
)) {
6394 ret
= btrfs_update_root(trans
, tree_root
,
6399 btrfs_end_transaction_throttle(trans
, tree_root
);
6400 trans
= btrfs_start_transaction(tree_root
, 0);
6401 BUG_ON(IS_ERR(trans
));
6403 trans
->block_rsv
= block_rsv
;
6406 btrfs_release_path(root
, path
);
6409 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
6412 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
6413 ret
= btrfs_find_last_root(tree_root
, root
->root_key
.objectid
,
6417 /* if we fail to delete the orphan item this time
6418 * around, it'll get picked up the next time.
6420 * The most common failure here is just -ENOENT.
6422 btrfs_del_orphan_item(trans
, tree_root
,
6423 root
->root_key
.objectid
);
6427 if (root
->in_radix
) {
6428 btrfs_free_fs_root(tree_root
->fs_info
, root
);
6430 free_extent_buffer(root
->node
);
6431 free_extent_buffer(root
->commit_root
);
6435 btrfs_end_transaction_throttle(trans
, tree_root
);
6437 btrfs_free_path(path
);
6442 * drop subtree rooted at tree block 'node'.
6444 * NOTE: this function will unlock and release tree block 'node'
6446 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
6447 struct btrfs_root
*root
,
6448 struct extent_buffer
*node
,
6449 struct extent_buffer
*parent
)
6451 struct btrfs_path
*path
;
6452 struct walk_control
*wc
;
6458 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
6460 path
= btrfs_alloc_path();
6464 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
6466 btrfs_free_path(path
);
6470 btrfs_assert_tree_locked(parent
);
6471 parent_level
= btrfs_header_level(parent
);
6472 extent_buffer_get(parent
);
6473 path
->nodes
[parent_level
] = parent
;
6474 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
6476 btrfs_assert_tree_locked(node
);
6477 level
= btrfs_header_level(node
);
6478 path
->nodes
[level
] = node
;
6479 path
->slots
[level
] = 0;
6480 path
->locks
[level
] = 1;
6482 wc
->refs
[parent_level
] = 1;
6483 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
6485 wc
->shared_level
= -1;
6486 wc
->stage
= DROP_REFERENCE
;
6489 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
6492 wret
= walk_down_tree(trans
, root
, path
, wc
);
6498 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
6506 btrfs_free_path(path
);
6511 static unsigned long calc_ra(unsigned long start
, unsigned long last
,
6514 return min(last
, start
+ nr
- 1);
6517 static noinline
int relocate_inode_pages(struct inode
*inode
, u64 start
,
6522 unsigned long first_index
;
6523 unsigned long last_index
;
6526 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
6527 struct file_ra_state
*ra
;
6528 struct btrfs_ordered_extent
*ordered
;
6529 unsigned int total_read
= 0;
6530 unsigned int total_dirty
= 0;
6533 ra
= kzalloc(sizeof(*ra
), GFP_NOFS
);
6537 mutex_lock(&inode
->i_mutex
);
6538 first_index
= start
>> PAGE_CACHE_SHIFT
;
6539 last_index
= (start
+ len
- 1) >> PAGE_CACHE_SHIFT
;
6541 /* make sure the dirty trick played by the caller work */
6542 ret
= invalidate_inode_pages2_range(inode
->i_mapping
,
6543 first_index
, last_index
);
6547 file_ra_state_init(ra
, inode
->i_mapping
);
6549 for (i
= first_index
; i
<= last_index
; i
++) {
6550 if (total_read
% ra
->ra_pages
== 0) {
6551 btrfs_force_ra(inode
->i_mapping
, ra
, NULL
, i
,
6552 calc_ra(i
, last_index
, ra
->ra_pages
));
6556 if (((u64
)i
<< PAGE_CACHE_SHIFT
) > i_size_read(inode
))
6558 page
= grab_cache_page(inode
->i_mapping
, i
);
6563 if (!PageUptodate(page
)) {
6564 btrfs_readpage(NULL
, page
);
6566 if (!PageUptodate(page
)) {
6568 page_cache_release(page
);
6573 wait_on_page_writeback(page
);
6575 page_start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
6576 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
6577 lock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
6579 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
6581 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
6583 page_cache_release(page
);
6584 btrfs_start_ordered_extent(inode
, ordered
, 1);
6585 btrfs_put_ordered_extent(ordered
);
6588 set_page_extent_mapped(page
);
6590 if (i
== first_index
)
6591 set_extent_bits(io_tree
, page_start
, page_end
,
6592 EXTENT_BOUNDARY
, GFP_NOFS
);
6593 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
6595 set_page_dirty(page
);
6598 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
6600 page_cache_release(page
);
6605 mutex_unlock(&inode
->i_mutex
);
6606 balance_dirty_pages_ratelimited_nr(inode
->i_mapping
, total_dirty
);
6610 static noinline
int relocate_data_extent(struct inode
*reloc_inode
,
6611 struct btrfs_key
*extent_key
,
6614 struct btrfs_root
*root
= BTRFS_I(reloc_inode
)->root
;
6615 struct extent_map_tree
*em_tree
= &BTRFS_I(reloc_inode
)->extent_tree
;
6616 struct extent_map
*em
;
6617 u64 start
= extent_key
->objectid
- offset
;
6618 u64 end
= start
+ extent_key
->offset
- 1;
6620 em
= alloc_extent_map(GFP_NOFS
);
6624 em
->len
= extent_key
->offset
;
6625 em
->block_len
= extent_key
->offset
;
6626 em
->block_start
= extent_key
->objectid
;
6627 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
6628 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
6630 /* setup extent map to cheat btrfs_readpage */
6631 lock_extent(&BTRFS_I(reloc_inode
)->io_tree
, start
, end
, GFP_NOFS
);
6634 write_lock(&em_tree
->lock
);
6635 ret
= add_extent_mapping(em_tree
, em
);
6636 write_unlock(&em_tree
->lock
);
6637 if (ret
!= -EEXIST
) {
6638 free_extent_map(em
);
6641 btrfs_drop_extent_cache(reloc_inode
, start
, end
, 0);
6643 unlock_extent(&BTRFS_I(reloc_inode
)->io_tree
, start
, end
, GFP_NOFS
);
6645 return relocate_inode_pages(reloc_inode
, start
, extent_key
->offset
);
6648 struct btrfs_ref_path
{
6650 u64 nodes
[BTRFS_MAX_LEVEL
];
6652 u64 root_generation
;
6659 struct btrfs_key node_keys
[BTRFS_MAX_LEVEL
];
6660 u64 new_nodes
[BTRFS_MAX_LEVEL
];
6663 struct disk_extent
{
6674 static int is_cowonly_root(u64 root_objectid
)
6676 if (root_objectid
== BTRFS_ROOT_TREE_OBJECTID
||
6677 root_objectid
== BTRFS_EXTENT_TREE_OBJECTID
||
6678 root_objectid
== BTRFS_CHUNK_TREE_OBJECTID
||
6679 root_objectid
== BTRFS_DEV_TREE_OBJECTID
||
6680 root_objectid
== BTRFS_TREE_LOG_OBJECTID
||
6681 root_objectid
== BTRFS_CSUM_TREE_OBJECTID
)
6686 static noinline
int __next_ref_path(struct btrfs_trans_handle
*trans
,
6687 struct btrfs_root
*extent_root
,
6688 struct btrfs_ref_path
*ref_path
,
6691 struct extent_buffer
*leaf
;
6692 struct btrfs_path
*path
;
6693 struct btrfs_extent_ref
*ref
;
6694 struct btrfs_key key
;
6695 struct btrfs_key found_key
;
6701 path
= btrfs_alloc_path();
6706 ref_path
->lowest_level
= -1;
6707 ref_path
->current_level
= -1;
6708 ref_path
->shared_level
= -1;
6712 level
= ref_path
->current_level
- 1;
6713 while (level
>= -1) {
6715 if (level
< ref_path
->lowest_level
)
6719 bytenr
= ref_path
->nodes
[level
];
6721 bytenr
= ref_path
->extent_start
;
6722 BUG_ON(bytenr
== 0);
6724 parent
= ref_path
->nodes
[level
+ 1];
6725 ref_path
->nodes
[level
+ 1] = 0;
6726 ref_path
->current_level
= level
;
6727 BUG_ON(parent
== 0);
6729 key
.objectid
= bytenr
;
6730 key
.offset
= parent
+ 1;
6731 key
.type
= BTRFS_EXTENT_REF_KEY
;
6733 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
, 0, 0);
6738 leaf
= path
->nodes
[0];
6739 nritems
= btrfs_header_nritems(leaf
);
6740 if (path
->slots
[0] >= nritems
) {
6741 ret
= btrfs_next_leaf(extent_root
, path
);
6746 leaf
= path
->nodes
[0];
6749 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6750 if (found_key
.objectid
== bytenr
&&
6751 found_key
.type
== BTRFS_EXTENT_REF_KEY
) {
6752 if (level
< ref_path
->shared_level
)
6753 ref_path
->shared_level
= level
;
6758 btrfs_release_path(extent_root
, path
);
6761 /* reached lowest level */
6765 level
= ref_path
->current_level
;
6766 while (level
< BTRFS_MAX_LEVEL
- 1) {
6770 bytenr
= ref_path
->nodes
[level
];
6772 bytenr
= ref_path
->extent_start
;
6774 BUG_ON(bytenr
== 0);
6776 key
.objectid
= bytenr
;
6778 key
.type
= BTRFS_EXTENT_REF_KEY
;
6780 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
, 0, 0);
6784 leaf
= path
->nodes
[0];
6785 nritems
= btrfs_header_nritems(leaf
);
6786 if (path
->slots
[0] >= nritems
) {
6787 ret
= btrfs_next_leaf(extent_root
, path
);
6791 /* the extent was freed by someone */
6792 if (ref_path
->lowest_level
== level
)
6794 btrfs_release_path(extent_root
, path
);
6797 leaf
= path
->nodes
[0];
6800 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6801 if (found_key
.objectid
!= bytenr
||
6802 found_key
.type
!= BTRFS_EXTENT_REF_KEY
) {
6803 /* the extent was freed by someone */
6804 if (ref_path
->lowest_level
== level
) {
6808 btrfs_release_path(extent_root
, path
);
6812 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
6813 struct btrfs_extent_ref
);
6814 ref_objectid
= btrfs_ref_objectid(leaf
, ref
);
6815 if (ref_objectid
< BTRFS_FIRST_FREE_OBJECTID
) {
6817 level
= (int)ref_objectid
;
6818 BUG_ON(level
>= BTRFS_MAX_LEVEL
);
6819 ref_path
->lowest_level
= level
;
6820 ref_path
->current_level
= level
;
6821 ref_path
->nodes
[level
] = bytenr
;
6823 WARN_ON(ref_objectid
!= level
);
6826 WARN_ON(level
!= -1);
6830 if (ref_path
->lowest_level
== level
) {
6831 ref_path
->owner_objectid
= ref_objectid
;
6832 ref_path
->num_refs
= btrfs_ref_num_refs(leaf
, ref
);
6836 * the block is tree root or the block isn't in reference
6839 if (found_key
.objectid
== found_key
.offset
||
6840 is_cowonly_root(btrfs_ref_root(leaf
, ref
))) {
6841 ref_path
->root_objectid
= btrfs_ref_root(leaf
, ref
);
6842 ref_path
->root_generation
=
6843 btrfs_ref_generation(leaf
, ref
);
6845 /* special reference from the tree log */
6846 ref_path
->nodes
[0] = found_key
.offset
;
6847 ref_path
->current_level
= 0;
6854 BUG_ON(ref_path
->nodes
[level
] != 0);
6855 ref_path
->nodes
[level
] = found_key
.offset
;
6856 ref_path
->current_level
= level
;
6859 * the reference was created in the running transaction,
6860 * no need to continue walking up.
6862 if (btrfs_ref_generation(leaf
, ref
) == trans
->transid
) {
6863 ref_path
->root_objectid
= btrfs_ref_root(leaf
, ref
);
6864 ref_path
->root_generation
=
6865 btrfs_ref_generation(leaf
, ref
);
6870 btrfs_release_path(extent_root
, path
);
6873 /* reached max tree level, but no tree root found. */
6876 btrfs_free_path(path
);
6880 static int btrfs_first_ref_path(struct btrfs_trans_handle
*trans
,
6881 struct btrfs_root
*extent_root
,
6882 struct btrfs_ref_path
*ref_path
,
6885 memset(ref_path
, 0, sizeof(*ref_path
));
6886 ref_path
->extent_start
= extent_start
;
6888 return __next_ref_path(trans
, extent_root
, ref_path
, 1);
6891 static int btrfs_next_ref_path(struct btrfs_trans_handle
*trans
,
6892 struct btrfs_root
*extent_root
,
6893 struct btrfs_ref_path
*ref_path
)
6895 return __next_ref_path(trans
, extent_root
, ref_path
, 0);
6898 static noinline
int get_new_locations(struct inode
*reloc_inode
,
6899 struct btrfs_key
*extent_key
,
6900 u64 offset
, int no_fragment
,
6901 struct disk_extent
**extents
,
6904 struct btrfs_root
*root
= BTRFS_I(reloc_inode
)->root
;
6905 struct btrfs_path
*path
;
6906 struct btrfs_file_extent_item
*fi
;
6907 struct extent_buffer
*leaf
;
6908 struct disk_extent
*exts
= *extents
;
6909 struct btrfs_key found_key
;
6914 int max
= *nr_extents
;
6917 WARN_ON(!no_fragment
&& *extents
);
6920 exts
= kmalloc(sizeof(*exts
) * max
, GFP_NOFS
);
6925 path
= btrfs_alloc_path();
6927 if (exts
!= *extents
)
6932 cur_pos
= extent_key
->objectid
- offset
;
6933 last_byte
= extent_key
->objectid
+ extent_key
->offset
;
6934 ret
= btrfs_lookup_file_extent(NULL
, root
, path
, reloc_inode
->i_ino
,
6944 leaf
= path
->nodes
[0];
6945 nritems
= btrfs_header_nritems(leaf
);
6946 if (path
->slots
[0] >= nritems
) {
6947 ret
= btrfs_next_leaf(root
, path
);
6952 leaf
= path
->nodes
[0];
6955 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6956 if (found_key
.offset
!= cur_pos
||
6957 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
||
6958 found_key
.objectid
!= reloc_inode
->i_ino
)
6961 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
6962 struct btrfs_file_extent_item
);
6963 if (btrfs_file_extent_type(leaf
, fi
) !=
6964 BTRFS_FILE_EXTENT_REG
||
6965 btrfs_file_extent_disk_bytenr(leaf
, fi
) == 0)
6969 struct disk_extent
*old
= exts
;
6971 exts
= kzalloc(sizeof(*exts
) * max
, GFP_NOFS
);
6972 memcpy(exts
, old
, sizeof(*exts
) * nr
);
6973 if (old
!= *extents
)
6977 exts
[nr
].disk_bytenr
=
6978 btrfs_file_extent_disk_bytenr(leaf
, fi
);
6979 exts
[nr
].disk_num_bytes
=
6980 btrfs_file_extent_disk_num_bytes(leaf
, fi
);
6981 exts
[nr
].offset
= btrfs_file_extent_offset(leaf
, fi
);
6982 exts
[nr
].num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
6983 exts
[nr
].ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
6984 exts
[nr
].compression
= btrfs_file_extent_compression(leaf
, fi
);
6985 exts
[nr
].encryption
= btrfs_file_extent_encryption(leaf
, fi
);
6986 exts
[nr
].other_encoding
= btrfs_file_extent_other_encoding(leaf
,
6988 BUG_ON(exts
[nr
].offset
> 0);
6989 BUG_ON(exts
[nr
].compression
|| exts
[nr
].encryption
);
6990 BUG_ON(exts
[nr
].num_bytes
!= exts
[nr
].disk_num_bytes
);
6992 cur_pos
+= exts
[nr
].num_bytes
;
6995 if (cur_pos
+ offset
>= last_byte
)
7005 BUG_ON(cur_pos
+ offset
> last_byte
);
7006 if (cur_pos
+ offset
< last_byte
) {
7012 btrfs_free_path(path
);
7014 if (exts
!= *extents
)
7023 static noinline
int replace_one_extent(struct btrfs_trans_handle
*trans
,
7024 struct btrfs_root
*root
,
7025 struct btrfs_path
*path
,
7026 struct btrfs_key
*extent_key
,
7027 struct btrfs_key
*leaf_key
,
7028 struct btrfs_ref_path
*ref_path
,
7029 struct disk_extent
*new_extents
,
7032 struct extent_buffer
*leaf
;
7033 struct btrfs_file_extent_item
*fi
;
7034 struct inode
*inode
= NULL
;
7035 struct btrfs_key key
;
7040 u64 search_end
= (u64
)-1;
7043 int extent_locked
= 0;
7047 memcpy(&key
, leaf_key
, sizeof(key
));
7048 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
) {
7049 if (key
.objectid
< ref_path
->owner_objectid
||
7050 (key
.objectid
== ref_path
->owner_objectid
&&
7051 key
.type
< BTRFS_EXTENT_DATA_KEY
)) {
7052 key
.objectid
= ref_path
->owner_objectid
;
7053 key
.type
= BTRFS_EXTENT_DATA_KEY
;
7059 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
7063 leaf
= path
->nodes
[0];
7064 nritems
= btrfs_header_nritems(leaf
);
7066 if (extent_locked
&& ret
> 0) {
7068 * the file extent item was modified by someone
7069 * before the extent got locked.
7071 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
7072 lock_end
, GFP_NOFS
);
7076 if (path
->slots
[0] >= nritems
) {
7077 if (++nr_scaned
> 2)
7080 BUG_ON(extent_locked
);
7081 ret
= btrfs_next_leaf(root
, path
);
7086 leaf
= path
->nodes
[0];
7087 nritems
= btrfs_header_nritems(leaf
);
7090 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
7092 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
) {
7093 if ((key
.objectid
> ref_path
->owner_objectid
) ||
7094 (key
.objectid
== ref_path
->owner_objectid
&&
7095 key
.type
> BTRFS_EXTENT_DATA_KEY
) ||
7096 key
.offset
>= search_end
)
7100 if (inode
&& key
.objectid
!= inode
->i_ino
) {
7101 BUG_ON(extent_locked
);
7102 btrfs_release_path(root
, path
);
7103 mutex_unlock(&inode
->i_mutex
);
7109 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
7114 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
7115 struct btrfs_file_extent_item
);
7116 extent_type
= btrfs_file_extent_type(leaf
, fi
);
7117 if ((extent_type
!= BTRFS_FILE_EXTENT_REG
&&
7118 extent_type
!= BTRFS_FILE_EXTENT_PREALLOC
) ||
7119 (btrfs_file_extent_disk_bytenr(leaf
, fi
) !=
7120 extent_key
->objectid
)) {
7126 num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
7127 ext_offset
= btrfs_file_extent_offset(leaf
, fi
);
7129 if (search_end
== (u64
)-1) {
7130 search_end
= key
.offset
- ext_offset
+
7131 btrfs_file_extent_ram_bytes(leaf
, fi
);
7134 if (!extent_locked
) {
7135 lock_start
= key
.offset
;
7136 lock_end
= lock_start
+ num_bytes
- 1;
7138 if (lock_start
> key
.offset
||
7139 lock_end
+ 1 < key
.offset
+ num_bytes
) {
7140 unlock_extent(&BTRFS_I(inode
)->io_tree
,
7141 lock_start
, lock_end
, GFP_NOFS
);
7147 btrfs_release_path(root
, path
);
7149 inode
= btrfs_iget_locked(root
->fs_info
->sb
,
7150 key
.objectid
, root
);
7151 if (inode
->i_state
& I_NEW
) {
7152 BTRFS_I(inode
)->root
= root
;
7153 BTRFS_I(inode
)->location
.objectid
=
7155 BTRFS_I(inode
)->location
.type
=
7156 BTRFS_INODE_ITEM_KEY
;
7157 BTRFS_I(inode
)->location
.offset
= 0;
7158 btrfs_read_locked_inode(inode
);
7159 unlock_new_inode(inode
);
7162 * some code call btrfs_commit_transaction while
7163 * holding the i_mutex, so we can't use mutex_lock
7166 if (is_bad_inode(inode
) ||
7167 !mutex_trylock(&inode
->i_mutex
)) {
7170 key
.offset
= (u64
)-1;
7175 if (!extent_locked
) {
7176 struct btrfs_ordered_extent
*ordered
;
7178 btrfs_release_path(root
, path
);
7180 lock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
7181 lock_end
, GFP_NOFS
);
7182 ordered
= btrfs_lookup_first_ordered_extent(inode
,
7185 ordered
->file_offset
<= lock_end
&&
7186 ordered
->file_offset
+ ordered
->len
> lock_start
) {
7187 unlock_extent(&BTRFS_I(inode
)->io_tree
,
7188 lock_start
, lock_end
, GFP_NOFS
);
7189 btrfs_start_ordered_extent(inode
, ordered
, 1);
7190 btrfs_put_ordered_extent(ordered
);
7191 key
.offset
+= num_bytes
;
7195 btrfs_put_ordered_extent(ordered
);
7201 if (nr_extents
== 1) {
7202 /* update extent pointer in place */
7203 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
7204 new_extents
[0].disk_bytenr
);
7205 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
7206 new_extents
[0].disk_num_bytes
);
7207 btrfs_mark_buffer_dirty(leaf
);
7209 btrfs_drop_extent_cache(inode
, key
.offset
,
7210 key
.offset
+ num_bytes
- 1, 0);
7212 ret
= btrfs_inc_extent_ref(trans
, root
,
7213 new_extents
[0].disk_bytenr
,
7214 new_extents
[0].disk_num_bytes
,
7216 root
->root_key
.objectid
,
7221 ret
= btrfs_free_extent(trans
, root
,
7222 extent_key
->objectid
,
7225 btrfs_header_owner(leaf
),
7226 btrfs_header_generation(leaf
),
7230 btrfs_release_path(root
, path
);
7231 key
.offset
+= num_bytes
;
7239 * drop old extent pointer at first, then insert the
7240 * new pointers one bye one
7242 btrfs_release_path(root
, path
);
7243 ret
= btrfs_drop_extents(trans
, root
, inode
, key
.offset
,
7244 key
.offset
+ num_bytes
,
7245 key
.offset
, &alloc_hint
);
7248 for (i
= 0; i
< nr_extents
; i
++) {
7249 if (ext_offset
>= new_extents
[i
].num_bytes
) {
7250 ext_offset
-= new_extents
[i
].num_bytes
;
7253 extent_len
= min(new_extents
[i
].num_bytes
-
7254 ext_offset
, num_bytes
);
7256 ret
= btrfs_insert_empty_item(trans
, root
,
7261 leaf
= path
->nodes
[0];
7262 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
7263 struct btrfs_file_extent_item
);
7264 btrfs_set_file_extent_generation(leaf
, fi
,
7266 btrfs_set_file_extent_type(leaf
, fi
,
7267 BTRFS_FILE_EXTENT_REG
);
7268 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
7269 new_extents
[i
].disk_bytenr
);
7270 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
7271 new_extents
[i
].disk_num_bytes
);
7272 btrfs_set_file_extent_ram_bytes(leaf
, fi
,
7273 new_extents
[i
].ram_bytes
);
7275 btrfs_set_file_extent_compression(leaf
, fi
,
7276 new_extents
[i
].compression
);
7277 btrfs_set_file_extent_encryption(leaf
, fi
,
7278 new_extents
[i
].encryption
);
7279 btrfs_set_file_extent_other_encoding(leaf
, fi
,
7280 new_extents
[i
].other_encoding
);
7282 btrfs_set_file_extent_num_bytes(leaf
, fi
,
7284 ext_offset
+= new_extents
[i
].offset
;
7285 btrfs_set_file_extent_offset(leaf
, fi
,
7287 btrfs_mark_buffer_dirty(leaf
);
7289 btrfs_drop_extent_cache(inode
, key
.offset
,
7290 key
.offset
+ extent_len
- 1, 0);
7292 ret
= btrfs_inc_extent_ref(trans
, root
,
7293 new_extents
[i
].disk_bytenr
,
7294 new_extents
[i
].disk_num_bytes
,
7296 root
->root_key
.objectid
,
7297 trans
->transid
, key
.objectid
);
7299 btrfs_release_path(root
, path
);
7301 inode_add_bytes(inode
, extent_len
);
7304 num_bytes
-= extent_len
;
7305 key
.offset
+= extent_len
;
7310 BUG_ON(i
>= nr_extents
);
7314 if (extent_locked
) {
7315 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
7316 lock_end
, GFP_NOFS
);
7320 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
&&
7321 key
.offset
>= search_end
)
7328 btrfs_release_path(root
, path
);
7330 mutex_unlock(&inode
->i_mutex
);
7331 if (extent_locked
) {
7332 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
7333 lock_end
, GFP_NOFS
);
7340 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle
*trans
,
7341 struct btrfs_root
*root
,
7342 struct extent_buffer
*buf
, u64 orig_start
)
7347 BUG_ON(btrfs_header_generation(buf
) != trans
->transid
);
7348 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
7350 level
= btrfs_header_level(buf
);
7352 struct btrfs_leaf_ref
*ref
;
7353 struct btrfs_leaf_ref
*orig_ref
;
7355 orig_ref
= btrfs_lookup_leaf_ref(root
, orig_start
);
7359 ref
= btrfs_alloc_leaf_ref(root
, orig_ref
->nritems
);
7361 btrfs_free_leaf_ref(root
, orig_ref
);
7365 ref
->nritems
= orig_ref
->nritems
;
7366 memcpy(ref
->extents
, orig_ref
->extents
,
7367 sizeof(ref
->extents
[0]) * ref
->nritems
);
7369 btrfs_free_leaf_ref(root
, orig_ref
);
7371 ref
->root_gen
= trans
->transid
;
7372 ref
->bytenr
= buf
->start
;
7373 ref
->owner
= btrfs_header_owner(buf
);
7374 ref
->generation
= btrfs_header_generation(buf
);
7376 ret
= btrfs_add_leaf_ref(root
, ref
, 0);
7378 btrfs_free_leaf_ref(root
, ref
);
7383 static noinline
int invalidate_extent_cache(struct btrfs_root
*root
,
7384 struct extent_buffer
*leaf
,
7385 struct btrfs_block_group_cache
*group
,
7386 struct btrfs_root
*target_root
)
7388 struct btrfs_key key
;
7389 struct inode
*inode
= NULL
;
7390 struct btrfs_file_extent_item
*fi
;
7391 struct extent_state
*cached_state
= NULL
;
7393 u64 skip_objectid
= 0;
7397 nritems
= btrfs_header_nritems(leaf
);
7398 for (i
= 0; i
< nritems
; i
++) {
7399 btrfs_item_key_to_cpu(leaf
, &key
, i
);
7400 if (key
.objectid
== skip_objectid
||
7401 key
.type
!= BTRFS_EXTENT_DATA_KEY
)
7403 fi
= btrfs_item_ptr(leaf
, i
, struct btrfs_file_extent_item
);
7404 if (btrfs_file_extent_type(leaf
, fi
) ==
7405 BTRFS_FILE_EXTENT_INLINE
)
7407 if (btrfs_file_extent_disk_bytenr(leaf
, fi
) == 0)
7409 if (!inode
|| inode
->i_ino
!= key
.objectid
) {
7411 inode
= btrfs_ilookup(target_root
->fs_info
->sb
,
7412 key
.objectid
, target_root
, 1);
7415 skip_objectid
= key
.objectid
;
7418 num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
7420 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, key
.offset
,
7421 key
.offset
+ num_bytes
- 1, 0, &cached_state
,
7423 btrfs_drop_extent_cache(inode
, key
.offset
,
7424 key
.offset
+ num_bytes
- 1, 1);
7425 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, key
.offset
,
7426 key
.offset
+ num_bytes
- 1, &cached_state
,
7434 static noinline
int replace_extents_in_leaf(struct btrfs_trans_handle
*trans
,
7435 struct btrfs_root
*root
,
7436 struct extent_buffer
*leaf
,
7437 struct btrfs_block_group_cache
*group
,
7438 struct inode
*reloc_inode
)
7440 struct btrfs_key key
;
7441 struct btrfs_key extent_key
;
7442 struct btrfs_file_extent_item
*fi
;
7443 struct btrfs_leaf_ref
*ref
;
7444 struct disk_extent
*new_extent
;
7453 new_extent
= kmalloc(sizeof(*new_extent
), GFP_NOFS
);
7457 ref
= btrfs_lookup_leaf_ref(root
, leaf
->start
);
7461 nritems
= btrfs_header_nritems(leaf
);
7462 for (i
= 0; i
< nritems
; i
++) {
7463 btrfs_item_key_to_cpu(leaf
, &key
, i
);
7464 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
7466 fi
= btrfs_item_ptr(leaf
, i
, struct btrfs_file_extent_item
);
7467 if (btrfs_file_extent_type(leaf
, fi
) ==
7468 BTRFS_FILE_EXTENT_INLINE
)
7470 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
7471 num_bytes
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
7476 if (bytenr
>= group
->key
.objectid
+ group
->key
.offset
||
7477 bytenr
+ num_bytes
<= group
->key
.objectid
)
7480 extent_key
.objectid
= bytenr
;
7481 extent_key
.offset
= num_bytes
;
7482 extent_key
.type
= BTRFS_EXTENT_ITEM_KEY
;
7484 ret
= get_new_locations(reloc_inode
, &extent_key
,
7485 group
->key
.objectid
, 1,
7486 &new_extent
, &nr_extent
);
7491 BUG_ON(ref
->extents
[ext_index
].bytenr
!= bytenr
);
7492 BUG_ON(ref
->extents
[ext_index
].num_bytes
!= num_bytes
);
7493 ref
->extents
[ext_index
].bytenr
= new_extent
->disk_bytenr
;
7494 ref
->extents
[ext_index
].num_bytes
= new_extent
->disk_num_bytes
;
7496 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
7497 new_extent
->disk_bytenr
);
7498 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
7499 new_extent
->disk_num_bytes
);
7500 btrfs_mark_buffer_dirty(leaf
);
7502 ret
= btrfs_inc_extent_ref(trans
, root
,
7503 new_extent
->disk_bytenr
,
7504 new_extent
->disk_num_bytes
,
7506 root
->root_key
.objectid
,
7507 trans
->transid
, key
.objectid
);
7510 ret
= btrfs_free_extent(trans
, root
,
7511 bytenr
, num_bytes
, leaf
->start
,
7512 btrfs_header_owner(leaf
),
7513 btrfs_header_generation(leaf
),
7519 BUG_ON(ext_index
+ 1 != ref
->nritems
);
7520 btrfs_free_leaf_ref(root
, ref
);
7524 int btrfs_free_reloc_root(struct btrfs_trans_handle
*trans
,
7525 struct btrfs_root
*root
)
7527 struct btrfs_root
*reloc_root
;
7530 if (root
->reloc_root
) {
7531 reloc_root
= root
->reloc_root
;
7532 root
->reloc_root
= NULL
;
7533 list_add(&reloc_root
->dead_list
,
7534 &root
->fs_info
->dead_reloc_roots
);
7536 btrfs_set_root_bytenr(&reloc_root
->root_item
,
7537 reloc_root
->node
->start
);
7538 btrfs_set_root_level(&root
->root_item
,
7539 btrfs_header_level(reloc_root
->node
));
7540 memset(&reloc_root
->root_item
.drop_progress
, 0,
7541 sizeof(struct btrfs_disk_key
));
7542 reloc_root
->root_item
.drop_level
= 0;
7544 ret
= btrfs_update_root(trans
, root
->fs_info
->tree_root
,
7545 &reloc_root
->root_key
,
7546 &reloc_root
->root_item
);
7552 int btrfs_drop_dead_reloc_roots(struct btrfs_root
*root
)
7554 struct btrfs_trans_handle
*trans
;
7555 struct btrfs_root
*reloc_root
;
7556 struct btrfs_root
*prev_root
= NULL
;
7557 struct list_head dead_roots
;
7561 INIT_LIST_HEAD(&dead_roots
);
7562 list_splice_init(&root
->fs_info
->dead_reloc_roots
, &dead_roots
);
7564 while (!list_empty(&dead_roots
)) {
7565 reloc_root
= list_entry(dead_roots
.prev
,
7566 struct btrfs_root
, dead_list
);
7567 list_del_init(&reloc_root
->dead_list
);
7569 BUG_ON(reloc_root
->commit_root
!= NULL
);
7571 trans
= btrfs_join_transaction(root
, 1);
7572 BUG_ON(IS_ERR(trans
));
7574 mutex_lock(&root
->fs_info
->drop_mutex
);
7575 ret
= btrfs_drop_snapshot(trans
, reloc_root
);
7578 mutex_unlock(&root
->fs_info
->drop_mutex
);
7580 nr
= trans
->blocks_used
;
7581 ret
= btrfs_end_transaction(trans
, root
);
7583 btrfs_btree_balance_dirty(root
, nr
);
7586 free_extent_buffer(reloc_root
->node
);
7588 ret
= btrfs_del_root(trans
, root
->fs_info
->tree_root
,
7589 &reloc_root
->root_key
);
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 prev_root
= reloc_root
;
7602 btrfs_remove_leaf_refs(prev_root
, (u64
)-1, 0);
7608 int btrfs_add_dead_reloc_root(struct btrfs_root
*root
)
7610 list_add(&root
->dead_list
, &root
->fs_info
->dead_reloc_roots
);
7614 int btrfs_cleanup_reloc_trees(struct btrfs_root
*root
)
7616 struct btrfs_root
*reloc_root
;
7617 struct btrfs_trans_handle
*trans
;
7618 struct btrfs_key location
;
7622 mutex_lock(&root
->fs_info
->tree_reloc_mutex
);
7623 ret
= btrfs_find_dead_roots(root
, BTRFS_TREE_RELOC_OBJECTID
, NULL
);
7625 found
= !list_empty(&root
->fs_info
->dead_reloc_roots
);
7626 mutex_unlock(&root
->fs_info
->tree_reloc_mutex
);
7629 trans
= btrfs_start_transaction(root
, 1);
7630 BUG_ON(IS_ERR(trans
));
7631 ret
= btrfs_commit_transaction(trans
, root
);
7635 location
.objectid
= BTRFS_DATA_RELOC_TREE_OBJECTID
;
7636 location
.offset
= (u64
)-1;
7637 location
.type
= BTRFS_ROOT_ITEM_KEY
;
7639 reloc_root
= btrfs_read_fs_root_no_name(root
->fs_info
, &location
);
7640 BUG_ON(!reloc_root
);
7641 ret
= btrfs_orphan_cleanup(reloc_root
);
7646 static noinline
int init_reloc_tree(struct btrfs_trans_handle
*trans
,
7647 struct btrfs_root
*root
)
7649 struct btrfs_root
*reloc_root
;
7650 struct extent_buffer
*eb
;
7651 struct btrfs_root_item
*root_item
;
7652 struct btrfs_key root_key
;
7655 BUG_ON(!root
->ref_cows
);
7656 if (root
->reloc_root
)
7659 root_item
= kmalloc(sizeof(*root_item
), GFP_NOFS
);
7663 ret
= btrfs_copy_root(trans
, root
, root
->commit_root
,
7664 &eb
, BTRFS_TREE_RELOC_OBJECTID
);
7667 root_key
.objectid
= BTRFS_TREE_RELOC_OBJECTID
;
7668 root_key
.offset
= root
->root_key
.objectid
;
7669 root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
7671 memcpy(root_item
, &root
->root_item
, sizeof(root_item
));
7672 btrfs_set_root_refs(root_item
, 0);
7673 btrfs_set_root_bytenr(root_item
, eb
->start
);
7674 btrfs_set_root_level(root_item
, btrfs_header_level(eb
));
7675 btrfs_set_root_generation(root_item
, trans
->transid
);
7677 btrfs_tree_unlock(eb
);
7678 free_extent_buffer(eb
);
7680 ret
= btrfs_insert_root(trans
, root
->fs_info
->tree_root
,
7681 &root_key
, root_item
);
7685 reloc_root
= btrfs_read_fs_root_no_radix(root
->fs_info
->tree_root
,
7687 BUG_ON(IS_ERR(reloc_root
));
7688 reloc_root
->last_trans
= trans
->transid
;
7689 reloc_root
->commit_root
= NULL
;
7690 reloc_root
->ref_tree
= &root
->fs_info
->reloc_ref_tree
;
7692 root
->reloc_root
= reloc_root
;
7697 * Core function of space balance.
7699 * The idea is using reloc trees to relocate tree blocks in reference
7700 * counted roots. There is one reloc tree for each subvol, and all
7701 * reloc trees share same root key objectid. Reloc trees are snapshots
7702 * of the latest committed roots of subvols (root->commit_root).
7704 * To relocate a tree block referenced by a subvol, there are two steps.
7705 * COW the block through subvol's reloc tree, then update block pointer
7706 * in the subvol to point to the new block. Since all reloc trees share
7707 * same root key objectid, doing special handing for tree blocks owned
7708 * by them is easy. Once a tree block has been COWed in one reloc tree,
7709 * we can use the resulting new block directly when the same block is
7710 * required to COW again through other reloc trees. By this way, relocated
7711 * tree blocks are shared between reloc trees, so they are also shared
7714 static noinline
int relocate_one_path(struct btrfs_trans_handle
*trans
,
7715 struct btrfs_root
*root
,
7716 struct btrfs_path
*path
,
7717 struct btrfs_key
*first_key
,
7718 struct btrfs_ref_path
*ref_path
,
7719 struct btrfs_block_group_cache
*group
,
7720 struct inode
*reloc_inode
)
7722 struct btrfs_root
*reloc_root
;
7723 struct extent_buffer
*eb
= NULL
;
7724 struct btrfs_key
*keys
;
7728 int lowest_level
= 0;
7731 if (ref_path
->owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
)
7732 lowest_level
= ref_path
->owner_objectid
;
7734 if (!root
->ref_cows
) {
7735 path
->lowest_level
= lowest_level
;
7736 ret
= btrfs_search_slot(trans
, root
, first_key
, path
, 0, 1);
7738 path
->lowest_level
= 0;
7739 btrfs_release_path(root
, path
);
7743 mutex_lock(&root
->fs_info
->tree_reloc_mutex
);
7744 ret
= init_reloc_tree(trans
, root
);
7746 reloc_root
= root
->reloc_root
;
7748 shared_level
= ref_path
->shared_level
;
7749 ref_path
->shared_level
= BTRFS_MAX_LEVEL
- 1;
7751 keys
= ref_path
->node_keys
;
7752 nodes
= ref_path
->new_nodes
;
7753 memset(&keys
[shared_level
+ 1], 0,
7754 sizeof(*keys
) * (BTRFS_MAX_LEVEL
- shared_level
- 1));
7755 memset(&nodes
[shared_level
+ 1], 0,
7756 sizeof(*nodes
) * (BTRFS_MAX_LEVEL
- shared_level
- 1));
7758 if (nodes
[lowest_level
] == 0) {
7759 path
->lowest_level
= lowest_level
;
7760 ret
= btrfs_search_slot(trans
, reloc_root
, first_key
, path
,
7763 for (level
= lowest_level
; level
< BTRFS_MAX_LEVEL
; level
++) {
7764 eb
= path
->nodes
[level
];
7765 if (!eb
|| eb
== reloc_root
->node
)
7767 nodes
[level
] = eb
->start
;
7769 btrfs_item_key_to_cpu(eb
, &keys
[level
], 0);
7771 btrfs_node_key_to_cpu(eb
, &keys
[level
], 0);
7774 ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
7775 eb
= path
->nodes
[0];
7776 ret
= replace_extents_in_leaf(trans
, reloc_root
, eb
,
7777 group
, reloc_inode
);
7780 btrfs_release_path(reloc_root
, path
);
7782 ret
= btrfs_merge_path(trans
, reloc_root
, keys
, nodes
,
7788 * replace tree blocks in the fs tree with tree blocks in
7791 ret
= btrfs_merge_path(trans
, root
, keys
, nodes
, lowest_level
);
7794 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
7795 ret
= btrfs_search_slot(trans
, reloc_root
, first_key
, path
,
7798 extent_buffer_get(path
->nodes
[0]);
7799 eb
= path
->nodes
[0];
7800 btrfs_release_path(reloc_root
, path
);
7801 ret
= invalidate_extent_cache(reloc_root
, eb
, group
, root
);
7803 free_extent_buffer(eb
);
7806 mutex_unlock(&root
->fs_info
->tree_reloc_mutex
);
7807 path
->lowest_level
= 0;
7811 static noinline
int relocate_tree_block(struct btrfs_trans_handle
*trans
,
7812 struct btrfs_root
*root
,
7813 struct btrfs_path
*path
,
7814 struct btrfs_key
*first_key
,
7815 struct btrfs_ref_path
*ref_path
)
7819 ret
= relocate_one_path(trans
, root
, path
, first_key
,
7820 ref_path
, NULL
, NULL
);
7826 static noinline
int del_extent_zero(struct btrfs_trans_handle
*trans
,
7827 struct btrfs_root
*extent_root
,
7828 struct btrfs_path
*path
,
7829 struct btrfs_key
*extent_key
)
7833 ret
= btrfs_search_slot(trans
, extent_root
, extent_key
, path
, -1, 1);
7836 ret
= btrfs_del_item(trans
, extent_root
, path
);
7838 btrfs_release_path(extent_root
, path
);
7842 static noinline
struct btrfs_root
*read_ref_root(struct btrfs_fs_info
*fs_info
,
7843 struct btrfs_ref_path
*ref_path
)
7845 struct btrfs_key root_key
;
7847 root_key
.objectid
= ref_path
->root_objectid
;
7848 root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
7849 if (is_cowonly_root(ref_path
->root_objectid
))
7850 root_key
.offset
= 0;
7852 root_key
.offset
= (u64
)-1;
7854 return btrfs_read_fs_root_no_name(fs_info
, &root_key
);
7857 static noinline
int relocate_one_extent(struct btrfs_root
*extent_root
,
7858 struct btrfs_path
*path
,
7859 struct btrfs_key
*extent_key
,
7860 struct btrfs_block_group_cache
*group
,
7861 struct inode
*reloc_inode
, int pass
)
7863 struct btrfs_trans_handle
*trans
;
7864 struct btrfs_root
*found_root
;
7865 struct btrfs_ref_path
*ref_path
= NULL
;
7866 struct disk_extent
*new_extents
= NULL
;
7871 struct btrfs_key first_key
;
7875 trans
= btrfs_start_transaction(extent_root
, 1);
7876 BUG_ON(IS_ERR(trans
));
7878 if (extent_key
->objectid
== 0) {
7879 ret
= del_extent_zero(trans
, extent_root
, path
, extent_key
);
7883 ref_path
= kmalloc(sizeof(*ref_path
), GFP_NOFS
);
7889 for (loops
= 0; ; loops
++) {
7891 ret
= btrfs_first_ref_path(trans
, extent_root
, ref_path
,
7892 extent_key
->objectid
);
7894 ret
= btrfs_next_ref_path(trans
, extent_root
, ref_path
);
7901 if (ref_path
->root_objectid
== BTRFS_TREE_LOG_OBJECTID
||
7902 ref_path
->root_objectid
== BTRFS_TREE_RELOC_OBJECTID
)
7905 found_root
= read_ref_root(extent_root
->fs_info
, ref_path
);
7906 BUG_ON(!found_root
);
7908 * for reference counted tree, only process reference paths
7909 * rooted at the latest committed root.
7911 if (found_root
->ref_cows
&&
7912 ref_path
->root_generation
!= found_root
->root_key
.offset
)
7915 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
7918 * copy data extents to new locations
7920 u64 group_start
= group
->key
.objectid
;
7921 ret
= relocate_data_extent(reloc_inode
,
7930 level
= ref_path
->owner_objectid
;
7933 if (prev_block
!= ref_path
->nodes
[level
]) {
7934 struct extent_buffer
*eb
;
7935 u64 block_start
= ref_path
->nodes
[level
];
7936 u64 block_size
= btrfs_level_size(found_root
, level
);
7938 eb
= read_tree_block(found_root
, block_start
,
7944 btrfs_tree_lock(eb
);
7945 BUG_ON(level
!= btrfs_header_level(eb
));
7948 btrfs_item_key_to_cpu(eb
, &first_key
, 0);
7950 btrfs_node_key_to_cpu(eb
, &first_key
, 0);
7952 btrfs_tree_unlock(eb
);
7953 free_extent_buffer(eb
);
7954 prev_block
= block_start
;
7957 mutex_lock(&extent_root
->fs_info
->trans_mutex
);
7958 btrfs_record_root_in_trans(found_root
);
7959 mutex_unlock(&extent_root
->fs_info
->trans_mutex
);
7960 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
7962 * try to update data extent references while
7963 * keeping metadata shared between snapshots.
7966 ret
= relocate_one_path(trans
, found_root
,
7967 path
, &first_key
, ref_path
,
7968 group
, reloc_inode
);
7974 * use fallback method to process the remaining
7978 u64 group_start
= group
->key
.objectid
;
7979 new_extents
= kmalloc(sizeof(*new_extents
),
7982 ret
= get_new_locations(reloc_inode
,
7990 ret
= replace_one_extent(trans
, found_root
,
7992 &first_key
, ref_path
,
7993 new_extents
, nr_extents
);
7995 ret
= relocate_tree_block(trans
, found_root
, path
,
7996 &first_key
, ref_path
);
8003 btrfs_end_transaction(trans
, extent_root
);
8010 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
8013 u64 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
8014 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
8017 * we add in the count of missing devices because we want
8018 * to make sure that any RAID levels on a degraded FS
8019 * continue to be honored.
8021 num_devices
= root
->fs_info
->fs_devices
->rw_devices
+
8022 root
->fs_info
->fs_devices
->missing_devices
;
8024 if (num_devices
== 1) {
8025 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
8026 stripped
= flags
& ~stripped
;
8028 /* turn raid0 into single device chunks */
8029 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
8032 /* turn mirroring into duplication */
8033 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
8034 BTRFS_BLOCK_GROUP_RAID10
))
8035 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
8038 /* they already had raid on here, just return */
8039 if (flags
& stripped
)
8042 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
8043 stripped
= flags
& ~stripped
;
8045 /* switch duplicated blocks with raid1 */
8046 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
8047 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
8049 /* turn single device chunks into raid0 */
8050 return stripped
| BTRFS_BLOCK_GROUP_RAID0
;
8055 static int set_block_group_ro(struct btrfs_block_group_cache
*cache
)
8057 struct btrfs_space_info
*sinfo
= cache
->space_info
;
8064 spin_lock(&sinfo
->lock
);
8065 spin_lock(&cache
->lock
);
8066 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
8067 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
8069 if (sinfo
->bytes_used
+ sinfo
->bytes_reserved
+ sinfo
->bytes_pinned
+
8070 sinfo
->bytes_may_use
+ sinfo
->bytes_readonly
+
8071 cache
->reserved_pinned
+ num_bytes
<= sinfo
->total_bytes
) {
8072 sinfo
->bytes_readonly
+= num_bytes
;
8073 sinfo
->bytes_reserved
+= cache
->reserved_pinned
;
8074 cache
->reserved_pinned
= 0;
8079 spin_unlock(&cache
->lock
);
8080 spin_unlock(&sinfo
->lock
);
8084 int btrfs_set_block_group_ro(struct btrfs_root
*root
,
8085 struct btrfs_block_group_cache
*cache
)
8088 struct btrfs_trans_handle
*trans
;
8094 trans
= btrfs_join_transaction(root
, 1);
8095 BUG_ON(IS_ERR(trans
));
8097 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
8098 if (alloc_flags
!= cache
->flags
)
8099 do_chunk_alloc(trans
, root
, 2 * 1024 * 1024, alloc_flags
, 1);
8101 ret
= set_block_group_ro(cache
);
8104 alloc_flags
= get_alloc_profile(root
, cache
->space_info
->flags
);
8105 ret
= do_chunk_alloc(trans
, root
, 2 * 1024 * 1024, alloc_flags
, 1);
8108 ret
= set_block_group_ro(cache
);
8110 btrfs_end_transaction(trans
, root
);
8114 int btrfs_force_chunk_alloc(struct btrfs_trans_handle
*trans
,
8115 struct btrfs_root
*root
, u64 type
)
8117 u64 alloc_flags
= get_alloc_profile(root
, type
);
8118 return do_chunk_alloc(trans
, root
, 2 * 1024 * 1024, alloc_flags
, 1);
8122 * helper to account the unused space of all the readonly block group in the
8123 * list. takes mirrors into account.
8125 static u64
__btrfs_get_ro_block_group_free_space(struct list_head
*groups_list
)
8127 struct btrfs_block_group_cache
*block_group
;
8131 list_for_each_entry(block_group
, groups_list
, list
) {
8132 spin_lock(&block_group
->lock
);
8134 if (!block_group
->ro
) {
8135 spin_unlock(&block_group
->lock
);
8139 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_RAID1
|
8140 BTRFS_BLOCK_GROUP_RAID10
|
8141 BTRFS_BLOCK_GROUP_DUP
))
8146 free_bytes
+= (block_group
->key
.offset
-
8147 btrfs_block_group_used(&block_group
->item
)) *
8150 spin_unlock(&block_group
->lock
);
8157 * helper to account the unused space of all the readonly block group in the
8158 * space_info. takes mirrors into account.
8160 u64
btrfs_account_ro_block_groups_free_space(struct btrfs_space_info
*sinfo
)
8165 spin_lock(&sinfo
->lock
);
8167 for(i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
8168 if (!list_empty(&sinfo
->block_groups
[i
]))
8169 free_bytes
+= __btrfs_get_ro_block_group_free_space(
8170 &sinfo
->block_groups
[i
]);
8172 spin_unlock(&sinfo
->lock
);
8177 int btrfs_set_block_group_rw(struct btrfs_root
*root
,
8178 struct btrfs_block_group_cache
*cache
)
8180 struct btrfs_space_info
*sinfo
= cache
->space_info
;
8185 spin_lock(&sinfo
->lock
);
8186 spin_lock(&cache
->lock
);
8187 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
8188 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
8189 sinfo
->bytes_readonly
-= num_bytes
;
8191 spin_unlock(&cache
->lock
);
8192 spin_unlock(&sinfo
->lock
);
8197 * checks to see if its even possible to relocate this block group.
8199 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8200 * ok to go ahead and try.
8202 int btrfs_can_relocate(struct btrfs_root
*root
, u64 bytenr
)
8204 struct btrfs_block_group_cache
*block_group
;
8205 struct btrfs_space_info
*space_info
;
8206 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
8207 struct btrfs_device
*device
;
8211 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
8213 /* odd, couldn't find the block group, leave it alone */
8217 /* no bytes used, we're good */
8218 if (!btrfs_block_group_used(&block_group
->item
))
8221 space_info
= block_group
->space_info
;
8222 spin_lock(&space_info
->lock
);
8224 full
= space_info
->full
;
8227 * if this is the last block group we have in this space, we can't
8228 * relocate it unless we're able to allocate a new chunk below.
8230 * Otherwise, we need to make sure we have room in the space to handle
8231 * all of the extents from this block group. If we can, we're good
8233 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
8234 (space_info
->bytes_used
+ space_info
->bytes_reserved
+
8235 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
8236 btrfs_block_group_used(&block_group
->item
) <
8237 space_info
->total_bytes
)) {
8238 spin_unlock(&space_info
->lock
);
8241 spin_unlock(&space_info
->lock
);
8244 * ok we don't have enough space, but maybe we have free space on our
8245 * devices to allocate new chunks for relocation, so loop through our
8246 * alloc devices and guess if we have enough space. However, if we
8247 * were marked as full, then we know there aren't enough chunks, and we
8254 mutex_lock(&root
->fs_info
->chunk_mutex
);
8255 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
8256 u64 min_free
= btrfs_block_group_used(&block_group
->item
);
8260 * check to make sure we can actually find a chunk with enough
8261 * space to fit our block group in.
8263 if (device
->total_bytes
> device
->bytes_used
+ min_free
) {
8264 ret
= find_free_dev_extent(NULL
, device
, min_free
,
8271 mutex_unlock(&root
->fs_info
->chunk_mutex
);
8273 btrfs_put_block_group(block_group
);
8277 static int find_first_block_group(struct btrfs_root
*root
,
8278 struct btrfs_path
*path
, struct btrfs_key
*key
)
8281 struct btrfs_key found_key
;
8282 struct extent_buffer
*leaf
;
8285 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
8290 slot
= path
->slots
[0];
8291 leaf
= path
->nodes
[0];
8292 if (slot
>= btrfs_header_nritems(leaf
)) {
8293 ret
= btrfs_next_leaf(root
, path
);
8300 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
8302 if (found_key
.objectid
>= key
->objectid
&&
8303 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
8313 void btrfs_put_block_group_cache(struct btrfs_fs_info
*info
)
8315 struct btrfs_block_group_cache
*block_group
;
8319 struct inode
*inode
;
8321 block_group
= btrfs_lookup_first_block_group(info
, last
);
8322 while (block_group
) {
8323 spin_lock(&block_group
->lock
);
8324 if (block_group
->iref
)
8326 spin_unlock(&block_group
->lock
);
8327 block_group
= next_block_group(info
->tree_root
,
8337 inode
= block_group
->inode
;
8338 block_group
->iref
= 0;
8339 block_group
->inode
= NULL
;
8340 spin_unlock(&block_group
->lock
);
8342 last
= block_group
->key
.objectid
+ block_group
->key
.offset
;
8343 btrfs_put_block_group(block_group
);
8347 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
8349 struct btrfs_block_group_cache
*block_group
;
8350 struct btrfs_space_info
*space_info
;
8351 struct btrfs_caching_control
*caching_ctl
;
8354 down_write(&info
->extent_commit_sem
);
8355 while (!list_empty(&info
->caching_block_groups
)) {
8356 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
8357 struct btrfs_caching_control
, list
);
8358 list_del(&caching_ctl
->list
);
8359 put_caching_control(caching_ctl
);
8361 up_write(&info
->extent_commit_sem
);
8363 spin_lock(&info
->block_group_cache_lock
);
8364 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
8365 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
8367 rb_erase(&block_group
->cache_node
,
8368 &info
->block_group_cache_tree
);
8369 spin_unlock(&info
->block_group_cache_lock
);
8371 down_write(&block_group
->space_info
->groups_sem
);
8372 list_del(&block_group
->list
);
8373 up_write(&block_group
->space_info
->groups_sem
);
8375 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
8376 wait_block_group_cache_done(block_group
);
8379 * We haven't cached this block group, which means we could
8380 * possibly have excluded extents on this block group.
8382 if (block_group
->cached
== BTRFS_CACHE_NO
)
8383 free_excluded_extents(info
->extent_root
, block_group
);
8385 btrfs_remove_free_space_cache(block_group
);
8386 btrfs_put_block_group(block_group
);
8388 spin_lock(&info
->block_group_cache_lock
);
8390 spin_unlock(&info
->block_group_cache_lock
);
8392 /* now that all the block groups are freed, go through and
8393 * free all the space_info structs. This is only called during
8394 * the final stages of unmount, and so we know nobody is
8395 * using them. We call synchronize_rcu() once before we start,
8396 * just to be on the safe side.
8400 release_global_block_rsv(info
);
8402 while(!list_empty(&info
->space_info
)) {
8403 space_info
= list_entry(info
->space_info
.next
,
8404 struct btrfs_space_info
,
8406 if (space_info
->bytes_pinned
> 0 ||
8407 space_info
->bytes_reserved
> 0) {
8409 dump_space_info(space_info
, 0, 0);
8411 list_del(&space_info
->list
);
8417 static void __link_block_group(struct btrfs_space_info
*space_info
,
8418 struct btrfs_block_group_cache
*cache
)
8420 int index
= get_block_group_index(cache
);
8422 down_write(&space_info
->groups_sem
);
8423 list_add_tail(&cache
->list
, &space_info
->block_groups
[index
]);
8424 up_write(&space_info
->groups_sem
);
8427 int btrfs_read_block_groups(struct btrfs_root
*root
)
8429 struct btrfs_path
*path
;
8431 struct btrfs_block_group_cache
*cache
;
8432 struct btrfs_fs_info
*info
= root
->fs_info
;
8433 struct btrfs_space_info
*space_info
;
8434 struct btrfs_key key
;
8435 struct btrfs_key found_key
;
8436 struct extent_buffer
*leaf
;
8440 root
= info
->extent_root
;
8443 btrfs_set_key_type(&key
, BTRFS_BLOCK_GROUP_ITEM_KEY
);
8444 path
= btrfs_alloc_path();
8448 cache_gen
= btrfs_super_cache_generation(&root
->fs_info
->super_copy
);
8449 if (cache_gen
!= 0 &&
8450 btrfs_super_generation(&root
->fs_info
->super_copy
) != cache_gen
)
8452 if (btrfs_test_opt(root
, CLEAR_CACHE
))
8454 if (!btrfs_test_opt(root
, SPACE_CACHE
) && cache_gen
)
8455 printk(KERN_INFO
"btrfs: disk space caching is enabled\n");
8458 ret
= find_first_block_group(root
, path
, &key
);
8463 leaf
= path
->nodes
[0];
8464 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
8465 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
8471 atomic_set(&cache
->count
, 1);
8472 spin_lock_init(&cache
->lock
);
8473 spin_lock_init(&cache
->tree_lock
);
8474 cache
->fs_info
= info
;
8475 INIT_LIST_HEAD(&cache
->list
);
8476 INIT_LIST_HEAD(&cache
->cluster_list
);
8479 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
8482 * we only want to have 32k of ram per block group for keeping
8483 * track of free space, and if we pass 1/2 of that we want to
8484 * start converting things over to using bitmaps
8486 cache
->extents_thresh
= ((1024 * 32) / 2) /
8487 sizeof(struct btrfs_free_space
);
8489 read_extent_buffer(leaf
, &cache
->item
,
8490 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
8491 sizeof(cache
->item
));
8492 memcpy(&cache
->key
, &found_key
, sizeof(found_key
));
8494 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
8495 btrfs_release_path(root
, path
);
8496 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
8497 cache
->sectorsize
= root
->sectorsize
;
8500 * We need to exclude the super stripes now so that the space
8501 * info has super bytes accounted for, otherwise we'll think
8502 * we have more space than we actually do.
8504 exclude_super_stripes(root
, cache
);
8507 * check for two cases, either we are full, and therefore
8508 * don't need to bother with the caching work since we won't
8509 * find any space, or we are empty, and we can just add all
8510 * the space in and be done with it. This saves us _alot_ of
8511 * time, particularly in the full case.
8513 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
8514 cache
->last_byte_to_unpin
= (u64
)-1;
8515 cache
->cached
= BTRFS_CACHE_FINISHED
;
8516 free_excluded_extents(root
, cache
);
8517 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
8518 cache
->last_byte_to_unpin
= (u64
)-1;
8519 cache
->cached
= BTRFS_CACHE_FINISHED
;
8520 add_new_free_space(cache
, root
->fs_info
,
8522 found_key
.objectid
+
8524 free_excluded_extents(root
, cache
);
8527 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
8528 btrfs_block_group_used(&cache
->item
),
8531 cache
->space_info
= space_info
;
8532 spin_lock(&cache
->space_info
->lock
);
8533 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
8534 spin_unlock(&cache
->space_info
->lock
);
8536 __link_block_group(space_info
, cache
);
8538 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
8541 set_avail_alloc_bits(root
->fs_info
, cache
->flags
);
8542 if (btrfs_chunk_readonly(root
, cache
->key
.objectid
))
8543 set_block_group_ro(cache
);
8546 list_for_each_entry_rcu(space_info
, &root
->fs_info
->space_info
, list
) {
8547 if (!(get_alloc_profile(root
, space_info
->flags
) &
8548 (BTRFS_BLOCK_GROUP_RAID10
|
8549 BTRFS_BLOCK_GROUP_RAID1
|
8550 BTRFS_BLOCK_GROUP_DUP
)))
8553 * avoid allocating from un-mirrored block group if there are
8554 * mirrored block groups.
8556 list_for_each_entry(cache
, &space_info
->block_groups
[3], list
)
8557 set_block_group_ro(cache
);
8558 list_for_each_entry(cache
, &space_info
->block_groups
[4], list
)
8559 set_block_group_ro(cache
);
8562 init_global_block_rsv(info
);
8565 btrfs_free_path(path
);
8569 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
8570 struct btrfs_root
*root
, u64 bytes_used
,
8571 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
8575 struct btrfs_root
*extent_root
;
8576 struct btrfs_block_group_cache
*cache
;
8578 extent_root
= root
->fs_info
->extent_root
;
8580 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
8582 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
8586 cache
->key
.objectid
= chunk_offset
;
8587 cache
->key
.offset
= size
;
8588 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
8589 cache
->sectorsize
= root
->sectorsize
;
8590 cache
->fs_info
= root
->fs_info
;
8593 * we only want to have 32k of ram per block group for keeping track
8594 * of free space, and if we pass 1/2 of that we want to start
8595 * converting things over to using bitmaps
8597 cache
->extents_thresh
= ((1024 * 32) / 2) /
8598 sizeof(struct btrfs_free_space
);
8599 atomic_set(&cache
->count
, 1);
8600 spin_lock_init(&cache
->lock
);
8601 spin_lock_init(&cache
->tree_lock
);
8602 INIT_LIST_HEAD(&cache
->list
);
8603 INIT_LIST_HEAD(&cache
->cluster_list
);
8605 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
8606 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
8607 cache
->flags
= type
;
8608 btrfs_set_block_group_flags(&cache
->item
, type
);
8610 cache
->last_byte_to_unpin
= (u64
)-1;
8611 cache
->cached
= BTRFS_CACHE_FINISHED
;
8612 exclude_super_stripes(root
, cache
);
8614 add_new_free_space(cache
, root
->fs_info
, chunk_offset
,
8615 chunk_offset
+ size
);
8617 free_excluded_extents(root
, cache
);
8619 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
8620 &cache
->space_info
);
8623 spin_lock(&cache
->space_info
->lock
);
8624 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
8625 spin_unlock(&cache
->space_info
->lock
);
8627 __link_block_group(cache
->space_info
, cache
);
8629 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
8632 ret
= btrfs_insert_item(trans
, extent_root
, &cache
->key
, &cache
->item
,
8633 sizeof(cache
->item
));
8636 set_avail_alloc_bits(extent_root
->fs_info
, type
);
8641 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
8642 struct btrfs_root
*root
, u64 group_start
)
8644 struct btrfs_path
*path
;
8645 struct btrfs_block_group_cache
*block_group
;
8646 struct btrfs_free_cluster
*cluster
;
8647 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
8648 struct btrfs_key key
;
8649 struct inode
*inode
;
8653 root
= root
->fs_info
->extent_root
;
8655 block_group
= btrfs_lookup_block_group(root
->fs_info
, group_start
);
8656 BUG_ON(!block_group
);
8657 BUG_ON(!block_group
->ro
);
8659 memcpy(&key
, &block_group
->key
, sizeof(key
));
8660 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
8661 BTRFS_BLOCK_GROUP_RAID1
|
8662 BTRFS_BLOCK_GROUP_RAID10
))
8667 /* make sure this block group isn't part of an allocation cluster */
8668 cluster
= &root
->fs_info
->data_alloc_cluster
;
8669 spin_lock(&cluster
->refill_lock
);
8670 btrfs_return_cluster_to_free_space(block_group
, cluster
);
8671 spin_unlock(&cluster
->refill_lock
);
8674 * make sure this block group isn't part of a metadata
8675 * allocation cluster
8677 cluster
= &root
->fs_info
->meta_alloc_cluster
;
8678 spin_lock(&cluster
->refill_lock
);
8679 btrfs_return_cluster_to_free_space(block_group
, cluster
);
8680 spin_unlock(&cluster
->refill_lock
);
8682 path
= btrfs_alloc_path();
8685 inode
= lookup_free_space_inode(root
, block_group
, path
);
8686 if (!IS_ERR(inode
)) {
8687 btrfs_orphan_add(trans
, inode
);
8689 /* One for the block groups ref */
8690 spin_lock(&block_group
->lock
);
8691 if (block_group
->iref
) {
8692 block_group
->iref
= 0;
8693 block_group
->inode
= NULL
;
8694 spin_unlock(&block_group
->lock
);
8697 spin_unlock(&block_group
->lock
);
8699 /* One for our lookup ref */
8703 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
8704 key
.offset
= block_group
->key
.objectid
;
8707 ret
= btrfs_search_slot(trans
, tree_root
, &key
, path
, -1, 1);
8711 btrfs_release_path(tree_root
, path
);
8713 ret
= btrfs_del_item(trans
, tree_root
, path
);
8716 btrfs_release_path(tree_root
, path
);
8719 spin_lock(&root
->fs_info
->block_group_cache_lock
);
8720 rb_erase(&block_group
->cache_node
,
8721 &root
->fs_info
->block_group_cache_tree
);
8722 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
8724 down_write(&block_group
->space_info
->groups_sem
);
8726 * we must use list_del_init so people can check to see if they
8727 * are still on the list after taking the semaphore
8729 list_del_init(&block_group
->list
);
8730 up_write(&block_group
->space_info
->groups_sem
);
8732 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
8733 wait_block_group_cache_done(block_group
);
8735 btrfs_remove_free_space_cache(block_group
);
8737 spin_lock(&block_group
->space_info
->lock
);
8738 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
8739 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
8740 block_group
->space_info
->disk_total
-= block_group
->key
.offset
* factor
;
8741 spin_unlock(&block_group
->space_info
->lock
);
8743 memcpy(&key
, &block_group
->key
, sizeof(key
));
8745 btrfs_clear_space_info_full(root
->fs_info
);
8747 btrfs_put_block_group(block_group
);
8748 btrfs_put_block_group(block_group
);
8750 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
8756 ret
= btrfs_del_item(trans
, root
, path
);
8758 btrfs_free_path(path
);
8762 int btrfs_error_unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
8764 return unpin_extent_range(root
, start
, end
);
8767 int btrfs_error_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
8770 return btrfs_discard_extent(root
, bytenr
, num_bytes
);