2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include <linux/slab.h>
30 #include "print-tree.h"
31 #include "transaction.h"
34 #include "free-space-cache.h"
36 /* control flags for do_chunk_alloc's force field
37 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
38 * if we really need one.
40 * CHUNK_ALLOC_FORCE means it must try to allocate one
42 * CHUNK_ALLOC_LIMITED means to only try and allocate one
43 * if we have very few chunks already allocated. This is
44 * used as part of the clustering code to help make sure
45 * we have a good pool of storage to cluster in, without
46 * filling the FS with empty chunks
50 CHUNK_ALLOC_NO_FORCE
= 0,
51 CHUNK_ALLOC_FORCE
= 1,
52 CHUNK_ALLOC_LIMITED
= 2,
55 static int update_block_group(struct btrfs_trans_handle
*trans
,
56 struct btrfs_root
*root
,
57 u64 bytenr
, u64 num_bytes
, int alloc
);
58 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
59 struct btrfs_root
*root
,
60 u64 bytenr
, u64 num_bytes
, u64 parent
,
61 u64 root_objectid
, u64 owner_objectid
,
62 u64 owner_offset
, int refs_to_drop
,
63 struct btrfs_delayed_extent_op
*extra_op
);
64 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
65 struct extent_buffer
*leaf
,
66 struct btrfs_extent_item
*ei
);
67 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
68 struct btrfs_root
*root
,
69 u64 parent
, u64 root_objectid
,
70 u64 flags
, u64 owner
, u64 offset
,
71 struct btrfs_key
*ins
, int ref_mod
);
72 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
73 struct btrfs_root
*root
,
74 u64 parent
, u64 root_objectid
,
75 u64 flags
, struct btrfs_disk_key
*key
,
76 int level
, struct btrfs_key
*ins
);
77 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
78 struct btrfs_root
*extent_root
, u64 alloc_bytes
,
79 u64 flags
, int force
);
80 static int find_next_key(struct btrfs_path
*path
, int level
,
81 struct btrfs_key
*key
);
82 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
83 int dump_block_groups
);
86 block_group_cache_done(struct btrfs_block_group_cache
*cache
)
89 return cache
->cached
== BTRFS_CACHE_FINISHED
;
92 static int block_group_bits(struct btrfs_block_group_cache
*cache
, u64 bits
)
94 return (cache
->flags
& bits
) == bits
;
97 void btrfs_get_block_group(struct btrfs_block_group_cache
*cache
)
99 atomic_inc(&cache
->count
);
102 void btrfs_put_block_group(struct btrfs_block_group_cache
*cache
)
104 if (atomic_dec_and_test(&cache
->count
)) {
105 WARN_ON(cache
->pinned
> 0);
106 WARN_ON(cache
->reserved
> 0);
107 WARN_ON(cache
->reserved_pinned
> 0);
108 kfree(cache
->free_space_ctl
);
114 * this adds the block group to the fs_info rb tree for the block group
117 static int btrfs_add_block_group_cache(struct btrfs_fs_info
*info
,
118 struct btrfs_block_group_cache
*block_group
)
121 struct rb_node
*parent
= NULL
;
122 struct btrfs_block_group_cache
*cache
;
124 spin_lock(&info
->block_group_cache_lock
);
125 p
= &info
->block_group_cache_tree
.rb_node
;
129 cache
= rb_entry(parent
, struct btrfs_block_group_cache
,
131 if (block_group
->key
.objectid
< cache
->key
.objectid
) {
133 } else if (block_group
->key
.objectid
> cache
->key
.objectid
) {
136 spin_unlock(&info
->block_group_cache_lock
);
141 rb_link_node(&block_group
->cache_node
, parent
, p
);
142 rb_insert_color(&block_group
->cache_node
,
143 &info
->block_group_cache_tree
);
144 spin_unlock(&info
->block_group_cache_lock
);
150 * This will return the block group at or after bytenr if contains is 0, else
151 * it will return the block group that contains the bytenr
153 static struct btrfs_block_group_cache
*
154 block_group_cache_tree_search(struct btrfs_fs_info
*info
, u64 bytenr
,
157 struct btrfs_block_group_cache
*cache
, *ret
= NULL
;
161 spin_lock(&info
->block_group_cache_lock
);
162 n
= info
->block_group_cache_tree
.rb_node
;
165 cache
= rb_entry(n
, struct btrfs_block_group_cache
,
167 end
= cache
->key
.objectid
+ cache
->key
.offset
- 1;
168 start
= cache
->key
.objectid
;
170 if (bytenr
< start
) {
171 if (!contains
&& (!ret
|| start
< ret
->key
.objectid
))
174 } else if (bytenr
> start
) {
175 if (contains
&& bytenr
<= end
) {
186 btrfs_get_block_group(ret
);
187 spin_unlock(&info
->block_group_cache_lock
);
192 static int add_excluded_extent(struct btrfs_root
*root
,
193 u64 start
, u64 num_bytes
)
195 u64 end
= start
+ num_bytes
- 1;
196 set_extent_bits(&root
->fs_info
->freed_extents
[0],
197 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
198 set_extent_bits(&root
->fs_info
->freed_extents
[1],
199 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
203 static void free_excluded_extents(struct btrfs_root
*root
,
204 struct btrfs_block_group_cache
*cache
)
208 start
= cache
->key
.objectid
;
209 end
= start
+ cache
->key
.offset
- 1;
211 clear_extent_bits(&root
->fs_info
->freed_extents
[0],
212 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
213 clear_extent_bits(&root
->fs_info
->freed_extents
[1],
214 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
217 static int exclude_super_stripes(struct btrfs_root
*root
,
218 struct btrfs_block_group_cache
*cache
)
225 if (cache
->key
.objectid
< BTRFS_SUPER_INFO_OFFSET
) {
226 stripe_len
= BTRFS_SUPER_INFO_OFFSET
- cache
->key
.objectid
;
227 cache
->bytes_super
+= stripe_len
;
228 ret
= add_excluded_extent(root
, cache
->key
.objectid
,
233 for (i
= 0; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
234 bytenr
= btrfs_sb_offset(i
);
235 ret
= btrfs_rmap_block(&root
->fs_info
->mapping_tree
,
236 cache
->key
.objectid
, bytenr
,
237 0, &logical
, &nr
, &stripe_len
);
241 cache
->bytes_super
+= stripe_len
;
242 ret
= add_excluded_extent(root
, logical
[nr
],
252 static struct btrfs_caching_control
*
253 get_caching_control(struct btrfs_block_group_cache
*cache
)
255 struct btrfs_caching_control
*ctl
;
257 spin_lock(&cache
->lock
);
258 if (cache
->cached
!= BTRFS_CACHE_STARTED
) {
259 spin_unlock(&cache
->lock
);
263 /* We're loading it the fast way, so we don't have a caching_ctl. */
264 if (!cache
->caching_ctl
) {
265 spin_unlock(&cache
->lock
);
269 ctl
= cache
->caching_ctl
;
270 atomic_inc(&ctl
->count
);
271 spin_unlock(&cache
->lock
);
275 static void put_caching_control(struct btrfs_caching_control
*ctl
)
277 if (atomic_dec_and_test(&ctl
->count
))
282 * this is only called by cache_block_group, since we could have freed extents
283 * we need to check the pinned_extents for any extents that can't be used yet
284 * since their free space will be released as soon as the transaction commits.
286 static u64
add_new_free_space(struct btrfs_block_group_cache
*block_group
,
287 struct btrfs_fs_info
*info
, u64 start
, u64 end
)
289 u64 extent_start
, extent_end
, size
, total_added
= 0;
292 while (start
< end
) {
293 ret
= find_first_extent_bit(info
->pinned_extents
, start
,
294 &extent_start
, &extent_end
,
295 EXTENT_DIRTY
| EXTENT_UPTODATE
);
299 if (extent_start
<= start
) {
300 start
= extent_end
+ 1;
301 } else if (extent_start
> start
&& extent_start
< end
) {
302 size
= extent_start
- start
;
304 ret
= btrfs_add_free_space(block_group
, start
,
307 start
= extent_end
+ 1;
316 ret
= btrfs_add_free_space(block_group
, start
, size
);
323 static int caching_kthread(void *data
)
325 struct btrfs_block_group_cache
*block_group
= data
;
326 struct btrfs_fs_info
*fs_info
= block_group
->fs_info
;
327 struct btrfs_caching_control
*caching_ctl
= block_group
->caching_ctl
;
328 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
329 struct btrfs_path
*path
;
330 struct extent_buffer
*leaf
;
331 struct btrfs_key key
;
337 path
= btrfs_alloc_path();
341 last
= max_t(u64
, block_group
->key
.objectid
, BTRFS_SUPER_INFO_OFFSET
);
344 * We don't want to deadlock with somebody trying to allocate a new
345 * extent for the extent root while also trying to search the extent
346 * root to add free space. So we skip locking and search the commit
347 * root, since its read-only
349 path
->skip_locking
= 1;
350 path
->search_commit_root
= 1;
355 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
357 mutex_lock(&caching_ctl
->mutex
);
358 /* need to make sure the commit_root doesn't disappear */
359 down_read(&fs_info
->extent_commit_sem
);
361 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
365 leaf
= path
->nodes
[0];
366 nritems
= btrfs_header_nritems(leaf
);
370 if (fs_info
->closing
> 1) {
375 if (path
->slots
[0] < nritems
) {
376 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
378 ret
= find_next_key(path
, 0, &key
);
382 caching_ctl
->progress
= last
;
383 btrfs_release_path(extent_root
, path
);
384 up_read(&fs_info
->extent_commit_sem
);
385 mutex_unlock(&caching_ctl
->mutex
);
386 if (btrfs_transaction_in_commit(fs_info
))
393 if (key
.objectid
< block_group
->key
.objectid
) {
398 if (key
.objectid
>= block_group
->key
.objectid
+
399 block_group
->key
.offset
)
402 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
) {
403 total_found
+= add_new_free_space(block_group
,
406 last
= key
.objectid
+ key
.offset
;
408 if (total_found
> (1024 * 1024 * 2)) {
410 wake_up(&caching_ctl
->wait
);
417 total_found
+= add_new_free_space(block_group
, fs_info
, last
,
418 block_group
->key
.objectid
+
419 block_group
->key
.offset
);
420 caching_ctl
->progress
= (u64
)-1;
422 spin_lock(&block_group
->lock
);
423 block_group
->caching_ctl
= NULL
;
424 block_group
->cached
= BTRFS_CACHE_FINISHED
;
425 spin_unlock(&block_group
->lock
);
428 btrfs_free_path(path
);
429 up_read(&fs_info
->extent_commit_sem
);
431 free_excluded_extents(extent_root
, block_group
);
433 mutex_unlock(&caching_ctl
->mutex
);
434 wake_up(&caching_ctl
->wait
);
436 put_caching_control(caching_ctl
);
437 atomic_dec(&block_group
->space_info
->caching_threads
);
438 btrfs_put_block_group(block_group
);
443 static int cache_block_group(struct btrfs_block_group_cache
*cache
,
444 struct btrfs_trans_handle
*trans
,
445 struct btrfs_root
*root
,
448 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
449 struct btrfs_caching_control
*caching_ctl
;
450 struct task_struct
*tsk
;
454 if (cache
->cached
!= BTRFS_CACHE_NO
)
458 * We can't do the read from on-disk cache during a commit since we need
459 * to have the normal tree locking. Also if we are currently trying to
460 * allocate blocks for the tree root we can't do the fast caching since
461 * we likely hold important locks.
463 if (trans
&& (!trans
->transaction
->in_commit
) &&
464 (root
&& root
!= root
->fs_info
->tree_root
)) {
465 spin_lock(&cache
->lock
);
466 if (cache
->cached
!= BTRFS_CACHE_NO
) {
467 spin_unlock(&cache
->lock
);
470 cache
->cached
= BTRFS_CACHE_STARTED
;
471 spin_unlock(&cache
->lock
);
473 ret
= load_free_space_cache(fs_info
, cache
);
475 spin_lock(&cache
->lock
);
477 cache
->cached
= BTRFS_CACHE_FINISHED
;
478 cache
->last_byte_to_unpin
= (u64
)-1;
480 cache
->cached
= BTRFS_CACHE_NO
;
482 spin_unlock(&cache
->lock
);
484 free_excluded_extents(fs_info
->extent_root
, cache
);
492 caching_ctl
= kzalloc(sizeof(*caching_ctl
), GFP_NOFS
);
493 BUG_ON(!caching_ctl
);
495 INIT_LIST_HEAD(&caching_ctl
->list
);
496 mutex_init(&caching_ctl
->mutex
);
497 init_waitqueue_head(&caching_ctl
->wait
);
498 caching_ctl
->block_group
= cache
;
499 caching_ctl
->progress
= cache
->key
.objectid
;
500 /* one for caching kthread, one for caching block group list */
501 atomic_set(&caching_ctl
->count
, 2);
503 spin_lock(&cache
->lock
);
504 if (cache
->cached
!= BTRFS_CACHE_NO
) {
505 spin_unlock(&cache
->lock
);
509 cache
->caching_ctl
= caching_ctl
;
510 cache
->cached
= BTRFS_CACHE_STARTED
;
511 spin_unlock(&cache
->lock
);
513 down_write(&fs_info
->extent_commit_sem
);
514 list_add_tail(&caching_ctl
->list
, &fs_info
->caching_block_groups
);
515 up_write(&fs_info
->extent_commit_sem
);
517 atomic_inc(&cache
->space_info
->caching_threads
);
518 btrfs_get_block_group(cache
);
520 tsk
= kthread_run(caching_kthread
, cache
, "btrfs-cache-%llu\n",
521 cache
->key
.objectid
);
524 printk(KERN_ERR
"error running thread %d\n", ret
);
532 * return the block group that starts at or after bytenr
534 static struct btrfs_block_group_cache
*
535 btrfs_lookup_first_block_group(struct btrfs_fs_info
*info
, u64 bytenr
)
537 struct btrfs_block_group_cache
*cache
;
539 cache
= block_group_cache_tree_search(info
, bytenr
, 0);
545 * return the block group that contains the given bytenr
547 struct btrfs_block_group_cache
*btrfs_lookup_block_group(
548 struct btrfs_fs_info
*info
,
551 struct btrfs_block_group_cache
*cache
;
553 cache
= block_group_cache_tree_search(info
, bytenr
, 1);
558 static struct btrfs_space_info
*__find_space_info(struct btrfs_fs_info
*info
,
561 struct list_head
*head
= &info
->space_info
;
562 struct btrfs_space_info
*found
;
564 flags
&= BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_SYSTEM
|
565 BTRFS_BLOCK_GROUP_METADATA
;
568 list_for_each_entry_rcu(found
, head
, list
) {
569 if (found
->flags
& flags
) {
579 * after adding space to the filesystem, we need to clear the full flags
580 * on all the space infos.
582 void btrfs_clear_space_info_full(struct btrfs_fs_info
*info
)
584 struct list_head
*head
= &info
->space_info
;
585 struct btrfs_space_info
*found
;
588 list_for_each_entry_rcu(found
, head
, list
)
593 static u64
div_factor(u64 num
, int factor
)
602 static u64
div_factor_fine(u64 num
, int factor
)
611 u64
btrfs_find_block_group(struct btrfs_root
*root
,
612 u64 search_start
, u64 search_hint
, int owner
)
614 struct btrfs_block_group_cache
*cache
;
616 u64 last
= max(search_hint
, search_start
);
623 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
627 spin_lock(&cache
->lock
);
628 last
= cache
->key
.objectid
+ cache
->key
.offset
;
629 used
= btrfs_block_group_used(&cache
->item
);
631 if ((full_search
|| !cache
->ro
) &&
632 block_group_bits(cache
, BTRFS_BLOCK_GROUP_METADATA
)) {
633 if (used
+ cache
->pinned
+ cache
->reserved
<
634 div_factor(cache
->key
.offset
, factor
)) {
635 group_start
= cache
->key
.objectid
;
636 spin_unlock(&cache
->lock
);
637 btrfs_put_block_group(cache
);
641 spin_unlock(&cache
->lock
);
642 btrfs_put_block_group(cache
);
650 if (!full_search
&& factor
< 10) {
660 /* simple helper to search for an existing extent at a given offset */
661 int btrfs_lookup_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
664 struct btrfs_key key
;
665 struct btrfs_path
*path
;
667 path
= btrfs_alloc_path();
669 key
.objectid
= start
;
671 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
672 ret
= btrfs_search_slot(NULL
, root
->fs_info
->extent_root
, &key
, path
,
674 btrfs_free_path(path
);
679 * helper function to lookup reference count and flags of extent.
681 * the head node for delayed ref is used to store the sum of all the
682 * reference count modifications queued up in the rbtree. the head
683 * node may also store the extent flags to set. This way you can check
684 * to see what the reference count and extent flags would be if all of
685 * the delayed refs are not processed.
687 int btrfs_lookup_extent_info(struct btrfs_trans_handle
*trans
,
688 struct btrfs_root
*root
, u64 bytenr
,
689 u64 num_bytes
, u64
*refs
, u64
*flags
)
691 struct btrfs_delayed_ref_head
*head
;
692 struct btrfs_delayed_ref_root
*delayed_refs
;
693 struct btrfs_path
*path
;
694 struct btrfs_extent_item
*ei
;
695 struct extent_buffer
*leaf
;
696 struct btrfs_key key
;
702 path
= btrfs_alloc_path();
706 key
.objectid
= bytenr
;
707 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
708 key
.offset
= num_bytes
;
710 path
->skip_locking
= 1;
711 path
->search_commit_root
= 1;
714 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
,
720 leaf
= path
->nodes
[0];
721 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
722 if (item_size
>= sizeof(*ei
)) {
723 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
724 struct btrfs_extent_item
);
725 num_refs
= btrfs_extent_refs(leaf
, ei
);
726 extent_flags
= btrfs_extent_flags(leaf
, ei
);
728 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
729 struct btrfs_extent_item_v0
*ei0
;
730 BUG_ON(item_size
!= sizeof(*ei0
));
731 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
732 struct btrfs_extent_item_v0
);
733 num_refs
= btrfs_extent_refs_v0(leaf
, ei0
);
734 /* FIXME: this isn't correct for data */
735 extent_flags
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
740 BUG_ON(num_refs
== 0);
750 delayed_refs
= &trans
->transaction
->delayed_refs
;
751 spin_lock(&delayed_refs
->lock
);
752 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
754 if (!mutex_trylock(&head
->mutex
)) {
755 atomic_inc(&head
->node
.refs
);
756 spin_unlock(&delayed_refs
->lock
);
758 btrfs_release_path(root
->fs_info
->extent_root
, path
);
760 mutex_lock(&head
->mutex
);
761 mutex_unlock(&head
->mutex
);
762 btrfs_put_delayed_ref(&head
->node
);
765 if (head
->extent_op
&& head
->extent_op
->update_flags
)
766 extent_flags
|= head
->extent_op
->flags_to_set
;
768 BUG_ON(num_refs
== 0);
770 num_refs
+= head
->node
.ref_mod
;
771 mutex_unlock(&head
->mutex
);
773 spin_unlock(&delayed_refs
->lock
);
775 WARN_ON(num_refs
== 0);
779 *flags
= extent_flags
;
781 btrfs_free_path(path
);
786 * Back reference rules. Back refs have three main goals:
788 * 1) differentiate between all holders of references to an extent so that
789 * when a reference is dropped we can make sure it was a valid reference
790 * before freeing the extent.
792 * 2) Provide enough information to quickly find the holders of an extent
793 * if we notice a given block is corrupted or bad.
795 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
796 * maintenance. This is actually the same as #2, but with a slightly
797 * different use case.
799 * There are two kinds of back refs. The implicit back refs is optimized
800 * for pointers in non-shared tree blocks. For a given pointer in a block,
801 * back refs of this kind provide information about the block's owner tree
802 * and the pointer's key. These information allow us to find the block by
803 * b-tree searching. The full back refs is for pointers in tree blocks not
804 * referenced by their owner trees. The location of tree block is recorded
805 * in the back refs. Actually the full back refs is generic, and can be
806 * used in all cases the implicit back refs is used. The major shortcoming
807 * of the full back refs is its overhead. Every time a tree block gets
808 * COWed, we have to update back refs entry for all pointers in it.
810 * For a newly allocated tree block, we use implicit back refs for
811 * pointers in it. This means most tree related operations only involve
812 * implicit back refs. For a tree block created in old transaction, the
813 * only way to drop a reference to it is COW it. So we can detect the
814 * event that tree block loses its owner tree's reference and do the
815 * back refs conversion.
817 * When a tree block is COW'd through a tree, there are four cases:
819 * The reference count of the block is one and the tree is the block's
820 * owner tree. Nothing to do in this case.
822 * The reference count of the block is one and the tree is not the
823 * block's owner tree. In this case, full back refs is used for pointers
824 * in the block. Remove these full back refs, add implicit back refs for
825 * every pointers in the new block.
827 * The reference count of the block is greater than one and the tree is
828 * the block's owner tree. In this case, implicit back refs is used for
829 * pointers in the block. Add full back refs for every pointers in the
830 * block, increase lower level extents' reference counts. The original
831 * implicit back refs are entailed to the new block.
833 * The reference count of the block is greater than one and the tree is
834 * not the block's owner tree. Add implicit back refs for every pointer in
835 * the new block, increase lower level extents' reference count.
837 * Back Reference Key composing:
839 * The key objectid corresponds to the first byte in the extent,
840 * The key type is used to differentiate between types of back refs.
841 * There are different meanings of the key offset for different types
844 * File extents can be referenced by:
846 * - multiple snapshots, subvolumes, or different generations in one subvol
847 * - different files inside a single subvolume
848 * - different offsets inside a file (bookend extents in file.c)
850 * The extent ref structure for the implicit back refs has fields for:
852 * - Objectid of the subvolume root
853 * - objectid of the file holding the reference
854 * - original offset in the file
855 * - how many bookend extents
857 * The key offset for the implicit back refs is hash of the first
860 * The extent ref structure for the full back refs has field for:
862 * - number of pointers in the tree leaf
864 * The key offset for the implicit back refs is the first byte of
867 * When a file extent is allocated, The implicit back refs is used.
868 * the fields are filled in:
870 * (root_key.objectid, inode objectid, offset in file, 1)
872 * When a file extent is removed file truncation, we find the
873 * corresponding implicit back refs and check the following fields:
875 * (btrfs_header_owner(leaf), inode objectid, offset in file)
877 * Btree extents can be referenced by:
879 * - Different subvolumes
881 * Both the implicit back refs and the full back refs for tree blocks
882 * only consist of key. The key offset for the implicit back refs is
883 * objectid of block's owner tree. The key offset for the full back refs
884 * is the first byte of parent block.
886 * When implicit back refs is used, information about the lowest key and
887 * level of the tree block are required. These information are stored in
888 * tree block info structure.
891 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
892 static int convert_extent_item_v0(struct btrfs_trans_handle
*trans
,
893 struct btrfs_root
*root
,
894 struct btrfs_path
*path
,
895 u64 owner
, u32 extra_size
)
897 struct btrfs_extent_item
*item
;
898 struct btrfs_extent_item_v0
*ei0
;
899 struct btrfs_extent_ref_v0
*ref0
;
900 struct btrfs_tree_block_info
*bi
;
901 struct extent_buffer
*leaf
;
902 struct btrfs_key key
;
903 struct btrfs_key found_key
;
904 u32 new_size
= sizeof(*item
);
908 leaf
= path
->nodes
[0];
909 BUG_ON(btrfs_item_size_nr(leaf
, path
->slots
[0]) != sizeof(*ei0
));
911 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
912 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
913 struct btrfs_extent_item_v0
);
914 refs
= btrfs_extent_refs_v0(leaf
, ei0
);
916 if (owner
== (u64
)-1) {
918 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
919 ret
= btrfs_next_leaf(root
, path
);
923 leaf
= path
->nodes
[0];
925 btrfs_item_key_to_cpu(leaf
, &found_key
,
927 BUG_ON(key
.objectid
!= found_key
.objectid
);
928 if (found_key
.type
!= BTRFS_EXTENT_REF_V0_KEY
) {
932 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
933 struct btrfs_extent_ref_v0
);
934 owner
= btrfs_ref_objectid_v0(leaf
, ref0
);
938 btrfs_release_path(root
, path
);
940 if (owner
< BTRFS_FIRST_FREE_OBJECTID
)
941 new_size
+= sizeof(*bi
);
943 new_size
-= sizeof(*ei0
);
944 ret
= btrfs_search_slot(trans
, root
, &key
, path
,
945 new_size
+ extra_size
, 1);
950 ret
= btrfs_extend_item(trans
, root
, path
, new_size
);
953 leaf
= path
->nodes
[0];
954 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
955 btrfs_set_extent_refs(leaf
, item
, refs
);
956 /* FIXME: get real generation */
957 btrfs_set_extent_generation(leaf
, item
, 0);
958 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
959 btrfs_set_extent_flags(leaf
, item
,
960 BTRFS_EXTENT_FLAG_TREE_BLOCK
|
961 BTRFS_BLOCK_FLAG_FULL_BACKREF
);
962 bi
= (struct btrfs_tree_block_info
*)(item
+ 1);
963 /* FIXME: get first key of the block */
964 memset_extent_buffer(leaf
, 0, (unsigned long)bi
, sizeof(*bi
));
965 btrfs_set_tree_block_level(leaf
, bi
, (int)owner
);
967 btrfs_set_extent_flags(leaf
, item
, BTRFS_EXTENT_FLAG_DATA
);
969 btrfs_mark_buffer_dirty(leaf
);
974 static u64
hash_extent_data_ref(u64 root_objectid
, u64 owner
, u64 offset
)
976 u32 high_crc
= ~(u32
)0;
977 u32 low_crc
= ~(u32
)0;
980 lenum
= cpu_to_le64(root_objectid
);
981 high_crc
= crc32c(high_crc
, &lenum
, sizeof(lenum
));
982 lenum
= cpu_to_le64(owner
);
983 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
984 lenum
= cpu_to_le64(offset
);
985 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
987 return ((u64
)high_crc
<< 31) ^ (u64
)low_crc
;
990 static u64
hash_extent_data_ref_item(struct extent_buffer
*leaf
,
991 struct btrfs_extent_data_ref
*ref
)
993 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf
, ref
),
994 btrfs_extent_data_ref_objectid(leaf
, ref
),
995 btrfs_extent_data_ref_offset(leaf
, ref
));
998 static int match_extent_data_ref(struct extent_buffer
*leaf
,
999 struct btrfs_extent_data_ref
*ref
,
1000 u64 root_objectid
, u64 owner
, u64 offset
)
1002 if (btrfs_extent_data_ref_root(leaf
, ref
) != root_objectid
||
1003 btrfs_extent_data_ref_objectid(leaf
, ref
) != owner
||
1004 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
1009 static noinline
int lookup_extent_data_ref(struct btrfs_trans_handle
*trans
,
1010 struct btrfs_root
*root
,
1011 struct btrfs_path
*path
,
1012 u64 bytenr
, u64 parent
,
1014 u64 owner
, u64 offset
)
1016 struct btrfs_key key
;
1017 struct btrfs_extent_data_ref
*ref
;
1018 struct extent_buffer
*leaf
;
1024 key
.objectid
= bytenr
;
1026 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1027 key
.offset
= parent
;
1029 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1030 key
.offset
= hash_extent_data_ref(root_objectid
,
1035 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1044 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1045 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1046 btrfs_release_path(root
, path
);
1047 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1058 leaf
= path
->nodes
[0];
1059 nritems
= btrfs_header_nritems(leaf
);
1061 if (path
->slots
[0] >= nritems
) {
1062 ret
= btrfs_next_leaf(root
, path
);
1068 leaf
= path
->nodes
[0];
1069 nritems
= btrfs_header_nritems(leaf
);
1073 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1074 if (key
.objectid
!= bytenr
||
1075 key
.type
!= BTRFS_EXTENT_DATA_REF_KEY
)
1078 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1079 struct btrfs_extent_data_ref
);
1081 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1084 btrfs_release_path(root
, path
);
1096 static noinline
int insert_extent_data_ref(struct btrfs_trans_handle
*trans
,
1097 struct btrfs_root
*root
,
1098 struct btrfs_path
*path
,
1099 u64 bytenr
, u64 parent
,
1100 u64 root_objectid
, u64 owner
,
1101 u64 offset
, int refs_to_add
)
1103 struct btrfs_key key
;
1104 struct extent_buffer
*leaf
;
1109 key
.objectid
= bytenr
;
1111 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1112 key
.offset
= parent
;
1113 size
= sizeof(struct btrfs_shared_data_ref
);
1115 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1116 key
.offset
= hash_extent_data_ref(root_objectid
,
1118 size
= sizeof(struct btrfs_extent_data_ref
);
1121 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, size
);
1122 if (ret
&& ret
!= -EEXIST
)
1125 leaf
= path
->nodes
[0];
1127 struct btrfs_shared_data_ref
*ref
;
1128 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1129 struct btrfs_shared_data_ref
);
1131 btrfs_set_shared_data_ref_count(leaf
, ref
, refs_to_add
);
1133 num_refs
= btrfs_shared_data_ref_count(leaf
, ref
);
1134 num_refs
+= refs_to_add
;
1135 btrfs_set_shared_data_ref_count(leaf
, ref
, num_refs
);
1138 struct btrfs_extent_data_ref
*ref
;
1139 while (ret
== -EEXIST
) {
1140 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1141 struct btrfs_extent_data_ref
);
1142 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1145 btrfs_release_path(root
, path
);
1147 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1149 if (ret
&& ret
!= -EEXIST
)
1152 leaf
= path
->nodes
[0];
1154 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1155 struct btrfs_extent_data_ref
);
1157 btrfs_set_extent_data_ref_root(leaf
, ref
,
1159 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
1160 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
1161 btrfs_set_extent_data_ref_count(leaf
, ref
, refs_to_add
);
1163 num_refs
= btrfs_extent_data_ref_count(leaf
, ref
);
1164 num_refs
+= refs_to_add
;
1165 btrfs_set_extent_data_ref_count(leaf
, ref
, num_refs
);
1168 btrfs_mark_buffer_dirty(leaf
);
1171 btrfs_release_path(root
, path
);
1175 static noinline
int remove_extent_data_ref(struct btrfs_trans_handle
*trans
,
1176 struct btrfs_root
*root
,
1177 struct btrfs_path
*path
,
1180 struct btrfs_key key
;
1181 struct btrfs_extent_data_ref
*ref1
= NULL
;
1182 struct btrfs_shared_data_ref
*ref2
= NULL
;
1183 struct extent_buffer
*leaf
;
1187 leaf
= path
->nodes
[0];
1188 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1190 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1191 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1192 struct btrfs_extent_data_ref
);
1193 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1194 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1195 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1196 struct btrfs_shared_data_ref
);
1197 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1198 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1199 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1200 struct btrfs_extent_ref_v0
*ref0
;
1201 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1202 struct btrfs_extent_ref_v0
);
1203 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1209 BUG_ON(num_refs
< refs_to_drop
);
1210 num_refs
-= refs_to_drop
;
1212 if (num_refs
== 0) {
1213 ret
= btrfs_del_item(trans
, root
, path
);
1215 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
)
1216 btrfs_set_extent_data_ref_count(leaf
, ref1
, num_refs
);
1217 else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
)
1218 btrfs_set_shared_data_ref_count(leaf
, ref2
, num_refs
);
1219 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1221 struct btrfs_extent_ref_v0
*ref0
;
1222 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1223 struct btrfs_extent_ref_v0
);
1224 btrfs_set_ref_count_v0(leaf
, ref0
, num_refs
);
1227 btrfs_mark_buffer_dirty(leaf
);
1232 static noinline u32
extent_data_ref_count(struct btrfs_root
*root
,
1233 struct btrfs_path
*path
,
1234 struct btrfs_extent_inline_ref
*iref
)
1236 struct btrfs_key key
;
1237 struct extent_buffer
*leaf
;
1238 struct btrfs_extent_data_ref
*ref1
;
1239 struct btrfs_shared_data_ref
*ref2
;
1242 leaf
= path
->nodes
[0];
1243 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1245 if (btrfs_extent_inline_ref_type(leaf
, iref
) ==
1246 BTRFS_EXTENT_DATA_REF_KEY
) {
1247 ref1
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1248 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1250 ref2
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1251 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1253 } else if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1254 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1255 struct btrfs_extent_data_ref
);
1256 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1257 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1258 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1259 struct btrfs_shared_data_ref
);
1260 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1261 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1262 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1263 struct btrfs_extent_ref_v0
*ref0
;
1264 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1265 struct btrfs_extent_ref_v0
);
1266 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1274 static noinline
int lookup_tree_block_ref(struct btrfs_trans_handle
*trans
,
1275 struct btrfs_root
*root
,
1276 struct btrfs_path
*path
,
1277 u64 bytenr
, u64 parent
,
1280 struct btrfs_key key
;
1283 key
.objectid
= bytenr
;
1285 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1286 key
.offset
= parent
;
1288 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1289 key
.offset
= root_objectid
;
1292 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1295 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1296 if (ret
== -ENOENT
&& parent
) {
1297 btrfs_release_path(root
, path
);
1298 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1299 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1307 static noinline
int insert_tree_block_ref(struct btrfs_trans_handle
*trans
,
1308 struct btrfs_root
*root
,
1309 struct btrfs_path
*path
,
1310 u64 bytenr
, u64 parent
,
1313 struct btrfs_key key
;
1316 key
.objectid
= bytenr
;
1318 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1319 key
.offset
= parent
;
1321 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1322 key
.offset
= root_objectid
;
1325 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
1326 btrfs_release_path(root
, path
);
1330 static inline int extent_ref_type(u64 parent
, u64 owner
)
1333 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1335 type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1337 type
= BTRFS_TREE_BLOCK_REF_KEY
;
1340 type
= BTRFS_SHARED_DATA_REF_KEY
;
1342 type
= BTRFS_EXTENT_DATA_REF_KEY
;
1347 static int find_next_key(struct btrfs_path
*path
, int level
,
1348 struct btrfs_key
*key
)
1351 for (; level
< BTRFS_MAX_LEVEL
; level
++) {
1352 if (!path
->nodes
[level
])
1354 if (path
->slots
[level
] + 1 >=
1355 btrfs_header_nritems(path
->nodes
[level
]))
1358 btrfs_item_key_to_cpu(path
->nodes
[level
], key
,
1359 path
->slots
[level
] + 1);
1361 btrfs_node_key_to_cpu(path
->nodes
[level
], key
,
1362 path
->slots
[level
] + 1);
1369 * look for inline back ref. if back ref is found, *ref_ret is set
1370 * to the address of inline back ref, and 0 is returned.
1372 * if back ref isn't found, *ref_ret is set to the address where it
1373 * should be inserted, and -ENOENT is returned.
1375 * if insert is true and there are too many inline back refs, the path
1376 * points to the extent item, and -EAGAIN is returned.
1378 * NOTE: inline back refs are ordered in the same way that back ref
1379 * items in the tree are ordered.
1381 static noinline_for_stack
1382 int lookup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1383 struct btrfs_root
*root
,
1384 struct btrfs_path
*path
,
1385 struct btrfs_extent_inline_ref
**ref_ret
,
1386 u64 bytenr
, u64 num_bytes
,
1387 u64 parent
, u64 root_objectid
,
1388 u64 owner
, u64 offset
, int insert
)
1390 struct btrfs_key key
;
1391 struct extent_buffer
*leaf
;
1392 struct btrfs_extent_item
*ei
;
1393 struct btrfs_extent_inline_ref
*iref
;
1404 key
.objectid
= bytenr
;
1405 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1406 key
.offset
= num_bytes
;
1408 want
= extent_ref_type(parent
, owner
);
1410 extra_size
= btrfs_extent_inline_ref_size(want
);
1411 path
->keep_locks
= 1;
1414 ret
= btrfs_search_slot(trans
, root
, &key
, path
, extra_size
, 1);
1421 leaf
= path
->nodes
[0];
1422 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1423 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1424 if (item_size
< sizeof(*ei
)) {
1429 ret
= convert_extent_item_v0(trans
, root
, path
, owner
,
1435 leaf
= path
->nodes
[0];
1436 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1439 BUG_ON(item_size
< sizeof(*ei
));
1441 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1442 flags
= btrfs_extent_flags(leaf
, ei
);
1444 ptr
= (unsigned long)(ei
+ 1);
1445 end
= (unsigned long)ei
+ item_size
;
1447 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
) {
1448 ptr
+= sizeof(struct btrfs_tree_block_info
);
1451 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_DATA
));
1460 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1461 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1465 ptr
+= btrfs_extent_inline_ref_size(type
);
1469 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1470 struct btrfs_extent_data_ref
*dref
;
1471 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1472 if (match_extent_data_ref(leaf
, dref
, root_objectid
,
1477 if (hash_extent_data_ref_item(leaf
, dref
) <
1478 hash_extent_data_ref(root_objectid
, owner
, offset
))
1482 ref_offset
= btrfs_extent_inline_ref_offset(leaf
, iref
);
1484 if (parent
== ref_offset
) {
1488 if (ref_offset
< parent
)
1491 if (root_objectid
== ref_offset
) {
1495 if (ref_offset
< root_objectid
)
1499 ptr
+= btrfs_extent_inline_ref_size(type
);
1501 if (err
== -ENOENT
&& insert
) {
1502 if (item_size
+ extra_size
>=
1503 BTRFS_MAX_EXTENT_ITEM_SIZE(root
)) {
1508 * To add new inline back ref, we have to make sure
1509 * there is no corresponding back ref item.
1510 * For simplicity, we just do not add new inline back
1511 * ref if there is any kind of item for this block
1513 if (find_next_key(path
, 0, &key
) == 0 &&
1514 key
.objectid
== bytenr
&&
1515 key
.type
< BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1520 *ref_ret
= (struct btrfs_extent_inline_ref
*)ptr
;
1523 path
->keep_locks
= 0;
1524 btrfs_unlock_up_safe(path
, 1);
1530 * helper to add new inline back ref
1532 static noinline_for_stack
1533 int setup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1534 struct btrfs_root
*root
,
1535 struct btrfs_path
*path
,
1536 struct btrfs_extent_inline_ref
*iref
,
1537 u64 parent
, u64 root_objectid
,
1538 u64 owner
, u64 offset
, int refs_to_add
,
1539 struct btrfs_delayed_extent_op
*extent_op
)
1541 struct extent_buffer
*leaf
;
1542 struct btrfs_extent_item
*ei
;
1545 unsigned long item_offset
;
1551 leaf
= path
->nodes
[0];
1552 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1553 item_offset
= (unsigned long)iref
- (unsigned long)ei
;
1555 type
= extent_ref_type(parent
, owner
);
1556 size
= btrfs_extent_inline_ref_size(type
);
1558 ret
= btrfs_extend_item(trans
, root
, path
, size
);
1561 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1562 refs
= btrfs_extent_refs(leaf
, ei
);
1563 refs
+= refs_to_add
;
1564 btrfs_set_extent_refs(leaf
, ei
, refs
);
1566 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1568 ptr
= (unsigned long)ei
+ item_offset
;
1569 end
= (unsigned long)ei
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
1570 if (ptr
< end
- size
)
1571 memmove_extent_buffer(leaf
, ptr
+ size
, ptr
,
1574 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1575 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
1576 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1577 struct btrfs_extent_data_ref
*dref
;
1578 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1579 btrfs_set_extent_data_ref_root(leaf
, dref
, root_objectid
);
1580 btrfs_set_extent_data_ref_objectid(leaf
, dref
, owner
);
1581 btrfs_set_extent_data_ref_offset(leaf
, dref
, offset
);
1582 btrfs_set_extent_data_ref_count(leaf
, dref
, refs_to_add
);
1583 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1584 struct btrfs_shared_data_ref
*sref
;
1585 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1586 btrfs_set_shared_data_ref_count(leaf
, sref
, refs_to_add
);
1587 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1588 } else if (type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
1589 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1591 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
1593 btrfs_mark_buffer_dirty(leaf
);
1597 static int lookup_extent_backref(struct btrfs_trans_handle
*trans
,
1598 struct btrfs_root
*root
,
1599 struct btrfs_path
*path
,
1600 struct btrfs_extent_inline_ref
**ref_ret
,
1601 u64 bytenr
, u64 num_bytes
, u64 parent
,
1602 u64 root_objectid
, u64 owner
, u64 offset
)
1606 ret
= lookup_inline_extent_backref(trans
, root
, path
, ref_ret
,
1607 bytenr
, num_bytes
, parent
,
1608 root_objectid
, owner
, offset
, 0);
1612 btrfs_release_path(root
, path
);
1615 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1616 ret
= lookup_tree_block_ref(trans
, root
, path
, bytenr
, parent
,
1619 ret
= lookup_extent_data_ref(trans
, root
, path
, bytenr
, parent
,
1620 root_objectid
, owner
, offset
);
1626 * helper to update/remove inline back ref
1628 static noinline_for_stack
1629 int update_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1630 struct btrfs_root
*root
,
1631 struct btrfs_path
*path
,
1632 struct btrfs_extent_inline_ref
*iref
,
1634 struct btrfs_delayed_extent_op
*extent_op
)
1636 struct extent_buffer
*leaf
;
1637 struct btrfs_extent_item
*ei
;
1638 struct btrfs_extent_data_ref
*dref
= NULL
;
1639 struct btrfs_shared_data_ref
*sref
= NULL
;
1648 leaf
= path
->nodes
[0];
1649 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1650 refs
= btrfs_extent_refs(leaf
, ei
);
1651 WARN_ON(refs_to_mod
< 0 && refs
+ refs_to_mod
<= 0);
1652 refs
+= refs_to_mod
;
1653 btrfs_set_extent_refs(leaf
, ei
, refs
);
1655 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1657 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1659 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1660 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1661 refs
= btrfs_extent_data_ref_count(leaf
, dref
);
1662 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1663 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1664 refs
= btrfs_shared_data_ref_count(leaf
, sref
);
1667 BUG_ON(refs_to_mod
!= -1);
1670 BUG_ON(refs_to_mod
< 0 && refs
< -refs_to_mod
);
1671 refs
+= refs_to_mod
;
1674 if (type
== BTRFS_EXTENT_DATA_REF_KEY
)
1675 btrfs_set_extent_data_ref_count(leaf
, dref
, refs
);
1677 btrfs_set_shared_data_ref_count(leaf
, sref
, refs
);
1679 size
= btrfs_extent_inline_ref_size(type
);
1680 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1681 ptr
= (unsigned long)iref
;
1682 end
= (unsigned long)ei
+ item_size
;
1683 if (ptr
+ size
< end
)
1684 memmove_extent_buffer(leaf
, ptr
, ptr
+ size
,
1687 ret
= btrfs_truncate_item(trans
, root
, path
, item_size
, 1);
1690 btrfs_mark_buffer_dirty(leaf
);
1694 static noinline_for_stack
1695 int insert_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1696 struct btrfs_root
*root
,
1697 struct btrfs_path
*path
,
1698 u64 bytenr
, u64 num_bytes
, u64 parent
,
1699 u64 root_objectid
, u64 owner
,
1700 u64 offset
, int refs_to_add
,
1701 struct btrfs_delayed_extent_op
*extent_op
)
1703 struct btrfs_extent_inline_ref
*iref
;
1706 ret
= lookup_inline_extent_backref(trans
, root
, path
, &iref
,
1707 bytenr
, num_bytes
, parent
,
1708 root_objectid
, owner
, offset
, 1);
1710 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
);
1711 ret
= update_inline_extent_backref(trans
, root
, path
, iref
,
1712 refs_to_add
, extent_op
);
1713 } else if (ret
== -ENOENT
) {
1714 ret
= setup_inline_extent_backref(trans
, root
, path
, iref
,
1715 parent
, root_objectid
,
1716 owner
, offset
, refs_to_add
,
1722 static int insert_extent_backref(struct btrfs_trans_handle
*trans
,
1723 struct btrfs_root
*root
,
1724 struct btrfs_path
*path
,
1725 u64 bytenr
, u64 parent
, u64 root_objectid
,
1726 u64 owner
, u64 offset
, int refs_to_add
)
1729 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1730 BUG_ON(refs_to_add
!= 1);
1731 ret
= insert_tree_block_ref(trans
, root
, path
, bytenr
,
1732 parent
, root_objectid
);
1734 ret
= insert_extent_data_ref(trans
, root
, path
, bytenr
,
1735 parent
, root_objectid
,
1736 owner
, offset
, refs_to_add
);
1741 static int remove_extent_backref(struct btrfs_trans_handle
*trans
,
1742 struct btrfs_root
*root
,
1743 struct btrfs_path
*path
,
1744 struct btrfs_extent_inline_ref
*iref
,
1745 int refs_to_drop
, int is_data
)
1749 BUG_ON(!is_data
&& refs_to_drop
!= 1);
1751 ret
= update_inline_extent_backref(trans
, root
, path
, iref
,
1752 -refs_to_drop
, NULL
);
1753 } else if (is_data
) {
1754 ret
= remove_extent_data_ref(trans
, root
, path
, refs_to_drop
);
1756 ret
= btrfs_del_item(trans
, root
, path
);
1761 static int btrfs_issue_discard(struct block_device
*bdev
,
1764 return blkdev_issue_discard(bdev
, start
>> 9, len
>> 9, GFP_NOFS
, 0);
1767 static int btrfs_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
1768 u64 num_bytes
, u64
*actual_bytes
)
1771 u64 discarded_bytes
= 0;
1772 struct btrfs_multi_bio
*multi
= NULL
;
1775 /* Tell the block device(s) that the sectors can be discarded */
1776 ret
= btrfs_map_block(&root
->fs_info
->mapping_tree
, REQ_DISCARD
,
1777 bytenr
, &num_bytes
, &multi
, 0);
1779 struct btrfs_bio_stripe
*stripe
= multi
->stripes
;
1783 for (i
= 0; i
< multi
->num_stripes
; i
++, stripe
++) {
1784 ret
= btrfs_issue_discard(stripe
->dev
->bdev
,
1788 discarded_bytes
+= stripe
->length
;
1789 else if (ret
!= -EOPNOTSUPP
)
1794 if (discarded_bytes
&& ret
== -EOPNOTSUPP
)
1798 *actual_bytes
= discarded_bytes
;
1804 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1805 struct btrfs_root
*root
,
1806 u64 bytenr
, u64 num_bytes
, u64 parent
,
1807 u64 root_objectid
, u64 owner
, u64 offset
)
1810 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
&&
1811 root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
1813 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1814 ret
= btrfs_add_delayed_tree_ref(trans
, bytenr
, num_bytes
,
1815 parent
, root_objectid
, (int)owner
,
1816 BTRFS_ADD_DELAYED_REF
, NULL
);
1818 ret
= btrfs_add_delayed_data_ref(trans
, bytenr
, num_bytes
,
1819 parent
, root_objectid
, owner
, offset
,
1820 BTRFS_ADD_DELAYED_REF
, NULL
);
1825 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1826 struct btrfs_root
*root
,
1827 u64 bytenr
, u64 num_bytes
,
1828 u64 parent
, u64 root_objectid
,
1829 u64 owner
, u64 offset
, int refs_to_add
,
1830 struct btrfs_delayed_extent_op
*extent_op
)
1832 struct btrfs_path
*path
;
1833 struct extent_buffer
*leaf
;
1834 struct btrfs_extent_item
*item
;
1839 path
= btrfs_alloc_path();
1844 path
->leave_spinning
= 1;
1845 /* this will setup the path even if it fails to insert the back ref */
1846 ret
= insert_inline_extent_backref(trans
, root
->fs_info
->extent_root
,
1847 path
, bytenr
, num_bytes
, parent
,
1848 root_objectid
, owner
, offset
,
1849 refs_to_add
, extent_op
);
1853 if (ret
!= -EAGAIN
) {
1858 leaf
= path
->nodes
[0];
1859 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1860 refs
= btrfs_extent_refs(leaf
, item
);
1861 btrfs_set_extent_refs(leaf
, item
, refs
+ refs_to_add
);
1863 __run_delayed_extent_op(extent_op
, leaf
, item
);
1865 btrfs_mark_buffer_dirty(leaf
);
1866 btrfs_release_path(root
->fs_info
->extent_root
, path
);
1869 path
->leave_spinning
= 1;
1871 /* now insert the actual backref */
1872 ret
= insert_extent_backref(trans
, root
->fs_info
->extent_root
,
1873 path
, bytenr
, parent
, root_objectid
,
1874 owner
, offset
, refs_to_add
);
1877 btrfs_free_path(path
);
1881 static int run_delayed_data_ref(struct btrfs_trans_handle
*trans
,
1882 struct btrfs_root
*root
,
1883 struct btrfs_delayed_ref_node
*node
,
1884 struct btrfs_delayed_extent_op
*extent_op
,
1885 int insert_reserved
)
1888 struct btrfs_delayed_data_ref
*ref
;
1889 struct btrfs_key ins
;
1894 ins
.objectid
= node
->bytenr
;
1895 ins
.offset
= node
->num_bytes
;
1896 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1898 ref
= btrfs_delayed_node_to_data_ref(node
);
1899 if (node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
1900 parent
= ref
->parent
;
1902 ref_root
= ref
->root
;
1904 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
1906 BUG_ON(extent_op
->update_key
);
1907 flags
|= extent_op
->flags_to_set
;
1909 ret
= alloc_reserved_file_extent(trans
, root
,
1910 parent
, ref_root
, flags
,
1911 ref
->objectid
, ref
->offset
,
1912 &ins
, node
->ref_mod
);
1913 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
1914 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
1915 node
->num_bytes
, parent
,
1916 ref_root
, ref
->objectid
,
1917 ref
->offset
, node
->ref_mod
,
1919 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
1920 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
1921 node
->num_bytes
, parent
,
1922 ref_root
, ref
->objectid
,
1923 ref
->offset
, node
->ref_mod
,
1931 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
1932 struct extent_buffer
*leaf
,
1933 struct btrfs_extent_item
*ei
)
1935 u64 flags
= btrfs_extent_flags(leaf
, ei
);
1936 if (extent_op
->update_flags
) {
1937 flags
|= extent_op
->flags_to_set
;
1938 btrfs_set_extent_flags(leaf
, ei
, flags
);
1941 if (extent_op
->update_key
) {
1942 struct btrfs_tree_block_info
*bi
;
1943 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
));
1944 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
1945 btrfs_set_tree_block_key(leaf
, bi
, &extent_op
->key
);
1949 static int run_delayed_extent_op(struct btrfs_trans_handle
*trans
,
1950 struct btrfs_root
*root
,
1951 struct btrfs_delayed_ref_node
*node
,
1952 struct btrfs_delayed_extent_op
*extent_op
)
1954 struct btrfs_key key
;
1955 struct btrfs_path
*path
;
1956 struct btrfs_extent_item
*ei
;
1957 struct extent_buffer
*leaf
;
1962 path
= btrfs_alloc_path();
1966 key
.objectid
= node
->bytenr
;
1967 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1968 key
.offset
= node
->num_bytes
;
1971 path
->leave_spinning
= 1;
1972 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
,
1983 leaf
= path
->nodes
[0];
1984 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1985 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1986 if (item_size
< sizeof(*ei
)) {
1987 ret
= convert_extent_item_v0(trans
, root
->fs_info
->extent_root
,
1993 leaf
= path
->nodes
[0];
1994 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1997 BUG_ON(item_size
< sizeof(*ei
));
1998 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1999 __run_delayed_extent_op(extent_op
, leaf
, ei
);
2001 btrfs_mark_buffer_dirty(leaf
);
2003 btrfs_free_path(path
);
2007 static int run_delayed_tree_ref(struct btrfs_trans_handle
*trans
,
2008 struct btrfs_root
*root
,
2009 struct btrfs_delayed_ref_node
*node
,
2010 struct btrfs_delayed_extent_op
*extent_op
,
2011 int insert_reserved
)
2014 struct btrfs_delayed_tree_ref
*ref
;
2015 struct btrfs_key ins
;
2019 ins
.objectid
= node
->bytenr
;
2020 ins
.offset
= node
->num_bytes
;
2021 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2023 ref
= btrfs_delayed_node_to_tree_ref(node
);
2024 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2025 parent
= ref
->parent
;
2027 ref_root
= ref
->root
;
2029 BUG_ON(node
->ref_mod
!= 1);
2030 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2031 BUG_ON(!extent_op
|| !extent_op
->update_flags
||
2032 !extent_op
->update_key
);
2033 ret
= alloc_reserved_tree_block(trans
, root
,
2035 extent_op
->flags_to_set
,
2038 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2039 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
2040 node
->num_bytes
, parent
, ref_root
,
2041 ref
->level
, 0, 1, extent_op
);
2042 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2043 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
2044 node
->num_bytes
, parent
, ref_root
,
2045 ref
->level
, 0, 1, extent_op
);
2052 /* helper function to actually process a single delayed ref entry */
2053 static int run_one_delayed_ref(struct btrfs_trans_handle
*trans
,
2054 struct btrfs_root
*root
,
2055 struct btrfs_delayed_ref_node
*node
,
2056 struct btrfs_delayed_extent_op
*extent_op
,
2057 int insert_reserved
)
2060 if (btrfs_delayed_ref_is_head(node
)) {
2061 struct btrfs_delayed_ref_head
*head
;
2063 * we've hit the end of the chain and we were supposed
2064 * to insert this extent into the tree. But, it got
2065 * deleted before we ever needed to insert it, so all
2066 * we have to do is clean up the accounting
2069 head
= btrfs_delayed_node_to_head(node
);
2070 if (insert_reserved
) {
2071 btrfs_pin_extent(root
, node
->bytenr
,
2072 node
->num_bytes
, 1);
2073 if (head
->is_data
) {
2074 ret
= btrfs_del_csums(trans
, root
,
2080 mutex_unlock(&head
->mutex
);
2084 if (node
->type
== BTRFS_TREE_BLOCK_REF_KEY
||
2085 node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2086 ret
= run_delayed_tree_ref(trans
, root
, node
, extent_op
,
2088 else if (node
->type
== BTRFS_EXTENT_DATA_REF_KEY
||
2089 node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2090 ret
= run_delayed_data_ref(trans
, root
, node
, extent_op
,
2097 static noinline
struct btrfs_delayed_ref_node
*
2098 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
2100 struct rb_node
*node
;
2101 struct btrfs_delayed_ref_node
*ref
;
2102 int action
= BTRFS_ADD_DELAYED_REF
;
2105 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2106 * this prevents ref count from going down to zero when
2107 * there still are pending delayed ref.
2109 node
= rb_prev(&head
->node
.rb_node
);
2113 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
2115 if (ref
->bytenr
!= head
->node
.bytenr
)
2117 if (ref
->action
== action
)
2119 node
= rb_prev(node
);
2121 if (action
== BTRFS_ADD_DELAYED_REF
) {
2122 action
= BTRFS_DROP_DELAYED_REF
;
2128 static noinline
int run_clustered_refs(struct btrfs_trans_handle
*trans
,
2129 struct btrfs_root
*root
,
2130 struct list_head
*cluster
)
2132 struct btrfs_delayed_ref_root
*delayed_refs
;
2133 struct btrfs_delayed_ref_node
*ref
;
2134 struct btrfs_delayed_ref_head
*locked_ref
= NULL
;
2135 struct btrfs_delayed_extent_op
*extent_op
;
2138 int must_insert_reserved
= 0;
2140 delayed_refs
= &trans
->transaction
->delayed_refs
;
2143 /* pick a new head ref from the cluster list */
2144 if (list_empty(cluster
))
2147 locked_ref
= list_entry(cluster
->next
,
2148 struct btrfs_delayed_ref_head
, cluster
);
2150 /* grab the lock that says we are going to process
2151 * all the refs for this head */
2152 ret
= btrfs_delayed_ref_lock(trans
, locked_ref
);
2155 * we may have dropped the spin lock to get the head
2156 * mutex lock, and that might have given someone else
2157 * time to free the head. If that's true, it has been
2158 * removed from our list and we can move on.
2160 if (ret
== -EAGAIN
) {
2168 * record the must insert reserved flag before we
2169 * drop the spin lock.
2171 must_insert_reserved
= locked_ref
->must_insert_reserved
;
2172 locked_ref
->must_insert_reserved
= 0;
2174 extent_op
= locked_ref
->extent_op
;
2175 locked_ref
->extent_op
= NULL
;
2178 * locked_ref is the head node, so we have to go one
2179 * node back for any delayed ref updates
2181 ref
= select_delayed_ref(locked_ref
);
2183 /* All delayed refs have been processed, Go ahead
2184 * and send the head node to run_one_delayed_ref,
2185 * so that any accounting fixes can happen
2187 ref
= &locked_ref
->node
;
2189 if (extent_op
&& must_insert_reserved
) {
2195 spin_unlock(&delayed_refs
->lock
);
2197 ret
= run_delayed_extent_op(trans
, root
,
2203 spin_lock(&delayed_refs
->lock
);
2207 list_del_init(&locked_ref
->cluster
);
2212 rb_erase(&ref
->rb_node
, &delayed_refs
->root
);
2213 delayed_refs
->num_entries
--;
2215 spin_unlock(&delayed_refs
->lock
);
2217 ret
= run_one_delayed_ref(trans
, root
, ref
, extent_op
,
2218 must_insert_reserved
);
2221 btrfs_put_delayed_ref(ref
);
2226 spin_lock(&delayed_refs
->lock
);
2232 * this starts processing the delayed reference count updates and
2233 * extent insertions we have queued up so far. count can be
2234 * 0, which means to process everything in the tree at the start
2235 * of the run (but not newly added entries), or it can be some target
2236 * number you'd like to process.
2238 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2239 struct btrfs_root
*root
, unsigned long count
)
2241 struct rb_node
*node
;
2242 struct btrfs_delayed_ref_root
*delayed_refs
;
2243 struct btrfs_delayed_ref_node
*ref
;
2244 struct list_head cluster
;
2246 int run_all
= count
== (unsigned long)-1;
2249 if (root
== root
->fs_info
->extent_root
)
2250 root
= root
->fs_info
->tree_root
;
2252 delayed_refs
= &trans
->transaction
->delayed_refs
;
2253 INIT_LIST_HEAD(&cluster
);
2255 spin_lock(&delayed_refs
->lock
);
2257 count
= delayed_refs
->num_entries
* 2;
2261 if (!(run_all
|| run_most
) &&
2262 delayed_refs
->num_heads_ready
< 64)
2266 * go find something we can process in the rbtree. We start at
2267 * the beginning of the tree, and then build a cluster
2268 * of refs to process starting at the first one we are able to
2271 ret
= btrfs_find_ref_cluster(trans
, &cluster
,
2272 delayed_refs
->run_delayed_start
);
2276 ret
= run_clustered_refs(trans
, root
, &cluster
);
2279 count
-= min_t(unsigned long, ret
, count
);
2286 node
= rb_first(&delayed_refs
->root
);
2289 count
= (unsigned long)-1;
2292 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
2294 if (btrfs_delayed_ref_is_head(ref
)) {
2295 struct btrfs_delayed_ref_head
*head
;
2297 head
= btrfs_delayed_node_to_head(ref
);
2298 atomic_inc(&ref
->refs
);
2300 spin_unlock(&delayed_refs
->lock
);
2301 mutex_lock(&head
->mutex
);
2302 mutex_unlock(&head
->mutex
);
2304 btrfs_put_delayed_ref(ref
);
2308 node
= rb_next(node
);
2310 spin_unlock(&delayed_refs
->lock
);
2311 schedule_timeout(1);
2315 spin_unlock(&delayed_refs
->lock
);
2319 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
2320 struct btrfs_root
*root
,
2321 u64 bytenr
, u64 num_bytes
, u64 flags
,
2324 struct btrfs_delayed_extent_op
*extent_op
;
2327 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
2331 extent_op
->flags_to_set
= flags
;
2332 extent_op
->update_flags
= 1;
2333 extent_op
->update_key
= 0;
2334 extent_op
->is_data
= is_data
? 1 : 0;
2336 ret
= btrfs_add_delayed_extent_op(trans
, bytenr
, num_bytes
, extent_op
);
2342 static noinline
int check_delayed_ref(struct btrfs_trans_handle
*trans
,
2343 struct btrfs_root
*root
,
2344 struct btrfs_path
*path
,
2345 u64 objectid
, u64 offset
, u64 bytenr
)
2347 struct btrfs_delayed_ref_head
*head
;
2348 struct btrfs_delayed_ref_node
*ref
;
2349 struct btrfs_delayed_data_ref
*data_ref
;
2350 struct btrfs_delayed_ref_root
*delayed_refs
;
2351 struct rb_node
*node
;
2355 delayed_refs
= &trans
->transaction
->delayed_refs
;
2356 spin_lock(&delayed_refs
->lock
);
2357 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
2361 if (!mutex_trylock(&head
->mutex
)) {
2362 atomic_inc(&head
->node
.refs
);
2363 spin_unlock(&delayed_refs
->lock
);
2365 btrfs_release_path(root
->fs_info
->extent_root
, path
);
2367 mutex_lock(&head
->mutex
);
2368 mutex_unlock(&head
->mutex
);
2369 btrfs_put_delayed_ref(&head
->node
);
2373 node
= rb_prev(&head
->node
.rb_node
);
2377 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2379 if (ref
->bytenr
!= bytenr
)
2383 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
)
2386 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
2388 node
= rb_prev(node
);
2390 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2391 if (ref
->bytenr
== bytenr
)
2395 if (data_ref
->root
!= root
->root_key
.objectid
||
2396 data_ref
->objectid
!= objectid
|| data_ref
->offset
!= offset
)
2401 mutex_unlock(&head
->mutex
);
2403 spin_unlock(&delayed_refs
->lock
);
2407 static noinline
int check_committed_ref(struct btrfs_trans_handle
*trans
,
2408 struct btrfs_root
*root
,
2409 struct btrfs_path
*path
,
2410 u64 objectid
, u64 offset
, u64 bytenr
)
2412 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2413 struct extent_buffer
*leaf
;
2414 struct btrfs_extent_data_ref
*ref
;
2415 struct btrfs_extent_inline_ref
*iref
;
2416 struct btrfs_extent_item
*ei
;
2417 struct btrfs_key key
;
2421 key
.objectid
= bytenr
;
2422 key
.offset
= (u64
)-1;
2423 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2425 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
2431 if (path
->slots
[0] == 0)
2435 leaf
= path
->nodes
[0];
2436 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2438 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
2442 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2443 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2444 if (item_size
< sizeof(*ei
)) {
2445 WARN_ON(item_size
!= sizeof(struct btrfs_extent_item_v0
));
2449 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2451 if (item_size
!= sizeof(*ei
) +
2452 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
2455 if (btrfs_extent_generation(leaf
, ei
) <=
2456 btrfs_root_last_snapshot(&root
->root_item
))
2459 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
2460 if (btrfs_extent_inline_ref_type(leaf
, iref
) !=
2461 BTRFS_EXTENT_DATA_REF_KEY
)
2464 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
2465 if (btrfs_extent_refs(leaf
, ei
) !=
2466 btrfs_extent_data_ref_count(leaf
, ref
) ||
2467 btrfs_extent_data_ref_root(leaf
, ref
) !=
2468 root
->root_key
.objectid
||
2469 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
2470 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
2478 int btrfs_cross_ref_exist(struct btrfs_trans_handle
*trans
,
2479 struct btrfs_root
*root
,
2480 u64 objectid
, u64 offset
, u64 bytenr
)
2482 struct btrfs_path
*path
;
2486 path
= btrfs_alloc_path();
2491 ret
= check_committed_ref(trans
, root
, path
, objectid
,
2493 if (ret
&& ret
!= -ENOENT
)
2496 ret2
= check_delayed_ref(trans
, root
, path
, objectid
,
2498 } while (ret2
== -EAGAIN
);
2500 if (ret2
&& ret2
!= -ENOENT
) {
2505 if (ret
!= -ENOENT
|| ret2
!= -ENOENT
)
2508 btrfs_free_path(path
);
2509 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
2515 int btrfs_cache_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2516 struct extent_buffer
*buf
, u32 nr_extents
)
2518 struct btrfs_key key
;
2519 struct btrfs_file_extent_item
*fi
;
2527 if (!root
->ref_cows
)
2530 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
2532 root_gen
= root
->root_key
.offset
;
2535 root_gen
= trans
->transid
- 1;
2538 level
= btrfs_header_level(buf
);
2539 nritems
= btrfs_header_nritems(buf
);
2542 struct btrfs_leaf_ref
*ref
;
2543 struct btrfs_extent_info
*info
;
2545 ref
= btrfs_alloc_leaf_ref(root
, nr_extents
);
2551 ref
->root_gen
= root_gen
;
2552 ref
->bytenr
= buf
->start
;
2553 ref
->owner
= btrfs_header_owner(buf
);
2554 ref
->generation
= btrfs_header_generation(buf
);
2555 ref
->nritems
= nr_extents
;
2556 info
= ref
->extents
;
2558 for (i
= 0; nr_extents
> 0 && i
< nritems
; i
++) {
2560 btrfs_item_key_to_cpu(buf
, &key
, i
);
2561 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
2563 fi
= btrfs_item_ptr(buf
, i
,
2564 struct btrfs_file_extent_item
);
2565 if (btrfs_file_extent_type(buf
, fi
) ==
2566 BTRFS_FILE_EXTENT_INLINE
)
2568 disk_bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
2569 if (disk_bytenr
== 0)
2572 info
->bytenr
= disk_bytenr
;
2574 btrfs_file_extent_disk_num_bytes(buf
, fi
);
2575 info
->objectid
= key
.objectid
;
2576 info
->offset
= key
.offset
;
2580 ret
= btrfs_add_leaf_ref(root
, ref
, shared
);
2581 if (ret
== -EEXIST
&& shared
) {
2582 struct btrfs_leaf_ref
*old
;
2583 old
= btrfs_lookup_leaf_ref(root
, ref
->bytenr
);
2585 btrfs_remove_leaf_ref(root
, old
);
2586 btrfs_free_leaf_ref(root
, old
);
2587 ret
= btrfs_add_leaf_ref(root
, ref
, shared
);
2590 btrfs_free_leaf_ref(root
, ref
);
2596 /* when a block goes through cow, we update the reference counts of
2597 * everything that block points to. The internal pointers of the block
2598 * can be in just about any order, and it is likely to have clusters of
2599 * things that are close together and clusters of things that are not.
2601 * To help reduce the seeks that come with updating all of these reference
2602 * counts, sort them by byte number before actual updates are done.
2604 * struct refsort is used to match byte number to slot in the btree block.
2605 * we sort based on the byte number and then use the slot to actually
2608 * struct refsort is smaller than strcut btrfs_item and smaller than
2609 * struct btrfs_key_ptr. Since we're currently limited to the page size
2610 * for a btree block, there's no way for a kmalloc of refsorts for a
2611 * single node to be bigger than a page.
2619 * for passing into sort()
2621 static int refsort_cmp(const void *a_void
, const void *b_void
)
2623 const struct refsort
*a
= a_void
;
2624 const struct refsort
*b
= b_void
;
2626 if (a
->bytenr
< b
->bytenr
)
2628 if (a
->bytenr
> b
->bytenr
)
2634 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
2635 struct btrfs_root
*root
,
2636 struct extent_buffer
*buf
,
2637 int full_backref
, int inc
)
2644 struct btrfs_key key
;
2645 struct btrfs_file_extent_item
*fi
;
2649 int (*process_func
)(struct btrfs_trans_handle
*, struct btrfs_root
*,
2650 u64
, u64
, u64
, u64
, u64
, u64
);
2652 ref_root
= btrfs_header_owner(buf
);
2653 nritems
= btrfs_header_nritems(buf
);
2654 level
= btrfs_header_level(buf
);
2656 if (!root
->ref_cows
&& level
== 0)
2660 process_func
= btrfs_inc_extent_ref
;
2662 process_func
= btrfs_free_extent
;
2665 parent
= buf
->start
;
2669 for (i
= 0; i
< nritems
; i
++) {
2671 btrfs_item_key_to_cpu(buf
, &key
, i
);
2672 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
2674 fi
= btrfs_item_ptr(buf
, i
,
2675 struct btrfs_file_extent_item
);
2676 if (btrfs_file_extent_type(buf
, fi
) ==
2677 BTRFS_FILE_EXTENT_INLINE
)
2679 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
2683 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
2684 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
2685 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
2686 parent
, ref_root
, key
.objectid
,
2691 bytenr
= btrfs_node_blockptr(buf
, i
);
2692 num_bytes
= btrfs_level_size(root
, level
- 1);
2693 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
2694 parent
, ref_root
, level
- 1, 0);
2705 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2706 struct extent_buffer
*buf
, int full_backref
)
2708 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1);
2711 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2712 struct extent_buffer
*buf
, int full_backref
)
2714 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0);
2717 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
2718 struct btrfs_root
*root
,
2719 struct btrfs_path
*path
,
2720 struct btrfs_block_group_cache
*cache
)
2723 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2725 struct extent_buffer
*leaf
;
2727 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
2732 leaf
= path
->nodes
[0];
2733 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
2734 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
2735 btrfs_mark_buffer_dirty(leaf
);
2736 btrfs_release_path(extent_root
, path
);
2744 static struct btrfs_block_group_cache
*
2745 next_block_group(struct btrfs_root
*root
,
2746 struct btrfs_block_group_cache
*cache
)
2748 struct rb_node
*node
;
2749 spin_lock(&root
->fs_info
->block_group_cache_lock
);
2750 node
= rb_next(&cache
->cache_node
);
2751 btrfs_put_block_group(cache
);
2753 cache
= rb_entry(node
, struct btrfs_block_group_cache
,
2755 btrfs_get_block_group(cache
);
2758 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
2762 static int cache_save_setup(struct btrfs_block_group_cache
*block_group
,
2763 struct btrfs_trans_handle
*trans
,
2764 struct btrfs_path
*path
)
2766 struct btrfs_root
*root
= block_group
->fs_info
->tree_root
;
2767 struct inode
*inode
= NULL
;
2769 int dcs
= BTRFS_DC_ERROR
;
2775 * If this block group is smaller than 100 megs don't bother caching the
2778 if (block_group
->key
.offset
< (100 * 1024 * 1024)) {
2779 spin_lock(&block_group
->lock
);
2780 block_group
->disk_cache_state
= BTRFS_DC_WRITTEN
;
2781 spin_unlock(&block_group
->lock
);
2786 inode
= lookup_free_space_inode(root
, block_group
, path
);
2787 if (IS_ERR(inode
) && PTR_ERR(inode
) != -ENOENT
) {
2788 ret
= PTR_ERR(inode
);
2789 btrfs_release_path(root
, path
);
2793 if (IS_ERR(inode
)) {
2797 if (block_group
->ro
)
2800 ret
= create_free_space_inode(root
, trans
, block_group
, path
);
2807 * We want to set the generation to 0, that way if anything goes wrong
2808 * from here on out we know not to trust this cache when we load up next
2811 BTRFS_I(inode
)->generation
= 0;
2812 ret
= btrfs_update_inode(trans
, root
, inode
);
2815 if (i_size_read(inode
) > 0) {
2816 ret
= btrfs_truncate_free_space_cache(root
, trans
, path
,
2822 spin_lock(&block_group
->lock
);
2823 if (block_group
->cached
!= BTRFS_CACHE_FINISHED
) {
2824 /* We're not cached, don't bother trying to write stuff out */
2825 dcs
= BTRFS_DC_WRITTEN
;
2826 spin_unlock(&block_group
->lock
);
2829 spin_unlock(&block_group
->lock
);
2831 num_pages
= (int)div64_u64(block_group
->key
.offset
, 1024 * 1024 * 1024);
2836 * Just to make absolutely sure we have enough space, we're going to
2837 * preallocate 12 pages worth of space for each block group. In
2838 * practice we ought to use at most 8, but we need extra space so we can
2839 * add our header and have a terminator between the extents and the
2843 num_pages
*= PAGE_CACHE_SIZE
;
2845 ret
= btrfs_check_data_free_space(inode
, num_pages
);
2849 ret
= btrfs_prealloc_file_range_trans(inode
, trans
, 0, 0, num_pages
,
2850 num_pages
, num_pages
,
2853 dcs
= BTRFS_DC_SETUP
;
2854 btrfs_free_reserved_data_space(inode
, num_pages
);
2858 btrfs_release_path(root
, path
);
2860 spin_lock(&block_group
->lock
);
2861 block_group
->disk_cache_state
= dcs
;
2862 spin_unlock(&block_group
->lock
);
2867 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
2868 struct btrfs_root
*root
)
2870 struct btrfs_block_group_cache
*cache
;
2872 struct btrfs_path
*path
;
2875 path
= btrfs_alloc_path();
2881 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
2883 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
)
2885 cache
= next_block_group(root
, cache
);
2893 err
= cache_save_setup(cache
, trans
, path
);
2894 last
= cache
->key
.objectid
+ cache
->key
.offset
;
2895 btrfs_put_block_group(cache
);
2900 err
= btrfs_run_delayed_refs(trans
, root
,
2905 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
2907 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
) {
2908 btrfs_put_block_group(cache
);
2914 cache
= next_block_group(root
, cache
);
2923 if (cache
->disk_cache_state
== BTRFS_DC_SETUP
)
2924 cache
->disk_cache_state
= BTRFS_DC_NEED_WRITE
;
2926 last
= cache
->key
.objectid
+ cache
->key
.offset
;
2928 err
= write_one_cache_group(trans
, root
, path
, cache
);
2930 btrfs_put_block_group(cache
);
2935 * I don't think this is needed since we're just marking our
2936 * preallocated extent as written, but just in case it can't
2940 err
= btrfs_run_delayed_refs(trans
, root
,
2945 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
2948 * Really this shouldn't happen, but it could if we
2949 * couldn't write the entire preallocated extent and
2950 * splitting the extent resulted in a new block.
2953 btrfs_put_block_group(cache
);
2956 if (cache
->disk_cache_state
== BTRFS_DC_NEED_WRITE
)
2958 cache
= next_block_group(root
, cache
);
2967 btrfs_write_out_cache(root
, trans
, cache
, path
);
2970 * If we didn't have an error then the cache state is still
2971 * NEED_WRITE, so we can set it to WRITTEN.
2973 if (cache
->disk_cache_state
== BTRFS_DC_NEED_WRITE
)
2974 cache
->disk_cache_state
= BTRFS_DC_WRITTEN
;
2975 last
= cache
->key
.objectid
+ cache
->key
.offset
;
2976 btrfs_put_block_group(cache
);
2979 btrfs_free_path(path
);
2983 int btrfs_extent_readonly(struct btrfs_root
*root
, u64 bytenr
)
2985 struct btrfs_block_group_cache
*block_group
;
2988 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
2989 if (!block_group
|| block_group
->ro
)
2992 btrfs_put_block_group(block_group
);
2996 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
2997 u64 total_bytes
, u64 bytes_used
,
2998 struct btrfs_space_info
**space_info
)
3000 struct btrfs_space_info
*found
;
3004 if (flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3005 BTRFS_BLOCK_GROUP_RAID10
))
3010 found
= __find_space_info(info
, flags
);
3012 spin_lock(&found
->lock
);
3013 found
->total_bytes
+= total_bytes
;
3014 found
->disk_total
+= total_bytes
* factor
;
3015 found
->bytes_used
+= bytes_used
;
3016 found
->disk_used
+= bytes_used
* factor
;
3018 spin_unlock(&found
->lock
);
3019 *space_info
= found
;
3022 found
= kzalloc(sizeof(*found
), GFP_NOFS
);
3026 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
3027 INIT_LIST_HEAD(&found
->block_groups
[i
]);
3028 init_rwsem(&found
->groups_sem
);
3029 spin_lock_init(&found
->lock
);
3030 found
->flags
= flags
& (BTRFS_BLOCK_GROUP_DATA
|
3031 BTRFS_BLOCK_GROUP_SYSTEM
|
3032 BTRFS_BLOCK_GROUP_METADATA
);
3033 found
->total_bytes
= total_bytes
;
3034 found
->disk_total
= total_bytes
* factor
;
3035 found
->bytes_used
= bytes_used
;
3036 found
->disk_used
= bytes_used
* factor
;
3037 found
->bytes_pinned
= 0;
3038 found
->bytes_reserved
= 0;
3039 found
->bytes_readonly
= 0;
3040 found
->bytes_may_use
= 0;
3042 found
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3043 found
->chunk_alloc
= 0;
3044 *space_info
= found
;
3045 list_add_rcu(&found
->list
, &info
->space_info
);
3046 atomic_set(&found
->caching_threads
, 0);
3050 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
3052 u64 extra_flags
= flags
& (BTRFS_BLOCK_GROUP_RAID0
|
3053 BTRFS_BLOCK_GROUP_RAID1
|
3054 BTRFS_BLOCK_GROUP_RAID10
|
3055 BTRFS_BLOCK_GROUP_DUP
);
3057 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3058 fs_info
->avail_data_alloc_bits
|= extra_flags
;
3059 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3060 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
3061 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3062 fs_info
->avail_system_alloc_bits
|= extra_flags
;
3066 u64
btrfs_reduce_alloc_profile(struct btrfs_root
*root
, u64 flags
)
3069 * we add in the count of missing devices because we want
3070 * to make sure that any RAID levels on a degraded FS
3071 * continue to be honored.
3073 u64 num_devices
= root
->fs_info
->fs_devices
->rw_devices
+
3074 root
->fs_info
->fs_devices
->missing_devices
;
3076 if (num_devices
== 1)
3077 flags
&= ~(BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID0
);
3078 if (num_devices
< 4)
3079 flags
&= ~BTRFS_BLOCK_GROUP_RAID10
;
3081 if ((flags
& BTRFS_BLOCK_GROUP_DUP
) &&
3082 (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
3083 BTRFS_BLOCK_GROUP_RAID10
))) {
3084 flags
&= ~BTRFS_BLOCK_GROUP_DUP
;
3087 if ((flags
& BTRFS_BLOCK_GROUP_RAID1
) &&
3088 (flags
& BTRFS_BLOCK_GROUP_RAID10
)) {
3089 flags
&= ~BTRFS_BLOCK_GROUP_RAID1
;
3092 if ((flags
& BTRFS_BLOCK_GROUP_RAID0
) &&
3093 ((flags
& BTRFS_BLOCK_GROUP_RAID1
) |
3094 (flags
& BTRFS_BLOCK_GROUP_RAID10
) |
3095 (flags
& BTRFS_BLOCK_GROUP_DUP
)))
3096 flags
&= ~BTRFS_BLOCK_GROUP_RAID0
;
3100 static u64
get_alloc_profile(struct btrfs_root
*root
, u64 flags
)
3102 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3103 flags
|= root
->fs_info
->avail_data_alloc_bits
&
3104 root
->fs_info
->data_alloc_profile
;
3105 else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3106 flags
|= root
->fs_info
->avail_system_alloc_bits
&
3107 root
->fs_info
->system_alloc_profile
;
3108 else if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3109 flags
|= root
->fs_info
->avail_metadata_alloc_bits
&
3110 root
->fs_info
->metadata_alloc_profile
;
3111 return btrfs_reduce_alloc_profile(root
, flags
);
3114 u64
btrfs_get_alloc_profile(struct btrfs_root
*root
, int data
)
3119 flags
= BTRFS_BLOCK_GROUP_DATA
;
3120 else if (root
== root
->fs_info
->chunk_root
)
3121 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
3123 flags
= BTRFS_BLOCK_GROUP_METADATA
;
3125 return get_alloc_profile(root
, flags
);
3128 void btrfs_set_inode_space_info(struct btrfs_root
*root
, struct inode
*inode
)
3130 BTRFS_I(inode
)->space_info
= __find_space_info(root
->fs_info
,
3131 BTRFS_BLOCK_GROUP_DATA
);
3135 * This will check the space that the inode allocates from to make sure we have
3136 * enough space for bytes.
3138 int btrfs_check_data_free_space(struct inode
*inode
, u64 bytes
)
3140 struct btrfs_space_info
*data_sinfo
;
3141 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3143 int ret
= 0, committed
= 0, alloc_chunk
= 1;
3145 /* make sure bytes are sectorsize aligned */
3146 bytes
= (bytes
+ root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
3148 if (root
== root
->fs_info
->tree_root
||
3149 BTRFS_I(inode
)->location
.objectid
== BTRFS_FREE_INO_OBJECTID
) {
3154 data_sinfo
= BTRFS_I(inode
)->space_info
;
3159 /* make sure we have enough space to handle the data first */
3160 spin_lock(&data_sinfo
->lock
);
3161 used
= data_sinfo
->bytes_used
+ data_sinfo
->bytes_reserved
+
3162 data_sinfo
->bytes_pinned
+ data_sinfo
->bytes_readonly
+
3163 data_sinfo
->bytes_may_use
;
3165 if (used
+ bytes
> data_sinfo
->total_bytes
) {
3166 struct btrfs_trans_handle
*trans
;
3169 * if we don't have enough free bytes in this space then we need
3170 * to alloc a new chunk.
3172 if (!data_sinfo
->full
&& alloc_chunk
) {
3175 data_sinfo
->force_alloc
= CHUNK_ALLOC_FORCE
;
3176 spin_unlock(&data_sinfo
->lock
);
3178 alloc_target
= btrfs_get_alloc_profile(root
, 1);
3179 trans
= btrfs_join_transaction(root
, 1);
3181 return PTR_ERR(trans
);
3183 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
3184 bytes
+ 2 * 1024 * 1024,
3186 CHUNK_ALLOC_NO_FORCE
);
3187 btrfs_end_transaction(trans
, root
);
3196 btrfs_set_inode_space_info(root
, inode
);
3197 data_sinfo
= BTRFS_I(inode
)->space_info
;
3201 spin_unlock(&data_sinfo
->lock
);
3203 /* commit the current transaction and try again */
3205 if (!committed
&& !root
->fs_info
->open_ioctl_trans
) {
3207 trans
= btrfs_join_transaction(root
, 1);
3209 return PTR_ERR(trans
);
3210 ret
= btrfs_commit_transaction(trans
, root
);
3216 #if 0 /* I hope we never need this code again, just in case */
3217 printk(KERN_ERR
"no space left, need %llu, %llu bytes_used, "
3218 "%llu bytes_reserved, " "%llu bytes_pinned, "
3219 "%llu bytes_readonly, %llu may use %llu total\n",
3220 (unsigned long long)bytes
,
3221 (unsigned long long)data_sinfo
->bytes_used
,
3222 (unsigned long long)data_sinfo
->bytes_reserved
,
3223 (unsigned long long)data_sinfo
->bytes_pinned
,
3224 (unsigned long long)data_sinfo
->bytes_readonly
,
3225 (unsigned long long)data_sinfo
->bytes_may_use
,
3226 (unsigned long long)data_sinfo
->total_bytes
);
3230 data_sinfo
->bytes_may_use
+= bytes
;
3231 BTRFS_I(inode
)->reserved_bytes
+= bytes
;
3232 spin_unlock(&data_sinfo
->lock
);
3238 * called when we are clearing an delalloc extent from the
3239 * inode's io_tree or there was an error for whatever reason
3240 * after calling btrfs_check_data_free_space
3242 void btrfs_free_reserved_data_space(struct inode
*inode
, u64 bytes
)
3244 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3245 struct btrfs_space_info
*data_sinfo
;
3247 /* make sure bytes are sectorsize aligned */
3248 bytes
= (bytes
+ root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
3250 data_sinfo
= BTRFS_I(inode
)->space_info
;
3251 spin_lock(&data_sinfo
->lock
);
3252 data_sinfo
->bytes_may_use
-= bytes
;
3253 BTRFS_I(inode
)->reserved_bytes
-= bytes
;
3254 spin_unlock(&data_sinfo
->lock
);
3257 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
3259 struct list_head
*head
= &info
->space_info
;
3260 struct btrfs_space_info
*found
;
3263 list_for_each_entry_rcu(found
, head
, list
) {
3264 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
3265 found
->force_alloc
= CHUNK_ALLOC_FORCE
;
3270 static int should_alloc_chunk(struct btrfs_root
*root
,
3271 struct btrfs_space_info
*sinfo
, u64 alloc_bytes
,
3274 u64 num_bytes
= sinfo
->total_bytes
- sinfo
->bytes_readonly
;
3275 u64 num_allocated
= sinfo
->bytes_used
+ sinfo
->bytes_reserved
;
3278 if (force
== CHUNK_ALLOC_FORCE
)
3282 * in limited mode, we want to have some free space up to
3283 * about 1% of the FS size.
3285 if (force
== CHUNK_ALLOC_LIMITED
) {
3286 thresh
= btrfs_super_total_bytes(&root
->fs_info
->super_copy
);
3287 thresh
= max_t(u64
, 64 * 1024 * 1024,
3288 div_factor_fine(thresh
, 1));
3290 if (num_bytes
- num_allocated
< thresh
)
3295 * we have two similar checks here, one based on percentage
3296 * and once based on a hard number of 256MB. The idea
3297 * is that if we have a good amount of free
3298 * room, don't allocate a chunk. A good mount is
3299 * less than 80% utilized of the chunks we have allocated,
3300 * or more than 256MB free
3302 if (num_allocated
+ alloc_bytes
+ 256 * 1024 * 1024 < num_bytes
)
3305 if (num_allocated
+ alloc_bytes
< div_factor(num_bytes
, 8))
3308 thresh
= btrfs_super_total_bytes(&root
->fs_info
->super_copy
);
3310 /* 256MB or 5% of the FS */
3311 thresh
= max_t(u64
, 256 * 1024 * 1024, div_factor_fine(thresh
, 5));
3313 if (num_bytes
> thresh
&& sinfo
->bytes_used
< div_factor(num_bytes
, 3))
3318 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
3319 struct btrfs_root
*extent_root
, u64 alloc_bytes
,
3320 u64 flags
, int force
)
3322 struct btrfs_space_info
*space_info
;
3323 struct btrfs_fs_info
*fs_info
= extent_root
->fs_info
;
3324 int wait_for_alloc
= 0;
3327 flags
= btrfs_reduce_alloc_profile(extent_root
, flags
);
3329 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
3331 ret
= update_space_info(extent_root
->fs_info
, flags
,
3335 BUG_ON(!space_info
);
3338 spin_lock(&space_info
->lock
);
3339 if (space_info
->force_alloc
)
3340 force
= space_info
->force_alloc
;
3341 if (space_info
->full
) {
3342 spin_unlock(&space_info
->lock
);
3346 if (!should_alloc_chunk(extent_root
, space_info
, alloc_bytes
, force
)) {
3347 spin_unlock(&space_info
->lock
);
3349 } else if (space_info
->chunk_alloc
) {
3352 space_info
->chunk_alloc
= 1;
3355 spin_unlock(&space_info
->lock
);
3357 mutex_lock(&fs_info
->chunk_mutex
);
3360 * The chunk_mutex is held throughout the entirety of a chunk
3361 * allocation, so once we've acquired the chunk_mutex we know that the
3362 * other guy is done and we need to recheck and see if we should
3365 if (wait_for_alloc
) {
3366 mutex_unlock(&fs_info
->chunk_mutex
);
3372 * If we have mixed data/metadata chunks we want to make sure we keep
3373 * allocating mixed chunks instead of individual chunks.
3375 if (btrfs_mixed_space_info(space_info
))
3376 flags
|= (BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
);
3379 * if we're doing a data chunk, go ahead and make sure that
3380 * we keep a reasonable number of metadata chunks allocated in the
3383 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
3384 fs_info
->data_chunk_allocations
++;
3385 if (!(fs_info
->data_chunk_allocations
%
3386 fs_info
->metadata_ratio
))
3387 force_metadata_allocation(fs_info
);
3390 ret
= btrfs_alloc_chunk(trans
, extent_root
, flags
);
3391 spin_lock(&space_info
->lock
);
3393 space_info
->full
= 1;
3397 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3398 space_info
->chunk_alloc
= 0;
3399 spin_unlock(&space_info
->lock
);
3400 mutex_unlock(&extent_root
->fs_info
->chunk_mutex
);
3405 * shrink metadata reservation for delalloc
3407 static int shrink_delalloc(struct btrfs_trans_handle
*trans
,
3408 struct btrfs_root
*root
, u64 to_reclaim
, int sync
)
3410 struct btrfs_block_rsv
*block_rsv
;
3411 struct btrfs_space_info
*space_info
;
3416 int nr_pages
= (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT
;
3418 unsigned long progress
;
3420 block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
3421 space_info
= block_rsv
->space_info
;
3424 reserved
= space_info
->bytes_reserved
;
3425 progress
= space_info
->reservation_progress
;
3430 max_reclaim
= min(reserved
, to_reclaim
);
3432 while (loops
< 1024) {
3433 /* have the flusher threads jump in and do some IO */
3435 nr_pages
= min_t(unsigned long, nr_pages
,
3436 root
->fs_info
->delalloc_bytes
>> PAGE_CACHE_SHIFT
);
3437 writeback_inodes_sb_nr_if_idle(root
->fs_info
->sb
, nr_pages
);
3439 spin_lock(&space_info
->lock
);
3440 if (reserved
> space_info
->bytes_reserved
)
3441 reclaimed
+= reserved
- space_info
->bytes_reserved
;
3442 reserved
= space_info
->bytes_reserved
;
3443 spin_unlock(&space_info
->lock
);
3447 if (reserved
== 0 || reclaimed
>= max_reclaim
)
3450 if (trans
&& trans
->transaction
->blocked
)
3453 time_left
= schedule_timeout_interruptible(1);
3455 /* We were interrupted, exit */
3459 /* we've kicked the IO a few times, if anything has been freed,
3460 * exit. There is no sense in looping here for a long time
3461 * when we really need to commit the transaction, or there are
3462 * just too many writers without enough free space
3467 if (progress
!= space_info
->reservation_progress
)
3472 return reclaimed
>= to_reclaim
;
3476 * Retries tells us how many times we've called reserve_metadata_bytes. The
3477 * idea is if this is the first call (retries == 0) then we will add to our
3478 * reserved count if we can't make the allocation in order to hold our place
3479 * while we go and try and free up space. That way for retries > 1 we don't try
3480 * and add space, we just check to see if the amount of unused space is >= the
3481 * total space, meaning that our reservation is valid.
3483 * However if we don't intend to retry this reservation, pass -1 as retries so
3484 * that it short circuits this logic.
3486 static int reserve_metadata_bytes(struct btrfs_trans_handle
*trans
,
3487 struct btrfs_root
*root
,
3488 struct btrfs_block_rsv
*block_rsv
,
3489 u64 orig_bytes
, int flush
)
3491 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
3493 u64 num_bytes
= orig_bytes
;
3496 bool reserved
= false;
3497 bool committed
= false;
3504 spin_lock(&space_info
->lock
);
3505 unused
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
3506 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
3507 space_info
->bytes_may_use
;
3510 * The idea here is that we've not already over-reserved the block group
3511 * then we can go ahead and save our reservation first and then start
3512 * flushing if we need to. Otherwise if we've already overcommitted
3513 * lets start flushing stuff first and then come back and try to make
3516 if (unused
<= space_info
->total_bytes
) {
3517 unused
= space_info
->total_bytes
- unused
;
3518 if (unused
>= num_bytes
) {
3520 space_info
->bytes_reserved
+= orig_bytes
;
3524 * Ok set num_bytes to orig_bytes since we aren't
3525 * overocmmitted, this way we only try and reclaim what
3528 num_bytes
= orig_bytes
;
3532 * Ok we're over committed, set num_bytes to the overcommitted
3533 * amount plus the amount of bytes that we need for this
3536 num_bytes
= unused
- space_info
->total_bytes
+
3537 (orig_bytes
* (retries
+ 1));
3541 * Couldn't make our reservation, save our place so while we're trying
3542 * to reclaim space we can actually use it instead of somebody else
3543 * stealing it from us.
3545 if (ret
&& !reserved
) {
3546 space_info
->bytes_reserved
+= orig_bytes
;
3550 spin_unlock(&space_info
->lock
);
3559 * We do synchronous shrinking since we don't actually unreserve
3560 * metadata until after the IO is completed.
3562 ret
= shrink_delalloc(trans
, root
, num_bytes
, 1);
3569 * So if we were overcommitted it's possible that somebody else flushed
3570 * out enough space and we simply didn't have enough space to reclaim,
3571 * so go back around and try again.
3578 spin_lock(&space_info
->lock
);
3580 * Not enough space to be reclaimed, don't bother committing the
3583 if (space_info
->bytes_pinned
< orig_bytes
)
3585 spin_unlock(&space_info
->lock
);
3590 if (trans
|| committed
)
3594 trans
= btrfs_join_transaction(root
, 1);
3597 ret
= btrfs_commit_transaction(trans
, root
);
3606 spin_lock(&space_info
->lock
);
3607 space_info
->bytes_reserved
-= orig_bytes
;
3608 spin_unlock(&space_info
->lock
);
3614 static struct btrfs_block_rsv
*get_block_rsv(struct btrfs_trans_handle
*trans
,
3615 struct btrfs_root
*root
)
3617 struct btrfs_block_rsv
*block_rsv
;
3619 block_rsv
= trans
->block_rsv
;
3621 block_rsv
= root
->block_rsv
;
3624 block_rsv
= &root
->fs_info
->empty_block_rsv
;
3629 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
3633 spin_lock(&block_rsv
->lock
);
3634 if (block_rsv
->reserved
>= num_bytes
) {
3635 block_rsv
->reserved
-= num_bytes
;
3636 if (block_rsv
->reserved
< block_rsv
->size
)
3637 block_rsv
->full
= 0;
3640 spin_unlock(&block_rsv
->lock
);
3644 static void block_rsv_add_bytes(struct btrfs_block_rsv
*block_rsv
,
3645 u64 num_bytes
, int update_size
)
3647 spin_lock(&block_rsv
->lock
);
3648 block_rsv
->reserved
+= num_bytes
;
3650 block_rsv
->size
+= num_bytes
;
3651 else if (block_rsv
->reserved
>= block_rsv
->size
)
3652 block_rsv
->full
= 1;
3653 spin_unlock(&block_rsv
->lock
);
3656 void block_rsv_release_bytes(struct btrfs_block_rsv
*block_rsv
,
3657 struct btrfs_block_rsv
*dest
, u64 num_bytes
)
3659 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
3661 spin_lock(&block_rsv
->lock
);
3662 if (num_bytes
== (u64
)-1)
3663 num_bytes
= block_rsv
->size
;
3664 block_rsv
->size
-= num_bytes
;
3665 if (block_rsv
->reserved
>= block_rsv
->size
) {
3666 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
3667 block_rsv
->reserved
= block_rsv
->size
;
3668 block_rsv
->full
= 1;
3672 spin_unlock(&block_rsv
->lock
);
3674 if (num_bytes
> 0) {
3676 spin_lock(&dest
->lock
);
3680 bytes_to_add
= dest
->size
- dest
->reserved
;
3681 bytes_to_add
= min(num_bytes
, bytes_to_add
);
3682 dest
->reserved
+= bytes_to_add
;
3683 if (dest
->reserved
>= dest
->size
)
3685 num_bytes
-= bytes_to_add
;
3687 spin_unlock(&dest
->lock
);
3690 spin_lock(&space_info
->lock
);
3691 space_info
->bytes_reserved
-= num_bytes
;
3692 space_info
->reservation_progress
++;
3693 spin_unlock(&space_info
->lock
);
3698 static int block_rsv_migrate_bytes(struct btrfs_block_rsv
*src
,
3699 struct btrfs_block_rsv
*dst
, u64 num_bytes
)
3703 ret
= block_rsv_use_bytes(src
, num_bytes
);
3707 block_rsv_add_bytes(dst
, num_bytes
, 1);
3711 void btrfs_init_block_rsv(struct btrfs_block_rsv
*rsv
)
3713 memset(rsv
, 0, sizeof(*rsv
));
3714 spin_lock_init(&rsv
->lock
);
3715 atomic_set(&rsv
->usage
, 1);
3717 INIT_LIST_HEAD(&rsv
->list
);
3720 struct btrfs_block_rsv
*btrfs_alloc_block_rsv(struct btrfs_root
*root
)
3722 struct btrfs_block_rsv
*block_rsv
;
3723 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3725 block_rsv
= kmalloc(sizeof(*block_rsv
), GFP_NOFS
);
3729 btrfs_init_block_rsv(block_rsv
);
3730 block_rsv
->space_info
= __find_space_info(fs_info
,
3731 BTRFS_BLOCK_GROUP_METADATA
);
3735 void btrfs_free_block_rsv(struct btrfs_root
*root
,
3736 struct btrfs_block_rsv
*rsv
)
3738 if (rsv
&& atomic_dec_and_test(&rsv
->usage
)) {
3739 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
3746 * make the block_rsv struct be able to capture freed space.
3747 * the captured space will re-add to the the block_rsv struct
3748 * after transaction commit
3750 void btrfs_add_durable_block_rsv(struct btrfs_fs_info
*fs_info
,
3751 struct btrfs_block_rsv
*block_rsv
)
3753 block_rsv
->durable
= 1;
3754 mutex_lock(&fs_info
->durable_block_rsv_mutex
);
3755 list_add_tail(&block_rsv
->list
, &fs_info
->durable_block_rsv_list
);
3756 mutex_unlock(&fs_info
->durable_block_rsv_mutex
);
3759 int btrfs_block_rsv_add(struct btrfs_trans_handle
*trans
,
3760 struct btrfs_root
*root
,
3761 struct btrfs_block_rsv
*block_rsv
,
3769 ret
= reserve_metadata_bytes(trans
, root
, block_rsv
, num_bytes
, 1);
3771 block_rsv_add_bytes(block_rsv
, num_bytes
, 1);
3778 int btrfs_block_rsv_check(struct btrfs_trans_handle
*trans
,
3779 struct btrfs_root
*root
,
3780 struct btrfs_block_rsv
*block_rsv
,
3781 u64 min_reserved
, int min_factor
)
3784 int commit_trans
= 0;
3790 spin_lock(&block_rsv
->lock
);
3792 num_bytes
= div_factor(block_rsv
->size
, min_factor
);
3793 if (min_reserved
> num_bytes
)
3794 num_bytes
= min_reserved
;
3796 if (block_rsv
->reserved
>= num_bytes
) {
3799 num_bytes
-= block_rsv
->reserved
;
3800 if (block_rsv
->durable
&&
3801 block_rsv
->freed
[0] + block_rsv
->freed
[1] >= num_bytes
)
3804 spin_unlock(&block_rsv
->lock
);
3808 if (block_rsv
->refill_used
) {
3809 ret
= reserve_metadata_bytes(trans
, root
, block_rsv
,
3812 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
3821 trans
= btrfs_join_transaction(root
, 1);
3822 BUG_ON(IS_ERR(trans
));
3823 ret
= btrfs_commit_transaction(trans
, root
);
3830 int btrfs_block_rsv_migrate(struct btrfs_block_rsv
*src_rsv
,
3831 struct btrfs_block_rsv
*dst_rsv
,
3834 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
3837 void btrfs_block_rsv_release(struct btrfs_root
*root
,
3838 struct btrfs_block_rsv
*block_rsv
,
3841 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
3842 if (global_rsv
->full
|| global_rsv
== block_rsv
||
3843 block_rsv
->space_info
!= global_rsv
->space_info
)
3845 block_rsv_release_bytes(block_rsv
, global_rsv
, num_bytes
);
3849 * helper to calculate size of global block reservation.
3850 * the desired value is sum of space used by extent tree,
3851 * checksum tree and root tree
3853 static u64
calc_global_metadata_size(struct btrfs_fs_info
*fs_info
)
3855 struct btrfs_space_info
*sinfo
;
3859 int csum_size
= btrfs_super_csum_size(&fs_info
->super_copy
);
3862 * per tree used space accounting can be inaccuracy, so we
3865 spin_lock(&fs_info
->extent_root
->accounting_lock
);
3866 num_bytes
= btrfs_root_used(&fs_info
->extent_root
->root_item
);
3867 spin_unlock(&fs_info
->extent_root
->accounting_lock
);
3869 spin_lock(&fs_info
->csum_root
->accounting_lock
);
3870 num_bytes
+= btrfs_root_used(&fs_info
->csum_root
->root_item
);
3871 spin_unlock(&fs_info
->csum_root
->accounting_lock
);
3873 spin_lock(&fs_info
->tree_root
->accounting_lock
);
3874 num_bytes
+= btrfs_root_used(&fs_info
->tree_root
->root_item
);
3875 spin_unlock(&fs_info
->tree_root
->accounting_lock
);
3877 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_DATA
);
3878 spin_lock(&sinfo
->lock
);
3879 data_used
= sinfo
->bytes_used
;
3880 spin_unlock(&sinfo
->lock
);
3882 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
3883 spin_lock(&sinfo
->lock
);
3884 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_DATA
)
3886 meta_used
= sinfo
->bytes_used
;
3887 spin_unlock(&sinfo
->lock
);
3889 num_bytes
= (data_used
>> fs_info
->sb
->s_blocksize_bits
) *
3891 num_bytes
+= div64_u64(data_used
+ meta_used
, 50);
3893 if (num_bytes
* 3 > meta_used
)
3894 num_bytes
= div64_u64(meta_used
, 3);
3896 return ALIGN(num_bytes
, fs_info
->extent_root
->leafsize
<< 10);
3899 static void update_global_block_rsv(struct btrfs_fs_info
*fs_info
)
3901 struct btrfs_block_rsv
*block_rsv
= &fs_info
->global_block_rsv
;
3902 struct btrfs_space_info
*sinfo
= block_rsv
->space_info
;
3905 num_bytes
= calc_global_metadata_size(fs_info
);
3907 spin_lock(&block_rsv
->lock
);
3908 spin_lock(&sinfo
->lock
);
3910 block_rsv
->size
= num_bytes
;
3912 num_bytes
= sinfo
->bytes_used
+ sinfo
->bytes_pinned
+
3913 sinfo
->bytes_reserved
+ sinfo
->bytes_readonly
+
3914 sinfo
->bytes_may_use
;
3916 if (sinfo
->total_bytes
> num_bytes
) {
3917 num_bytes
= sinfo
->total_bytes
- num_bytes
;
3918 block_rsv
->reserved
+= num_bytes
;
3919 sinfo
->bytes_reserved
+= num_bytes
;
3922 if (block_rsv
->reserved
>= block_rsv
->size
) {
3923 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
3924 sinfo
->bytes_reserved
-= num_bytes
;
3925 sinfo
->reservation_progress
++;
3926 block_rsv
->reserved
= block_rsv
->size
;
3927 block_rsv
->full
= 1;
3930 printk(KERN_INFO
"global block rsv size %llu reserved %llu\n",
3931 block_rsv
->size
, block_rsv
->reserved
);
3933 spin_unlock(&sinfo
->lock
);
3934 spin_unlock(&block_rsv
->lock
);
3937 static void init_global_block_rsv(struct btrfs_fs_info
*fs_info
)
3939 struct btrfs_space_info
*space_info
;
3941 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
3942 fs_info
->chunk_block_rsv
.space_info
= space_info
;
3943 fs_info
->chunk_block_rsv
.priority
= 10;
3945 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
3946 fs_info
->global_block_rsv
.space_info
= space_info
;
3947 fs_info
->global_block_rsv
.priority
= 10;
3948 fs_info
->global_block_rsv
.refill_used
= 1;
3949 fs_info
->delalloc_block_rsv
.space_info
= space_info
;
3950 fs_info
->trans_block_rsv
.space_info
= space_info
;
3951 fs_info
->empty_block_rsv
.space_info
= space_info
;
3952 fs_info
->empty_block_rsv
.priority
= 10;
3954 fs_info
->extent_root
->block_rsv
= &fs_info
->global_block_rsv
;
3955 fs_info
->csum_root
->block_rsv
= &fs_info
->global_block_rsv
;
3956 fs_info
->dev_root
->block_rsv
= &fs_info
->global_block_rsv
;
3957 fs_info
->tree_root
->block_rsv
= &fs_info
->global_block_rsv
;
3958 fs_info
->chunk_root
->block_rsv
= &fs_info
->chunk_block_rsv
;
3960 btrfs_add_durable_block_rsv(fs_info
, &fs_info
->global_block_rsv
);
3962 btrfs_add_durable_block_rsv(fs_info
, &fs_info
->delalloc_block_rsv
);
3964 update_global_block_rsv(fs_info
);
3967 static void release_global_block_rsv(struct btrfs_fs_info
*fs_info
)
3969 block_rsv_release_bytes(&fs_info
->global_block_rsv
, NULL
, (u64
)-1);
3970 WARN_ON(fs_info
->delalloc_block_rsv
.size
> 0);
3971 WARN_ON(fs_info
->delalloc_block_rsv
.reserved
> 0);
3972 WARN_ON(fs_info
->trans_block_rsv
.size
> 0);
3973 WARN_ON(fs_info
->trans_block_rsv
.reserved
> 0);
3974 WARN_ON(fs_info
->chunk_block_rsv
.size
> 0);
3975 WARN_ON(fs_info
->chunk_block_rsv
.reserved
> 0);
3978 int btrfs_trans_reserve_metadata(struct btrfs_trans_handle
*trans
,
3979 struct btrfs_root
*root
,
3985 if (num_items
== 0 || root
->fs_info
->chunk_root
== root
)
3988 num_bytes
= btrfs_calc_trans_metadata_size(root
, num_items
);
3989 ret
= btrfs_block_rsv_add(trans
, root
, &root
->fs_info
->trans_block_rsv
,
3992 trans
->bytes_reserved
+= num_bytes
;
3993 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
3998 void btrfs_trans_release_metadata(struct btrfs_trans_handle
*trans
,
3999 struct btrfs_root
*root
)
4001 if (!trans
->bytes_reserved
)
4004 BUG_ON(trans
->block_rsv
!= &root
->fs_info
->trans_block_rsv
);
4005 btrfs_block_rsv_release(root
, trans
->block_rsv
,
4006 trans
->bytes_reserved
);
4007 trans
->bytes_reserved
= 0;
4010 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle
*trans
,
4011 struct inode
*inode
)
4013 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4014 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
4015 struct btrfs_block_rsv
*dst_rsv
= root
->orphan_block_rsv
;
4018 * one for deleting orphan item, one for updating inode and
4019 * two for calling btrfs_truncate_inode_items.
4021 * btrfs_truncate_inode_items is a delete operation, it frees
4022 * more space than it uses in most cases. So two units of
4023 * metadata space should be enough for calling it many times.
4024 * If all of the metadata space is used, we can commit
4025 * transaction and use space it freed.
4027 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 4);
4028 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
4031 void btrfs_orphan_release_metadata(struct inode
*inode
)
4033 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4034 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 4);
4035 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
, num_bytes
);
4038 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle
*trans
,
4039 struct btrfs_pending_snapshot
*pending
)
4041 struct btrfs_root
*root
= pending
->root
;
4042 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
4043 struct btrfs_block_rsv
*dst_rsv
= &pending
->block_rsv
;
4045 * two for root back/forward refs, two for directory entries
4046 * and one for root of the snapshot.
4048 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 5);
4049 dst_rsv
->space_info
= src_rsv
->space_info
;
4050 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
4053 static u64
calc_csum_metadata_size(struct inode
*inode
, u64 num_bytes
)
4055 return num_bytes
>>= 3;
4058 int btrfs_delalloc_reserve_metadata(struct inode
*inode
, u64 num_bytes
)
4060 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4061 struct btrfs_block_rsv
*block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
4064 int reserved_extents
;
4067 if (btrfs_transaction_in_commit(root
->fs_info
))
4068 schedule_timeout(1);
4070 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
4072 nr_extents
= atomic_read(&BTRFS_I(inode
)->outstanding_extents
) + 1;
4073 reserved_extents
= atomic_read(&BTRFS_I(inode
)->reserved_extents
);
4075 if (nr_extents
> reserved_extents
) {
4076 nr_extents
-= reserved_extents
;
4077 to_reserve
= btrfs_calc_trans_metadata_size(root
, nr_extents
);
4083 to_reserve
+= calc_csum_metadata_size(inode
, num_bytes
);
4084 ret
= reserve_metadata_bytes(NULL
, root
, block_rsv
, to_reserve
, 1);
4088 atomic_add(nr_extents
, &BTRFS_I(inode
)->reserved_extents
);
4089 atomic_inc(&BTRFS_I(inode
)->outstanding_extents
);
4091 block_rsv_add_bytes(block_rsv
, to_reserve
, 1);
4093 if (block_rsv
->size
> 512 * 1024 * 1024)
4094 shrink_delalloc(NULL
, root
, to_reserve
, 0);
4099 void btrfs_delalloc_release_metadata(struct inode
*inode
, u64 num_bytes
)
4101 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4104 int reserved_extents
;
4106 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
4107 atomic_dec(&BTRFS_I(inode
)->outstanding_extents
);
4108 WARN_ON(atomic_read(&BTRFS_I(inode
)->outstanding_extents
) < 0);
4110 reserved_extents
= atomic_read(&BTRFS_I(inode
)->reserved_extents
);
4114 nr_extents
= atomic_read(&BTRFS_I(inode
)->outstanding_extents
);
4115 if (nr_extents
>= reserved_extents
) {
4119 old
= reserved_extents
;
4120 nr_extents
= reserved_extents
- nr_extents
;
4121 new = reserved_extents
- nr_extents
;
4122 old
= atomic_cmpxchg(&BTRFS_I(inode
)->reserved_extents
,
4123 reserved_extents
, new);
4124 if (likely(old
== reserved_extents
))
4126 reserved_extents
= old
;
4129 to_free
= calc_csum_metadata_size(inode
, num_bytes
);
4131 to_free
+= btrfs_calc_trans_metadata_size(root
, nr_extents
);
4133 btrfs_block_rsv_release(root
, &root
->fs_info
->delalloc_block_rsv
,
4137 int btrfs_delalloc_reserve_space(struct inode
*inode
, u64 num_bytes
)
4141 ret
= btrfs_check_data_free_space(inode
, num_bytes
);
4145 ret
= btrfs_delalloc_reserve_metadata(inode
, num_bytes
);
4147 btrfs_free_reserved_data_space(inode
, num_bytes
);
4154 void btrfs_delalloc_release_space(struct inode
*inode
, u64 num_bytes
)
4156 btrfs_delalloc_release_metadata(inode
, num_bytes
);
4157 btrfs_free_reserved_data_space(inode
, num_bytes
);
4160 static int update_block_group(struct btrfs_trans_handle
*trans
,
4161 struct btrfs_root
*root
,
4162 u64 bytenr
, u64 num_bytes
, int alloc
)
4164 struct btrfs_block_group_cache
*cache
= NULL
;
4165 struct btrfs_fs_info
*info
= root
->fs_info
;
4166 u64 total
= num_bytes
;
4171 /* block accounting for super block */
4172 spin_lock(&info
->delalloc_lock
);
4173 old_val
= btrfs_super_bytes_used(&info
->super_copy
);
4175 old_val
+= num_bytes
;
4177 old_val
-= num_bytes
;
4178 btrfs_set_super_bytes_used(&info
->super_copy
, old_val
);
4179 spin_unlock(&info
->delalloc_lock
);
4182 cache
= btrfs_lookup_block_group(info
, bytenr
);
4185 if (cache
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
4186 BTRFS_BLOCK_GROUP_RAID1
|
4187 BTRFS_BLOCK_GROUP_RAID10
))
4192 * If this block group has free space cache written out, we
4193 * need to make sure to load it if we are removing space. This
4194 * is because we need the unpinning stage to actually add the
4195 * space back to the block group, otherwise we will leak space.
4197 if (!alloc
&& cache
->cached
== BTRFS_CACHE_NO
)
4198 cache_block_group(cache
, trans
, NULL
, 1);
4200 byte_in_group
= bytenr
- cache
->key
.objectid
;
4201 WARN_ON(byte_in_group
> cache
->key
.offset
);
4203 spin_lock(&cache
->space_info
->lock
);
4204 spin_lock(&cache
->lock
);
4206 if (btrfs_super_cache_generation(&info
->super_copy
) != 0 &&
4207 cache
->disk_cache_state
< BTRFS_DC_CLEAR
)
4208 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
4211 old_val
= btrfs_block_group_used(&cache
->item
);
4212 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
4214 old_val
+= num_bytes
;
4215 btrfs_set_block_group_used(&cache
->item
, old_val
);
4216 cache
->reserved
-= num_bytes
;
4217 cache
->space_info
->bytes_reserved
-= num_bytes
;
4218 cache
->space_info
->reservation_progress
++;
4219 cache
->space_info
->bytes_used
+= num_bytes
;
4220 cache
->space_info
->disk_used
+= num_bytes
* factor
;
4221 spin_unlock(&cache
->lock
);
4222 spin_unlock(&cache
->space_info
->lock
);
4224 old_val
-= num_bytes
;
4225 btrfs_set_block_group_used(&cache
->item
, old_val
);
4226 cache
->pinned
+= num_bytes
;
4227 cache
->space_info
->bytes_pinned
+= num_bytes
;
4228 cache
->space_info
->bytes_used
-= num_bytes
;
4229 cache
->space_info
->disk_used
-= num_bytes
* factor
;
4230 spin_unlock(&cache
->lock
);
4231 spin_unlock(&cache
->space_info
->lock
);
4233 set_extent_dirty(info
->pinned_extents
,
4234 bytenr
, bytenr
+ num_bytes
- 1,
4235 GFP_NOFS
| __GFP_NOFAIL
);
4237 btrfs_put_block_group(cache
);
4239 bytenr
+= num_bytes
;
4244 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
4246 struct btrfs_block_group_cache
*cache
;
4249 cache
= btrfs_lookup_first_block_group(root
->fs_info
, search_start
);
4253 bytenr
= cache
->key
.objectid
;
4254 btrfs_put_block_group(cache
);
4259 static int pin_down_extent(struct btrfs_root
*root
,
4260 struct btrfs_block_group_cache
*cache
,
4261 u64 bytenr
, u64 num_bytes
, int reserved
)
4263 spin_lock(&cache
->space_info
->lock
);
4264 spin_lock(&cache
->lock
);
4265 cache
->pinned
+= num_bytes
;
4266 cache
->space_info
->bytes_pinned
+= num_bytes
;
4268 cache
->reserved
-= num_bytes
;
4269 cache
->space_info
->bytes_reserved
-= num_bytes
;
4270 cache
->space_info
->reservation_progress
++;
4272 spin_unlock(&cache
->lock
);
4273 spin_unlock(&cache
->space_info
->lock
);
4275 set_extent_dirty(root
->fs_info
->pinned_extents
, bytenr
,
4276 bytenr
+ num_bytes
- 1, GFP_NOFS
| __GFP_NOFAIL
);
4281 * this function must be called within transaction
4283 int btrfs_pin_extent(struct btrfs_root
*root
,
4284 u64 bytenr
, u64 num_bytes
, int reserved
)
4286 struct btrfs_block_group_cache
*cache
;
4288 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
4291 pin_down_extent(root
, cache
, bytenr
, num_bytes
, reserved
);
4293 btrfs_put_block_group(cache
);
4298 * update size of reserved extents. this function may return -EAGAIN
4299 * if 'reserve' is true or 'sinfo' is false.
4301 int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
4302 u64 num_bytes
, int reserve
, int sinfo
)
4306 struct btrfs_space_info
*space_info
= cache
->space_info
;
4307 spin_lock(&space_info
->lock
);
4308 spin_lock(&cache
->lock
);
4313 cache
->reserved
+= num_bytes
;
4314 space_info
->bytes_reserved
+= num_bytes
;
4318 space_info
->bytes_readonly
+= num_bytes
;
4319 cache
->reserved
-= num_bytes
;
4320 space_info
->bytes_reserved
-= num_bytes
;
4321 space_info
->reservation_progress
++;
4323 spin_unlock(&cache
->lock
);
4324 spin_unlock(&space_info
->lock
);
4326 spin_lock(&cache
->lock
);
4331 cache
->reserved
+= num_bytes
;
4333 cache
->reserved
-= num_bytes
;
4335 spin_unlock(&cache
->lock
);
4340 int btrfs_prepare_extent_commit(struct btrfs_trans_handle
*trans
,
4341 struct btrfs_root
*root
)
4343 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4344 struct btrfs_caching_control
*next
;
4345 struct btrfs_caching_control
*caching_ctl
;
4346 struct btrfs_block_group_cache
*cache
;
4348 down_write(&fs_info
->extent_commit_sem
);
4350 list_for_each_entry_safe(caching_ctl
, next
,
4351 &fs_info
->caching_block_groups
, list
) {
4352 cache
= caching_ctl
->block_group
;
4353 if (block_group_cache_done(cache
)) {
4354 cache
->last_byte_to_unpin
= (u64
)-1;
4355 list_del_init(&caching_ctl
->list
);
4356 put_caching_control(caching_ctl
);
4358 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
4362 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
4363 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
4365 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
4367 up_write(&fs_info
->extent_commit_sem
);
4369 update_global_block_rsv(fs_info
);
4373 static int unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
4375 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4376 struct btrfs_block_group_cache
*cache
= NULL
;
4379 while (start
<= end
) {
4381 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
4383 btrfs_put_block_group(cache
);
4384 cache
= btrfs_lookup_block_group(fs_info
, start
);
4388 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
4389 len
= min(len
, end
+ 1 - start
);
4391 if (start
< cache
->last_byte_to_unpin
) {
4392 len
= min(len
, cache
->last_byte_to_unpin
- start
);
4393 btrfs_add_free_space(cache
, start
, len
);
4398 spin_lock(&cache
->space_info
->lock
);
4399 spin_lock(&cache
->lock
);
4400 cache
->pinned
-= len
;
4401 cache
->space_info
->bytes_pinned
-= len
;
4403 cache
->space_info
->bytes_readonly
+= len
;
4404 } else if (cache
->reserved_pinned
> 0) {
4405 len
= min(len
, cache
->reserved_pinned
);
4406 cache
->reserved_pinned
-= len
;
4407 cache
->space_info
->bytes_reserved
+= len
;
4409 spin_unlock(&cache
->lock
);
4410 spin_unlock(&cache
->space_info
->lock
);
4414 btrfs_put_block_group(cache
);
4418 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
4419 struct btrfs_root
*root
)
4421 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4422 struct extent_io_tree
*unpin
;
4423 struct btrfs_block_rsv
*block_rsv
;
4424 struct btrfs_block_rsv
*next_rsv
;
4430 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
4431 unpin
= &fs_info
->freed_extents
[1];
4433 unpin
= &fs_info
->freed_extents
[0];
4436 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
4441 if (btrfs_test_opt(root
, DISCARD
))
4442 ret
= btrfs_discard_extent(root
, start
,
4443 end
+ 1 - start
, NULL
);
4445 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
4446 unpin_extent_range(root
, start
, end
);
4450 mutex_lock(&fs_info
->durable_block_rsv_mutex
);
4451 list_for_each_entry_safe(block_rsv
, next_rsv
,
4452 &fs_info
->durable_block_rsv_list
, list
) {
4454 idx
= trans
->transid
& 0x1;
4455 if (block_rsv
->freed
[idx
] > 0) {
4456 block_rsv_add_bytes(block_rsv
,
4457 block_rsv
->freed
[idx
], 0);
4458 block_rsv
->freed
[idx
] = 0;
4460 if (atomic_read(&block_rsv
->usage
) == 0) {
4461 btrfs_block_rsv_release(root
, block_rsv
, (u64
)-1);
4463 if (block_rsv
->freed
[0] == 0 &&
4464 block_rsv
->freed
[1] == 0) {
4465 list_del_init(&block_rsv
->list
);
4469 btrfs_block_rsv_release(root
, block_rsv
, 0);
4472 mutex_unlock(&fs_info
->durable_block_rsv_mutex
);
4477 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
4478 struct btrfs_root
*root
,
4479 u64 bytenr
, u64 num_bytes
, u64 parent
,
4480 u64 root_objectid
, u64 owner_objectid
,
4481 u64 owner_offset
, int refs_to_drop
,
4482 struct btrfs_delayed_extent_op
*extent_op
)
4484 struct btrfs_key key
;
4485 struct btrfs_path
*path
;
4486 struct btrfs_fs_info
*info
= root
->fs_info
;
4487 struct btrfs_root
*extent_root
= info
->extent_root
;
4488 struct extent_buffer
*leaf
;
4489 struct btrfs_extent_item
*ei
;
4490 struct btrfs_extent_inline_ref
*iref
;
4493 int extent_slot
= 0;
4494 int found_extent
= 0;
4499 path
= btrfs_alloc_path();
4504 path
->leave_spinning
= 1;
4506 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
4507 BUG_ON(!is_data
&& refs_to_drop
!= 1);
4509 ret
= lookup_extent_backref(trans
, extent_root
, path
, &iref
,
4510 bytenr
, num_bytes
, parent
,
4511 root_objectid
, owner_objectid
,
4514 extent_slot
= path
->slots
[0];
4515 while (extent_slot
>= 0) {
4516 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
4518 if (key
.objectid
!= bytenr
)
4520 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
4521 key
.offset
== num_bytes
) {
4525 if (path
->slots
[0] - extent_slot
> 5)
4529 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4530 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
4531 if (found_extent
&& item_size
< sizeof(*ei
))
4534 if (!found_extent
) {
4536 ret
= remove_extent_backref(trans
, extent_root
, path
,
4540 btrfs_release_path(extent_root
, path
);
4541 path
->leave_spinning
= 1;
4543 key
.objectid
= bytenr
;
4544 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
4545 key
.offset
= num_bytes
;
4547 ret
= btrfs_search_slot(trans
, extent_root
,
4550 printk(KERN_ERR
"umm, got %d back from search"
4551 ", was looking for %llu\n", ret
,
4552 (unsigned long long)bytenr
);
4553 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
4556 extent_slot
= path
->slots
[0];
4559 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
4561 printk(KERN_ERR
"btrfs unable to find ref byte nr %llu "
4562 "parent %llu root %llu owner %llu offset %llu\n",
4563 (unsigned long long)bytenr
,
4564 (unsigned long long)parent
,
4565 (unsigned long long)root_objectid
,
4566 (unsigned long long)owner_objectid
,
4567 (unsigned long long)owner_offset
);
4570 leaf
= path
->nodes
[0];
4571 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
4572 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4573 if (item_size
< sizeof(*ei
)) {
4574 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
4575 ret
= convert_extent_item_v0(trans
, extent_root
, path
,
4579 btrfs_release_path(extent_root
, path
);
4580 path
->leave_spinning
= 1;
4582 key
.objectid
= bytenr
;
4583 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
4584 key
.offset
= num_bytes
;
4586 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
4589 printk(KERN_ERR
"umm, got %d back from search"
4590 ", was looking for %llu\n", ret
,
4591 (unsigned long long)bytenr
);
4592 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
4595 extent_slot
= path
->slots
[0];
4596 leaf
= path
->nodes
[0];
4597 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
4600 BUG_ON(item_size
< sizeof(*ei
));
4601 ei
= btrfs_item_ptr(leaf
, extent_slot
,
4602 struct btrfs_extent_item
);
4603 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
) {
4604 struct btrfs_tree_block_info
*bi
;
4605 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
4606 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
4607 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
4610 refs
= btrfs_extent_refs(leaf
, ei
);
4611 BUG_ON(refs
< refs_to_drop
);
4612 refs
-= refs_to_drop
;
4616 __run_delayed_extent_op(extent_op
, leaf
, ei
);
4618 * In the case of inline back ref, reference count will
4619 * be updated by remove_extent_backref
4622 BUG_ON(!found_extent
);
4624 btrfs_set_extent_refs(leaf
, ei
, refs
);
4625 btrfs_mark_buffer_dirty(leaf
);
4628 ret
= remove_extent_backref(trans
, extent_root
, path
,
4635 BUG_ON(is_data
&& refs_to_drop
!=
4636 extent_data_ref_count(root
, path
, iref
));
4638 BUG_ON(path
->slots
[0] != extent_slot
);
4640 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
4641 path
->slots
[0] = extent_slot
;
4646 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
4649 btrfs_release_path(extent_root
, path
);
4652 ret
= btrfs_del_csums(trans
, root
, bytenr
, num_bytes
);
4655 invalidate_mapping_pages(info
->btree_inode
->i_mapping
,
4656 bytenr
>> PAGE_CACHE_SHIFT
,
4657 (bytenr
+ num_bytes
- 1) >> PAGE_CACHE_SHIFT
);
4660 ret
= update_block_group(trans
, root
, bytenr
, num_bytes
, 0);
4663 btrfs_free_path(path
);
4668 * when we free an block, it is possible (and likely) that we free the last
4669 * delayed ref for that extent as well. This searches the delayed ref tree for
4670 * a given extent, and if there are no other delayed refs to be processed, it
4671 * removes it from the tree.
4673 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
4674 struct btrfs_root
*root
, u64 bytenr
)
4676 struct btrfs_delayed_ref_head
*head
;
4677 struct btrfs_delayed_ref_root
*delayed_refs
;
4678 struct btrfs_delayed_ref_node
*ref
;
4679 struct rb_node
*node
;
4682 delayed_refs
= &trans
->transaction
->delayed_refs
;
4683 spin_lock(&delayed_refs
->lock
);
4684 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
4688 node
= rb_prev(&head
->node
.rb_node
);
4692 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
4694 /* there are still entries for this ref, we can't drop it */
4695 if (ref
->bytenr
== bytenr
)
4698 if (head
->extent_op
) {
4699 if (!head
->must_insert_reserved
)
4701 kfree(head
->extent_op
);
4702 head
->extent_op
= NULL
;
4706 * waiting for the lock here would deadlock. If someone else has it
4707 * locked they are already in the process of dropping it anyway
4709 if (!mutex_trylock(&head
->mutex
))
4713 * at this point we have a head with no other entries. Go
4714 * ahead and process it.
4716 head
->node
.in_tree
= 0;
4717 rb_erase(&head
->node
.rb_node
, &delayed_refs
->root
);
4719 delayed_refs
->num_entries
--;
4722 * we don't take a ref on the node because we're removing it from the
4723 * tree, so we just steal the ref the tree was holding.
4725 delayed_refs
->num_heads
--;
4726 if (list_empty(&head
->cluster
))
4727 delayed_refs
->num_heads_ready
--;
4729 list_del_init(&head
->cluster
);
4730 spin_unlock(&delayed_refs
->lock
);
4732 BUG_ON(head
->extent_op
);
4733 if (head
->must_insert_reserved
)
4736 mutex_unlock(&head
->mutex
);
4737 btrfs_put_delayed_ref(&head
->node
);
4740 spin_unlock(&delayed_refs
->lock
);
4744 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
4745 struct btrfs_root
*root
,
4746 struct extent_buffer
*buf
,
4747 u64 parent
, int last_ref
)
4749 struct btrfs_block_rsv
*block_rsv
;
4750 struct btrfs_block_group_cache
*cache
= NULL
;
4753 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
4754 ret
= btrfs_add_delayed_tree_ref(trans
, buf
->start
, buf
->len
,
4755 parent
, root
->root_key
.objectid
,
4756 btrfs_header_level(buf
),
4757 BTRFS_DROP_DELAYED_REF
, NULL
);
4764 block_rsv
= get_block_rsv(trans
, root
);
4765 cache
= btrfs_lookup_block_group(root
->fs_info
, buf
->start
);
4766 if (block_rsv
->space_info
!= cache
->space_info
)
4769 if (btrfs_header_generation(buf
) == trans
->transid
) {
4770 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
4771 ret
= check_ref_cleanup(trans
, root
, buf
->start
);
4776 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
4777 pin_down_extent(root
, cache
, buf
->start
, buf
->len
, 1);
4781 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
4783 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
4784 ret
= btrfs_update_reserved_bytes(cache
, buf
->len
, 0, 0);
4785 if (ret
== -EAGAIN
) {
4786 /* block group became read-only */
4787 btrfs_update_reserved_bytes(cache
, buf
->len
, 0, 1);
4792 spin_lock(&block_rsv
->lock
);
4793 if (block_rsv
->reserved
< block_rsv
->size
) {
4794 block_rsv
->reserved
+= buf
->len
;
4797 spin_unlock(&block_rsv
->lock
);
4800 spin_lock(&cache
->space_info
->lock
);
4801 cache
->space_info
->bytes_reserved
-= buf
->len
;
4802 cache
->space_info
->reservation_progress
++;
4803 spin_unlock(&cache
->space_info
->lock
);
4808 if (block_rsv
->durable
&& !cache
->ro
) {
4810 spin_lock(&cache
->lock
);
4812 cache
->reserved_pinned
+= buf
->len
;
4815 spin_unlock(&cache
->lock
);
4818 spin_lock(&block_rsv
->lock
);
4819 block_rsv
->freed
[trans
->transid
& 0x1] += buf
->len
;
4820 spin_unlock(&block_rsv
->lock
);
4825 * Deleting the buffer, clear the corrupt flag since it doesn't matter
4828 clear_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
);
4829 btrfs_put_block_group(cache
);
4832 int btrfs_free_extent(struct btrfs_trans_handle
*trans
,
4833 struct btrfs_root
*root
,
4834 u64 bytenr
, u64 num_bytes
, u64 parent
,
4835 u64 root_objectid
, u64 owner
, u64 offset
)
4840 * tree log blocks never actually go into the extent allocation
4841 * tree, just update pinning info and exit early.
4843 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
4844 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
4845 /* unlocks the pinned mutex */
4846 btrfs_pin_extent(root
, bytenr
, num_bytes
, 1);
4848 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
4849 ret
= btrfs_add_delayed_tree_ref(trans
, bytenr
, num_bytes
,
4850 parent
, root_objectid
, (int)owner
,
4851 BTRFS_DROP_DELAYED_REF
, NULL
);
4854 ret
= btrfs_add_delayed_data_ref(trans
, bytenr
, num_bytes
,
4855 parent
, root_objectid
, owner
,
4856 offset
, BTRFS_DROP_DELAYED_REF
, NULL
);
4862 static u64
stripe_align(struct btrfs_root
*root
, u64 val
)
4864 u64 mask
= ((u64
)root
->stripesize
- 1);
4865 u64 ret
= (val
+ mask
) & ~mask
;
4870 * when we wait for progress in the block group caching, its because
4871 * our allocation attempt failed at least once. So, we must sleep
4872 * and let some progress happen before we try again.
4874 * This function will sleep at least once waiting for new free space to
4875 * show up, and then it will check the block group free space numbers
4876 * for our min num_bytes. Another option is to have it go ahead
4877 * and look in the rbtree for a free extent of a given size, but this
4881 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
4884 struct btrfs_caching_control
*caching_ctl
;
4887 caching_ctl
= get_caching_control(cache
);
4891 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
4892 (cache
->free_space_ctl
->free_space
>= num_bytes
));
4894 put_caching_control(caching_ctl
);
4899 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
4901 struct btrfs_caching_control
*caching_ctl
;
4904 caching_ctl
= get_caching_control(cache
);
4908 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
4910 put_caching_control(caching_ctl
);
4914 static int get_block_group_index(struct btrfs_block_group_cache
*cache
)
4917 if (cache
->flags
& BTRFS_BLOCK_GROUP_RAID10
)
4919 else if (cache
->flags
& BTRFS_BLOCK_GROUP_RAID1
)
4921 else if (cache
->flags
& BTRFS_BLOCK_GROUP_DUP
)
4923 else if (cache
->flags
& BTRFS_BLOCK_GROUP_RAID0
)
4930 enum btrfs_loop_type
{
4931 LOOP_FIND_IDEAL
= 0,
4932 LOOP_CACHING_NOWAIT
= 1,
4933 LOOP_CACHING_WAIT
= 2,
4934 LOOP_ALLOC_CHUNK
= 3,
4935 LOOP_NO_EMPTY_SIZE
= 4,
4939 * walks the btree of allocated extents and find a hole of a given size.
4940 * The key ins is changed to record the hole:
4941 * ins->objectid == block start
4942 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4943 * ins->offset == number of blocks
4944 * Any available blocks before search_start are skipped.
4946 static noinline
int find_free_extent(struct btrfs_trans_handle
*trans
,
4947 struct btrfs_root
*orig_root
,
4948 u64 num_bytes
, u64 empty_size
,
4949 u64 search_start
, u64 search_end
,
4950 u64 hint_byte
, struct btrfs_key
*ins
,
4954 struct btrfs_root
*root
= orig_root
->fs_info
->extent_root
;
4955 struct btrfs_free_cluster
*last_ptr
= NULL
;
4956 struct btrfs_block_group_cache
*block_group
= NULL
;
4957 int empty_cluster
= 2 * 1024 * 1024;
4958 int allowed_chunk_alloc
= 0;
4959 int done_chunk_alloc
= 0;
4960 struct btrfs_space_info
*space_info
;
4961 int last_ptr_loop
= 0;
4964 bool found_uncached_bg
= false;
4965 bool failed_cluster_refill
= false;
4966 bool failed_alloc
= false;
4967 bool use_cluster
= true;
4968 u64 ideal_cache_percent
= 0;
4969 u64 ideal_cache_offset
= 0;
4971 WARN_ON(num_bytes
< root
->sectorsize
);
4972 btrfs_set_key_type(ins
, BTRFS_EXTENT_ITEM_KEY
);
4976 space_info
= __find_space_info(root
->fs_info
, data
);
4978 printk(KERN_ERR
"No space info for %d\n", data
);
4983 * If the space info is for both data and metadata it means we have a
4984 * small filesystem and we can't use the clustering stuff.
4986 if (btrfs_mixed_space_info(space_info
))
4987 use_cluster
= false;
4989 if (orig_root
->ref_cows
|| empty_size
)
4990 allowed_chunk_alloc
= 1;
4992 if (data
& BTRFS_BLOCK_GROUP_METADATA
&& use_cluster
) {
4993 last_ptr
= &root
->fs_info
->meta_alloc_cluster
;
4994 if (!btrfs_test_opt(root
, SSD
))
4995 empty_cluster
= 64 * 1024;
4998 if ((data
& BTRFS_BLOCK_GROUP_DATA
) && use_cluster
&&
4999 btrfs_test_opt(root
, SSD
)) {
5000 last_ptr
= &root
->fs_info
->data_alloc_cluster
;
5004 spin_lock(&last_ptr
->lock
);
5005 if (last_ptr
->block_group
)
5006 hint_byte
= last_ptr
->window_start
;
5007 spin_unlock(&last_ptr
->lock
);
5010 search_start
= max(search_start
, first_logical_byte(root
, 0));
5011 search_start
= max(search_start
, hint_byte
);
5016 if (search_start
== hint_byte
) {
5018 block_group
= btrfs_lookup_block_group(root
->fs_info
,
5021 * we don't want to use the block group if it doesn't match our
5022 * allocation bits, or if its not cached.
5024 * However if we are re-searching with an ideal block group
5025 * picked out then we don't care that the block group is cached.
5027 if (block_group
&& block_group_bits(block_group
, data
) &&
5028 (block_group
->cached
!= BTRFS_CACHE_NO
||
5029 search_start
== ideal_cache_offset
)) {
5030 down_read(&space_info
->groups_sem
);
5031 if (list_empty(&block_group
->list
) ||
5034 * someone is removing this block group,
5035 * we can't jump into the have_block_group
5036 * target because our list pointers are not
5039 btrfs_put_block_group(block_group
);
5040 up_read(&space_info
->groups_sem
);
5042 index
= get_block_group_index(block_group
);
5043 goto have_block_group
;
5045 } else if (block_group
) {
5046 btrfs_put_block_group(block_group
);
5050 down_read(&space_info
->groups_sem
);
5051 list_for_each_entry(block_group
, &space_info
->block_groups
[index
],
5056 btrfs_get_block_group(block_group
);
5057 search_start
= block_group
->key
.objectid
;
5060 * this can happen if we end up cycling through all the
5061 * raid types, but we want to make sure we only allocate
5062 * for the proper type.
5064 if (!block_group_bits(block_group
, data
)) {
5065 u64 extra
= BTRFS_BLOCK_GROUP_DUP
|
5066 BTRFS_BLOCK_GROUP_RAID1
|
5067 BTRFS_BLOCK_GROUP_RAID10
;
5070 * if they asked for extra copies and this block group
5071 * doesn't provide them, bail. This does allow us to
5072 * fill raid0 from raid1.
5074 if ((data
& extra
) && !(block_group
->flags
& extra
))
5079 if (unlikely(block_group
->cached
== BTRFS_CACHE_NO
)) {
5082 ret
= cache_block_group(block_group
, trans
,
5084 if (block_group
->cached
== BTRFS_CACHE_FINISHED
)
5085 goto have_block_group
;
5087 free_percent
= btrfs_block_group_used(&block_group
->item
);
5088 free_percent
*= 100;
5089 free_percent
= div64_u64(free_percent
,
5090 block_group
->key
.offset
);
5091 free_percent
= 100 - free_percent
;
5092 if (free_percent
> ideal_cache_percent
&&
5093 likely(!block_group
->ro
)) {
5094 ideal_cache_offset
= block_group
->key
.objectid
;
5095 ideal_cache_percent
= free_percent
;
5099 * We only want to start kthread caching if we are at
5100 * the point where we will wait for caching to make
5101 * progress, or if our ideal search is over and we've
5102 * found somebody to start caching.
5104 if (loop
> LOOP_CACHING_NOWAIT
||
5105 (loop
> LOOP_FIND_IDEAL
&&
5106 atomic_read(&space_info
->caching_threads
) < 2)) {
5107 ret
= cache_block_group(block_group
, trans
,
5111 found_uncached_bg
= true;
5114 * If loop is set for cached only, try the next block
5117 if (loop
== LOOP_FIND_IDEAL
)
5121 cached
= block_group_cache_done(block_group
);
5122 if (unlikely(!cached
))
5123 found_uncached_bg
= true;
5125 if (unlikely(block_group
->ro
))
5129 * Ok we want to try and use the cluster allocator, so lets look
5130 * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
5131 * have tried the cluster allocator plenty of times at this
5132 * point and not have found anything, so we are likely way too
5133 * fragmented for the clustering stuff to find anything, so lets
5134 * just skip it and let the allocator find whatever block it can
5137 if (last_ptr
&& loop
< LOOP_NO_EMPTY_SIZE
) {
5139 * the refill lock keeps out other
5140 * people trying to start a new cluster
5142 spin_lock(&last_ptr
->refill_lock
);
5143 if (last_ptr
->block_group
&&
5144 (last_ptr
->block_group
->ro
||
5145 !block_group_bits(last_ptr
->block_group
, data
))) {
5147 goto refill_cluster
;
5150 offset
= btrfs_alloc_from_cluster(block_group
, last_ptr
,
5151 num_bytes
, search_start
);
5153 /* we have a block, we're done */
5154 spin_unlock(&last_ptr
->refill_lock
);
5158 spin_lock(&last_ptr
->lock
);
5160 * whoops, this cluster doesn't actually point to
5161 * this block group. Get a ref on the block
5162 * group is does point to and try again
5164 if (!last_ptr_loop
&& last_ptr
->block_group
&&
5165 last_ptr
->block_group
!= block_group
) {
5167 btrfs_put_block_group(block_group
);
5168 block_group
= last_ptr
->block_group
;
5169 btrfs_get_block_group(block_group
);
5170 spin_unlock(&last_ptr
->lock
);
5171 spin_unlock(&last_ptr
->refill_lock
);
5174 search_start
= block_group
->key
.objectid
;
5176 * we know this block group is properly
5177 * in the list because
5178 * btrfs_remove_block_group, drops the
5179 * cluster before it removes the block
5180 * group from the list
5182 goto have_block_group
;
5184 spin_unlock(&last_ptr
->lock
);
5187 * this cluster didn't work out, free it and
5190 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
5194 /* allocate a cluster in this block group */
5195 ret
= btrfs_find_space_cluster(trans
, root
,
5196 block_group
, last_ptr
,
5198 empty_cluster
+ empty_size
);
5201 * now pull our allocation out of this
5204 offset
= btrfs_alloc_from_cluster(block_group
,
5205 last_ptr
, num_bytes
,
5208 /* we found one, proceed */
5209 spin_unlock(&last_ptr
->refill_lock
);
5212 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
5213 && !failed_cluster_refill
) {
5214 spin_unlock(&last_ptr
->refill_lock
);
5216 failed_cluster_refill
= true;
5217 wait_block_group_cache_progress(block_group
,
5218 num_bytes
+ empty_cluster
+ empty_size
);
5219 goto have_block_group
;
5223 * at this point we either didn't find a cluster
5224 * or we weren't able to allocate a block from our
5225 * cluster. Free the cluster we've been trying
5226 * to use, and go to the next block group
5228 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
5229 spin_unlock(&last_ptr
->refill_lock
);
5233 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
5234 num_bytes
, empty_size
);
5236 * If we didn't find a chunk, and we haven't failed on this
5237 * block group before, and this block group is in the middle of
5238 * caching and we are ok with waiting, then go ahead and wait
5239 * for progress to be made, and set failed_alloc to true.
5241 * If failed_alloc is true then we've already waited on this
5242 * block group once and should move on to the next block group.
5244 if (!offset
&& !failed_alloc
&& !cached
&&
5245 loop
> LOOP_CACHING_NOWAIT
) {
5246 wait_block_group_cache_progress(block_group
,
5247 num_bytes
+ empty_size
);
5248 failed_alloc
= true;
5249 goto have_block_group
;
5250 } else if (!offset
) {
5254 search_start
= stripe_align(root
, offset
);
5255 /* move on to the next group */
5256 if (search_start
+ num_bytes
>= search_end
) {
5257 btrfs_add_free_space(block_group
, offset
, num_bytes
);
5261 /* move on to the next group */
5262 if (search_start
+ num_bytes
>
5263 block_group
->key
.objectid
+ block_group
->key
.offset
) {
5264 btrfs_add_free_space(block_group
, offset
, num_bytes
);
5268 ins
->objectid
= search_start
;
5269 ins
->offset
= num_bytes
;
5271 if (offset
< search_start
)
5272 btrfs_add_free_space(block_group
, offset
,
5273 search_start
- offset
);
5274 BUG_ON(offset
> search_start
);
5276 ret
= btrfs_update_reserved_bytes(block_group
, num_bytes
, 1,
5277 (data
& BTRFS_BLOCK_GROUP_DATA
));
5278 if (ret
== -EAGAIN
) {
5279 btrfs_add_free_space(block_group
, offset
, num_bytes
);
5283 /* we are all good, lets return */
5284 ins
->objectid
= search_start
;
5285 ins
->offset
= num_bytes
;
5287 if (offset
< search_start
)
5288 btrfs_add_free_space(block_group
, offset
,
5289 search_start
- offset
);
5290 BUG_ON(offset
> search_start
);
5293 failed_cluster_refill
= false;
5294 failed_alloc
= false;
5295 BUG_ON(index
!= get_block_group_index(block_group
));
5296 btrfs_put_block_group(block_group
);
5298 up_read(&space_info
->groups_sem
);
5300 if (!ins
->objectid
&& ++index
< BTRFS_NR_RAID_TYPES
)
5303 /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5304 * for them to make caching progress. Also
5305 * determine the best possible bg to cache
5306 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5307 * caching kthreads as we move along
5308 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5309 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5310 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5313 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
&&
5314 (found_uncached_bg
|| empty_size
|| empty_cluster
||
5315 allowed_chunk_alloc
)) {
5317 if (loop
== LOOP_FIND_IDEAL
&& found_uncached_bg
) {
5318 found_uncached_bg
= false;
5320 if (!ideal_cache_percent
&&
5321 atomic_read(&space_info
->caching_threads
))
5325 * 1 of the following 2 things have happened so far
5327 * 1) We found an ideal block group for caching that
5328 * is mostly full and will cache quickly, so we might
5329 * as well wait for it.
5331 * 2) We searched for cached only and we didn't find
5332 * anything, and we didn't start any caching kthreads
5333 * either, so chances are we will loop through and
5334 * start a couple caching kthreads, and then come back
5335 * around and just wait for them. This will be slower
5336 * because we will have 2 caching kthreads reading at
5337 * the same time when we could have just started one
5338 * and waited for it to get far enough to give us an
5339 * allocation, so go ahead and go to the wait caching
5342 loop
= LOOP_CACHING_WAIT
;
5343 search_start
= ideal_cache_offset
;
5344 ideal_cache_percent
= 0;
5346 } else if (loop
== LOOP_FIND_IDEAL
) {
5348 * Didn't find a uncached bg, wait on anything we find
5351 loop
= LOOP_CACHING_WAIT
;
5355 if (loop
< LOOP_CACHING_WAIT
) {
5360 if (loop
== LOOP_ALLOC_CHUNK
) {
5365 if (allowed_chunk_alloc
) {
5366 ret
= do_chunk_alloc(trans
, root
, num_bytes
+
5367 2 * 1024 * 1024, data
,
5368 CHUNK_ALLOC_LIMITED
);
5369 allowed_chunk_alloc
= 0;
5370 done_chunk_alloc
= 1;
5371 } else if (!done_chunk_alloc
&&
5372 space_info
->force_alloc
== CHUNK_ALLOC_NO_FORCE
) {
5373 space_info
->force_alloc
= CHUNK_ALLOC_LIMITED
;
5376 if (loop
< LOOP_NO_EMPTY_SIZE
) {
5381 } else if (!ins
->objectid
) {
5385 /* we found what we needed */
5386 if (ins
->objectid
) {
5387 if (!(data
& BTRFS_BLOCK_GROUP_DATA
))
5388 trans
->block_group
= block_group
->key
.objectid
;
5390 btrfs_put_block_group(block_group
);
5397 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
5398 int dump_block_groups
)
5400 struct btrfs_block_group_cache
*cache
;
5403 spin_lock(&info
->lock
);
5404 printk(KERN_INFO
"space_info has %llu free, is %sfull\n",
5405 (unsigned long long)(info
->total_bytes
- info
->bytes_used
-
5406 info
->bytes_pinned
- info
->bytes_reserved
-
5407 info
->bytes_readonly
),
5408 (info
->full
) ? "" : "not ");
5409 printk(KERN_INFO
"space_info total=%llu, used=%llu, pinned=%llu, "
5410 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5411 (unsigned long long)info
->total_bytes
,
5412 (unsigned long long)info
->bytes_used
,
5413 (unsigned long long)info
->bytes_pinned
,
5414 (unsigned long long)info
->bytes_reserved
,
5415 (unsigned long long)info
->bytes_may_use
,
5416 (unsigned long long)info
->bytes_readonly
);
5417 spin_unlock(&info
->lock
);
5419 if (!dump_block_groups
)
5422 down_read(&info
->groups_sem
);
5424 list_for_each_entry(cache
, &info
->block_groups
[index
], list
) {
5425 spin_lock(&cache
->lock
);
5426 printk(KERN_INFO
"block group %llu has %llu bytes, %llu used "
5427 "%llu pinned %llu reserved\n",
5428 (unsigned long long)cache
->key
.objectid
,
5429 (unsigned long long)cache
->key
.offset
,
5430 (unsigned long long)btrfs_block_group_used(&cache
->item
),
5431 (unsigned long long)cache
->pinned
,
5432 (unsigned long long)cache
->reserved
);
5433 btrfs_dump_free_space(cache
, bytes
);
5434 spin_unlock(&cache
->lock
);
5436 if (++index
< BTRFS_NR_RAID_TYPES
)
5438 up_read(&info
->groups_sem
);
5441 int btrfs_reserve_extent(struct btrfs_trans_handle
*trans
,
5442 struct btrfs_root
*root
,
5443 u64 num_bytes
, u64 min_alloc_size
,
5444 u64 empty_size
, u64 hint_byte
,
5445 u64 search_end
, struct btrfs_key
*ins
,
5449 u64 search_start
= 0;
5451 data
= btrfs_get_alloc_profile(root
, data
);
5454 * the only place that sets empty_size is btrfs_realloc_node, which
5455 * is not called recursively on allocations
5457 if (empty_size
|| root
->ref_cows
)
5458 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
5459 num_bytes
+ 2 * 1024 * 1024, data
,
5460 CHUNK_ALLOC_NO_FORCE
);
5462 WARN_ON(num_bytes
< root
->sectorsize
);
5463 ret
= find_free_extent(trans
, root
, num_bytes
, empty_size
,
5464 search_start
, search_end
, hint_byte
,
5467 if (ret
== -ENOSPC
&& num_bytes
> min_alloc_size
) {
5468 num_bytes
= num_bytes
>> 1;
5469 num_bytes
= num_bytes
& ~(root
->sectorsize
- 1);
5470 num_bytes
= max(num_bytes
, min_alloc_size
);
5471 do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
5472 num_bytes
, data
, CHUNK_ALLOC_FORCE
);
5475 if (ret
== -ENOSPC
&& btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
5476 struct btrfs_space_info
*sinfo
;
5478 sinfo
= __find_space_info(root
->fs_info
, data
);
5479 printk(KERN_ERR
"btrfs allocation failed flags %llu, "
5480 "wanted %llu\n", (unsigned long long)data
,
5481 (unsigned long long)num_bytes
);
5482 dump_space_info(sinfo
, num_bytes
, 1);
5485 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, ins
->offset
);
5490 int btrfs_free_reserved_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
5492 struct btrfs_block_group_cache
*cache
;
5495 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
5497 printk(KERN_ERR
"Unable to find block group for %llu\n",
5498 (unsigned long long)start
);
5502 if (btrfs_test_opt(root
, DISCARD
))
5503 ret
= btrfs_discard_extent(root
, start
, len
, NULL
);
5505 btrfs_add_free_space(cache
, start
, len
);
5506 btrfs_update_reserved_bytes(cache
, len
, 0, 1);
5507 btrfs_put_block_group(cache
);
5509 trace_btrfs_reserved_extent_free(root
, start
, len
);
5514 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
5515 struct btrfs_root
*root
,
5516 u64 parent
, u64 root_objectid
,
5517 u64 flags
, u64 owner
, u64 offset
,
5518 struct btrfs_key
*ins
, int ref_mod
)
5521 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5522 struct btrfs_extent_item
*extent_item
;
5523 struct btrfs_extent_inline_ref
*iref
;
5524 struct btrfs_path
*path
;
5525 struct extent_buffer
*leaf
;
5530 type
= BTRFS_SHARED_DATA_REF_KEY
;
5532 type
= BTRFS_EXTENT_DATA_REF_KEY
;
5534 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
5536 path
= btrfs_alloc_path();
5540 path
->leave_spinning
= 1;
5541 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
5545 leaf
= path
->nodes
[0];
5546 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
5547 struct btrfs_extent_item
);
5548 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
5549 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
5550 btrfs_set_extent_flags(leaf
, extent_item
,
5551 flags
| BTRFS_EXTENT_FLAG_DATA
);
5553 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
5554 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
5556 struct btrfs_shared_data_ref
*ref
;
5557 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
5558 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
5559 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
5561 struct btrfs_extent_data_ref
*ref
;
5562 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
5563 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
5564 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
5565 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
5566 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
5569 btrfs_mark_buffer_dirty(path
->nodes
[0]);
5570 btrfs_free_path(path
);
5572 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1);
5574 printk(KERN_ERR
"btrfs update block group failed for %llu "
5575 "%llu\n", (unsigned long long)ins
->objectid
,
5576 (unsigned long long)ins
->offset
);
5582 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
5583 struct btrfs_root
*root
,
5584 u64 parent
, u64 root_objectid
,
5585 u64 flags
, struct btrfs_disk_key
*key
,
5586 int level
, struct btrfs_key
*ins
)
5589 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5590 struct btrfs_extent_item
*extent_item
;
5591 struct btrfs_tree_block_info
*block_info
;
5592 struct btrfs_extent_inline_ref
*iref
;
5593 struct btrfs_path
*path
;
5594 struct extent_buffer
*leaf
;
5595 u32 size
= sizeof(*extent_item
) + sizeof(*block_info
) + sizeof(*iref
);
5597 path
= btrfs_alloc_path();
5600 path
->leave_spinning
= 1;
5601 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
5605 leaf
= path
->nodes
[0];
5606 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
5607 struct btrfs_extent_item
);
5608 btrfs_set_extent_refs(leaf
, extent_item
, 1);
5609 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
5610 btrfs_set_extent_flags(leaf
, extent_item
,
5611 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
5612 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
5614 btrfs_set_tree_block_key(leaf
, block_info
, key
);
5615 btrfs_set_tree_block_level(leaf
, block_info
, level
);
5617 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
5619 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
5620 btrfs_set_extent_inline_ref_type(leaf
, iref
,
5621 BTRFS_SHARED_BLOCK_REF_KEY
);
5622 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
5624 btrfs_set_extent_inline_ref_type(leaf
, iref
,
5625 BTRFS_TREE_BLOCK_REF_KEY
);
5626 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
5629 btrfs_mark_buffer_dirty(leaf
);
5630 btrfs_free_path(path
);
5632 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1);
5634 printk(KERN_ERR
"btrfs update block group failed for %llu "
5635 "%llu\n", (unsigned long long)ins
->objectid
,
5636 (unsigned long long)ins
->offset
);
5642 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
5643 struct btrfs_root
*root
,
5644 u64 root_objectid
, u64 owner
,
5645 u64 offset
, struct btrfs_key
*ins
)
5649 BUG_ON(root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
5651 ret
= btrfs_add_delayed_data_ref(trans
, ins
->objectid
, ins
->offset
,
5652 0, root_objectid
, owner
, offset
,
5653 BTRFS_ADD_DELAYED_EXTENT
, NULL
);
5658 * this is used by the tree logging recovery code. It records that
5659 * an extent has been allocated and makes sure to clear the free
5660 * space cache bits as well
5662 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
5663 struct btrfs_root
*root
,
5664 u64 root_objectid
, u64 owner
, u64 offset
,
5665 struct btrfs_key
*ins
)
5668 struct btrfs_block_group_cache
*block_group
;
5669 struct btrfs_caching_control
*caching_ctl
;
5670 u64 start
= ins
->objectid
;
5671 u64 num_bytes
= ins
->offset
;
5673 block_group
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
5674 cache_block_group(block_group
, trans
, NULL
, 0);
5675 caching_ctl
= get_caching_control(block_group
);
5678 BUG_ON(!block_group_cache_done(block_group
));
5679 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
5682 mutex_lock(&caching_ctl
->mutex
);
5684 if (start
>= caching_ctl
->progress
) {
5685 ret
= add_excluded_extent(root
, start
, num_bytes
);
5687 } else if (start
+ num_bytes
<= caching_ctl
->progress
) {
5688 ret
= btrfs_remove_free_space(block_group
,
5692 num_bytes
= caching_ctl
->progress
- start
;
5693 ret
= btrfs_remove_free_space(block_group
,
5697 start
= caching_ctl
->progress
;
5698 num_bytes
= ins
->objectid
+ ins
->offset
-
5699 caching_ctl
->progress
;
5700 ret
= add_excluded_extent(root
, start
, num_bytes
);
5704 mutex_unlock(&caching_ctl
->mutex
);
5705 put_caching_control(caching_ctl
);
5708 ret
= btrfs_update_reserved_bytes(block_group
, ins
->offset
, 1, 1);
5710 btrfs_put_block_group(block_group
);
5711 ret
= alloc_reserved_file_extent(trans
, root
, 0, root_objectid
,
5712 0, owner
, offset
, ins
, 1);
5716 struct extent_buffer
*btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
,
5717 struct btrfs_root
*root
,
5718 u64 bytenr
, u32 blocksize
,
5721 struct extent_buffer
*buf
;
5723 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
5725 return ERR_PTR(-ENOMEM
);
5726 btrfs_set_header_generation(buf
, trans
->transid
);
5727 btrfs_set_buffer_lockdep_class(buf
, level
);
5728 btrfs_tree_lock(buf
);
5729 clean_tree_block(trans
, root
, buf
);
5731 btrfs_set_lock_blocking(buf
);
5732 btrfs_set_buffer_uptodate(buf
);
5734 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
5736 * we allow two log transactions at a time, use different
5737 * EXENT bit to differentiate dirty pages.
5739 if (root
->log_transid
% 2 == 0)
5740 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
5741 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
5743 set_extent_new(&root
->dirty_log_pages
, buf
->start
,
5744 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
5746 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
5747 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
5749 trans
->blocks_used
++;
5750 /* this returns a buffer locked for blocking */
5754 static struct btrfs_block_rsv
*
5755 use_block_rsv(struct btrfs_trans_handle
*trans
,
5756 struct btrfs_root
*root
, u32 blocksize
)
5758 struct btrfs_block_rsv
*block_rsv
;
5759 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
5762 block_rsv
= get_block_rsv(trans
, root
);
5764 if (block_rsv
->size
== 0) {
5765 ret
= reserve_metadata_bytes(trans
, root
, block_rsv
,
5768 * If we couldn't reserve metadata bytes try and use some from
5769 * the global reserve.
5771 if (ret
&& block_rsv
!= global_rsv
) {
5772 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
5775 return ERR_PTR(ret
);
5777 return ERR_PTR(ret
);
5782 ret
= block_rsv_use_bytes(block_rsv
, blocksize
);
5787 ret
= reserve_metadata_bytes(trans
, root
, block_rsv
, blocksize
,
5790 spin_lock(&block_rsv
->lock
);
5791 block_rsv
->size
+= blocksize
;
5792 spin_unlock(&block_rsv
->lock
);
5794 } else if (ret
&& block_rsv
!= global_rsv
) {
5795 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
5801 return ERR_PTR(-ENOSPC
);
5804 static void unuse_block_rsv(struct btrfs_block_rsv
*block_rsv
, u32 blocksize
)
5806 block_rsv_add_bytes(block_rsv
, blocksize
, 0);
5807 block_rsv_release_bytes(block_rsv
, NULL
, 0);
5811 * finds a free extent and does all the dirty work required for allocation
5812 * returns the key for the extent through ins, and a tree buffer for
5813 * the first block of the extent through buf.
5815 * returns the tree buffer or NULL.
5817 struct extent_buffer
*btrfs_alloc_free_block(struct btrfs_trans_handle
*trans
,
5818 struct btrfs_root
*root
, u32 blocksize
,
5819 u64 parent
, u64 root_objectid
,
5820 struct btrfs_disk_key
*key
, int level
,
5821 u64 hint
, u64 empty_size
)
5823 struct btrfs_key ins
;
5824 struct btrfs_block_rsv
*block_rsv
;
5825 struct extent_buffer
*buf
;
5830 block_rsv
= use_block_rsv(trans
, root
, blocksize
);
5831 if (IS_ERR(block_rsv
))
5832 return ERR_CAST(block_rsv
);
5834 ret
= btrfs_reserve_extent(trans
, root
, blocksize
, blocksize
,
5835 empty_size
, hint
, (u64
)-1, &ins
, 0);
5837 unuse_block_rsv(block_rsv
, blocksize
);
5838 return ERR_PTR(ret
);
5841 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
,
5843 BUG_ON(IS_ERR(buf
));
5845 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
5847 parent
= ins
.objectid
;
5848 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
5852 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
5853 struct btrfs_delayed_extent_op
*extent_op
;
5854 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
5857 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
5859 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
5860 extent_op
->flags_to_set
= flags
;
5861 extent_op
->update_key
= 1;
5862 extent_op
->update_flags
= 1;
5863 extent_op
->is_data
= 0;
5865 ret
= btrfs_add_delayed_tree_ref(trans
, ins
.objectid
,
5866 ins
.offset
, parent
, root_objectid
,
5867 level
, BTRFS_ADD_DELAYED_EXTENT
,
5874 struct walk_control
{
5875 u64 refs
[BTRFS_MAX_LEVEL
];
5876 u64 flags
[BTRFS_MAX_LEVEL
];
5877 struct btrfs_key update_progress
;
5887 #define DROP_REFERENCE 1
5888 #define UPDATE_BACKREF 2
5890 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
5891 struct btrfs_root
*root
,
5892 struct walk_control
*wc
,
5893 struct btrfs_path
*path
)
5901 struct btrfs_key key
;
5902 struct extent_buffer
*eb
;
5907 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
5908 wc
->reada_count
= wc
->reada_count
* 2 / 3;
5909 wc
->reada_count
= max(wc
->reada_count
, 2);
5911 wc
->reada_count
= wc
->reada_count
* 3 / 2;
5912 wc
->reada_count
= min_t(int, wc
->reada_count
,
5913 BTRFS_NODEPTRS_PER_BLOCK(root
));
5916 eb
= path
->nodes
[wc
->level
];
5917 nritems
= btrfs_header_nritems(eb
);
5918 blocksize
= btrfs_level_size(root
, wc
->level
- 1);
5920 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
5921 if (nread
>= wc
->reada_count
)
5925 bytenr
= btrfs_node_blockptr(eb
, slot
);
5926 generation
= btrfs_node_ptr_generation(eb
, slot
);
5928 if (slot
== path
->slots
[wc
->level
])
5931 if (wc
->stage
== UPDATE_BACKREF
&&
5932 generation
<= root
->root_key
.offset
)
5935 /* We don't lock the tree block, it's OK to be racy here */
5936 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, blocksize
,
5941 if (wc
->stage
== DROP_REFERENCE
) {
5945 if (wc
->level
== 1 &&
5946 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
5948 if (!wc
->update_ref
||
5949 generation
<= root
->root_key
.offset
)
5951 btrfs_node_key_to_cpu(eb
, &key
, slot
);
5952 ret
= btrfs_comp_cpu_keys(&key
,
5953 &wc
->update_progress
);
5957 if (wc
->level
== 1 &&
5958 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
5962 ret
= readahead_tree_block(root
, bytenr
, blocksize
,
5968 wc
->reada_slot
= slot
;
5972 * hepler to process tree block while walking down the tree.
5974 * when wc->stage == UPDATE_BACKREF, this function updates
5975 * back refs for pointers in the block.
5977 * NOTE: return value 1 means we should stop walking down.
5979 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
5980 struct btrfs_root
*root
,
5981 struct btrfs_path
*path
,
5982 struct walk_control
*wc
, int lookup_info
)
5984 int level
= wc
->level
;
5985 struct extent_buffer
*eb
= path
->nodes
[level
];
5986 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
5989 if (wc
->stage
== UPDATE_BACKREF
&&
5990 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
5994 * when reference count of tree block is 1, it won't increase
5995 * again. once full backref flag is set, we never clear it.
5998 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
5999 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
6000 BUG_ON(!path
->locks
[level
]);
6001 ret
= btrfs_lookup_extent_info(trans
, root
,
6006 BUG_ON(wc
->refs
[level
] == 0);
6009 if (wc
->stage
== DROP_REFERENCE
) {
6010 if (wc
->refs
[level
] > 1)
6013 if (path
->locks
[level
] && !wc
->keep_locks
) {
6014 btrfs_tree_unlock(eb
);
6015 path
->locks
[level
] = 0;
6020 /* wc->stage == UPDATE_BACKREF */
6021 if (!(wc
->flags
[level
] & flag
)) {
6022 BUG_ON(!path
->locks
[level
]);
6023 ret
= btrfs_inc_ref(trans
, root
, eb
, 1);
6025 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
6027 ret
= btrfs_set_disk_extent_flags(trans
, root
, eb
->start
,
6030 wc
->flags
[level
] |= flag
;
6034 * the block is shared by multiple trees, so it's not good to
6035 * keep the tree lock
6037 if (path
->locks
[level
] && level
> 0) {
6038 btrfs_tree_unlock(eb
);
6039 path
->locks
[level
] = 0;
6045 * hepler to process tree block pointer.
6047 * when wc->stage == DROP_REFERENCE, this function checks
6048 * reference count of the block pointed to. if the block
6049 * is shared and we need update back refs for the subtree
6050 * rooted at the block, this function changes wc->stage to
6051 * UPDATE_BACKREF. if the block is shared and there is no
6052 * need to update back, this function drops the reference
6055 * NOTE: return value 1 means we should stop walking down.
6057 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
6058 struct btrfs_root
*root
,
6059 struct btrfs_path
*path
,
6060 struct walk_control
*wc
, int *lookup_info
)
6066 struct btrfs_key key
;
6067 struct extent_buffer
*next
;
6068 int level
= wc
->level
;
6072 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
6073 path
->slots
[level
]);
6075 * if the lower level block was created before the snapshot
6076 * was created, we know there is no need to update back refs
6079 if (wc
->stage
== UPDATE_BACKREF
&&
6080 generation
<= root
->root_key
.offset
) {
6085 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
6086 blocksize
= btrfs_level_size(root
, level
- 1);
6088 next
= btrfs_find_tree_block(root
, bytenr
, blocksize
);
6090 next
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
6095 btrfs_tree_lock(next
);
6096 btrfs_set_lock_blocking(next
);
6098 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, blocksize
,
6099 &wc
->refs
[level
- 1],
6100 &wc
->flags
[level
- 1]);
6102 BUG_ON(wc
->refs
[level
- 1] == 0);
6105 if (wc
->stage
== DROP_REFERENCE
) {
6106 if (wc
->refs
[level
- 1] > 1) {
6108 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
6111 if (!wc
->update_ref
||
6112 generation
<= root
->root_key
.offset
)
6115 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
6116 path
->slots
[level
]);
6117 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
6121 wc
->stage
= UPDATE_BACKREF
;
6122 wc
->shared_level
= level
- 1;
6126 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
6130 if (!btrfs_buffer_uptodate(next
, generation
)) {
6131 btrfs_tree_unlock(next
);
6132 free_extent_buffer(next
);
6138 if (reada
&& level
== 1)
6139 reada_walk_down(trans
, root
, wc
, path
);
6140 next
= read_tree_block(root
, bytenr
, blocksize
, generation
);
6143 btrfs_tree_lock(next
);
6144 btrfs_set_lock_blocking(next
);
6148 BUG_ON(level
!= btrfs_header_level(next
));
6149 path
->nodes
[level
] = next
;
6150 path
->slots
[level
] = 0;
6151 path
->locks
[level
] = 1;
6157 wc
->refs
[level
- 1] = 0;
6158 wc
->flags
[level
- 1] = 0;
6159 if (wc
->stage
== DROP_REFERENCE
) {
6160 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
6161 parent
= path
->nodes
[level
]->start
;
6163 BUG_ON(root
->root_key
.objectid
!=
6164 btrfs_header_owner(path
->nodes
[level
]));
6168 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
, parent
,
6169 root
->root_key
.objectid
, level
- 1, 0);
6172 btrfs_tree_unlock(next
);
6173 free_extent_buffer(next
);
6179 * hepler to process tree block while walking up the tree.
6181 * when wc->stage == DROP_REFERENCE, this function drops
6182 * reference count on the block.
6184 * when wc->stage == UPDATE_BACKREF, this function changes
6185 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6186 * to UPDATE_BACKREF previously while processing the block.
6188 * NOTE: return value 1 means we should stop walking up.
6190 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
6191 struct btrfs_root
*root
,
6192 struct btrfs_path
*path
,
6193 struct walk_control
*wc
)
6196 int level
= wc
->level
;
6197 struct extent_buffer
*eb
= path
->nodes
[level
];
6200 if (wc
->stage
== UPDATE_BACKREF
) {
6201 BUG_ON(wc
->shared_level
< level
);
6202 if (level
< wc
->shared_level
)
6205 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
6209 wc
->stage
= DROP_REFERENCE
;
6210 wc
->shared_level
= -1;
6211 path
->slots
[level
] = 0;
6214 * check reference count again if the block isn't locked.
6215 * we should start walking down the tree again if reference
6218 if (!path
->locks
[level
]) {
6220 btrfs_tree_lock(eb
);
6221 btrfs_set_lock_blocking(eb
);
6222 path
->locks
[level
] = 1;
6224 ret
= btrfs_lookup_extent_info(trans
, root
,
6229 BUG_ON(wc
->refs
[level
] == 0);
6230 if (wc
->refs
[level
] == 1) {
6231 btrfs_tree_unlock(eb
);
6232 path
->locks
[level
] = 0;
6238 /* wc->stage == DROP_REFERENCE */
6239 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
6241 if (wc
->refs
[level
] == 1) {
6243 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
6244 ret
= btrfs_dec_ref(trans
, root
, eb
, 1);
6246 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
6249 /* make block locked assertion in clean_tree_block happy */
6250 if (!path
->locks
[level
] &&
6251 btrfs_header_generation(eb
) == trans
->transid
) {
6252 btrfs_tree_lock(eb
);
6253 btrfs_set_lock_blocking(eb
);
6254 path
->locks
[level
] = 1;
6256 clean_tree_block(trans
, root
, eb
);
6259 if (eb
== root
->node
) {
6260 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
6263 BUG_ON(root
->root_key
.objectid
!=
6264 btrfs_header_owner(eb
));
6266 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
6267 parent
= path
->nodes
[level
+ 1]->start
;
6269 BUG_ON(root
->root_key
.objectid
!=
6270 btrfs_header_owner(path
->nodes
[level
+ 1]));
6273 btrfs_free_tree_block(trans
, root
, eb
, parent
, wc
->refs
[level
] == 1);
6275 wc
->refs
[level
] = 0;
6276 wc
->flags
[level
] = 0;
6280 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
6281 struct btrfs_root
*root
,
6282 struct btrfs_path
*path
,
6283 struct walk_control
*wc
)
6285 int level
= wc
->level
;
6286 int lookup_info
= 1;
6289 while (level
>= 0) {
6290 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
6297 if (path
->slots
[level
] >=
6298 btrfs_header_nritems(path
->nodes
[level
]))
6301 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
6303 path
->slots
[level
]++;
6312 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
6313 struct btrfs_root
*root
,
6314 struct btrfs_path
*path
,
6315 struct walk_control
*wc
, int max_level
)
6317 int level
= wc
->level
;
6320 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
6321 while (level
< max_level
&& path
->nodes
[level
]) {
6323 if (path
->slots
[level
] + 1 <
6324 btrfs_header_nritems(path
->nodes
[level
])) {
6325 path
->slots
[level
]++;
6328 ret
= walk_up_proc(trans
, root
, path
, wc
);
6332 if (path
->locks
[level
]) {
6333 btrfs_tree_unlock(path
->nodes
[level
]);
6334 path
->locks
[level
] = 0;
6336 free_extent_buffer(path
->nodes
[level
]);
6337 path
->nodes
[level
] = NULL
;
6345 * drop a subvolume tree.
6347 * this function traverses the tree freeing any blocks that only
6348 * referenced by the tree.
6350 * when a shared tree block is found. this function decreases its
6351 * reference count by one. if update_ref is true, this function
6352 * also make sure backrefs for the shared block and all lower level
6353 * blocks are properly updated.
6355 int btrfs_drop_snapshot(struct btrfs_root
*root
,
6356 struct btrfs_block_rsv
*block_rsv
, int update_ref
)
6358 struct btrfs_path
*path
;
6359 struct btrfs_trans_handle
*trans
;
6360 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
6361 struct btrfs_root_item
*root_item
= &root
->root_item
;
6362 struct walk_control
*wc
;
6363 struct btrfs_key key
;
6368 path
= btrfs_alloc_path();
6371 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
6374 trans
= btrfs_start_transaction(tree_root
, 0);
6375 BUG_ON(IS_ERR(trans
));
6378 trans
->block_rsv
= block_rsv
;
6380 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
6381 level
= btrfs_header_level(root
->node
);
6382 path
->nodes
[level
] = btrfs_lock_root_node(root
);
6383 btrfs_set_lock_blocking(path
->nodes
[level
]);
6384 path
->slots
[level
] = 0;
6385 path
->locks
[level
] = 1;
6386 memset(&wc
->update_progress
, 0,
6387 sizeof(wc
->update_progress
));
6389 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
6390 memcpy(&wc
->update_progress
, &key
,
6391 sizeof(wc
->update_progress
));
6393 level
= root_item
->drop_level
;
6395 path
->lowest_level
= level
;
6396 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6397 path
->lowest_level
= 0;
6405 * unlock our path, this is safe because only this
6406 * function is allowed to delete this snapshot
6408 btrfs_unlock_up_safe(path
, 0);
6410 level
= btrfs_header_level(root
->node
);
6412 btrfs_tree_lock(path
->nodes
[level
]);
6413 btrfs_set_lock_blocking(path
->nodes
[level
]);
6415 ret
= btrfs_lookup_extent_info(trans
, root
,
6416 path
->nodes
[level
]->start
,
6417 path
->nodes
[level
]->len
,
6421 BUG_ON(wc
->refs
[level
] == 0);
6423 if (level
== root_item
->drop_level
)
6426 btrfs_tree_unlock(path
->nodes
[level
]);
6427 WARN_ON(wc
->refs
[level
] != 1);
6433 wc
->shared_level
= -1;
6434 wc
->stage
= DROP_REFERENCE
;
6435 wc
->update_ref
= update_ref
;
6437 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
6440 ret
= walk_down_tree(trans
, root
, path
, wc
);
6446 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
6453 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
6457 if (wc
->stage
== DROP_REFERENCE
) {
6459 btrfs_node_key(path
->nodes
[level
],
6460 &root_item
->drop_progress
,
6461 path
->slots
[level
]);
6462 root_item
->drop_level
= level
;
6465 BUG_ON(wc
->level
== 0);
6466 if (btrfs_should_end_transaction(trans
, tree_root
)) {
6467 ret
= btrfs_update_root(trans
, tree_root
,
6472 btrfs_end_transaction_throttle(trans
, tree_root
);
6473 trans
= btrfs_start_transaction(tree_root
, 0);
6474 BUG_ON(IS_ERR(trans
));
6476 trans
->block_rsv
= block_rsv
;
6479 btrfs_release_path(root
, path
);
6482 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
6485 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
6486 ret
= btrfs_find_last_root(tree_root
, root
->root_key
.objectid
,
6490 /* if we fail to delete the orphan item this time
6491 * around, it'll get picked up the next time.
6493 * The most common failure here is just -ENOENT.
6495 btrfs_del_orphan_item(trans
, tree_root
,
6496 root
->root_key
.objectid
);
6500 if (root
->in_radix
) {
6501 btrfs_free_fs_root(tree_root
->fs_info
, root
);
6503 free_extent_buffer(root
->node
);
6504 free_extent_buffer(root
->commit_root
);
6508 btrfs_end_transaction_throttle(trans
, tree_root
);
6510 btrfs_free_path(path
);
6515 * drop subtree rooted at tree block 'node'.
6517 * NOTE: this function will unlock and release tree block 'node'
6519 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
6520 struct btrfs_root
*root
,
6521 struct extent_buffer
*node
,
6522 struct extent_buffer
*parent
)
6524 struct btrfs_path
*path
;
6525 struct walk_control
*wc
;
6531 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
6533 path
= btrfs_alloc_path();
6537 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
6539 btrfs_free_path(path
);
6543 btrfs_assert_tree_locked(parent
);
6544 parent_level
= btrfs_header_level(parent
);
6545 extent_buffer_get(parent
);
6546 path
->nodes
[parent_level
] = parent
;
6547 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
6549 btrfs_assert_tree_locked(node
);
6550 level
= btrfs_header_level(node
);
6551 path
->nodes
[level
] = node
;
6552 path
->slots
[level
] = 0;
6553 path
->locks
[level
] = 1;
6555 wc
->refs
[parent_level
] = 1;
6556 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
6558 wc
->shared_level
= -1;
6559 wc
->stage
= DROP_REFERENCE
;
6562 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
6565 wret
= walk_down_tree(trans
, root
, path
, wc
);
6571 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
6579 btrfs_free_path(path
);
6584 static unsigned long calc_ra(unsigned long start
, unsigned long last
,
6587 return min(last
, start
+ nr
- 1);
6590 static noinline
int relocate_inode_pages(struct inode
*inode
, u64 start
,
6595 unsigned long first_index
;
6596 unsigned long last_index
;
6599 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
6600 struct file_ra_state
*ra
;
6601 struct btrfs_ordered_extent
*ordered
;
6602 unsigned int total_read
= 0;
6603 unsigned int total_dirty
= 0;
6606 ra
= kzalloc(sizeof(*ra
), GFP_NOFS
);
6610 mutex_lock(&inode
->i_mutex
);
6611 first_index
= start
>> PAGE_CACHE_SHIFT
;
6612 last_index
= (start
+ len
- 1) >> PAGE_CACHE_SHIFT
;
6614 /* make sure the dirty trick played by the caller work */
6615 ret
= invalidate_inode_pages2_range(inode
->i_mapping
,
6616 first_index
, last_index
);
6620 file_ra_state_init(ra
, inode
->i_mapping
);
6622 for (i
= first_index
; i
<= last_index
; i
++) {
6623 if (total_read
% ra
->ra_pages
== 0) {
6624 btrfs_force_ra(inode
->i_mapping
, ra
, NULL
, i
,
6625 calc_ra(i
, last_index
, ra
->ra_pages
));
6629 if (((u64
)i
<< PAGE_CACHE_SHIFT
) > i_size_read(inode
))
6631 page
= grab_cache_page(inode
->i_mapping
, i
);
6636 if (!PageUptodate(page
)) {
6637 btrfs_readpage(NULL
, page
);
6639 if (!PageUptodate(page
)) {
6641 page_cache_release(page
);
6646 wait_on_page_writeback(page
);
6648 page_start
= (u64
)page
->index
<< PAGE_CACHE_SHIFT
;
6649 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
6650 lock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
6652 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
6654 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
6656 page_cache_release(page
);
6657 btrfs_start_ordered_extent(inode
, ordered
, 1);
6658 btrfs_put_ordered_extent(ordered
);
6661 set_page_extent_mapped(page
);
6663 if (i
== first_index
)
6664 set_extent_bits(io_tree
, page_start
, page_end
,
6665 EXTENT_BOUNDARY
, GFP_NOFS
);
6666 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
6668 set_page_dirty(page
);
6671 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
6673 page_cache_release(page
);
6678 mutex_unlock(&inode
->i_mutex
);
6679 balance_dirty_pages_ratelimited_nr(inode
->i_mapping
, total_dirty
);
6683 static noinline
int relocate_data_extent(struct inode
*reloc_inode
,
6684 struct btrfs_key
*extent_key
,
6687 struct btrfs_root
*root
= BTRFS_I(reloc_inode
)->root
;
6688 struct extent_map_tree
*em_tree
= &BTRFS_I(reloc_inode
)->extent_tree
;
6689 struct extent_map
*em
;
6690 u64 start
= extent_key
->objectid
- offset
;
6691 u64 end
= start
+ extent_key
->offset
- 1;
6693 em
= alloc_extent_map(GFP_NOFS
);
6697 em
->len
= extent_key
->offset
;
6698 em
->block_len
= extent_key
->offset
;
6699 em
->block_start
= extent_key
->objectid
;
6700 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
6701 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
6703 /* setup extent map to cheat btrfs_readpage */
6704 lock_extent(&BTRFS_I(reloc_inode
)->io_tree
, start
, end
, GFP_NOFS
);
6707 write_lock(&em_tree
->lock
);
6708 ret
= add_extent_mapping(em_tree
, em
);
6709 write_unlock(&em_tree
->lock
);
6710 if (ret
!= -EEXIST
) {
6711 free_extent_map(em
);
6714 btrfs_drop_extent_cache(reloc_inode
, start
, end
, 0);
6716 unlock_extent(&BTRFS_I(reloc_inode
)->io_tree
, start
, end
, GFP_NOFS
);
6718 return relocate_inode_pages(reloc_inode
, start
, extent_key
->offset
);
6721 struct btrfs_ref_path
{
6723 u64 nodes
[BTRFS_MAX_LEVEL
];
6725 u64 root_generation
;
6732 struct btrfs_key node_keys
[BTRFS_MAX_LEVEL
];
6733 u64 new_nodes
[BTRFS_MAX_LEVEL
];
6736 struct disk_extent
{
6747 static int is_cowonly_root(u64 root_objectid
)
6749 if (root_objectid
== BTRFS_ROOT_TREE_OBJECTID
||
6750 root_objectid
== BTRFS_EXTENT_TREE_OBJECTID
||
6751 root_objectid
== BTRFS_CHUNK_TREE_OBJECTID
||
6752 root_objectid
== BTRFS_DEV_TREE_OBJECTID
||
6753 root_objectid
== BTRFS_TREE_LOG_OBJECTID
||
6754 root_objectid
== BTRFS_CSUM_TREE_OBJECTID
)
6759 static noinline
int __next_ref_path(struct btrfs_trans_handle
*trans
,
6760 struct btrfs_root
*extent_root
,
6761 struct btrfs_ref_path
*ref_path
,
6764 struct extent_buffer
*leaf
;
6765 struct btrfs_path
*path
;
6766 struct btrfs_extent_ref
*ref
;
6767 struct btrfs_key key
;
6768 struct btrfs_key found_key
;
6774 path
= btrfs_alloc_path();
6779 ref_path
->lowest_level
= -1;
6780 ref_path
->current_level
= -1;
6781 ref_path
->shared_level
= -1;
6785 level
= ref_path
->current_level
- 1;
6786 while (level
>= -1) {
6788 if (level
< ref_path
->lowest_level
)
6792 bytenr
= ref_path
->nodes
[level
];
6794 bytenr
= ref_path
->extent_start
;
6795 BUG_ON(bytenr
== 0);
6797 parent
= ref_path
->nodes
[level
+ 1];
6798 ref_path
->nodes
[level
+ 1] = 0;
6799 ref_path
->current_level
= level
;
6800 BUG_ON(parent
== 0);
6802 key
.objectid
= bytenr
;
6803 key
.offset
= parent
+ 1;
6804 key
.type
= BTRFS_EXTENT_REF_KEY
;
6806 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
, 0, 0);
6811 leaf
= path
->nodes
[0];
6812 nritems
= btrfs_header_nritems(leaf
);
6813 if (path
->slots
[0] >= nritems
) {
6814 ret
= btrfs_next_leaf(extent_root
, path
);
6819 leaf
= path
->nodes
[0];
6822 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6823 if (found_key
.objectid
== bytenr
&&
6824 found_key
.type
== BTRFS_EXTENT_REF_KEY
) {
6825 if (level
< ref_path
->shared_level
)
6826 ref_path
->shared_level
= level
;
6831 btrfs_release_path(extent_root
, path
);
6834 /* reached lowest level */
6838 level
= ref_path
->current_level
;
6839 while (level
< BTRFS_MAX_LEVEL
- 1) {
6843 bytenr
= ref_path
->nodes
[level
];
6845 bytenr
= ref_path
->extent_start
;
6847 BUG_ON(bytenr
== 0);
6849 key
.objectid
= bytenr
;
6851 key
.type
= BTRFS_EXTENT_REF_KEY
;
6853 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
, 0, 0);
6857 leaf
= path
->nodes
[0];
6858 nritems
= btrfs_header_nritems(leaf
);
6859 if (path
->slots
[0] >= nritems
) {
6860 ret
= btrfs_next_leaf(extent_root
, path
);
6864 /* the extent was freed by someone */
6865 if (ref_path
->lowest_level
== level
)
6867 btrfs_release_path(extent_root
, path
);
6870 leaf
= path
->nodes
[0];
6873 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6874 if (found_key
.objectid
!= bytenr
||
6875 found_key
.type
!= BTRFS_EXTENT_REF_KEY
) {
6876 /* the extent was freed by someone */
6877 if (ref_path
->lowest_level
== level
) {
6881 btrfs_release_path(extent_root
, path
);
6885 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
6886 struct btrfs_extent_ref
);
6887 ref_objectid
= btrfs_ref_objectid(leaf
, ref
);
6888 if (ref_objectid
< BTRFS_FIRST_FREE_OBJECTID
) {
6890 level
= (int)ref_objectid
;
6891 BUG_ON(level
>= BTRFS_MAX_LEVEL
);
6892 ref_path
->lowest_level
= level
;
6893 ref_path
->current_level
= level
;
6894 ref_path
->nodes
[level
] = bytenr
;
6896 WARN_ON(ref_objectid
!= level
);
6899 WARN_ON(level
!= -1);
6903 if (ref_path
->lowest_level
== level
) {
6904 ref_path
->owner_objectid
= ref_objectid
;
6905 ref_path
->num_refs
= btrfs_ref_num_refs(leaf
, ref
);
6909 * the block is tree root or the block isn't in reference
6912 if (found_key
.objectid
== found_key
.offset
||
6913 is_cowonly_root(btrfs_ref_root(leaf
, ref
))) {
6914 ref_path
->root_objectid
= btrfs_ref_root(leaf
, ref
);
6915 ref_path
->root_generation
=
6916 btrfs_ref_generation(leaf
, ref
);
6918 /* special reference from the tree log */
6919 ref_path
->nodes
[0] = found_key
.offset
;
6920 ref_path
->current_level
= 0;
6927 BUG_ON(ref_path
->nodes
[level
] != 0);
6928 ref_path
->nodes
[level
] = found_key
.offset
;
6929 ref_path
->current_level
= level
;
6932 * the reference was created in the running transaction,
6933 * no need to continue walking up.
6935 if (btrfs_ref_generation(leaf
, ref
) == trans
->transid
) {
6936 ref_path
->root_objectid
= btrfs_ref_root(leaf
, ref
);
6937 ref_path
->root_generation
=
6938 btrfs_ref_generation(leaf
, ref
);
6943 btrfs_release_path(extent_root
, path
);
6946 /* reached max tree level, but no tree root found. */
6949 btrfs_free_path(path
);
6953 static int btrfs_first_ref_path(struct btrfs_trans_handle
*trans
,
6954 struct btrfs_root
*extent_root
,
6955 struct btrfs_ref_path
*ref_path
,
6958 memset(ref_path
, 0, sizeof(*ref_path
));
6959 ref_path
->extent_start
= extent_start
;
6961 return __next_ref_path(trans
, extent_root
, ref_path
, 1);
6964 static int btrfs_next_ref_path(struct btrfs_trans_handle
*trans
,
6965 struct btrfs_root
*extent_root
,
6966 struct btrfs_ref_path
*ref_path
)
6968 return __next_ref_path(trans
, extent_root
, ref_path
, 0);
6971 static noinline
int get_new_locations(struct inode
*reloc_inode
,
6972 struct btrfs_key
*extent_key
,
6973 u64 offset
, int no_fragment
,
6974 struct disk_extent
**extents
,
6977 struct btrfs_root
*root
= BTRFS_I(reloc_inode
)->root
;
6978 struct btrfs_path
*path
;
6979 struct btrfs_file_extent_item
*fi
;
6980 struct extent_buffer
*leaf
;
6981 struct disk_extent
*exts
= *extents
;
6982 struct btrfs_key found_key
;
6987 int max
= *nr_extents
;
6990 WARN_ON(!no_fragment
&& *extents
);
6993 exts
= kmalloc(sizeof(*exts
) * max
, GFP_NOFS
);
6998 path
= btrfs_alloc_path();
7000 if (exts
!= *extents
)
7005 cur_pos
= extent_key
->objectid
- offset
;
7006 last_byte
= extent_key
->objectid
+ extent_key
->offset
;
7007 ret
= btrfs_lookup_file_extent(NULL
, root
, path
,
7008 btrfs_ino(reloc_inode
), cur_pos
, 0);
7017 leaf
= path
->nodes
[0];
7018 nritems
= btrfs_header_nritems(leaf
);
7019 if (path
->slots
[0] >= nritems
) {
7020 ret
= btrfs_next_leaf(root
, path
);
7025 leaf
= path
->nodes
[0];
7028 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
7029 if (found_key
.offset
!= cur_pos
||
7030 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
||
7031 found_key
.objectid
!= btrfs_ino(reloc_inode
))
7034 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
7035 struct btrfs_file_extent_item
);
7036 if (btrfs_file_extent_type(leaf
, fi
) !=
7037 BTRFS_FILE_EXTENT_REG
||
7038 btrfs_file_extent_disk_bytenr(leaf
, fi
) == 0)
7042 struct disk_extent
*old
= exts
;
7044 exts
= kzalloc(sizeof(*exts
) * max
, GFP_NOFS
);
7049 memcpy(exts
, old
, sizeof(*exts
) * nr
);
7050 if (old
!= *extents
)
7054 exts
[nr
].disk_bytenr
=
7055 btrfs_file_extent_disk_bytenr(leaf
, fi
);
7056 exts
[nr
].disk_num_bytes
=
7057 btrfs_file_extent_disk_num_bytes(leaf
, fi
);
7058 exts
[nr
].offset
= btrfs_file_extent_offset(leaf
, fi
);
7059 exts
[nr
].num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
7060 exts
[nr
].ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
7061 exts
[nr
].compression
= btrfs_file_extent_compression(leaf
, fi
);
7062 exts
[nr
].encryption
= btrfs_file_extent_encryption(leaf
, fi
);
7063 exts
[nr
].other_encoding
= btrfs_file_extent_other_encoding(leaf
,
7065 BUG_ON(exts
[nr
].offset
> 0);
7066 BUG_ON(exts
[nr
].compression
|| exts
[nr
].encryption
);
7067 BUG_ON(exts
[nr
].num_bytes
!= exts
[nr
].disk_num_bytes
);
7069 cur_pos
+= exts
[nr
].num_bytes
;
7072 if (cur_pos
+ offset
>= last_byte
)
7082 BUG_ON(cur_pos
+ offset
> last_byte
);
7083 if (cur_pos
+ offset
< last_byte
) {
7089 btrfs_free_path(path
);
7091 if (exts
!= *extents
)
7100 static noinline
int replace_one_extent(struct btrfs_trans_handle
*trans
,
7101 struct btrfs_root
*root
,
7102 struct btrfs_path
*path
,
7103 struct btrfs_key
*extent_key
,
7104 struct btrfs_key
*leaf_key
,
7105 struct btrfs_ref_path
*ref_path
,
7106 struct disk_extent
*new_extents
,
7109 struct extent_buffer
*leaf
;
7110 struct btrfs_file_extent_item
*fi
;
7111 struct inode
*inode
= NULL
;
7112 struct btrfs_key key
;
7117 u64 search_end
= (u64
)-1;
7120 int extent_locked
= 0;
7124 memcpy(&key
, leaf_key
, sizeof(key
));
7125 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
) {
7126 if (key
.objectid
< ref_path
->owner_objectid
||
7127 (key
.objectid
== ref_path
->owner_objectid
&&
7128 key
.type
< BTRFS_EXTENT_DATA_KEY
)) {
7129 key
.objectid
= ref_path
->owner_objectid
;
7130 key
.type
= BTRFS_EXTENT_DATA_KEY
;
7136 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
7140 leaf
= path
->nodes
[0];
7141 nritems
= btrfs_header_nritems(leaf
);
7143 if (extent_locked
&& ret
> 0) {
7145 * the file extent item was modified by someone
7146 * before the extent got locked.
7148 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
7149 lock_end
, GFP_NOFS
);
7153 if (path
->slots
[0] >= nritems
) {
7154 if (++nr_scaned
> 2)
7157 BUG_ON(extent_locked
);
7158 ret
= btrfs_next_leaf(root
, path
);
7163 leaf
= path
->nodes
[0];
7164 nritems
= btrfs_header_nritems(leaf
);
7167 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
7169 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
) {
7170 if ((key
.objectid
> ref_path
->owner_objectid
) ||
7171 (key
.objectid
== ref_path
->owner_objectid
&&
7172 key
.type
> BTRFS_EXTENT_DATA_KEY
) ||
7173 key
.offset
>= search_end
)
7177 if (inode
&& key
.objectid
!= btrfs_ino(inode
)) {
7178 BUG_ON(extent_locked
);
7179 btrfs_release_path(root
, path
);
7180 mutex_unlock(&inode
->i_mutex
);
7186 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
7191 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
7192 struct btrfs_file_extent_item
);
7193 extent_type
= btrfs_file_extent_type(leaf
, fi
);
7194 if ((extent_type
!= BTRFS_FILE_EXTENT_REG
&&
7195 extent_type
!= BTRFS_FILE_EXTENT_PREALLOC
) ||
7196 (btrfs_file_extent_disk_bytenr(leaf
, fi
) !=
7197 extent_key
->objectid
)) {
7203 num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
7204 ext_offset
= btrfs_file_extent_offset(leaf
, fi
);
7206 if (search_end
== (u64
)-1) {
7207 search_end
= key
.offset
- ext_offset
+
7208 btrfs_file_extent_ram_bytes(leaf
, fi
);
7211 if (!extent_locked
) {
7212 lock_start
= key
.offset
;
7213 lock_end
= lock_start
+ num_bytes
- 1;
7215 if (lock_start
> key
.offset
||
7216 lock_end
+ 1 < key
.offset
+ num_bytes
) {
7217 unlock_extent(&BTRFS_I(inode
)->io_tree
,
7218 lock_start
, lock_end
, GFP_NOFS
);
7224 btrfs_release_path(root
, path
);
7226 inode
= btrfs_iget_locked(root
->fs_info
->sb
,
7227 key
.objectid
, root
);
7228 if (inode
->i_state
& I_NEW
) {
7229 BTRFS_I(inode
)->root
= root
;
7230 BTRFS_I(inode
)->location
.objectid
=
7232 BTRFS_I(inode
)->location
.type
=
7233 BTRFS_INODE_ITEM_KEY
;
7234 BTRFS_I(inode
)->location
.offset
= 0;
7235 btrfs_read_locked_inode(inode
);
7236 unlock_new_inode(inode
);
7239 * some code call btrfs_commit_transaction while
7240 * holding the i_mutex, so we can't use mutex_lock
7243 if (is_bad_inode(inode
) ||
7244 !mutex_trylock(&inode
->i_mutex
)) {
7247 key
.offset
= (u64
)-1;
7252 if (!extent_locked
) {
7253 struct btrfs_ordered_extent
*ordered
;
7255 btrfs_release_path(root
, path
);
7257 lock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
7258 lock_end
, GFP_NOFS
);
7259 ordered
= btrfs_lookup_first_ordered_extent(inode
,
7262 ordered
->file_offset
<= lock_end
&&
7263 ordered
->file_offset
+ ordered
->len
> lock_start
) {
7264 unlock_extent(&BTRFS_I(inode
)->io_tree
,
7265 lock_start
, lock_end
, GFP_NOFS
);
7266 btrfs_start_ordered_extent(inode
, ordered
, 1);
7267 btrfs_put_ordered_extent(ordered
);
7268 key
.offset
+= num_bytes
;
7272 btrfs_put_ordered_extent(ordered
);
7278 if (nr_extents
== 1) {
7279 /* update extent pointer in place */
7280 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
7281 new_extents
[0].disk_bytenr
);
7282 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
7283 new_extents
[0].disk_num_bytes
);
7284 btrfs_mark_buffer_dirty(leaf
);
7286 btrfs_drop_extent_cache(inode
, key
.offset
,
7287 key
.offset
+ num_bytes
- 1, 0);
7289 ret
= btrfs_inc_extent_ref(trans
, root
,
7290 new_extents
[0].disk_bytenr
,
7291 new_extents
[0].disk_num_bytes
,
7293 root
->root_key
.objectid
,
7298 ret
= btrfs_free_extent(trans
, root
,
7299 extent_key
->objectid
,
7302 btrfs_header_owner(leaf
),
7303 btrfs_header_generation(leaf
),
7307 btrfs_release_path(root
, path
);
7308 key
.offset
+= num_bytes
;
7316 * drop old extent pointer at first, then insert the
7317 * new pointers one bye one
7319 btrfs_release_path(root
, path
);
7320 ret
= btrfs_drop_extents(trans
, root
, inode
, key
.offset
,
7321 key
.offset
+ num_bytes
,
7322 key
.offset
, &alloc_hint
);
7325 for (i
= 0; i
< nr_extents
; i
++) {
7326 if (ext_offset
>= new_extents
[i
].num_bytes
) {
7327 ext_offset
-= new_extents
[i
].num_bytes
;
7330 extent_len
= min(new_extents
[i
].num_bytes
-
7331 ext_offset
, num_bytes
);
7333 ret
= btrfs_insert_empty_item(trans
, root
,
7338 leaf
= path
->nodes
[0];
7339 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
7340 struct btrfs_file_extent_item
);
7341 btrfs_set_file_extent_generation(leaf
, fi
,
7343 btrfs_set_file_extent_type(leaf
, fi
,
7344 BTRFS_FILE_EXTENT_REG
);
7345 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
7346 new_extents
[i
].disk_bytenr
);
7347 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
7348 new_extents
[i
].disk_num_bytes
);
7349 btrfs_set_file_extent_ram_bytes(leaf
, fi
,
7350 new_extents
[i
].ram_bytes
);
7352 btrfs_set_file_extent_compression(leaf
, fi
,
7353 new_extents
[i
].compression
);
7354 btrfs_set_file_extent_encryption(leaf
, fi
,
7355 new_extents
[i
].encryption
);
7356 btrfs_set_file_extent_other_encoding(leaf
, fi
,
7357 new_extents
[i
].other_encoding
);
7359 btrfs_set_file_extent_num_bytes(leaf
, fi
,
7361 ext_offset
+= new_extents
[i
].offset
;
7362 btrfs_set_file_extent_offset(leaf
, fi
,
7364 btrfs_mark_buffer_dirty(leaf
);
7366 btrfs_drop_extent_cache(inode
, key
.offset
,
7367 key
.offset
+ extent_len
- 1, 0);
7369 ret
= btrfs_inc_extent_ref(trans
, root
,
7370 new_extents
[i
].disk_bytenr
,
7371 new_extents
[i
].disk_num_bytes
,
7373 root
->root_key
.objectid
,
7374 trans
->transid
, key
.objectid
);
7376 btrfs_release_path(root
, path
);
7378 inode_add_bytes(inode
, extent_len
);
7381 num_bytes
-= extent_len
;
7382 key
.offset
+= extent_len
;
7387 BUG_ON(i
>= nr_extents
);
7391 if (extent_locked
) {
7392 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
7393 lock_end
, GFP_NOFS
);
7397 if (ref_path
->owner_objectid
!= BTRFS_MULTIPLE_OBJECTIDS
&&
7398 key
.offset
>= search_end
)
7405 btrfs_release_path(root
, path
);
7407 mutex_unlock(&inode
->i_mutex
);
7408 if (extent_locked
) {
7409 unlock_extent(&BTRFS_I(inode
)->io_tree
, lock_start
,
7410 lock_end
, GFP_NOFS
);
7417 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle
*trans
,
7418 struct btrfs_root
*root
,
7419 struct extent_buffer
*buf
, u64 orig_start
)
7424 BUG_ON(btrfs_header_generation(buf
) != trans
->transid
);
7425 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
7427 level
= btrfs_header_level(buf
);
7429 struct btrfs_leaf_ref
*ref
;
7430 struct btrfs_leaf_ref
*orig_ref
;
7432 orig_ref
= btrfs_lookup_leaf_ref(root
, orig_start
);
7436 ref
= btrfs_alloc_leaf_ref(root
, orig_ref
->nritems
);
7438 btrfs_free_leaf_ref(root
, orig_ref
);
7442 ref
->nritems
= orig_ref
->nritems
;
7443 memcpy(ref
->extents
, orig_ref
->extents
,
7444 sizeof(ref
->extents
[0]) * ref
->nritems
);
7446 btrfs_free_leaf_ref(root
, orig_ref
);
7448 ref
->root_gen
= trans
->transid
;
7449 ref
->bytenr
= buf
->start
;
7450 ref
->owner
= btrfs_header_owner(buf
);
7451 ref
->generation
= btrfs_header_generation(buf
);
7453 ret
= btrfs_add_leaf_ref(root
, ref
, 0);
7455 btrfs_free_leaf_ref(root
, ref
);
7460 static noinline
int invalidate_extent_cache(struct btrfs_root
*root
,
7461 struct extent_buffer
*leaf
,
7462 struct btrfs_block_group_cache
*group
,
7463 struct btrfs_root
*target_root
)
7465 struct btrfs_key key
;
7466 struct inode
*inode
= NULL
;
7467 struct btrfs_file_extent_item
*fi
;
7468 struct extent_state
*cached_state
= NULL
;
7470 u64 skip_objectid
= 0;
7474 nritems
= btrfs_header_nritems(leaf
);
7475 for (i
= 0; i
< nritems
; i
++) {
7476 btrfs_item_key_to_cpu(leaf
, &key
, i
);
7477 if (key
.objectid
== skip_objectid
||
7478 key
.type
!= BTRFS_EXTENT_DATA_KEY
)
7480 fi
= btrfs_item_ptr(leaf
, i
, struct btrfs_file_extent_item
);
7481 if (btrfs_file_extent_type(leaf
, fi
) ==
7482 BTRFS_FILE_EXTENT_INLINE
)
7484 if (btrfs_file_extent_disk_bytenr(leaf
, fi
) == 0)
7486 if (!inode
|| btrfs_ino(inode
) != key
.objectid
) {
7488 inode
= btrfs_ilookup(target_root
->fs_info
->sb
,
7489 key
.objectid
, target_root
, 1);
7492 skip_objectid
= key
.objectid
;
7495 num_bytes
= btrfs_file_extent_num_bytes(leaf
, fi
);
7497 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, key
.offset
,
7498 key
.offset
+ num_bytes
- 1, 0, &cached_state
,
7500 btrfs_drop_extent_cache(inode
, key
.offset
,
7501 key
.offset
+ num_bytes
- 1, 1);
7502 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, key
.offset
,
7503 key
.offset
+ num_bytes
- 1, &cached_state
,
7511 static noinline
int replace_extents_in_leaf(struct btrfs_trans_handle
*trans
,
7512 struct btrfs_root
*root
,
7513 struct extent_buffer
*leaf
,
7514 struct btrfs_block_group_cache
*group
,
7515 struct inode
*reloc_inode
)
7517 struct btrfs_key key
;
7518 struct btrfs_key extent_key
;
7519 struct btrfs_file_extent_item
*fi
;
7520 struct btrfs_leaf_ref
*ref
;
7521 struct disk_extent
*new_extent
;
7530 new_extent
= kmalloc(sizeof(*new_extent
), GFP_NOFS
);
7534 ref
= btrfs_lookup_leaf_ref(root
, leaf
->start
);
7538 nritems
= btrfs_header_nritems(leaf
);
7539 for (i
= 0; i
< nritems
; i
++) {
7540 btrfs_item_key_to_cpu(leaf
, &key
, i
);
7541 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
7543 fi
= btrfs_item_ptr(leaf
, i
, struct btrfs_file_extent_item
);
7544 if (btrfs_file_extent_type(leaf
, fi
) ==
7545 BTRFS_FILE_EXTENT_INLINE
)
7547 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
7548 num_bytes
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
7553 if (bytenr
>= group
->key
.objectid
+ group
->key
.offset
||
7554 bytenr
+ num_bytes
<= group
->key
.objectid
)
7557 extent_key
.objectid
= bytenr
;
7558 extent_key
.offset
= num_bytes
;
7559 extent_key
.type
= BTRFS_EXTENT_ITEM_KEY
;
7561 ret
= get_new_locations(reloc_inode
, &extent_key
,
7562 group
->key
.objectid
, 1,
7563 &new_extent
, &nr_extent
);
7568 BUG_ON(ref
->extents
[ext_index
].bytenr
!= bytenr
);
7569 BUG_ON(ref
->extents
[ext_index
].num_bytes
!= num_bytes
);
7570 ref
->extents
[ext_index
].bytenr
= new_extent
->disk_bytenr
;
7571 ref
->extents
[ext_index
].num_bytes
= new_extent
->disk_num_bytes
;
7573 btrfs_set_file_extent_disk_bytenr(leaf
, fi
,
7574 new_extent
->disk_bytenr
);
7575 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
,
7576 new_extent
->disk_num_bytes
);
7577 btrfs_mark_buffer_dirty(leaf
);
7579 ret
= btrfs_inc_extent_ref(trans
, root
,
7580 new_extent
->disk_bytenr
,
7581 new_extent
->disk_num_bytes
,
7583 root
->root_key
.objectid
,
7584 trans
->transid
, key
.objectid
);
7587 ret
= btrfs_free_extent(trans
, root
,
7588 bytenr
, num_bytes
, leaf
->start
,
7589 btrfs_header_owner(leaf
),
7590 btrfs_header_generation(leaf
),
7596 BUG_ON(ext_index
+ 1 != ref
->nritems
);
7597 btrfs_free_leaf_ref(root
, ref
);
7601 int btrfs_free_reloc_root(struct btrfs_trans_handle
*trans
,
7602 struct btrfs_root
*root
)
7604 struct btrfs_root
*reloc_root
;
7607 if (root
->reloc_root
) {
7608 reloc_root
= root
->reloc_root
;
7609 root
->reloc_root
= NULL
;
7610 list_add(&reloc_root
->dead_list
,
7611 &root
->fs_info
->dead_reloc_roots
);
7613 btrfs_set_root_bytenr(&reloc_root
->root_item
,
7614 reloc_root
->node
->start
);
7615 btrfs_set_root_level(&root
->root_item
,
7616 btrfs_header_level(reloc_root
->node
));
7617 memset(&reloc_root
->root_item
.drop_progress
, 0,
7618 sizeof(struct btrfs_disk_key
));
7619 reloc_root
->root_item
.drop_level
= 0;
7621 ret
= btrfs_update_root(trans
, root
->fs_info
->tree_root
,
7622 &reloc_root
->root_key
,
7623 &reloc_root
->root_item
);
7629 int btrfs_drop_dead_reloc_roots(struct btrfs_root
*root
)
7631 struct btrfs_trans_handle
*trans
;
7632 struct btrfs_root
*reloc_root
;
7633 struct btrfs_root
*prev_root
= NULL
;
7634 struct list_head dead_roots
;
7638 INIT_LIST_HEAD(&dead_roots
);
7639 list_splice_init(&root
->fs_info
->dead_reloc_roots
, &dead_roots
);
7641 while (!list_empty(&dead_roots
)) {
7642 reloc_root
= list_entry(dead_roots
.prev
,
7643 struct btrfs_root
, dead_list
);
7644 list_del_init(&reloc_root
->dead_list
);
7646 BUG_ON(reloc_root
->commit_root
!= NULL
);
7648 trans
= btrfs_join_transaction(root
, 1);
7649 BUG_ON(IS_ERR(trans
));
7651 mutex_lock(&root
->fs_info
->drop_mutex
);
7652 ret
= btrfs_drop_snapshot(trans
, reloc_root
);
7655 mutex_unlock(&root
->fs_info
->drop_mutex
);
7657 nr
= trans
->blocks_used
;
7658 ret
= btrfs_end_transaction(trans
, root
);
7660 btrfs_btree_balance_dirty(root
, nr
);
7663 free_extent_buffer(reloc_root
->node
);
7665 ret
= btrfs_del_root(trans
, root
->fs_info
->tree_root
,
7666 &reloc_root
->root_key
);
7668 mutex_unlock(&root
->fs_info
->drop_mutex
);
7670 nr
= trans
->blocks_used
;
7671 ret
= btrfs_end_transaction(trans
, root
);
7673 btrfs_btree_balance_dirty(root
, nr
);
7676 prev_root
= reloc_root
;
7679 btrfs_remove_leaf_refs(prev_root
, (u64
)-1, 0);
7685 int btrfs_add_dead_reloc_root(struct btrfs_root
*root
)
7687 list_add(&root
->dead_list
, &root
->fs_info
->dead_reloc_roots
);
7691 int btrfs_cleanup_reloc_trees(struct btrfs_root
*root
)
7693 struct btrfs_root
*reloc_root
;
7694 struct btrfs_trans_handle
*trans
;
7695 struct btrfs_key location
;
7699 mutex_lock(&root
->fs_info
->tree_reloc_mutex
);
7700 ret
= btrfs_find_dead_roots(root
, BTRFS_TREE_RELOC_OBJECTID
, NULL
);
7702 found
= !list_empty(&root
->fs_info
->dead_reloc_roots
);
7703 mutex_unlock(&root
->fs_info
->tree_reloc_mutex
);
7706 trans
= btrfs_start_transaction(root
, 1);
7707 BUG_ON(IS_ERR(trans
));
7708 ret
= btrfs_commit_transaction(trans
, root
);
7712 location
.objectid
= BTRFS_DATA_RELOC_TREE_OBJECTID
;
7713 location
.offset
= (u64
)-1;
7714 location
.type
= BTRFS_ROOT_ITEM_KEY
;
7716 reloc_root
= btrfs_read_fs_root_no_name(root
->fs_info
, &location
);
7717 BUG_ON(!reloc_root
);
7718 ret
= btrfs_orphan_cleanup(reloc_root
);
7723 static noinline
int init_reloc_tree(struct btrfs_trans_handle
*trans
,
7724 struct btrfs_root
*root
)
7726 struct btrfs_root
*reloc_root
;
7727 struct extent_buffer
*eb
;
7728 struct btrfs_root_item
*root_item
;
7729 struct btrfs_key root_key
;
7732 BUG_ON(!root
->ref_cows
);
7733 if (root
->reloc_root
)
7736 root_item
= kmalloc(sizeof(*root_item
), GFP_NOFS
);
7740 ret
= btrfs_copy_root(trans
, root
, root
->commit_root
,
7741 &eb
, BTRFS_TREE_RELOC_OBJECTID
);
7744 root_key
.objectid
= BTRFS_TREE_RELOC_OBJECTID
;
7745 root_key
.offset
= root
->root_key
.objectid
;
7746 root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
7748 memcpy(root_item
, &root
->root_item
, sizeof(root_item
));
7749 btrfs_set_root_refs(root_item
, 0);
7750 btrfs_set_root_bytenr(root_item
, eb
->start
);
7751 btrfs_set_root_level(root_item
, btrfs_header_level(eb
));
7752 btrfs_set_root_generation(root_item
, trans
->transid
);
7754 btrfs_tree_unlock(eb
);
7755 free_extent_buffer(eb
);
7757 ret
= btrfs_insert_root(trans
, root
->fs_info
->tree_root
,
7758 &root_key
, root_item
);
7762 reloc_root
= btrfs_read_fs_root_no_radix(root
->fs_info
->tree_root
,
7764 BUG_ON(IS_ERR(reloc_root
));
7765 reloc_root
->last_trans
= trans
->transid
;
7766 reloc_root
->commit_root
= NULL
;
7767 reloc_root
->ref_tree
= &root
->fs_info
->reloc_ref_tree
;
7769 root
->reloc_root
= reloc_root
;
7774 * Core function of space balance.
7776 * The idea is using reloc trees to relocate tree blocks in reference
7777 * counted roots. There is one reloc tree for each subvol, and all
7778 * reloc trees share same root key objectid. Reloc trees are snapshots
7779 * of the latest committed roots of subvols (root->commit_root).
7781 * To relocate a tree block referenced by a subvol, there are two steps.
7782 * COW the block through subvol's reloc tree, then update block pointer
7783 * in the subvol to point to the new block. Since all reloc trees share
7784 * same root key objectid, doing special handing for tree blocks owned
7785 * by them is easy. Once a tree block has been COWed in one reloc tree,
7786 * we can use the resulting new block directly when the same block is
7787 * required to COW again through other reloc trees. By this way, relocated
7788 * tree blocks are shared between reloc trees, so they are also shared
7791 static noinline
int relocate_one_path(struct btrfs_trans_handle
*trans
,
7792 struct btrfs_root
*root
,
7793 struct btrfs_path
*path
,
7794 struct btrfs_key
*first_key
,
7795 struct btrfs_ref_path
*ref_path
,
7796 struct btrfs_block_group_cache
*group
,
7797 struct inode
*reloc_inode
)
7799 struct btrfs_root
*reloc_root
;
7800 struct extent_buffer
*eb
= NULL
;
7801 struct btrfs_key
*keys
;
7805 int lowest_level
= 0;
7808 if (ref_path
->owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
)
7809 lowest_level
= ref_path
->owner_objectid
;
7811 if (!root
->ref_cows
) {
7812 path
->lowest_level
= lowest_level
;
7813 ret
= btrfs_search_slot(trans
, root
, first_key
, path
, 0, 1);
7815 path
->lowest_level
= 0;
7816 btrfs_release_path(root
, path
);
7820 mutex_lock(&root
->fs_info
->tree_reloc_mutex
);
7821 ret
= init_reloc_tree(trans
, root
);
7823 reloc_root
= root
->reloc_root
;
7825 shared_level
= ref_path
->shared_level
;
7826 ref_path
->shared_level
= BTRFS_MAX_LEVEL
- 1;
7828 keys
= ref_path
->node_keys
;
7829 nodes
= ref_path
->new_nodes
;
7830 memset(&keys
[shared_level
+ 1], 0,
7831 sizeof(*keys
) * (BTRFS_MAX_LEVEL
- shared_level
- 1));
7832 memset(&nodes
[shared_level
+ 1], 0,
7833 sizeof(*nodes
) * (BTRFS_MAX_LEVEL
- shared_level
- 1));
7835 if (nodes
[lowest_level
] == 0) {
7836 path
->lowest_level
= lowest_level
;
7837 ret
= btrfs_search_slot(trans
, reloc_root
, first_key
, path
,
7840 for (level
= lowest_level
; level
< BTRFS_MAX_LEVEL
; level
++) {
7841 eb
= path
->nodes
[level
];
7842 if (!eb
|| eb
== reloc_root
->node
)
7844 nodes
[level
] = eb
->start
;
7846 btrfs_item_key_to_cpu(eb
, &keys
[level
], 0);
7848 btrfs_node_key_to_cpu(eb
, &keys
[level
], 0);
7851 ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
7852 eb
= path
->nodes
[0];
7853 ret
= replace_extents_in_leaf(trans
, reloc_root
, eb
,
7854 group
, reloc_inode
);
7857 btrfs_release_path(reloc_root
, path
);
7859 ret
= btrfs_merge_path(trans
, reloc_root
, keys
, nodes
,
7865 * replace tree blocks in the fs tree with tree blocks in
7868 ret
= btrfs_merge_path(trans
, root
, keys
, nodes
, lowest_level
);
7871 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
7872 ret
= btrfs_search_slot(trans
, reloc_root
, first_key
, path
,
7875 extent_buffer_get(path
->nodes
[0]);
7876 eb
= path
->nodes
[0];
7877 btrfs_release_path(reloc_root
, path
);
7878 ret
= invalidate_extent_cache(reloc_root
, eb
, group
, root
);
7880 free_extent_buffer(eb
);
7883 mutex_unlock(&root
->fs_info
->tree_reloc_mutex
);
7884 path
->lowest_level
= 0;
7888 static noinline
int relocate_tree_block(struct btrfs_trans_handle
*trans
,
7889 struct btrfs_root
*root
,
7890 struct btrfs_path
*path
,
7891 struct btrfs_key
*first_key
,
7892 struct btrfs_ref_path
*ref_path
)
7896 ret
= relocate_one_path(trans
, root
, path
, first_key
,
7897 ref_path
, NULL
, NULL
);
7903 static noinline
int del_extent_zero(struct btrfs_trans_handle
*trans
,
7904 struct btrfs_root
*extent_root
,
7905 struct btrfs_path
*path
,
7906 struct btrfs_key
*extent_key
)
7910 ret
= btrfs_search_slot(trans
, extent_root
, extent_key
, path
, -1, 1);
7913 ret
= btrfs_del_item(trans
, extent_root
, path
);
7915 btrfs_release_path(extent_root
, path
);
7919 static noinline
struct btrfs_root
*read_ref_root(struct btrfs_fs_info
*fs_info
,
7920 struct btrfs_ref_path
*ref_path
)
7922 struct btrfs_key root_key
;
7924 root_key
.objectid
= ref_path
->root_objectid
;
7925 root_key
.type
= BTRFS_ROOT_ITEM_KEY
;
7926 if (is_cowonly_root(ref_path
->root_objectid
))
7927 root_key
.offset
= 0;
7929 root_key
.offset
= (u64
)-1;
7931 return btrfs_read_fs_root_no_name(fs_info
, &root_key
);
7934 static noinline
int relocate_one_extent(struct btrfs_root
*extent_root
,
7935 struct btrfs_path
*path
,
7936 struct btrfs_key
*extent_key
,
7937 struct btrfs_block_group_cache
*group
,
7938 struct inode
*reloc_inode
, int pass
)
7940 struct btrfs_trans_handle
*trans
;
7941 struct btrfs_root
*found_root
;
7942 struct btrfs_ref_path
*ref_path
= NULL
;
7943 struct disk_extent
*new_extents
= NULL
;
7948 struct btrfs_key first_key
;
7952 trans
= btrfs_start_transaction(extent_root
, 1);
7953 BUG_ON(IS_ERR(trans
));
7955 if (extent_key
->objectid
== 0) {
7956 ret
= del_extent_zero(trans
, extent_root
, path
, extent_key
);
7960 ref_path
= kmalloc(sizeof(*ref_path
), GFP_NOFS
);
7966 for (loops
= 0; ; loops
++) {
7968 ret
= btrfs_first_ref_path(trans
, extent_root
, ref_path
,
7969 extent_key
->objectid
);
7971 ret
= btrfs_next_ref_path(trans
, extent_root
, ref_path
);
7978 if (ref_path
->root_objectid
== BTRFS_TREE_LOG_OBJECTID
||
7979 ref_path
->root_objectid
== BTRFS_TREE_RELOC_OBJECTID
)
7982 found_root
= read_ref_root(extent_root
->fs_info
, ref_path
);
7983 BUG_ON(!found_root
);
7985 * for reference counted tree, only process reference paths
7986 * rooted at the latest committed root.
7988 if (found_root
->ref_cows
&&
7989 ref_path
->root_generation
!= found_root
->root_key
.offset
)
7992 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
7995 * copy data extents to new locations
7997 u64 group_start
= group
->key
.objectid
;
7998 ret
= relocate_data_extent(reloc_inode
,
8007 level
= ref_path
->owner_objectid
;
8010 if (prev_block
!= ref_path
->nodes
[level
]) {
8011 struct extent_buffer
*eb
;
8012 u64 block_start
= ref_path
->nodes
[level
];
8013 u64 block_size
= btrfs_level_size(found_root
, level
);
8015 eb
= read_tree_block(found_root
, block_start
,
8021 btrfs_tree_lock(eb
);
8022 BUG_ON(level
!= btrfs_header_level(eb
));
8025 btrfs_item_key_to_cpu(eb
, &first_key
, 0);
8027 btrfs_node_key_to_cpu(eb
, &first_key
, 0);
8029 btrfs_tree_unlock(eb
);
8030 free_extent_buffer(eb
);
8031 prev_block
= block_start
;
8034 mutex_lock(&extent_root
->fs_info
->trans_mutex
);
8035 btrfs_record_root_in_trans(found_root
);
8036 mutex_unlock(&extent_root
->fs_info
->trans_mutex
);
8037 if (ref_path
->owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
) {
8039 * try to update data extent references while
8040 * keeping metadata shared between snapshots.
8043 ret
= relocate_one_path(trans
, found_root
,
8044 path
, &first_key
, ref_path
,
8045 group
, reloc_inode
);
8051 * use fallback method to process the remaining
8055 u64 group_start
= group
->key
.objectid
;
8056 new_extents
= kmalloc(sizeof(*new_extents
),
8063 ret
= get_new_locations(reloc_inode
,
8071 ret
= replace_one_extent(trans
, found_root
,
8073 &first_key
, ref_path
,
8074 new_extents
, nr_extents
);
8076 ret
= relocate_tree_block(trans
, found_root
, path
,
8077 &first_key
, ref_path
);
8084 btrfs_end_transaction(trans
, extent_root
);
8091 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
8094 u64 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
8095 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
8098 * we add in the count of missing devices because we want
8099 * to make sure that any RAID levels on a degraded FS
8100 * continue to be honored.
8102 num_devices
= root
->fs_info
->fs_devices
->rw_devices
+
8103 root
->fs_info
->fs_devices
->missing_devices
;
8105 if (num_devices
== 1) {
8106 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
8107 stripped
= flags
& ~stripped
;
8109 /* turn raid0 into single device chunks */
8110 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
8113 /* turn mirroring into duplication */
8114 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
8115 BTRFS_BLOCK_GROUP_RAID10
))
8116 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
8119 /* they already had raid on here, just return */
8120 if (flags
& stripped
)
8123 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
8124 stripped
= flags
& ~stripped
;
8126 /* switch duplicated blocks with raid1 */
8127 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
8128 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
8130 /* turn single device chunks into raid0 */
8131 return stripped
| BTRFS_BLOCK_GROUP_RAID0
;
8136 static int set_block_group_ro(struct btrfs_block_group_cache
*cache
)
8138 struct btrfs_space_info
*sinfo
= cache
->space_info
;
8145 spin_lock(&sinfo
->lock
);
8146 spin_lock(&cache
->lock
);
8147 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
8148 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
8150 if (sinfo
->bytes_used
+ sinfo
->bytes_reserved
+ sinfo
->bytes_pinned
+
8151 sinfo
->bytes_may_use
+ sinfo
->bytes_readonly
+
8152 cache
->reserved_pinned
+ num_bytes
<= sinfo
->total_bytes
) {
8153 sinfo
->bytes_readonly
+= num_bytes
;
8154 sinfo
->bytes_reserved
+= cache
->reserved_pinned
;
8155 cache
->reserved_pinned
= 0;
8160 spin_unlock(&cache
->lock
);
8161 spin_unlock(&sinfo
->lock
);
8165 int btrfs_set_block_group_ro(struct btrfs_root
*root
,
8166 struct btrfs_block_group_cache
*cache
)
8169 struct btrfs_trans_handle
*trans
;
8175 trans
= btrfs_join_transaction(root
, 1);
8176 BUG_ON(IS_ERR(trans
));
8178 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
8179 if (alloc_flags
!= cache
->flags
)
8180 do_chunk_alloc(trans
, root
, 2 * 1024 * 1024, alloc_flags
,
8183 ret
= set_block_group_ro(cache
);
8186 alloc_flags
= get_alloc_profile(root
, cache
->space_info
->flags
);
8187 ret
= do_chunk_alloc(trans
, root
, 2 * 1024 * 1024, alloc_flags
,
8191 ret
= set_block_group_ro(cache
);
8193 btrfs_end_transaction(trans
, root
);
8197 int btrfs_force_chunk_alloc(struct btrfs_trans_handle
*trans
,
8198 struct btrfs_root
*root
, u64 type
)
8200 u64 alloc_flags
= get_alloc_profile(root
, type
);
8201 return do_chunk_alloc(trans
, root
, 2 * 1024 * 1024, alloc_flags
,
8206 * helper to account the unused space of all the readonly block group in the
8207 * list. takes mirrors into account.
8209 static u64
__btrfs_get_ro_block_group_free_space(struct list_head
*groups_list
)
8211 struct btrfs_block_group_cache
*block_group
;
8215 list_for_each_entry(block_group
, groups_list
, list
) {
8216 spin_lock(&block_group
->lock
);
8218 if (!block_group
->ro
) {
8219 spin_unlock(&block_group
->lock
);
8223 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_RAID1
|
8224 BTRFS_BLOCK_GROUP_RAID10
|
8225 BTRFS_BLOCK_GROUP_DUP
))
8230 free_bytes
+= (block_group
->key
.offset
-
8231 btrfs_block_group_used(&block_group
->item
)) *
8234 spin_unlock(&block_group
->lock
);
8241 * helper to account the unused space of all the readonly block group in the
8242 * space_info. takes mirrors into account.
8244 u64
btrfs_account_ro_block_groups_free_space(struct btrfs_space_info
*sinfo
)
8249 spin_lock(&sinfo
->lock
);
8251 for(i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
8252 if (!list_empty(&sinfo
->block_groups
[i
]))
8253 free_bytes
+= __btrfs_get_ro_block_group_free_space(
8254 &sinfo
->block_groups
[i
]);
8256 spin_unlock(&sinfo
->lock
);
8261 int btrfs_set_block_group_rw(struct btrfs_root
*root
,
8262 struct btrfs_block_group_cache
*cache
)
8264 struct btrfs_space_info
*sinfo
= cache
->space_info
;
8269 spin_lock(&sinfo
->lock
);
8270 spin_lock(&cache
->lock
);
8271 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
8272 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
8273 sinfo
->bytes_readonly
-= num_bytes
;
8275 spin_unlock(&cache
->lock
);
8276 spin_unlock(&sinfo
->lock
);
8281 * checks to see if its even possible to relocate this block group.
8283 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8284 * ok to go ahead and try.
8286 int btrfs_can_relocate(struct btrfs_root
*root
, u64 bytenr
)
8288 struct btrfs_block_group_cache
*block_group
;
8289 struct btrfs_space_info
*space_info
;
8290 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
8291 struct btrfs_device
*device
;
8295 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
8297 /* odd, couldn't find the block group, leave it alone */
8301 /* no bytes used, we're good */
8302 if (!btrfs_block_group_used(&block_group
->item
))
8305 space_info
= block_group
->space_info
;
8306 spin_lock(&space_info
->lock
);
8308 full
= space_info
->full
;
8311 * if this is the last block group we have in this space, we can't
8312 * relocate it unless we're able to allocate a new chunk below.
8314 * Otherwise, we need to make sure we have room in the space to handle
8315 * all of the extents from this block group. If we can, we're good
8317 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
8318 (space_info
->bytes_used
+ space_info
->bytes_reserved
+
8319 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
8320 btrfs_block_group_used(&block_group
->item
) <
8321 space_info
->total_bytes
)) {
8322 spin_unlock(&space_info
->lock
);
8325 spin_unlock(&space_info
->lock
);
8328 * ok we don't have enough space, but maybe we have free space on our
8329 * devices to allocate new chunks for relocation, so loop through our
8330 * alloc devices and guess if we have enough space. However, if we
8331 * were marked as full, then we know there aren't enough chunks, and we
8338 mutex_lock(&root
->fs_info
->chunk_mutex
);
8339 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
8340 u64 min_free
= btrfs_block_group_used(&block_group
->item
);
8344 * check to make sure we can actually find a chunk with enough
8345 * space to fit our block group in.
8347 if (device
->total_bytes
> device
->bytes_used
+ min_free
) {
8348 ret
= find_free_dev_extent(NULL
, device
, min_free
,
8355 mutex_unlock(&root
->fs_info
->chunk_mutex
);
8357 btrfs_put_block_group(block_group
);
8361 static int find_first_block_group(struct btrfs_root
*root
,
8362 struct btrfs_path
*path
, struct btrfs_key
*key
)
8365 struct btrfs_key found_key
;
8366 struct extent_buffer
*leaf
;
8369 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
8374 slot
= path
->slots
[0];
8375 leaf
= path
->nodes
[0];
8376 if (slot
>= btrfs_header_nritems(leaf
)) {
8377 ret
= btrfs_next_leaf(root
, path
);
8384 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
8386 if (found_key
.objectid
>= key
->objectid
&&
8387 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
8397 void btrfs_put_block_group_cache(struct btrfs_fs_info
*info
)
8399 struct btrfs_block_group_cache
*block_group
;
8403 struct inode
*inode
;
8405 block_group
= btrfs_lookup_first_block_group(info
, last
);
8406 while (block_group
) {
8407 spin_lock(&block_group
->lock
);
8408 if (block_group
->iref
)
8410 spin_unlock(&block_group
->lock
);
8411 block_group
= next_block_group(info
->tree_root
,
8421 inode
= block_group
->inode
;
8422 block_group
->iref
= 0;
8423 block_group
->inode
= NULL
;
8424 spin_unlock(&block_group
->lock
);
8426 last
= block_group
->key
.objectid
+ block_group
->key
.offset
;
8427 btrfs_put_block_group(block_group
);
8431 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
8433 struct btrfs_block_group_cache
*block_group
;
8434 struct btrfs_space_info
*space_info
;
8435 struct btrfs_caching_control
*caching_ctl
;
8438 down_write(&info
->extent_commit_sem
);
8439 while (!list_empty(&info
->caching_block_groups
)) {
8440 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
8441 struct btrfs_caching_control
, list
);
8442 list_del(&caching_ctl
->list
);
8443 put_caching_control(caching_ctl
);
8445 up_write(&info
->extent_commit_sem
);
8447 spin_lock(&info
->block_group_cache_lock
);
8448 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
8449 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
8451 rb_erase(&block_group
->cache_node
,
8452 &info
->block_group_cache_tree
);
8453 spin_unlock(&info
->block_group_cache_lock
);
8455 down_write(&block_group
->space_info
->groups_sem
);
8456 list_del(&block_group
->list
);
8457 up_write(&block_group
->space_info
->groups_sem
);
8459 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
8460 wait_block_group_cache_done(block_group
);
8463 * We haven't cached this block group, which means we could
8464 * possibly have excluded extents on this block group.
8466 if (block_group
->cached
== BTRFS_CACHE_NO
)
8467 free_excluded_extents(info
->extent_root
, block_group
);
8469 btrfs_remove_free_space_cache(block_group
);
8470 btrfs_put_block_group(block_group
);
8472 spin_lock(&info
->block_group_cache_lock
);
8474 spin_unlock(&info
->block_group_cache_lock
);
8476 /* now that all the block groups are freed, go through and
8477 * free all the space_info structs. This is only called during
8478 * the final stages of unmount, and so we know nobody is
8479 * using them. We call synchronize_rcu() once before we start,
8480 * just to be on the safe side.
8484 release_global_block_rsv(info
);
8486 while(!list_empty(&info
->space_info
)) {
8487 space_info
= list_entry(info
->space_info
.next
,
8488 struct btrfs_space_info
,
8490 if (space_info
->bytes_pinned
> 0 ||
8491 space_info
->bytes_reserved
> 0) {
8493 dump_space_info(space_info
, 0, 0);
8495 list_del(&space_info
->list
);
8501 static void __link_block_group(struct btrfs_space_info
*space_info
,
8502 struct btrfs_block_group_cache
*cache
)
8504 int index
= get_block_group_index(cache
);
8506 down_write(&space_info
->groups_sem
);
8507 list_add_tail(&cache
->list
, &space_info
->block_groups
[index
]);
8508 up_write(&space_info
->groups_sem
);
8511 int btrfs_read_block_groups(struct btrfs_root
*root
)
8513 struct btrfs_path
*path
;
8515 struct btrfs_block_group_cache
*cache
;
8516 struct btrfs_fs_info
*info
= root
->fs_info
;
8517 struct btrfs_space_info
*space_info
;
8518 struct btrfs_key key
;
8519 struct btrfs_key found_key
;
8520 struct extent_buffer
*leaf
;
8524 root
= info
->extent_root
;
8527 btrfs_set_key_type(&key
, BTRFS_BLOCK_GROUP_ITEM_KEY
);
8528 path
= btrfs_alloc_path();
8532 cache_gen
= btrfs_super_cache_generation(&root
->fs_info
->super_copy
);
8533 if (cache_gen
!= 0 &&
8534 btrfs_super_generation(&root
->fs_info
->super_copy
) != cache_gen
)
8536 if (btrfs_test_opt(root
, CLEAR_CACHE
))
8538 if (!btrfs_test_opt(root
, SPACE_CACHE
) && cache_gen
)
8539 printk(KERN_INFO
"btrfs: disk space caching is enabled\n");
8542 ret
= find_first_block_group(root
, path
, &key
);
8547 leaf
= path
->nodes
[0];
8548 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
8549 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
8554 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
8556 if (!cache
->free_space_ctl
) {
8562 atomic_set(&cache
->count
, 1);
8563 spin_lock_init(&cache
->lock
);
8564 cache
->fs_info
= info
;
8565 INIT_LIST_HEAD(&cache
->list
);
8566 INIT_LIST_HEAD(&cache
->cluster_list
);
8569 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
8571 read_extent_buffer(leaf
, &cache
->item
,
8572 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
8573 sizeof(cache
->item
));
8574 memcpy(&cache
->key
, &found_key
, sizeof(found_key
));
8576 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
8577 btrfs_release_path(root
, path
);
8578 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
8579 cache
->sectorsize
= root
->sectorsize
;
8581 btrfs_init_free_space_ctl(cache
);
8584 * We need to exclude the super stripes now so that the space
8585 * info has super bytes accounted for, otherwise we'll think
8586 * we have more space than we actually do.
8588 exclude_super_stripes(root
, cache
);
8591 * check for two cases, either we are full, and therefore
8592 * don't need to bother with the caching work since we won't
8593 * find any space, or we are empty, and we can just add all
8594 * the space in and be done with it. This saves us _alot_ of
8595 * time, particularly in the full case.
8597 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
8598 cache
->last_byte_to_unpin
= (u64
)-1;
8599 cache
->cached
= BTRFS_CACHE_FINISHED
;
8600 free_excluded_extents(root
, cache
);
8601 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
8602 cache
->last_byte_to_unpin
= (u64
)-1;
8603 cache
->cached
= BTRFS_CACHE_FINISHED
;
8604 add_new_free_space(cache
, root
->fs_info
,
8606 found_key
.objectid
+
8608 free_excluded_extents(root
, cache
);
8611 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
8612 btrfs_block_group_used(&cache
->item
),
8615 cache
->space_info
= space_info
;
8616 spin_lock(&cache
->space_info
->lock
);
8617 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
8618 spin_unlock(&cache
->space_info
->lock
);
8620 __link_block_group(space_info
, cache
);
8622 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
8625 set_avail_alloc_bits(root
->fs_info
, cache
->flags
);
8626 if (btrfs_chunk_readonly(root
, cache
->key
.objectid
))
8627 set_block_group_ro(cache
);
8630 list_for_each_entry_rcu(space_info
, &root
->fs_info
->space_info
, list
) {
8631 if (!(get_alloc_profile(root
, space_info
->flags
) &
8632 (BTRFS_BLOCK_GROUP_RAID10
|
8633 BTRFS_BLOCK_GROUP_RAID1
|
8634 BTRFS_BLOCK_GROUP_DUP
)))
8637 * avoid allocating from un-mirrored block group if there are
8638 * mirrored block groups.
8640 list_for_each_entry(cache
, &space_info
->block_groups
[3], list
)
8641 set_block_group_ro(cache
);
8642 list_for_each_entry(cache
, &space_info
->block_groups
[4], list
)
8643 set_block_group_ro(cache
);
8646 init_global_block_rsv(info
);
8649 btrfs_free_path(path
);
8653 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
8654 struct btrfs_root
*root
, u64 bytes_used
,
8655 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
8659 struct btrfs_root
*extent_root
;
8660 struct btrfs_block_group_cache
*cache
;
8662 extent_root
= root
->fs_info
->extent_root
;
8664 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
8666 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
8669 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
8671 if (!cache
->free_space_ctl
) {
8676 cache
->key
.objectid
= chunk_offset
;
8677 cache
->key
.offset
= size
;
8678 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
8679 cache
->sectorsize
= root
->sectorsize
;
8680 cache
->fs_info
= root
->fs_info
;
8682 atomic_set(&cache
->count
, 1);
8683 spin_lock_init(&cache
->lock
);
8684 INIT_LIST_HEAD(&cache
->list
);
8685 INIT_LIST_HEAD(&cache
->cluster_list
);
8687 btrfs_init_free_space_ctl(cache
);
8689 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
8690 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
8691 cache
->flags
= type
;
8692 btrfs_set_block_group_flags(&cache
->item
, type
);
8694 cache
->last_byte_to_unpin
= (u64
)-1;
8695 cache
->cached
= BTRFS_CACHE_FINISHED
;
8696 exclude_super_stripes(root
, cache
);
8698 add_new_free_space(cache
, root
->fs_info
, chunk_offset
,
8699 chunk_offset
+ size
);
8701 free_excluded_extents(root
, cache
);
8703 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
8704 &cache
->space_info
);
8707 spin_lock(&cache
->space_info
->lock
);
8708 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
8709 spin_unlock(&cache
->space_info
->lock
);
8711 __link_block_group(cache
->space_info
, cache
);
8713 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
8716 ret
= btrfs_insert_item(trans
, extent_root
, &cache
->key
, &cache
->item
,
8717 sizeof(cache
->item
));
8720 set_avail_alloc_bits(extent_root
->fs_info
, type
);
8725 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
8726 struct btrfs_root
*root
, u64 group_start
)
8728 struct btrfs_path
*path
;
8729 struct btrfs_block_group_cache
*block_group
;
8730 struct btrfs_free_cluster
*cluster
;
8731 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
8732 struct btrfs_key key
;
8733 struct inode
*inode
;
8737 root
= root
->fs_info
->extent_root
;
8739 block_group
= btrfs_lookup_block_group(root
->fs_info
, group_start
);
8740 BUG_ON(!block_group
);
8741 BUG_ON(!block_group
->ro
);
8744 * Free the reserved super bytes from this block group before
8747 free_excluded_extents(root
, block_group
);
8749 memcpy(&key
, &block_group
->key
, sizeof(key
));
8750 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
8751 BTRFS_BLOCK_GROUP_RAID1
|
8752 BTRFS_BLOCK_GROUP_RAID10
))
8757 /* make sure this block group isn't part of an allocation cluster */
8758 cluster
= &root
->fs_info
->data_alloc_cluster
;
8759 spin_lock(&cluster
->refill_lock
);
8760 btrfs_return_cluster_to_free_space(block_group
, cluster
);
8761 spin_unlock(&cluster
->refill_lock
);
8764 * make sure this block group isn't part of a metadata
8765 * allocation cluster
8767 cluster
= &root
->fs_info
->meta_alloc_cluster
;
8768 spin_lock(&cluster
->refill_lock
);
8769 btrfs_return_cluster_to_free_space(block_group
, cluster
);
8770 spin_unlock(&cluster
->refill_lock
);
8772 path
= btrfs_alloc_path();
8775 inode
= lookup_free_space_inode(root
, block_group
, path
);
8776 if (!IS_ERR(inode
)) {
8777 btrfs_orphan_add(trans
, inode
);
8779 /* One for the block groups ref */
8780 spin_lock(&block_group
->lock
);
8781 if (block_group
->iref
) {
8782 block_group
->iref
= 0;
8783 block_group
->inode
= NULL
;
8784 spin_unlock(&block_group
->lock
);
8787 spin_unlock(&block_group
->lock
);
8789 /* One for our lookup ref */
8793 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
8794 key
.offset
= block_group
->key
.objectid
;
8797 ret
= btrfs_search_slot(trans
, tree_root
, &key
, path
, -1, 1);
8801 btrfs_release_path(tree_root
, path
);
8803 ret
= btrfs_del_item(trans
, tree_root
, path
);
8806 btrfs_release_path(tree_root
, path
);
8809 spin_lock(&root
->fs_info
->block_group_cache_lock
);
8810 rb_erase(&block_group
->cache_node
,
8811 &root
->fs_info
->block_group_cache_tree
);
8812 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
8814 down_write(&block_group
->space_info
->groups_sem
);
8816 * we must use list_del_init so people can check to see if they
8817 * are still on the list after taking the semaphore
8819 list_del_init(&block_group
->list
);
8820 up_write(&block_group
->space_info
->groups_sem
);
8822 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
8823 wait_block_group_cache_done(block_group
);
8825 btrfs_remove_free_space_cache(block_group
);
8827 spin_lock(&block_group
->space_info
->lock
);
8828 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
8829 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
8830 block_group
->space_info
->disk_total
-= block_group
->key
.offset
* factor
;
8831 spin_unlock(&block_group
->space_info
->lock
);
8833 memcpy(&key
, &block_group
->key
, sizeof(key
));
8835 btrfs_clear_space_info_full(root
->fs_info
);
8837 btrfs_put_block_group(block_group
);
8838 btrfs_put_block_group(block_group
);
8840 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
8846 ret
= btrfs_del_item(trans
, root
, path
);
8848 btrfs_free_path(path
);
8852 int btrfs_init_space_info(struct btrfs_fs_info
*fs_info
)
8854 struct btrfs_space_info
*space_info
;
8855 struct btrfs_super_block
*disk_super
;
8861 disk_super
= &fs_info
->super_copy
;
8862 if (!btrfs_super_root(disk_super
))
8865 features
= btrfs_super_incompat_flags(disk_super
);
8866 if (features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
8869 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
8870 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
8875 flags
= BTRFS_BLOCK_GROUP_METADATA
| BTRFS_BLOCK_GROUP_DATA
;
8876 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
8878 flags
= BTRFS_BLOCK_GROUP_METADATA
;
8879 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
8883 flags
= BTRFS_BLOCK_GROUP_DATA
;
8884 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
8890 int btrfs_error_unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
8892 return unpin_extent_range(root
, start
, end
);
8895 int btrfs_error_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
8896 u64 num_bytes
, u64
*actual_bytes
)
8898 return btrfs_discard_extent(root
, bytenr
, num_bytes
, actual_bytes
);
8901 int btrfs_trim_fs(struct btrfs_root
*root
, struct fstrim_range
*range
)
8903 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8904 struct btrfs_block_group_cache
*cache
= NULL
;
8911 cache
= btrfs_lookup_block_group(fs_info
, range
->start
);
8914 if (cache
->key
.objectid
>= (range
->start
+ range
->len
)) {
8915 btrfs_put_block_group(cache
);
8919 start
= max(range
->start
, cache
->key
.objectid
);
8920 end
= min(range
->start
+ range
->len
,
8921 cache
->key
.objectid
+ cache
->key
.offset
);
8923 if (end
- start
>= range
->minlen
) {
8924 if (!block_group_cache_done(cache
)) {
8925 ret
= cache_block_group(cache
, NULL
, root
, 0);
8927 wait_block_group_cache_done(cache
);
8929 ret
= btrfs_trim_block_group(cache
,
8935 trimmed
+= group_trimmed
;
8937 btrfs_put_block_group(cache
);
8942 cache
= next_block_group(fs_info
->tree_root
, cache
);
8945 range
->len
= trimmed
;