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 update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
40 u64 num_bytes
, int reserve
, int sinfo
);
41 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
42 struct btrfs_root
*root
,
43 u64 bytenr
, u64 num_bytes
, u64 parent
,
44 u64 root_objectid
, u64 owner_objectid
,
45 u64 owner_offset
, int refs_to_drop
,
46 struct btrfs_delayed_extent_op
*extra_op
);
47 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
48 struct extent_buffer
*leaf
,
49 struct btrfs_extent_item
*ei
);
50 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
51 struct btrfs_root
*root
,
52 u64 parent
, u64 root_objectid
,
53 u64 flags
, u64 owner
, u64 offset
,
54 struct btrfs_key
*ins
, int ref_mod
);
55 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
56 struct btrfs_root
*root
,
57 u64 parent
, u64 root_objectid
,
58 u64 flags
, struct btrfs_disk_key
*key
,
59 int level
, struct btrfs_key
*ins
);
60 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
61 struct btrfs_root
*extent_root
, u64 alloc_bytes
,
62 u64 flags
, int force
);
63 static int find_next_key(struct btrfs_path
*path
, int level
,
64 struct btrfs_key
*key
);
65 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
66 int dump_block_groups
);
69 block_group_cache_done(struct btrfs_block_group_cache
*cache
)
72 return cache
->cached
== BTRFS_CACHE_FINISHED
;
75 static int block_group_bits(struct btrfs_block_group_cache
*cache
, u64 bits
)
77 return (cache
->flags
& bits
) == bits
;
80 void btrfs_get_block_group(struct btrfs_block_group_cache
*cache
)
82 atomic_inc(&cache
->count
);
85 void btrfs_put_block_group(struct btrfs_block_group_cache
*cache
)
87 if (atomic_dec_and_test(&cache
->count
)) {
88 WARN_ON(cache
->pinned
> 0);
89 WARN_ON(cache
->reserved
> 0);
90 WARN_ON(cache
->reserved_pinned
> 0);
96 * this adds the block group to the fs_info rb tree for the block group
99 static int btrfs_add_block_group_cache(struct btrfs_fs_info
*info
,
100 struct btrfs_block_group_cache
*block_group
)
103 struct rb_node
*parent
= NULL
;
104 struct btrfs_block_group_cache
*cache
;
106 spin_lock(&info
->block_group_cache_lock
);
107 p
= &info
->block_group_cache_tree
.rb_node
;
111 cache
= rb_entry(parent
, struct btrfs_block_group_cache
,
113 if (block_group
->key
.objectid
< cache
->key
.objectid
) {
115 } else if (block_group
->key
.objectid
> cache
->key
.objectid
) {
118 spin_unlock(&info
->block_group_cache_lock
);
123 rb_link_node(&block_group
->cache_node
, parent
, p
);
124 rb_insert_color(&block_group
->cache_node
,
125 &info
->block_group_cache_tree
);
126 spin_unlock(&info
->block_group_cache_lock
);
132 * This will return the block group at or after bytenr if contains is 0, else
133 * it will return the block group that contains the bytenr
135 static struct btrfs_block_group_cache
*
136 block_group_cache_tree_search(struct btrfs_fs_info
*info
, u64 bytenr
,
139 struct btrfs_block_group_cache
*cache
, *ret
= NULL
;
143 spin_lock(&info
->block_group_cache_lock
);
144 n
= info
->block_group_cache_tree
.rb_node
;
147 cache
= rb_entry(n
, struct btrfs_block_group_cache
,
149 end
= cache
->key
.objectid
+ cache
->key
.offset
- 1;
150 start
= cache
->key
.objectid
;
152 if (bytenr
< start
) {
153 if (!contains
&& (!ret
|| start
< ret
->key
.objectid
))
156 } else if (bytenr
> start
) {
157 if (contains
&& bytenr
<= end
) {
168 btrfs_get_block_group(ret
);
169 spin_unlock(&info
->block_group_cache_lock
);
174 static int add_excluded_extent(struct btrfs_root
*root
,
175 u64 start
, u64 num_bytes
)
177 u64 end
= start
+ num_bytes
- 1;
178 set_extent_bits(&root
->fs_info
->freed_extents
[0],
179 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
180 set_extent_bits(&root
->fs_info
->freed_extents
[1],
181 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
185 static void free_excluded_extents(struct btrfs_root
*root
,
186 struct btrfs_block_group_cache
*cache
)
190 start
= cache
->key
.objectid
;
191 end
= start
+ cache
->key
.offset
- 1;
193 clear_extent_bits(&root
->fs_info
->freed_extents
[0],
194 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
195 clear_extent_bits(&root
->fs_info
->freed_extents
[1],
196 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
199 static int exclude_super_stripes(struct btrfs_root
*root
,
200 struct btrfs_block_group_cache
*cache
)
207 if (cache
->key
.objectid
< BTRFS_SUPER_INFO_OFFSET
) {
208 stripe_len
= BTRFS_SUPER_INFO_OFFSET
- cache
->key
.objectid
;
209 cache
->bytes_super
+= stripe_len
;
210 ret
= add_excluded_extent(root
, cache
->key
.objectid
,
215 for (i
= 0; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
216 bytenr
= btrfs_sb_offset(i
);
217 ret
= btrfs_rmap_block(&root
->fs_info
->mapping_tree
,
218 cache
->key
.objectid
, bytenr
,
219 0, &logical
, &nr
, &stripe_len
);
223 cache
->bytes_super
+= stripe_len
;
224 ret
= add_excluded_extent(root
, logical
[nr
],
234 static struct btrfs_caching_control
*
235 get_caching_control(struct btrfs_block_group_cache
*cache
)
237 struct btrfs_caching_control
*ctl
;
239 spin_lock(&cache
->lock
);
240 if (cache
->cached
!= BTRFS_CACHE_STARTED
) {
241 spin_unlock(&cache
->lock
);
245 /* We're loading it the fast way, so we don't have a caching_ctl. */
246 if (!cache
->caching_ctl
) {
247 spin_unlock(&cache
->lock
);
251 ctl
= cache
->caching_ctl
;
252 atomic_inc(&ctl
->count
);
253 spin_unlock(&cache
->lock
);
257 static void put_caching_control(struct btrfs_caching_control
*ctl
)
259 if (atomic_dec_and_test(&ctl
->count
))
264 * this is only called by cache_block_group, since we could have freed extents
265 * we need to check the pinned_extents for any extents that can't be used yet
266 * since their free space will be released as soon as the transaction commits.
268 static u64
add_new_free_space(struct btrfs_block_group_cache
*block_group
,
269 struct btrfs_fs_info
*info
, u64 start
, u64 end
)
271 u64 extent_start
, extent_end
, size
, total_added
= 0;
274 while (start
< end
) {
275 ret
= find_first_extent_bit(info
->pinned_extents
, start
,
276 &extent_start
, &extent_end
,
277 EXTENT_DIRTY
| EXTENT_UPTODATE
);
281 if (extent_start
<= start
) {
282 start
= extent_end
+ 1;
283 } else if (extent_start
> start
&& extent_start
< end
) {
284 size
= extent_start
- start
;
286 ret
= btrfs_add_free_space(block_group
, start
,
289 start
= extent_end
+ 1;
298 ret
= btrfs_add_free_space(block_group
, start
, size
);
305 static int caching_kthread(void *data
)
307 struct btrfs_block_group_cache
*block_group
= data
;
308 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
309 struct btrfs_caching_control
*caching_ctl
= block_group
->caching_ctl
;
310 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
311 struct btrfs_path
*path
;
312 struct extent_buffer
*leaf
;
313 struct btrfs_key key
;
319 path
= btrfs_alloc_path();
323 exclude_super_stripes(extent_root
, block_group
);
324 spin_lock(&block_group
->space_info
->lock
);
325 block_group
->space_info
->bytes_readonly
+= block_group
->bytes_super
;
326 spin_unlock(&block_group
->space_info
->lock
);
328 last
= max_t(u64
, block_group
->key
.objectid
, BTRFS_SUPER_INFO_OFFSET
);
331 * We don't want to deadlock with somebody trying to allocate a new
332 * extent for the extent root while also trying to search the extent
333 * root to add free space. So we skip locking and search the commit
334 * root, since its read-only
336 path
->skip_locking
= 1;
337 path
->search_commit_root
= 1;
342 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
344 mutex_lock(&caching_ctl
->mutex
);
345 /* need to make sure the commit_root doesn't disappear */
346 down_read(&fs_info
->extent_commit_sem
);
348 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
352 leaf
= path
->nodes
[0];
353 nritems
= btrfs_header_nritems(leaf
);
357 if (fs_info
->closing
> 1) {
362 if (path
->slots
[0] < nritems
) {
363 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
365 ret
= find_next_key(path
, 0, &key
);
369 caching_ctl
->progress
= last
;
370 btrfs_release_path(extent_root
, path
);
371 up_read(&fs_info
->extent_commit_sem
);
372 mutex_unlock(&caching_ctl
->mutex
);
373 if (btrfs_transaction_in_commit(fs_info
))
380 if (key
.objectid
< block_group
->key
.objectid
) {
385 if (key
.objectid
>= block_group
->key
.objectid
+
386 block_group
->key
.offset
)
389 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
) {
390 total_found
+= add_new_free_space(block_group
,
393 last
= key
.objectid
+ key
.offset
;
395 if (total_found
> (1024 * 1024 * 2)) {
397 wake_up(&caching_ctl
->wait
);
404 total_found
+= add_new_free_space(block_group
, fs_info
, last
,
405 block_group
->key
.objectid
+
406 block_group
->key
.offset
);
407 caching_ctl
->progress
= (u64
)-1;
409 spin_lock(&block_group
->lock
);
410 block_group
->caching_ctl
= NULL
;
411 block_group
->cached
= BTRFS_CACHE_FINISHED
;
412 spin_unlock(&block_group
->lock
);
415 btrfs_free_path(path
);
416 up_read(&fs_info
->extent_commit_sem
);
418 free_excluded_extents(extent_root
, block_group
);
420 mutex_unlock(&caching_ctl
->mutex
);
421 wake_up(&caching_ctl
->wait
);
423 put_caching_control(caching_ctl
);
424 atomic_dec(&block_group
->space_info
->caching_threads
);
425 btrfs_put_block_group(block_group
);
430 static int cache_block_group(struct btrfs_block_group_cache
*cache
,
431 struct btrfs_trans_handle
*trans
,
434 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
435 struct btrfs_caching_control
*caching_ctl
;
436 struct task_struct
*tsk
;
440 if (cache
->cached
!= BTRFS_CACHE_NO
)
444 * We can't do the read from on-disk cache during a commit since we need
445 * to have the normal tree locking.
447 if (!trans
->transaction
->in_commit
) {
448 spin_lock(&cache
->lock
);
449 if (cache
->cached
!= BTRFS_CACHE_NO
) {
450 spin_unlock(&cache
->lock
);
453 cache
->cached
= BTRFS_CACHE_STARTED
;
454 spin_unlock(&cache
->lock
);
456 ret
= load_free_space_cache(fs_info
, cache
);
458 spin_lock(&cache
->lock
);
460 cache
->cached
= BTRFS_CACHE_FINISHED
;
461 cache
->last_byte_to_unpin
= (u64
)-1;
463 cache
->cached
= BTRFS_CACHE_NO
;
465 spin_unlock(&cache
->lock
);
473 caching_ctl
= kzalloc(sizeof(*caching_ctl
), GFP_KERNEL
);
474 BUG_ON(!caching_ctl
);
476 INIT_LIST_HEAD(&caching_ctl
->list
);
477 mutex_init(&caching_ctl
->mutex
);
478 init_waitqueue_head(&caching_ctl
->wait
);
479 caching_ctl
->block_group
= cache
;
480 caching_ctl
->progress
= cache
->key
.objectid
;
481 /* one for caching kthread, one for caching block group list */
482 atomic_set(&caching_ctl
->count
, 2);
484 spin_lock(&cache
->lock
);
485 if (cache
->cached
!= BTRFS_CACHE_NO
) {
486 spin_unlock(&cache
->lock
);
490 cache
->caching_ctl
= caching_ctl
;
491 cache
->cached
= BTRFS_CACHE_STARTED
;
492 spin_unlock(&cache
->lock
);
494 down_write(&fs_info
->extent_commit_sem
);
495 list_add_tail(&caching_ctl
->list
, &fs_info
->caching_block_groups
);
496 up_write(&fs_info
->extent_commit_sem
);
498 atomic_inc(&cache
->space_info
->caching_threads
);
499 btrfs_get_block_group(cache
);
501 tsk
= kthread_run(caching_kthread
, cache
, "btrfs-cache-%llu\n",
502 cache
->key
.objectid
);
505 printk(KERN_ERR
"error running thread %d\n", ret
);
513 * return the block group that starts at or after bytenr
515 static struct btrfs_block_group_cache
*
516 btrfs_lookup_first_block_group(struct btrfs_fs_info
*info
, u64 bytenr
)
518 struct btrfs_block_group_cache
*cache
;
520 cache
= block_group_cache_tree_search(info
, bytenr
, 0);
526 * return the block group that contains the given bytenr
528 struct btrfs_block_group_cache
*btrfs_lookup_block_group(
529 struct btrfs_fs_info
*info
,
532 struct btrfs_block_group_cache
*cache
;
534 cache
= block_group_cache_tree_search(info
, bytenr
, 1);
539 static struct btrfs_space_info
*__find_space_info(struct btrfs_fs_info
*info
,
542 struct list_head
*head
= &info
->space_info
;
543 struct btrfs_space_info
*found
;
545 flags
&= BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_SYSTEM
|
546 BTRFS_BLOCK_GROUP_METADATA
;
549 list_for_each_entry_rcu(found
, head
, list
) {
550 if (found
->flags
& flags
) {
560 * after adding space to the filesystem, we need to clear the full flags
561 * on all the space infos.
563 void btrfs_clear_space_info_full(struct btrfs_fs_info
*info
)
565 struct list_head
*head
= &info
->space_info
;
566 struct btrfs_space_info
*found
;
569 list_for_each_entry_rcu(found
, head
, list
)
574 static u64
div_factor(u64 num
, int factor
)
583 static u64
div_factor_fine(u64 num
, int factor
)
592 u64
btrfs_find_block_group(struct btrfs_root
*root
,
593 u64 search_start
, u64 search_hint
, int owner
)
595 struct btrfs_block_group_cache
*cache
;
597 u64 last
= max(search_hint
, search_start
);
604 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
608 spin_lock(&cache
->lock
);
609 last
= cache
->key
.objectid
+ cache
->key
.offset
;
610 used
= btrfs_block_group_used(&cache
->item
);
612 if ((full_search
|| !cache
->ro
) &&
613 block_group_bits(cache
, BTRFS_BLOCK_GROUP_METADATA
)) {
614 if (used
+ cache
->pinned
+ cache
->reserved
<
615 div_factor(cache
->key
.offset
, factor
)) {
616 group_start
= cache
->key
.objectid
;
617 spin_unlock(&cache
->lock
);
618 btrfs_put_block_group(cache
);
622 spin_unlock(&cache
->lock
);
623 btrfs_put_block_group(cache
);
631 if (!full_search
&& factor
< 10) {
641 /* simple helper to search for an existing extent at a given offset */
642 int btrfs_lookup_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
645 struct btrfs_key key
;
646 struct btrfs_path
*path
;
648 path
= btrfs_alloc_path();
650 key
.objectid
= start
;
652 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
653 ret
= btrfs_search_slot(NULL
, root
->fs_info
->extent_root
, &key
, path
,
655 btrfs_free_path(path
);
660 * helper function to lookup reference count and flags of extent.
662 * the head node for delayed ref is used to store the sum of all the
663 * reference count modifications queued up in the rbtree. the head
664 * node may also store the extent flags to set. This way you can check
665 * to see what the reference count and extent flags would be if all of
666 * the delayed refs are not processed.
668 int btrfs_lookup_extent_info(struct btrfs_trans_handle
*trans
,
669 struct btrfs_root
*root
, u64 bytenr
,
670 u64 num_bytes
, u64
*refs
, u64
*flags
)
672 struct btrfs_delayed_ref_head
*head
;
673 struct btrfs_delayed_ref_root
*delayed_refs
;
674 struct btrfs_path
*path
;
675 struct btrfs_extent_item
*ei
;
676 struct extent_buffer
*leaf
;
677 struct btrfs_key key
;
683 path
= btrfs_alloc_path();
687 key
.objectid
= bytenr
;
688 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
689 key
.offset
= num_bytes
;
691 path
->skip_locking
= 1;
692 path
->search_commit_root
= 1;
695 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
,
701 leaf
= path
->nodes
[0];
702 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
703 if (item_size
>= sizeof(*ei
)) {
704 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
705 struct btrfs_extent_item
);
706 num_refs
= btrfs_extent_refs(leaf
, ei
);
707 extent_flags
= btrfs_extent_flags(leaf
, ei
);
709 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
710 struct btrfs_extent_item_v0
*ei0
;
711 BUG_ON(item_size
!= sizeof(*ei0
));
712 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
713 struct btrfs_extent_item_v0
);
714 num_refs
= btrfs_extent_refs_v0(leaf
, ei0
);
715 /* FIXME: this isn't correct for data */
716 extent_flags
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
721 BUG_ON(num_refs
== 0);
731 delayed_refs
= &trans
->transaction
->delayed_refs
;
732 spin_lock(&delayed_refs
->lock
);
733 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
735 if (!mutex_trylock(&head
->mutex
)) {
736 atomic_inc(&head
->node
.refs
);
737 spin_unlock(&delayed_refs
->lock
);
739 btrfs_release_path(root
->fs_info
->extent_root
, path
);
741 mutex_lock(&head
->mutex
);
742 mutex_unlock(&head
->mutex
);
743 btrfs_put_delayed_ref(&head
->node
);
746 if (head
->extent_op
&& head
->extent_op
->update_flags
)
747 extent_flags
|= head
->extent_op
->flags_to_set
;
749 BUG_ON(num_refs
== 0);
751 num_refs
+= head
->node
.ref_mod
;
752 mutex_unlock(&head
->mutex
);
754 spin_unlock(&delayed_refs
->lock
);
756 WARN_ON(num_refs
== 0);
760 *flags
= extent_flags
;
762 btrfs_free_path(path
);
767 * Back reference rules. Back refs have three main goals:
769 * 1) differentiate between all holders of references to an extent so that
770 * when a reference is dropped we can make sure it was a valid reference
771 * before freeing the extent.
773 * 2) Provide enough information to quickly find the holders of an extent
774 * if we notice a given block is corrupted or bad.
776 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
777 * maintenance. This is actually the same as #2, but with a slightly
778 * different use case.
780 * There are two kinds of back refs. The implicit back refs is optimized
781 * for pointers in non-shared tree blocks. For a given pointer in a block,
782 * back refs of this kind provide information about the block's owner tree
783 * and the pointer's key. These information allow us to find the block by
784 * b-tree searching. The full back refs is for pointers in tree blocks not
785 * referenced by their owner trees. The location of tree block is recorded
786 * in the back refs. Actually the full back refs is generic, and can be
787 * used in all cases the implicit back refs is used. The major shortcoming
788 * of the full back refs is its overhead. Every time a tree block gets
789 * COWed, we have to update back refs entry for all pointers in it.
791 * For a newly allocated tree block, we use implicit back refs for
792 * pointers in it. This means most tree related operations only involve
793 * implicit back refs. For a tree block created in old transaction, the
794 * only way to drop a reference to it is COW it. So we can detect the
795 * event that tree block loses its owner tree's reference and do the
796 * back refs conversion.
798 * When a tree block is COW'd through a tree, there are four cases:
800 * The reference count of the block is one and the tree is the block's
801 * owner tree. Nothing to do in this case.
803 * The reference count of the block is one and the tree is not the
804 * block's owner tree. In this case, full back refs is used for pointers
805 * in the block. Remove these full back refs, add implicit back refs for
806 * every pointers in the new block.
808 * The reference count of the block is greater than one and the tree is
809 * the block's owner tree. In this case, implicit back refs is used for
810 * pointers in the block. Add full back refs for every pointers in the
811 * block, increase lower level extents' reference counts. The original
812 * implicit back refs are entailed to the new block.
814 * The reference count of the block is greater than one and the tree is
815 * not the block's owner tree. Add implicit back refs for every pointer in
816 * the new block, increase lower level extents' reference count.
818 * Back Reference Key composing:
820 * The key objectid corresponds to the first byte in the extent,
821 * The key type is used to differentiate between types of back refs.
822 * There are different meanings of the key offset for different types
825 * File extents can be referenced by:
827 * - multiple snapshots, subvolumes, or different generations in one subvol
828 * - different files inside a single subvolume
829 * - different offsets inside a file (bookend extents in file.c)
831 * The extent ref structure for the implicit back refs has fields for:
833 * - Objectid of the subvolume root
834 * - objectid of the file holding the reference
835 * - original offset in the file
836 * - how many bookend extents
838 * The key offset for the implicit back refs is hash of the first
841 * The extent ref structure for the full back refs has field for:
843 * - number of pointers in the tree leaf
845 * The key offset for the implicit back refs is the first byte of
848 * When a file extent is allocated, The implicit back refs is used.
849 * the fields are filled in:
851 * (root_key.objectid, inode objectid, offset in file, 1)
853 * When a file extent is removed file truncation, we find the
854 * corresponding implicit back refs and check the following fields:
856 * (btrfs_header_owner(leaf), inode objectid, offset in file)
858 * Btree extents can be referenced by:
860 * - Different subvolumes
862 * Both the implicit back refs and the full back refs for tree blocks
863 * only consist of key. The key offset for the implicit back refs is
864 * objectid of block's owner tree. The key offset for the full back refs
865 * is the first byte of parent block.
867 * When implicit back refs is used, information about the lowest key and
868 * level of the tree block are required. These information are stored in
869 * tree block info structure.
872 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
873 static int convert_extent_item_v0(struct btrfs_trans_handle
*trans
,
874 struct btrfs_root
*root
,
875 struct btrfs_path
*path
,
876 u64 owner
, u32 extra_size
)
878 struct btrfs_extent_item
*item
;
879 struct btrfs_extent_item_v0
*ei0
;
880 struct btrfs_extent_ref_v0
*ref0
;
881 struct btrfs_tree_block_info
*bi
;
882 struct extent_buffer
*leaf
;
883 struct btrfs_key key
;
884 struct btrfs_key found_key
;
885 u32 new_size
= sizeof(*item
);
889 leaf
= path
->nodes
[0];
890 BUG_ON(btrfs_item_size_nr(leaf
, path
->slots
[0]) != sizeof(*ei0
));
892 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
893 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
894 struct btrfs_extent_item_v0
);
895 refs
= btrfs_extent_refs_v0(leaf
, ei0
);
897 if (owner
== (u64
)-1) {
899 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
900 ret
= btrfs_next_leaf(root
, path
);
904 leaf
= path
->nodes
[0];
906 btrfs_item_key_to_cpu(leaf
, &found_key
,
908 BUG_ON(key
.objectid
!= found_key
.objectid
);
909 if (found_key
.type
!= BTRFS_EXTENT_REF_V0_KEY
) {
913 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
914 struct btrfs_extent_ref_v0
);
915 owner
= btrfs_ref_objectid_v0(leaf
, ref0
);
919 btrfs_release_path(root
, path
);
921 if (owner
< BTRFS_FIRST_FREE_OBJECTID
)
922 new_size
+= sizeof(*bi
);
924 new_size
-= sizeof(*ei0
);
925 ret
= btrfs_search_slot(trans
, root
, &key
, path
,
926 new_size
+ extra_size
, 1);
931 ret
= btrfs_extend_item(trans
, root
, path
, new_size
);
934 leaf
= path
->nodes
[0];
935 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
936 btrfs_set_extent_refs(leaf
, item
, refs
);
937 /* FIXME: get real generation */
938 btrfs_set_extent_generation(leaf
, item
, 0);
939 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
940 btrfs_set_extent_flags(leaf
, item
,
941 BTRFS_EXTENT_FLAG_TREE_BLOCK
|
942 BTRFS_BLOCK_FLAG_FULL_BACKREF
);
943 bi
= (struct btrfs_tree_block_info
*)(item
+ 1);
944 /* FIXME: get first key of the block */
945 memset_extent_buffer(leaf
, 0, (unsigned long)bi
, sizeof(*bi
));
946 btrfs_set_tree_block_level(leaf
, bi
, (int)owner
);
948 btrfs_set_extent_flags(leaf
, item
, BTRFS_EXTENT_FLAG_DATA
);
950 btrfs_mark_buffer_dirty(leaf
);
955 static u64
hash_extent_data_ref(u64 root_objectid
, u64 owner
, u64 offset
)
957 u32 high_crc
= ~(u32
)0;
958 u32 low_crc
= ~(u32
)0;
961 lenum
= cpu_to_le64(root_objectid
);
962 high_crc
= crc32c(high_crc
, &lenum
, sizeof(lenum
));
963 lenum
= cpu_to_le64(owner
);
964 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
965 lenum
= cpu_to_le64(offset
);
966 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
968 return ((u64
)high_crc
<< 31) ^ (u64
)low_crc
;
971 static u64
hash_extent_data_ref_item(struct extent_buffer
*leaf
,
972 struct btrfs_extent_data_ref
*ref
)
974 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf
, ref
),
975 btrfs_extent_data_ref_objectid(leaf
, ref
),
976 btrfs_extent_data_ref_offset(leaf
, ref
));
979 static int match_extent_data_ref(struct extent_buffer
*leaf
,
980 struct btrfs_extent_data_ref
*ref
,
981 u64 root_objectid
, u64 owner
, u64 offset
)
983 if (btrfs_extent_data_ref_root(leaf
, ref
) != root_objectid
||
984 btrfs_extent_data_ref_objectid(leaf
, ref
) != owner
||
985 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
990 static noinline
int lookup_extent_data_ref(struct btrfs_trans_handle
*trans
,
991 struct btrfs_root
*root
,
992 struct btrfs_path
*path
,
993 u64 bytenr
, u64 parent
,
995 u64 owner
, u64 offset
)
997 struct btrfs_key key
;
998 struct btrfs_extent_data_ref
*ref
;
999 struct extent_buffer
*leaf
;
1005 key
.objectid
= bytenr
;
1007 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1008 key
.offset
= parent
;
1010 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1011 key
.offset
= hash_extent_data_ref(root_objectid
,
1016 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1025 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1026 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1027 btrfs_release_path(root
, path
);
1028 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1039 leaf
= path
->nodes
[0];
1040 nritems
= btrfs_header_nritems(leaf
);
1042 if (path
->slots
[0] >= nritems
) {
1043 ret
= btrfs_next_leaf(root
, path
);
1049 leaf
= path
->nodes
[0];
1050 nritems
= btrfs_header_nritems(leaf
);
1054 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1055 if (key
.objectid
!= bytenr
||
1056 key
.type
!= BTRFS_EXTENT_DATA_REF_KEY
)
1059 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1060 struct btrfs_extent_data_ref
);
1062 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1065 btrfs_release_path(root
, path
);
1077 static noinline
int insert_extent_data_ref(struct btrfs_trans_handle
*trans
,
1078 struct btrfs_root
*root
,
1079 struct btrfs_path
*path
,
1080 u64 bytenr
, u64 parent
,
1081 u64 root_objectid
, u64 owner
,
1082 u64 offset
, int refs_to_add
)
1084 struct btrfs_key key
;
1085 struct extent_buffer
*leaf
;
1090 key
.objectid
= bytenr
;
1092 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1093 key
.offset
= parent
;
1094 size
= sizeof(struct btrfs_shared_data_ref
);
1096 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1097 key
.offset
= hash_extent_data_ref(root_objectid
,
1099 size
= sizeof(struct btrfs_extent_data_ref
);
1102 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, size
);
1103 if (ret
&& ret
!= -EEXIST
)
1106 leaf
= path
->nodes
[0];
1108 struct btrfs_shared_data_ref
*ref
;
1109 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1110 struct btrfs_shared_data_ref
);
1112 btrfs_set_shared_data_ref_count(leaf
, ref
, refs_to_add
);
1114 num_refs
= btrfs_shared_data_ref_count(leaf
, ref
);
1115 num_refs
+= refs_to_add
;
1116 btrfs_set_shared_data_ref_count(leaf
, ref
, num_refs
);
1119 struct btrfs_extent_data_ref
*ref
;
1120 while (ret
== -EEXIST
) {
1121 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1122 struct btrfs_extent_data_ref
);
1123 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1126 btrfs_release_path(root
, path
);
1128 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1130 if (ret
&& ret
!= -EEXIST
)
1133 leaf
= path
->nodes
[0];
1135 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1136 struct btrfs_extent_data_ref
);
1138 btrfs_set_extent_data_ref_root(leaf
, ref
,
1140 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
1141 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
1142 btrfs_set_extent_data_ref_count(leaf
, ref
, refs_to_add
);
1144 num_refs
= btrfs_extent_data_ref_count(leaf
, ref
);
1145 num_refs
+= refs_to_add
;
1146 btrfs_set_extent_data_ref_count(leaf
, ref
, num_refs
);
1149 btrfs_mark_buffer_dirty(leaf
);
1152 btrfs_release_path(root
, path
);
1156 static noinline
int remove_extent_data_ref(struct btrfs_trans_handle
*trans
,
1157 struct btrfs_root
*root
,
1158 struct btrfs_path
*path
,
1161 struct btrfs_key key
;
1162 struct btrfs_extent_data_ref
*ref1
= NULL
;
1163 struct btrfs_shared_data_ref
*ref2
= NULL
;
1164 struct extent_buffer
*leaf
;
1168 leaf
= path
->nodes
[0];
1169 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1171 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1172 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1173 struct btrfs_extent_data_ref
);
1174 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1175 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1176 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1177 struct btrfs_shared_data_ref
);
1178 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1179 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1180 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1181 struct btrfs_extent_ref_v0
*ref0
;
1182 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1183 struct btrfs_extent_ref_v0
);
1184 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1190 BUG_ON(num_refs
< refs_to_drop
);
1191 num_refs
-= refs_to_drop
;
1193 if (num_refs
== 0) {
1194 ret
= btrfs_del_item(trans
, root
, path
);
1196 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
)
1197 btrfs_set_extent_data_ref_count(leaf
, ref1
, num_refs
);
1198 else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
)
1199 btrfs_set_shared_data_ref_count(leaf
, ref2
, num_refs
);
1200 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1202 struct btrfs_extent_ref_v0
*ref0
;
1203 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1204 struct btrfs_extent_ref_v0
);
1205 btrfs_set_ref_count_v0(leaf
, ref0
, num_refs
);
1208 btrfs_mark_buffer_dirty(leaf
);
1213 static noinline u32
extent_data_ref_count(struct btrfs_root
*root
,
1214 struct btrfs_path
*path
,
1215 struct btrfs_extent_inline_ref
*iref
)
1217 struct btrfs_key key
;
1218 struct extent_buffer
*leaf
;
1219 struct btrfs_extent_data_ref
*ref1
;
1220 struct btrfs_shared_data_ref
*ref2
;
1223 leaf
= path
->nodes
[0];
1224 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1226 if (btrfs_extent_inline_ref_type(leaf
, iref
) ==
1227 BTRFS_EXTENT_DATA_REF_KEY
) {
1228 ref1
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1229 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1231 ref2
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1232 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1234 } else if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1235 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1236 struct btrfs_extent_data_ref
);
1237 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1238 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1239 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1240 struct btrfs_shared_data_ref
);
1241 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1242 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1243 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1244 struct btrfs_extent_ref_v0
*ref0
;
1245 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1246 struct btrfs_extent_ref_v0
);
1247 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1255 static noinline
int lookup_tree_block_ref(struct btrfs_trans_handle
*trans
,
1256 struct btrfs_root
*root
,
1257 struct btrfs_path
*path
,
1258 u64 bytenr
, u64 parent
,
1261 struct btrfs_key key
;
1264 key
.objectid
= bytenr
;
1266 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1267 key
.offset
= parent
;
1269 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1270 key
.offset
= root_objectid
;
1273 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1276 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1277 if (ret
== -ENOENT
&& parent
) {
1278 btrfs_release_path(root
, path
);
1279 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1280 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1288 static noinline
int insert_tree_block_ref(struct btrfs_trans_handle
*trans
,
1289 struct btrfs_root
*root
,
1290 struct btrfs_path
*path
,
1291 u64 bytenr
, u64 parent
,
1294 struct btrfs_key key
;
1297 key
.objectid
= bytenr
;
1299 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1300 key
.offset
= parent
;
1302 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1303 key
.offset
= root_objectid
;
1306 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
1307 btrfs_release_path(root
, path
);
1311 static inline int extent_ref_type(u64 parent
, u64 owner
)
1314 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1316 type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1318 type
= BTRFS_TREE_BLOCK_REF_KEY
;
1321 type
= BTRFS_SHARED_DATA_REF_KEY
;
1323 type
= BTRFS_EXTENT_DATA_REF_KEY
;
1328 static int find_next_key(struct btrfs_path
*path
, int level
,
1329 struct btrfs_key
*key
)
1332 for (; level
< BTRFS_MAX_LEVEL
; level
++) {
1333 if (!path
->nodes
[level
])
1335 if (path
->slots
[level
] + 1 >=
1336 btrfs_header_nritems(path
->nodes
[level
]))
1339 btrfs_item_key_to_cpu(path
->nodes
[level
], key
,
1340 path
->slots
[level
] + 1);
1342 btrfs_node_key_to_cpu(path
->nodes
[level
], key
,
1343 path
->slots
[level
] + 1);
1350 * look for inline back ref. if back ref is found, *ref_ret is set
1351 * to the address of inline back ref, and 0 is returned.
1353 * if back ref isn't found, *ref_ret is set to the address where it
1354 * should be inserted, and -ENOENT is returned.
1356 * if insert is true and there are too many inline back refs, the path
1357 * points to the extent item, and -EAGAIN is returned.
1359 * NOTE: inline back refs are ordered in the same way that back ref
1360 * items in the tree are ordered.
1362 static noinline_for_stack
1363 int lookup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1364 struct btrfs_root
*root
,
1365 struct btrfs_path
*path
,
1366 struct btrfs_extent_inline_ref
**ref_ret
,
1367 u64 bytenr
, u64 num_bytes
,
1368 u64 parent
, u64 root_objectid
,
1369 u64 owner
, u64 offset
, int insert
)
1371 struct btrfs_key key
;
1372 struct extent_buffer
*leaf
;
1373 struct btrfs_extent_item
*ei
;
1374 struct btrfs_extent_inline_ref
*iref
;
1385 key
.objectid
= bytenr
;
1386 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1387 key
.offset
= num_bytes
;
1389 want
= extent_ref_type(parent
, owner
);
1391 extra_size
= btrfs_extent_inline_ref_size(want
);
1392 path
->keep_locks
= 1;
1395 ret
= btrfs_search_slot(trans
, root
, &key
, path
, extra_size
, 1);
1402 leaf
= path
->nodes
[0];
1403 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1404 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1405 if (item_size
< sizeof(*ei
)) {
1410 ret
= convert_extent_item_v0(trans
, root
, path
, owner
,
1416 leaf
= path
->nodes
[0];
1417 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1420 BUG_ON(item_size
< sizeof(*ei
));
1422 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1423 flags
= btrfs_extent_flags(leaf
, ei
);
1425 ptr
= (unsigned long)(ei
+ 1);
1426 end
= (unsigned long)ei
+ item_size
;
1428 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
) {
1429 ptr
+= sizeof(struct btrfs_tree_block_info
);
1432 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_DATA
));
1441 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1442 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1446 ptr
+= btrfs_extent_inline_ref_size(type
);
1450 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1451 struct btrfs_extent_data_ref
*dref
;
1452 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1453 if (match_extent_data_ref(leaf
, dref
, root_objectid
,
1458 if (hash_extent_data_ref_item(leaf
, dref
) <
1459 hash_extent_data_ref(root_objectid
, owner
, offset
))
1463 ref_offset
= btrfs_extent_inline_ref_offset(leaf
, iref
);
1465 if (parent
== ref_offset
) {
1469 if (ref_offset
< parent
)
1472 if (root_objectid
== ref_offset
) {
1476 if (ref_offset
< root_objectid
)
1480 ptr
+= btrfs_extent_inline_ref_size(type
);
1482 if (err
== -ENOENT
&& insert
) {
1483 if (item_size
+ extra_size
>=
1484 BTRFS_MAX_EXTENT_ITEM_SIZE(root
)) {
1489 * To add new inline back ref, we have to make sure
1490 * there is no corresponding back ref item.
1491 * For simplicity, we just do not add new inline back
1492 * ref if there is any kind of item for this block
1494 if (find_next_key(path
, 0, &key
) == 0 &&
1495 key
.objectid
== bytenr
&&
1496 key
.type
< BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1501 *ref_ret
= (struct btrfs_extent_inline_ref
*)ptr
;
1504 path
->keep_locks
= 0;
1505 btrfs_unlock_up_safe(path
, 1);
1511 * helper to add new inline back ref
1513 static noinline_for_stack
1514 int setup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1515 struct btrfs_root
*root
,
1516 struct btrfs_path
*path
,
1517 struct btrfs_extent_inline_ref
*iref
,
1518 u64 parent
, u64 root_objectid
,
1519 u64 owner
, u64 offset
, int refs_to_add
,
1520 struct btrfs_delayed_extent_op
*extent_op
)
1522 struct extent_buffer
*leaf
;
1523 struct btrfs_extent_item
*ei
;
1526 unsigned long item_offset
;
1532 leaf
= path
->nodes
[0];
1533 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1534 item_offset
= (unsigned long)iref
- (unsigned long)ei
;
1536 type
= extent_ref_type(parent
, owner
);
1537 size
= btrfs_extent_inline_ref_size(type
);
1539 ret
= btrfs_extend_item(trans
, root
, path
, size
);
1542 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1543 refs
= btrfs_extent_refs(leaf
, ei
);
1544 refs
+= refs_to_add
;
1545 btrfs_set_extent_refs(leaf
, ei
, refs
);
1547 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1549 ptr
= (unsigned long)ei
+ item_offset
;
1550 end
= (unsigned long)ei
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
1551 if (ptr
< end
- size
)
1552 memmove_extent_buffer(leaf
, ptr
+ size
, ptr
,
1555 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1556 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
1557 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1558 struct btrfs_extent_data_ref
*dref
;
1559 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1560 btrfs_set_extent_data_ref_root(leaf
, dref
, root_objectid
);
1561 btrfs_set_extent_data_ref_objectid(leaf
, dref
, owner
);
1562 btrfs_set_extent_data_ref_offset(leaf
, dref
, offset
);
1563 btrfs_set_extent_data_ref_count(leaf
, dref
, refs_to_add
);
1564 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1565 struct btrfs_shared_data_ref
*sref
;
1566 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1567 btrfs_set_shared_data_ref_count(leaf
, sref
, refs_to_add
);
1568 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1569 } else if (type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
1570 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1572 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
1574 btrfs_mark_buffer_dirty(leaf
);
1578 static int lookup_extent_backref(struct btrfs_trans_handle
*trans
,
1579 struct btrfs_root
*root
,
1580 struct btrfs_path
*path
,
1581 struct btrfs_extent_inline_ref
**ref_ret
,
1582 u64 bytenr
, u64 num_bytes
, u64 parent
,
1583 u64 root_objectid
, u64 owner
, u64 offset
)
1587 ret
= lookup_inline_extent_backref(trans
, root
, path
, ref_ret
,
1588 bytenr
, num_bytes
, parent
,
1589 root_objectid
, owner
, offset
, 0);
1593 btrfs_release_path(root
, path
);
1596 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1597 ret
= lookup_tree_block_ref(trans
, root
, path
, bytenr
, parent
,
1600 ret
= lookup_extent_data_ref(trans
, root
, path
, bytenr
, parent
,
1601 root_objectid
, owner
, offset
);
1607 * helper to update/remove inline back ref
1609 static noinline_for_stack
1610 int update_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1611 struct btrfs_root
*root
,
1612 struct btrfs_path
*path
,
1613 struct btrfs_extent_inline_ref
*iref
,
1615 struct btrfs_delayed_extent_op
*extent_op
)
1617 struct extent_buffer
*leaf
;
1618 struct btrfs_extent_item
*ei
;
1619 struct btrfs_extent_data_ref
*dref
= NULL
;
1620 struct btrfs_shared_data_ref
*sref
= NULL
;
1629 leaf
= path
->nodes
[0];
1630 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1631 refs
= btrfs_extent_refs(leaf
, ei
);
1632 WARN_ON(refs_to_mod
< 0 && refs
+ refs_to_mod
<= 0);
1633 refs
+= refs_to_mod
;
1634 btrfs_set_extent_refs(leaf
, ei
, refs
);
1636 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1638 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1640 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1641 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1642 refs
= btrfs_extent_data_ref_count(leaf
, dref
);
1643 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1644 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1645 refs
= btrfs_shared_data_ref_count(leaf
, sref
);
1648 BUG_ON(refs_to_mod
!= -1);
1651 BUG_ON(refs_to_mod
< 0 && refs
< -refs_to_mod
);
1652 refs
+= refs_to_mod
;
1655 if (type
== BTRFS_EXTENT_DATA_REF_KEY
)
1656 btrfs_set_extent_data_ref_count(leaf
, dref
, refs
);
1658 btrfs_set_shared_data_ref_count(leaf
, sref
, refs
);
1660 size
= btrfs_extent_inline_ref_size(type
);
1661 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1662 ptr
= (unsigned long)iref
;
1663 end
= (unsigned long)ei
+ item_size
;
1664 if (ptr
+ size
< end
)
1665 memmove_extent_buffer(leaf
, ptr
, ptr
+ size
,
1668 ret
= btrfs_truncate_item(trans
, root
, path
, item_size
, 1);
1671 btrfs_mark_buffer_dirty(leaf
);
1675 static noinline_for_stack
1676 int insert_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1677 struct btrfs_root
*root
,
1678 struct btrfs_path
*path
,
1679 u64 bytenr
, u64 num_bytes
, u64 parent
,
1680 u64 root_objectid
, u64 owner
,
1681 u64 offset
, int refs_to_add
,
1682 struct btrfs_delayed_extent_op
*extent_op
)
1684 struct btrfs_extent_inline_ref
*iref
;
1687 ret
= lookup_inline_extent_backref(trans
, root
, path
, &iref
,
1688 bytenr
, num_bytes
, parent
,
1689 root_objectid
, owner
, offset
, 1);
1691 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
);
1692 ret
= update_inline_extent_backref(trans
, root
, path
, iref
,
1693 refs_to_add
, extent_op
);
1694 } else if (ret
== -ENOENT
) {
1695 ret
= setup_inline_extent_backref(trans
, root
, path
, iref
,
1696 parent
, root_objectid
,
1697 owner
, offset
, refs_to_add
,
1703 static int insert_extent_backref(struct btrfs_trans_handle
*trans
,
1704 struct btrfs_root
*root
,
1705 struct btrfs_path
*path
,
1706 u64 bytenr
, u64 parent
, u64 root_objectid
,
1707 u64 owner
, u64 offset
, int refs_to_add
)
1710 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1711 BUG_ON(refs_to_add
!= 1);
1712 ret
= insert_tree_block_ref(trans
, root
, path
, bytenr
,
1713 parent
, root_objectid
);
1715 ret
= insert_extent_data_ref(trans
, root
, path
, bytenr
,
1716 parent
, root_objectid
,
1717 owner
, offset
, refs_to_add
);
1722 static int remove_extent_backref(struct btrfs_trans_handle
*trans
,
1723 struct btrfs_root
*root
,
1724 struct btrfs_path
*path
,
1725 struct btrfs_extent_inline_ref
*iref
,
1726 int refs_to_drop
, int is_data
)
1730 BUG_ON(!is_data
&& refs_to_drop
!= 1);
1732 ret
= update_inline_extent_backref(trans
, root
, path
, iref
,
1733 -refs_to_drop
, NULL
);
1734 } else if (is_data
) {
1735 ret
= remove_extent_data_ref(trans
, root
, path
, refs_to_drop
);
1737 ret
= btrfs_del_item(trans
, root
, path
);
1742 static void btrfs_issue_discard(struct block_device
*bdev
,
1745 blkdev_issue_discard(bdev
, start
>> 9, len
>> 9, GFP_KERNEL
,
1746 BLKDEV_IFL_WAIT
| BLKDEV_IFL_BARRIER
);
1749 static int btrfs_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
1753 u64 map_length
= num_bytes
;
1754 struct btrfs_multi_bio
*multi
= NULL
;
1756 if (!btrfs_test_opt(root
, DISCARD
))
1759 /* Tell the block device(s) that the sectors can be discarded */
1760 ret
= btrfs_map_block(&root
->fs_info
->mapping_tree
, READ
,
1761 bytenr
, &map_length
, &multi
, 0);
1763 struct btrfs_bio_stripe
*stripe
= multi
->stripes
;
1766 if (map_length
> num_bytes
)
1767 map_length
= num_bytes
;
1769 for (i
= 0; i
< multi
->num_stripes
; i
++, stripe
++) {
1770 btrfs_issue_discard(stripe
->dev
->bdev
,
1780 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1781 struct btrfs_root
*root
,
1782 u64 bytenr
, u64 num_bytes
, u64 parent
,
1783 u64 root_objectid
, u64 owner
, u64 offset
)
1786 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
&&
1787 root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
1789 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1790 ret
= btrfs_add_delayed_tree_ref(trans
, bytenr
, num_bytes
,
1791 parent
, root_objectid
, (int)owner
,
1792 BTRFS_ADD_DELAYED_REF
, NULL
);
1794 ret
= btrfs_add_delayed_data_ref(trans
, bytenr
, num_bytes
,
1795 parent
, root_objectid
, owner
, offset
,
1796 BTRFS_ADD_DELAYED_REF
, NULL
);
1801 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1802 struct btrfs_root
*root
,
1803 u64 bytenr
, u64 num_bytes
,
1804 u64 parent
, u64 root_objectid
,
1805 u64 owner
, u64 offset
, int refs_to_add
,
1806 struct btrfs_delayed_extent_op
*extent_op
)
1808 struct btrfs_path
*path
;
1809 struct extent_buffer
*leaf
;
1810 struct btrfs_extent_item
*item
;
1815 path
= btrfs_alloc_path();
1820 path
->leave_spinning
= 1;
1821 /* this will setup the path even if it fails to insert the back ref */
1822 ret
= insert_inline_extent_backref(trans
, root
->fs_info
->extent_root
,
1823 path
, bytenr
, num_bytes
, parent
,
1824 root_objectid
, owner
, offset
,
1825 refs_to_add
, extent_op
);
1829 if (ret
!= -EAGAIN
) {
1834 leaf
= path
->nodes
[0];
1835 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1836 refs
= btrfs_extent_refs(leaf
, item
);
1837 btrfs_set_extent_refs(leaf
, item
, refs
+ refs_to_add
);
1839 __run_delayed_extent_op(extent_op
, leaf
, item
);
1841 btrfs_mark_buffer_dirty(leaf
);
1842 btrfs_release_path(root
->fs_info
->extent_root
, path
);
1845 path
->leave_spinning
= 1;
1847 /* now insert the actual backref */
1848 ret
= insert_extent_backref(trans
, root
->fs_info
->extent_root
,
1849 path
, bytenr
, parent
, root_objectid
,
1850 owner
, offset
, refs_to_add
);
1853 btrfs_free_path(path
);
1857 static int run_delayed_data_ref(struct btrfs_trans_handle
*trans
,
1858 struct btrfs_root
*root
,
1859 struct btrfs_delayed_ref_node
*node
,
1860 struct btrfs_delayed_extent_op
*extent_op
,
1861 int insert_reserved
)
1864 struct btrfs_delayed_data_ref
*ref
;
1865 struct btrfs_key ins
;
1870 ins
.objectid
= node
->bytenr
;
1871 ins
.offset
= node
->num_bytes
;
1872 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1874 ref
= btrfs_delayed_node_to_data_ref(node
);
1875 if (node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
1876 parent
= ref
->parent
;
1878 ref_root
= ref
->root
;
1880 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
1882 BUG_ON(extent_op
->update_key
);
1883 flags
|= extent_op
->flags_to_set
;
1885 ret
= alloc_reserved_file_extent(trans
, root
,
1886 parent
, ref_root
, flags
,
1887 ref
->objectid
, ref
->offset
,
1888 &ins
, node
->ref_mod
);
1889 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
1890 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
1891 node
->num_bytes
, parent
,
1892 ref_root
, ref
->objectid
,
1893 ref
->offset
, node
->ref_mod
,
1895 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
1896 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
1897 node
->num_bytes
, parent
,
1898 ref_root
, ref
->objectid
,
1899 ref
->offset
, node
->ref_mod
,
1907 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
1908 struct extent_buffer
*leaf
,
1909 struct btrfs_extent_item
*ei
)
1911 u64 flags
= btrfs_extent_flags(leaf
, ei
);
1912 if (extent_op
->update_flags
) {
1913 flags
|= extent_op
->flags_to_set
;
1914 btrfs_set_extent_flags(leaf
, ei
, flags
);
1917 if (extent_op
->update_key
) {
1918 struct btrfs_tree_block_info
*bi
;
1919 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
));
1920 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
1921 btrfs_set_tree_block_key(leaf
, bi
, &extent_op
->key
);
1925 static int run_delayed_extent_op(struct btrfs_trans_handle
*trans
,
1926 struct btrfs_root
*root
,
1927 struct btrfs_delayed_ref_node
*node
,
1928 struct btrfs_delayed_extent_op
*extent_op
)
1930 struct btrfs_key key
;
1931 struct btrfs_path
*path
;
1932 struct btrfs_extent_item
*ei
;
1933 struct extent_buffer
*leaf
;
1938 path
= btrfs_alloc_path();
1942 key
.objectid
= node
->bytenr
;
1943 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1944 key
.offset
= node
->num_bytes
;
1947 path
->leave_spinning
= 1;
1948 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
,
1959 leaf
= path
->nodes
[0];
1960 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1961 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1962 if (item_size
< sizeof(*ei
)) {
1963 ret
= convert_extent_item_v0(trans
, root
->fs_info
->extent_root
,
1969 leaf
= path
->nodes
[0];
1970 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1973 BUG_ON(item_size
< sizeof(*ei
));
1974 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1975 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1977 btrfs_mark_buffer_dirty(leaf
);
1979 btrfs_free_path(path
);
1983 static int run_delayed_tree_ref(struct btrfs_trans_handle
*trans
,
1984 struct btrfs_root
*root
,
1985 struct btrfs_delayed_ref_node
*node
,
1986 struct btrfs_delayed_extent_op
*extent_op
,
1987 int insert_reserved
)
1990 struct btrfs_delayed_tree_ref
*ref
;
1991 struct btrfs_key ins
;
1995 ins
.objectid
= node
->bytenr
;
1996 ins
.offset
= node
->num_bytes
;
1997 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1999 ref
= btrfs_delayed_node_to_tree_ref(node
);
2000 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2001 parent
= ref
->parent
;
2003 ref_root
= ref
->root
;
2005 BUG_ON(node
->ref_mod
!= 1);
2006 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2007 BUG_ON(!extent_op
|| !extent_op
->update_flags
||
2008 !extent_op
->update_key
);
2009 ret
= alloc_reserved_tree_block(trans
, root
,
2011 extent_op
->flags_to_set
,
2014 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2015 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
2016 node
->num_bytes
, parent
, ref_root
,
2017 ref
->level
, 0, 1, extent_op
);
2018 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2019 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
2020 node
->num_bytes
, parent
, ref_root
,
2021 ref
->level
, 0, 1, extent_op
);
2028 /* helper function to actually process a single delayed ref entry */
2029 static int run_one_delayed_ref(struct btrfs_trans_handle
*trans
,
2030 struct btrfs_root
*root
,
2031 struct btrfs_delayed_ref_node
*node
,
2032 struct btrfs_delayed_extent_op
*extent_op
,
2033 int insert_reserved
)
2036 if (btrfs_delayed_ref_is_head(node
)) {
2037 struct btrfs_delayed_ref_head
*head
;
2039 * we've hit the end of the chain and we were supposed
2040 * to insert this extent into the tree. But, it got
2041 * deleted before we ever needed to insert it, so all
2042 * we have to do is clean up the accounting
2045 head
= btrfs_delayed_node_to_head(node
);
2046 if (insert_reserved
) {
2047 btrfs_pin_extent(root
, node
->bytenr
,
2048 node
->num_bytes
, 1);
2049 if (head
->is_data
) {
2050 ret
= btrfs_del_csums(trans
, root
,
2056 mutex_unlock(&head
->mutex
);
2060 if (node
->type
== BTRFS_TREE_BLOCK_REF_KEY
||
2061 node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2062 ret
= run_delayed_tree_ref(trans
, root
, node
, extent_op
,
2064 else if (node
->type
== BTRFS_EXTENT_DATA_REF_KEY
||
2065 node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2066 ret
= run_delayed_data_ref(trans
, root
, node
, extent_op
,
2073 static noinline
struct btrfs_delayed_ref_node
*
2074 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
2076 struct rb_node
*node
;
2077 struct btrfs_delayed_ref_node
*ref
;
2078 int action
= BTRFS_ADD_DELAYED_REF
;
2081 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2082 * this prevents ref count from going down to zero when
2083 * there still are pending delayed ref.
2085 node
= rb_prev(&head
->node
.rb_node
);
2089 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
2091 if (ref
->bytenr
!= head
->node
.bytenr
)
2093 if (ref
->action
== action
)
2095 node
= rb_prev(node
);
2097 if (action
== BTRFS_ADD_DELAYED_REF
) {
2098 action
= BTRFS_DROP_DELAYED_REF
;
2104 static noinline
int run_clustered_refs(struct btrfs_trans_handle
*trans
,
2105 struct btrfs_root
*root
,
2106 struct list_head
*cluster
)
2108 struct btrfs_delayed_ref_root
*delayed_refs
;
2109 struct btrfs_delayed_ref_node
*ref
;
2110 struct btrfs_delayed_ref_head
*locked_ref
= NULL
;
2111 struct btrfs_delayed_extent_op
*extent_op
;
2114 int must_insert_reserved
= 0;
2116 delayed_refs
= &trans
->transaction
->delayed_refs
;
2119 /* pick a new head ref from the cluster list */
2120 if (list_empty(cluster
))
2123 locked_ref
= list_entry(cluster
->next
,
2124 struct btrfs_delayed_ref_head
, cluster
);
2126 /* grab the lock that says we are going to process
2127 * all the refs for this head */
2128 ret
= btrfs_delayed_ref_lock(trans
, locked_ref
);
2131 * we may have dropped the spin lock to get the head
2132 * mutex lock, and that might have given someone else
2133 * time to free the head. If that's true, it has been
2134 * removed from our list and we can move on.
2136 if (ret
== -EAGAIN
) {
2144 * record the must insert reserved flag before we
2145 * drop the spin lock.
2147 must_insert_reserved
= locked_ref
->must_insert_reserved
;
2148 locked_ref
->must_insert_reserved
= 0;
2150 extent_op
= locked_ref
->extent_op
;
2151 locked_ref
->extent_op
= NULL
;
2154 * locked_ref is the head node, so we have to go one
2155 * node back for any delayed ref updates
2157 ref
= select_delayed_ref(locked_ref
);
2159 /* All delayed refs have been processed, Go ahead
2160 * and send the head node to run_one_delayed_ref,
2161 * so that any accounting fixes can happen
2163 ref
= &locked_ref
->node
;
2165 if (extent_op
&& must_insert_reserved
) {
2171 spin_unlock(&delayed_refs
->lock
);
2173 ret
= run_delayed_extent_op(trans
, root
,
2179 spin_lock(&delayed_refs
->lock
);
2183 list_del_init(&locked_ref
->cluster
);
2188 rb_erase(&ref
->rb_node
, &delayed_refs
->root
);
2189 delayed_refs
->num_entries
--;
2191 spin_unlock(&delayed_refs
->lock
);
2193 ret
= run_one_delayed_ref(trans
, root
, ref
, extent_op
,
2194 must_insert_reserved
);
2197 btrfs_put_delayed_ref(ref
);
2202 spin_lock(&delayed_refs
->lock
);
2208 * this starts processing the delayed reference count updates and
2209 * extent insertions we have queued up so far. count can be
2210 * 0, which means to process everything in the tree at the start
2211 * of the run (but not newly added entries), or it can be some target
2212 * number you'd like to process.
2214 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2215 struct btrfs_root
*root
, unsigned long count
)
2217 struct rb_node
*node
;
2218 struct btrfs_delayed_ref_root
*delayed_refs
;
2219 struct btrfs_delayed_ref_node
*ref
;
2220 struct list_head cluster
;
2222 int run_all
= count
== (unsigned long)-1;
2225 if (root
== root
->fs_info
->extent_root
)
2226 root
= root
->fs_info
->tree_root
;
2228 delayed_refs
= &trans
->transaction
->delayed_refs
;
2229 INIT_LIST_HEAD(&cluster
);
2231 spin_lock(&delayed_refs
->lock
);
2233 count
= delayed_refs
->num_entries
* 2;
2237 if (!(run_all
|| run_most
) &&
2238 delayed_refs
->num_heads_ready
< 64)
2242 * go find something we can process in the rbtree. We start at
2243 * the beginning of the tree, and then build a cluster
2244 * of refs to process starting at the first one we are able to
2247 ret
= btrfs_find_ref_cluster(trans
, &cluster
,
2248 delayed_refs
->run_delayed_start
);
2252 ret
= run_clustered_refs(trans
, root
, &cluster
);
2255 count
-= min_t(unsigned long, ret
, count
);
2262 node
= rb_first(&delayed_refs
->root
);
2265 count
= (unsigned long)-1;
2268 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
2270 if (btrfs_delayed_ref_is_head(ref
)) {
2271 struct btrfs_delayed_ref_head
*head
;
2273 head
= btrfs_delayed_node_to_head(ref
);
2274 atomic_inc(&ref
->refs
);
2276 spin_unlock(&delayed_refs
->lock
);
2277 mutex_lock(&head
->mutex
);
2278 mutex_unlock(&head
->mutex
);
2280 btrfs_put_delayed_ref(ref
);
2284 node
= rb_next(node
);
2286 spin_unlock(&delayed_refs
->lock
);
2287 schedule_timeout(1);
2291 spin_unlock(&delayed_refs
->lock
);
2295 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
2296 struct btrfs_root
*root
,
2297 u64 bytenr
, u64 num_bytes
, u64 flags
,
2300 struct btrfs_delayed_extent_op
*extent_op
;
2303 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
2307 extent_op
->flags_to_set
= flags
;
2308 extent_op
->update_flags
= 1;
2309 extent_op
->update_key
= 0;
2310 extent_op
->is_data
= is_data
? 1 : 0;
2312 ret
= btrfs_add_delayed_extent_op(trans
, bytenr
, num_bytes
, extent_op
);
2318 static noinline
int check_delayed_ref(struct btrfs_trans_handle
*trans
,
2319 struct btrfs_root
*root
,
2320 struct btrfs_path
*path
,
2321 u64 objectid
, u64 offset
, u64 bytenr
)
2323 struct btrfs_delayed_ref_head
*head
;
2324 struct btrfs_delayed_ref_node
*ref
;
2325 struct btrfs_delayed_data_ref
*data_ref
;
2326 struct btrfs_delayed_ref_root
*delayed_refs
;
2327 struct rb_node
*node
;
2331 delayed_refs
= &trans
->transaction
->delayed_refs
;
2332 spin_lock(&delayed_refs
->lock
);
2333 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
2337 if (!mutex_trylock(&head
->mutex
)) {
2338 atomic_inc(&head
->node
.refs
);
2339 spin_unlock(&delayed_refs
->lock
);
2341 btrfs_release_path(root
->fs_info
->extent_root
, path
);
2343 mutex_lock(&head
->mutex
);
2344 mutex_unlock(&head
->mutex
);
2345 btrfs_put_delayed_ref(&head
->node
);
2349 node
= rb_prev(&head
->node
.rb_node
);
2353 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2355 if (ref
->bytenr
!= bytenr
)
2359 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
)
2362 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
2364 node
= rb_prev(node
);
2366 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2367 if (ref
->bytenr
== bytenr
)
2371 if (data_ref
->root
!= root
->root_key
.objectid
||
2372 data_ref
->objectid
!= objectid
|| data_ref
->offset
!= offset
)
2377 mutex_unlock(&head
->mutex
);
2379 spin_unlock(&delayed_refs
->lock
);
2383 static noinline
int check_committed_ref(struct btrfs_trans_handle
*trans
,
2384 struct btrfs_root
*root
,
2385 struct btrfs_path
*path
,
2386 u64 objectid
, u64 offset
, u64 bytenr
)
2388 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2389 struct extent_buffer
*leaf
;
2390 struct btrfs_extent_data_ref
*ref
;
2391 struct btrfs_extent_inline_ref
*iref
;
2392 struct btrfs_extent_item
*ei
;
2393 struct btrfs_key key
;
2397 key
.objectid
= bytenr
;
2398 key
.offset
= (u64
)-1;
2399 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2401 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
2407 if (path
->slots
[0] == 0)
2411 leaf
= path
->nodes
[0];
2412 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2414 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
2418 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2419 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2420 if (item_size
< sizeof(*ei
)) {
2421 WARN_ON(item_size
!= sizeof(struct btrfs_extent_item_v0
));
2425 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2427 if (item_size
!= sizeof(*ei
) +
2428 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
2431 if (btrfs_extent_generation(leaf
, ei
) <=
2432 btrfs_root_last_snapshot(&root
->root_item
))
2435 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
2436 if (btrfs_extent_inline_ref_type(leaf
, iref
) !=
2437 BTRFS_EXTENT_DATA_REF_KEY
)
2440 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
2441 if (btrfs_extent_refs(leaf
, ei
) !=
2442 btrfs_extent_data_ref_count(leaf
, ref
) ||
2443 btrfs_extent_data_ref_root(leaf
, ref
) !=
2444 root
->root_key
.objectid
||
2445 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
2446 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
2454 int btrfs_cross_ref_exist(struct btrfs_trans_handle
*trans
,
2455 struct btrfs_root
*root
,
2456 u64 objectid
, u64 offset
, u64 bytenr
)
2458 struct btrfs_path
*path
;
2462 path
= btrfs_alloc_path();
2467 ret
= check_committed_ref(trans
, root
, path
, objectid
,
2469 if (ret
&& ret
!= -ENOENT
)
2472 ret2
= check_delayed_ref(trans
, root
, path
, objectid
,
2474 } while (ret2
== -EAGAIN
);
2476 if (ret2
&& ret2
!= -ENOENT
) {
2481 if (ret
!= -ENOENT
|| ret2
!= -ENOENT
)
2484 btrfs_free_path(path
);
2485 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
2491 int btrfs_cache_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2492 struct extent_buffer
*buf
, u32 nr_extents
)
2494 struct btrfs_key key
;
2495 struct btrfs_file_extent_item
*fi
;
2503 if (!root
->ref_cows
)
2506 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
2508 root_gen
= root
->root_key
.offset
;
2511 root_gen
= trans
->transid
- 1;
2514 level
= btrfs_header_level(buf
);
2515 nritems
= btrfs_header_nritems(buf
);
2518 struct btrfs_leaf_ref
*ref
;
2519 struct btrfs_extent_info
*info
;
2521 ref
= btrfs_alloc_leaf_ref(root
, nr_extents
);
2527 ref
->root_gen
= root_gen
;
2528 ref
->bytenr
= buf
->start
;
2529 ref
->owner
= btrfs_header_owner(buf
);
2530 ref
->generation
= btrfs_header_generation(buf
);
2531 ref
->nritems
= nr_extents
;
2532 info
= ref
->extents
;
2534 for (i
= 0; nr_extents
> 0 && i
< nritems
; i
++) {
2536 btrfs_item_key_to_cpu(buf
, &key
, i
);
2537 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
2539 fi
= btrfs_item_ptr(buf
, i
,
2540 struct btrfs_file_extent_item
);
2541 if (btrfs_file_extent_type(buf
, fi
) ==
2542 BTRFS_FILE_EXTENT_INLINE
)
2544 disk_bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
2545 if (disk_bytenr
== 0)
2548 info
->bytenr
= disk_bytenr
;
2550 btrfs_file_extent_disk_num_bytes(buf
, fi
);
2551 info
->objectid
= key
.objectid
;
2552 info
->offset
= key
.offset
;
2556 ret
= btrfs_add_leaf_ref(root
, ref
, shared
);
2557 if (ret
== -EEXIST
&& shared
) {
2558 struct btrfs_leaf_ref
*old
;
2559 old
= btrfs_lookup_leaf_ref(root
, ref
->bytenr
);
2561 btrfs_remove_leaf_ref(root
, old
);
2562 btrfs_free_leaf_ref(root
, old
);
2563 ret
= btrfs_add_leaf_ref(root
, ref
, shared
);
2566 btrfs_free_leaf_ref(root
, ref
);
2572 /* when a block goes through cow, we update the reference counts of
2573 * everything that block points to. The internal pointers of the block
2574 * can be in just about any order, and it is likely to have clusters of
2575 * things that are close together and clusters of things that are not.
2577 * To help reduce the seeks that come with updating all of these reference
2578 * counts, sort them by byte number before actual updates are done.
2580 * struct refsort is used to match byte number to slot in the btree block.
2581 * we sort based on the byte number and then use the slot to actually
2584 * struct refsort is smaller than strcut btrfs_item and smaller than
2585 * struct btrfs_key_ptr. Since we're currently limited to the page size
2586 * for a btree block, there's no way for a kmalloc of refsorts for a
2587 * single node to be bigger than a page.
2595 * for passing into sort()
2597 static int refsort_cmp(const void *a_void
, const void *b_void
)
2599 const struct refsort
*a
= a_void
;
2600 const struct refsort
*b
= b_void
;
2602 if (a
->bytenr
< b
->bytenr
)
2604 if (a
->bytenr
> b
->bytenr
)
2610 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
2611 struct btrfs_root
*root
,
2612 struct extent_buffer
*buf
,
2613 int full_backref
, int inc
)
2620 struct btrfs_key key
;
2621 struct btrfs_file_extent_item
*fi
;
2625 int (*process_func
)(struct btrfs_trans_handle
*, struct btrfs_root
*,
2626 u64
, u64
, u64
, u64
, u64
, u64
);
2628 ref_root
= btrfs_header_owner(buf
);
2629 nritems
= btrfs_header_nritems(buf
);
2630 level
= btrfs_header_level(buf
);
2632 if (!root
->ref_cows
&& level
== 0)
2636 process_func
= btrfs_inc_extent_ref
;
2638 process_func
= btrfs_free_extent
;
2641 parent
= buf
->start
;
2645 for (i
= 0; i
< nritems
; i
++) {
2647 btrfs_item_key_to_cpu(buf
, &key
, i
);
2648 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
2650 fi
= btrfs_item_ptr(buf
, i
,
2651 struct btrfs_file_extent_item
);
2652 if (btrfs_file_extent_type(buf
, fi
) ==
2653 BTRFS_FILE_EXTENT_INLINE
)
2655 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
2659 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
2660 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
2661 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
2662 parent
, ref_root
, key
.objectid
,
2667 bytenr
= btrfs_node_blockptr(buf
, i
);
2668 num_bytes
= btrfs_level_size(root
, level
- 1);
2669 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
2670 parent
, ref_root
, level
- 1, 0);
2681 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2682 struct extent_buffer
*buf
, int full_backref
)
2684 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1);
2687 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2688 struct extent_buffer
*buf
, int full_backref
)
2690 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0);
2693 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
2694 struct btrfs_root
*root
,
2695 struct btrfs_path
*path
,
2696 struct btrfs_block_group_cache
*cache
)
2699 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2701 struct extent_buffer
*leaf
;
2703 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
2708 leaf
= path
->nodes
[0];
2709 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
2710 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
2711 btrfs_mark_buffer_dirty(leaf
);
2712 btrfs_release_path(extent_root
, path
);
2720 static struct btrfs_block_group_cache
*
2721 next_block_group(struct btrfs_root
*root
,
2722 struct btrfs_block_group_cache
*cache
)
2724 struct rb_node
*node
;
2725 spin_lock(&root
->fs_info
->block_group_cache_lock
);
2726 node
= rb_next(&cache
->cache_node
);
2727 btrfs_put_block_group(cache
);
2729 cache
= rb_entry(node
, struct btrfs_block_group_cache
,
2731 btrfs_get_block_group(cache
);
2734 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
2738 static int cache_save_setup(struct btrfs_block_group_cache
*block_group
,
2739 struct btrfs_trans_handle
*trans
,
2740 struct btrfs_path
*path
)
2742 struct btrfs_root
*root
= block_group
->fs_info
->tree_root
;
2743 struct inode
*inode
= NULL
;
2750 * If this block group is smaller than 100 megs don't bother caching the
2753 if (block_group
->key
.offset
< (100 * 1024 * 1024)) {
2754 spin_lock(&block_group
->lock
);
2755 block_group
->disk_cache_state
= BTRFS_DC_WRITTEN
;
2756 spin_unlock(&block_group
->lock
);
2761 inode
= lookup_free_space_inode(root
, block_group
, path
);
2762 if (IS_ERR(inode
) && PTR_ERR(inode
) != -ENOENT
) {
2763 ret
= PTR_ERR(inode
);
2764 btrfs_release_path(root
, path
);
2768 if (IS_ERR(inode
)) {
2772 if (block_group
->ro
)
2775 ret
= create_free_space_inode(root
, trans
, block_group
, path
);
2782 * We want to set the generation to 0, that way if anything goes wrong
2783 * from here on out we know not to trust this cache when we load up next
2786 BTRFS_I(inode
)->generation
= 0;
2787 ret
= btrfs_update_inode(trans
, root
, inode
);
2790 if (i_size_read(inode
) > 0) {
2791 ret
= btrfs_truncate_free_space_cache(root
, trans
, path
,
2797 spin_lock(&block_group
->lock
);
2798 if (block_group
->cached
!= BTRFS_CACHE_FINISHED
) {
2799 spin_unlock(&block_group
->lock
);
2802 spin_unlock(&block_group
->lock
);
2804 num_pages
= (int)div64_u64(block_group
->key
.offset
, 1024 * 1024 * 1024);
2809 * Just to make absolutely sure we have enough space, we're going to
2810 * preallocate 12 pages worth of space for each block group. In
2811 * practice we ought to use at most 8, but we need extra space so we can
2812 * add our header and have a terminator between the extents and the
2816 num_pages
*= PAGE_CACHE_SIZE
;
2818 ret
= btrfs_check_data_free_space(inode
, num_pages
);
2822 ret
= btrfs_prealloc_file_range_trans(inode
, trans
, 0, 0, num_pages
,
2823 num_pages
, num_pages
,
2825 btrfs_free_reserved_data_space(inode
, num_pages
);
2829 btrfs_release_path(root
, path
);
2831 spin_lock(&block_group
->lock
);
2833 block_group
->disk_cache_state
= BTRFS_DC_ERROR
;
2835 block_group
->disk_cache_state
= BTRFS_DC_SETUP
;
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
)
3041 u64 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
3043 if (num_devices
== 1)
3044 flags
&= ~(BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID0
);
3045 if (num_devices
< 4)
3046 flags
&= ~BTRFS_BLOCK_GROUP_RAID10
;
3048 if ((flags
& BTRFS_BLOCK_GROUP_DUP
) &&
3049 (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
3050 BTRFS_BLOCK_GROUP_RAID10
))) {
3051 flags
&= ~BTRFS_BLOCK_GROUP_DUP
;
3054 if ((flags
& BTRFS_BLOCK_GROUP_RAID1
) &&
3055 (flags
& BTRFS_BLOCK_GROUP_RAID10
)) {
3056 flags
&= ~BTRFS_BLOCK_GROUP_RAID1
;
3059 if ((flags
& BTRFS_BLOCK_GROUP_RAID0
) &&
3060 ((flags
& BTRFS_BLOCK_GROUP_RAID1
) |
3061 (flags
& BTRFS_BLOCK_GROUP_RAID10
) |
3062 (flags
& BTRFS_BLOCK_GROUP_DUP
)))
3063 flags
&= ~BTRFS_BLOCK_GROUP_RAID0
;
3067 static u64
get_alloc_profile(struct btrfs_root
*root
, u64 flags
)
3069 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3070 flags
|= root
->fs_info
->avail_data_alloc_bits
&
3071 root
->fs_info
->data_alloc_profile
;
3072 else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3073 flags
|= root
->fs_info
->avail_system_alloc_bits
&
3074 root
->fs_info
->system_alloc_profile
;
3075 else if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3076 flags
|= root
->fs_info
->avail_metadata_alloc_bits
&
3077 root
->fs_info
->metadata_alloc_profile
;
3078 return btrfs_reduce_alloc_profile(root
, flags
);
3081 static u64
btrfs_get_alloc_profile(struct btrfs_root
*root
, int data
)
3086 flags
= BTRFS_BLOCK_GROUP_DATA
;
3087 else if (root
== root
->fs_info
->chunk_root
)
3088 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
3090 flags
= BTRFS_BLOCK_GROUP_METADATA
;
3092 return get_alloc_profile(root
, flags
);
3095 void btrfs_set_inode_space_info(struct btrfs_root
*root
, struct inode
*inode
)
3097 BTRFS_I(inode
)->space_info
= __find_space_info(root
->fs_info
,
3098 BTRFS_BLOCK_GROUP_DATA
);
3102 * This will check the space that the inode allocates from to make sure we have
3103 * enough space for bytes.
3105 int btrfs_check_data_free_space(struct inode
*inode
, u64 bytes
)
3107 struct btrfs_space_info
*data_sinfo
;
3108 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3110 int ret
= 0, committed
= 0, alloc_chunk
= 1;
3112 /* make sure bytes are sectorsize aligned */
3113 bytes
= (bytes
+ root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
3115 if (root
== root
->fs_info
->tree_root
) {
3120 data_sinfo
= BTRFS_I(inode
)->space_info
;
3125 /* make sure we have enough space to handle the data first */
3126 spin_lock(&data_sinfo
->lock
);
3127 used
= data_sinfo
->bytes_used
+ data_sinfo
->bytes_reserved
+
3128 data_sinfo
->bytes_pinned
+ data_sinfo
->bytes_readonly
+
3129 data_sinfo
->bytes_may_use
;
3131 if (used
+ bytes
> data_sinfo
->total_bytes
) {
3132 struct btrfs_trans_handle
*trans
;
3135 * if we don't have enough free bytes in this space then we need
3136 * to alloc a new chunk.
3138 if (!data_sinfo
->full
&& alloc_chunk
) {
3141 data_sinfo
->force_alloc
= 1;
3142 spin_unlock(&data_sinfo
->lock
);
3144 alloc_target
= btrfs_get_alloc_profile(root
, 1);
3145 trans
= btrfs_join_transaction(root
, 1);
3147 return PTR_ERR(trans
);
3149 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
3150 bytes
+ 2 * 1024 * 1024,
3152 btrfs_end_transaction(trans
, root
);
3157 btrfs_set_inode_space_info(root
, inode
);
3158 data_sinfo
= BTRFS_I(inode
)->space_info
;
3162 spin_unlock(&data_sinfo
->lock
);
3164 /* commit the current transaction and try again */
3165 if (!committed
&& !root
->fs_info
->open_ioctl_trans
) {
3167 trans
= btrfs_join_transaction(root
, 1);
3169 return PTR_ERR(trans
);
3170 ret
= btrfs_commit_transaction(trans
, root
);
3176 #if 0 /* I hope we never need this code again, just in case */
3177 printk(KERN_ERR
"no space left, need %llu, %llu bytes_used, "
3178 "%llu bytes_reserved, " "%llu bytes_pinned, "
3179 "%llu bytes_readonly, %llu may use %llu total\n",
3180 (unsigned long long)bytes
,
3181 (unsigned long long)data_sinfo
->bytes_used
,
3182 (unsigned long long)data_sinfo
->bytes_reserved
,
3183 (unsigned long long)data_sinfo
->bytes_pinned
,
3184 (unsigned long long)data_sinfo
->bytes_readonly
,
3185 (unsigned long long)data_sinfo
->bytes_may_use
,
3186 (unsigned long long)data_sinfo
->total_bytes
);
3190 data_sinfo
->bytes_may_use
+= bytes
;
3191 BTRFS_I(inode
)->reserved_bytes
+= bytes
;
3192 spin_unlock(&data_sinfo
->lock
);
3198 * called when we are clearing an delalloc extent from the
3199 * inode's io_tree or there was an error for whatever reason
3200 * after calling btrfs_check_data_free_space
3202 void btrfs_free_reserved_data_space(struct inode
*inode
, u64 bytes
)
3204 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3205 struct btrfs_space_info
*data_sinfo
;
3207 /* make sure bytes are sectorsize aligned */
3208 bytes
= (bytes
+ root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
3210 data_sinfo
= BTRFS_I(inode
)->space_info
;
3211 spin_lock(&data_sinfo
->lock
);
3212 data_sinfo
->bytes_may_use
-= bytes
;
3213 BTRFS_I(inode
)->reserved_bytes
-= bytes
;
3214 spin_unlock(&data_sinfo
->lock
);
3217 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
3219 struct list_head
*head
= &info
->space_info
;
3220 struct btrfs_space_info
*found
;
3223 list_for_each_entry_rcu(found
, head
, list
) {
3224 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
3225 found
->force_alloc
= 1;
3230 static int should_alloc_chunk(struct btrfs_root
*root
,
3231 struct btrfs_space_info
*sinfo
, u64 alloc_bytes
)
3233 u64 num_bytes
= sinfo
->total_bytes
- sinfo
->bytes_readonly
;
3236 if (sinfo
->bytes_used
+ sinfo
->bytes_reserved
+
3237 alloc_bytes
+ 256 * 1024 * 1024 < num_bytes
)
3240 if (sinfo
->bytes_used
+ sinfo
->bytes_reserved
+
3241 alloc_bytes
< div_factor(num_bytes
, 8))
3244 thresh
= btrfs_super_total_bytes(&root
->fs_info
->super_copy
);
3245 thresh
= max_t(u64
, 256 * 1024 * 1024, div_factor_fine(thresh
, 5));
3247 if (num_bytes
> thresh
&& sinfo
->bytes_used
< div_factor(num_bytes
, 3))
3253 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
3254 struct btrfs_root
*extent_root
, u64 alloc_bytes
,
3255 u64 flags
, int force
)
3257 struct btrfs_space_info
*space_info
;
3258 struct btrfs_fs_info
*fs_info
= extent_root
->fs_info
;
3261 mutex_lock(&fs_info
->chunk_mutex
);
3263 flags
= btrfs_reduce_alloc_profile(extent_root
, flags
);
3265 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
3267 ret
= update_space_info(extent_root
->fs_info
, flags
,
3271 BUG_ON(!space_info
);
3273 spin_lock(&space_info
->lock
);
3274 if (space_info
->force_alloc
)
3276 if (space_info
->full
) {
3277 spin_unlock(&space_info
->lock
);
3281 if (!force
&& !should_alloc_chunk(extent_root
, space_info
,
3283 spin_unlock(&space_info
->lock
);
3286 spin_unlock(&space_info
->lock
);
3289 * If we have mixed data/metadata chunks we want to make sure we keep
3290 * allocating mixed chunks instead of individual chunks.
3292 if (btrfs_mixed_space_info(space_info
))
3293 flags
|= (BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
);
3296 * if we're doing a data chunk, go ahead and make sure that
3297 * we keep a reasonable number of metadata chunks allocated in the
3300 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
3301 fs_info
->data_chunk_allocations
++;
3302 if (!(fs_info
->data_chunk_allocations
%
3303 fs_info
->metadata_ratio
))
3304 force_metadata_allocation(fs_info
);
3307 ret
= btrfs_alloc_chunk(trans
, extent_root
, flags
);
3308 spin_lock(&space_info
->lock
);
3310 space_info
->full
= 1;
3313 space_info
->force_alloc
= 0;
3314 spin_unlock(&space_info
->lock
);
3316 mutex_unlock(&extent_root
->fs_info
->chunk_mutex
);
3321 * shrink metadata reservation for delalloc
3323 static int shrink_delalloc(struct btrfs_trans_handle
*trans
,
3324 struct btrfs_root
*root
, u64 to_reclaim
, int sync
)
3326 struct btrfs_block_rsv
*block_rsv
;
3327 struct btrfs_space_info
*space_info
;
3335 block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
3336 space_info
= block_rsv
->space_info
;
3337 spin_lock(&space_info
->lock
);
3338 reserved
= space_info
->bytes_reserved
;
3339 spin_unlock(&space_info
->lock
);
3344 max_reclaim
= min(reserved
, to_reclaim
);
3347 ret
= btrfs_start_one_delalloc_inode(root
, trans
? 1 : 0, sync
);
3352 __set_current_state(TASK_INTERRUPTIBLE
);
3353 schedule_timeout(pause
);
3355 if (pause
> HZ
/ 10)
3362 spin_lock(&space_info
->lock
);
3363 if (reserved
> space_info
->bytes_reserved
)
3364 reclaimed
+= reserved
- space_info
->bytes_reserved
;
3365 reserved
= space_info
->bytes_reserved
;
3366 spin_unlock(&space_info
->lock
);
3368 if (reserved
== 0 || reclaimed
>= max_reclaim
)
3371 if (trans
&& trans
->transaction
->blocked
)
3374 return reclaimed
>= to_reclaim
;
3378 * Retries tells us how many times we've called reserve_metadata_bytes. The
3379 * idea is if this is the first call (retries == 0) then we will add to our
3380 * reserved count if we can't make the allocation in order to hold our place
3381 * while we go and try and free up space. That way for retries > 1 we don't try
3382 * and add space, we just check to see if the amount of unused space is >= the
3383 * total space, meaning that our reservation is valid.
3385 * However if we don't intend to retry this reservation, pass -1 as retries so
3386 * that it short circuits this logic.
3388 static int reserve_metadata_bytes(struct btrfs_trans_handle
*trans
,
3389 struct btrfs_root
*root
,
3390 struct btrfs_block_rsv
*block_rsv
,
3391 u64 orig_bytes
, int flush
)
3393 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
3395 u64 num_bytes
= orig_bytes
;
3398 bool reserved
= false;
3399 bool committed
= false;
3406 spin_lock(&space_info
->lock
);
3407 unused
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
3408 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
3409 space_info
->bytes_may_use
;
3412 * The idea here is that we've not already over-reserved the block group
3413 * then we can go ahead and save our reservation first and then start
3414 * flushing if we need to. Otherwise if we've already overcommitted
3415 * lets start flushing stuff first and then come back and try to make
3418 if (unused
<= space_info
->total_bytes
) {
3419 unused
-= space_info
->total_bytes
;
3420 if (unused
>= num_bytes
) {
3422 space_info
->bytes_reserved
+= orig_bytes
;
3426 * Ok set num_bytes to orig_bytes since we aren't
3427 * overocmmitted, this way we only try and reclaim what
3430 num_bytes
= orig_bytes
;
3434 * Ok we're over committed, set num_bytes to the overcommitted
3435 * amount plus the amount of bytes that we need for this
3438 num_bytes
= unused
- space_info
->total_bytes
+
3439 (orig_bytes
* (retries
+ 1));
3443 * Couldn't make our reservation, save our place so while we're trying
3444 * to reclaim space we can actually use it instead of somebody else
3445 * stealing it from us.
3447 if (ret
&& !reserved
) {
3448 space_info
->bytes_reserved
+= orig_bytes
;
3452 spin_unlock(&space_info
->lock
);
3461 * We do synchronous shrinking since we don't actually unreserve
3462 * metadata until after the IO is completed.
3464 ret
= shrink_delalloc(trans
, root
, num_bytes
, 1);
3471 * So if we were overcommitted it's possible that somebody else flushed
3472 * out enough space and we simply didn't have enough space to reclaim,
3473 * so go back around and try again.
3480 spin_lock(&space_info
->lock
);
3482 * Not enough space to be reclaimed, don't bother committing the
3485 if (space_info
->bytes_pinned
< orig_bytes
)
3487 spin_unlock(&space_info
->lock
);
3492 if (trans
|| committed
)
3496 trans
= btrfs_join_transaction(root
, 1);
3499 ret
= btrfs_commit_transaction(trans
, root
);
3508 spin_lock(&space_info
->lock
);
3509 space_info
->bytes_reserved
-= orig_bytes
;
3510 spin_unlock(&space_info
->lock
);
3516 static struct btrfs_block_rsv
*get_block_rsv(struct btrfs_trans_handle
*trans
,
3517 struct btrfs_root
*root
)
3519 struct btrfs_block_rsv
*block_rsv
;
3521 block_rsv
= trans
->block_rsv
;
3523 block_rsv
= root
->block_rsv
;
3526 block_rsv
= &root
->fs_info
->empty_block_rsv
;
3531 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
3535 spin_lock(&block_rsv
->lock
);
3536 if (block_rsv
->reserved
>= num_bytes
) {
3537 block_rsv
->reserved
-= num_bytes
;
3538 if (block_rsv
->reserved
< block_rsv
->size
)
3539 block_rsv
->full
= 0;
3542 spin_unlock(&block_rsv
->lock
);
3546 static void block_rsv_add_bytes(struct btrfs_block_rsv
*block_rsv
,
3547 u64 num_bytes
, int update_size
)
3549 spin_lock(&block_rsv
->lock
);
3550 block_rsv
->reserved
+= num_bytes
;
3552 block_rsv
->size
+= num_bytes
;
3553 else if (block_rsv
->reserved
>= block_rsv
->size
)
3554 block_rsv
->full
= 1;
3555 spin_unlock(&block_rsv
->lock
);
3558 void block_rsv_release_bytes(struct btrfs_block_rsv
*block_rsv
,
3559 struct btrfs_block_rsv
*dest
, u64 num_bytes
)
3561 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
3563 spin_lock(&block_rsv
->lock
);
3564 if (num_bytes
== (u64
)-1)
3565 num_bytes
= block_rsv
->size
;
3566 block_rsv
->size
-= num_bytes
;
3567 if (block_rsv
->reserved
>= block_rsv
->size
) {
3568 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
3569 block_rsv
->reserved
= block_rsv
->size
;
3570 block_rsv
->full
= 1;
3574 spin_unlock(&block_rsv
->lock
);
3576 if (num_bytes
> 0) {
3578 block_rsv_add_bytes(dest
, num_bytes
, 0);
3580 spin_lock(&space_info
->lock
);
3581 space_info
->bytes_reserved
-= num_bytes
;
3582 spin_unlock(&space_info
->lock
);
3587 static int block_rsv_migrate_bytes(struct btrfs_block_rsv
*src
,
3588 struct btrfs_block_rsv
*dst
, u64 num_bytes
)
3592 ret
= block_rsv_use_bytes(src
, num_bytes
);
3596 block_rsv_add_bytes(dst
, num_bytes
, 1);
3600 void btrfs_init_block_rsv(struct btrfs_block_rsv
*rsv
)
3602 memset(rsv
, 0, sizeof(*rsv
));
3603 spin_lock_init(&rsv
->lock
);
3604 atomic_set(&rsv
->usage
, 1);
3606 INIT_LIST_HEAD(&rsv
->list
);
3609 struct btrfs_block_rsv
*btrfs_alloc_block_rsv(struct btrfs_root
*root
)
3611 struct btrfs_block_rsv
*block_rsv
;
3612 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3615 block_rsv
= kmalloc(sizeof(*block_rsv
), GFP_NOFS
);
3619 btrfs_init_block_rsv(block_rsv
);
3621 alloc_target
= btrfs_get_alloc_profile(root
, 0);
3622 block_rsv
->space_info
= __find_space_info(fs_info
,
3623 BTRFS_BLOCK_GROUP_METADATA
);
3628 void btrfs_free_block_rsv(struct btrfs_root
*root
,
3629 struct btrfs_block_rsv
*rsv
)
3631 if (rsv
&& atomic_dec_and_test(&rsv
->usage
)) {
3632 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
3639 * make the block_rsv struct be able to capture freed space.
3640 * the captured space will re-add to the the block_rsv struct
3641 * after transaction commit
3643 void btrfs_add_durable_block_rsv(struct btrfs_fs_info
*fs_info
,
3644 struct btrfs_block_rsv
*block_rsv
)
3646 block_rsv
->durable
= 1;
3647 mutex_lock(&fs_info
->durable_block_rsv_mutex
);
3648 list_add_tail(&block_rsv
->list
, &fs_info
->durable_block_rsv_list
);
3649 mutex_unlock(&fs_info
->durable_block_rsv_mutex
);
3652 int btrfs_block_rsv_add(struct btrfs_trans_handle
*trans
,
3653 struct btrfs_root
*root
,
3654 struct btrfs_block_rsv
*block_rsv
,
3662 ret
= reserve_metadata_bytes(trans
, root
, block_rsv
, num_bytes
, 1);
3664 block_rsv_add_bytes(block_rsv
, num_bytes
, 1);
3671 int btrfs_block_rsv_check(struct btrfs_trans_handle
*trans
,
3672 struct btrfs_root
*root
,
3673 struct btrfs_block_rsv
*block_rsv
,
3674 u64 min_reserved
, int min_factor
)
3677 int commit_trans
= 0;
3683 spin_lock(&block_rsv
->lock
);
3685 num_bytes
= div_factor(block_rsv
->size
, min_factor
);
3686 if (min_reserved
> num_bytes
)
3687 num_bytes
= min_reserved
;
3689 if (block_rsv
->reserved
>= num_bytes
) {
3692 num_bytes
-= block_rsv
->reserved
;
3693 if (block_rsv
->durable
&&
3694 block_rsv
->freed
[0] + block_rsv
->freed
[1] >= num_bytes
)
3697 spin_unlock(&block_rsv
->lock
);
3701 if (block_rsv
->refill_used
) {
3702 ret
= reserve_metadata_bytes(trans
, root
, block_rsv
,
3705 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
3714 trans
= btrfs_join_transaction(root
, 1);
3715 BUG_ON(IS_ERR(trans
));
3716 ret
= btrfs_commit_transaction(trans
, root
);
3721 printk(KERN_INFO
"block_rsv size %llu reserved %llu freed %llu %llu\n",
3722 block_rsv
->size
, block_rsv
->reserved
,
3723 block_rsv
->freed
[0], block_rsv
->freed
[1]);
3728 int btrfs_block_rsv_migrate(struct btrfs_block_rsv
*src_rsv
,
3729 struct btrfs_block_rsv
*dst_rsv
,
3732 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
3735 void btrfs_block_rsv_release(struct btrfs_root
*root
,
3736 struct btrfs_block_rsv
*block_rsv
,
3739 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
3740 if (global_rsv
->full
|| global_rsv
== block_rsv
||
3741 block_rsv
->space_info
!= global_rsv
->space_info
)
3743 block_rsv_release_bytes(block_rsv
, global_rsv
, num_bytes
);
3747 * helper to calculate size of global block reservation.
3748 * the desired value is sum of space used by extent tree,
3749 * checksum tree and root tree
3751 static u64
calc_global_metadata_size(struct btrfs_fs_info
*fs_info
)
3753 struct btrfs_space_info
*sinfo
;
3757 int csum_size
= btrfs_super_csum_size(&fs_info
->super_copy
);
3760 * per tree used space accounting can be inaccuracy, so we
3763 spin_lock(&fs_info
->extent_root
->accounting_lock
);
3764 num_bytes
= btrfs_root_used(&fs_info
->extent_root
->root_item
);
3765 spin_unlock(&fs_info
->extent_root
->accounting_lock
);
3767 spin_lock(&fs_info
->csum_root
->accounting_lock
);
3768 num_bytes
+= btrfs_root_used(&fs_info
->csum_root
->root_item
);
3769 spin_unlock(&fs_info
->csum_root
->accounting_lock
);
3771 spin_lock(&fs_info
->tree_root
->accounting_lock
);
3772 num_bytes
+= btrfs_root_used(&fs_info
->tree_root
->root_item
);
3773 spin_unlock(&fs_info
->tree_root
->accounting_lock
);
3775 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_DATA
);
3776 spin_lock(&sinfo
->lock
);
3777 data_used
= sinfo
->bytes_used
;
3778 spin_unlock(&sinfo
->lock
);
3780 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
3781 spin_lock(&sinfo
->lock
);
3782 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_DATA
)
3784 meta_used
= sinfo
->bytes_used
;
3785 spin_unlock(&sinfo
->lock
);
3787 num_bytes
= (data_used
>> fs_info
->sb
->s_blocksize_bits
) *
3789 num_bytes
+= div64_u64(data_used
+ meta_used
, 50);
3791 if (num_bytes
* 3 > meta_used
)
3792 num_bytes
= div64_u64(meta_used
, 3);
3794 return ALIGN(num_bytes
, fs_info
->extent_root
->leafsize
<< 10);
3797 static void update_global_block_rsv(struct btrfs_fs_info
*fs_info
)
3799 struct btrfs_block_rsv
*block_rsv
= &fs_info
->global_block_rsv
;
3800 struct btrfs_space_info
*sinfo
= block_rsv
->space_info
;
3803 num_bytes
= calc_global_metadata_size(fs_info
);
3805 spin_lock(&block_rsv
->lock
);
3806 spin_lock(&sinfo
->lock
);
3808 block_rsv
->size
= num_bytes
;
3810 num_bytes
= sinfo
->bytes_used
+ sinfo
->bytes_pinned
+
3811 sinfo
->bytes_reserved
+ sinfo
->bytes_readonly
+
3812 sinfo
->bytes_may_use
;
3814 if (sinfo
->total_bytes
> num_bytes
) {
3815 num_bytes
= sinfo
->total_bytes
- num_bytes
;
3816 block_rsv
->reserved
+= num_bytes
;
3817 sinfo
->bytes_reserved
+= num_bytes
;
3820 if (block_rsv
->reserved
>= block_rsv
->size
) {
3821 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
3822 sinfo
->bytes_reserved
-= num_bytes
;
3823 block_rsv
->reserved
= block_rsv
->size
;
3824 block_rsv
->full
= 1;
3827 printk(KERN_INFO
"global block rsv size %llu reserved %llu\n",
3828 block_rsv
->size
, block_rsv
->reserved
);
3830 spin_unlock(&sinfo
->lock
);
3831 spin_unlock(&block_rsv
->lock
);
3834 static void init_global_block_rsv(struct btrfs_fs_info
*fs_info
)
3836 struct btrfs_space_info
*space_info
;
3838 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
3839 fs_info
->chunk_block_rsv
.space_info
= space_info
;
3840 fs_info
->chunk_block_rsv
.priority
= 10;
3842 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
3843 fs_info
->global_block_rsv
.space_info
= space_info
;
3844 fs_info
->global_block_rsv
.priority
= 10;
3845 fs_info
->global_block_rsv
.refill_used
= 1;
3846 fs_info
->delalloc_block_rsv
.space_info
= space_info
;
3847 fs_info
->trans_block_rsv
.space_info
= space_info
;
3848 fs_info
->empty_block_rsv
.space_info
= space_info
;
3849 fs_info
->empty_block_rsv
.priority
= 10;
3851 fs_info
->extent_root
->block_rsv
= &fs_info
->global_block_rsv
;
3852 fs_info
->csum_root
->block_rsv
= &fs_info
->global_block_rsv
;
3853 fs_info
->dev_root
->block_rsv
= &fs_info
->global_block_rsv
;
3854 fs_info
->tree_root
->block_rsv
= &fs_info
->global_block_rsv
;
3855 fs_info
->chunk_root
->block_rsv
= &fs_info
->chunk_block_rsv
;
3857 btrfs_add_durable_block_rsv(fs_info
, &fs_info
->global_block_rsv
);
3859 btrfs_add_durable_block_rsv(fs_info
, &fs_info
->delalloc_block_rsv
);
3861 update_global_block_rsv(fs_info
);
3864 static void release_global_block_rsv(struct btrfs_fs_info
*fs_info
)
3866 block_rsv_release_bytes(&fs_info
->global_block_rsv
, NULL
, (u64
)-1);
3867 WARN_ON(fs_info
->delalloc_block_rsv
.size
> 0);
3868 WARN_ON(fs_info
->delalloc_block_rsv
.reserved
> 0);
3869 WARN_ON(fs_info
->trans_block_rsv
.size
> 0);
3870 WARN_ON(fs_info
->trans_block_rsv
.reserved
> 0);
3871 WARN_ON(fs_info
->chunk_block_rsv
.size
> 0);
3872 WARN_ON(fs_info
->chunk_block_rsv
.reserved
> 0);
3875 static u64
calc_trans_metadata_size(struct btrfs_root
*root
, int num_items
)
3877 return (root
->leafsize
+ root
->nodesize
* (BTRFS_MAX_LEVEL
- 1)) *
3881 int btrfs_trans_reserve_metadata(struct btrfs_trans_handle
*trans
,
3882 struct btrfs_root
*root
,
3888 if (num_items
== 0 || root
->fs_info
->chunk_root
== root
)
3891 num_bytes
= calc_trans_metadata_size(root
, num_items
);
3892 ret
= btrfs_block_rsv_add(trans
, root
, &root
->fs_info
->trans_block_rsv
,
3895 trans
->bytes_reserved
+= num_bytes
;
3896 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
3901 void btrfs_trans_release_metadata(struct btrfs_trans_handle
*trans
,
3902 struct btrfs_root
*root
)
3904 if (!trans
->bytes_reserved
)
3907 BUG_ON(trans
->block_rsv
!= &root
->fs_info
->trans_block_rsv
);
3908 btrfs_block_rsv_release(root
, trans
->block_rsv
,
3909 trans
->bytes_reserved
);
3910 trans
->bytes_reserved
= 0;
3913 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle
*trans
,
3914 struct inode
*inode
)
3916 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3917 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
3918 struct btrfs_block_rsv
*dst_rsv
= root
->orphan_block_rsv
;
3921 * one for deleting orphan item, one for updating inode and
3922 * two for calling btrfs_truncate_inode_items.
3924 * btrfs_truncate_inode_items is a delete operation, it frees
3925 * more space than it uses in most cases. So two units of
3926 * metadata space should be enough for calling it many times.
3927 * If all of the metadata space is used, we can commit
3928 * transaction and use space it freed.
3930 u64 num_bytes
= calc_trans_metadata_size(root
, 4);
3931 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
3934 void btrfs_orphan_release_metadata(struct inode
*inode
)
3936 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3937 u64 num_bytes
= calc_trans_metadata_size(root
, 4);
3938 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
, num_bytes
);
3941 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle
*trans
,
3942 struct btrfs_pending_snapshot
*pending
)
3944 struct btrfs_root
*root
= pending
->root
;
3945 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
3946 struct btrfs_block_rsv
*dst_rsv
= &pending
->block_rsv
;
3948 * two for root back/forward refs, two for directory entries
3949 * and one for root of the snapshot.
3951 u64 num_bytes
= calc_trans_metadata_size(root
, 5);
3952 dst_rsv
->space_info
= src_rsv
->space_info
;
3953 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
3956 static u64
calc_csum_metadata_size(struct inode
*inode
, u64 num_bytes
)
3958 return num_bytes
>>= 3;
3961 int btrfs_delalloc_reserve_metadata(struct inode
*inode
, u64 num_bytes
)
3963 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3964 struct btrfs_block_rsv
*block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
3969 if (btrfs_transaction_in_commit(root
->fs_info
))
3970 schedule_timeout(1);
3972 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
3974 spin_lock(&BTRFS_I(inode
)->accounting_lock
);
3975 nr_extents
= atomic_read(&BTRFS_I(inode
)->outstanding_extents
) + 1;
3976 if (nr_extents
> BTRFS_I(inode
)->reserved_extents
) {
3977 nr_extents
-= BTRFS_I(inode
)->reserved_extents
;
3978 to_reserve
= calc_trans_metadata_size(root
, nr_extents
);
3983 spin_unlock(&BTRFS_I(inode
)->accounting_lock
);
3985 to_reserve
+= calc_csum_metadata_size(inode
, num_bytes
);
3986 ret
= reserve_metadata_bytes(NULL
, root
, block_rsv
, to_reserve
, 1);
3990 spin_lock(&BTRFS_I(inode
)->accounting_lock
);
3991 BTRFS_I(inode
)->reserved_extents
+= nr_extents
;
3992 atomic_inc(&BTRFS_I(inode
)->outstanding_extents
);
3993 spin_unlock(&BTRFS_I(inode
)->accounting_lock
);
3995 block_rsv_add_bytes(block_rsv
, to_reserve
, 1);
3997 if (block_rsv
->size
> 512 * 1024 * 1024)
3998 shrink_delalloc(NULL
, root
, to_reserve
, 0);
4003 void btrfs_delalloc_release_metadata(struct inode
*inode
, u64 num_bytes
)
4005 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4009 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
4010 atomic_dec(&BTRFS_I(inode
)->outstanding_extents
);
4012 spin_lock(&BTRFS_I(inode
)->accounting_lock
);
4013 nr_extents
= atomic_read(&BTRFS_I(inode
)->outstanding_extents
);
4014 if (nr_extents
< BTRFS_I(inode
)->reserved_extents
) {
4015 nr_extents
= BTRFS_I(inode
)->reserved_extents
- nr_extents
;
4016 BTRFS_I(inode
)->reserved_extents
-= nr_extents
;
4020 spin_unlock(&BTRFS_I(inode
)->accounting_lock
);
4022 to_free
= calc_csum_metadata_size(inode
, num_bytes
);
4024 to_free
+= calc_trans_metadata_size(root
, nr_extents
);
4026 btrfs_block_rsv_release(root
, &root
->fs_info
->delalloc_block_rsv
,
4030 int btrfs_delalloc_reserve_space(struct inode
*inode
, u64 num_bytes
)
4034 ret
= btrfs_check_data_free_space(inode
, num_bytes
);
4038 ret
= btrfs_delalloc_reserve_metadata(inode
, num_bytes
);
4040 btrfs_free_reserved_data_space(inode
, num_bytes
);
4047 void btrfs_delalloc_release_space(struct inode
*inode
, u64 num_bytes
)
4049 btrfs_delalloc_release_metadata(inode
, num_bytes
);
4050 btrfs_free_reserved_data_space(inode
, num_bytes
);
4053 static int update_block_group(struct btrfs_trans_handle
*trans
,
4054 struct btrfs_root
*root
,
4055 u64 bytenr
, u64 num_bytes
, int alloc
)
4057 struct btrfs_block_group_cache
*cache
= NULL
;
4058 struct btrfs_fs_info
*info
= root
->fs_info
;
4059 u64 total
= num_bytes
;
4064 /* block accounting for super block */
4065 spin_lock(&info
->delalloc_lock
);
4066 old_val
= btrfs_super_bytes_used(&info
->super_copy
);
4068 old_val
+= num_bytes
;
4070 old_val
-= num_bytes
;
4071 btrfs_set_super_bytes_used(&info
->super_copy
, old_val
);
4072 spin_unlock(&info
->delalloc_lock
);
4075 cache
= btrfs_lookup_block_group(info
, bytenr
);
4078 if (cache
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
4079 BTRFS_BLOCK_GROUP_RAID1
|
4080 BTRFS_BLOCK_GROUP_RAID10
))
4085 * If this block group has free space cache written out, we
4086 * need to make sure to load it if we are removing space. This
4087 * is because we need the unpinning stage to actually add the
4088 * space back to the block group, otherwise we will leak space.
4090 if (!alloc
&& cache
->cached
== BTRFS_CACHE_NO
)
4091 cache_block_group(cache
, trans
, 1);
4093 byte_in_group
= bytenr
- cache
->key
.objectid
;
4094 WARN_ON(byte_in_group
> cache
->key
.offset
);
4096 spin_lock(&cache
->space_info
->lock
);
4097 spin_lock(&cache
->lock
);
4099 if (btrfs_super_cache_generation(&info
->super_copy
) != 0 &&
4100 cache
->disk_cache_state
< BTRFS_DC_CLEAR
)
4101 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
4104 old_val
= btrfs_block_group_used(&cache
->item
);
4105 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
4107 old_val
+= num_bytes
;
4108 btrfs_set_block_group_used(&cache
->item
, old_val
);
4109 cache
->reserved
-= num_bytes
;
4110 cache
->space_info
->bytes_reserved
-= num_bytes
;
4111 cache
->space_info
->bytes_used
+= num_bytes
;
4112 cache
->space_info
->disk_used
+= num_bytes
* factor
;
4113 spin_unlock(&cache
->lock
);
4114 spin_unlock(&cache
->space_info
->lock
);
4116 old_val
-= num_bytes
;
4117 btrfs_set_block_group_used(&cache
->item
, old_val
);
4118 cache
->pinned
+= num_bytes
;
4119 cache
->space_info
->bytes_pinned
+= num_bytes
;
4120 cache
->space_info
->bytes_used
-= num_bytes
;
4121 cache
->space_info
->disk_used
-= num_bytes
* factor
;
4122 spin_unlock(&cache
->lock
);
4123 spin_unlock(&cache
->space_info
->lock
);
4125 set_extent_dirty(info
->pinned_extents
,
4126 bytenr
, bytenr
+ num_bytes
- 1,
4127 GFP_NOFS
| __GFP_NOFAIL
);
4129 btrfs_put_block_group(cache
);
4131 bytenr
+= num_bytes
;
4136 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
4138 struct btrfs_block_group_cache
*cache
;
4141 cache
= btrfs_lookup_first_block_group(root
->fs_info
, search_start
);
4145 bytenr
= cache
->key
.objectid
;
4146 btrfs_put_block_group(cache
);
4151 static int pin_down_extent(struct btrfs_root
*root
,
4152 struct btrfs_block_group_cache
*cache
,
4153 u64 bytenr
, u64 num_bytes
, int reserved
)
4155 spin_lock(&cache
->space_info
->lock
);
4156 spin_lock(&cache
->lock
);
4157 cache
->pinned
+= num_bytes
;
4158 cache
->space_info
->bytes_pinned
+= num_bytes
;
4160 cache
->reserved
-= num_bytes
;
4161 cache
->space_info
->bytes_reserved
-= num_bytes
;
4163 spin_unlock(&cache
->lock
);
4164 spin_unlock(&cache
->space_info
->lock
);
4166 set_extent_dirty(root
->fs_info
->pinned_extents
, bytenr
,
4167 bytenr
+ num_bytes
- 1, GFP_NOFS
| __GFP_NOFAIL
);
4172 * this function must be called within transaction
4174 int btrfs_pin_extent(struct btrfs_root
*root
,
4175 u64 bytenr
, u64 num_bytes
, int reserved
)
4177 struct btrfs_block_group_cache
*cache
;
4179 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
4182 pin_down_extent(root
, cache
, bytenr
, num_bytes
, reserved
);
4184 btrfs_put_block_group(cache
);
4189 * update size of reserved extents. this function may return -EAGAIN
4190 * if 'reserve' is true or 'sinfo' is false.
4192 static int update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
4193 u64 num_bytes
, int reserve
, int sinfo
)
4197 struct btrfs_space_info
*space_info
= cache
->space_info
;
4198 spin_lock(&space_info
->lock
);
4199 spin_lock(&cache
->lock
);
4204 cache
->reserved
+= num_bytes
;
4205 space_info
->bytes_reserved
+= num_bytes
;
4209 space_info
->bytes_readonly
+= num_bytes
;
4210 cache
->reserved
-= num_bytes
;
4211 space_info
->bytes_reserved
-= num_bytes
;
4213 spin_unlock(&cache
->lock
);
4214 spin_unlock(&space_info
->lock
);
4216 spin_lock(&cache
->lock
);
4221 cache
->reserved
+= num_bytes
;
4223 cache
->reserved
-= num_bytes
;
4225 spin_unlock(&cache
->lock
);
4230 int btrfs_prepare_extent_commit(struct btrfs_trans_handle
*trans
,
4231 struct btrfs_root
*root
)
4233 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4234 struct btrfs_caching_control
*next
;
4235 struct btrfs_caching_control
*caching_ctl
;
4236 struct btrfs_block_group_cache
*cache
;
4238 down_write(&fs_info
->extent_commit_sem
);
4240 list_for_each_entry_safe(caching_ctl
, next
,
4241 &fs_info
->caching_block_groups
, list
) {
4242 cache
= caching_ctl
->block_group
;
4243 if (block_group_cache_done(cache
)) {
4244 cache
->last_byte_to_unpin
= (u64
)-1;
4245 list_del_init(&caching_ctl
->list
);
4246 put_caching_control(caching_ctl
);
4248 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
4252 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
4253 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
4255 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
4257 up_write(&fs_info
->extent_commit_sem
);
4259 update_global_block_rsv(fs_info
);
4263 static int unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
4265 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4266 struct btrfs_block_group_cache
*cache
= NULL
;
4269 while (start
<= end
) {
4271 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
4273 btrfs_put_block_group(cache
);
4274 cache
= btrfs_lookup_block_group(fs_info
, start
);
4278 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
4279 len
= min(len
, end
+ 1 - start
);
4281 if (start
< cache
->last_byte_to_unpin
) {
4282 len
= min(len
, cache
->last_byte_to_unpin
- start
);
4283 btrfs_add_free_space(cache
, start
, len
);
4288 spin_lock(&cache
->space_info
->lock
);
4289 spin_lock(&cache
->lock
);
4290 cache
->pinned
-= len
;
4291 cache
->space_info
->bytes_pinned
-= len
;
4293 cache
->space_info
->bytes_readonly
+= len
;
4294 } else if (cache
->reserved_pinned
> 0) {
4295 len
= min(len
, cache
->reserved_pinned
);
4296 cache
->reserved_pinned
-= len
;
4297 cache
->space_info
->bytes_reserved
+= len
;
4299 spin_unlock(&cache
->lock
);
4300 spin_unlock(&cache
->space_info
->lock
);
4304 btrfs_put_block_group(cache
);
4308 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
4309 struct btrfs_root
*root
)
4311 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4312 struct extent_io_tree
*unpin
;
4313 struct btrfs_block_rsv
*block_rsv
;
4314 struct btrfs_block_rsv
*next_rsv
;
4320 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
4321 unpin
= &fs_info
->freed_extents
[1];
4323 unpin
= &fs_info
->freed_extents
[0];
4326 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
4331 ret
= btrfs_discard_extent(root
, start
, end
+ 1 - start
);
4333 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
4334 unpin_extent_range(root
, start
, end
);
4338 mutex_lock(&fs_info
->durable_block_rsv_mutex
);
4339 list_for_each_entry_safe(block_rsv
, next_rsv
,
4340 &fs_info
->durable_block_rsv_list
, list
) {
4342 idx
= trans
->transid
& 0x1;
4343 if (block_rsv
->freed
[idx
] > 0) {
4344 block_rsv_add_bytes(block_rsv
,
4345 block_rsv
->freed
[idx
], 0);
4346 block_rsv
->freed
[idx
] = 0;
4348 if (atomic_read(&block_rsv
->usage
) == 0) {
4349 btrfs_block_rsv_release(root
, block_rsv
, (u64
)-1);
4351 if (block_rsv
->freed
[0] == 0 &&
4352 block_rsv
->freed
[1] == 0) {
4353 list_del_init(&block_rsv
->list
);
4357 btrfs_block_rsv_release(root
, block_rsv
, 0);
4360 mutex_unlock(&fs_info
->durable_block_rsv_mutex
);
4365 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
4366 struct btrfs_root
*root
,
4367 u64 bytenr
, u64 num_bytes
, u64 parent
,
4368 u64 root_objectid
, u64 owner_objectid
,
4369 u64 owner_offset
, int refs_to_drop
,
4370 struct btrfs_delayed_extent_op
*extent_op
)
4372 struct btrfs_key key
;
4373 struct btrfs_path
*path
;
4374 struct btrfs_fs_info
*info
= root
->fs_info
;
4375 struct btrfs_root
*extent_root
= info
->extent_root
;
4376 struct extent_buffer
*leaf
;
4377 struct btrfs_extent_item
*ei
;
4378 struct btrfs_extent_inline_ref
*iref
;
4381 int extent_slot
= 0;
4382 int found_extent
= 0;
4387 path
= btrfs_alloc_path();
4392 path
->leave_spinning
= 1;
4394 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
4395 BUG_ON(!is_data
&& refs_to_drop
!= 1);
4397 ret
= lookup_extent_backref(trans
, extent_root
, path
, &iref
,
4398 bytenr
, num_bytes
, parent
,
4399 root_objectid
, owner_objectid
,
4402 extent_slot
= path
->slots
[0];
4403 while (extent_slot
>= 0) {
4404 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
4406 if (key
.objectid
!= bytenr
)
4408 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
4409 key
.offset
== num_bytes
) {
4413 if (path
->slots
[0] - extent_slot
> 5)
4417 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4418 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
4419 if (found_extent
&& item_size
< sizeof(*ei
))
4422 if (!found_extent
) {
4424 ret
= remove_extent_backref(trans
, extent_root
, path
,
4428 btrfs_release_path(extent_root
, path
);
4429 path
->leave_spinning
= 1;
4431 key
.objectid
= bytenr
;
4432 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
4433 key
.offset
= num_bytes
;
4435 ret
= btrfs_search_slot(trans
, extent_root
,
4438 printk(KERN_ERR
"umm, got %d back from search"
4439 ", was looking for %llu\n", ret
,
4440 (unsigned long long)bytenr
);
4441 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
4444 extent_slot
= path
->slots
[0];
4447 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
4449 printk(KERN_ERR
"btrfs unable to find ref byte nr %llu "
4450 "parent %llu root %llu owner %llu offset %llu\n",
4451 (unsigned long long)bytenr
,
4452 (unsigned long long)parent
,
4453 (unsigned long long)root_objectid
,
4454 (unsigned long long)owner_objectid
,
4455 (unsigned long long)owner_offset
);
4458 leaf
= path
->nodes
[0];
4459 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
4460 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4461 if (item_size
< sizeof(*ei
)) {
4462 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
4463 ret
= convert_extent_item_v0(trans
, extent_root
, path
,
4467 btrfs_release_path(extent_root
, path
);
4468 path
->leave_spinning
= 1;
4470 key
.objectid
= bytenr
;
4471 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
4472 key
.offset
= num_bytes
;
4474 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
4477 printk(KERN_ERR
"umm, got %d back from search"
4478 ", was looking for %llu\n", ret
,
4479 (unsigned long long)bytenr
);
4480 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
4483 extent_slot
= path
->slots
[0];
4484 leaf
= path
->nodes
[0];
4485 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
4488 BUG_ON(item_size
< sizeof(*ei
));
4489 ei
= btrfs_item_ptr(leaf
, extent_slot
,
4490 struct btrfs_extent_item
);
4491 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
) {
4492 struct btrfs_tree_block_info
*bi
;
4493 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
4494 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
4495 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
4498 refs
= btrfs_extent_refs(leaf
, ei
);
4499 BUG_ON(refs
< refs_to_drop
);
4500 refs
-= refs_to_drop
;
4504 __run_delayed_extent_op(extent_op
, leaf
, ei
);
4506 * In the case of inline back ref, reference count will
4507 * be updated by remove_extent_backref
4510 BUG_ON(!found_extent
);
4512 btrfs_set_extent_refs(leaf
, ei
, refs
);
4513 btrfs_mark_buffer_dirty(leaf
);
4516 ret
= remove_extent_backref(trans
, extent_root
, path
,
4523 BUG_ON(is_data
&& refs_to_drop
!=
4524 extent_data_ref_count(root
, path
, iref
));
4526 BUG_ON(path
->slots
[0] != extent_slot
);
4528 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
4529 path
->slots
[0] = extent_slot
;
4534 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
4537 btrfs_release_path(extent_root
, path
);
4540 ret
= btrfs_del_csums(trans
, root
, bytenr
, num_bytes
);
4543 invalidate_mapping_pages(info
->btree_inode
->i_mapping
,
4544 bytenr
>> PAGE_CACHE_SHIFT
,
4545 (bytenr
+ num_bytes
- 1) >> PAGE_CACHE_SHIFT
);
4548 ret
= update_block_group(trans
, root
, bytenr
, num_bytes
, 0);
4551 btrfs_free_path(path
);
4556 * when we free an block, it is possible (and likely) that we free the last
4557 * delayed ref for that extent as well. This searches the delayed ref tree for
4558 * a given extent, and if there are no other delayed refs to be processed, it
4559 * removes it from the tree.
4561 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
4562 struct btrfs_root
*root
, u64 bytenr
)
4564 struct btrfs_delayed_ref_head
*head
;
4565 struct btrfs_delayed_ref_root
*delayed_refs
;
4566 struct btrfs_delayed_ref_node
*ref
;
4567 struct rb_node
*node
;
4570 delayed_refs
= &trans
->transaction
->delayed_refs
;
4571 spin_lock(&delayed_refs
->lock
);
4572 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
4576 node
= rb_prev(&head
->node
.rb_node
);
4580 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
4582 /* there are still entries for this ref, we can't drop it */
4583 if (ref
->bytenr
== bytenr
)
4586 if (head
->extent_op
) {
4587 if (!head
->must_insert_reserved
)
4589 kfree(head
->extent_op
);
4590 head
->extent_op
= NULL
;
4594 * waiting for the lock here would deadlock. If someone else has it
4595 * locked they are already in the process of dropping it anyway
4597 if (!mutex_trylock(&head
->mutex
))
4601 * at this point we have a head with no other entries. Go
4602 * ahead and process it.
4604 head
->node
.in_tree
= 0;
4605 rb_erase(&head
->node
.rb_node
, &delayed_refs
->root
);
4607 delayed_refs
->num_entries
--;
4610 * we don't take a ref on the node because we're removing it from the
4611 * tree, so we just steal the ref the tree was holding.
4613 delayed_refs
->num_heads
--;
4614 if (list_empty(&head
->cluster
))
4615 delayed_refs
->num_heads_ready
--;
4617 list_del_init(&head
->cluster
);
4618 spin_unlock(&delayed_refs
->lock
);
4620 BUG_ON(head
->extent_op
);
4621 if (head
->must_insert_reserved
)
4624 mutex_unlock(&head
->mutex
);
4625 btrfs_put_delayed_ref(&head
->node
);
4628 spin_unlock(&delayed_refs
->lock
);
4632 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
4633 struct btrfs_root
*root
,
4634 struct extent_buffer
*buf
,
4635 u64 parent
, int last_ref
)
4637 struct btrfs_block_rsv
*block_rsv
;
4638 struct btrfs_block_group_cache
*cache
= NULL
;
4641 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
4642 ret
= btrfs_add_delayed_tree_ref(trans
, buf
->start
, buf
->len
,
4643 parent
, root
->root_key
.objectid
,
4644 btrfs_header_level(buf
),
4645 BTRFS_DROP_DELAYED_REF
, NULL
);
4652 block_rsv
= get_block_rsv(trans
, root
);
4653 cache
= btrfs_lookup_block_group(root
->fs_info
, buf
->start
);
4654 if (block_rsv
->space_info
!= cache
->space_info
)
4657 if (btrfs_header_generation(buf
) == trans
->transid
) {
4658 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
4659 ret
= check_ref_cleanup(trans
, root
, buf
->start
);
4664 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
4665 pin_down_extent(root
, cache
, buf
->start
, buf
->len
, 1);
4669 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
4671 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
4672 ret
= update_reserved_bytes(cache
, buf
->len
, 0, 0);
4673 if (ret
== -EAGAIN
) {
4674 /* block group became read-only */
4675 update_reserved_bytes(cache
, buf
->len
, 0, 1);
4680 spin_lock(&block_rsv
->lock
);
4681 if (block_rsv
->reserved
< block_rsv
->size
) {
4682 block_rsv
->reserved
+= buf
->len
;
4685 spin_unlock(&block_rsv
->lock
);
4688 spin_lock(&cache
->space_info
->lock
);
4689 cache
->space_info
->bytes_reserved
-= buf
->len
;
4690 spin_unlock(&cache
->space_info
->lock
);
4695 if (block_rsv
->durable
&& !cache
->ro
) {
4697 spin_lock(&cache
->lock
);
4699 cache
->reserved_pinned
+= buf
->len
;
4702 spin_unlock(&cache
->lock
);
4705 spin_lock(&block_rsv
->lock
);
4706 block_rsv
->freed
[trans
->transid
& 0x1] += buf
->len
;
4707 spin_unlock(&block_rsv
->lock
);
4711 btrfs_put_block_group(cache
);
4714 int btrfs_free_extent(struct btrfs_trans_handle
*trans
,
4715 struct btrfs_root
*root
,
4716 u64 bytenr
, u64 num_bytes
, u64 parent
,
4717 u64 root_objectid
, u64 owner
, u64 offset
)
4722 * tree log blocks never actually go into the extent allocation
4723 * tree, just update pinning info and exit early.
4725 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
4726 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
4727 /* unlocks the pinned mutex */
4728 btrfs_pin_extent(root
, bytenr
, num_bytes
, 1);
4730 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
4731 ret
= btrfs_add_delayed_tree_ref(trans
, bytenr
, num_bytes
,
4732 parent
, root_objectid
, (int)owner
,
4733 BTRFS_DROP_DELAYED_REF
, NULL
);
4736 ret
= btrfs_add_delayed_data_ref(trans
, bytenr
, num_bytes
,
4737 parent
, root_objectid
, owner
,
4738 offset
, BTRFS_DROP_DELAYED_REF
, NULL
);
4744 static u64
stripe_align(struct btrfs_root
*root
, u64 val
)
4746 u64 mask
= ((u64
)root
->stripesize
- 1);
4747 u64 ret
= (val
+ mask
) & ~mask
;
4752 * when we wait for progress in the block group caching, its because
4753 * our allocation attempt failed at least once. So, we must sleep
4754 * and let some progress happen before we try again.
4756 * This function will sleep at least once waiting for new free space to
4757 * show up, and then it will check the block group free space numbers
4758 * for our min num_bytes. Another option is to have it go ahead
4759 * and look in the rbtree for a free extent of a given size, but this
4763 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
4766 struct btrfs_caching_control
*caching_ctl
;
4769 caching_ctl
= get_caching_control(cache
);
4773 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
4774 (cache
->free_space
>= num_bytes
));
4776 put_caching_control(caching_ctl
);
4781 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
4783 struct btrfs_caching_control
*caching_ctl
;
4786 caching_ctl
= get_caching_control(cache
);
4790 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
4792 put_caching_control(caching_ctl
);
4796 static int get_block_group_index(struct btrfs_block_group_cache
*cache
)
4799 if (cache
->flags
& BTRFS_BLOCK_GROUP_RAID10
)
4801 else if (cache
->flags
& BTRFS_BLOCK_GROUP_RAID1
)
4803 else if (cache
->flags
& BTRFS_BLOCK_GROUP_DUP
)
4805 else if (cache
->flags
& BTRFS_BLOCK_GROUP_RAID0
)
4812 enum btrfs_loop_type
{
4813 LOOP_FIND_IDEAL
= 0,
4814 LOOP_CACHING_NOWAIT
= 1,
4815 LOOP_CACHING_WAIT
= 2,
4816 LOOP_ALLOC_CHUNK
= 3,
4817 LOOP_NO_EMPTY_SIZE
= 4,
4821 * walks the btree of allocated extents and find a hole of a given size.
4822 * The key ins is changed to record the hole:
4823 * ins->objectid == block start
4824 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4825 * ins->offset == number of blocks
4826 * Any available blocks before search_start are skipped.
4828 static noinline
int find_free_extent(struct btrfs_trans_handle
*trans
,
4829 struct btrfs_root
*orig_root
,
4830 u64 num_bytes
, u64 empty_size
,
4831 u64 search_start
, u64 search_end
,
4832 u64 hint_byte
, struct btrfs_key
*ins
,
4836 struct btrfs_root
*root
= orig_root
->fs_info
->extent_root
;
4837 struct btrfs_free_cluster
*last_ptr
= NULL
;
4838 struct btrfs_block_group_cache
*block_group
= NULL
;
4839 int empty_cluster
= 2 * 1024 * 1024;
4840 int allowed_chunk_alloc
= 0;
4841 int done_chunk_alloc
= 0;
4842 struct btrfs_space_info
*space_info
;
4843 int last_ptr_loop
= 0;
4846 bool found_uncached_bg
= false;
4847 bool failed_cluster_refill
= false;
4848 bool failed_alloc
= false;
4849 bool use_cluster
= true;
4850 u64 ideal_cache_percent
= 0;
4851 u64 ideal_cache_offset
= 0;
4853 WARN_ON(num_bytes
< root
->sectorsize
);
4854 btrfs_set_key_type(ins
, BTRFS_EXTENT_ITEM_KEY
);
4858 space_info
= __find_space_info(root
->fs_info
, data
);
4860 printk(KERN_ERR
"No space info for %d\n", data
);
4865 * If the space info is for both data and metadata it means we have a
4866 * small filesystem and we can't use the clustering stuff.
4868 if (btrfs_mixed_space_info(space_info
))
4869 use_cluster
= false;
4871 if (orig_root
->ref_cows
|| empty_size
)
4872 allowed_chunk_alloc
= 1;
4874 if (data
& BTRFS_BLOCK_GROUP_METADATA
&& use_cluster
) {
4875 last_ptr
= &root
->fs_info
->meta_alloc_cluster
;
4876 if (!btrfs_test_opt(root
, SSD
))
4877 empty_cluster
= 64 * 1024;
4880 if ((data
& BTRFS_BLOCK_GROUP_DATA
) && use_cluster
&&
4881 btrfs_test_opt(root
, SSD
)) {
4882 last_ptr
= &root
->fs_info
->data_alloc_cluster
;
4886 spin_lock(&last_ptr
->lock
);
4887 if (last_ptr
->block_group
)
4888 hint_byte
= last_ptr
->window_start
;
4889 spin_unlock(&last_ptr
->lock
);
4892 search_start
= max(search_start
, first_logical_byte(root
, 0));
4893 search_start
= max(search_start
, hint_byte
);
4898 if (search_start
== hint_byte
) {
4900 block_group
= btrfs_lookup_block_group(root
->fs_info
,
4903 * we don't want to use the block group if it doesn't match our
4904 * allocation bits, or if its not cached.
4906 * However if we are re-searching with an ideal block group
4907 * picked out then we don't care that the block group is cached.
4909 if (block_group
&& block_group_bits(block_group
, data
) &&
4910 (block_group
->cached
!= BTRFS_CACHE_NO
||
4911 search_start
== ideal_cache_offset
)) {
4912 down_read(&space_info
->groups_sem
);
4913 if (list_empty(&block_group
->list
) ||
4916 * someone is removing this block group,
4917 * we can't jump into the have_block_group
4918 * target because our list pointers are not
4921 btrfs_put_block_group(block_group
);
4922 up_read(&space_info
->groups_sem
);
4924 index
= get_block_group_index(block_group
);
4925 goto have_block_group
;
4927 } else if (block_group
) {
4928 btrfs_put_block_group(block_group
);
4932 down_read(&space_info
->groups_sem
);
4933 list_for_each_entry(block_group
, &space_info
->block_groups
[index
],
4938 btrfs_get_block_group(block_group
);
4939 search_start
= block_group
->key
.objectid
;
4942 if (unlikely(block_group
->cached
== BTRFS_CACHE_NO
)) {
4945 ret
= cache_block_group(block_group
, trans
, 1);
4946 if (block_group
->cached
== BTRFS_CACHE_FINISHED
)
4947 goto have_block_group
;
4949 free_percent
= btrfs_block_group_used(&block_group
->item
);
4950 free_percent
*= 100;
4951 free_percent
= div64_u64(free_percent
,
4952 block_group
->key
.offset
);
4953 free_percent
= 100 - free_percent
;
4954 if (free_percent
> ideal_cache_percent
&&
4955 likely(!block_group
->ro
)) {
4956 ideal_cache_offset
= block_group
->key
.objectid
;
4957 ideal_cache_percent
= free_percent
;
4961 * We only want to start kthread caching if we are at
4962 * the point where we will wait for caching to make
4963 * progress, or if our ideal search is over and we've
4964 * found somebody to start caching.
4966 if (loop
> LOOP_CACHING_NOWAIT
||
4967 (loop
> LOOP_FIND_IDEAL
&&
4968 atomic_read(&space_info
->caching_threads
) < 2)) {
4969 ret
= cache_block_group(block_group
, trans
, 0);
4972 found_uncached_bg
= true;
4975 * If loop is set for cached only, try the next block
4978 if (loop
== LOOP_FIND_IDEAL
)
4982 cached
= block_group_cache_done(block_group
);
4983 if (unlikely(!cached
))
4984 found_uncached_bg
= true;
4986 if (unlikely(block_group
->ro
))
4990 * Ok we want to try and use the cluster allocator, so lets look
4991 * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
4992 * have tried the cluster allocator plenty of times at this
4993 * point and not have found anything, so we are likely way too
4994 * fragmented for the clustering stuff to find anything, so lets
4995 * just skip it and let the allocator find whatever block it can
4998 if (last_ptr
&& loop
< LOOP_NO_EMPTY_SIZE
) {
5000 * the refill lock keeps out other
5001 * people trying to start a new cluster
5003 spin_lock(&last_ptr
->refill_lock
);
5004 if (last_ptr
->block_group
&&
5005 (last_ptr
->block_group
->ro
||
5006 !block_group_bits(last_ptr
->block_group
, data
))) {
5008 goto refill_cluster
;
5011 offset
= btrfs_alloc_from_cluster(block_group
, last_ptr
,
5012 num_bytes
, search_start
);
5014 /* we have a block, we're done */
5015 spin_unlock(&last_ptr
->refill_lock
);
5019 spin_lock(&last_ptr
->lock
);
5021 * whoops, this cluster doesn't actually point to
5022 * this block group. Get a ref on the block
5023 * group is does point to and try again
5025 if (!last_ptr_loop
&& last_ptr
->block_group
&&
5026 last_ptr
->block_group
!= block_group
) {
5028 btrfs_put_block_group(block_group
);
5029 block_group
= last_ptr
->block_group
;
5030 btrfs_get_block_group(block_group
);
5031 spin_unlock(&last_ptr
->lock
);
5032 spin_unlock(&last_ptr
->refill_lock
);
5035 search_start
= block_group
->key
.objectid
;
5037 * we know this block group is properly
5038 * in the list because
5039 * btrfs_remove_block_group, drops the
5040 * cluster before it removes the block
5041 * group from the list
5043 goto have_block_group
;
5045 spin_unlock(&last_ptr
->lock
);
5048 * this cluster didn't work out, free it and
5051 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
5055 /* allocate a cluster in this block group */
5056 ret
= btrfs_find_space_cluster(trans
, root
,
5057 block_group
, last_ptr
,
5059 empty_cluster
+ empty_size
);
5062 * now pull our allocation out of this
5065 offset
= btrfs_alloc_from_cluster(block_group
,
5066 last_ptr
, num_bytes
,
5069 /* we found one, proceed */
5070 spin_unlock(&last_ptr
->refill_lock
);
5073 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
5074 && !failed_cluster_refill
) {
5075 spin_unlock(&last_ptr
->refill_lock
);
5077 failed_cluster_refill
= true;
5078 wait_block_group_cache_progress(block_group
,
5079 num_bytes
+ empty_cluster
+ empty_size
);
5080 goto have_block_group
;
5084 * at this point we either didn't find a cluster
5085 * or we weren't able to allocate a block from our
5086 * cluster. Free the cluster we've been trying
5087 * to use, and go to the next block group
5089 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
5090 spin_unlock(&last_ptr
->refill_lock
);
5094 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
5095 num_bytes
, empty_size
);
5097 * If we didn't find a chunk, and we haven't failed on this
5098 * block group before, and this block group is in the middle of
5099 * caching and we are ok with waiting, then go ahead and wait
5100 * for progress to be made, and set failed_alloc to true.
5102 * If failed_alloc is true then we've already waited on this
5103 * block group once and should move on to the next block group.
5105 if (!offset
&& !failed_alloc
&& !cached
&&
5106 loop
> LOOP_CACHING_NOWAIT
) {
5107 wait_block_group_cache_progress(block_group
,
5108 num_bytes
+ empty_size
);
5109 failed_alloc
= true;
5110 goto have_block_group
;
5111 } else if (!offset
) {
5115 search_start
= stripe_align(root
, offset
);
5116 /* move on to the next group */
5117 if (search_start
+ num_bytes
>= search_end
) {
5118 btrfs_add_free_space(block_group
, offset
, num_bytes
);
5122 /* move on to the next group */
5123 if (search_start
+ num_bytes
>
5124 block_group
->key
.objectid
+ block_group
->key
.offset
) {
5125 btrfs_add_free_space(block_group
, offset
, num_bytes
);
5129 ins
->objectid
= search_start
;
5130 ins
->offset
= num_bytes
;
5132 if (offset
< search_start
)
5133 btrfs_add_free_space(block_group
, offset
,
5134 search_start
- offset
);
5135 BUG_ON(offset
> search_start
);
5137 ret
= update_reserved_bytes(block_group
, num_bytes
, 1,
5138 (data
& BTRFS_BLOCK_GROUP_DATA
));
5139 if (ret
== -EAGAIN
) {
5140 btrfs_add_free_space(block_group
, offset
, num_bytes
);
5144 /* we are all good, lets return */
5145 ins
->objectid
= search_start
;
5146 ins
->offset
= num_bytes
;
5148 if (offset
< search_start
)
5149 btrfs_add_free_space(block_group
, offset
,
5150 search_start
- offset
);
5151 BUG_ON(offset
> search_start
);
5154 failed_cluster_refill
= false;
5155 failed_alloc
= false;
5156 BUG_ON(index
!= get_block_group_index(block_group
));
5157 btrfs_put_block_group(block_group
);
5159 up_read(&space_info
->groups_sem
);
5161 if (!ins
->objectid
&& ++index
< BTRFS_NR_RAID_TYPES
)
5164 /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5165 * for them to make caching progress. Also
5166 * determine the best possible bg to cache
5167 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5168 * caching kthreads as we move along
5169 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5170 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5171 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5174 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
&&
5175 (found_uncached_bg
|| empty_size
|| empty_cluster
||
5176 allowed_chunk_alloc
)) {
5178 if (loop
== LOOP_FIND_IDEAL
&& found_uncached_bg
) {
5179 found_uncached_bg
= false;
5181 if (!ideal_cache_percent
&&
5182 atomic_read(&space_info
->caching_threads
))
5186 * 1 of the following 2 things have happened so far
5188 * 1) We found an ideal block group for caching that
5189 * is mostly full and will cache quickly, so we might
5190 * as well wait for it.
5192 * 2) We searched for cached only and we didn't find
5193 * anything, and we didn't start any caching kthreads
5194 * either, so chances are we will loop through and
5195 * start a couple caching kthreads, and then come back
5196 * around and just wait for them. This will be slower
5197 * because we will have 2 caching kthreads reading at
5198 * the same time when we could have just started one
5199 * and waited for it to get far enough to give us an
5200 * allocation, so go ahead and go to the wait caching
5203 loop
= LOOP_CACHING_WAIT
;
5204 search_start
= ideal_cache_offset
;
5205 ideal_cache_percent
= 0;
5207 } else if (loop
== LOOP_FIND_IDEAL
) {
5209 * Didn't find a uncached bg, wait on anything we find
5212 loop
= LOOP_CACHING_WAIT
;
5216 if (loop
< LOOP_CACHING_WAIT
) {
5221 if (loop
== LOOP_ALLOC_CHUNK
) {
5226 if (allowed_chunk_alloc
) {
5227 ret
= do_chunk_alloc(trans
, root
, num_bytes
+
5228 2 * 1024 * 1024, data
, 1);
5229 allowed_chunk_alloc
= 0;
5230 done_chunk_alloc
= 1;
5231 } else if (!done_chunk_alloc
) {
5232 space_info
->force_alloc
= 1;
5235 if (loop
< LOOP_NO_EMPTY_SIZE
) {
5240 } else if (!ins
->objectid
) {
5244 /* we found what we needed */
5245 if (ins
->objectid
) {
5246 if (!(data
& BTRFS_BLOCK_GROUP_DATA
))
5247 trans
->block_group
= block_group
->key
.objectid
;
5249 btrfs_put_block_group(block_group
);
5256 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
5257 int dump_block_groups
)
5259 struct btrfs_block_group_cache
*cache
;
5262 spin_lock(&info
->lock
);
5263 printk(KERN_INFO
"space_info has %llu free, is %sfull\n",
5264 (unsigned long long)(info
->total_bytes
- info
->bytes_used
-
5265 info
->bytes_pinned
- info
->bytes_reserved
-
5266 info
->bytes_readonly
),
5267 (info
->full
) ? "" : "not ");
5268 printk(KERN_INFO
"space_info total=%llu, used=%llu, pinned=%llu, "
5269 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5270 (unsigned long long)info
->total_bytes
,
5271 (unsigned long long)info
->bytes_used
,
5272 (unsigned long long)info
->bytes_pinned
,
5273 (unsigned long long)info
->bytes_reserved
,
5274 (unsigned long long)info
->bytes_may_use
,
5275 (unsigned long long)info
->bytes_readonly
);
5276 spin_unlock(&info
->lock
);
5278 if (!dump_block_groups
)
5281 down_read(&info
->groups_sem
);
5283 list_for_each_entry(cache
, &info
->block_groups
[index
], list
) {
5284 spin_lock(&cache
->lock
);
5285 printk(KERN_INFO
"block group %llu has %llu bytes, %llu used "
5286 "%llu pinned %llu reserved\n",
5287 (unsigned long long)cache
->key
.objectid
,
5288 (unsigned long long)cache
->key
.offset
,
5289 (unsigned long long)btrfs_block_group_used(&cache
->item
),
5290 (unsigned long long)cache
->pinned
,
5291 (unsigned long long)cache
->reserved
);
5292 btrfs_dump_free_space(cache
, bytes
);
5293 spin_unlock(&cache
->lock
);
5295 if (++index
< BTRFS_NR_RAID_TYPES
)
5297 up_read(&info
->groups_sem
);
5300 int btrfs_reserve_extent(struct btrfs_trans_handle
*trans
,
5301 struct btrfs_root
*root
,
5302 u64 num_bytes
, u64 min_alloc_size
,
5303 u64 empty_size
, u64 hint_byte
,
5304 u64 search_end
, struct btrfs_key
*ins
,
5308 u64 search_start
= 0;
5310 data
= btrfs_get_alloc_profile(root
, data
);
5313 * the only place that sets empty_size is btrfs_realloc_node, which
5314 * is not called recursively on allocations
5316 if (empty_size
|| root
->ref_cows
)
5317 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
5318 num_bytes
+ 2 * 1024 * 1024, data
, 0);
5320 WARN_ON(num_bytes
< root
->sectorsize
);
5321 ret
= find_free_extent(trans
, root
, num_bytes
, empty_size
,
5322 search_start
, search_end
, hint_byte
,
5325 if (ret
== -ENOSPC
&& num_bytes
> min_alloc_size
) {
5326 num_bytes
= num_bytes
>> 1;
5327 num_bytes
= num_bytes
& ~(root
->sectorsize
- 1);
5328 num_bytes
= max(num_bytes
, min_alloc_size
);
5329 do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
5330 num_bytes
, data
, 1);
5333 if (ret
== -ENOSPC
) {
5334 struct btrfs_space_info
*sinfo
;
5336 sinfo
= __find_space_info(root
->fs_info
, data
);
5337 printk(KERN_ERR
"btrfs allocation failed flags %llu, "
5338 "wanted %llu\n", (unsigned long long)data
,
5339 (unsigned long long)num_bytes
);
5340 dump_space_info(sinfo
, num_bytes
, 1);
5346 int btrfs_free_reserved_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
5348 struct btrfs_block_group_cache
*cache
;
5351 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
5353 printk(KERN_ERR
"Unable to find block group for %llu\n",
5354 (unsigned long long)start
);
5358 ret
= btrfs_discard_extent(root
, start
, len
);
5360 btrfs_add_free_space(cache
, start
, len
);
5361 update_reserved_bytes(cache
, len
, 0, 1);
5362 btrfs_put_block_group(cache
);
5367 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
5368 struct btrfs_root
*root
,
5369 u64 parent
, u64 root_objectid
,
5370 u64 flags
, u64 owner
, u64 offset
,
5371 struct btrfs_key
*ins
, int ref_mod
)
5374 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5375 struct btrfs_extent_item
*extent_item
;
5376 struct btrfs_extent_inline_ref
*iref
;
5377 struct btrfs_path
*path
;
5378 struct extent_buffer
*leaf
;
5383 type
= BTRFS_SHARED_DATA_REF_KEY
;
5385 type
= BTRFS_EXTENT_DATA_REF_KEY
;
5387 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
5389 path
= btrfs_alloc_path();
5392 path
->leave_spinning
= 1;
5393 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
5397 leaf
= path
->nodes
[0];
5398 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
5399 struct btrfs_extent_item
);
5400 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
5401 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
5402 btrfs_set_extent_flags(leaf
, extent_item
,
5403 flags
| BTRFS_EXTENT_FLAG_DATA
);
5405 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
5406 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
5408 struct btrfs_shared_data_ref
*ref
;
5409 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
5410 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
5411 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
5413 struct btrfs_extent_data_ref
*ref
;
5414 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
5415 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
5416 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
5417 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
5418 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
5421 btrfs_mark_buffer_dirty(path
->nodes
[0]);
5422 btrfs_free_path(path
);
5424 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1);
5426 printk(KERN_ERR
"btrfs update block group failed for %llu "
5427 "%llu\n", (unsigned long long)ins
->objectid
,
5428 (unsigned long long)ins
->offset
);
5434 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
5435 struct btrfs_root
*root
,
5436 u64 parent
, u64 root_objectid
,
5437 u64 flags
, struct btrfs_disk_key
*key
,
5438 int level
, struct btrfs_key
*ins
)
5441 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5442 struct btrfs_extent_item
*extent_item
;
5443 struct btrfs_tree_block_info
*block_info
;
5444 struct btrfs_extent_inline_ref
*iref
;
5445 struct btrfs_path
*path
;
5446 struct extent_buffer
*leaf
;
5447 u32 size
= sizeof(*extent_item
) + sizeof(*block_info
) + sizeof(*iref
);
5449 path
= btrfs_alloc_path();
5452 path
->leave_spinning
= 1;
5453 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
5457 leaf
= path
->nodes
[0];
5458 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
5459 struct btrfs_extent_item
);
5460 btrfs_set_extent_refs(leaf
, extent_item
, 1);
5461 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
5462 btrfs_set_extent_flags(leaf
, extent_item
,
5463 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
5464 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
5466 btrfs_set_tree_block_key(leaf
, block_info
, key
);
5467 btrfs_set_tree_block_level(leaf
, block_info
, level
);
5469 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
5471 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
5472 btrfs_set_extent_inline_ref_type(leaf
, iref
,
5473 BTRFS_SHARED_BLOCK_REF_KEY
);
5474 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
5476 btrfs_set_extent_inline_ref_type(leaf
, iref
,
5477 BTRFS_TREE_BLOCK_REF_KEY
);
5478 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
5481 btrfs_mark_buffer_dirty(leaf
);
5482 btrfs_free_path(path
);
5484 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1);
5486 printk(KERN_ERR
"btrfs update block group failed for %llu "
5487 "%llu\n", (unsigned long long)ins
->objectid
,
5488 (unsigned long long)ins
->offset
);
5494 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
5495 struct btrfs_root
*root
,
5496 u64 root_objectid
, u64 owner
,
5497 u64 offset
, struct btrfs_key
*ins
)
5501 BUG_ON(root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
5503 ret
= btrfs_add_delayed_data_ref(trans
, ins
->objectid
, ins
->offset
,
5504 0, root_objectid
, owner
, offset
,
5505 BTRFS_ADD_DELAYED_EXTENT
, NULL
);
5510 * this is used by the tree logging recovery code. It records that
5511 * an extent has been allocated and makes sure to clear the free
5512 * space cache bits as well
5514 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
5515 struct btrfs_root
*root
,
5516 u64 root_objectid
, u64 owner
, u64 offset
,
5517 struct btrfs_key
*ins
)
5520 struct btrfs_block_group_cache
*block_group
;
5521 struct btrfs_caching_control
*caching_ctl
;
5522 u64 start
= ins
->objectid
;
5523 u64 num_bytes
= ins
->offset
;
5525 block_group
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
5526 cache_block_group(block_group
, trans
, 0);
5527 caching_ctl
= get_caching_control(block_group
);
5530 BUG_ON(!block_group_cache_done(block_group
));
5531 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
5534 mutex_lock(&caching_ctl
->mutex
);
5536 if (start
>= caching_ctl
->progress
) {
5537 ret
= add_excluded_extent(root
, start
, num_bytes
);
5539 } else if (start
+ num_bytes
<= caching_ctl
->progress
) {
5540 ret
= btrfs_remove_free_space(block_group
,
5544 num_bytes
= caching_ctl
->progress
- start
;
5545 ret
= btrfs_remove_free_space(block_group
,
5549 start
= caching_ctl
->progress
;
5550 num_bytes
= ins
->objectid
+ ins
->offset
-
5551 caching_ctl
->progress
;
5552 ret
= add_excluded_extent(root
, start
, num_bytes
);
5556 mutex_unlock(&caching_ctl
->mutex
);
5557 put_caching_control(caching_ctl
);
5560 ret
= update_reserved_bytes(block_group
, ins
->offset
, 1, 1);
5562 btrfs_put_block_group(block_group
);
5563 ret
= alloc_reserved_file_extent(trans
, root
, 0, root_objectid
,
5564 0, owner
, offset
, ins
, 1);
5568 struct extent_buffer
*btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
,
5569 struct btrfs_root
*root
,
5570 u64 bytenr
, u32 blocksize
,
5573 struct extent_buffer
*buf
;
5575 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
5577 return ERR_PTR(-ENOMEM
);
5578 btrfs_set_header_generation(buf
, trans
->transid
);
5579 btrfs_set_buffer_lockdep_class(buf
, level
);
5580 btrfs_tree_lock(buf
);
5581 clean_tree_block(trans
, root
, buf
);
5583 btrfs_set_lock_blocking(buf
);
5584 btrfs_set_buffer_uptodate(buf
);
5586 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
5588 * we allow two log transactions at a time, use different
5589 * EXENT bit to differentiate dirty pages.
5591 if (root
->log_transid
% 2 == 0)
5592 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
5593 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
5595 set_extent_new(&root
->dirty_log_pages
, buf
->start
,
5596 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
5598 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
5599 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
5601 trans
->blocks_used
++;
5602 /* this returns a buffer locked for blocking */
5606 static struct btrfs_block_rsv
*
5607 use_block_rsv(struct btrfs_trans_handle
*trans
,
5608 struct btrfs_root
*root
, u32 blocksize
)
5610 struct btrfs_block_rsv
*block_rsv
;
5613 block_rsv
= get_block_rsv(trans
, root
);
5615 if (block_rsv
->size
== 0) {
5616 ret
= reserve_metadata_bytes(trans
, root
, block_rsv
,
5619 return ERR_PTR(ret
);
5623 ret
= block_rsv_use_bytes(block_rsv
, blocksize
);
5627 return ERR_PTR(-ENOSPC
);
5630 static void unuse_block_rsv(struct btrfs_block_rsv
*block_rsv
, u32 blocksize
)
5632 block_rsv_add_bytes(block_rsv
, blocksize
, 0);
5633 block_rsv_release_bytes(block_rsv
, NULL
, 0);
5637 * finds a free extent and does all the dirty work required for allocation
5638 * returns the key for the extent through ins, and a tree buffer for
5639 * the first block of the extent through buf.
5641 * returns the tree buffer or NULL.
5643 struct extent_buffer
*btrfs_alloc_free_block(struct btrfs_trans_handle
*trans
,
5644 struct btrfs_root
*root
, u32 blocksize
,
5645 u64 parent
, u64 root_objectid
,
5646 struct btrfs_disk_key
*key
, int level
,
5647 u64 hint
, u64 empty_size
)
5649 struct btrfs_key ins
;
5650 struct btrfs_block_rsv
*block_rsv
;
5651 struct extent_buffer
*buf
;
5656 block_rsv
= use_block_rsv(trans
, root
, blocksize
);
5657 if (IS_ERR(block_rsv
))
5658 return ERR_CAST(block_rsv
);
5660 ret
= btrfs_reserve_extent(trans
, root
, blocksize
, blocksize
,
5661 empty_size
, hint
, (u64
)-1, &ins
, 0);
5663 unuse_block_rsv(block_rsv
, blocksize
);
5664 return ERR_PTR(ret
);
5667 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
,
5669 BUG_ON(IS_ERR(buf
));
5671 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
5673 parent
= ins
.objectid
;
5674 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
5678 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
5679 struct btrfs_delayed_extent_op
*extent_op
;
5680 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
5683 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
5685 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
5686 extent_op
->flags_to_set
= flags
;
5687 extent_op
->update_key
= 1;
5688 extent_op
->update_flags
= 1;
5689 extent_op
->is_data
= 0;
5691 ret
= btrfs_add_delayed_tree_ref(trans
, ins
.objectid
,
5692 ins
.offset
, parent
, root_objectid
,
5693 level
, BTRFS_ADD_DELAYED_EXTENT
,
5700 struct walk_control
{
5701 u64 refs
[BTRFS_MAX_LEVEL
];
5702 u64 flags
[BTRFS_MAX_LEVEL
];
5703 struct btrfs_key update_progress
;
5713 #define DROP_REFERENCE 1
5714 #define UPDATE_BACKREF 2
5716 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
5717 struct btrfs_root
*root
,
5718 struct walk_control
*wc
,
5719 struct btrfs_path
*path
)
5728 struct btrfs_key key
;
5729 struct extent_buffer
*eb
;
5734 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
5735 wc
->reada_count
= wc
->reada_count
* 2 / 3;
5736 wc
->reada_count
= max(wc
->reada_count
, 2);
5738 wc
->reada_count
= wc
->reada_count
* 3 / 2;
5739 wc
->reada_count
= min_t(int, wc
->reada_count
,
5740 BTRFS_NODEPTRS_PER_BLOCK(root
));
5743 eb
= path
->nodes
[wc
->level
];
5744 nritems
= btrfs_header_nritems(eb
);
5745 blocksize
= btrfs_level_size(root
, wc
->level
- 1);
5747 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
5748 if (nread
>= wc
->reada_count
)
5752 bytenr
= btrfs_node_blockptr(eb
, slot
);
5753 generation
= btrfs_node_ptr_generation(eb
, slot
);
5755 if (slot
== path
->slots
[wc
->level
])
5758 if (wc
->stage
== UPDATE_BACKREF
&&
5759 generation
<= root
->root_key
.offset
)
5762 /* We don't lock the tree block, it's OK to be racy here */
5763 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, blocksize
,
5768 if (wc
->stage
== DROP_REFERENCE
) {
5772 if (wc
->level
== 1 &&
5773 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
5775 if (!wc
->update_ref
||
5776 generation
<= root
->root_key
.offset
)
5778 btrfs_node_key_to_cpu(eb
, &key
, slot
);
5779 ret
= btrfs_comp_cpu_keys(&key
,
5780 &wc
->update_progress
);
5784 if (wc
->level
== 1 &&
5785 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
5789 ret
= readahead_tree_block(root
, bytenr
, blocksize
,
5793 last
= bytenr
+ blocksize
;
5796 wc
->reada_slot
= slot
;
5800 * hepler to process tree block while walking down the tree.
5802 * when wc->stage == UPDATE_BACKREF, this function updates
5803 * back refs for pointers in the block.
5805 * NOTE: return value 1 means we should stop walking down.
5807 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
5808 struct btrfs_root
*root
,
5809 struct btrfs_path
*path
,
5810 struct walk_control
*wc
, int lookup_info
)
5812 int level
= wc
->level
;
5813 struct extent_buffer
*eb
= path
->nodes
[level
];
5814 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
5817 if (wc
->stage
== UPDATE_BACKREF
&&
5818 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
5822 * when reference count of tree block is 1, it won't increase
5823 * again. once full backref flag is set, we never clear it.
5826 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
5827 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
5828 BUG_ON(!path
->locks
[level
]);
5829 ret
= btrfs_lookup_extent_info(trans
, root
,
5834 BUG_ON(wc
->refs
[level
] == 0);
5837 if (wc
->stage
== DROP_REFERENCE
) {
5838 if (wc
->refs
[level
] > 1)
5841 if (path
->locks
[level
] && !wc
->keep_locks
) {
5842 btrfs_tree_unlock(eb
);
5843 path
->locks
[level
] = 0;
5848 /* wc->stage == UPDATE_BACKREF */
5849 if (!(wc
->flags
[level
] & flag
)) {
5850 BUG_ON(!path
->locks
[level
]);
5851 ret
= btrfs_inc_ref(trans
, root
, eb
, 1);
5853 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
5855 ret
= btrfs_set_disk_extent_flags(trans
, root
, eb
->start
,
5858 wc
->flags
[level
] |= flag
;
5862 * the block is shared by multiple trees, so it's not good to
5863 * keep the tree lock
5865 if (path
->locks
[level
] && level
> 0) {
5866 btrfs_tree_unlock(eb
);
5867 path
->locks
[level
] = 0;
5873 * hepler to process tree block pointer.
5875 * when wc->stage == DROP_REFERENCE, this function checks
5876 * reference count of the block pointed to. if the block
5877 * is shared and we need update back refs for the subtree
5878 * rooted at the block, this function changes wc->stage to
5879 * UPDATE_BACKREF. if the block is shared and there is no
5880 * need to update back, this function drops the reference
5883 * NOTE: return value 1 means we should stop walking down.
5885 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
5886 struct btrfs_root
*root
,
5887 struct btrfs_path
*path
,
5888 struct walk_control
*wc
, int *lookup_info
)
5894 struct btrfs_key key
;
5895 struct extent_buffer
*next
;
5896 int level
= wc
->level
;
5900 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
5901 path
->slots
[level
]);
5903 * if the lower level block was created before the snapshot
5904 * was created, we know there is no need to update back refs
5907 if (wc
->stage
== UPDATE_BACKREF
&&
5908 generation
<= root
->root_key
.offset
) {
5913 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
5914 blocksize
= btrfs_level_size(root
, level
- 1);
5916 next
= btrfs_find_tree_block(root
, bytenr
, blocksize
);
5918 next
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
5923 btrfs_tree_lock(next
);
5924 btrfs_set_lock_blocking(next
);
5926 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, blocksize
,
5927 &wc
->refs
[level
- 1],
5928 &wc
->flags
[level
- 1]);
5930 BUG_ON(wc
->refs
[level
- 1] == 0);
5933 if (wc
->stage
== DROP_REFERENCE
) {
5934 if (wc
->refs
[level
- 1] > 1) {
5936 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
5939 if (!wc
->update_ref
||
5940 generation
<= root
->root_key
.offset
)
5943 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
5944 path
->slots
[level
]);
5945 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
5949 wc
->stage
= UPDATE_BACKREF
;
5950 wc
->shared_level
= level
- 1;
5954 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
5958 if (!btrfs_buffer_uptodate(next
, generation
)) {
5959 btrfs_tree_unlock(next
);
5960 free_extent_buffer(next
);
5966 if (reada
&& level
== 1)
5967 reada_walk_down(trans
, root
, wc
, path
);
5968 next
= read_tree_block(root
, bytenr
, blocksize
, generation
);
5969 btrfs_tree_lock(next
);
5970 btrfs_set_lock_blocking(next
);
5974 BUG_ON(level
!= btrfs_header_level(next
));
5975 path
->nodes
[level
] = next
;
5976 path
->slots
[level
] = 0;
5977 path
->locks
[level
] = 1;
5983 wc
->refs
[level
- 1] = 0;
5984 wc
->flags
[level
- 1] = 0;
5985 if (wc
->stage
== DROP_REFERENCE
) {
5986 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
5987 parent
= path
->nodes
[level
]->start
;
5989 BUG_ON(root
->root_key
.objectid
!=
5990 btrfs_header_owner(path
->nodes
[level
]));
5994 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
, parent
,
5995 root
->root_key
.objectid
, level
- 1, 0);
5998 btrfs_tree_unlock(next
);
5999 free_extent_buffer(next
);
6005 * hepler to process tree block while walking up the tree.
6007 * when wc->stage == DROP_REFERENCE, this function drops
6008 * reference count on the block.
6010 * when wc->stage == UPDATE_BACKREF, this function changes
6011 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6012 * to UPDATE_BACKREF previously while processing the block.
6014 * NOTE: return value 1 means we should stop walking up.
6016 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
6017 struct btrfs_root
*root
,
6018 struct btrfs_path
*path
,
6019 struct walk_control
*wc
)
6022 int level
= wc
->level
;
6023 struct extent_buffer
*eb
= path
->nodes
[level
];
6026 if (wc
->stage
== UPDATE_BACKREF
) {
6027 BUG_ON(wc
->shared_level
< level
);
6028 if (level
< wc
->shared_level
)
6031 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
6035 wc
->stage
= DROP_REFERENCE
;
6036 wc
->shared_level
= -1;
6037 path
->slots
[level
] = 0;
6040 * check reference count again if the block isn't locked.
6041 * we should start walking down the tree again if reference
6044 if (!path
->locks
[level
]) {
6046 btrfs_tree_lock(eb
);
6047 btrfs_set_lock_blocking(eb
);
6048 path
->locks
[level
] = 1;
6050 ret
= btrfs_lookup_extent_info(trans
, root
,
6055 BUG_ON(wc
->refs
[level
] == 0);
6056 if (wc
->refs
[level
] == 1) {
6057 btrfs_tree_unlock(eb
);
6058 path
->locks
[level
] = 0;
6064 /* wc->stage == DROP_REFERENCE */
6065 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
6067 if (wc
->refs
[level
] == 1) {
6069 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
6070 ret
= btrfs_dec_ref(trans
, root
, eb
, 1);
6072 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
6075 /* make block locked assertion in clean_tree_block happy */
6076 if (!path
->locks
[level
] &&
6077 btrfs_header_generation(eb
) == trans
->transid
) {
6078 btrfs_tree_lock(eb
);
6079 btrfs_set_lock_blocking(eb
);
6080 path
->locks
[level
] = 1;
6082 clean_tree_block(trans
, root
, eb
);
6085 if (eb
== root
->node
) {
6086 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
6089 BUG_ON(root
->root_key
.objectid
!=
6090 btrfs_header_owner(eb
));
6092 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
6093 parent
= path
->nodes
[level
+ 1]->start
;
6095 BUG_ON(root
->root_key
.objectid
!=
6096 btrfs_header_owner(path
->nodes
[level
+ 1]));
6099 btrfs_free_tree_block(trans
, root
, eb
, parent
, wc
->refs
[level
] == 1);
6101 wc
->refs
[level
] = 0;
6102 wc
->flags
[level
] = 0;
6106 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
6107 struct btrfs_root
*root
,
6108 struct btrfs_path
*path
,
6109 struct walk_control
*wc
)
6111 int level
= wc
->level
;
6112 int lookup_info
= 1;
6115 while (level
>= 0) {
6116 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
6123 if (path
->slots
[level
] >=
6124 btrfs_header_nritems(path
->nodes
[level
]))
6127 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
6129 path
->slots
[level
]++;
6138 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
6139 struct btrfs_root
*root
,
6140 struct btrfs_path
*path
,
6141 struct walk_control
*wc
, int max_level
)
6143 int level
= wc
->level
;
6146 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
6147 while (level
< max_level
&& path
->nodes
[level
]) {
6149 if (path
->slots
[level
] + 1 <
6150 btrfs_header_nritems(path
->nodes
[level
])) {
6151 path
->slots
[level
]++;
6154 ret
= walk_up_proc(trans
, root
, path
, wc
);
6158 if (path
->locks
[level
]) {
6159 btrfs_tree_unlock(path
->nodes
[level
]);
6160 path
->locks
[level
] = 0;
6162 free_extent_buffer(path
->nodes
[level
]);
6163 path
->nodes
[level
] = NULL
;
6171 * drop a subvolume tree.
6173 * this function traverses the tree freeing any blocks that only
6174 * referenced by the tree.
6176 * when a shared tree block is found. this function decreases its
6177 * reference count by one. if update_ref is true, this function
6178 * also make sure backrefs for the shared block and all lower level
6179 * blocks are properly updated.
6181 int btrfs_drop_snapshot(struct btrfs_root
*root
,
6182 struct btrfs_block_rsv
*block_rsv
, int update_ref
)
6184 struct btrfs_path
*path
;
6185 struct btrfs_trans_handle
*trans
;
6186 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
6187 struct btrfs_root_item
*root_item
= &root
->root_item
;
6188 struct walk_control
*wc
;
6189 struct btrfs_key key
;
6194 path
= btrfs_alloc_path();
6197 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
6200 trans
= btrfs_start_transaction(tree_root
, 0);
6202 trans
->block_rsv
= block_rsv
;
6204 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
6205 level
= btrfs_header_level(root
->node
);
6206 path
->nodes
[level
] = btrfs_lock_root_node(root
);
6207 btrfs_set_lock_blocking(path
->nodes
[level
]);
6208 path
->slots
[level
] = 0;
6209 path
->locks
[level
] = 1;
6210 memset(&wc
->update_progress
, 0,
6211 sizeof(wc
->update_progress
));
6213 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
6214 memcpy(&wc
->update_progress
, &key
,
6215 sizeof(wc
->update_progress
));
6217 level
= root_item
->drop_level
;
6219 path
->lowest_level
= level
;
6220 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6221 path
->lowest_level
= 0;
6229 * unlock our path, this is safe because only this
6230 * function is allowed to delete this snapshot
6232 btrfs_unlock_up_safe(path
, 0);
6234 level
= btrfs_header_level(root
->node
);
6236 btrfs_tree_lock(path
->nodes
[level
]);
6237 btrfs_set_lock_blocking(path
->nodes
[level
]);
6239 ret
= btrfs_lookup_extent_info(trans
, root
,
6240 path
->nodes
[level
]->start
,
6241 path
->nodes
[level
]->len
,
6245 BUG_ON(wc
->refs
[level
] == 0);
6247 if (level
== root_item
->drop_level
)
6250 btrfs_tree_unlock(path
->nodes
[level
]);
6251 WARN_ON(wc
->refs
[level
] != 1);
6257 wc
->shared_level
= -1;
6258 wc
->stage
= DROP_REFERENCE
;
6259 wc
->update_ref
= update_ref
;
6261 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
6264 ret
= walk_down_tree(trans
, root
, path
, wc
);
6270 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
6277 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
6281 if (wc
->stage
== DROP_REFERENCE
) {
6283 btrfs_node_key(path
->nodes
[level
],
6284 &root_item
->drop_progress
,
6285 path
->slots
[level
]);
6286 root_item
->drop_level
= level
;
6289 BUG_ON(wc
->level
== 0);
6290 if (btrfs_should_end_transaction(trans
, tree_root
)) {
6291 ret
= btrfs_update_root(trans
, tree_root
,
6296 btrfs_end_transaction_throttle(trans
, tree_root
);
6297 trans
= btrfs_start_transaction(tree_root
, 0);
6299 trans
->block_rsv
= block_rsv
;
6302 btrfs_release_path(root
, path
);
6305 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
6308 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
6309 ret
= btrfs_find_last_root(tree_root
, root
->root_key
.objectid
,
6313 ret
= btrfs_del_orphan_item(trans
, tree_root
,
6314 root
->root_key
.objectid
);
6319 if (root
->in_radix
) {
6320 btrfs_free_fs_root(tree_root
->fs_info
, root
);
6322 free_extent_buffer(root
->node
);
6323 free_extent_buffer(root
->commit_root
);
6327 btrfs_end_transaction_throttle(trans
, tree_root
);
6329 btrfs_free_path(path
);
6334 * drop subtree rooted at tree block 'node'.
6336 * NOTE: this function will unlock and release tree block 'node'
6338 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
6339 struct btrfs_root
*root
,
6340 struct extent_buffer
*node
,
6341 struct extent_buffer
*parent
)
6343 struct btrfs_path
*path
;
6344 struct walk_control
*wc
;
6350 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
6352 path
= btrfs_alloc_path();
6355 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
6358 btrfs_assert_tree_locked(parent
);
6359 parent_level
= btrfs_header_level(parent
);
6360 extent_buffer_get(parent
);
6361 path
->nodes
[parent_level
] = parent
;
6362 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
6364 btrfs_assert_tree_locked(node
);
6365 level
= btrfs_header_level(node
);
6366 path
->nodes
[level
] = node
;
6367 path
->slots
[level
] = 0;
6368 path
->locks
[level
] = 1;
6370 wc
->refs
[parent_level
] = 1;
6371 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
6373 wc
->shared_level
= -1;
6374 wc
->stage
= DROP_REFERENCE
;
6377 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
6380 wret
= walk_down_tree(trans
, root
, path
, wc
);
6386 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
6394 btrfs_free_path(path
);
6399 static unsigned long calc_ra(unsigned long start
, unsigned long last
,
6402 return min(last
, start
+ nr
- 1);
6405 static noinline
int relocate_inode_pages(struct inode
*inode
, u64 start
,
6410 unsigned long first_index
;
6411 unsigned long last_index
;
6414 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
6415 struct file_ra_state
*ra
;
6416 struct btrfs_ordered_extent
*ordered
;
6417 unsigned int total_read
= 0;
6418 unsigned int total_dirty
= 0;
6421 ra
= kzalloc(sizeof(*ra
), GFP_NOFS
);
6423 mutex_lock(&inode
->i_mutex
);
6424 first_index
= start
>> PAGE_CACHE_SHIFT
;
6425 last_index
= (start
+ len
- 1) >> PAGE_CACHE_SHIFT
;
6427 /* make sure the dirty trick played by the caller work */
6428 ret
= invalidate_inode_pages2_range(inode
->i_mapping
,
6429 first_index
, last_index
);
6433 file_ra_state_init(ra
, inode
->i_mapping
);
6435 for (i
= first_index
; i
<= last_index
; i
++) {
6436 if (total_read
% ra
->ra_pages
== 0) {
6437 btrfs_force_ra(inode
->i_mapping
, ra
, NULL
, i
,
6438 calc_ra(i
, last_index
, ra
->ra_pages
));
6442 if (((u64
)i
<< PAGE_CACHE_SHIFT
) > i_size_read(inode
))
6444 page
= grab_cache_page(inode
->i_mapping
, i
);
6449 if (!PageUptodate(page
)) {
6450 btrfs_readpage(NULL
, page
);
6452 if (!PageUptodate(page
)) {
6454 page_cache_release(page
);
6459 wait_on_page_writeback(page
);
6461 page_start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
6462 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
6463 lock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
6465 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
6467 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
6469 page_cache_release(page
);
6470 btrfs_start_ordered_extent(inode
, ordered
, 1);
6471 btrfs_put_ordered_extent(ordered
);
6474 set_page_extent_mapped(page
);
6476 if (i
== first_index
)
6477 set_extent_bits(io_tree
, page_start
, page_end
,
6478 EXTENT_BOUNDARY
, GFP_NOFS
);
6479 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
6481 set_page_dirty(page
);
6484 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
6486 page_cache_release(page
);
6491 mutex_unlock(&inode
->i_mutex
);
6492 balance_dirty_pages_ratelimited_nr(inode
->i_mapping
, total_dirty
);
6496 static noinline
int relocate_data_extent(struct inode
*reloc_inode
,
6497 struct btrfs_key
*extent_key
,
6500 struct btrfs_root
*root
= BTRFS_I(reloc_inode
)->root
;
6501 struct extent_map_tree
*em_tree
= &BTRFS_I(reloc_inode
)->extent_tree
;
6502 struct extent_map
*em
;
6503 u64 start
= extent_key
->objectid
- offset
;
6504 u64 end
= start
+ extent_key
->offset
- 1;
6506 em
= alloc_extent_map(GFP_NOFS
);
6507 BUG_ON(!em
|| IS_ERR(em
));
6510 em
->len
= extent_key
->offset
;
6511 em
->block_len
= extent_key
->offset
;
6512 em
->block_start
= extent_key
->objectid
;
6513 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
6514 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
6516 /* setup extent map to cheat btrfs_readpage */
6517 lock_extent(&BTRFS_I(reloc_inode
)->io_tree
, start
, end
, GFP_NOFS
);
6520 write_lock(&em_tree
->lock
);
6521 ret
= add_extent_mapping(em_tree
, em
);
6522 write_unlock(&em_tree
->lock
);
6523 if (ret
!= -EEXIST
) {
6524 free_extent_map(em
);
6527 btrfs_drop_extent_cache(reloc_inode
, start
, end
, 0);
6529 unlock_extent(&BTRFS_I(reloc_inode
)->io_tree
, start
, end
, GFP_NOFS
);
6531 return relocate_inode_pages(reloc_inode
, start
, extent_key
->offset
);
6534 struct btrfs_ref_path
{
6536 u64 nodes
[BTRFS_MAX_LEVEL
];
6538 u64 root_generation
;
6545 struct btrfs_key node_keys
[BTRFS_MAX_LEVEL
];
6546 u64 new_nodes
[BTRFS_MAX_LEVEL
];
6549 struct disk_extent
{
6560 static int is_cowonly_root(u64 root_objectid
)
6562 if (root_objectid
== BTRFS_ROOT_TREE_OBJECTID
||
6563 root_objectid
== BTRFS_EXTENT_TREE_OBJECTID
||
6564 root_objectid
== BTRFS_CHUNK_TREE_OBJECTID
||
6565 root_objectid
== BTRFS_DEV_TREE_OBJECTID
||
6566 root_objectid
== BTRFS_TREE_LOG_OBJECTID
||
6567 root_objectid
== BTRFS_CSUM_TREE_OBJECTID
)
6572 static noinline
int __next_ref_path(struct btrfs_trans_handle
*trans
,
6573 struct btrfs_root
*extent_root
,
6574 struct btrfs_ref_path
*ref_path
,
6577 struct extent_buffer
*leaf
;
6578 struct btrfs_path
*path
;
6579 struct btrfs_extent_ref
*ref
;
6580 struct btrfs_key key
;
6581 struct btrfs_key found_key
;
6587 path
= btrfs_alloc_path();
6592 ref_path
->lowest_level
= -1;
6593 ref_path
->current_level
= -1;
6594 ref_path
->shared_level
= -1;
6598 level
= ref_path
->current_level
- 1;
6599 while (level
>= -1) {
6601 if (level
< ref_path
->lowest_level
)
6605 bytenr
= ref_path
->nodes
[level
];
6607 bytenr
= ref_path
->extent_start
;
6608 BUG_ON(bytenr
== 0);
6610 parent
= ref_path
->nodes
[level
+ 1];
6611 ref_path
->nodes
[level
+ 1] = 0;
6612 ref_path
->current_level
= level
;
6613 BUG_ON(parent
== 0);
6615 key
.objectid
= bytenr
;
6616 key
.offset
= parent
+ 1;
6617 key
.type
= BTRFS_EXTENT_REF_KEY
;
6619 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
, 0, 0);
6624 leaf
= path
->nodes
[0];
6625 nritems
= btrfs_header_nritems(leaf
);
6626 if (path
->slots
[0] >= nritems
) {
6627 ret
= btrfs_next_leaf(extent_root
, path
);
6632 leaf
= path
->nodes
[0];
6635 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6636 if (found_key
.objectid
== bytenr
&&
6637 found_key
.type
== BTRFS_EXTENT_REF_KEY
) {
6638 if (level
< ref_path
->shared_level
)
6639 ref_path
->shared_level
= level
;
6644 btrfs_release_path(extent_root
, path
);
6647 /* reached lowest level */
6651 level
= ref_path
->current_level
;
6652 while (level
< BTRFS_MAX_LEVEL
- 1) {
6656 bytenr
= ref_path
->nodes
[level
];
6658 bytenr
= ref_path
->extent_start
;
6660 BUG_ON(bytenr
== 0);
6662 key
.objectid
= bytenr
;
6664 key
.type
= BTRFS_EXTENT_REF_KEY
;
6666 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
, 0, 0);
6670 leaf
= path
->nodes
[0];
6671 nritems
= btrfs_header_nritems(leaf
);
6672 if (path
->slots
[0] >= nritems
) {
6673 ret
= btrfs_next_leaf(extent_root
, path
);
6677 /* the extent was freed by someone */
6678 if (ref_path
->lowest_level
== level
)
6680 btrfs_release_path(extent_root
, path
);
6683 leaf
= path
->nodes
[0];
6686 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6687 if (found_key
.objectid
!= bytenr
||
6688 found_key
.type
!= BTRFS_EXTENT_REF_KEY
) {
6689 /* the extent was freed by someone */
6690 if (ref_path
->lowest_level
== level
) {
6694 btrfs_release_path(extent_root
, path
);
6698 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
6699 struct btrfs_extent_ref
);
6700 ref_objectid
= btrfs_ref_objectid(leaf
, ref
);
6701 if (ref_objectid
< BTRFS_FIRST_FREE_OBJECTID
) {
6703 level
= (int)ref_objectid
;
6704 BUG_ON(level
>= BTRFS_MAX_LEVEL
);
6705 ref_path
->lowest_level
= level
;
6706 ref_path
->current_level
= level
;
6707 ref_path
->nodes
[level
] = bytenr
;
6709 WARN_ON(ref_objectid
!= level
);
6712 WARN_ON(level
!= -1);
6716 if (ref_path
->lowest_level
== level
) {
6717 ref_path
->owner_objectid
= ref_objectid
;
6718 ref_path
->num_refs
= btrfs_ref_num_refs(leaf
, ref
);
6722 * the block is tree root or the block isn't in reference
6725 if (found_key
.objectid
== found_key
.offset
||
6726 is_cowonly_root(btrfs_ref_root(leaf
, ref
))) {
6727 ref_path
->root_objectid
= btrfs_ref_root(leaf
, ref
);
6728 ref_path
->root_generation
=
6729 btrfs_ref_generation(leaf
, ref
);
6731 /* special reference from the tree log */
6732 ref_path
->nodes
[0] = found_key
.offset
;
6733 ref_path
->current_level
= 0;
6740 BUG_ON(ref_path
->nodes
[level
] != 0);
6741 ref_path
->nodes
[level
] = found_key
.offset
;
6742 ref_path
->current_level
= level
;
6745 * the reference was created in the running transaction,
6746 * no need to continue walking up.
6748 if (btrfs_ref_generation(leaf
, ref
) == trans
->transid
) {
6749 ref_path
->root_objectid
= btrfs_ref_root(leaf
, ref
);
6750 ref_path
->root_generation
=
6751 btrfs_ref_generation(leaf
, ref
);
6756 btrfs_release_path(extent_root
, path
);
6759 /* reached max tree level, but no tree root found. */
6762 btrfs_free_path(path
);
6766 static int btrfs_first_ref_path(struct btrfs_trans_handle
*trans
,
6767 struct btrfs_root
*extent_root
,
6768 struct btrfs_ref_path
*ref_path
,
6771 memset(ref_path
, 0, sizeof(*ref_path
));
6772 ref_path
->extent_start
= extent_start
;
6774 return __next_ref_path(trans
, extent_root
, ref_path
, 1);
6777 static int btrfs_next_ref_path(struct btrfs_trans_handle
*trans
,
6778 struct btrfs_root
*extent_root
,
6779 struct btrfs_ref_path
*ref_path
)
6781 return __next_ref_path(trans
, extent_root
, ref_path
, 0);
6784 static noinline
int get_new_locations(struct inode
*reloc_inode
,
6785 struct btrfs_key
*extent_key
,
6786 u64 offset
, int no_fragment
,
6787 struct disk_extent
**extents
,
6790 struct btrfs_root
*root
= BTRFS_I(reloc_inode
)->root
;
6791 struct btrfs_path
*path
;
6792 struct btrfs_file_extent_item
*fi
;
6793 struct extent_buffer
*leaf
;
6794 struct disk_extent
*exts
= *extents
;
6795 struct btrfs_key found_key
;
6800 int max
= *nr_extents
;
6803 WARN_ON(!no_fragment
&& *extents
);
6806 exts
= kmalloc(sizeof(*exts
) * max
, GFP_NOFS
);
6811 path
= btrfs_alloc_path();
6814 cur_pos
= extent_key
->objectid
- offset
;
6815 last_byte
= extent_key
->objectid
+ extent_key
->offset
;
6816 ret
= btrfs_lookup_file_extent(NULL
, root
, path
, reloc_inode
->i_ino
,
6826 leaf
= path
->nodes
[0];
6827 nritems
= btrfs_header_nritems(leaf
);
6828 if (path
->slots
[0] >= nritems
) {
6829 ret
= btrfs_next_leaf(root
, path
);
6834 leaf
= path
->nodes
[0];
6837 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6838 if (found_key
.offset
!= cur_pos
||
6839 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
||
6840 found_key
.objectid
!= reloc_inode
->i_ino
)
6843 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
6844 struct btrfs_file_extent_item
);
6845 if (btrfs_file_extent_type(leaf
, fi
) !=
6846 BTRFS_FILE_EXTENT_REG
||
6847 btrfs_file_extent_disk_bytenr(leaf
, fi
) == 0)
6851 struct disk_extent
*old
= exts
;
6853 exts
= kzalloc(sizeof(*exts
) * max
, GFP_NOFS
);
6854 memcpy(exts
, old
, sizeof(*exts
) * nr
);
6855 if (old
!= *extents
)
6859 exts
[nr
].disk_bytenr
=
6860 btrfs_file_extent_disk_bytenr(leaf
, fi
);
6861 exts
[nr
].disk_num_bytes
=
6862 btrfs_file_extent_disk_num_bytes(leaf
, fi
);
6863 exts
[nr
].offset
= btrfs_file_extent_offset(leaf
, fi
);
6864 exts
[nr
].num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
6865 exts
[nr
].ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
6866 exts
[nr
].compression
= btrfs_file_extent_compression(leaf
, fi
);
6867 exts
[nr
].encryption
= btrfs_file_extent_encryption(leaf
, fi
);
6868 exts
[nr
].other_encoding
= btrfs_file_extent_other_encoding(leaf
,
6870 BUG_ON(exts
[nr
].offset
> 0);
6871 BUG_ON(exts
[nr
].compression
|| exts
[nr
].encryption
);
6872 BUG_ON(exts
[nr
].num_bytes
!= exts
[nr
].disk_num_bytes
);
6874 cur_pos
+= exts
[nr
].num_bytes
;
6877 if (cur_pos
+ offset
>= last_byte
)
6887 BUG_ON(cur_pos
+ offset
> last_byte
);
6888 if (cur_pos
+ offset
< last_byte
) {
6894 btrfs_free_path(path
);
6896 if (exts
!= *extents
)
6905 static noinline
int replace_one_extent(struct btrfs_trans_handle
*trans
,
6906 struct btrfs_root
*root
,
6907 struct btrfs_path
*path
,
6908 struct btrfs_key
*extent_key
,
6909 struct btrfs_key
*leaf_key
,
6910 struct btrfs_ref_path
*ref_path
,
6911 struct disk_extent
*new_extents
,
6914 struct extent_buffer
*leaf
;
6915 struct btrfs_file_extent_item
*fi
;
6916 struct inode
*inode
= NULL
;
6917 struct btrfs_key key
;
6922 u64 search_end
= (u64
)-1;
6925 int extent_locked
= 0;
6929 memcpy(&key
, leaf_key
, sizeof(key
));
6930 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
) {
6931 if (key
.objectid
< ref_path
->owner_objectid
||
6932 (key
.objectid
== ref_path
->owner_objectid
&&
6933 key
.type
< BTRFS_EXTENT_DATA_KEY
)) {
6934 key
.objectid
= ref_path
->owner_objectid
;
6935 key
.type
= BTRFS_EXTENT_DATA_KEY
;
6941 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
6945 leaf
= path
->nodes
[0];
6946 nritems
= btrfs_header_nritems(leaf
);
6948 if (extent_locked
&& ret
> 0) {
6950 * the file extent item was modified by someone
6951 * before the extent got locked.
6953 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
6954 lock_end
, GFP_NOFS
);
6958 if (path
->slots
[0] >= nritems
) {
6959 if (++nr_scaned
> 2)
6962 BUG_ON(extent_locked
);
6963 ret
= btrfs_next_leaf(root
, path
);
6968 leaf
= path
->nodes
[0];
6969 nritems
= btrfs_header_nritems(leaf
);
6972 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
6974 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
) {
6975 if ((key
.objectid
> ref_path
->owner_objectid
) ||
6976 (key
.objectid
== ref_path
->owner_objectid
&&
6977 key
.type
> BTRFS_EXTENT_DATA_KEY
) ||
6978 key
.offset
>= search_end
)
6982 if (inode
&& key
.objectid
!= inode
->i_ino
) {
6983 BUG_ON(extent_locked
);
6984 btrfs_release_path(root
, path
);
6985 mutex_unlock(&inode
->i_mutex
);
6991 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
6996 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
6997 struct btrfs_file_extent_item
);
6998 extent_type
= btrfs_file_extent_type(leaf
, fi
);
6999 if ((extent_type
!= BTRFS_FILE_EXTENT_REG
&&
7000 extent_type
!= BTRFS_FILE_EXTENT_PREALLOC
) ||
7001 (btrfs_file_extent_disk_bytenr(leaf
, fi
) !=
7002 extent_key
->objectid
)) {
7008 num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
7009 ext_offset
= btrfs_file_extent_offset(leaf
, fi
);
7011 if (search_end
== (u64
)-1) {
7012 search_end
= key
.offset
- ext_offset
+
7013 btrfs_file_extent_ram_bytes(leaf
, fi
);
7016 if (!extent_locked
) {
7017 lock_start
= key
.offset
;
7018 lock_end
= lock_start
+ num_bytes
- 1;
7020 if (lock_start
> key
.offset
||
7021 lock_end
+ 1 < key
.offset
+ num_bytes
) {
7022 unlock_extent(&BTRFS_I(inode
)->io_tree
,
7023 lock_start
, lock_end
, GFP_NOFS
);
7029 btrfs_release_path(root
, path
);
7031 inode
= btrfs_iget_locked(root
->fs_info
->sb
,
7032 key
.objectid
, root
);
7033 if (inode
->i_state
& I_NEW
) {
7034 BTRFS_I(inode
)->root
= root
;
7035 BTRFS_I(inode
)->location
.objectid
=
7037 BTRFS_I(inode
)->location
.type
=
7038 BTRFS_INODE_ITEM_KEY
;
7039 BTRFS_I(inode
)->location
.offset
= 0;
7040 btrfs_read_locked_inode(inode
);
7041 unlock_new_inode(inode
);
7044 * some code call btrfs_commit_transaction while
7045 * holding the i_mutex, so we can't use mutex_lock
7048 if (is_bad_inode(inode
) ||
7049 !mutex_trylock(&inode
->i_mutex
)) {
7052 key
.offset
= (u64
)-1;
7057 if (!extent_locked
) {
7058 struct btrfs_ordered_extent
*ordered
;
7060 btrfs_release_path(root
, path
);
7062 lock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
7063 lock_end
, GFP_NOFS
);
7064 ordered
= btrfs_lookup_first_ordered_extent(inode
,
7067 ordered
->file_offset
<= lock_end
&&
7068 ordered
->file_offset
+ ordered
->len
> lock_start
) {
7069 unlock_extent(&BTRFS_I(inode
)->io_tree
,
7070 lock_start
, lock_end
, GFP_NOFS
);
7071 btrfs_start_ordered_extent(inode
, ordered
, 1);
7072 btrfs_put_ordered_extent(ordered
);
7073 key
.offset
+= num_bytes
;
7077 btrfs_put_ordered_extent(ordered
);
7083 if (nr_extents
== 1) {
7084 /* update extent pointer in place */
7085 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
7086 new_extents
[0].disk_bytenr
);
7087 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
7088 new_extents
[0].disk_num_bytes
);
7089 btrfs_mark_buffer_dirty(leaf
);
7091 btrfs_drop_extent_cache(inode
, key
.offset
,
7092 key
.offset
+ num_bytes
- 1, 0);
7094 ret
= btrfs_inc_extent_ref(trans
, root
,
7095 new_extents
[0].disk_bytenr
,
7096 new_extents
[0].disk_num_bytes
,
7098 root
->root_key
.objectid
,
7103 ret
= btrfs_free_extent(trans
, root
,
7104 extent_key
->objectid
,
7107 btrfs_header_owner(leaf
),
7108 btrfs_header_generation(leaf
),
7112 btrfs_release_path(root
, path
);
7113 key
.offset
+= num_bytes
;
7121 * drop old extent pointer at first, then insert the
7122 * new pointers one bye one
7124 btrfs_release_path(root
, path
);
7125 ret
= btrfs_drop_extents(trans
, root
, inode
, key
.offset
,
7126 key
.offset
+ num_bytes
,
7127 key
.offset
, &alloc_hint
);
7130 for (i
= 0; i
< nr_extents
; i
++) {
7131 if (ext_offset
>= new_extents
[i
].num_bytes
) {
7132 ext_offset
-= new_extents
[i
].num_bytes
;
7135 extent_len
= min(new_extents
[i
].num_bytes
-
7136 ext_offset
, num_bytes
);
7138 ret
= btrfs_insert_empty_item(trans
, root
,
7143 leaf
= path
->nodes
[0];
7144 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
7145 struct btrfs_file_extent_item
);
7146 btrfs_set_file_extent_generation(leaf
, fi
,
7148 btrfs_set_file_extent_type(leaf
, fi
,
7149 BTRFS_FILE_EXTENT_REG
);
7150 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
7151 new_extents
[i
].disk_bytenr
);
7152 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
7153 new_extents
[i
].disk_num_bytes
);
7154 btrfs_set_file_extent_ram_bytes(leaf
, fi
,
7155 new_extents
[i
].ram_bytes
);
7157 btrfs_set_file_extent_compression(leaf
, fi
,
7158 new_extents
[i
].compression
);
7159 btrfs_set_file_extent_encryption(leaf
, fi
,
7160 new_extents
[i
].encryption
);
7161 btrfs_set_file_extent_other_encoding(leaf
, fi
,
7162 new_extents
[i
].other_encoding
);
7164 btrfs_set_file_extent_num_bytes(leaf
, fi
,
7166 ext_offset
+= new_extents
[i
].offset
;
7167 btrfs_set_file_extent_offset(leaf
, fi
,
7169 btrfs_mark_buffer_dirty(leaf
);
7171 btrfs_drop_extent_cache(inode
, key
.offset
,
7172 key
.offset
+ extent_len
- 1, 0);
7174 ret
= btrfs_inc_extent_ref(trans
, root
,
7175 new_extents
[i
].disk_bytenr
,
7176 new_extents
[i
].disk_num_bytes
,
7178 root
->root_key
.objectid
,
7179 trans
->transid
, key
.objectid
);
7181 btrfs_release_path(root
, path
);
7183 inode_add_bytes(inode
, extent_len
);
7186 num_bytes
-= extent_len
;
7187 key
.offset
+= extent_len
;
7192 BUG_ON(i
>= nr_extents
);
7196 if (extent_locked
) {
7197 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
7198 lock_end
, GFP_NOFS
);
7202 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
&&
7203 key
.offset
>= search_end
)
7210 btrfs_release_path(root
, path
);
7212 mutex_unlock(&inode
->i_mutex
);
7213 if (extent_locked
) {
7214 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
7215 lock_end
, GFP_NOFS
);
7222 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle
*trans
,
7223 struct btrfs_root
*root
,
7224 struct extent_buffer
*buf
, u64 orig_start
)
7229 BUG_ON(btrfs_header_generation(buf
) != trans
->transid
);
7230 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
7232 level
= btrfs_header_level(buf
);
7234 struct btrfs_leaf_ref
*ref
;
7235 struct btrfs_leaf_ref
*orig_ref
;
7237 orig_ref
= btrfs_lookup_leaf_ref(root
, orig_start
);
7241 ref
= btrfs_alloc_leaf_ref(root
, orig_ref
->nritems
);
7243 btrfs_free_leaf_ref(root
, orig_ref
);
7247 ref
->nritems
= orig_ref
->nritems
;
7248 memcpy(ref
->extents
, orig_ref
->extents
,
7249 sizeof(ref
->extents
[0]) * ref
->nritems
);
7251 btrfs_free_leaf_ref(root
, orig_ref
);
7253 ref
->root_gen
= trans
->transid
;
7254 ref
->bytenr
= buf
->start
;
7255 ref
->owner
= btrfs_header_owner(buf
);
7256 ref
->generation
= btrfs_header_generation(buf
);
7258 ret
= btrfs_add_leaf_ref(root
, ref
, 0);
7260 btrfs_free_leaf_ref(root
, ref
);
7265 static noinline
int invalidate_extent_cache(struct btrfs_root
*root
,
7266 struct extent_buffer
*leaf
,
7267 struct btrfs_block_group_cache
*group
,
7268 struct btrfs_root
*target_root
)
7270 struct btrfs_key key
;
7271 struct inode
*inode
= NULL
;
7272 struct btrfs_file_extent_item
*fi
;
7273 struct extent_state
*cached_state
= NULL
;
7275 u64 skip_objectid
= 0;
7279 nritems
= btrfs_header_nritems(leaf
);
7280 for (i
= 0; i
< nritems
; i
++) {
7281 btrfs_item_key_to_cpu(leaf
, &key
, i
);
7282 if (key
.objectid
== skip_objectid
||
7283 key
.type
!= BTRFS_EXTENT_DATA_KEY
)
7285 fi
= btrfs_item_ptr(leaf
, i
, struct btrfs_file_extent_item
);
7286 if (btrfs_file_extent_type(leaf
, fi
) ==
7287 BTRFS_FILE_EXTENT_INLINE
)
7289 if (btrfs_file_extent_disk_bytenr(leaf
, fi
) == 0)
7291 if (!inode
|| inode
->i_ino
!= key
.objectid
) {
7293 inode
= btrfs_ilookup(target_root
->fs_info
->sb
,
7294 key
.objectid
, target_root
, 1);
7297 skip_objectid
= key
.objectid
;
7300 num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
7302 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, key
.offset
,
7303 key
.offset
+ num_bytes
- 1, 0, &cached_state
,
7305 btrfs_drop_extent_cache(inode
, key
.offset
,
7306 key
.offset
+ num_bytes
- 1, 1);
7307 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, key
.offset
,
7308 key
.offset
+ num_bytes
- 1, &cached_state
,
7316 static noinline
int replace_extents_in_leaf(struct btrfs_trans_handle
*trans
,
7317 struct btrfs_root
*root
,
7318 struct extent_buffer
*leaf
,
7319 struct btrfs_block_group_cache
*group
,
7320 struct inode
*reloc_inode
)
7322 struct btrfs_key key
;
7323 struct btrfs_key extent_key
;
7324 struct btrfs_file_extent_item
*fi
;
7325 struct btrfs_leaf_ref
*ref
;
7326 struct disk_extent
*new_extent
;
7335 new_extent
= kmalloc(sizeof(*new_extent
), GFP_NOFS
);
7336 BUG_ON(!new_extent
);
7338 ref
= btrfs_lookup_leaf_ref(root
, leaf
->start
);
7342 nritems
= btrfs_header_nritems(leaf
);
7343 for (i
= 0; i
< nritems
; i
++) {
7344 btrfs_item_key_to_cpu(leaf
, &key
, i
);
7345 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
7347 fi
= btrfs_item_ptr(leaf
, i
, struct btrfs_file_extent_item
);
7348 if (btrfs_file_extent_type(leaf
, fi
) ==
7349 BTRFS_FILE_EXTENT_INLINE
)
7351 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
7352 num_bytes
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
7357 if (bytenr
>= group
->key
.objectid
+ group
->key
.offset
||
7358 bytenr
+ num_bytes
<= group
->key
.objectid
)
7361 extent_key
.objectid
= bytenr
;
7362 extent_key
.offset
= num_bytes
;
7363 extent_key
.type
= BTRFS_EXTENT_ITEM_KEY
;
7365 ret
= get_new_locations(reloc_inode
, &extent_key
,
7366 group
->key
.objectid
, 1,
7367 &new_extent
, &nr_extent
);
7372 BUG_ON(ref
->extents
[ext_index
].bytenr
!= bytenr
);
7373 BUG_ON(ref
->extents
[ext_index
].num_bytes
!= num_bytes
);
7374 ref
->extents
[ext_index
].bytenr
= new_extent
->disk_bytenr
;
7375 ref
->extents
[ext_index
].num_bytes
= new_extent
->disk_num_bytes
;
7377 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
7378 new_extent
->disk_bytenr
);
7379 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
7380 new_extent
->disk_num_bytes
);
7381 btrfs_mark_buffer_dirty(leaf
);
7383 ret
= btrfs_inc_extent_ref(trans
, root
,
7384 new_extent
->disk_bytenr
,
7385 new_extent
->disk_num_bytes
,
7387 root
->root_key
.objectid
,
7388 trans
->transid
, key
.objectid
);
7391 ret
= btrfs_free_extent(trans
, root
,
7392 bytenr
, num_bytes
, leaf
->start
,
7393 btrfs_header_owner(leaf
),
7394 btrfs_header_generation(leaf
),
7400 BUG_ON(ext_index
+ 1 != ref
->nritems
);
7401 btrfs_free_leaf_ref(root
, ref
);
7405 int btrfs_free_reloc_root(struct btrfs_trans_handle
*trans
,
7406 struct btrfs_root
*root
)
7408 struct btrfs_root
*reloc_root
;
7411 if (root
->reloc_root
) {
7412 reloc_root
= root
->reloc_root
;
7413 root
->reloc_root
= NULL
;
7414 list_add(&reloc_root
->dead_list
,
7415 &root
->fs_info
->dead_reloc_roots
);
7417 btrfs_set_root_bytenr(&reloc_root
->root_item
,
7418 reloc_root
->node
->start
);
7419 btrfs_set_root_level(&root
->root_item
,
7420 btrfs_header_level(reloc_root
->node
));
7421 memset(&reloc_root
->root_item
.drop_progress
, 0,
7422 sizeof(struct btrfs_disk_key
));
7423 reloc_root
->root_item
.drop_level
= 0;
7425 ret
= btrfs_update_root(trans
, root
->fs_info
->tree_root
,
7426 &reloc_root
->root_key
,
7427 &reloc_root
->root_item
);
7433 int btrfs_drop_dead_reloc_roots(struct btrfs_root
*root
)
7435 struct btrfs_trans_handle
*trans
;
7436 struct btrfs_root
*reloc_root
;
7437 struct btrfs_root
*prev_root
= NULL
;
7438 struct list_head dead_roots
;
7442 INIT_LIST_HEAD(&dead_roots
);
7443 list_splice_init(&root
->fs_info
->dead_reloc_roots
, &dead_roots
);
7445 while (!list_empty(&dead_roots
)) {
7446 reloc_root
= list_entry(dead_roots
.prev
,
7447 struct btrfs_root
, dead_list
);
7448 list_del_init(&reloc_root
->dead_list
);
7450 BUG_ON(reloc_root
->commit_root
!= NULL
);
7452 trans
= btrfs_join_transaction(root
, 1);
7455 mutex_lock(&root
->fs_info
->drop_mutex
);
7456 ret
= btrfs_drop_snapshot(trans
, reloc_root
);
7459 mutex_unlock(&root
->fs_info
->drop_mutex
);
7461 nr
= trans
->blocks_used
;
7462 ret
= btrfs_end_transaction(trans
, root
);
7464 btrfs_btree_balance_dirty(root
, nr
);
7467 free_extent_buffer(reloc_root
->node
);
7469 ret
= btrfs_del_root(trans
, root
->fs_info
->tree_root
,
7470 &reloc_root
->root_key
);
7472 mutex_unlock(&root
->fs_info
->drop_mutex
);
7474 nr
= trans
->blocks_used
;
7475 ret
= btrfs_end_transaction(trans
, root
);
7477 btrfs_btree_balance_dirty(root
, nr
);
7480 prev_root
= reloc_root
;
7483 btrfs_remove_leaf_refs(prev_root
, (u64
)-1, 0);
7489 int btrfs_add_dead_reloc_root(struct btrfs_root
*root
)
7491 list_add(&root
->dead_list
, &root
->fs_info
->dead_reloc_roots
);
7495 int btrfs_cleanup_reloc_trees(struct btrfs_root
*root
)
7497 struct btrfs_root
*reloc_root
;
7498 struct btrfs_trans_handle
*trans
;
7499 struct btrfs_key location
;
7503 mutex_lock(&root
->fs_info
->tree_reloc_mutex
);
7504 ret
= btrfs_find_dead_roots(root
, BTRFS_TREE_RELOC_OBJECTID
, NULL
);
7506 found
= !list_empty(&root
->fs_info
->dead_reloc_roots
);
7507 mutex_unlock(&root
->fs_info
->tree_reloc_mutex
);
7510 trans
= btrfs_start_transaction(root
, 1);
7512 ret
= btrfs_commit_transaction(trans
, root
);
7516 location
.objectid
= BTRFS_DATA_RELOC_TREE_OBJECTID
;
7517 location
.offset
= (u64
)-1;
7518 location
.type
= BTRFS_ROOT_ITEM_KEY
;
7520 reloc_root
= btrfs_read_fs_root_no_name(root
->fs_info
, &location
);
7521 BUG_ON(!reloc_root
);
7522 btrfs_orphan_cleanup(reloc_root
);
7526 static noinline
int init_reloc_tree(struct btrfs_trans_handle
*trans
,
7527 struct btrfs_root
*root
)
7529 struct btrfs_root
*reloc_root
;
7530 struct extent_buffer
*eb
;
7531 struct btrfs_root_item
*root_item
;
7532 struct btrfs_key root_key
;
7535 BUG_ON(!root
->ref_cows
);
7536 if (root
->reloc_root
)
7539 root_item
= kmalloc(sizeof(*root_item
), GFP_NOFS
);
7542 ret
= btrfs_copy_root(trans
, root
, root
->commit_root
,
7543 &eb
, BTRFS_TREE_RELOC_OBJECTID
);
7546 root_key
.objectid
= BTRFS_TREE_RELOC_OBJECTID
;
7547 root_key
.offset
= root
->root_key
.objectid
;
7548 root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
7550 memcpy(root_item
, &root
->root_item
, sizeof(root_item
));
7551 btrfs_set_root_refs(root_item
, 0);
7552 btrfs_set_root_bytenr(root_item
, eb
->start
);
7553 btrfs_set_root_level(root_item
, btrfs_header_level(eb
));
7554 btrfs_set_root_generation(root_item
, trans
->transid
);
7556 btrfs_tree_unlock(eb
);
7557 free_extent_buffer(eb
);
7559 ret
= btrfs_insert_root(trans
, root
->fs_info
->tree_root
,
7560 &root_key
, root_item
);
7564 reloc_root
= btrfs_read_fs_root_no_radix(root
->fs_info
->tree_root
,
7566 BUG_ON(!reloc_root
);
7567 reloc_root
->last_trans
= trans
->transid
;
7568 reloc_root
->commit_root
= NULL
;
7569 reloc_root
->ref_tree
= &root
->fs_info
->reloc_ref_tree
;
7571 root
->reloc_root
= reloc_root
;
7576 * Core function of space balance.
7578 * The idea is using reloc trees to relocate tree blocks in reference
7579 * counted roots. There is one reloc tree for each subvol, and all
7580 * reloc trees share same root key objectid. Reloc trees are snapshots
7581 * of the latest committed roots of subvols (root->commit_root).
7583 * To relocate a tree block referenced by a subvol, there are two steps.
7584 * COW the block through subvol's reloc tree, then update block pointer
7585 * in the subvol to point to the new block. Since all reloc trees share
7586 * same root key objectid, doing special handing for tree blocks owned
7587 * by them is easy. Once a tree block has been COWed in one reloc tree,
7588 * we can use the resulting new block directly when the same block is
7589 * required to COW again through other reloc trees. By this way, relocated
7590 * tree blocks are shared between reloc trees, so they are also shared
7593 static noinline
int relocate_one_path(struct btrfs_trans_handle
*trans
,
7594 struct btrfs_root
*root
,
7595 struct btrfs_path
*path
,
7596 struct btrfs_key
*first_key
,
7597 struct btrfs_ref_path
*ref_path
,
7598 struct btrfs_block_group_cache
*group
,
7599 struct inode
*reloc_inode
)
7601 struct btrfs_root
*reloc_root
;
7602 struct extent_buffer
*eb
= NULL
;
7603 struct btrfs_key
*keys
;
7607 int lowest_level
= 0;
7610 if (ref_path
->owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
)
7611 lowest_level
= ref_path
->owner_objectid
;
7613 if (!root
->ref_cows
) {
7614 path
->lowest_level
= lowest_level
;
7615 ret
= btrfs_search_slot(trans
, root
, first_key
, path
, 0, 1);
7617 path
->lowest_level
= 0;
7618 btrfs_release_path(root
, path
);
7622 mutex_lock(&root
->fs_info
->tree_reloc_mutex
);
7623 ret
= init_reloc_tree(trans
, root
);
7625 reloc_root
= root
->reloc_root
;
7627 shared_level
= ref_path
->shared_level
;
7628 ref_path
->shared_level
= BTRFS_MAX_LEVEL
- 1;
7630 keys
= ref_path
->node_keys
;
7631 nodes
= ref_path
->new_nodes
;
7632 memset(&keys
[shared_level
+ 1], 0,
7633 sizeof(*keys
) * (BTRFS_MAX_LEVEL
- shared_level
- 1));
7634 memset(&nodes
[shared_level
+ 1], 0,
7635 sizeof(*nodes
) * (BTRFS_MAX_LEVEL
- shared_level
- 1));
7637 if (nodes
[lowest_level
] == 0) {
7638 path
->lowest_level
= lowest_level
;
7639 ret
= btrfs_search_slot(trans
, reloc_root
, first_key
, path
,
7642 for (level
= lowest_level
; level
< BTRFS_MAX_LEVEL
; level
++) {
7643 eb
= path
->nodes
[level
];
7644 if (!eb
|| eb
== reloc_root
->node
)
7646 nodes
[level
] = eb
->start
;
7648 btrfs_item_key_to_cpu(eb
, &keys
[level
], 0);
7650 btrfs_node_key_to_cpu(eb
, &keys
[level
], 0);
7653 ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
7654 eb
= path
->nodes
[0];
7655 ret
= replace_extents_in_leaf(trans
, reloc_root
, eb
,
7656 group
, reloc_inode
);
7659 btrfs_release_path(reloc_root
, path
);
7661 ret
= btrfs_merge_path(trans
, reloc_root
, keys
, nodes
,
7667 * replace tree blocks in the fs tree with tree blocks in
7670 ret
= btrfs_merge_path(trans
, root
, keys
, nodes
, lowest_level
);
7673 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
7674 ret
= btrfs_search_slot(trans
, reloc_root
, first_key
, path
,
7677 extent_buffer_get(path
->nodes
[0]);
7678 eb
= path
->nodes
[0];
7679 btrfs_release_path(reloc_root
, path
);
7680 ret
= invalidate_extent_cache(reloc_root
, eb
, group
, root
);
7682 free_extent_buffer(eb
);
7685 mutex_unlock(&root
->fs_info
->tree_reloc_mutex
);
7686 path
->lowest_level
= 0;
7690 static noinline
int relocate_tree_block(struct btrfs_trans_handle
*trans
,
7691 struct btrfs_root
*root
,
7692 struct btrfs_path
*path
,
7693 struct btrfs_key
*first_key
,
7694 struct btrfs_ref_path
*ref_path
)
7698 ret
= relocate_one_path(trans
, root
, path
, first_key
,
7699 ref_path
, NULL
, NULL
);
7705 static noinline
int del_extent_zero(struct btrfs_trans_handle
*trans
,
7706 struct btrfs_root
*extent_root
,
7707 struct btrfs_path
*path
,
7708 struct btrfs_key
*extent_key
)
7712 ret
= btrfs_search_slot(trans
, extent_root
, extent_key
, path
, -1, 1);
7715 ret
= btrfs_del_item(trans
, extent_root
, path
);
7717 btrfs_release_path(extent_root
, path
);
7721 static noinline
struct btrfs_root
*read_ref_root(struct btrfs_fs_info
*fs_info
,
7722 struct btrfs_ref_path
*ref_path
)
7724 struct btrfs_key root_key
;
7726 root_key
.objectid
= ref_path
->root_objectid
;
7727 root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
7728 if (is_cowonly_root(ref_path
->root_objectid
))
7729 root_key
.offset
= 0;
7731 root_key
.offset
= (u64
)-1;
7733 return btrfs_read_fs_root_no_name(fs_info
, &root_key
);
7736 static noinline
int relocate_one_extent(struct btrfs_root
*extent_root
,
7737 struct btrfs_path
*path
,
7738 struct btrfs_key
*extent_key
,
7739 struct btrfs_block_group_cache
*group
,
7740 struct inode
*reloc_inode
, int pass
)
7742 struct btrfs_trans_handle
*trans
;
7743 struct btrfs_root
*found_root
;
7744 struct btrfs_ref_path
*ref_path
= NULL
;
7745 struct disk_extent
*new_extents
= NULL
;
7750 struct btrfs_key first_key
;
7754 trans
= btrfs_start_transaction(extent_root
, 1);
7757 if (extent_key
->objectid
== 0) {
7758 ret
= del_extent_zero(trans
, extent_root
, path
, extent_key
);
7762 ref_path
= kmalloc(sizeof(*ref_path
), GFP_NOFS
);
7768 for (loops
= 0; ; loops
++) {
7770 ret
= btrfs_first_ref_path(trans
, extent_root
, ref_path
,
7771 extent_key
->objectid
);
7773 ret
= btrfs_next_ref_path(trans
, extent_root
, ref_path
);
7780 if (ref_path
->root_objectid
== BTRFS_TREE_LOG_OBJECTID
||
7781 ref_path
->root_objectid
== BTRFS_TREE_RELOC_OBJECTID
)
7784 found_root
= read_ref_root(extent_root
->fs_info
, ref_path
);
7785 BUG_ON(!found_root
);
7787 * for reference counted tree, only process reference paths
7788 * rooted at the latest committed root.
7790 if (found_root
->ref_cows
&&
7791 ref_path
->root_generation
!= found_root
->root_key
.offset
)
7794 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
7797 * copy data extents to new locations
7799 u64 group_start
= group
->key
.objectid
;
7800 ret
= relocate_data_extent(reloc_inode
,
7809 level
= ref_path
->owner_objectid
;
7812 if (prev_block
!= ref_path
->nodes
[level
]) {
7813 struct extent_buffer
*eb
;
7814 u64 block_start
= ref_path
->nodes
[level
];
7815 u64 block_size
= btrfs_level_size(found_root
, level
);
7817 eb
= read_tree_block(found_root
, block_start
,
7819 btrfs_tree_lock(eb
);
7820 BUG_ON(level
!= btrfs_header_level(eb
));
7823 btrfs_item_key_to_cpu(eb
, &first_key
, 0);
7825 btrfs_node_key_to_cpu(eb
, &first_key
, 0);
7827 btrfs_tree_unlock(eb
);
7828 free_extent_buffer(eb
);
7829 prev_block
= block_start
;
7832 mutex_lock(&extent_root
->fs_info
->trans_mutex
);
7833 btrfs_record_root_in_trans(found_root
);
7834 mutex_unlock(&extent_root
->fs_info
->trans_mutex
);
7835 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
7837 * try to update data extent references while
7838 * keeping metadata shared between snapshots.
7841 ret
= relocate_one_path(trans
, found_root
,
7842 path
, &first_key
, ref_path
,
7843 group
, reloc_inode
);
7849 * use fallback method to process the remaining
7853 u64 group_start
= group
->key
.objectid
;
7854 new_extents
= kmalloc(sizeof(*new_extents
),
7857 ret
= get_new_locations(reloc_inode
,
7865 ret
= replace_one_extent(trans
, found_root
,
7867 &first_key
, ref_path
,
7868 new_extents
, nr_extents
);
7870 ret
= relocate_tree_block(trans
, found_root
, path
,
7871 &first_key
, ref_path
);
7878 btrfs_end_transaction(trans
, extent_root
);
7885 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
7888 u64 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
7889 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
7891 num_devices
= root
->fs_info
->fs_devices
->rw_devices
;
7892 if (num_devices
== 1) {
7893 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
7894 stripped
= flags
& ~stripped
;
7896 /* turn raid0 into single device chunks */
7897 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
7900 /* turn mirroring into duplication */
7901 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
7902 BTRFS_BLOCK_GROUP_RAID10
))
7903 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
7906 /* they already had raid on here, just return */
7907 if (flags
& stripped
)
7910 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
7911 stripped
= flags
& ~stripped
;
7913 /* switch duplicated blocks with raid1 */
7914 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
7915 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
7917 /* turn single device chunks into raid0 */
7918 return stripped
| BTRFS_BLOCK_GROUP_RAID0
;
7923 static int set_block_group_ro(struct btrfs_block_group_cache
*cache
)
7925 struct btrfs_space_info
*sinfo
= cache
->space_info
;
7932 spin_lock(&sinfo
->lock
);
7933 spin_lock(&cache
->lock
);
7934 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
7935 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
7937 if (sinfo
->bytes_used
+ sinfo
->bytes_reserved
+ sinfo
->bytes_pinned
+
7938 sinfo
->bytes_may_use
+ sinfo
->bytes_readonly
+
7939 cache
->reserved_pinned
+ num_bytes
< sinfo
->total_bytes
) {
7940 sinfo
->bytes_readonly
+= num_bytes
;
7941 sinfo
->bytes_reserved
+= cache
->reserved_pinned
;
7942 cache
->reserved_pinned
= 0;
7946 spin_unlock(&cache
->lock
);
7947 spin_unlock(&sinfo
->lock
);
7951 int btrfs_set_block_group_ro(struct btrfs_root
*root
,
7952 struct btrfs_block_group_cache
*cache
)
7955 struct btrfs_trans_handle
*trans
;
7961 trans
= btrfs_join_transaction(root
, 1);
7962 BUG_ON(IS_ERR(trans
));
7964 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
7965 if (alloc_flags
!= cache
->flags
)
7966 do_chunk_alloc(trans
, root
, 2 * 1024 * 1024, alloc_flags
, 1);
7968 ret
= set_block_group_ro(cache
);
7971 alloc_flags
= get_alloc_profile(root
, cache
->space_info
->flags
);
7972 ret
= do_chunk_alloc(trans
, root
, 2 * 1024 * 1024, alloc_flags
, 1);
7975 ret
= set_block_group_ro(cache
);
7977 btrfs_end_transaction(trans
, root
);
7981 int btrfs_set_block_group_rw(struct btrfs_root
*root
,
7982 struct btrfs_block_group_cache
*cache
)
7984 struct btrfs_space_info
*sinfo
= cache
->space_info
;
7989 spin_lock(&sinfo
->lock
);
7990 spin_lock(&cache
->lock
);
7991 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
7992 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
7993 sinfo
->bytes_readonly
-= num_bytes
;
7995 spin_unlock(&cache
->lock
);
7996 spin_unlock(&sinfo
->lock
);
8001 * checks to see if its even possible to relocate this block group.
8003 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8004 * ok to go ahead and try.
8006 int btrfs_can_relocate(struct btrfs_root
*root
, u64 bytenr
)
8008 struct btrfs_block_group_cache
*block_group
;
8009 struct btrfs_space_info
*space_info
;
8010 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
8011 struct btrfs_device
*device
;
8015 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
8017 /* odd, couldn't find the block group, leave it alone */
8021 /* no bytes used, we're good */
8022 if (!btrfs_block_group_used(&block_group
->item
))
8025 space_info
= block_group
->space_info
;
8026 spin_lock(&space_info
->lock
);
8028 full
= space_info
->full
;
8031 * if this is the last block group we have in this space, we can't
8032 * relocate it unless we're able to allocate a new chunk below.
8034 * Otherwise, we need to make sure we have room in the space to handle
8035 * all of the extents from this block group. If we can, we're good
8037 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
8038 (space_info
->bytes_used
+ space_info
->bytes_reserved
+
8039 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
8040 btrfs_block_group_used(&block_group
->item
) <
8041 space_info
->total_bytes
)) {
8042 spin_unlock(&space_info
->lock
);
8045 spin_unlock(&space_info
->lock
);
8048 * ok we don't have enough space, but maybe we have free space on our
8049 * devices to allocate new chunks for relocation, so loop through our
8050 * alloc devices and guess if we have enough space. However, if we
8051 * were marked as full, then we know there aren't enough chunks, and we
8058 mutex_lock(&root
->fs_info
->chunk_mutex
);
8059 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
8060 u64 min_free
= btrfs_block_group_used(&block_group
->item
);
8061 u64 dev_offset
, max_avail
;
8064 * check to make sure we can actually find a chunk with enough
8065 * space to fit our block group in.
8067 if (device
->total_bytes
> device
->bytes_used
+ min_free
) {
8068 ret
= find_free_dev_extent(NULL
, device
, min_free
,
8069 &dev_offset
, &max_avail
);
8075 mutex_unlock(&root
->fs_info
->chunk_mutex
);
8077 btrfs_put_block_group(block_group
);
8081 static int find_first_block_group(struct btrfs_root
*root
,
8082 struct btrfs_path
*path
, struct btrfs_key
*key
)
8085 struct btrfs_key found_key
;
8086 struct extent_buffer
*leaf
;
8089 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
8094 slot
= path
->slots
[0];
8095 leaf
= path
->nodes
[0];
8096 if (slot
>= btrfs_header_nritems(leaf
)) {
8097 ret
= btrfs_next_leaf(root
, path
);
8104 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
8106 if (found_key
.objectid
>= key
->objectid
&&
8107 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
8117 void btrfs_put_block_group_cache(struct btrfs_fs_info
*info
)
8119 struct btrfs_block_group_cache
*block_group
;
8123 struct inode
*inode
;
8125 block_group
= btrfs_lookup_first_block_group(info
, last
);
8126 while (block_group
) {
8127 spin_lock(&block_group
->lock
);
8128 if (block_group
->iref
)
8130 spin_unlock(&block_group
->lock
);
8131 block_group
= next_block_group(info
->tree_root
,
8141 inode
= block_group
->inode
;
8142 block_group
->iref
= 0;
8143 block_group
->inode
= NULL
;
8144 spin_unlock(&block_group
->lock
);
8146 last
= block_group
->key
.objectid
+ block_group
->key
.offset
;
8147 btrfs_put_block_group(block_group
);
8151 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
8153 struct btrfs_block_group_cache
*block_group
;
8154 struct btrfs_space_info
*space_info
;
8155 struct btrfs_caching_control
*caching_ctl
;
8158 down_write(&info
->extent_commit_sem
);
8159 while (!list_empty(&info
->caching_block_groups
)) {
8160 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
8161 struct btrfs_caching_control
, list
);
8162 list_del(&caching_ctl
->list
);
8163 put_caching_control(caching_ctl
);
8165 up_write(&info
->extent_commit_sem
);
8167 spin_lock(&info
->block_group_cache_lock
);
8168 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
8169 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
8171 rb_erase(&block_group
->cache_node
,
8172 &info
->block_group_cache_tree
);
8173 spin_unlock(&info
->block_group_cache_lock
);
8175 down_write(&block_group
->space_info
->groups_sem
);
8176 list_del(&block_group
->list
);
8177 up_write(&block_group
->space_info
->groups_sem
);
8179 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
8180 wait_block_group_cache_done(block_group
);
8182 btrfs_remove_free_space_cache(block_group
);
8183 btrfs_put_block_group(block_group
);
8185 spin_lock(&info
->block_group_cache_lock
);
8187 spin_unlock(&info
->block_group_cache_lock
);
8189 /* now that all the block groups are freed, go through and
8190 * free all the space_info structs. This is only called during
8191 * the final stages of unmount, and so we know nobody is
8192 * using them. We call synchronize_rcu() once before we start,
8193 * just to be on the safe side.
8197 release_global_block_rsv(info
);
8199 while(!list_empty(&info
->space_info
)) {
8200 space_info
= list_entry(info
->space_info
.next
,
8201 struct btrfs_space_info
,
8203 if (space_info
->bytes_pinned
> 0 ||
8204 space_info
->bytes_reserved
> 0) {
8206 dump_space_info(space_info
, 0, 0);
8208 list_del(&space_info
->list
);
8214 static void __link_block_group(struct btrfs_space_info
*space_info
,
8215 struct btrfs_block_group_cache
*cache
)
8217 int index
= get_block_group_index(cache
);
8219 down_write(&space_info
->groups_sem
);
8220 list_add_tail(&cache
->list
, &space_info
->block_groups
[index
]);
8221 up_write(&space_info
->groups_sem
);
8224 int btrfs_read_block_groups(struct btrfs_root
*root
)
8226 struct btrfs_path
*path
;
8228 struct btrfs_block_group_cache
*cache
;
8229 struct btrfs_fs_info
*info
= root
->fs_info
;
8230 struct btrfs_space_info
*space_info
;
8231 struct btrfs_key key
;
8232 struct btrfs_key found_key
;
8233 struct extent_buffer
*leaf
;
8237 root
= info
->extent_root
;
8240 btrfs_set_key_type(&key
, BTRFS_BLOCK_GROUP_ITEM_KEY
);
8241 path
= btrfs_alloc_path();
8245 cache_gen
= btrfs_super_cache_generation(&root
->fs_info
->super_copy
);
8246 if (cache_gen
!= 0 &&
8247 btrfs_super_generation(&root
->fs_info
->super_copy
) != cache_gen
)
8249 if (btrfs_test_opt(root
, CLEAR_CACHE
))
8251 if (!btrfs_test_opt(root
, SPACE_CACHE
) && cache_gen
)
8252 printk(KERN_INFO
"btrfs: disk space caching is enabled\n");
8255 ret
= find_first_block_group(root
, path
, &key
);
8261 leaf
= path
->nodes
[0];
8262 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
8263 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
8269 atomic_set(&cache
->count
, 1);
8270 spin_lock_init(&cache
->lock
);
8271 spin_lock_init(&cache
->tree_lock
);
8272 cache
->fs_info
= info
;
8273 INIT_LIST_HEAD(&cache
->list
);
8274 INIT_LIST_HEAD(&cache
->cluster_list
);
8277 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
8280 * we only want to have 32k of ram per block group for keeping
8281 * track of free space, and if we pass 1/2 of that we want to
8282 * start converting things over to using bitmaps
8284 cache
->extents_thresh
= ((1024 * 32) / 2) /
8285 sizeof(struct btrfs_free_space
);
8287 read_extent_buffer(leaf
, &cache
->item
,
8288 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
8289 sizeof(cache
->item
));
8290 memcpy(&cache
->key
, &found_key
, sizeof(found_key
));
8292 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
8293 btrfs_release_path(root
, path
);
8294 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
8295 cache
->sectorsize
= root
->sectorsize
;
8298 * check for two cases, either we are full, and therefore
8299 * don't need to bother with the caching work since we won't
8300 * find any space, or we are empty, and we can just add all
8301 * the space in and be done with it. This saves us _alot_ of
8302 * time, particularly in the full case.
8304 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
8305 exclude_super_stripes(root
, cache
);
8306 cache
->last_byte_to_unpin
= (u64
)-1;
8307 cache
->cached
= BTRFS_CACHE_FINISHED
;
8308 free_excluded_extents(root
, cache
);
8309 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
8310 exclude_super_stripes(root
, cache
);
8311 cache
->last_byte_to_unpin
= (u64
)-1;
8312 cache
->cached
= BTRFS_CACHE_FINISHED
;
8313 add_new_free_space(cache
, root
->fs_info
,
8315 found_key
.objectid
+
8317 free_excluded_extents(root
, cache
);
8320 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
8321 btrfs_block_group_used(&cache
->item
),
8324 cache
->space_info
= space_info
;
8325 spin_lock(&cache
->space_info
->lock
);
8326 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
8327 spin_unlock(&cache
->space_info
->lock
);
8329 __link_block_group(space_info
, cache
);
8331 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
8334 set_avail_alloc_bits(root
->fs_info
, cache
->flags
);
8335 if (btrfs_chunk_readonly(root
, cache
->key
.objectid
))
8336 set_block_group_ro(cache
);
8339 list_for_each_entry_rcu(space_info
, &root
->fs_info
->space_info
, list
) {
8340 if (!(get_alloc_profile(root
, space_info
->flags
) &
8341 (BTRFS_BLOCK_GROUP_RAID10
|
8342 BTRFS_BLOCK_GROUP_RAID1
|
8343 BTRFS_BLOCK_GROUP_DUP
)))
8346 * avoid allocating from un-mirrored block group if there are
8347 * mirrored block groups.
8349 list_for_each_entry(cache
, &space_info
->block_groups
[3], list
)
8350 set_block_group_ro(cache
);
8351 list_for_each_entry(cache
, &space_info
->block_groups
[4], list
)
8352 set_block_group_ro(cache
);
8355 init_global_block_rsv(info
);
8358 btrfs_free_path(path
);
8362 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
8363 struct btrfs_root
*root
, u64 bytes_used
,
8364 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
8368 struct btrfs_root
*extent_root
;
8369 struct btrfs_block_group_cache
*cache
;
8371 extent_root
= root
->fs_info
->extent_root
;
8373 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
8375 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
8379 cache
->key
.objectid
= chunk_offset
;
8380 cache
->key
.offset
= size
;
8381 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
8382 cache
->sectorsize
= root
->sectorsize
;
8383 cache
->fs_info
= root
->fs_info
;
8386 * we only want to have 32k of ram per block group for keeping track
8387 * of free space, and if we pass 1/2 of that we want to start
8388 * converting things over to using bitmaps
8390 cache
->extents_thresh
= ((1024 * 32) / 2) /
8391 sizeof(struct btrfs_free_space
);
8392 atomic_set(&cache
->count
, 1);
8393 spin_lock_init(&cache
->lock
);
8394 spin_lock_init(&cache
->tree_lock
);
8395 INIT_LIST_HEAD(&cache
->list
);
8396 INIT_LIST_HEAD(&cache
->cluster_list
);
8398 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
8399 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
8400 cache
->flags
= type
;
8401 btrfs_set_block_group_flags(&cache
->item
, type
);
8403 cache
->last_byte_to_unpin
= (u64
)-1;
8404 cache
->cached
= BTRFS_CACHE_FINISHED
;
8405 exclude_super_stripes(root
, cache
);
8407 add_new_free_space(cache
, root
->fs_info
, chunk_offset
,
8408 chunk_offset
+ size
);
8410 free_excluded_extents(root
, cache
);
8412 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
8413 &cache
->space_info
);
8416 spin_lock(&cache
->space_info
->lock
);
8417 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
8418 spin_unlock(&cache
->space_info
->lock
);
8420 __link_block_group(cache
->space_info
, cache
);
8422 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
8425 ret
= btrfs_insert_item(trans
, extent_root
, &cache
->key
, &cache
->item
,
8426 sizeof(cache
->item
));
8429 set_avail_alloc_bits(extent_root
->fs_info
, type
);
8434 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
8435 struct btrfs_root
*root
, u64 group_start
)
8437 struct btrfs_path
*path
;
8438 struct btrfs_block_group_cache
*block_group
;
8439 struct btrfs_free_cluster
*cluster
;
8440 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
8441 struct btrfs_key key
;
8442 struct inode
*inode
;
8446 root
= root
->fs_info
->extent_root
;
8448 block_group
= btrfs_lookup_block_group(root
->fs_info
, group_start
);
8449 BUG_ON(!block_group
);
8450 BUG_ON(!block_group
->ro
);
8452 memcpy(&key
, &block_group
->key
, sizeof(key
));
8453 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
8454 BTRFS_BLOCK_GROUP_RAID1
|
8455 BTRFS_BLOCK_GROUP_RAID10
))
8460 /* make sure this block group isn't part of an allocation cluster */
8461 cluster
= &root
->fs_info
->data_alloc_cluster
;
8462 spin_lock(&cluster
->refill_lock
);
8463 btrfs_return_cluster_to_free_space(block_group
, cluster
);
8464 spin_unlock(&cluster
->refill_lock
);
8467 * make sure this block group isn't part of a metadata
8468 * allocation cluster
8470 cluster
= &root
->fs_info
->meta_alloc_cluster
;
8471 spin_lock(&cluster
->refill_lock
);
8472 btrfs_return_cluster_to_free_space(block_group
, cluster
);
8473 spin_unlock(&cluster
->refill_lock
);
8475 path
= btrfs_alloc_path();
8478 inode
= lookup_free_space_inode(root
, block_group
, path
);
8479 if (!IS_ERR(inode
)) {
8480 btrfs_orphan_add(trans
, inode
);
8482 /* One for the block groups ref */
8483 spin_lock(&block_group
->lock
);
8484 if (block_group
->iref
) {
8485 block_group
->iref
= 0;
8486 block_group
->inode
= NULL
;
8487 spin_unlock(&block_group
->lock
);
8490 spin_unlock(&block_group
->lock
);
8492 /* One for our lookup ref */
8496 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
8497 key
.offset
= block_group
->key
.objectid
;
8500 ret
= btrfs_search_slot(trans
, tree_root
, &key
, path
, -1, 1);
8504 btrfs_release_path(tree_root
, path
);
8506 ret
= btrfs_del_item(trans
, tree_root
, path
);
8509 btrfs_release_path(tree_root
, path
);
8512 spin_lock(&root
->fs_info
->block_group_cache_lock
);
8513 rb_erase(&block_group
->cache_node
,
8514 &root
->fs_info
->block_group_cache_tree
);
8515 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
8517 down_write(&block_group
->space_info
->groups_sem
);
8519 * we must use list_del_init so people can check to see if they
8520 * are still on the list after taking the semaphore
8522 list_del_init(&block_group
->list
);
8523 up_write(&block_group
->space_info
->groups_sem
);
8525 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
8526 wait_block_group_cache_done(block_group
);
8528 btrfs_remove_free_space_cache(block_group
);
8530 spin_lock(&block_group
->space_info
->lock
);
8531 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
8532 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
8533 block_group
->space_info
->disk_total
-= block_group
->key
.offset
* factor
;
8534 spin_unlock(&block_group
->space_info
->lock
);
8536 memcpy(&key
, &block_group
->key
, sizeof(key
));
8538 btrfs_clear_space_info_full(root
->fs_info
);
8540 btrfs_put_block_group(block_group
);
8541 btrfs_put_block_group(block_group
);
8543 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
8549 ret
= btrfs_del_item(trans
, root
, path
);
8551 btrfs_free_path(path
);