2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include <linux/slab.h>
30 #include "print-tree.h"
31 #include "transaction.h"
34 #include "free-space-cache.h"
36 /* control flags for do_chunk_alloc's force field
37 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
38 * if we really need one.
40 * CHUNK_ALLOC_FORCE means it must try to allocate one
42 * CHUNK_ALLOC_LIMITED means to only try and allocate one
43 * if we have very few chunks already allocated. This is
44 * used as part of the clustering code to help make sure
45 * we have a good pool of storage to cluster in, without
46 * filling the FS with empty chunks
50 CHUNK_ALLOC_NO_FORCE
= 0,
51 CHUNK_ALLOC_FORCE
= 1,
52 CHUNK_ALLOC_LIMITED
= 2,
55 static int update_block_group(struct btrfs_trans_handle
*trans
,
56 struct btrfs_root
*root
,
57 u64 bytenr
, u64 num_bytes
, int alloc
);
58 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
59 struct btrfs_root
*root
,
60 u64 bytenr
, u64 num_bytes
, u64 parent
,
61 u64 root_objectid
, u64 owner_objectid
,
62 u64 owner_offset
, int refs_to_drop
,
63 struct btrfs_delayed_extent_op
*extra_op
);
64 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
65 struct extent_buffer
*leaf
,
66 struct btrfs_extent_item
*ei
);
67 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
68 struct btrfs_root
*root
,
69 u64 parent
, u64 root_objectid
,
70 u64 flags
, u64 owner
, u64 offset
,
71 struct btrfs_key
*ins
, int ref_mod
);
72 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
73 struct btrfs_root
*root
,
74 u64 parent
, u64 root_objectid
,
75 u64 flags
, struct btrfs_disk_key
*key
,
76 int level
, struct btrfs_key
*ins
);
77 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
78 struct btrfs_root
*extent_root
, u64 alloc_bytes
,
79 u64 flags
, int force
);
80 static int find_next_key(struct btrfs_path
*path
, int level
,
81 struct btrfs_key
*key
);
82 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
83 int dump_block_groups
);
86 block_group_cache_done(struct btrfs_block_group_cache
*cache
)
89 return cache
->cached
== BTRFS_CACHE_FINISHED
;
92 static int block_group_bits(struct btrfs_block_group_cache
*cache
, u64 bits
)
94 return (cache
->flags
& bits
) == bits
;
97 static void btrfs_get_block_group(struct btrfs_block_group_cache
*cache
)
99 atomic_inc(&cache
->count
);
102 void btrfs_put_block_group(struct btrfs_block_group_cache
*cache
)
104 if (atomic_dec_and_test(&cache
->count
)) {
105 WARN_ON(cache
->pinned
> 0);
106 WARN_ON(cache
->reserved
> 0);
107 WARN_ON(cache
->reserved_pinned
> 0);
108 kfree(cache
->free_space_ctl
);
114 * this adds the block group to the fs_info rb tree for the block group
117 static int btrfs_add_block_group_cache(struct btrfs_fs_info
*info
,
118 struct btrfs_block_group_cache
*block_group
)
121 struct rb_node
*parent
= NULL
;
122 struct btrfs_block_group_cache
*cache
;
124 spin_lock(&info
->block_group_cache_lock
);
125 p
= &info
->block_group_cache_tree
.rb_node
;
129 cache
= rb_entry(parent
, struct btrfs_block_group_cache
,
131 if (block_group
->key
.objectid
< cache
->key
.objectid
) {
133 } else if (block_group
->key
.objectid
> cache
->key
.objectid
) {
136 spin_unlock(&info
->block_group_cache_lock
);
141 rb_link_node(&block_group
->cache_node
, parent
, p
);
142 rb_insert_color(&block_group
->cache_node
,
143 &info
->block_group_cache_tree
);
144 spin_unlock(&info
->block_group_cache_lock
);
150 * This will return the block group at or after bytenr if contains is 0, else
151 * it will return the block group that contains the bytenr
153 static struct btrfs_block_group_cache
*
154 block_group_cache_tree_search(struct btrfs_fs_info
*info
, u64 bytenr
,
157 struct btrfs_block_group_cache
*cache
, *ret
= NULL
;
161 spin_lock(&info
->block_group_cache_lock
);
162 n
= info
->block_group_cache_tree
.rb_node
;
165 cache
= rb_entry(n
, struct btrfs_block_group_cache
,
167 end
= cache
->key
.objectid
+ cache
->key
.offset
- 1;
168 start
= cache
->key
.objectid
;
170 if (bytenr
< start
) {
171 if (!contains
&& (!ret
|| start
< ret
->key
.objectid
))
174 } else if (bytenr
> start
) {
175 if (contains
&& bytenr
<= end
) {
186 btrfs_get_block_group(ret
);
187 spin_unlock(&info
->block_group_cache_lock
);
192 static int add_excluded_extent(struct btrfs_root
*root
,
193 u64 start
, u64 num_bytes
)
195 u64 end
= start
+ num_bytes
- 1;
196 set_extent_bits(&root
->fs_info
->freed_extents
[0],
197 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
198 set_extent_bits(&root
->fs_info
->freed_extents
[1],
199 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
203 static void free_excluded_extents(struct btrfs_root
*root
,
204 struct btrfs_block_group_cache
*cache
)
208 start
= cache
->key
.objectid
;
209 end
= start
+ cache
->key
.offset
- 1;
211 clear_extent_bits(&root
->fs_info
->freed_extents
[0],
212 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
213 clear_extent_bits(&root
->fs_info
->freed_extents
[1],
214 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
217 static int exclude_super_stripes(struct btrfs_root
*root
,
218 struct btrfs_block_group_cache
*cache
)
225 if (cache
->key
.objectid
< BTRFS_SUPER_INFO_OFFSET
) {
226 stripe_len
= BTRFS_SUPER_INFO_OFFSET
- cache
->key
.objectid
;
227 cache
->bytes_super
+= stripe_len
;
228 ret
= add_excluded_extent(root
, cache
->key
.objectid
,
233 for (i
= 0; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
234 bytenr
= btrfs_sb_offset(i
);
235 ret
= btrfs_rmap_block(&root
->fs_info
->mapping_tree
,
236 cache
->key
.objectid
, bytenr
,
237 0, &logical
, &nr
, &stripe_len
);
241 cache
->bytes_super
+= stripe_len
;
242 ret
= add_excluded_extent(root
, logical
[nr
],
252 static struct btrfs_caching_control
*
253 get_caching_control(struct btrfs_block_group_cache
*cache
)
255 struct btrfs_caching_control
*ctl
;
257 spin_lock(&cache
->lock
);
258 if (cache
->cached
!= BTRFS_CACHE_STARTED
) {
259 spin_unlock(&cache
->lock
);
263 /* We're loading it the fast way, so we don't have a caching_ctl. */
264 if (!cache
->caching_ctl
) {
265 spin_unlock(&cache
->lock
);
269 ctl
= cache
->caching_ctl
;
270 atomic_inc(&ctl
->count
);
271 spin_unlock(&cache
->lock
);
275 static void put_caching_control(struct btrfs_caching_control
*ctl
)
277 if (atomic_dec_and_test(&ctl
->count
))
282 * this is only called by cache_block_group, since we could have freed extents
283 * we need to check the pinned_extents for any extents that can't be used yet
284 * since their free space will be released as soon as the transaction commits.
286 static u64
add_new_free_space(struct btrfs_block_group_cache
*block_group
,
287 struct btrfs_fs_info
*info
, u64 start
, u64 end
)
289 u64 extent_start
, extent_end
, size
, total_added
= 0;
292 while (start
< end
) {
293 ret
= find_first_extent_bit(info
->pinned_extents
, start
,
294 &extent_start
, &extent_end
,
295 EXTENT_DIRTY
| EXTENT_UPTODATE
);
299 if (extent_start
<= start
) {
300 start
= extent_end
+ 1;
301 } else if (extent_start
> start
&& extent_start
< end
) {
302 size
= extent_start
- start
;
304 ret
= btrfs_add_free_space(block_group
, start
,
307 start
= extent_end
+ 1;
316 ret
= btrfs_add_free_space(block_group
, start
, size
);
323 static 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
);
369 if (btrfs_fs_closing(fs_info
) > 1) {
374 if (path
->slots
[0] < nritems
) {
375 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
377 ret
= find_next_key(path
, 0, &key
);
381 if (need_resched() ||
382 btrfs_next_leaf(extent_root
, path
)) {
383 caching_ctl
->progress
= last
;
384 btrfs_release_path(path
);
385 up_read(&fs_info
->extent_commit_sem
);
386 mutex_unlock(&caching_ctl
->mutex
);
390 leaf
= path
->nodes
[0];
391 nritems
= btrfs_header_nritems(leaf
);
395 if (key
.objectid
< block_group
->key
.objectid
) {
400 if (key
.objectid
>= block_group
->key
.objectid
+
401 block_group
->key
.offset
)
404 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
) {
405 total_found
+= add_new_free_space(block_group
,
408 last
= key
.objectid
+ key
.offset
;
410 if (total_found
> (1024 * 1024 * 2)) {
412 wake_up(&caching_ctl
->wait
);
419 total_found
+= add_new_free_space(block_group
, fs_info
, last
,
420 block_group
->key
.objectid
+
421 block_group
->key
.offset
);
422 caching_ctl
->progress
= (u64
)-1;
424 spin_lock(&block_group
->lock
);
425 block_group
->caching_ctl
= NULL
;
426 block_group
->cached
= BTRFS_CACHE_FINISHED
;
427 spin_unlock(&block_group
->lock
);
430 btrfs_free_path(path
);
431 up_read(&fs_info
->extent_commit_sem
);
433 free_excluded_extents(extent_root
, block_group
);
435 mutex_unlock(&caching_ctl
->mutex
);
436 wake_up(&caching_ctl
->wait
);
438 put_caching_control(caching_ctl
);
439 atomic_dec(&block_group
->space_info
->caching_threads
);
440 btrfs_put_block_group(block_group
);
445 static int cache_block_group(struct btrfs_block_group_cache
*cache
,
446 struct btrfs_trans_handle
*trans
,
447 struct btrfs_root
*root
,
450 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
451 struct btrfs_caching_control
*caching_ctl
;
452 struct task_struct
*tsk
;
456 if (cache
->cached
!= BTRFS_CACHE_NO
)
460 * We can't do the read from on-disk cache during a commit since we need
461 * to have the normal tree locking. Also if we are currently trying to
462 * allocate blocks for the tree root we can't do the fast caching since
463 * we likely hold important locks.
465 if (trans
&& (!trans
->transaction
->in_commit
) &&
466 (root
&& root
!= root
->fs_info
->tree_root
)) {
467 spin_lock(&cache
->lock
);
468 if (cache
->cached
!= BTRFS_CACHE_NO
) {
469 spin_unlock(&cache
->lock
);
472 cache
->cached
= BTRFS_CACHE_STARTED
;
473 spin_unlock(&cache
->lock
);
475 ret
= load_free_space_cache(fs_info
, cache
);
477 spin_lock(&cache
->lock
);
479 cache
->cached
= BTRFS_CACHE_FINISHED
;
480 cache
->last_byte_to_unpin
= (u64
)-1;
482 cache
->cached
= BTRFS_CACHE_NO
;
484 spin_unlock(&cache
->lock
);
486 free_excluded_extents(fs_info
->extent_root
, cache
);
494 caching_ctl
= kzalloc(sizeof(*caching_ctl
), GFP_NOFS
);
495 BUG_ON(!caching_ctl
);
497 INIT_LIST_HEAD(&caching_ctl
->list
);
498 mutex_init(&caching_ctl
->mutex
);
499 init_waitqueue_head(&caching_ctl
->wait
);
500 caching_ctl
->block_group
= cache
;
501 caching_ctl
->progress
= cache
->key
.objectid
;
502 /* one for caching kthread, one for caching block group list */
503 atomic_set(&caching_ctl
->count
, 2);
505 spin_lock(&cache
->lock
);
506 if (cache
->cached
!= BTRFS_CACHE_NO
) {
507 spin_unlock(&cache
->lock
);
511 cache
->caching_ctl
= caching_ctl
;
512 cache
->cached
= BTRFS_CACHE_STARTED
;
513 spin_unlock(&cache
->lock
);
515 down_write(&fs_info
->extent_commit_sem
);
516 list_add_tail(&caching_ctl
->list
, &fs_info
->caching_block_groups
);
517 up_write(&fs_info
->extent_commit_sem
);
519 atomic_inc(&cache
->space_info
->caching_threads
);
520 btrfs_get_block_group(cache
);
522 tsk
= kthread_run(caching_kthread
, cache
, "btrfs-cache-%llu\n",
523 cache
->key
.objectid
);
526 printk(KERN_ERR
"error running thread %d\n", ret
);
534 * return the block group that starts at or after bytenr
536 static struct btrfs_block_group_cache
*
537 btrfs_lookup_first_block_group(struct btrfs_fs_info
*info
, u64 bytenr
)
539 struct btrfs_block_group_cache
*cache
;
541 cache
= block_group_cache_tree_search(info
, bytenr
, 0);
547 * return the block group that contains the given bytenr
549 struct btrfs_block_group_cache
*btrfs_lookup_block_group(
550 struct btrfs_fs_info
*info
,
553 struct btrfs_block_group_cache
*cache
;
555 cache
= block_group_cache_tree_search(info
, bytenr
, 1);
560 static struct btrfs_space_info
*__find_space_info(struct btrfs_fs_info
*info
,
563 struct list_head
*head
= &info
->space_info
;
564 struct btrfs_space_info
*found
;
566 flags
&= BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_SYSTEM
|
567 BTRFS_BLOCK_GROUP_METADATA
;
570 list_for_each_entry_rcu(found
, head
, list
) {
571 if (found
->flags
& flags
) {
581 * after adding space to the filesystem, we need to clear the full flags
582 * on all the space infos.
584 void btrfs_clear_space_info_full(struct btrfs_fs_info
*info
)
586 struct list_head
*head
= &info
->space_info
;
587 struct btrfs_space_info
*found
;
590 list_for_each_entry_rcu(found
, head
, list
)
595 static u64
div_factor(u64 num
, int factor
)
604 static u64
div_factor_fine(u64 num
, int factor
)
613 u64
btrfs_find_block_group(struct btrfs_root
*root
,
614 u64 search_start
, u64 search_hint
, int owner
)
616 struct btrfs_block_group_cache
*cache
;
618 u64 last
= max(search_hint
, search_start
);
625 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
629 spin_lock(&cache
->lock
);
630 last
= cache
->key
.objectid
+ cache
->key
.offset
;
631 used
= btrfs_block_group_used(&cache
->item
);
633 if ((full_search
|| !cache
->ro
) &&
634 block_group_bits(cache
, BTRFS_BLOCK_GROUP_METADATA
)) {
635 if (used
+ cache
->pinned
+ cache
->reserved
<
636 div_factor(cache
->key
.offset
, factor
)) {
637 group_start
= cache
->key
.objectid
;
638 spin_unlock(&cache
->lock
);
639 btrfs_put_block_group(cache
);
643 spin_unlock(&cache
->lock
);
644 btrfs_put_block_group(cache
);
652 if (!full_search
&& factor
< 10) {
662 /* simple helper to search for an existing extent at a given offset */
663 int btrfs_lookup_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
666 struct btrfs_key key
;
667 struct btrfs_path
*path
;
669 path
= btrfs_alloc_path();
671 key
.objectid
= start
;
673 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
674 ret
= btrfs_search_slot(NULL
, root
->fs_info
->extent_root
, &key
, path
,
676 btrfs_free_path(path
);
681 * helper function to lookup reference count and flags of extent.
683 * the head node for delayed ref is used to store the sum of all the
684 * reference count modifications queued up in the rbtree. the head
685 * node may also store the extent flags to set. This way you can check
686 * to see what the reference count and extent flags would be if all of
687 * the delayed refs are not processed.
689 int btrfs_lookup_extent_info(struct btrfs_trans_handle
*trans
,
690 struct btrfs_root
*root
, u64 bytenr
,
691 u64 num_bytes
, u64
*refs
, u64
*flags
)
693 struct btrfs_delayed_ref_head
*head
;
694 struct btrfs_delayed_ref_root
*delayed_refs
;
695 struct btrfs_path
*path
;
696 struct btrfs_extent_item
*ei
;
697 struct extent_buffer
*leaf
;
698 struct btrfs_key key
;
704 path
= btrfs_alloc_path();
708 key
.objectid
= bytenr
;
709 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
710 key
.offset
= num_bytes
;
712 path
->skip_locking
= 1;
713 path
->search_commit_root
= 1;
716 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
,
722 leaf
= path
->nodes
[0];
723 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
724 if (item_size
>= sizeof(*ei
)) {
725 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
726 struct btrfs_extent_item
);
727 num_refs
= btrfs_extent_refs(leaf
, ei
);
728 extent_flags
= btrfs_extent_flags(leaf
, ei
);
730 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
731 struct btrfs_extent_item_v0
*ei0
;
732 BUG_ON(item_size
!= sizeof(*ei0
));
733 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
734 struct btrfs_extent_item_v0
);
735 num_refs
= btrfs_extent_refs_v0(leaf
, ei0
);
736 /* FIXME: this isn't correct for data */
737 extent_flags
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
742 BUG_ON(num_refs
== 0);
752 delayed_refs
= &trans
->transaction
->delayed_refs
;
753 spin_lock(&delayed_refs
->lock
);
754 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
756 if (!mutex_trylock(&head
->mutex
)) {
757 atomic_inc(&head
->node
.refs
);
758 spin_unlock(&delayed_refs
->lock
);
760 btrfs_release_path(path
);
763 * Mutex was contended, block until it's released and try
766 mutex_lock(&head
->mutex
);
767 mutex_unlock(&head
->mutex
);
768 btrfs_put_delayed_ref(&head
->node
);
771 if (head
->extent_op
&& head
->extent_op
->update_flags
)
772 extent_flags
|= head
->extent_op
->flags_to_set
;
774 BUG_ON(num_refs
== 0);
776 num_refs
+= head
->node
.ref_mod
;
777 mutex_unlock(&head
->mutex
);
779 spin_unlock(&delayed_refs
->lock
);
781 WARN_ON(num_refs
== 0);
785 *flags
= extent_flags
;
787 btrfs_free_path(path
);
792 * Back reference rules. Back refs have three main goals:
794 * 1) differentiate between all holders of references to an extent so that
795 * when a reference is dropped we can make sure it was a valid reference
796 * before freeing the extent.
798 * 2) Provide enough information to quickly find the holders of an extent
799 * if we notice a given block is corrupted or bad.
801 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
802 * maintenance. This is actually the same as #2, but with a slightly
803 * different use case.
805 * There are two kinds of back refs. The implicit back refs is optimized
806 * for pointers in non-shared tree blocks. For a given pointer in a block,
807 * back refs of this kind provide information about the block's owner tree
808 * and the pointer's key. These information allow us to find the block by
809 * b-tree searching. The full back refs is for pointers in tree blocks not
810 * referenced by their owner trees. The location of tree block is recorded
811 * in the back refs. Actually the full back refs is generic, and can be
812 * used in all cases the implicit back refs is used. The major shortcoming
813 * of the full back refs is its overhead. Every time a tree block gets
814 * COWed, we have to update back refs entry for all pointers in it.
816 * For a newly allocated tree block, we use implicit back refs for
817 * pointers in it. This means most tree related operations only involve
818 * implicit back refs. For a tree block created in old transaction, the
819 * only way to drop a reference to it is COW it. So we can detect the
820 * event that tree block loses its owner tree's reference and do the
821 * back refs conversion.
823 * When a tree block is COW'd through a tree, there are four cases:
825 * The reference count of the block is one and the tree is the block's
826 * owner tree. Nothing to do in this case.
828 * The reference count of the block is one and the tree is not the
829 * block's owner tree. In this case, full back refs is used for pointers
830 * in the block. Remove these full back refs, add implicit back refs for
831 * every pointers in the new block.
833 * The reference count of the block is greater than one and the tree is
834 * the block's owner tree. In this case, implicit back refs is used for
835 * pointers in the block. Add full back refs for every pointers in the
836 * block, increase lower level extents' reference counts. The original
837 * implicit back refs are entailed to the new block.
839 * The reference count of the block is greater than one and the tree is
840 * not the block's owner tree. Add implicit back refs for every pointer in
841 * the new block, increase lower level extents' reference count.
843 * Back Reference Key composing:
845 * The key objectid corresponds to the first byte in the extent,
846 * The key type is used to differentiate between types of back refs.
847 * There are different meanings of the key offset for different types
850 * File extents can be referenced by:
852 * - multiple snapshots, subvolumes, or different generations in one subvol
853 * - different files inside a single subvolume
854 * - different offsets inside a file (bookend extents in file.c)
856 * The extent ref structure for the implicit back refs has fields for:
858 * - Objectid of the subvolume root
859 * - objectid of the file holding the reference
860 * - original offset in the file
861 * - how many bookend extents
863 * The key offset for the implicit back refs is hash of the first
866 * The extent ref structure for the full back refs has field for:
868 * - number of pointers in the tree leaf
870 * The key offset for the implicit back refs is the first byte of
873 * When a file extent is allocated, The implicit back refs is used.
874 * the fields are filled in:
876 * (root_key.objectid, inode objectid, offset in file, 1)
878 * When a file extent is removed file truncation, we find the
879 * corresponding implicit back refs and check the following fields:
881 * (btrfs_header_owner(leaf), inode objectid, offset in file)
883 * Btree extents can be referenced by:
885 * - Different subvolumes
887 * Both the implicit back refs and the full back refs for tree blocks
888 * only consist of key. The key offset for the implicit back refs is
889 * objectid of block's owner tree. The key offset for the full back refs
890 * is the first byte of parent block.
892 * When implicit back refs is used, information about the lowest key and
893 * level of the tree block are required. These information are stored in
894 * tree block info structure.
897 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
898 static int convert_extent_item_v0(struct btrfs_trans_handle
*trans
,
899 struct btrfs_root
*root
,
900 struct btrfs_path
*path
,
901 u64 owner
, u32 extra_size
)
903 struct btrfs_extent_item
*item
;
904 struct btrfs_extent_item_v0
*ei0
;
905 struct btrfs_extent_ref_v0
*ref0
;
906 struct btrfs_tree_block_info
*bi
;
907 struct extent_buffer
*leaf
;
908 struct btrfs_key key
;
909 struct btrfs_key found_key
;
910 u32 new_size
= sizeof(*item
);
914 leaf
= path
->nodes
[0];
915 BUG_ON(btrfs_item_size_nr(leaf
, path
->slots
[0]) != sizeof(*ei0
));
917 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
918 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
919 struct btrfs_extent_item_v0
);
920 refs
= btrfs_extent_refs_v0(leaf
, ei0
);
922 if (owner
== (u64
)-1) {
924 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
925 ret
= btrfs_next_leaf(root
, path
);
929 leaf
= path
->nodes
[0];
931 btrfs_item_key_to_cpu(leaf
, &found_key
,
933 BUG_ON(key
.objectid
!= found_key
.objectid
);
934 if (found_key
.type
!= BTRFS_EXTENT_REF_V0_KEY
) {
938 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
939 struct btrfs_extent_ref_v0
);
940 owner
= btrfs_ref_objectid_v0(leaf
, ref0
);
944 btrfs_release_path(path
);
946 if (owner
< BTRFS_FIRST_FREE_OBJECTID
)
947 new_size
+= sizeof(*bi
);
949 new_size
-= sizeof(*ei0
);
950 ret
= btrfs_search_slot(trans
, root
, &key
, path
,
951 new_size
+ extra_size
, 1);
956 ret
= btrfs_extend_item(trans
, root
, path
, new_size
);
958 leaf
= path
->nodes
[0];
959 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
960 btrfs_set_extent_refs(leaf
, item
, refs
);
961 /* FIXME: get real generation */
962 btrfs_set_extent_generation(leaf
, item
, 0);
963 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
964 btrfs_set_extent_flags(leaf
, item
,
965 BTRFS_EXTENT_FLAG_TREE_BLOCK
|
966 BTRFS_BLOCK_FLAG_FULL_BACKREF
);
967 bi
= (struct btrfs_tree_block_info
*)(item
+ 1);
968 /* FIXME: get first key of the block */
969 memset_extent_buffer(leaf
, 0, (unsigned long)bi
, sizeof(*bi
));
970 btrfs_set_tree_block_level(leaf
, bi
, (int)owner
);
972 btrfs_set_extent_flags(leaf
, item
, BTRFS_EXTENT_FLAG_DATA
);
974 btrfs_mark_buffer_dirty(leaf
);
979 static u64
hash_extent_data_ref(u64 root_objectid
, u64 owner
, u64 offset
)
981 u32 high_crc
= ~(u32
)0;
982 u32 low_crc
= ~(u32
)0;
985 lenum
= cpu_to_le64(root_objectid
);
986 high_crc
= crc32c(high_crc
, &lenum
, sizeof(lenum
));
987 lenum
= cpu_to_le64(owner
);
988 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
989 lenum
= cpu_to_le64(offset
);
990 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
992 return ((u64
)high_crc
<< 31) ^ (u64
)low_crc
;
995 static u64
hash_extent_data_ref_item(struct extent_buffer
*leaf
,
996 struct btrfs_extent_data_ref
*ref
)
998 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf
, ref
),
999 btrfs_extent_data_ref_objectid(leaf
, ref
),
1000 btrfs_extent_data_ref_offset(leaf
, ref
));
1003 static int match_extent_data_ref(struct extent_buffer
*leaf
,
1004 struct btrfs_extent_data_ref
*ref
,
1005 u64 root_objectid
, u64 owner
, u64 offset
)
1007 if (btrfs_extent_data_ref_root(leaf
, ref
) != root_objectid
||
1008 btrfs_extent_data_ref_objectid(leaf
, ref
) != owner
||
1009 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
1014 static noinline
int lookup_extent_data_ref(struct btrfs_trans_handle
*trans
,
1015 struct btrfs_root
*root
,
1016 struct btrfs_path
*path
,
1017 u64 bytenr
, u64 parent
,
1019 u64 owner
, u64 offset
)
1021 struct btrfs_key key
;
1022 struct btrfs_extent_data_ref
*ref
;
1023 struct extent_buffer
*leaf
;
1029 key
.objectid
= bytenr
;
1031 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1032 key
.offset
= parent
;
1034 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1035 key
.offset
= hash_extent_data_ref(root_objectid
,
1040 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1049 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1050 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1051 btrfs_release_path(path
);
1052 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1063 leaf
= path
->nodes
[0];
1064 nritems
= btrfs_header_nritems(leaf
);
1066 if (path
->slots
[0] >= nritems
) {
1067 ret
= btrfs_next_leaf(root
, path
);
1073 leaf
= path
->nodes
[0];
1074 nritems
= btrfs_header_nritems(leaf
);
1078 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1079 if (key
.objectid
!= bytenr
||
1080 key
.type
!= BTRFS_EXTENT_DATA_REF_KEY
)
1083 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1084 struct btrfs_extent_data_ref
);
1086 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1089 btrfs_release_path(path
);
1101 static noinline
int insert_extent_data_ref(struct btrfs_trans_handle
*trans
,
1102 struct btrfs_root
*root
,
1103 struct btrfs_path
*path
,
1104 u64 bytenr
, u64 parent
,
1105 u64 root_objectid
, u64 owner
,
1106 u64 offset
, int refs_to_add
)
1108 struct btrfs_key key
;
1109 struct extent_buffer
*leaf
;
1114 key
.objectid
= bytenr
;
1116 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1117 key
.offset
= parent
;
1118 size
= sizeof(struct btrfs_shared_data_ref
);
1120 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1121 key
.offset
= hash_extent_data_ref(root_objectid
,
1123 size
= sizeof(struct btrfs_extent_data_ref
);
1126 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, size
);
1127 if (ret
&& ret
!= -EEXIST
)
1130 leaf
= path
->nodes
[0];
1132 struct btrfs_shared_data_ref
*ref
;
1133 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1134 struct btrfs_shared_data_ref
);
1136 btrfs_set_shared_data_ref_count(leaf
, ref
, refs_to_add
);
1138 num_refs
= btrfs_shared_data_ref_count(leaf
, ref
);
1139 num_refs
+= refs_to_add
;
1140 btrfs_set_shared_data_ref_count(leaf
, ref
, num_refs
);
1143 struct btrfs_extent_data_ref
*ref
;
1144 while (ret
== -EEXIST
) {
1145 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1146 struct btrfs_extent_data_ref
);
1147 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1150 btrfs_release_path(path
);
1152 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1154 if (ret
&& ret
!= -EEXIST
)
1157 leaf
= path
->nodes
[0];
1159 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1160 struct btrfs_extent_data_ref
);
1162 btrfs_set_extent_data_ref_root(leaf
, ref
,
1164 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
1165 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
1166 btrfs_set_extent_data_ref_count(leaf
, ref
, refs_to_add
);
1168 num_refs
= btrfs_extent_data_ref_count(leaf
, ref
);
1169 num_refs
+= refs_to_add
;
1170 btrfs_set_extent_data_ref_count(leaf
, ref
, num_refs
);
1173 btrfs_mark_buffer_dirty(leaf
);
1176 btrfs_release_path(path
);
1180 static noinline
int remove_extent_data_ref(struct btrfs_trans_handle
*trans
,
1181 struct btrfs_root
*root
,
1182 struct btrfs_path
*path
,
1185 struct btrfs_key key
;
1186 struct btrfs_extent_data_ref
*ref1
= NULL
;
1187 struct btrfs_shared_data_ref
*ref2
= NULL
;
1188 struct extent_buffer
*leaf
;
1192 leaf
= path
->nodes
[0];
1193 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1195 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1196 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1197 struct btrfs_extent_data_ref
);
1198 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1199 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1200 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1201 struct btrfs_shared_data_ref
);
1202 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1203 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1204 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1205 struct btrfs_extent_ref_v0
*ref0
;
1206 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1207 struct btrfs_extent_ref_v0
);
1208 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1214 BUG_ON(num_refs
< refs_to_drop
);
1215 num_refs
-= refs_to_drop
;
1217 if (num_refs
== 0) {
1218 ret
= btrfs_del_item(trans
, root
, path
);
1220 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
)
1221 btrfs_set_extent_data_ref_count(leaf
, ref1
, num_refs
);
1222 else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
)
1223 btrfs_set_shared_data_ref_count(leaf
, ref2
, num_refs
);
1224 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1226 struct btrfs_extent_ref_v0
*ref0
;
1227 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1228 struct btrfs_extent_ref_v0
);
1229 btrfs_set_ref_count_v0(leaf
, ref0
, num_refs
);
1232 btrfs_mark_buffer_dirty(leaf
);
1237 static noinline u32
extent_data_ref_count(struct btrfs_root
*root
,
1238 struct btrfs_path
*path
,
1239 struct btrfs_extent_inline_ref
*iref
)
1241 struct btrfs_key key
;
1242 struct extent_buffer
*leaf
;
1243 struct btrfs_extent_data_ref
*ref1
;
1244 struct btrfs_shared_data_ref
*ref2
;
1247 leaf
= path
->nodes
[0];
1248 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1250 if (btrfs_extent_inline_ref_type(leaf
, iref
) ==
1251 BTRFS_EXTENT_DATA_REF_KEY
) {
1252 ref1
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1253 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1255 ref2
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1256 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1258 } else if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1259 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1260 struct btrfs_extent_data_ref
);
1261 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1262 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1263 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1264 struct btrfs_shared_data_ref
);
1265 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1266 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1267 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1268 struct btrfs_extent_ref_v0
*ref0
;
1269 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1270 struct btrfs_extent_ref_v0
);
1271 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1279 static noinline
int lookup_tree_block_ref(struct btrfs_trans_handle
*trans
,
1280 struct btrfs_root
*root
,
1281 struct btrfs_path
*path
,
1282 u64 bytenr
, u64 parent
,
1285 struct btrfs_key key
;
1288 key
.objectid
= bytenr
;
1290 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1291 key
.offset
= parent
;
1293 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1294 key
.offset
= root_objectid
;
1297 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1300 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1301 if (ret
== -ENOENT
&& parent
) {
1302 btrfs_release_path(path
);
1303 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1304 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1312 static noinline
int insert_tree_block_ref(struct btrfs_trans_handle
*trans
,
1313 struct btrfs_root
*root
,
1314 struct btrfs_path
*path
,
1315 u64 bytenr
, u64 parent
,
1318 struct btrfs_key key
;
1321 key
.objectid
= bytenr
;
1323 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1324 key
.offset
= parent
;
1326 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1327 key
.offset
= root_objectid
;
1330 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
1331 btrfs_release_path(path
);
1335 static inline int extent_ref_type(u64 parent
, u64 owner
)
1338 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1340 type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1342 type
= BTRFS_TREE_BLOCK_REF_KEY
;
1345 type
= BTRFS_SHARED_DATA_REF_KEY
;
1347 type
= BTRFS_EXTENT_DATA_REF_KEY
;
1352 static int find_next_key(struct btrfs_path
*path
, int level
,
1353 struct btrfs_key
*key
)
1356 for (; level
< BTRFS_MAX_LEVEL
; level
++) {
1357 if (!path
->nodes
[level
])
1359 if (path
->slots
[level
] + 1 >=
1360 btrfs_header_nritems(path
->nodes
[level
]))
1363 btrfs_item_key_to_cpu(path
->nodes
[level
], key
,
1364 path
->slots
[level
] + 1);
1366 btrfs_node_key_to_cpu(path
->nodes
[level
], key
,
1367 path
->slots
[level
] + 1);
1374 * look for inline back ref. if back ref is found, *ref_ret is set
1375 * to the address of inline back ref, and 0 is returned.
1377 * if back ref isn't found, *ref_ret is set to the address where it
1378 * should be inserted, and -ENOENT is returned.
1380 * if insert is true and there are too many inline back refs, the path
1381 * points to the extent item, and -EAGAIN is returned.
1383 * NOTE: inline back refs are ordered in the same way that back ref
1384 * items in the tree are ordered.
1386 static noinline_for_stack
1387 int lookup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1388 struct btrfs_root
*root
,
1389 struct btrfs_path
*path
,
1390 struct btrfs_extent_inline_ref
**ref_ret
,
1391 u64 bytenr
, u64 num_bytes
,
1392 u64 parent
, u64 root_objectid
,
1393 u64 owner
, u64 offset
, int insert
)
1395 struct btrfs_key key
;
1396 struct extent_buffer
*leaf
;
1397 struct btrfs_extent_item
*ei
;
1398 struct btrfs_extent_inline_ref
*iref
;
1409 key
.objectid
= bytenr
;
1410 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1411 key
.offset
= num_bytes
;
1413 want
= extent_ref_type(parent
, owner
);
1415 extra_size
= btrfs_extent_inline_ref_size(want
);
1416 path
->keep_locks
= 1;
1419 ret
= btrfs_search_slot(trans
, root
, &key
, path
, extra_size
, 1);
1426 leaf
= path
->nodes
[0];
1427 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1428 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1429 if (item_size
< sizeof(*ei
)) {
1434 ret
= convert_extent_item_v0(trans
, root
, path
, owner
,
1440 leaf
= path
->nodes
[0];
1441 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1444 BUG_ON(item_size
< sizeof(*ei
));
1446 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1447 flags
= btrfs_extent_flags(leaf
, ei
);
1449 ptr
= (unsigned long)(ei
+ 1);
1450 end
= (unsigned long)ei
+ item_size
;
1452 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
) {
1453 ptr
+= sizeof(struct btrfs_tree_block_info
);
1456 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_DATA
));
1465 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1466 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1470 ptr
+= btrfs_extent_inline_ref_size(type
);
1474 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1475 struct btrfs_extent_data_ref
*dref
;
1476 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1477 if (match_extent_data_ref(leaf
, dref
, root_objectid
,
1482 if (hash_extent_data_ref_item(leaf
, dref
) <
1483 hash_extent_data_ref(root_objectid
, owner
, offset
))
1487 ref_offset
= btrfs_extent_inline_ref_offset(leaf
, iref
);
1489 if (parent
== ref_offset
) {
1493 if (ref_offset
< parent
)
1496 if (root_objectid
== ref_offset
) {
1500 if (ref_offset
< root_objectid
)
1504 ptr
+= btrfs_extent_inline_ref_size(type
);
1506 if (err
== -ENOENT
&& insert
) {
1507 if (item_size
+ extra_size
>=
1508 BTRFS_MAX_EXTENT_ITEM_SIZE(root
)) {
1513 * To add new inline back ref, we have to make sure
1514 * there is no corresponding back ref item.
1515 * For simplicity, we just do not add new inline back
1516 * ref if there is any kind of item for this block
1518 if (find_next_key(path
, 0, &key
) == 0 &&
1519 key
.objectid
== bytenr
&&
1520 key
.type
< BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1525 *ref_ret
= (struct btrfs_extent_inline_ref
*)ptr
;
1528 path
->keep_locks
= 0;
1529 btrfs_unlock_up_safe(path
, 1);
1535 * helper to add new inline back ref
1537 static noinline_for_stack
1538 int setup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1539 struct btrfs_root
*root
,
1540 struct btrfs_path
*path
,
1541 struct btrfs_extent_inline_ref
*iref
,
1542 u64 parent
, u64 root_objectid
,
1543 u64 owner
, u64 offset
, int refs_to_add
,
1544 struct btrfs_delayed_extent_op
*extent_op
)
1546 struct extent_buffer
*leaf
;
1547 struct btrfs_extent_item
*ei
;
1550 unsigned long item_offset
;
1556 leaf
= path
->nodes
[0];
1557 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1558 item_offset
= (unsigned long)iref
- (unsigned long)ei
;
1560 type
= extent_ref_type(parent
, owner
);
1561 size
= btrfs_extent_inline_ref_size(type
);
1563 ret
= btrfs_extend_item(trans
, root
, path
, size
);
1565 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1566 refs
= btrfs_extent_refs(leaf
, ei
);
1567 refs
+= refs_to_add
;
1568 btrfs_set_extent_refs(leaf
, ei
, refs
);
1570 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1572 ptr
= (unsigned long)ei
+ item_offset
;
1573 end
= (unsigned long)ei
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
1574 if (ptr
< end
- size
)
1575 memmove_extent_buffer(leaf
, ptr
+ size
, ptr
,
1578 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1579 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
1580 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1581 struct btrfs_extent_data_ref
*dref
;
1582 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1583 btrfs_set_extent_data_ref_root(leaf
, dref
, root_objectid
);
1584 btrfs_set_extent_data_ref_objectid(leaf
, dref
, owner
);
1585 btrfs_set_extent_data_ref_offset(leaf
, dref
, offset
);
1586 btrfs_set_extent_data_ref_count(leaf
, dref
, refs_to_add
);
1587 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1588 struct btrfs_shared_data_ref
*sref
;
1589 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1590 btrfs_set_shared_data_ref_count(leaf
, sref
, refs_to_add
);
1591 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1592 } else if (type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
1593 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1595 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
1597 btrfs_mark_buffer_dirty(leaf
);
1601 static int lookup_extent_backref(struct btrfs_trans_handle
*trans
,
1602 struct btrfs_root
*root
,
1603 struct btrfs_path
*path
,
1604 struct btrfs_extent_inline_ref
**ref_ret
,
1605 u64 bytenr
, u64 num_bytes
, u64 parent
,
1606 u64 root_objectid
, u64 owner
, u64 offset
)
1610 ret
= lookup_inline_extent_backref(trans
, root
, path
, ref_ret
,
1611 bytenr
, num_bytes
, parent
,
1612 root_objectid
, owner
, offset
, 0);
1616 btrfs_release_path(path
);
1619 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1620 ret
= lookup_tree_block_ref(trans
, root
, path
, bytenr
, parent
,
1623 ret
= lookup_extent_data_ref(trans
, root
, path
, bytenr
, parent
,
1624 root_objectid
, owner
, offset
);
1630 * helper to update/remove inline back ref
1632 static noinline_for_stack
1633 int update_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1634 struct btrfs_root
*root
,
1635 struct btrfs_path
*path
,
1636 struct btrfs_extent_inline_ref
*iref
,
1638 struct btrfs_delayed_extent_op
*extent_op
)
1640 struct extent_buffer
*leaf
;
1641 struct btrfs_extent_item
*ei
;
1642 struct btrfs_extent_data_ref
*dref
= NULL
;
1643 struct btrfs_shared_data_ref
*sref
= NULL
;
1652 leaf
= path
->nodes
[0];
1653 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1654 refs
= btrfs_extent_refs(leaf
, ei
);
1655 WARN_ON(refs_to_mod
< 0 && refs
+ refs_to_mod
<= 0);
1656 refs
+= refs_to_mod
;
1657 btrfs_set_extent_refs(leaf
, ei
, refs
);
1659 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1661 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1663 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1664 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1665 refs
= btrfs_extent_data_ref_count(leaf
, dref
);
1666 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1667 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1668 refs
= btrfs_shared_data_ref_count(leaf
, sref
);
1671 BUG_ON(refs_to_mod
!= -1);
1674 BUG_ON(refs_to_mod
< 0 && refs
< -refs_to_mod
);
1675 refs
+= refs_to_mod
;
1678 if (type
== BTRFS_EXTENT_DATA_REF_KEY
)
1679 btrfs_set_extent_data_ref_count(leaf
, dref
, refs
);
1681 btrfs_set_shared_data_ref_count(leaf
, sref
, refs
);
1683 size
= btrfs_extent_inline_ref_size(type
);
1684 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1685 ptr
= (unsigned long)iref
;
1686 end
= (unsigned long)ei
+ item_size
;
1687 if (ptr
+ size
< end
)
1688 memmove_extent_buffer(leaf
, ptr
, ptr
+ size
,
1691 ret
= btrfs_truncate_item(trans
, root
, path
, item_size
, 1);
1693 btrfs_mark_buffer_dirty(leaf
);
1697 static noinline_for_stack
1698 int insert_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1699 struct btrfs_root
*root
,
1700 struct btrfs_path
*path
,
1701 u64 bytenr
, u64 num_bytes
, u64 parent
,
1702 u64 root_objectid
, u64 owner
,
1703 u64 offset
, int refs_to_add
,
1704 struct btrfs_delayed_extent_op
*extent_op
)
1706 struct btrfs_extent_inline_ref
*iref
;
1709 ret
= lookup_inline_extent_backref(trans
, root
, path
, &iref
,
1710 bytenr
, num_bytes
, parent
,
1711 root_objectid
, owner
, offset
, 1);
1713 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
);
1714 ret
= update_inline_extent_backref(trans
, root
, path
, iref
,
1715 refs_to_add
, extent_op
);
1716 } else if (ret
== -ENOENT
) {
1717 ret
= setup_inline_extent_backref(trans
, root
, path
, iref
,
1718 parent
, root_objectid
,
1719 owner
, offset
, refs_to_add
,
1725 static int insert_extent_backref(struct btrfs_trans_handle
*trans
,
1726 struct btrfs_root
*root
,
1727 struct btrfs_path
*path
,
1728 u64 bytenr
, u64 parent
, u64 root_objectid
,
1729 u64 owner
, u64 offset
, int refs_to_add
)
1732 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1733 BUG_ON(refs_to_add
!= 1);
1734 ret
= insert_tree_block_ref(trans
, root
, path
, bytenr
,
1735 parent
, root_objectid
);
1737 ret
= insert_extent_data_ref(trans
, root
, path
, bytenr
,
1738 parent
, root_objectid
,
1739 owner
, offset
, refs_to_add
);
1744 static int remove_extent_backref(struct btrfs_trans_handle
*trans
,
1745 struct btrfs_root
*root
,
1746 struct btrfs_path
*path
,
1747 struct btrfs_extent_inline_ref
*iref
,
1748 int refs_to_drop
, int is_data
)
1752 BUG_ON(!is_data
&& refs_to_drop
!= 1);
1754 ret
= update_inline_extent_backref(trans
, root
, path
, iref
,
1755 -refs_to_drop
, NULL
);
1756 } else if (is_data
) {
1757 ret
= remove_extent_data_ref(trans
, root
, path
, refs_to_drop
);
1759 ret
= btrfs_del_item(trans
, root
, path
);
1764 static int btrfs_issue_discard(struct block_device
*bdev
,
1767 return blkdev_issue_discard(bdev
, start
>> 9, len
>> 9, GFP_NOFS
, 0);
1770 static int btrfs_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
1771 u64 num_bytes
, u64
*actual_bytes
)
1774 u64 discarded_bytes
= 0;
1775 struct btrfs_multi_bio
*multi
= NULL
;
1778 /* Tell the block device(s) that the sectors can be discarded */
1779 ret
= btrfs_map_block(&root
->fs_info
->mapping_tree
, REQ_DISCARD
,
1780 bytenr
, &num_bytes
, &multi
, 0);
1782 struct btrfs_bio_stripe
*stripe
= multi
->stripes
;
1786 for (i
= 0; i
< multi
->num_stripes
; i
++, stripe
++) {
1787 if (!stripe
->dev
->can_discard
)
1790 ret
= btrfs_issue_discard(stripe
->dev
->bdev
,
1794 discarded_bytes
+= stripe
->length
;
1795 else if (ret
!= -EOPNOTSUPP
)
1799 * Just in case we get back EOPNOTSUPP for some reason,
1800 * just ignore the return value so we don't screw up
1801 * people calling discard_extent.
1809 *actual_bytes
= discarded_bytes
;
1815 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1816 struct btrfs_root
*root
,
1817 u64 bytenr
, u64 num_bytes
, u64 parent
,
1818 u64 root_objectid
, u64 owner
, u64 offset
)
1821 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
&&
1822 root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
1824 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1825 ret
= btrfs_add_delayed_tree_ref(trans
, bytenr
, num_bytes
,
1826 parent
, root_objectid
, (int)owner
,
1827 BTRFS_ADD_DELAYED_REF
, NULL
);
1829 ret
= btrfs_add_delayed_data_ref(trans
, bytenr
, num_bytes
,
1830 parent
, root_objectid
, owner
, offset
,
1831 BTRFS_ADD_DELAYED_REF
, NULL
);
1836 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1837 struct btrfs_root
*root
,
1838 u64 bytenr
, u64 num_bytes
,
1839 u64 parent
, u64 root_objectid
,
1840 u64 owner
, u64 offset
, int refs_to_add
,
1841 struct btrfs_delayed_extent_op
*extent_op
)
1843 struct btrfs_path
*path
;
1844 struct extent_buffer
*leaf
;
1845 struct btrfs_extent_item
*item
;
1850 path
= btrfs_alloc_path();
1855 path
->leave_spinning
= 1;
1856 /* this will setup the path even if it fails to insert the back ref */
1857 ret
= insert_inline_extent_backref(trans
, root
->fs_info
->extent_root
,
1858 path
, bytenr
, num_bytes
, parent
,
1859 root_objectid
, owner
, offset
,
1860 refs_to_add
, extent_op
);
1864 if (ret
!= -EAGAIN
) {
1869 leaf
= path
->nodes
[0];
1870 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1871 refs
= btrfs_extent_refs(leaf
, item
);
1872 btrfs_set_extent_refs(leaf
, item
, refs
+ refs_to_add
);
1874 __run_delayed_extent_op(extent_op
, leaf
, item
);
1876 btrfs_mark_buffer_dirty(leaf
);
1877 btrfs_release_path(path
);
1880 path
->leave_spinning
= 1;
1882 /* now insert the actual backref */
1883 ret
= insert_extent_backref(trans
, root
->fs_info
->extent_root
,
1884 path
, bytenr
, parent
, root_objectid
,
1885 owner
, offset
, refs_to_add
);
1888 btrfs_free_path(path
);
1892 static int run_delayed_data_ref(struct btrfs_trans_handle
*trans
,
1893 struct btrfs_root
*root
,
1894 struct btrfs_delayed_ref_node
*node
,
1895 struct btrfs_delayed_extent_op
*extent_op
,
1896 int insert_reserved
)
1899 struct btrfs_delayed_data_ref
*ref
;
1900 struct btrfs_key ins
;
1905 ins
.objectid
= node
->bytenr
;
1906 ins
.offset
= node
->num_bytes
;
1907 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1909 ref
= btrfs_delayed_node_to_data_ref(node
);
1910 if (node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
1911 parent
= ref
->parent
;
1913 ref_root
= ref
->root
;
1915 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
1917 BUG_ON(extent_op
->update_key
);
1918 flags
|= extent_op
->flags_to_set
;
1920 ret
= alloc_reserved_file_extent(trans
, root
,
1921 parent
, ref_root
, flags
,
1922 ref
->objectid
, ref
->offset
,
1923 &ins
, node
->ref_mod
);
1924 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
1925 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
1926 node
->num_bytes
, parent
,
1927 ref_root
, ref
->objectid
,
1928 ref
->offset
, node
->ref_mod
,
1930 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
1931 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
1932 node
->num_bytes
, parent
,
1933 ref_root
, ref
->objectid
,
1934 ref
->offset
, node
->ref_mod
,
1942 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
1943 struct extent_buffer
*leaf
,
1944 struct btrfs_extent_item
*ei
)
1946 u64 flags
= btrfs_extent_flags(leaf
, ei
);
1947 if (extent_op
->update_flags
) {
1948 flags
|= extent_op
->flags_to_set
;
1949 btrfs_set_extent_flags(leaf
, ei
, flags
);
1952 if (extent_op
->update_key
) {
1953 struct btrfs_tree_block_info
*bi
;
1954 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
));
1955 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
1956 btrfs_set_tree_block_key(leaf
, bi
, &extent_op
->key
);
1960 static int run_delayed_extent_op(struct btrfs_trans_handle
*trans
,
1961 struct btrfs_root
*root
,
1962 struct btrfs_delayed_ref_node
*node
,
1963 struct btrfs_delayed_extent_op
*extent_op
)
1965 struct btrfs_key key
;
1966 struct btrfs_path
*path
;
1967 struct btrfs_extent_item
*ei
;
1968 struct extent_buffer
*leaf
;
1973 path
= btrfs_alloc_path();
1977 key
.objectid
= node
->bytenr
;
1978 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1979 key
.offset
= node
->num_bytes
;
1982 path
->leave_spinning
= 1;
1983 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
,
1994 leaf
= path
->nodes
[0];
1995 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1996 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1997 if (item_size
< sizeof(*ei
)) {
1998 ret
= convert_extent_item_v0(trans
, root
->fs_info
->extent_root
,
2004 leaf
= path
->nodes
[0];
2005 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2008 BUG_ON(item_size
< sizeof(*ei
));
2009 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2010 __run_delayed_extent_op(extent_op
, leaf
, ei
);
2012 btrfs_mark_buffer_dirty(leaf
);
2014 btrfs_free_path(path
);
2018 static int run_delayed_tree_ref(struct btrfs_trans_handle
*trans
,
2019 struct btrfs_root
*root
,
2020 struct btrfs_delayed_ref_node
*node
,
2021 struct btrfs_delayed_extent_op
*extent_op
,
2022 int insert_reserved
)
2025 struct btrfs_delayed_tree_ref
*ref
;
2026 struct btrfs_key ins
;
2030 ins
.objectid
= node
->bytenr
;
2031 ins
.offset
= node
->num_bytes
;
2032 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2034 ref
= btrfs_delayed_node_to_tree_ref(node
);
2035 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2036 parent
= ref
->parent
;
2038 ref_root
= ref
->root
;
2040 BUG_ON(node
->ref_mod
!= 1);
2041 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2042 BUG_ON(!extent_op
|| !extent_op
->update_flags
||
2043 !extent_op
->update_key
);
2044 ret
= alloc_reserved_tree_block(trans
, root
,
2046 extent_op
->flags_to_set
,
2049 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2050 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
2051 node
->num_bytes
, parent
, ref_root
,
2052 ref
->level
, 0, 1, extent_op
);
2053 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2054 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
2055 node
->num_bytes
, parent
, ref_root
,
2056 ref
->level
, 0, 1, extent_op
);
2063 /* helper function to actually process a single delayed ref entry */
2064 static int run_one_delayed_ref(struct btrfs_trans_handle
*trans
,
2065 struct btrfs_root
*root
,
2066 struct btrfs_delayed_ref_node
*node
,
2067 struct btrfs_delayed_extent_op
*extent_op
,
2068 int insert_reserved
)
2071 if (btrfs_delayed_ref_is_head(node
)) {
2072 struct btrfs_delayed_ref_head
*head
;
2074 * we've hit the end of the chain and we were supposed
2075 * to insert this extent into the tree. But, it got
2076 * deleted before we ever needed to insert it, so all
2077 * we have to do is clean up the accounting
2080 head
= btrfs_delayed_node_to_head(node
);
2081 if (insert_reserved
) {
2082 btrfs_pin_extent(root
, node
->bytenr
,
2083 node
->num_bytes
, 1);
2084 if (head
->is_data
) {
2085 ret
= btrfs_del_csums(trans
, root
,
2091 mutex_unlock(&head
->mutex
);
2095 if (node
->type
== BTRFS_TREE_BLOCK_REF_KEY
||
2096 node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2097 ret
= run_delayed_tree_ref(trans
, root
, node
, extent_op
,
2099 else if (node
->type
== BTRFS_EXTENT_DATA_REF_KEY
||
2100 node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2101 ret
= run_delayed_data_ref(trans
, root
, node
, extent_op
,
2108 static noinline
struct btrfs_delayed_ref_node
*
2109 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
2111 struct rb_node
*node
;
2112 struct btrfs_delayed_ref_node
*ref
;
2113 int action
= BTRFS_ADD_DELAYED_REF
;
2116 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2117 * this prevents ref count from going down to zero when
2118 * there still are pending delayed ref.
2120 node
= rb_prev(&head
->node
.rb_node
);
2124 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
2126 if (ref
->bytenr
!= head
->node
.bytenr
)
2128 if (ref
->action
== action
)
2130 node
= rb_prev(node
);
2132 if (action
== BTRFS_ADD_DELAYED_REF
) {
2133 action
= BTRFS_DROP_DELAYED_REF
;
2139 static noinline
int run_clustered_refs(struct btrfs_trans_handle
*trans
,
2140 struct btrfs_root
*root
,
2141 struct list_head
*cluster
)
2143 struct btrfs_delayed_ref_root
*delayed_refs
;
2144 struct btrfs_delayed_ref_node
*ref
;
2145 struct btrfs_delayed_ref_head
*locked_ref
= NULL
;
2146 struct btrfs_delayed_extent_op
*extent_op
;
2149 int must_insert_reserved
= 0;
2151 delayed_refs
= &trans
->transaction
->delayed_refs
;
2154 /* pick a new head ref from the cluster list */
2155 if (list_empty(cluster
))
2158 locked_ref
= list_entry(cluster
->next
,
2159 struct btrfs_delayed_ref_head
, cluster
);
2161 /* grab the lock that says we are going to process
2162 * all the refs for this head */
2163 ret
= btrfs_delayed_ref_lock(trans
, locked_ref
);
2166 * we may have dropped the spin lock to get the head
2167 * mutex lock, and that might have given someone else
2168 * time to free the head. If that's true, it has been
2169 * removed from our list and we can move on.
2171 if (ret
== -EAGAIN
) {
2179 * record the must insert reserved flag before we
2180 * drop the spin lock.
2182 must_insert_reserved
= locked_ref
->must_insert_reserved
;
2183 locked_ref
->must_insert_reserved
= 0;
2185 extent_op
= locked_ref
->extent_op
;
2186 locked_ref
->extent_op
= NULL
;
2189 * locked_ref is the head node, so we have to go one
2190 * node back for any delayed ref updates
2192 ref
= select_delayed_ref(locked_ref
);
2194 /* All delayed refs have been processed, Go ahead
2195 * and send the head node to run_one_delayed_ref,
2196 * so that any accounting fixes can happen
2198 ref
= &locked_ref
->node
;
2200 if (extent_op
&& must_insert_reserved
) {
2206 spin_unlock(&delayed_refs
->lock
);
2208 ret
= run_delayed_extent_op(trans
, root
,
2214 spin_lock(&delayed_refs
->lock
);
2218 list_del_init(&locked_ref
->cluster
);
2223 rb_erase(&ref
->rb_node
, &delayed_refs
->root
);
2224 delayed_refs
->num_entries
--;
2226 spin_unlock(&delayed_refs
->lock
);
2228 ret
= run_one_delayed_ref(trans
, root
, ref
, extent_op
,
2229 must_insert_reserved
);
2232 btrfs_put_delayed_ref(ref
);
2237 spin_lock(&delayed_refs
->lock
);
2243 * this starts processing the delayed reference count updates and
2244 * extent insertions we have queued up so far. count can be
2245 * 0, which means to process everything in the tree at the start
2246 * of the run (but not newly added entries), or it can be some target
2247 * number you'd like to process.
2249 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2250 struct btrfs_root
*root
, unsigned long count
)
2252 struct rb_node
*node
;
2253 struct btrfs_delayed_ref_root
*delayed_refs
;
2254 struct btrfs_delayed_ref_node
*ref
;
2255 struct list_head cluster
;
2257 int run_all
= count
== (unsigned long)-1;
2260 if (root
== root
->fs_info
->extent_root
)
2261 root
= root
->fs_info
->tree_root
;
2263 delayed_refs
= &trans
->transaction
->delayed_refs
;
2264 INIT_LIST_HEAD(&cluster
);
2266 spin_lock(&delayed_refs
->lock
);
2268 count
= delayed_refs
->num_entries
* 2;
2272 if (!(run_all
|| run_most
) &&
2273 delayed_refs
->num_heads_ready
< 64)
2277 * go find something we can process in the rbtree. We start at
2278 * the beginning of the tree, and then build a cluster
2279 * of refs to process starting at the first one we are able to
2282 ret
= btrfs_find_ref_cluster(trans
, &cluster
,
2283 delayed_refs
->run_delayed_start
);
2287 ret
= run_clustered_refs(trans
, root
, &cluster
);
2290 count
-= min_t(unsigned long, ret
, count
);
2297 node
= rb_first(&delayed_refs
->root
);
2300 count
= (unsigned long)-1;
2303 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
2305 if (btrfs_delayed_ref_is_head(ref
)) {
2306 struct btrfs_delayed_ref_head
*head
;
2308 head
= btrfs_delayed_node_to_head(ref
);
2309 atomic_inc(&ref
->refs
);
2311 spin_unlock(&delayed_refs
->lock
);
2313 * Mutex was contended, block until it's
2314 * released and try again
2316 mutex_lock(&head
->mutex
);
2317 mutex_unlock(&head
->mutex
);
2319 btrfs_put_delayed_ref(ref
);
2323 node
= rb_next(node
);
2325 spin_unlock(&delayed_refs
->lock
);
2326 schedule_timeout(1);
2330 spin_unlock(&delayed_refs
->lock
);
2334 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
2335 struct btrfs_root
*root
,
2336 u64 bytenr
, u64 num_bytes
, u64 flags
,
2339 struct btrfs_delayed_extent_op
*extent_op
;
2342 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
2346 extent_op
->flags_to_set
= flags
;
2347 extent_op
->update_flags
= 1;
2348 extent_op
->update_key
= 0;
2349 extent_op
->is_data
= is_data
? 1 : 0;
2351 ret
= btrfs_add_delayed_extent_op(trans
, bytenr
, num_bytes
, extent_op
);
2357 static noinline
int check_delayed_ref(struct btrfs_trans_handle
*trans
,
2358 struct btrfs_root
*root
,
2359 struct btrfs_path
*path
,
2360 u64 objectid
, u64 offset
, u64 bytenr
)
2362 struct btrfs_delayed_ref_head
*head
;
2363 struct btrfs_delayed_ref_node
*ref
;
2364 struct btrfs_delayed_data_ref
*data_ref
;
2365 struct btrfs_delayed_ref_root
*delayed_refs
;
2366 struct rb_node
*node
;
2370 delayed_refs
= &trans
->transaction
->delayed_refs
;
2371 spin_lock(&delayed_refs
->lock
);
2372 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
2376 if (!mutex_trylock(&head
->mutex
)) {
2377 atomic_inc(&head
->node
.refs
);
2378 spin_unlock(&delayed_refs
->lock
);
2380 btrfs_release_path(path
);
2383 * Mutex was contended, block until it's released and let
2386 mutex_lock(&head
->mutex
);
2387 mutex_unlock(&head
->mutex
);
2388 btrfs_put_delayed_ref(&head
->node
);
2392 node
= rb_prev(&head
->node
.rb_node
);
2396 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2398 if (ref
->bytenr
!= bytenr
)
2402 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
)
2405 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
2407 node
= rb_prev(node
);
2409 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2410 if (ref
->bytenr
== bytenr
)
2414 if (data_ref
->root
!= root
->root_key
.objectid
||
2415 data_ref
->objectid
!= objectid
|| data_ref
->offset
!= offset
)
2420 mutex_unlock(&head
->mutex
);
2422 spin_unlock(&delayed_refs
->lock
);
2426 static noinline
int check_committed_ref(struct btrfs_trans_handle
*trans
,
2427 struct btrfs_root
*root
,
2428 struct btrfs_path
*path
,
2429 u64 objectid
, u64 offset
, u64 bytenr
)
2431 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2432 struct extent_buffer
*leaf
;
2433 struct btrfs_extent_data_ref
*ref
;
2434 struct btrfs_extent_inline_ref
*iref
;
2435 struct btrfs_extent_item
*ei
;
2436 struct btrfs_key key
;
2440 key
.objectid
= bytenr
;
2441 key
.offset
= (u64
)-1;
2442 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2444 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
2450 if (path
->slots
[0] == 0)
2454 leaf
= path
->nodes
[0];
2455 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2457 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
2461 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2462 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2463 if (item_size
< sizeof(*ei
)) {
2464 WARN_ON(item_size
!= sizeof(struct btrfs_extent_item_v0
));
2468 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2470 if (item_size
!= sizeof(*ei
) +
2471 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
2474 if (btrfs_extent_generation(leaf
, ei
) <=
2475 btrfs_root_last_snapshot(&root
->root_item
))
2478 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
2479 if (btrfs_extent_inline_ref_type(leaf
, iref
) !=
2480 BTRFS_EXTENT_DATA_REF_KEY
)
2483 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
2484 if (btrfs_extent_refs(leaf
, ei
) !=
2485 btrfs_extent_data_ref_count(leaf
, ref
) ||
2486 btrfs_extent_data_ref_root(leaf
, ref
) !=
2487 root
->root_key
.objectid
||
2488 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
2489 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
2497 int btrfs_cross_ref_exist(struct btrfs_trans_handle
*trans
,
2498 struct btrfs_root
*root
,
2499 u64 objectid
, u64 offset
, u64 bytenr
)
2501 struct btrfs_path
*path
;
2505 path
= btrfs_alloc_path();
2510 ret
= check_committed_ref(trans
, root
, path
, objectid
,
2512 if (ret
&& ret
!= -ENOENT
)
2515 ret2
= check_delayed_ref(trans
, root
, path
, objectid
,
2517 } while (ret2
== -EAGAIN
);
2519 if (ret2
&& ret2
!= -ENOENT
) {
2524 if (ret
!= -ENOENT
|| ret2
!= -ENOENT
)
2527 btrfs_free_path(path
);
2528 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
2533 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
2534 struct btrfs_root
*root
,
2535 struct extent_buffer
*buf
,
2536 int full_backref
, int inc
)
2543 struct btrfs_key key
;
2544 struct btrfs_file_extent_item
*fi
;
2548 int (*process_func
)(struct btrfs_trans_handle
*, struct btrfs_root
*,
2549 u64
, u64
, u64
, u64
, u64
, u64
);
2551 ref_root
= btrfs_header_owner(buf
);
2552 nritems
= btrfs_header_nritems(buf
);
2553 level
= btrfs_header_level(buf
);
2555 if (!root
->ref_cows
&& level
== 0)
2559 process_func
= btrfs_inc_extent_ref
;
2561 process_func
= btrfs_free_extent
;
2564 parent
= buf
->start
;
2568 for (i
= 0; i
< nritems
; i
++) {
2570 btrfs_item_key_to_cpu(buf
, &key
, i
);
2571 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
2573 fi
= btrfs_item_ptr(buf
, i
,
2574 struct btrfs_file_extent_item
);
2575 if (btrfs_file_extent_type(buf
, fi
) ==
2576 BTRFS_FILE_EXTENT_INLINE
)
2578 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
2582 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
2583 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
2584 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
2585 parent
, ref_root
, key
.objectid
,
2590 bytenr
= btrfs_node_blockptr(buf
, i
);
2591 num_bytes
= btrfs_level_size(root
, level
- 1);
2592 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
2593 parent
, ref_root
, level
- 1, 0);
2604 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2605 struct extent_buffer
*buf
, int full_backref
)
2607 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1);
2610 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2611 struct extent_buffer
*buf
, int full_backref
)
2613 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0);
2616 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
2617 struct btrfs_root
*root
,
2618 struct btrfs_path
*path
,
2619 struct btrfs_block_group_cache
*cache
)
2622 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2624 struct extent_buffer
*leaf
;
2626 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
2631 leaf
= path
->nodes
[0];
2632 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
2633 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
2634 btrfs_mark_buffer_dirty(leaf
);
2635 btrfs_release_path(path
);
2643 static struct btrfs_block_group_cache
*
2644 next_block_group(struct btrfs_root
*root
,
2645 struct btrfs_block_group_cache
*cache
)
2647 struct rb_node
*node
;
2648 spin_lock(&root
->fs_info
->block_group_cache_lock
);
2649 node
= rb_next(&cache
->cache_node
);
2650 btrfs_put_block_group(cache
);
2652 cache
= rb_entry(node
, struct btrfs_block_group_cache
,
2654 btrfs_get_block_group(cache
);
2657 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
2661 static int cache_save_setup(struct btrfs_block_group_cache
*block_group
,
2662 struct btrfs_trans_handle
*trans
,
2663 struct btrfs_path
*path
)
2665 struct btrfs_root
*root
= block_group
->fs_info
->tree_root
;
2666 struct inode
*inode
= NULL
;
2668 int dcs
= BTRFS_DC_ERROR
;
2674 * If this block group is smaller than 100 megs don't bother caching the
2677 if (block_group
->key
.offset
< (100 * 1024 * 1024)) {
2678 spin_lock(&block_group
->lock
);
2679 block_group
->disk_cache_state
= BTRFS_DC_WRITTEN
;
2680 spin_unlock(&block_group
->lock
);
2685 inode
= lookup_free_space_inode(root
, block_group
, path
);
2686 if (IS_ERR(inode
) && PTR_ERR(inode
) != -ENOENT
) {
2687 ret
= PTR_ERR(inode
);
2688 btrfs_release_path(path
);
2692 if (IS_ERR(inode
)) {
2696 if (block_group
->ro
)
2699 ret
= create_free_space_inode(root
, trans
, block_group
, path
);
2706 * We want to set the generation to 0, that way if anything goes wrong
2707 * from here on out we know not to trust this cache when we load up next
2710 BTRFS_I(inode
)->generation
= 0;
2711 ret
= btrfs_update_inode(trans
, root
, inode
);
2714 if (i_size_read(inode
) > 0) {
2715 ret
= btrfs_truncate_free_space_cache(root
, trans
, path
,
2721 spin_lock(&block_group
->lock
);
2722 if (block_group
->cached
!= BTRFS_CACHE_FINISHED
) {
2723 /* We're not cached, don't bother trying to write stuff out */
2724 dcs
= BTRFS_DC_WRITTEN
;
2725 spin_unlock(&block_group
->lock
);
2728 spin_unlock(&block_group
->lock
);
2730 num_pages
= (int)div64_u64(block_group
->key
.offset
, 1024 * 1024 * 1024);
2735 * Just to make absolutely sure we have enough space, we're going to
2736 * preallocate 12 pages worth of space for each block group. In
2737 * practice we ought to use at most 8, but we need extra space so we can
2738 * add our header and have a terminator between the extents and the
2742 num_pages
*= PAGE_CACHE_SIZE
;
2744 ret
= btrfs_check_data_free_space(inode
, num_pages
);
2748 ret
= btrfs_prealloc_file_range_trans(inode
, trans
, 0, 0, num_pages
,
2749 num_pages
, num_pages
,
2752 dcs
= BTRFS_DC_SETUP
;
2753 btrfs_free_reserved_data_space(inode
, num_pages
);
2757 btrfs_release_path(path
);
2759 spin_lock(&block_group
->lock
);
2760 block_group
->disk_cache_state
= dcs
;
2761 spin_unlock(&block_group
->lock
);
2766 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
2767 struct btrfs_root
*root
)
2769 struct btrfs_block_group_cache
*cache
;
2771 struct btrfs_path
*path
;
2774 path
= btrfs_alloc_path();
2780 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
2782 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
)
2784 cache
= next_block_group(root
, cache
);
2792 err
= cache_save_setup(cache
, trans
, path
);
2793 last
= cache
->key
.objectid
+ cache
->key
.offset
;
2794 btrfs_put_block_group(cache
);
2799 err
= btrfs_run_delayed_refs(trans
, root
,
2804 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
2806 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
) {
2807 btrfs_put_block_group(cache
);
2813 cache
= next_block_group(root
, cache
);
2822 if (cache
->disk_cache_state
== BTRFS_DC_SETUP
)
2823 cache
->disk_cache_state
= BTRFS_DC_NEED_WRITE
;
2825 last
= cache
->key
.objectid
+ cache
->key
.offset
;
2827 err
= write_one_cache_group(trans
, root
, path
, cache
);
2829 btrfs_put_block_group(cache
);
2834 * I don't think this is needed since we're just marking our
2835 * preallocated extent as written, but just in case it can't
2839 err
= btrfs_run_delayed_refs(trans
, root
,
2844 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
2847 * Really this shouldn't happen, but it could if we
2848 * couldn't write the entire preallocated extent and
2849 * splitting the extent resulted in a new block.
2852 btrfs_put_block_group(cache
);
2855 if (cache
->disk_cache_state
== BTRFS_DC_NEED_WRITE
)
2857 cache
= next_block_group(root
, cache
);
2866 btrfs_write_out_cache(root
, trans
, cache
, path
);
2869 * If we didn't have an error then the cache state is still
2870 * NEED_WRITE, so we can set it to WRITTEN.
2872 if (cache
->disk_cache_state
== BTRFS_DC_NEED_WRITE
)
2873 cache
->disk_cache_state
= BTRFS_DC_WRITTEN
;
2874 last
= cache
->key
.objectid
+ cache
->key
.offset
;
2875 btrfs_put_block_group(cache
);
2878 btrfs_free_path(path
);
2882 int btrfs_extent_readonly(struct btrfs_root
*root
, u64 bytenr
)
2884 struct btrfs_block_group_cache
*block_group
;
2887 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
2888 if (!block_group
|| block_group
->ro
)
2891 btrfs_put_block_group(block_group
);
2895 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
2896 u64 total_bytes
, u64 bytes_used
,
2897 struct btrfs_space_info
**space_info
)
2899 struct btrfs_space_info
*found
;
2903 if (flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
2904 BTRFS_BLOCK_GROUP_RAID10
))
2909 found
= __find_space_info(info
, flags
);
2911 spin_lock(&found
->lock
);
2912 found
->total_bytes
+= total_bytes
;
2913 found
->disk_total
+= total_bytes
* factor
;
2914 found
->bytes_used
+= bytes_used
;
2915 found
->disk_used
+= bytes_used
* factor
;
2917 spin_unlock(&found
->lock
);
2918 *space_info
= found
;
2921 found
= kzalloc(sizeof(*found
), GFP_NOFS
);
2925 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
2926 INIT_LIST_HEAD(&found
->block_groups
[i
]);
2927 init_rwsem(&found
->groups_sem
);
2928 spin_lock_init(&found
->lock
);
2929 found
->flags
= flags
& (BTRFS_BLOCK_GROUP_DATA
|
2930 BTRFS_BLOCK_GROUP_SYSTEM
|
2931 BTRFS_BLOCK_GROUP_METADATA
);
2932 found
->total_bytes
= total_bytes
;
2933 found
->disk_total
= total_bytes
* factor
;
2934 found
->bytes_used
= bytes_used
;
2935 found
->disk_used
= bytes_used
* factor
;
2936 found
->bytes_pinned
= 0;
2937 found
->bytes_reserved
= 0;
2938 found
->bytes_readonly
= 0;
2939 found
->bytes_may_use
= 0;
2941 found
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
2942 found
->chunk_alloc
= 0;
2943 *space_info
= found
;
2944 list_add_rcu(&found
->list
, &info
->space_info
);
2945 atomic_set(&found
->caching_threads
, 0);
2949 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
2951 u64 extra_flags
= flags
& (BTRFS_BLOCK_GROUP_RAID0
|
2952 BTRFS_BLOCK_GROUP_RAID1
|
2953 BTRFS_BLOCK_GROUP_RAID10
|
2954 BTRFS_BLOCK_GROUP_DUP
);
2956 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
2957 fs_info
->avail_data_alloc_bits
|= extra_flags
;
2958 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
2959 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
2960 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
2961 fs_info
->avail_system_alloc_bits
|= extra_flags
;
2965 u64
btrfs_reduce_alloc_profile(struct btrfs_root
*root
, u64 flags
)
2968 * we add in the count of missing devices because we want
2969 * to make sure that any RAID levels on a degraded FS
2970 * continue to be honored.
2972 u64 num_devices
= root
->fs_info
->fs_devices
->rw_devices
+
2973 root
->fs_info
->fs_devices
->missing_devices
;
2975 if (num_devices
== 1)
2976 flags
&= ~(BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID0
);
2977 if (num_devices
< 4)
2978 flags
&= ~BTRFS_BLOCK_GROUP_RAID10
;
2980 if ((flags
& BTRFS_BLOCK_GROUP_DUP
) &&
2981 (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
2982 BTRFS_BLOCK_GROUP_RAID10
))) {
2983 flags
&= ~BTRFS_BLOCK_GROUP_DUP
;
2986 if ((flags
& BTRFS_BLOCK_GROUP_RAID1
) &&
2987 (flags
& BTRFS_BLOCK_GROUP_RAID10
)) {
2988 flags
&= ~BTRFS_BLOCK_GROUP_RAID1
;
2991 if ((flags
& BTRFS_BLOCK_GROUP_RAID0
) &&
2992 ((flags
& BTRFS_BLOCK_GROUP_RAID1
) |
2993 (flags
& BTRFS_BLOCK_GROUP_RAID10
) |
2994 (flags
& BTRFS_BLOCK_GROUP_DUP
)))
2995 flags
&= ~BTRFS_BLOCK_GROUP_RAID0
;
2999 static u64
get_alloc_profile(struct btrfs_root
*root
, u64 flags
)
3001 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3002 flags
|= root
->fs_info
->avail_data_alloc_bits
&
3003 root
->fs_info
->data_alloc_profile
;
3004 else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3005 flags
|= root
->fs_info
->avail_system_alloc_bits
&
3006 root
->fs_info
->system_alloc_profile
;
3007 else if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3008 flags
|= root
->fs_info
->avail_metadata_alloc_bits
&
3009 root
->fs_info
->metadata_alloc_profile
;
3010 return btrfs_reduce_alloc_profile(root
, flags
);
3013 u64
btrfs_get_alloc_profile(struct btrfs_root
*root
, int data
)
3018 flags
= BTRFS_BLOCK_GROUP_DATA
;
3019 else if (root
== root
->fs_info
->chunk_root
)
3020 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
3022 flags
= BTRFS_BLOCK_GROUP_METADATA
;
3024 return get_alloc_profile(root
, flags
);
3027 void btrfs_set_inode_space_info(struct btrfs_root
*root
, struct inode
*inode
)
3029 BTRFS_I(inode
)->space_info
= __find_space_info(root
->fs_info
,
3030 BTRFS_BLOCK_GROUP_DATA
);
3034 * This will check the space that the inode allocates from to make sure we have
3035 * enough space for bytes.
3037 int btrfs_check_data_free_space(struct inode
*inode
, u64 bytes
)
3039 struct btrfs_space_info
*data_sinfo
;
3040 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3042 int ret
= 0, committed
= 0, alloc_chunk
= 1;
3044 /* make sure bytes are sectorsize aligned */
3045 bytes
= (bytes
+ root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
3047 if (root
== root
->fs_info
->tree_root
||
3048 BTRFS_I(inode
)->location
.objectid
== BTRFS_FREE_INO_OBJECTID
) {
3053 data_sinfo
= BTRFS_I(inode
)->space_info
;
3058 /* make sure we have enough space to handle the data first */
3059 spin_lock(&data_sinfo
->lock
);
3060 used
= data_sinfo
->bytes_used
+ data_sinfo
->bytes_reserved
+
3061 data_sinfo
->bytes_pinned
+ data_sinfo
->bytes_readonly
+
3062 data_sinfo
->bytes_may_use
;
3064 if (used
+ bytes
> data_sinfo
->total_bytes
) {
3065 struct btrfs_trans_handle
*trans
;
3068 * if we don't have enough free bytes in this space then we need
3069 * to alloc a new chunk.
3071 if (!data_sinfo
->full
&& alloc_chunk
) {
3074 data_sinfo
->force_alloc
= CHUNK_ALLOC_FORCE
;
3075 spin_unlock(&data_sinfo
->lock
);
3077 alloc_target
= btrfs_get_alloc_profile(root
, 1);
3078 trans
= btrfs_join_transaction(root
);
3080 return PTR_ERR(trans
);
3082 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
3083 bytes
+ 2 * 1024 * 1024,
3085 CHUNK_ALLOC_NO_FORCE
);
3086 btrfs_end_transaction(trans
, root
);
3095 btrfs_set_inode_space_info(root
, inode
);
3096 data_sinfo
= BTRFS_I(inode
)->space_info
;
3102 * If we have less pinned bytes than we want to allocate then
3103 * don't bother committing the transaction, it won't help us.
3105 if (data_sinfo
->bytes_pinned
< bytes
)
3107 spin_unlock(&data_sinfo
->lock
);
3109 /* commit the current transaction and try again */
3112 !atomic_read(&root
->fs_info
->open_ioctl_trans
)) {
3114 trans
= btrfs_join_transaction(root
);
3116 return PTR_ERR(trans
);
3117 ret
= btrfs_commit_transaction(trans
, root
);
3125 data_sinfo
->bytes_may_use
+= bytes
;
3126 BTRFS_I(inode
)->reserved_bytes
+= bytes
;
3127 spin_unlock(&data_sinfo
->lock
);
3133 * called when we are clearing an delalloc extent from the
3134 * inode's io_tree or there was an error for whatever reason
3135 * after calling btrfs_check_data_free_space
3137 void btrfs_free_reserved_data_space(struct inode
*inode
, u64 bytes
)
3139 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3140 struct btrfs_space_info
*data_sinfo
;
3142 /* make sure bytes are sectorsize aligned */
3143 bytes
= (bytes
+ root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
3145 data_sinfo
= BTRFS_I(inode
)->space_info
;
3146 spin_lock(&data_sinfo
->lock
);
3147 data_sinfo
->bytes_may_use
-= bytes
;
3148 BTRFS_I(inode
)->reserved_bytes
-= bytes
;
3149 spin_unlock(&data_sinfo
->lock
);
3152 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
3154 struct list_head
*head
= &info
->space_info
;
3155 struct btrfs_space_info
*found
;
3158 list_for_each_entry_rcu(found
, head
, list
) {
3159 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
3160 found
->force_alloc
= CHUNK_ALLOC_FORCE
;
3165 static int should_alloc_chunk(struct btrfs_root
*root
,
3166 struct btrfs_space_info
*sinfo
, u64 alloc_bytes
,
3169 u64 num_bytes
= sinfo
->total_bytes
- sinfo
->bytes_readonly
;
3170 u64 num_allocated
= sinfo
->bytes_used
+ sinfo
->bytes_reserved
;
3173 if (force
== CHUNK_ALLOC_FORCE
)
3177 * in limited mode, we want to have some free space up to
3178 * about 1% of the FS size.
3180 if (force
== CHUNK_ALLOC_LIMITED
) {
3181 thresh
= btrfs_super_total_bytes(&root
->fs_info
->super_copy
);
3182 thresh
= max_t(u64
, 64 * 1024 * 1024,
3183 div_factor_fine(thresh
, 1));
3185 if (num_bytes
- num_allocated
< thresh
)
3190 * we have two similar checks here, one based on percentage
3191 * and once based on a hard number of 256MB. The idea
3192 * is that if we have a good amount of free
3193 * room, don't allocate a chunk. A good mount is
3194 * less than 80% utilized of the chunks we have allocated,
3195 * or more than 256MB free
3197 if (num_allocated
+ alloc_bytes
+ 256 * 1024 * 1024 < num_bytes
)
3200 if (num_allocated
+ alloc_bytes
< div_factor(num_bytes
, 8))
3203 thresh
= btrfs_super_total_bytes(&root
->fs_info
->super_copy
);
3205 /* 256MB or 5% of the FS */
3206 thresh
= max_t(u64
, 256 * 1024 * 1024, div_factor_fine(thresh
, 5));
3208 if (num_bytes
> thresh
&& sinfo
->bytes_used
< div_factor(num_bytes
, 3))
3213 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
3214 struct btrfs_root
*extent_root
, u64 alloc_bytes
,
3215 u64 flags
, int force
)
3217 struct btrfs_space_info
*space_info
;
3218 struct btrfs_fs_info
*fs_info
= extent_root
->fs_info
;
3219 int wait_for_alloc
= 0;
3222 flags
= btrfs_reduce_alloc_profile(extent_root
, flags
);
3224 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
3226 ret
= update_space_info(extent_root
->fs_info
, flags
,
3230 BUG_ON(!space_info
);
3233 spin_lock(&space_info
->lock
);
3234 if (space_info
->force_alloc
)
3235 force
= space_info
->force_alloc
;
3236 if (space_info
->full
) {
3237 spin_unlock(&space_info
->lock
);
3241 if (!should_alloc_chunk(extent_root
, space_info
, alloc_bytes
, force
)) {
3242 spin_unlock(&space_info
->lock
);
3244 } else if (space_info
->chunk_alloc
) {
3247 space_info
->chunk_alloc
= 1;
3250 spin_unlock(&space_info
->lock
);
3252 mutex_lock(&fs_info
->chunk_mutex
);
3255 * The chunk_mutex is held throughout the entirety of a chunk
3256 * allocation, so once we've acquired the chunk_mutex we know that the
3257 * other guy is done and we need to recheck and see if we should
3260 if (wait_for_alloc
) {
3261 mutex_unlock(&fs_info
->chunk_mutex
);
3267 * If we have mixed data/metadata chunks we want to make sure we keep
3268 * allocating mixed chunks instead of individual chunks.
3270 if (btrfs_mixed_space_info(space_info
))
3271 flags
|= (BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
);
3274 * if we're doing a data chunk, go ahead and make sure that
3275 * we keep a reasonable number of metadata chunks allocated in the
3278 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
3279 fs_info
->data_chunk_allocations
++;
3280 if (!(fs_info
->data_chunk_allocations
%
3281 fs_info
->metadata_ratio
))
3282 force_metadata_allocation(fs_info
);
3285 ret
= btrfs_alloc_chunk(trans
, extent_root
, flags
);
3286 spin_lock(&space_info
->lock
);
3288 space_info
->full
= 1;
3292 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3293 space_info
->chunk_alloc
= 0;
3294 spin_unlock(&space_info
->lock
);
3295 mutex_unlock(&extent_root
->fs_info
->chunk_mutex
);
3300 * shrink metadata reservation for delalloc
3302 static int shrink_delalloc(struct btrfs_trans_handle
*trans
,
3303 struct btrfs_root
*root
, u64 to_reclaim
, int sync
)
3305 struct btrfs_block_rsv
*block_rsv
;
3306 struct btrfs_space_info
*space_info
;
3311 int nr_pages
= (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT
;
3313 unsigned long progress
;
3315 block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
3316 space_info
= block_rsv
->space_info
;
3319 reserved
= space_info
->bytes_reserved
;
3320 progress
= space_info
->reservation_progress
;
3325 max_reclaim
= min(reserved
, to_reclaim
);
3327 while (loops
< 1024) {
3328 /* have the flusher threads jump in and do some IO */
3330 nr_pages
= min_t(unsigned long, nr_pages
,
3331 root
->fs_info
->delalloc_bytes
>> PAGE_CACHE_SHIFT
);
3332 writeback_inodes_sb_nr_if_idle(root
->fs_info
->sb
, nr_pages
);
3334 spin_lock(&space_info
->lock
);
3335 if (reserved
> space_info
->bytes_reserved
)
3336 reclaimed
+= reserved
- space_info
->bytes_reserved
;
3337 reserved
= space_info
->bytes_reserved
;
3338 spin_unlock(&space_info
->lock
);
3342 if (reserved
== 0 || reclaimed
>= max_reclaim
)
3345 if (trans
&& trans
->transaction
->blocked
)
3348 time_left
= schedule_timeout_interruptible(1);
3350 /* We were interrupted, exit */
3354 /* we've kicked the IO a few times, if anything has been freed,
3355 * exit. There is no sense in looping here for a long time
3356 * when we really need to commit the transaction, or there are
3357 * just too many writers without enough free space
3362 if (progress
!= space_info
->reservation_progress
)
3367 return reclaimed
>= to_reclaim
;
3371 * Retries tells us how many times we've called reserve_metadata_bytes. The
3372 * idea is if this is the first call (retries == 0) then we will add to our
3373 * reserved count if we can't make the allocation in order to hold our place
3374 * while we go and try and free up space. That way for retries > 1 we don't try
3375 * and add space, we just check to see if the amount of unused space is >= the
3376 * total space, meaning that our reservation is valid.
3378 * However if we don't intend to retry this reservation, pass -1 as retries so
3379 * that it short circuits this logic.
3381 static int reserve_metadata_bytes(struct btrfs_trans_handle
*trans
,
3382 struct btrfs_root
*root
,
3383 struct btrfs_block_rsv
*block_rsv
,
3384 u64 orig_bytes
, int flush
)
3386 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
3388 u64 num_bytes
= orig_bytes
;
3391 bool reserved
= false;
3392 bool committed
= false;
3399 spin_lock(&space_info
->lock
);
3400 unused
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
3401 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
3402 space_info
->bytes_may_use
;
3405 * The idea here is that we've not already over-reserved the block group
3406 * then we can go ahead and save our reservation first and then start
3407 * flushing if we need to. Otherwise if we've already overcommitted
3408 * lets start flushing stuff first and then come back and try to make
3411 if (unused
<= space_info
->total_bytes
) {
3412 unused
= space_info
->total_bytes
- unused
;
3413 if (unused
>= num_bytes
) {
3415 space_info
->bytes_reserved
+= orig_bytes
;
3419 * Ok set num_bytes to orig_bytes since we aren't
3420 * overocmmitted, this way we only try and reclaim what
3423 num_bytes
= orig_bytes
;
3427 * Ok we're over committed, set num_bytes to the overcommitted
3428 * amount plus the amount of bytes that we need for this
3431 num_bytes
= unused
- space_info
->total_bytes
+
3432 (orig_bytes
* (retries
+ 1));
3436 * Couldn't make our reservation, save our place so while we're trying
3437 * to reclaim space we can actually use it instead of somebody else
3438 * stealing it from us.
3440 if (ret
&& !reserved
) {
3441 space_info
->bytes_reserved
+= orig_bytes
;
3445 spin_unlock(&space_info
->lock
);
3454 * We do synchronous shrinking since we don't actually unreserve
3455 * metadata until after the IO is completed.
3457 ret
= shrink_delalloc(trans
, root
, num_bytes
, 1);
3464 * So if we were overcommitted it's possible that somebody else flushed
3465 * out enough space and we simply didn't have enough space to reclaim,
3466 * so go back around and try again.
3473 spin_lock(&space_info
->lock
);
3475 * Not enough space to be reclaimed, don't bother committing the
3478 if (space_info
->bytes_pinned
< orig_bytes
)
3480 spin_unlock(&space_info
->lock
);
3485 if (trans
|| committed
)
3489 trans
= btrfs_join_transaction(root
);
3492 ret
= btrfs_commit_transaction(trans
, root
);
3501 spin_lock(&space_info
->lock
);
3502 space_info
->bytes_reserved
-= orig_bytes
;
3503 spin_unlock(&space_info
->lock
);
3509 static struct btrfs_block_rsv
*get_block_rsv(struct btrfs_trans_handle
*trans
,
3510 struct btrfs_root
*root
)
3512 struct btrfs_block_rsv
*block_rsv
;
3514 block_rsv
= trans
->block_rsv
;
3516 block_rsv
= root
->block_rsv
;
3519 block_rsv
= &root
->fs_info
->empty_block_rsv
;
3524 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
3528 spin_lock(&block_rsv
->lock
);
3529 if (block_rsv
->reserved
>= num_bytes
) {
3530 block_rsv
->reserved
-= num_bytes
;
3531 if (block_rsv
->reserved
< block_rsv
->size
)
3532 block_rsv
->full
= 0;
3535 spin_unlock(&block_rsv
->lock
);
3539 static void block_rsv_add_bytes(struct btrfs_block_rsv
*block_rsv
,
3540 u64 num_bytes
, int update_size
)
3542 spin_lock(&block_rsv
->lock
);
3543 block_rsv
->reserved
+= num_bytes
;
3545 block_rsv
->size
+= num_bytes
;
3546 else if (block_rsv
->reserved
>= block_rsv
->size
)
3547 block_rsv
->full
= 1;
3548 spin_unlock(&block_rsv
->lock
);
3551 static void block_rsv_release_bytes(struct btrfs_block_rsv
*block_rsv
,
3552 struct btrfs_block_rsv
*dest
, u64 num_bytes
)
3554 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
3556 spin_lock(&block_rsv
->lock
);
3557 if (num_bytes
== (u64
)-1)
3558 num_bytes
= block_rsv
->size
;
3559 block_rsv
->size
-= num_bytes
;
3560 if (block_rsv
->reserved
>= block_rsv
->size
) {
3561 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
3562 block_rsv
->reserved
= block_rsv
->size
;
3563 block_rsv
->full
= 1;
3567 spin_unlock(&block_rsv
->lock
);
3569 if (num_bytes
> 0) {
3571 spin_lock(&dest
->lock
);
3575 bytes_to_add
= dest
->size
- dest
->reserved
;
3576 bytes_to_add
= min(num_bytes
, bytes_to_add
);
3577 dest
->reserved
+= bytes_to_add
;
3578 if (dest
->reserved
>= dest
->size
)
3580 num_bytes
-= bytes_to_add
;
3582 spin_unlock(&dest
->lock
);
3585 spin_lock(&space_info
->lock
);
3586 space_info
->bytes_reserved
-= num_bytes
;
3587 space_info
->reservation_progress
++;
3588 spin_unlock(&space_info
->lock
);
3593 static int block_rsv_migrate_bytes(struct btrfs_block_rsv
*src
,
3594 struct btrfs_block_rsv
*dst
, u64 num_bytes
)
3598 ret
= block_rsv_use_bytes(src
, num_bytes
);
3602 block_rsv_add_bytes(dst
, num_bytes
, 1);
3606 void btrfs_init_block_rsv(struct btrfs_block_rsv
*rsv
)
3608 memset(rsv
, 0, sizeof(*rsv
));
3609 spin_lock_init(&rsv
->lock
);
3610 atomic_set(&rsv
->usage
, 1);
3612 INIT_LIST_HEAD(&rsv
->list
);
3615 struct btrfs_block_rsv
*btrfs_alloc_block_rsv(struct btrfs_root
*root
)
3617 struct btrfs_block_rsv
*block_rsv
;
3618 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3620 block_rsv
= kmalloc(sizeof(*block_rsv
), GFP_NOFS
);
3624 btrfs_init_block_rsv(block_rsv
);
3625 block_rsv
->space_info
= __find_space_info(fs_info
,
3626 BTRFS_BLOCK_GROUP_METADATA
);
3630 void btrfs_free_block_rsv(struct btrfs_root
*root
,
3631 struct btrfs_block_rsv
*rsv
)
3633 if (rsv
&& atomic_dec_and_test(&rsv
->usage
)) {
3634 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
3641 * make the block_rsv struct be able to capture freed space.
3642 * the captured space will re-add to the the block_rsv struct
3643 * after transaction commit
3645 void btrfs_add_durable_block_rsv(struct btrfs_fs_info
*fs_info
,
3646 struct btrfs_block_rsv
*block_rsv
)
3648 block_rsv
->durable
= 1;
3649 mutex_lock(&fs_info
->durable_block_rsv_mutex
);
3650 list_add_tail(&block_rsv
->list
, &fs_info
->durable_block_rsv_list
);
3651 mutex_unlock(&fs_info
->durable_block_rsv_mutex
);
3654 int btrfs_block_rsv_add(struct btrfs_trans_handle
*trans
,
3655 struct btrfs_root
*root
,
3656 struct btrfs_block_rsv
*block_rsv
,
3664 ret
= reserve_metadata_bytes(trans
, root
, block_rsv
, num_bytes
, 1);
3666 block_rsv_add_bytes(block_rsv
, num_bytes
, 1);
3673 int btrfs_block_rsv_check(struct btrfs_trans_handle
*trans
,
3674 struct btrfs_root
*root
,
3675 struct btrfs_block_rsv
*block_rsv
,
3676 u64 min_reserved
, int min_factor
)
3679 int commit_trans
= 0;
3685 spin_lock(&block_rsv
->lock
);
3687 num_bytes
= div_factor(block_rsv
->size
, min_factor
);
3688 if (min_reserved
> num_bytes
)
3689 num_bytes
= min_reserved
;
3691 if (block_rsv
->reserved
>= num_bytes
) {
3694 num_bytes
-= block_rsv
->reserved
;
3695 if (block_rsv
->durable
&&
3696 block_rsv
->freed
[0] + block_rsv
->freed
[1] >= num_bytes
)
3699 spin_unlock(&block_rsv
->lock
);
3703 if (block_rsv
->refill_used
) {
3704 ret
= reserve_metadata_bytes(trans
, root
, block_rsv
,
3707 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
3716 trans
= btrfs_join_transaction(root
);
3717 BUG_ON(IS_ERR(trans
));
3718 ret
= btrfs_commit_transaction(trans
, root
);
3725 int btrfs_block_rsv_migrate(struct btrfs_block_rsv
*src_rsv
,
3726 struct btrfs_block_rsv
*dst_rsv
,
3729 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
3732 void btrfs_block_rsv_release(struct btrfs_root
*root
,
3733 struct btrfs_block_rsv
*block_rsv
,
3736 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
3737 if (global_rsv
->full
|| global_rsv
== block_rsv
||
3738 block_rsv
->space_info
!= global_rsv
->space_info
)
3740 block_rsv_release_bytes(block_rsv
, global_rsv
, num_bytes
);
3744 * helper to calculate size of global block reservation.
3745 * the desired value is sum of space used by extent tree,
3746 * checksum tree and root tree
3748 static u64
calc_global_metadata_size(struct btrfs_fs_info
*fs_info
)
3750 struct btrfs_space_info
*sinfo
;
3754 int csum_size
= btrfs_super_csum_size(&fs_info
->super_copy
);
3756 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_DATA
);
3757 spin_lock(&sinfo
->lock
);
3758 data_used
= sinfo
->bytes_used
;
3759 spin_unlock(&sinfo
->lock
);
3761 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
3762 spin_lock(&sinfo
->lock
);
3763 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_DATA
)
3765 meta_used
= sinfo
->bytes_used
;
3766 spin_unlock(&sinfo
->lock
);
3768 num_bytes
= (data_used
>> fs_info
->sb
->s_blocksize_bits
) *
3770 num_bytes
+= div64_u64(data_used
+ meta_used
, 50);
3772 if (num_bytes
* 3 > meta_used
)
3773 num_bytes
= div64_u64(meta_used
, 3);
3775 return ALIGN(num_bytes
, fs_info
->extent_root
->leafsize
<< 10);
3778 static void update_global_block_rsv(struct btrfs_fs_info
*fs_info
)
3780 struct btrfs_block_rsv
*block_rsv
= &fs_info
->global_block_rsv
;
3781 struct btrfs_space_info
*sinfo
= block_rsv
->space_info
;
3784 num_bytes
= calc_global_metadata_size(fs_info
);
3786 spin_lock(&block_rsv
->lock
);
3787 spin_lock(&sinfo
->lock
);
3789 block_rsv
->size
= num_bytes
;
3791 num_bytes
= sinfo
->bytes_used
+ sinfo
->bytes_pinned
+
3792 sinfo
->bytes_reserved
+ sinfo
->bytes_readonly
+
3793 sinfo
->bytes_may_use
;
3795 if (sinfo
->total_bytes
> num_bytes
) {
3796 num_bytes
= sinfo
->total_bytes
- num_bytes
;
3797 block_rsv
->reserved
+= num_bytes
;
3798 sinfo
->bytes_reserved
+= num_bytes
;
3801 if (block_rsv
->reserved
>= block_rsv
->size
) {
3802 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
3803 sinfo
->bytes_reserved
-= num_bytes
;
3804 sinfo
->reservation_progress
++;
3805 block_rsv
->reserved
= block_rsv
->size
;
3806 block_rsv
->full
= 1;
3809 spin_unlock(&sinfo
->lock
);
3810 spin_unlock(&block_rsv
->lock
);
3813 static void init_global_block_rsv(struct btrfs_fs_info
*fs_info
)
3815 struct btrfs_space_info
*space_info
;
3817 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
3818 fs_info
->chunk_block_rsv
.space_info
= space_info
;
3819 fs_info
->chunk_block_rsv
.priority
= 10;
3821 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
3822 fs_info
->global_block_rsv
.space_info
= space_info
;
3823 fs_info
->global_block_rsv
.priority
= 10;
3824 fs_info
->global_block_rsv
.refill_used
= 1;
3825 fs_info
->delalloc_block_rsv
.space_info
= space_info
;
3826 fs_info
->trans_block_rsv
.space_info
= space_info
;
3827 fs_info
->empty_block_rsv
.space_info
= space_info
;
3828 fs_info
->empty_block_rsv
.priority
= 10;
3830 fs_info
->extent_root
->block_rsv
= &fs_info
->global_block_rsv
;
3831 fs_info
->csum_root
->block_rsv
= &fs_info
->global_block_rsv
;
3832 fs_info
->dev_root
->block_rsv
= &fs_info
->global_block_rsv
;
3833 fs_info
->tree_root
->block_rsv
= &fs_info
->global_block_rsv
;
3834 fs_info
->chunk_root
->block_rsv
= &fs_info
->chunk_block_rsv
;
3836 btrfs_add_durable_block_rsv(fs_info
, &fs_info
->global_block_rsv
);
3838 btrfs_add_durable_block_rsv(fs_info
, &fs_info
->delalloc_block_rsv
);
3840 update_global_block_rsv(fs_info
);
3843 static void release_global_block_rsv(struct btrfs_fs_info
*fs_info
)
3845 block_rsv_release_bytes(&fs_info
->global_block_rsv
, NULL
, (u64
)-1);
3846 WARN_ON(fs_info
->delalloc_block_rsv
.size
> 0);
3847 WARN_ON(fs_info
->delalloc_block_rsv
.reserved
> 0);
3848 WARN_ON(fs_info
->trans_block_rsv
.size
> 0);
3849 WARN_ON(fs_info
->trans_block_rsv
.reserved
> 0);
3850 WARN_ON(fs_info
->chunk_block_rsv
.size
> 0);
3851 WARN_ON(fs_info
->chunk_block_rsv
.reserved
> 0);
3854 int btrfs_truncate_reserve_metadata(struct btrfs_trans_handle
*trans
,
3855 struct btrfs_root
*root
,
3856 struct btrfs_block_rsv
*rsv
)
3858 struct btrfs_block_rsv
*trans_rsv
= &root
->fs_info
->trans_block_rsv
;
3863 * Truncate should be freeing data, but give us 2 items just in case it
3864 * needs to use some space. We may want to be smarter about this in the
3867 num_bytes
= btrfs_calc_trans_metadata_size(root
, 2);
3869 /* We already have enough bytes, just return */
3870 if (rsv
->reserved
>= num_bytes
)
3873 num_bytes
-= rsv
->reserved
;
3876 * You should have reserved enough space before hand to do this, so this
3879 ret
= block_rsv_migrate_bytes(trans_rsv
, rsv
, num_bytes
);
3885 int btrfs_trans_reserve_metadata(struct btrfs_trans_handle
*trans
,
3886 struct btrfs_root
*root
,
3892 if (num_items
== 0 || root
->fs_info
->chunk_root
== root
)
3895 num_bytes
= btrfs_calc_trans_metadata_size(root
, num_items
);
3896 ret
= btrfs_block_rsv_add(trans
, root
, &root
->fs_info
->trans_block_rsv
,
3899 trans
->bytes_reserved
+= num_bytes
;
3900 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
3905 void btrfs_trans_release_metadata(struct btrfs_trans_handle
*trans
,
3906 struct btrfs_root
*root
)
3908 if (!trans
->bytes_reserved
)
3911 BUG_ON(trans
->block_rsv
!= &root
->fs_info
->trans_block_rsv
);
3912 btrfs_block_rsv_release(root
, trans
->block_rsv
,
3913 trans
->bytes_reserved
);
3914 trans
->bytes_reserved
= 0;
3917 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle
*trans
,
3918 struct inode
*inode
)
3920 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3921 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
3922 struct btrfs_block_rsv
*dst_rsv
= root
->orphan_block_rsv
;
3925 * We need to hold space in order to delete our orphan item once we've
3926 * added it, so this takes the reservation so we can release it later
3927 * when we are truly done with the orphan item.
3929 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
3930 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
3933 void btrfs_orphan_release_metadata(struct inode
*inode
)
3935 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3936 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
3937 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
, num_bytes
);
3940 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle
*trans
,
3941 struct btrfs_pending_snapshot
*pending
)
3943 struct btrfs_root
*root
= pending
->root
;
3944 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
3945 struct btrfs_block_rsv
*dst_rsv
= &pending
->block_rsv
;
3947 * two for root back/forward refs, two for directory entries
3948 * and one for root of the snapshot.
3950 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 5);
3951 dst_rsv
->space_info
= src_rsv
->space_info
;
3952 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
3955 static u64
calc_csum_metadata_size(struct inode
*inode
, u64 num_bytes
)
3957 return num_bytes
>>= 3;
3960 int btrfs_delalloc_reserve_metadata(struct inode
*inode
, u64 num_bytes
)
3962 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3963 struct btrfs_block_rsv
*block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
3966 int reserved_extents
;
3969 if (btrfs_transaction_in_commit(root
->fs_info
))
3970 schedule_timeout(1);
3972 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
3974 nr_extents
= atomic_read(&BTRFS_I(inode
)->outstanding_extents
) + 1;
3975 reserved_extents
= atomic_read(&BTRFS_I(inode
)->reserved_extents
);
3977 if (nr_extents
> reserved_extents
) {
3978 nr_extents
-= reserved_extents
;
3979 to_reserve
= btrfs_calc_trans_metadata_size(root
, nr_extents
);
3985 to_reserve
+= calc_csum_metadata_size(inode
, num_bytes
);
3986 ret
= reserve_metadata_bytes(NULL
, root
, block_rsv
, to_reserve
, 1);
3990 atomic_add(nr_extents
, &BTRFS_I(inode
)->reserved_extents
);
3991 atomic_inc(&BTRFS_I(inode
)->outstanding_extents
);
3993 block_rsv_add_bytes(block_rsv
, to_reserve
, 1);
3995 if (block_rsv
->size
> 512 * 1024 * 1024)
3996 shrink_delalloc(NULL
, root
, to_reserve
, 0);
4001 void btrfs_delalloc_release_metadata(struct inode
*inode
, u64 num_bytes
)
4003 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4006 int reserved_extents
;
4008 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
4009 atomic_dec(&BTRFS_I(inode
)->outstanding_extents
);
4010 WARN_ON(atomic_read(&BTRFS_I(inode
)->outstanding_extents
) < 0);
4012 reserved_extents
= atomic_read(&BTRFS_I(inode
)->reserved_extents
);
4016 nr_extents
= atomic_read(&BTRFS_I(inode
)->outstanding_extents
);
4017 if (nr_extents
>= reserved_extents
) {
4021 old
= reserved_extents
;
4022 nr_extents
= reserved_extents
- nr_extents
;
4023 new = reserved_extents
- nr_extents
;
4024 old
= atomic_cmpxchg(&BTRFS_I(inode
)->reserved_extents
,
4025 reserved_extents
, new);
4026 if (likely(old
== reserved_extents
))
4028 reserved_extents
= old
;
4031 to_free
= calc_csum_metadata_size(inode
, num_bytes
);
4033 to_free
+= btrfs_calc_trans_metadata_size(root
, nr_extents
);
4035 btrfs_block_rsv_release(root
, &root
->fs_info
->delalloc_block_rsv
,
4039 int btrfs_delalloc_reserve_space(struct inode
*inode
, u64 num_bytes
)
4043 ret
= btrfs_check_data_free_space(inode
, num_bytes
);
4047 ret
= btrfs_delalloc_reserve_metadata(inode
, num_bytes
);
4049 btrfs_free_reserved_data_space(inode
, num_bytes
);
4056 void btrfs_delalloc_release_space(struct inode
*inode
, u64 num_bytes
)
4058 btrfs_delalloc_release_metadata(inode
, num_bytes
);
4059 btrfs_free_reserved_data_space(inode
, num_bytes
);
4062 static int update_block_group(struct btrfs_trans_handle
*trans
,
4063 struct btrfs_root
*root
,
4064 u64 bytenr
, u64 num_bytes
, int alloc
)
4066 struct btrfs_block_group_cache
*cache
= NULL
;
4067 struct btrfs_fs_info
*info
= root
->fs_info
;
4068 u64 total
= num_bytes
;
4073 /* block accounting for super block */
4074 spin_lock(&info
->delalloc_lock
);
4075 old_val
= btrfs_super_bytes_used(&info
->super_copy
);
4077 old_val
+= num_bytes
;
4079 old_val
-= num_bytes
;
4080 btrfs_set_super_bytes_used(&info
->super_copy
, old_val
);
4081 spin_unlock(&info
->delalloc_lock
);
4084 cache
= btrfs_lookup_block_group(info
, bytenr
);
4087 if (cache
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
4088 BTRFS_BLOCK_GROUP_RAID1
|
4089 BTRFS_BLOCK_GROUP_RAID10
))
4094 * If this block group has free space cache written out, we
4095 * need to make sure to load it if we are removing space. This
4096 * is because we need the unpinning stage to actually add the
4097 * space back to the block group, otherwise we will leak space.
4099 if (!alloc
&& cache
->cached
== BTRFS_CACHE_NO
)
4100 cache_block_group(cache
, trans
, NULL
, 1);
4102 byte_in_group
= bytenr
- cache
->key
.objectid
;
4103 WARN_ON(byte_in_group
> cache
->key
.offset
);
4105 spin_lock(&cache
->space_info
->lock
);
4106 spin_lock(&cache
->lock
);
4108 if (btrfs_super_cache_generation(&info
->super_copy
) != 0 &&
4109 cache
->disk_cache_state
< BTRFS_DC_CLEAR
)
4110 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
4113 old_val
= btrfs_block_group_used(&cache
->item
);
4114 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
4116 old_val
+= num_bytes
;
4117 btrfs_set_block_group_used(&cache
->item
, old_val
);
4118 cache
->reserved
-= num_bytes
;
4119 cache
->space_info
->bytes_reserved
-= num_bytes
;
4120 cache
->space_info
->reservation_progress
++;
4121 cache
->space_info
->bytes_used
+= num_bytes
;
4122 cache
->space_info
->disk_used
+= num_bytes
* factor
;
4123 spin_unlock(&cache
->lock
);
4124 spin_unlock(&cache
->space_info
->lock
);
4126 old_val
-= num_bytes
;
4127 btrfs_set_block_group_used(&cache
->item
, old_val
);
4128 cache
->pinned
+= num_bytes
;
4129 cache
->space_info
->bytes_pinned
+= num_bytes
;
4130 cache
->space_info
->bytes_used
-= num_bytes
;
4131 cache
->space_info
->disk_used
-= num_bytes
* factor
;
4132 spin_unlock(&cache
->lock
);
4133 spin_unlock(&cache
->space_info
->lock
);
4135 set_extent_dirty(info
->pinned_extents
,
4136 bytenr
, bytenr
+ num_bytes
- 1,
4137 GFP_NOFS
| __GFP_NOFAIL
);
4139 btrfs_put_block_group(cache
);
4141 bytenr
+= num_bytes
;
4146 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
4148 struct btrfs_block_group_cache
*cache
;
4151 cache
= btrfs_lookup_first_block_group(root
->fs_info
, search_start
);
4155 bytenr
= cache
->key
.objectid
;
4156 btrfs_put_block_group(cache
);
4161 static int pin_down_extent(struct btrfs_root
*root
,
4162 struct btrfs_block_group_cache
*cache
,
4163 u64 bytenr
, u64 num_bytes
, int reserved
)
4165 spin_lock(&cache
->space_info
->lock
);
4166 spin_lock(&cache
->lock
);
4167 cache
->pinned
+= num_bytes
;
4168 cache
->space_info
->bytes_pinned
+= num_bytes
;
4170 cache
->reserved
-= num_bytes
;
4171 cache
->space_info
->bytes_reserved
-= num_bytes
;
4172 cache
->space_info
->reservation_progress
++;
4174 spin_unlock(&cache
->lock
);
4175 spin_unlock(&cache
->space_info
->lock
);
4177 set_extent_dirty(root
->fs_info
->pinned_extents
, bytenr
,
4178 bytenr
+ num_bytes
- 1, GFP_NOFS
| __GFP_NOFAIL
);
4183 * this function must be called within transaction
4185 int btrfs_pin_extent(struct btrfs_root
*root
,
4186 u64 bytenr
, u64 num_bytes
, int reserved
)
4188 struct btrfs_block_group_cache
*cache
;
4190 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
4193 pin_down_extent(root
, cache
, bytenr
, num_bytes
, reserved
);
4195 btrfs_put_block_group(cache
);
4200 * update size of reserved extents. this function may return -EAGAIN
4201 * if 'reserve' is true or 'sinfo' is false.
4203 int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
4204 u64 num_bytes
, int reserve
, int sinfo
)
4208 struct btrfs_space_info
*space_info
= cache
->space_info
;
4209 spin_lock(&space_info
->lock
);
4210 spin_lock(&cache
->lock
);
4215 cache
->reserved
+= num_bytes
;
4216 space_info
->bytes_reserved
+= num_bytes
;
4220 space_info
->bytes_readonly
+= num_bytes
;
4221 cache
->reserved
-= num_bytes
;
4222 space_info
->bytes_reserved
-= num_bytes
;
4223 space_info
->reservation_progress
++;
4225 spin_unlock(&cache
->lock
);
4226 spin_unlock(&space_info
->lock
);
4228 spin_lock(&cache
->lock
);
4233 cache
->reserved
+= num_bytes
;
4235 cache
->reserved
-= num_bytes
;
4237 spin_unlock(&cache
->lock
);
4242 int btrfs_prepare_extent_commit(struct btrfs_trans_handle
*trans
,
4243 struct btrfs_root
*root
)
4245 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4246 struct btrfs_caching_control
*next
;
4247 struct btrfs_caching_control
*caching_ctl
;
4248 struct btrfs_block_group_cache
*cache
;
4250 down_write(&fs_info
->extent_commit_sem
);
4252 list_for_each_entry_safe(caching_ctl
, next
,
4253 &fs_info
->caching_block_groups
, list
) {
4254 cache
= caching_ctl
->block_group
;
4255 if (block_group_cache_done(cache
)) {
4256 cache
->last_byte_to_unpin
= (u64
)-1;
4257 list_del_init(&caching_ctl
->list
);
4258 put_caching_control(caching_ctl
);
4260 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
4264 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
4265 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
4267 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
4269 up_write(&fs_info
->extent_commit_sem
);
4271 update_global_block_rsv(fs_info
);
4275 static int unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
4277 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4278 struct btrfs_block_group_cache
*cache
= NULL
;
4281 while (start
<= end
) {
4283 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
4285 btrfs_put_block_group(cache
);
4286 cache
= btrfs_lookup_block_group(fs_info
, start
);
4290 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
4291 len
= min(len
, end
+ 1 - start
);
4293 if (start
< cache
->last_byte_to_unpin
) {
4294 len
= min(len
, cache
->last_byte_to_unpin
- start
);
4295 btrfs_add_free_space(cache
, start
, len
);
4300 spin_lock(&cache
->space_info
->lock
);
4301 spin_lock(&cache
->lock
);
4302 cache
->pinned
-= len
;
4303 cache
->space_info
->bytes_pinned
-= len
;
4305 cache
->space_info
->bytes_readonly
+= len
;
4306 } else if (cache
->reserved_pinned
> 0) {
4307 len
= min(len
, cache
->reserved_pinned
);
4308 cache
->reserved_pinned
-= len
;
4309 cache
->space_info
->bytes_reserved
+= len
;
4311 spin_unlock(&cache
->lock
);
4312 spin_unlock(&cache
->space_info
->lock
);
4316 btrfs_put_block_group(cache
);
4320 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
4321 struct btrfs_root
*root
)
4323 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4324 struct extent_io_tree
*unpin
;
4325 struct btrfs_block_rsv
*block_rsv
;
4326 struct btrfs_block_rsv
*next_rsv
;
4332 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
4333 unpin
= &fs_info
->freed_extents
[1];
4335 unpin
= &fs_info
->freed_extents
[0];
4338 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
4343 if (btrfs_test_opt(root
, DISCARD
))
4344 ret
= btrfs_discard_extent(root
, start
,
4345 end
+ 1 - start
, NULL
);
4347 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
4348 unpin_extent_range(root
, start
, end
);
4352 mutex_lock(&fs_info
->durable_block_rsv_mutex
);
4353 list_for_each_entry_safe(block_rsv
, next_rsv
,
4354 &fs_info
->durable_block_rsv_list
, list
) {
4356 idx
= trans
->transid
& 0x1;
4357 if (block_rsv
->freed
[idx
] > 0) {
4358 block_rsv_add_bytes(block_rsv
,
4359 block_rsv
->freed
[idx
], 0);
4360 block_rsv
->freed
[idx
] = 0;
4362 if (atomic_read(&block_rsv
->usage
) == 0) {
4363 btrfs_block_rsv_release(root
, block_rsv
, (u64
)-1);
4365 if (block_rsv
->freed
[0] == 0 &&
4366 block_rsv
->freed
[1] == 0) {
4367 list_del_init(&block_rsv
->list
);
4371 btrfs_block_rsv_release(root
, block_rsv
, 0);
4374 mutex_unlock(&fs_info
->durable_block_rsv_mutex
);
4379 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
4380 struct btrfs_root
*root
,
4381 u64 bytenr
, u64 num_bytes
, u64 parent
,
4382 u64 root_objectid
, u64 owner_objectid
,
4383 u64 owner_offset
, int refs_to_drop
,
4384 struct btrfs_delayed_extent_op
*extent_op
)
4386 struct btrfs_key key
;
4387 struct btrfs_path
*path
;
4388 struct btrfs_fs_info
*info
= root
->fs_info
;
4389 struct btrfs_root
*extent_root
= info
->extent_root
;
4390 struct extent_buffer
*leaf
;
4391 struct btrfs_extent_item
*ei
;
4392 struct btrfs_extent_inline_ref
*iref
;
4395 int extent_slot
= 0;
4396 int found_extent
= 0;
4401 path
= btrfs_alloc_path();
4406 path
->leave_spinning
= 1;
4408 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
4409 BUG_ON(!is_data
&& refs_to_drop
!= 1);
4411 ret
= lookup_extent_backref(trans
, extent_root
, path
, &iref
,
4412 bytenr
, num_bytes
, parent
,
4413 root_objectid
, owner_objectid
,
4416 extent_slot
= path
->slots
[0];
4417 while (extent_slot
>= 0) {
4418 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
4420 if (key
.objectid
!= bytenr
)
4422 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
4423 key
.offset
== num_bytes
) {
4427 if (path
->slots
[0] - extent_slot
> 5)
4431 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4432 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
4433 if (found_extent
&& item_size
< sizeof(*ei
))
4436 if (!found_extent
) {
4438 ret
= remove_extent_backref(trans
, extent_root
, path
,
4442 btrfs_release_path(path
);
4443 path
->leave_spinning
= 1;
4445 key
.objectid
= bytenr
;
4446 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
4447 key
.offset
= num_bytes
;
4449 ret
= btrfs_search_slot(trans
, extent_root
,
4452 printk(KERN_ERR
"umm, got %d back from search"
4453 ", was looking for %llu\n", ret
,
4454 (unsigned long long)bytenr
);
4455 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
4458 extent_slot
= path
->slots
[0];
4461 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
4463 printk(KERN_ERR
"btrfs unable to find ref byte nr %llu "
4464 "parent %llu root %llu owner %llu offset %llu\n",
4465 (unsigned long long)bytenr
,
4466 (unsigned long long)parent
,
4467 (unsigned long long)root_objectid
,
4468 (unsigned long long)owner_objectid
,
4469 (unsigned long long)owner_offset
);
4472 leaf
= path
->nodes
[0];
4473 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
4474 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4475 if (item_size
< sizeof(*ei
)) {
4476 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
4477 ret
= convert_extent_item_v0(trans
, extent_root
, path
,
4481 btrfs_release_path(path
);
4482 path
->leave_spinning
= 1;
4484 key
.objectid
= bytenr
;
4485 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
4486 key
.offset
= num_bytes
;
4488 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
4491 printk(KERN_ERR
"umm, got %d back from search"
4492 ", was looking for %llu\n", ret
,
4493 (unsigned long long)bytenr
);
4494 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
4497 extent_slot
= path
->slots
[0];
4498 leaf
= path
->nodes
[0];
4499 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
4502 BUG_ON(item_size
< sizeof(*ei
));
4503 ei
= btrfs_item_ptr(leaf
, extent_slot
,
4504 struct btrfs_extent_item
);
4505 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
) {
4506 struct btrfs_tree_block_info
*bi
;
4507 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
4508 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
4509 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
4512 refs
= btrfs_extent_refs(leaf
, ei
);
4513 BUG_ON(refs
< refs_to_drop
);
4514 refs
-= refs_to_drop
;
4518 __run_delayed_extent_op(extent_op
, leaf
, ei
);
4520 * In the case of inline back ref, reference count will
4521 * be updated by remove_extent_backref
4524 BUG_ON(!found_extent
);
4526 btrfs_set_extent_refs(leaf
, ei
, refs
);
4527 btrfs_mark_buffer_dirty(leaf
);
4530 ret
= remove_extent_backref(trans
, extent_root
, path
,
4537 BUG_ON(is_data
&& refs_to_drop
!=
4538 extent_data_ref_count(root
, path
, iref
));
4540 BUG_ON(path
->slots
[0] != extent_slot
);
4542 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
4543 path
->slots
[0] = extent_slot
;
4548 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
4551 btrfs_release_path(path
);
4554 ret
= btrfs_del_csums(trans
, root
, bytenr
, num_bytes
);
4557 invalidate_mapping_pages(info
->btree_inode
->i_mapping
,
4558 bytenr
>> PAGE_CACHE_SHIFT
,
4559 (bytenr
+ num_bytes
- 1) >> PAGE_CACHE_SHIFT
);
4562 ret
= update_block_group(trans
, root
, bytenr
, num_bytes
, 0);
4565 btrfs_free_path(path
);
4570 * when we free an block, it is possible (and likely) that we free the last
4571 * delayed ref for that extent as well. This searches the delayed ref tree for
4572 * a given extent, and if there are no other delayed refs to be processed, it
4573 * removes it from the tree.
4575 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
4576 struct btrfs_root
*root
, u64 bytenr
)
4578 struct btrfs_delayed_ref_head
*head
;
4579 struct btrfs_delayed_ref_root
*delayed_refs
;
4580 struct btrfs_delayed_ref_node
*ref
;
4581 struct rb_node
*node
;
4584 delayed_refs
= &trans
->transaction
->delayed_refs
;
4585 spin_lock(&delayed_refs
->lock
);
4586 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
4590 node
= rb_prev(&head
->node
.rb_node
);
4594 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
4596 /* there are still entries for this ref, we can't drop it */
4597 if (ref
->bytenr
== bytenr
)
4600 if (head
->extent_op
) {
4601 if (!head
->must_insert_reserved
)
4603 kfree(head
->extent_op
);
4604 head
->extent_op
= NULL
;
4608 * waiting for the lock here would deadlock. If someone else has it
4609 * locked they are already in the process of dropping it anyway
4611 if (!mutex_trylock(&head
->mutex
))
4615 * at this point we have a head with no other entries. Go
4616 * ahead and process it.
4618 head
->node
.in_tree
= 0;
4619 rb_erase(&head
->node
.rb_node
, &delayed_refs
->root
);
4621 delayed_refs
->num_entries
--;
4624 * we don't take a ref on the node because we're removing it from the
4625 * tree, so we just steal the ref the tree was holding.
4627 delayed_refs
->num_heads
--;
4628 if (list_empty(&head
->cluster
))
4629 delayed_refs
->num_heads_ready
--;
4631 list_del_init(&head
->cluster
);
4632 spin_unlock(&delayed_refs
->lock
);
4634 BUG_ON(head
->extent_op
);
4635 if (head
->must_insert_reserved
)
4638 mutex_unlock(&head
->mutex
);
4639 btrfs_put_delayed_ref(&head
->node
);
4642 spin_unlock(&delayed_refs
->lock
);
4646 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
4647 struct btrfs_root
*root
,
4648 struct extent_buffer
*buf
,
4649 u64 parent
, int last_ref
)
4651 struct btrfs_block_rsv
*block_rsv
;
4652 struct btrfs_block_group_cache
*cache
= NULL
;
4655 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
4656 ret
= btrfs_add_delayed_tree_ref(trans
, buf
->start
, buf
->len
,
4657 parent
, root
->root_key
.objectid
,
4658 btrfs_header_level(buf
),
4659 BTRFS_DROP_DELAYED_REF
, NULL
);
4666 block_rsv
= get_block_rsv(trans
, root
);
4667 cache
= btrfs_lookup_block_group(root
->fs_info
, buf
->start
);
4668 if (block_rsv
->space_info
!= cache
->space_info
)
4671 if (btrfs_header_generation(buf
) == trans
->transid
) {
4672 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
4673 ret
= check_ref_cleanup(trans
, root
, buf
->start
);
4678 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
4679 pin_down_extent(root
, cache
, buf
->start
, buf
->len
, 1);
4683 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
4685 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
4686 ret
= btrfs_update_reserved_bytes(cache
, buf
->len
, 0, 0);
4687 if (ret
== -EAGAIN
) {
4688 /* block group became read-only */
4689 btrfs_update_reserved_bytes(cache
, buf
->len
, 0, 1);
4694 spin_lock(&block_rsv
->lock
);
4695 if (block_rsv
->reserved
< block_rsv
->size
) {
4696 block_rsv
->reserved
+= buf
->len
;
4699 spin_unlock(&block_rsv
->lock
);
4702 spin_lock(&cache
->space_info
->lock
);
4703 cache
->space_info
->bytes_reserved
-= buf
->len
;
4704 cache
->space_info
->reservation_progress
++;
4705 spin_unlock(&cache
->space_info
->lock
);
4710 if (block_rsv
->durable
&& !cache
->ro
) {
4712 spin_lock(&cache
->lock
);
4714 cache
->reserved_pinned
+= buf
->len
;
4717 spin_unlock(&cache
->lock
);
4720 spin_lock(&block_rsv
->lock
);
4721 block_rsv
->freed
[trans
->transid
& 0x1] += buf
->len
;
4722 spin_unlock(&block_rsv
->lock
);
4727 * Deleting the buffer, clear the corrupt flag since it doesn't matter
4730 clear_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
);
4731 btrfs_put_block_group(cache
);
4734 int btrfs_free_extent(struct btrfs_trans_handle
*trans
,
4735 struct btrfs_root
*root
,
4736 u64 bytenr
, u64 num_bytes
, u64 parent
,
4737 u64 root_objectid
, u64 owner
, u64 offset
)
4742 * tree log blocks never actually go into the extent allocation
4743 * tree, just update pinning info and exit early.
4745 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
4746 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
4747 /* unlocks the pinned mutex */
4748 btrfs_pin_extent(root
, bytenr
, num_bytes
, 1);
4750 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
4751 ret
= btrfs_add_delayed_tree_ref(trans
, bytenr
, num_bytes
,
4752 parent
, root_objectid
, (int)owner
,
4753 BTRFS_DROP_DELAYED_REF
, NULL
);
4756 ret
= btrfs_add_delayed_data_ref(trans
, bytenr
, num_bytes
,
4757 parent
, root_objectid
, owner
,
4758 offset
, BTRFS_DROP_DELAYED_REF
, NULL
);
4764 static u64
stripe_align(struct btrfs_root
*root
, u64 val
)
4766 u64 mask
= ((u64
)root
->stripesize
- 1);
4767 u64 ret
= (val
+ mask
) & ~mask
;
4772 * when we wait for progress in the block group caching, its because
4773 * our allocation attempt failed at least once. So, we must sleep
4774 * and let some progress happen before we try again.
4776 * This function will sleep at least once waiting for new free space to
4777 * show up, and then it will check the block group free space numbers
4778 * for our min num_bytes. Another option is to have it go ahead
4779 * and look in the rbtree for a free extent of a given size, but this
4783 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
4786 struct btrfs_caching_control
*caching_ctl
;
4789 caching_ctl
= get_caching_control(cache
);
4793 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
4794 (cache
->free_space_ctl
->free_space
>= num_bytes
));
4796 put_caching_control(caching_ctl
);
4801 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
4803 struct btrfs_caching_control
*caching_ctl
;
4806 caching_ctl
= get_caching_control(cache
);
4810 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
4812 put_caching_control(caching_ctl
);
4816 static int get_block_group_index(struct btrfs_block_group_cache
*cache
)
4819 if (cache
->flags
& BTRFS_BLOCK_GROUP_RAID10
)
4821 else if (cache
->flags
& BTRFS_BLOCK_GROUP_RAID1
)
4823 else if (cache
->flags
& BTRFS_BLOCK_GROUP_DUP
)
4825 else if (cache
->flags
& BTRFS_BLOCK_GROUP_RAID0
)
4832 enum btrfs_loop_type
{
4833 LOOP_FIND_IDEAL
= 0,
4834 LOOP_CACHING_NOWAIT
= 1,
4835 LOOP_CACHING_WAIT
= 2,
4836 LOOP_ALLOC_CHUNK
= 3,
4837 LOOP_NO_EMPTY_SIZE
= 4,
4841 * walks the btree of allocated extents and find a hole of a given size.
4842 * The key ins is changed to record the hole:
4843 * ins->objectid == block start
4844 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4845 * ins->offset == number of blocks
4846 * Any available blocks before search_start are skipped.
4848 static noinline
int find_free_extent(struct btrfs_trans_handle
*trans
,
4849 struct btrfs_root
*orig_root
,
4850 u64 num_bytes
, u64 empty_size
,
4851 u64 search_start
, u64 search_end
,
4852 u64 hint_byte
, struct btrfs_key
*ins
,
4856 struct btrfs_root
*root
= orig_root
->fs_info
->extent_root
;
4857 struct btrfs_free_cluster
*last_ptr
= NULL
;
4858 struct btrfs_block_group_cache
*block_group
= NULL
;
4859 int empty_cluster
= 2 * 1024 * 1024;
4860 int allowed_chunk_alloc
= 0;
4861 int done_chunk_alloc
= 0;
4862 struct btrfs_space_info
*space_info
;
4863 int last_ptr_loop
= 0;
4866 bool found_uncached_bg
= false;
4867 bool failed_cluster_refill
= false;
4868 bool failed_alloc
= false;
4869 bool use_cluster
= true;
4870 u64 ideal_cache_percent
= 0;
4871 u64 ideal_cache_offset
= 0;
4873 WARN_ON(num_bytes
< root
->sectorsize
);
4874 btrfs_set_key_type(ins
, BTRFS_EXTENT_ITEM_KEY
);
4878 space_info
= __find_space_info(root
->fs_info
, data
);
4880 printk(KERN_ERR
"No space info for %llu\n", data
);
4885 * If the space info is for both data and metadata it means we have a
4886 * small filesystem and we can't use the clustering stuff.
4888 if (btrfs_mixed_space_info(space_info
))
4889 use_cluster
= false;
4891 if (orig_root
->ref_cows
|| empty_size
)
4892 allowed_chunk_alloc
= 1;
4894 if (data
& BTRFS_BLOCK_GROUP_METADATA
&& use_cluster
) {
4895 last_ptr
= &root
->fs_info
->meta_alloc_cluster
;
4896 if (!btrfs_test_opt(root
, SSD
))
4897 empty_cluster
= 64 * 1024;
4900 if ((data
& BTRFS_BLOCK_GROUP_DATA
) && use_cluster
&&
4901 btrfs_test_opt(root
, SSD
)) {
4902 last_ptr
= &root
->fs_info
->data_alloc_cluster
;
4906 spin_lock(&last_ptr
->lock
);
4907 if (last_ptr
->block_group
)
4908 hint_byte
= last_ptr
->window_start
;
4909 spin_unlock(&last_ptr
->lock
);
4912 search_start
= max(search_start
, first_logical_byte(root
, 0));
4913 search_start
= max(search_start
, hint_byte
);
4918 if (search_start
== hint_byte
) {
4920 block_group
= btrfs_lookup_block_group(root
->fs_info
,
4923 * we don't want to use the block group if it doesn't match our
4924 * allocation bits, or if its not cached.
4926 * However if we are re-searching with an ideal block group
4927 * picked out then we don't care that the block group is cached.
4929 if (block_group
&& block_group_bits(block_group
, data
) &&
4930 (block_group
->cached
!= BTRFS_CACHE_NO
||
4931 search_start
== ideal_cache_offset
)) {
4932 down_read(&space_info
->groups_sem
);
4933 if (list_empty(&block_group
->list
) ||
4936 * someone is removing this block group,
4937 * we can't jump into the have_block_group
4938 * target because our list pointers are not
4941 btrfs_put_block_group(block_group
);
4942 up_read(&space_info
->groups_sem
);
4944 index
= get_block_group_index(block_group
);
4945 goto have_block_group
;
4947 } else if (block_group
) {
4948 btrfs_put_block_group(block_group
);
4952 down_read(&space_info
->groups_sem
);
4953 list_for_each_entry(block_group
, &space_info
->block_groups
[index
],
4958 btrfs_get_block_group(block_group
);
4959 search_start
= block_group
->key
.objectid
;
4962 * this can happen if we end up cycling through all the
4963 * raid types, but we want to make sure we only allocate
4964 * for the proper type.
4966 if (!block_group_bits(block_group
, data
)) {
4967 u64 extra
= BTRFS_BLOCK_GROUP_DUP
|
4968 BTRFS_BLOCK_GROUP_RAID1
|
4969 BTRFS_BLOCK_GROUP_RAID10
;
4972 * if they asked for extra copies and this block group
4973 * doesn't provide them, bail. This does allow us to
4974 * fill raid0 from raid1.
4976 if ((data
& extra
) && !(block_group
->flags
& extra
))
4981 if (unlikely(block_group
->cached
== BTRFS_CACHE_NO
)) {
4984 ret
= cache_block_group(block_group
, trans
,
4986 if (block_group
->cached
== BTRFS_CACHE_FINISHED
)
4987 goto have_block_group
;
4989 free_percent
= btrfs_block_group_used(&block_group
->item
);
4990 free_percent
*= 100;
4991 free_percent
= div64_u64(free_percent
,
4992 block_group
->key
.offset
);
4993 free_percent
= 100 - free_percent
;
4994 if (free_percent
> ideal_cache_percent
&&
4995 likely(!block_group
->ro
)) {
4996 ideal_cache_offset
= block_group
->key
.objectid
;
4997 ideal_cache_percent
= free_percent
;
5001 * We only want to start kthread caching if we are at
5002 * the point where we will wait for caching to make
5003 * progress, or if our ideal search is over and we've
5004 * found somebody to start caching.
5006 if (loop
> LOOP_CACHING_NOWAIT
||
5007 (loop
> LOOP_FIND_IDEAL
&&
5008 atomic_read(&space_info
->caching_threads
) < 2)) {
5009 ret
= cache_block_group(block_group
, trans
,
5013 found_uncached_bg
= true;
5016 * If loop is set for cached only, try the next block
5019 if (loop
== LOOP_FIND_IDEAL
)
5023 cached
= block_group_cache_done(block_group
);
5024 if (unlikely(!cached
))
5025 found_uncached_bg
= true;
5027 if (unlikely(block_group
->ro
))
5030 spin_lock(&block_group
->free_space_ctl
->tree_lock
);
5032 block_group
->free_space_ctl
->free_space
<
5033 num_bytes
+ empty_size
) {
5034 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
5037 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
5040 * Ok we want to try and use the cluster allocator, so lets look
5041 * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
5042 * have tried the cluster allocator plenty of times at this
5043 * point and not have found anything, so we are likely way too
5044 * fragmented for the clustering stuff to find anything, so lets
5045 * just skip it and let the allocator find whatever block it can
5048 if (last_ptr
&& loop
< LOOP_NO_EMPTY_SIZE
) {
5050 * the refill lock keeps out other
5051 * people trying to start a new cluster
5053 spin_lock(&last_ptr
->refill_lock
);
5054 if (last_ptr
->block_group
&&
5055 (last_ptr
->block_group
->ro
||
5056 !block_group_bits(last_ptr
->block_group
, data
))) {
5058 goto refill_cluster
;
5061 offset
= btrfs_alloc_from_cluster(block_group
, last_ptr
,
5062 num_bytes
, search_start
);
5064 /* we have a block, we're done */
5065 spin_unlock(&last_ptr
->refill_lock
);
5069 spin_lock(&last_ptr
->lock
);
5071 * whoops, this cluster doesn't actually point to
5072 * this block group. Get a ref on the block
5073 * group is does point to and try again
5075 if (!last_ptr_loop
&& last_ptr
->block_group
&&
5076 last_ptr
->block_group
!= block_group
) {
5078 btrfs_put_block_group(block_group
);
5079 block_group
= last_ptr
->block_group
;
5080 btrfs_get_block_group(block_group
);
5081 spin_unlock(&last_ptr
->lock
);
5082 spin_unlock(&last_ptr
->refill_lock
);
5085 search_start
= block_group
->key
.objectid
;
5087 * we know this block group is properly
5088 * in the list because
5089 * btrfs_remove_block_group, drops the
5090 * cluster before it removes the block
5091 * group from the list
5093 goto have_block_group
;
5095 spin_unlock(&last_ptr
->lock
);
5098 * this cluster didn't work out, free it and
5101 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
5105 /* allocate a cluster in this block group */
5106 ret
= btrfs_find_space_cluster(trans
, root
,
5107 block_group
, last_ptr
,
5109 empty_cluster
+ empty_size
);
5112 * now pull our allocation out of this
5115 offset
= btrfs_alloc_from_cluster(block_group
,
5116 last_ptr
, num_bytes
,
5119 /* we found one, proceed */
5120 spin_unlock(&last_ptr
->refill_lock
);
5123 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
5124 && !failed_cluster_refill
) {
5125 spin_unlock(&last_ptr
->refill_lock
);
5127 failed_cluster_refill
= true;
5128 wait_block_group_cache_progress(block_group
,
5129 num_bytes
+ empty_cluster
+ empty_size
);
5130 goto have_block_group
;
5134 * at this point we either didn't find a cluster
5135 * or we weren't able to allocate a block from our
5136 * cluster. Free the cluster we've been trying
5137 * to use, and go to the next block group
5139 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
5140 spin_unlock(&last_ptr
->refill_lock
);
5144 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
5145 num_bytes
, empty_size
);
5147 * If we didn't find a chunk, and we haven't failed on this
5148 * block group before, and this block group is in the middle of
5149 * caching and we are ok with waiting, then go ahead and wait
5150 * for progress to be made, and set failed_alloc to true.
5152 * If failed_alloc is true then we've already waited on this
5153 * block group once and should move on to the next block group.
5155 if (!offset
&& !failed_alloc
&& !cached
&&
5156 loop
> LOOP_CACHING_NOWAIT
) {
5157 wait_block_group_cache_progress(block_group
,
5158 num_bytes
+ empty_size
);
5159 failed_alloc
= true;
5160 goto have_block_group
;
5161 } else if (!offset
) {
5165 search_start
= stripe_align(root
, offset
);
5166 /* move on to the next group */
5167 if (search_start
+ num_bytes
>= search_end
) {
5168 btrfs_add_free_space(block_group
, offset
, num_bytes
);
5172 /* move on to the next group */
5173 if (search_start
+ num_bytes
>
5174 block_group
->key
.objectid
+ block_group
->key
.offset
) {
5175 btrfs_add_free_space(block_group
, offset
, num_bytes
);
5179 ins
->objectid
= search_start
;
5180 ins
->offset
= num_bytes
;
5182 if (offset
< search_start
)
5183 btrfs_add_free_space(block_group
, offset
,
5184 search_start
- offset
);
5185 BUG_ON(offset
> search_start
);
5187 ret
= btrfs_update_reserved_bytes(block_group
, num_bytes
, 1,
5188 (data
& BTRFS_BLOCK_GROUP_DATA
));
5189 if (ret
== -EAGAIN
) {
5190 btrfs_add_free_space(block_group
, offset
, num_bytes
);
5194 /* we are all good, lets return */
5195 ins
->objectid
= search_start
;
5196 ins
->offset
= num_bytes
;
5198 if (offset
< search_start
)
5199 btrfs_add_free_space(block_group
, offset
,
5200 search_start
- offset
);
5201 BUG_ON(offset
> search_start
);
5202 btrfs_put_block_group(block_group
);
5205 failed_cluster_refill
= false;
5206 failed_alloc
= false;
5207 BUG_ON(index
!= get_block_group_index(block_group
));
5208 btrfs_put_block_group(block_group
);
5210 up_read(&space_info
->groups_sem
);
5212 if (!ins
->objectid
&& ++index
< BTRFS_NR_RAID_TYPES
)
5215 /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5216 * for them to make caching progress. Also
5217 * determine the best possible bg to cache
5218 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5219 * caching kthreads as we move along
5220 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5221 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5222 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5225 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
) {
5227 if (loop
== LOOP_FIND_IDEAL
&& found_uncached_bg
) {
5228 found_uncached_bg
= false;
5230 if (!ideal_cache_percent
&&
5231 atomic_read(&space_info
->caching_threads
))
5235 * 1 of the following 2 things have happened so far
5237 * 1) We found an ideal block group for caching that
5238 * is mostly full and will cache quickly, so we might
5239 * as well wait for it.
5241 * 2) We searched for cached only and we didn't find
5242 * anything, and we didn't start any caching kthreads
5243 * either, so chances are we will loop through and
5244 * start a couple caching kthreads, and then come back
5245 * around and just wait for them. This will be slower
5246 * because we will have 2 caching kthreads reading at
5247 * the same time when we could have just started one
5248 * and waited for it to get far enough to give us an
5249 * allocation, so go ahead and go to the wait caching
5252 loop
= LOOP_CACHING_WAIT
;
5253 search_start
= ideal_cache_offset
;
5254 ideal_cache_percent
= 0;
5256 } else if (loop
== LOOP_FIND_IDEAL
) {
5258 * Didn't find a uncached bg, wait on anything we find
5261 loop
= LOOP_CACHING_WAIT
;
5267 if (loop
== LOOP_ALLOC_CHUNK
) {
5268 if (allowed_chunk_alloc
) {
5269 ret
= do_chunk_alloc(trans
, root
, num_bytes
+
5270 2 * 1024 * 1024, data
,
5271 CHUNK_ALLOC_LIMITED
);
5272 allowed_chunk_alloc
= 0;
5274 done_chunk_alloc
= 1;
5275 } else if (!done_chunk_alloc
&&
5276 space_info
->force_alloc
==
5277 CHUNK_ALLOC_NO_FORCE
) {
5278 space_info
->force_alloc
= CHUNK_ALLOC_LIMITED
;
5282 * We didn't allocate a chunk, go ahead and drop the
5283 * empty size and loop again.
5285 if (!done_chunk_alloc
)
5286 loop
= LOOP_NO_EMPTY_SIZE
;
5289 if (loop
== LOOP_NO_EMPTY_SIZE
) {
5295 } else if (!ins
->objectid
) {
5297 } else if (ins
->objectid
) {
5304 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
5305 int dump_block_groups
)
5307 struct btrfs_block_group_cache
*cache
;
5310 spin_lock(&info
->lock
);
5311 printk(KERN_INFO
"space_info has %llu free, is %sfull\n",
5312 (unsigned long long)(info
->total_bytes
- info
->bytes_used
-
5313 info
->bytes_pinned
- info
->bytes_reserved
-
5314 info
->bytes_readonly
),
5315 (info
->full
) ? "" : "not ");
5316 printk(KERN_INFO
"space_info total=%llu, used=%llu, pinned=%llu, "
5317 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5318 (unsigned long long)info
->total_bytes
,
5319 (unsigned long long)info
->bytes_used
,
5320 (unsigned long long)info
->bytes_pinned
,
5321 (unsigned long long)info
->bytes_reserved
,
5322 (unsigned long long)info
->bytes_may_use
,
5323 (unsigned long long)info
->bytes_readonly
);
5324 spin_unlock(&info
->lock
);
5326 if (!dump_block_groups
)
5329 down_read(&info
->groups_sem
);
5331 list_for_each_entry(cache
, &info
->block_groups
[index
], list
) {
5332 spin_lock(&cache
->lock
);
5333 printk(KERN_INFO
"block group %llu has %llu bytes, %llu used "
5334 "%llu pinned %llu reserved\n",
5335 (unsigned long long)cache
->key
.objectid
,
5336 (unsigned long long)cache
->key
.offset
,
5337 (unsigned long long)btrfs_block_group_used(&cache
->item
),
5338 (unsigned long long)cache
->pinned
,
5339 (unsigned long long)cache
->reserved
);
5340 btrfs_dump_free_space(cache
, bytes
);
5341 spin_unlock(&cache
->lock
);
5343 if (++index
< BTRFS_NR_RAID_TYPES
)
5345 up_read(&info
->groups_sem
);
5348 int btrfs_reserve_extent(struct btrfs_trans_handle
*trans
,
5349 struct btrfs_root
*root
,
5350 u64 num_bytes
, u64 min_alloc_size
,
5351 u64 empty_size
, u64 hint_byte
,
5352 u64 search_end
, struct btrfs_key
*ins
,
5356 u64 search_start
= 0;
5358 data
= btrfs_get_alloc_profile(root
, data
);
5361 * the only place that sets empty_size is btrfs_realloc_node, which
5362 * is not called recursively on allocations
5364 if (empty_size
|| root
->ref_cows
)
5365 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
5366 num_bytes
+ 2 * 1024 * 1024, data
,
5367 CHUNK_ALLOC_NO_FORCE
);
5369 WARN_ON(num_bytes
< root
->sectorsize
);
5370 ret
= find_free_extent(trans
, root
, num_bytes
, empty_size
,
5371 search_start
, search_end
, hint_byte
,
5374 if (ret
== -ENOSPC
&& num_bytes
> min_alloc_size
) {
5375 num_bytes
= num_bytes
>> 1;
5376 num_bytes
= num_bytes
& ~(root
->sectorsize
- 1);
5377 num_bytes
= max(num_bytes
, min_alloc_size
);
5378 do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
5379 num_bytes
, data
, CHUNK_ALLOC_FORCE
);
5382 if (ret
== -ENOSPC
&& btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
5383 struct btrfs_space_info
*sinfo
;
5385 sinfo
= __find_space_info(root
->fs_info
, data
);
5386 printk(KERN_ERR
"btrfs allocation failed flags %llu, "
5387 "wanted %llu\n", (unsigned long long)data
,
5388 (unsigned long long)num_bytes
);
5389 dump_space_info(sinfo
, num_bytes
, 1);
5392 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, ins
->offset
);
5397 int btrfs_free_reserved_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
5399 struct btrfs_block_group_cache
*cache
;
5402 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
5404 printk(KERN_ERR
"Unable to find block group for %llu\n",
5405 (unsigned long long)start
);
5409 if (btrfs_test_opt(root
, DISCARD
))
5410 ret
= btrfs_discard_extent(root
, start
, len
, NULL
);
5412 btrfs_add_free_space(cache
, start
, len
);
5413 btrfs_update_reserved_bytes(cache
, len
, 0, 1);
5414 btrfs_put_block_group(cache
);
5416 trace_btrfs_reserved_extent_free(root
, start
, len
);
5421 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
5422 struct btrfs_root
*root
,
5423 u64 parent
, u64 root_objectid
,
5424 u64 flags
, u64 owner
, u64 offset
,
5425 struct btrfs_key
*ins
, int ref_mod
)
5428 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5429 struct btrfs_extent_item
*extent_item
;
5430 struct btrfs_extent_inline_ref
*iref
;
5431 struct btrfs_path
*path
;
5432 struct extent_buffer
*leaf
;
5437 type
= BTRFS_SHARED_DATA_REF_KEY
;
5439 type
= BTRFS_EXTENT_DATA_REF_KEY
;
5441 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
5443 path
= btrfs_alloc_path();
5447 path
->leave_spinning
= 1;
5448 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
5452 leaf
= path
->nodes
[0];
5453 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
5454 struct btrfs_extent_item
);
5455 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
5456 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
5457 btrfs_set_extent_flags(leaf
, extent_item
,
5458 flags
| BTRFS_EXTENT_FLAG_DATA
);
5460 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
5461 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
5463 struct btrfs_shared_data_ref
*ref
;
5464 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
5465 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
5466 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
5468 struct btrfs_extent_data_ref
*ref
;
5469 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
5470 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
5471 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
5472 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
5473 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
5476 btrfs_mark_buffer_dirty(path
->nodes
[0]);
5477 btrfs_free_path(path
);
5479 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1);
5481 printk(KERN_ERR
"btrfs update block group failed for %llu "
5482 "%llu\n", (unsigned long long)ins
->objectid
,
5483 (unsigned long long)ins
->offset
);
5489 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
5490 struct btrfs_root
*root
,
5491 u64 parent
, u64 root_objectid
,
5492 u64 flags
, struct btrfs_disk_key
*key
,
5493 int level
, struct btrfs_key
*ins
)
5496 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5497 struct btrfs_extent_item
*extent_item
;
5498 struct btrfs_tree_block_info
*block_info
;
5499 struct btrfs_extent_inline_ref
*iref
;
5500 struct btrfs_path
*path
;
5501 struct extent_buffer
*leaf
;
5502 u32 size
= sizeof(*extent_item
) + sizeof(*block_info
) + sizeof(*iref
);
5504 path
= btrfs_alloc_path();
5507 path
->leave_spinning
= 1;
5508 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
5512 leaf
= path
->nodes
[0];
5513 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
5514 struct btrfs_extent_item
);
5515 btrfs_set_extent_refs(leaf
, extent_item
, 1);
5516 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
5517 btrfs_set_extent_flags(leaf
, extent_item
,
5518 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
5519 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
5521 btrfs_set_tree_block_key(leaf
, block_info
, key
);
5522 btrfs_set_tree_block_level(leaf
, block_info
, level
);
5524 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
5526 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
5527 btrfs_set_extent_inline_ref_type(leaf
, iref
,
5528 BTRFS_SHARED_BLOCK_REF_KEY
);
5529 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
5531 btrfs_set_extent_inline_ref_type(leaf
, iref
,
5532 BTRFS_TREE_BLOCK_REF_KEY
);
5533 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
5536 btrfs_mark_buffer_dirty(leaf
);
5537 btrfs_free_path(path
);
5539 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1);
5541 printk(KERN_ERR
"btrfs update block group failed for %llu "
5542 "%llu\n", (unsigned long long)ins
->objectid
,
5543 (unsigned long long)ins
->offset
);
5549 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
5550 struct btrfs_root
*root
,
5551 u64 root_objectid
, u64 owner
,
5552 u64 offset
, struct btrfs_key
*ins
)
5556 BUG_ON(root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
5558 ret
= btrfs_add_delayed_data_ref(trans
, ins
->objectid
, ins
->offset
,
5559 0, root_objectid
, owner
, offset
,
5560 BTRFS_ADD_DELAYED_EXTENT
, NULL
);
5565 * this is used by the tree logging recovery code. It records that
5566 * an extent has been allocated and makes sure to clear the free
5567 * space cache bits as well
5569 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
5570 struct btrfs_root
*root
,
5571 u64 root_objectid
, u64 owner
, u64 offset
,
5572 struct btrfs_key
*ins
)
5575 struct btrfs_block_group_cache
*block_group
;
5576 struct btrfs_caching_control
*caching_ctl
;
5577 u64 start
= ins
->objectid
;
5578 u64 num_bytes
= ins
->offset
;
5580 block_group
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
5581 cache_block_group(block_group
, trans
, NULL
, 0);
5582 caching_ctl
= get_caching_control(block_group
);
5585 BUG_ON(!block_group_cache_done(block_group
));
5586 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
5589 mutex_lock(&caching_ctl
->mutex
);
5591 if (start
>= caching_ctl
->progress
) {
5592 ret
= add_excluded_extent(root
, start
, num_bytes
);
5594 } else if (start
+ num_bytes
<= caching_ctl
->progress
) {
5595 ret
= btrfs_remove_free_space(block_group
,
5599 num_bytes
= caching_ctl
->progress
- start
;
5600 ret
= btrfs_remove_free_space(block_group
,
5604 start
= caching_ctl
->progress
;
5605 num_bytes
= ins
->objectid
+ ins
->offset
-
5606 caching_ctl
->progress
;
5607 ret
= add_excluded_extent(root
, start
, num_bytes
);
5611 mutex_unlock(&caching_ctl
->mutex
);
5612 put_caching_control(caching_ctl
);
5615 ret
= btrfs_update_reserved_bytes(block_group
, ins
->offset
, 1, 1);
5617 btrfs_put_block_group(block_group
);
5618 ret
= alloc_reserved_file_extent(trans
, root
, 0, root_objectid
,
5619 0, owner
, offset
, ins
, 1);
5623 struct extent_buffer
*btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
,
5624 struct btrfs_root
*root
,
5625 u64 bytenr
, u32 blocksize
,
5628 struct extent_buffer
*buf
;
5630 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
5632 return ERR_PTR(-ENOMEM
);
5633 btrfs_set_header_generation(buf
, trans
->transid
);
5634 btrfs_set_buffer_lockdep_class(buf
, level
);
5635 btrfs_tree_lock(buf
);
5636 clean_tree_block(trans
, root
, buf
);
5638 btrfs_set_lock_blocking(buf
);
5639 btrfs_set_buffer_uptodate(buf
);
5641 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
5643 * we allow two log transactions at a time, use different
5644 * EXENT bit to differentiate dirty pages.
5646 if (root
->log_transid
% 2 == 0)
5647 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
5648 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
5650 set_extent_new(&root
->dirty_log_pages
, buf
->start
,
5651 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
5653 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
5654 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
5656 trans
->blocks_used
++;
5657 /* this returns a buffer locked for blocking */
5661 static struct btrfs_block_rsv
*
5662 use_block_rsv(struct btrfs_trans_handle
*trans
,
5663 struct btrfs_root
*root
, u32 blocksize
)
5665 struct btrfs_block_rsv
*block_rsv
;
5666 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
5669 block_rsv
= get_block_rsv(trans
, root
);
5671 if (block_rsv
->size
== 0) {
5672 ret
= reserve_metadata_bytes(trans
, root
, block_rsv
,
5675 * If we couldn't reserve metadata bytes try and use some from
5676 * the global reserve.
5678 if (ret
&& block_rsv
!= global_rsv
) {
5679 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
5682 return ERR_PTR(ret
);
5684 return ERR_PTR(ret
);
5689 ret
= block_rsv_use_bytes(block_rsv
, blocksize
);
5694 ret
= reserve_metadata_bytes(trans
, root
, block_rsv
, blocksize
,
5697 spin_lock(&block_rsv
->lock
);
5698 block_rsv
->size
+= blocksize
;
5699 spin_unlock(&block_rsv
->lock
);
5701 } else if (ret
&& block_rsv
!= global_rsv
) {
5702 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
5708 return ERR_PTR(-ENOSPC
);
5711 static void unuse_block_rsv(struct btrfs_block_rsv
*block_rsv
, u32 blocksize
)
5713 block_rsv_add_bytes(block_rsv
, blocksize
, 0);
5714 block_rsv_release_bytes(block_rsv
, NULL
, 0);
5718 * finds a free extent and does all the dirty work required for allocation
5719 * returns the key for the extent through ins, and a tree buffer for
5720 * the first block of the extent through buf.
5722 * returns the tree buffer or NULL.
5724 struct extent_buffer
*btrfs_alloc_free_block(struct btrfs_trans_handle
*trans
,
5725 struct btrfs_root
*root
, u32 blocksize
,
5726 u64 parent
, u64 root_objectid
,
5727 struct btrfs_disk_key
*key
, int level
,
5728 u64 hint
, u64 empty_size
)
5730 struct btrfs_key ins
;
5731 struct btrfs_block_rsv
*block_rsv
;
5732 struct extent_buffer
*buf
;
5737 block_rsv
= use_block_rsv(trans
, root
, blocksize
);
5738 if (IS_ERR(block_rsv
))
5739 return ERR_CAST(block_rsv
);
5741 ret
= btrfs_reserve_extent(trans
, root
, blocksize
, blocksize
,
5742 empty_size
, hint
, (u64
)-1, &ins
, 0);
5744 unuse_block_rsv(block_rsv
, blocksize
);
5745 return ERR_PTR(ret
);
5748 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
,
5750 BUG_ON(IS_ERR(buf
));
5752 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
5754 parent
= ins
.objectid
;
5755 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
5759 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
5760 struct btrfs_delayed_extent_op
*extent_op
;
5761 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
5764 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
5766 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
5767 extent_op
->flags_to_set
= flags
;
5768 extent_op
->update_key
= 1;
5769 extent_op
->update_flags
= 1;
5770 extent_op
->is_data
= 0;
5772 ret
= btrfs_add_delayed_tree_ref(trans
, ins
.objectid
,
5773 ins
.offset
, parent
, root_objectid
,
5774 level
, BTRFS_ADD_DELAYED_EXTENT
,
5781 struct walk_control
{
5782 u64 refs
[BTRFS_MAX_LEVEL
];
5783 u64 flags
[BTRFS_MAX_LEVEL
];
5784 struct btrfs_key update_progress
;
5794 #define DROP_REFERENCE 1
5795 #define UPDATE_BACKREF 2
5797 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
5798 struct btrfs_root
*root
,
5799 struct walk_control
*wc
,
5800 struct btrfs_path
*path
)
5808 struct btrfs_key key
;
5809 struct extent_buffer
*eb
;
5814 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
5815 wc
->reada_count
= wc
->reada_count
* 2 / 3;
5816 wc
->reada_count
= max(wc
->reada_count
, 2);
5818 wc
->reada_count
= wc
->reada_count
* 3 / 2;
5819 wc
->reada_count
= min_t(int, wc
->reada_count
,
5820 BTRFS_NODEPTRS_PER_BLOCK(root
));
5823 eb
= path
->nodes
[wc
->level
];
5824 nritems
= btrfs_header_nritems(eb
);
5825 blocksize
= btrfs_level_size(root
, wc
->level
- 1);
5827 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
5828 if (nread
>= wc
->reada_count
)
5832 bytenr
= btrfs_node_blockptr(eb
, slot
);
5833 generation
= btrfs_node_ptr_generation(eb
, slot
);
5835 if (slot
== path
->slots
[wc
->level
])
5838 if (wc
->stage
== UPDATE_BACKREF
&&
5839 generation
<= root
->root_key
.offset
)
5842 /* We don't lock the tree block, it's OK to be racy here */
5843 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, blocksize
,
5848 if (wc
->stage
== DROP_REFERENCE
) {
5852 if (wc
->level
== 1 &&
5853 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
5855 if (!wc
->update_ref
||
5856 generation
<= root
->root_key
.offset
)
5858 btrfs_node_key_to_cpu(eb
, &key
, slot
);
5859 ret
= btrfs_comp_cpu_keys(&key
,
5860 &wc
->update_progress
);
5864 if (wc
->level
== 1 &&
5865 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
5869 ret
= readahead_tree_block(root
, bytenr
, blocksize
,
5875 wc
->reada_slot
= slot
;
5879 * hepler to process tree block while walking down the tree.
5881 * when wc->stage == UPDATE_BACKREF, this function updates
5882 * back refs for pointers in the block.
5884 * NOTE: return value 1 means we should stop walking down.
5886 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
5887 struct btrfs_root
*root
,
5888 struct btrfs_path
*path
,
5889 struct walk_control
*wc
, int lookup_info
)
5891 int level
= wc
->level
;
5892 struct extent_buffer
*eb
= path
->nodes
[level
];
5893 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
5896 if (wc
->stage
== UPDATE_BACKREF
&&
5897 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
5901 * when reference count of tree block is 1, it won't increase
5902 * again. once full backref flag is set, we never clear it.
5905 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
5906 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
5907 BUG_ON(!path
->locks
[level
]);
5908 ret
= btrfs_lookup_extent_info(trans
, root
,
5913 BUG_ON(wc
->refs
[level
] == 0);
5916 if (wc
->stage
== DROP_REFERENCE
) {
5917 if (wc
->refs
[level
] > 1)
5920 if (path
->locks
[level
] && !wc
->keep_locks
) {
5921 btrfs_tree_unlock(eb
);
5922 path
->locks
[level
] = 0;
5927 /* wc->stage == UPDATE_BACKREF */
5928 if (!(wc
->flags
[level
] & flag
)) {
5929 BUG_ON(!path
->locks
[level
]);
5930 ret
= btrfs_inc_ref(trans
, root
, eb
, 1);
5932 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
5934 ret
= btrfs_set_disk_extent_flags(trans
, root
, eb
->start
,
5937 wc
->flags
[level
] |= flag
;
5941 * the block is shared by multiple trees, so it's not good to
5942 * keep the tree lock
5944 if (path
->locks
[level
] && level
> 0) {
5945 btrfs_tree_unlock(eb
);
5946 path
->locks
[level
] = 0;
5952 * hepler to process tree block pointer.
5954 * when wc->stage == DROP_REFERENCE, this function checks
5955 * reference count of the block pointed to. if the block
5956 * is shared and we need update back refs for the subtree
5957 * rooted at the block, this function changes wc->stage to
5958 * UPDATE_BACKREF. if the block is shared and there is no
5959 * need to update back, this function drops the reference
5962 * NOTE: return value 1 means we should stop walking down.
5964 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
5965 struct btrfs_root
*root
,
5966 struct btrfs_path
*path
,
5967 struct walk_control
*wc
, int *lookup_info
)
5973 struct btrfs_key key
;
5974 struct extent_buffer
*next
;
5975 int level
= wc
->level
;
5979 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
5980 path
->slots
[level
]);
5982 * if the lower level block was created before the snapshot
5983 * was created, we know there is no need to update back refs
5986 if (wc
->stage
== UPDATE_BACKREF
&&
5987 generation
<= root
->root_key
.offset
) {
5992 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
5993 blocksize
= btrfs_level_size(root
, level
- 1);
5995 next
= btrfs_find_tree_block(root
, bytenr
, blocksize
);
5997 next
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
6002 btrfs_tree_lock(next
);
6003 btrfs_set_lock_blocking(next
);
6005 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, blocksize
,
6006 &wc
->refs
[level
- 1],
6007 &wc
->flags
[level
- 1]);
6009 BUG_ON(wc
->refs
[level
- 1] == 0);
6012 if (wc
->stage
== DROP_REFERENCE
) {
6013 if (wc
->refs
[level
- 1] > 1) {
6015 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
6018 if (!wc
->update_ref
||
6019 generation
<= root
->root_key
.offset
)
6022 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
6023 path
->slots
[level
]);
6024 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
6028 wc
->stage
= UPDATE_BACKREF
;
6029 wc
->shared_level
= level
- 1;
6033 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
6037 if (!btrfs_buffer_uptodate(next
, generation
)) {
6038 btrfs_tree_unlock(next
);
6039 free_extent_buffer(next
);
6045 if (reada
&& level
== 1)
6046 reada_walk_down(trans
, root
, wc
, path
);
6047 next
= read_tree_block(root
, bytenr
, blocksize
, generation
);
6050 btrfs_tree_lock(next
);
6051 btrfs_set_lock_blocking(next
);
6055 BUG_ON(level
!= btrfs_header_level(next
));
6056 path
->nodes
[level
] = next
;
6057 path
->slots
[level
] = 0;
6058 path
->locks
[level
] = 1;
6064 wc
->refs
[level
- 1] = 0;
6065 wc
->flags
[level
- 1] = 0;
6066 if (wc
->stage
== DROP_REFERENCE
) {
6067 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
6068 parent
= path
->nodes
[level
]->start
;
6070 BUG_ON(root
->root_key
.objectid
!=
6071 btrfs_header_owner(path
->nodes
[level
]));
6075 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
, parent
,
6076 root
->root_key
.objectid
, level
- 1, 0);
6079 btrfs_tree_unlock(next
);
6080 free_extent_buffer(next
);
6086 * hepler to process tree block while walking up the tree.
6088 * when wc->stage == DROP_REFERENCE, this function drops
6089 * reference count on the block.
6091 * when wc->stage == UPDATE_BACKREF, this function changes
6092 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6093 * to UPDATE_BACKREF previously while processing the block.
6095 * NOTE: return value 1 means we should stop walking up.
6097 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
6098 struct btrfs_root
*root
,
6099 struct btrfs_path
*path
,
6100 struct walk_control
*wc
)
6103 int level
= wc
->level
;
6104 struct extent_buffer
*eb
= path
->nodes
[level
];
6107 if (wc
->stage
== UPDATE_BACKREF
) {
6108 BUG_ON(wc
->shared_level
< level
);
6109 if (level
< wc
->shared_level
)
6112 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
6116 wc
->stage
= DROP_REFERENCE
;
6117 wc
->shared_level
= -1;
6118 path
->slots
[level
] = 0;
6121 * check reference count again if the block isn't locked.
6122 * we should start walking down the tree again if reference
6125 if (!path
->locks
[level
]) {
6127 btrfs_tree_lock(eb
);
6128 btrfs_set_lock_blocking(eb
);
6129 path
->locks
[level
] = 1;
6131 ret
= btrfs_lookup_extent_info(trans
, root
,
6136 BUG_ON(wc
->refs
[level
] == 0);
6137 if (wc
->refs
[level
] == 1) {
6138 btrfs_tree_unlock(eb
);
6139 path
->locks
[level
] = 0;
6145 /* wc->stage == DROP_REFERENCE */
6146 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
6148 if (wc
->refs
[level
] == 1) {
6150 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
6151 ret
= btrfs_dec_ref(trans
, root
, eb
, 1);
6153 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
6156 /* make block locked assertion in clean_tree_block happy */
6157 if (!path
->locks
[level
] &&
6158 btrfs_header_generation(eb
) == trans
->transid
) {
6159 btrfs_tree_lock(eb
);
6160 btrfs_set_lock_blocking(eb
);
6161 path
->locks
[level
] = 1;
6163 clean_tree_block(trans
, root
, eb
);
6166 if (eb
== root
->node
) {
6167 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
6170 BUG_ON(root
->root_key
.objectid
!=
6171 btrfs_header_owner(eb
));
6173 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
6174 parent
= path
->nodes
[level
+ 1]->start
;
6176 BUG_ON(root
->root_key
.objectid
!=
6177 btrfs_header_owner(path
->nodes
[level
+ 1]));
6180 btrfs_free_tree_block(trans
, root
, eb
, parent
, wc
->refs
[level
] == 1);
6182 wc
->refs
[level
] = 0;
6183 wc
->flags
[level
] = 0;
6187 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
6188 struct btrfs_root
*root
,
6189 struct btrfs_path
*path
,
6190 struct walk_control
*wc
)
6192 int level
= wc
->level
;
6193 int lookup_info
= 1;
6196 while (level
>= 0) {
6197 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
6204 if (path
->slots
[level
] >=
6205 btrfs_header_nritems(path
->nodes
[level
]))
6208 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
6210 path
->slots
[level
]++;
6219 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
6220 struct btrfs_root
*root
,
6221 struct btrfs_path
*path
,
6222 struct walk_control
*wc
, int max_level
)
6224 int level
= wc
->level
;
6227 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
6228 while (level
< max_level
&& path
->nodes
[level
]) {
6230 if (path
->slots
[level
] + 1 <
6231 btrfs_header_nritems(path
->nodes
[level
])) {
6232 path
->slots
[level
]++;
6235 ret
= walk_up_proc(trans
, root
, path
, wc
);
6239 if (path
->locks
[level
]) {
6240 btrfs_tree_unlock(path
->nodes
[level
]);
6241 path
->locks
[level
] = 0;
6243 free_extent_buffer(path
->nodes
[level
]);
6244 path
->nodes
[level
] = NULL
;
6252 * drop a subvolume tree.
6254 * this function traverses the tree freeing any blocks that only
6255 * referenced by the tree.
6257 * when a shared tree block is found. this function decreases its
6258 * reference count by one. if update_ref is true, this function
6259 * also make sure backrefs for the shared block and all lower level
6260 * blocks are properly updated.
6262 int btrfs_drop_snapshot(struct btrfs_root
*root
,
6263 struct btrfs_block_rsv
*block_rsv
, int update_ref
)
6265 struct btrfs_path
*path
;
6266 struct btrfs_trans_handle
*trans
;
6267 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
6268 struct btrfs_root_item
*root_item
= &root
->root_item
;
6269 struct walk_control
*wc
;
6270 struct btrfs_key key
;
6275 path
= btrfs_alloc_path();
6278 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
6281 trans
= btrfs_start_transaction(tree_root
, 0);
6282 BUG_ON(IS_ERR(trans
));
6285 trans
->block_rsv
= block_rsv
;
6287 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
6288 level
= btrfs_header_level(root
->node
);
6289 path
->nodes
[level
] = btrfs_lock_root_node(root
);
6290 btrfs_set_lock_blocking(path
->nodes
[level
]);
6291 path
->slots
[level
] = 0;
6292 path
->locks
[level
] = 1;
6293 memset(&wc
->update_progress
, 0,
6294 sizeof(wc
->update_progress
));
6296 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
6297 memcpy(&wc
->update_progress
, &key
,
6298 sizeof(wc
->update_progress
));
6300 level
= root_item
->drop_level
;
6302 path
->lowest_level
= level
;
6303 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6304 path
->lowest_level
= 0;
6312 * unlock our path, this is safe because only this
6313 * function is allowed to delete this snapshot
6315 btrfs_unlock_up_safe(path
, 0);
6317 level
= btrfs_header_level(root
->node
);
6319 btrfs_tree_lock(path
->nodes
[level
]);
6320 btrfs_set_lock_blocking(path
->nodes
[level
]);
6322 ret
= btrfs_lookup_extent_info(trans
, root
,
6323 path
->nodes
[level
]->start
,
6324 path
->nodes
[level
]->len
,
6328 BUG_ON(wc
->refs
[level
] == 0);
6330 if (level
== root_item
->drop_level
)
6333 btrfs_tree_unlock(path
->nodes
[level
]);
6334 WARN_ON(wc
->refs
[level
] != 1);
6340 wc
->shared_level
= -1;
6341 wc
->stage
= DROP_REFERENCE
;
6342 wc
->update_ref
= update_ref
;
6344 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
6347 ret
= walk_down_tree(trans
, root
, path
, wc
);
6353 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
6360 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
6364 if (wc
->stage
== DROP_REFERENCE
) {
6366 btrfs_node_key(path
->nodes
[level
],
6367 &root_item
->drop_progress
,
6368 path
->slots
[level
]);
6369 root_item
->drop_level
= level
;
6372 BUG_ON(wc
->level
== 0);
6373 if (btrfs_should_end_transaction(trans
, tree_root
)) {
6374 ret
= btrfs_update_root(trans
, tree_root
,
6379 btrfs_end_transaction_throttle(trans
, tree_root
);
6380 trans
= btrfs_start_transaction(tree_root
, 0);
6381 BUG_ON(IS_ERR(trans
));
6383 trans
->block_rsv
= block_rsv
;
6386 btrfs_release_path(path
);
6389 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
6392 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
6393 ret
= btrfs_find_last_root(tree_root
, root
->root_key
.objectid
,
6397 /* if we fail to delete the orphan item this time
6398 * around, it'll get picked up the next time.
6400 * The most common failure here is just -ENOENT.
6402 btrfs_del_orphan_item(trans
, tree_root
,
6403 root
->root_key
.objectid
);
6407 if (root
->in_radix
) {
6408 btrfs_free_fs_root(tree_root
->fs_info
, root
);
6410 free_extent_buffer(root
->node
);
6411 free_extent_buffer(root
->commit_root
);
6415 btrfs_end_transaction_throttle(trans
, tree_root
);
6417 btrfs_free_path(path
);
6422 * drop subtree rooted at tree block 'node'.
6424 * NOTE: this function will unlock and release tree block 'node'
6426 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
6427 struct btrfs_root
*root
,
6428 struct extent_buffer
*node
,
6429 struct extent_buffer
*parent
)
6431 struct btrfs_path
*path
;
6432 struct walk_control
*wc
;
6438 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
6440 path
= btrfs_alloc_path();
6444 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
6446 btrfs_free_path(path
);
6450 btrfs_assert_tree_locked(parent
);
6451 parent_level
= btrfs_header_level(parent
);
6452 extent_buffer_get(parent
);
6453 path
->nodes
[parent_level
] = parent
;
6454 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
6456 btrfs_assert_tree_locked(node
);
6457 level
= btrfs_header_level(node
);
6458 path
->nodes
[level
] = node
;
6459 path
->slots
[level
] = 0;
6460 path
->locks
[level
] = 1;
6462 wc
->refs
[parent_level
] = 1;
6463 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
6465 wc
->shared_level
= -1;
6466 wc
->stage
= DROP_REFERENCE
;
6469 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
6472 wret
= walk_down_tree(trans
, root
, path
, wc
);
6478 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
6486 btrfs_free_path(path
);
6490 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
6493 u64 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
6494 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
6497 * we add in the count of missing devices because we want
6498 * to make sure that any RAID levels on a degraded FS
6499 * continue to be honored.
6501 num_devices
= root
->fs_info
->fs_devices
->rw_devices
+
6502 root
->fs_info
->fs_devices
->missing_devices
;
6504 if (num_devices
== 1) {
6505 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
6506 stripped
= flags
& ~stripped
;
6508 /* turn raid0 into single device chunks */
6509 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
6512 /* turn mirroring into duplication */
6513 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
6514 BTRFS_BLOCK_GROUP_RAID10
))
6515 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
6518 /* they already had raid on here, just return */
6519 if (flags
& stripped
)
6522 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
6523 stripped
= flags
& ~stripped
;
6525 /* switch duplicated blocks with raid1 */
6526 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
6527 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
6529 /* turn single device chunks into raid0 */
6530 return stripped
| BTRFS_BLOCK_GROUP_RAID0
;
6535 static int set_block_group_ro(struct btrfs_block_group_cache
*cache
)
6537 struct btrfs_space_info
*sinfo
= cache
->space_info
;
6544 spin_lock(&sinfo
->lock
);
6545 spin_lock(&cache
->lock
);
6546 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
6547 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
6549 if (sinfo
->bytes_used
+ sinfo
->bytes_reserved
+ sinfo
->bytes_pinned
+
6550 sinfo
->bytes_may_use
+ sinfo
->bytes_readonly
+
6551 cache
->reserved_pinned
+ num_bytes
<= sinfo
->total_bytes
) {
6552 sinfo
->bytes_readonly
+= num_bytes
;
6553 sinfo
->bytes_reserved
+= cache
->reserved_pinned
;
6554 cache
->reserved_pinned
= 0;
6559 spin_unlock(&cache
->lock
);
6560 spin_unlock(&sinfo
->lock
);
6564 int btrfs_set_block_group_ro(struct btrfs_root
*root
,
6565 struct btrfs_block_group_cache
*cache
)
6568 struct btrfs_trans_handle
*trans
;
6574 trans
= btrfs_join_transaction(root
);
6575 BUG_ON(IS_ERR(trans
));
6577 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
6578 if (alloc_flags
!= cache
->flags
)
6579 do_chunk_alloc(trans
, root
, 2 * 1024 * 1024, alloc_flags
,
6582 ret
= set_block_group_ro(cache
);
6585 alloc_flags
= get_alloc_profile(root
, cache
->space_info
->flags
);
6586 ret
= do_chunk_alloc(trans
, root
, 2 * 1024 * 1024, alloc_flags
,
6590 ret
= set_block_group_ro(cache
);
6592 btrfs_end_transaction(trans
, root
);
6596 int btrfs_force_chunk_alloc(struct btrfs_trans_handle
*trans
,
6597 struct btrfs_root
*root
, u64 type
)
6599 u64 alloc_flags
= get_alloc_profile(root
, type
);
6600 return do_chunk_alloc(trans
, root
, 2 * 1024 * 1024, alloc_flags
,
6605 * helper to account the unused space of all the readonly block group in the
6606 * list. takes mirrors into account.
6608 static u64
__btrfs_get_ro_block_group_free_space(struct list_head
*groups_list
)
6610 struct btrfs_block_group_cache
*block_group
;
6614 list_for_each_entry(block_group
, groups_list
, list
) {
6615 spin_lock(&block_group
->lock
);
6617 if (!block_group
->ro
) {
6618 spin_unlock(&block_group
->lock
);
6622 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_RAID1
|
6623 BTRFS_BLOCK_GROUP_RAID10
|
6624 BTRFS_BLOCK_GROUP_DUP
))
6629 free_bytes
+= (block_group
->key
.offset
-
6630 btrfs_block_group_used(&block_group
->item
)) *
6633 spin_unlock(&block_group
->lock
);
6640 * helper to account the unused space of all the readonly block group in the
6641 * space_info. takes mirrors into account.
6643 u64
btrfs_account_ro_block_groups_free_space(struct btrfs_space_info
*sinfo
)
6648 spin_lock(&sinfo
->lock
);
6650 for(i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
6651 if (!list_empty(&sinfo
->block_groups
[i
]))
6652 free_bytes
+= __btrfs_get_ro_block_group_free_space(
6653 &sinfo
->block_groups
[i
]);
6655 spin_unlock(&sinfo
->lock
);
6660 int btrfs_set_block_group_rw(struct btrfs_root
*root
,
6661 struct btrfs_block_group_cache
*cache
)
6663 struct btrfs_space_info
*sinfo
= cache
->space_info
;
6668 spin_lock(&sinfo
->lock
);
6669 spin_lock(&cache
->lock
);
6670 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
6671 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
6672 sinfo
->bytes_readonly
-= num_bytes
;
6674 spin_unlock(&cache
->lock
);
6675 spin_unlock(&sinfo
->lock
);
6680 * checks to see if its even possible to relocate this block group.
6682 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
6683 * ok to go ahead and try.
6685 int btrfs_can_relocate(struct btrfs_root
*root
, u64 bytenr
)
6687 struct btrfs_block_group_cache
*block_group
;
6688 struct btrfs_space_info
*space_info
;
6689 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
6690 struct btrfs_device
*device
;
6694 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
6696 /* odd, couldn't find the block group, leave it alone */
6700 /* no bytes used, we're good */
6701 if (!btrfs_block_group_used(&block_group
->item
))
6704 space_info
= block_group
->space_info
;
6705 spin_lock(&space_info
->lock
);
6707 full
= space_info
->full
;
6710 * if this is the last block group we have in this space, we can't
6711 * relocate it unless we're able to allocate a new chunk below.
6713 * Otherwise, we need to make sure we have room in the space to handle
6714 * all of the extents from this block group. If we can, we're good
6716 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
6717 (space_info
->bytes_used
+ space_info
->bytes_reserved
+
6718 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
6719 btrfs_block_group_used(&block_group
->item
) <
6720 space_info
->total_bytes
)) {
6721 spin_unlock(&space_info
->lock
);
6724 spin_unlock(&space_info
->lock
);
6727 * ok we don't have enough space, but maybe we have free space on our
6728 * devices to allocate new chunks for relocation, so loop through our
6729 * alloc devices and guess if we have enough space. However, if we
6730 * were marked as full, then we know there aren't enough chunks, and we
6737 mutex_lock(&root
->fs_info
->chunk_mutex
);
6738 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
6739 u64 min_free
= btrfs_block_group_used(&block_group
->item
);
6743 * check to make sure we can actually find a chunk with enough
6744 * space to fit our block group in.
6746 if (device
->total_bytes
> device
->bytes_used
+ min_free
) {
6747 ret
= find_free_dev_extent(NULL
, device
, min_free
,
6754 mutex_unlock(&root
->fs_info
->chunk_mutex
);
6756 btrfs_put_block_group(block_group
);
6760 static int find_first_block_group(struct btrfs_root
*root
,
6761 struct btrfs_path
*path
, struct btrfs_key
*key
)
6764 struct btrfs_key found_key
;
6765 struct extent_buffer
*leaf
;
6768 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
6773 slot
= path
->slots
[0];
6774 leaf
= path
->nodes
[0];
6775 if (slot
>= btrfs_header_nritems(leaf
)) {
6776 ret
= btrfs_next_leaf(root
, path
);
6783 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
6785 if (found_key
.objectid
>= key
->objectid
&&
6786 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
6796 void btrfs_put_block_group_cache(struct btrfs_fs_info
*info
)
6798 struct btrfs_block_group_cache
*block_group
;
6802 struct inode
*inode
;
6804 block_group
= btrfs_lookup_first_block_group(info
, last
);
6805 while (block_group
) {
6806 spin_lock(&block_group
->lock
);
6807 if (block_group
->iref
)
6809 spin_unlock(&block_group
->lock
);
6810 block_group
= next_block_group(info
->tree_root
,
6820 inode
= block_group
->inode
;
6821 block_group
->iref
= 0;
6822 block_group
->inode
= NULL
;
6823 spin_unlock(&block_group
->lock
);
6825 last
= block_group
->key
.objectid
+ block_group
->key
.offset
;
6826 btrfs_put_block_group(block_group
);
6830 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
6832 struct btrfs_block_group_cache
*block_group
;
6833 struct btrfs_space_info
*space_info
;
6834 struct btrfs_caching_control
*caching_ctl
;
6837 down_write(&info
->extent_commit_sem
);
6838 while (!list_empty(&info
->caching_block_groups
)) {
6839 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
6840 struct btrfs_caching_control
, list
);
6841 list_del(&caching_ctl
->list
);
6842 put_caching_control(caching_ctl
);
6844 up_write(&info
->extent_commit_sem
);
6846 spin_lock(&info
->block_group_cache_lock
);
6847 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
6848 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
6850 rb_erase(&block_group
->cache_node
,
6851 &info
->block_group_cache_tree
);
6852 spin_unlock(&info
->block_group_cache_lock
);
6854 down_write(&block_group
->space_info
->groups_sem
);
6855 list_del(&block_group
->list
);
6856 up_write(&block_group
->space_info
->groups_sem
);
6858 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
6859 wait_block_group_cache_done(block_group
);
6862 * We haven't cached this block group, which means we could
6863 * possibly have excluded extents on this block group.
6865 if (block_group
->cached
== BTRFS_CACHE_NO
)
6866 free_excluded_extents(info
->extent_root
, block_group
);
6868 btrfs_remove_free_space_cache(block_group
);
6869 btrfs_put_block_group(block_group
);
6871 spin_lock(&info
->block_group_cache_lock
);
6873 spin_unlock(&info
->block_group_cache_lock
);
6875 /* now that all the block groups are freed, go through and
6876 * free all the space_info structs. This is only called during
6877 * the final stages of unmount, and so we know nobody is
6878 * using them. We call synchronize_rcu() once before we start,
6879 * just to be on the safe side.
6883 release_global_block_rsv(info
);
6885 while(!list_empty(&info
->space_info
)) {
6886 space_info
= list_entry(info
->space_info
.next
,
6887 struct btrfs_space_info
,
6889 if (space_info
->bytes_pinned
> 0 ||
6890 space_info
->bytes_reserved
> 0) {
6892 dump_space_info(space_info
, 0, 0);
6894 list_del(&space_info
->list
);
6900 static void __link_block_group(struct btrfs_space_info
*space_info
,
6901 struct btrfs_block_group_cache
*cache
)
6903 int index
= get_block_group_index(cache
);
6905 down_write(&space_info
->groups_sem
);
6906 list_add_tail(&cache
->list
, &space_info
->block_groups
[index
]);
6907 up_write(&space_info
->groups_sem
);
6910 int btrfs_read_block_groups(struct btrfs_root
*root
)
6912 struct btrfs_path
*path
;
6914 struct btrfs_block_group_cache
*cache
;
6915 struct btrfs_fs_info
*info
= root
->fs_info
;
6916 struct btrfs_space_info
*space_info
;
6917 struct btrfs_key key
;
6918 struct btrfs_key found_key
;
6919 struct extent_buffer
*leaf
;
6923 root
= info
->extent_root
;
6926 btrfs_set_key_type(&key
, BTRFS_BLOCK_GROUP_ITEM_KEY
);
6927 path
= btrfs_alloc_path();
6932 cache_gen
= btrfs_super_cache_generation(&root
->fs_info
->super_copy
);
6933 if (cache_gen
!= 0 &&
6934 btrfs_super_generation(&root
->fs_info
->super_copy
) != cache_gen
)
6936 if (btrfs_test_opt(root
, CLEAR_CACHE
))
6938 if (!btrfs_test_opt(root
, SPACE_CACHE
) && cache_gen
)
6939 printk(KERN_INFO
"btrfs: disk space caching is enabled\n");
6942 ret
= find_first_block_group(root
, path
, &key
);
6947 leaf
= path
->nodes
[0];
6948 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6949 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
6954 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
6956 if (!cache
->free_space_ctl
) {
6962 atomic_set(&cache
->count
, 1);
6963 spin_lock_init(&cache
->lock
);
6964 cache
->fs_info
= info
;
6965 INIT_LIST_HEAD(&cache
->list
);
6966 INIT_LIST_HEAD(&cache
->cluster_list
);
6969 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
6971 read_extent_buffer(leaf
, &cache
->item
,
6972 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
6973 sizeof(cache
->item
));
6974 memcpy(&cache
->key
, &found_key
, sizeof(found_key
));
6976 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
6977 btrfs_release_path(path
);
6978 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
6979 cache
->sectorsize
= root
->sectorsize
;
6981 btrfs_init_free_space_ctl(cache
);
6984 * We need to exclude the super stripes now so that the space
6985 * info has super bytes accounted for, otherwise we'll think
6986 * we have more space than we actually do.
6988 exclude_super_stripes(root
, cache
);
6991 * check for two cases, either we are full, and therefore
6992 * don't need to bother with the caching work since we won't
6993 * find any space, or we are empty, and we can just add all
6994 * the space in and be done with it. This saves us _alot_ of
6995 * time, particularly in the full case.
6997 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
6998 cache
->last_byte_to_unpin
= (u64
)-1;
6999 cache
->cached
= BTRFS_CACHE_FINISHED
;
7000 free_excluded_extents(root
, cache
);
7001 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
7002 cache
->last_byte_to_unpin
= (u64
)-1;
7003 cache
->cached
= BTRFS_CACHE_FINISHED
;
7004 add_new_free_space(cache
, root
->fs_info
,
7006 found_key
.objectid
+
7008 free_excluded_extents(root
, cache
);
7011 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
7012 btrfs_block_group_used(&cache
->item
),
7015 cache
->space_info
= space_info
;
7016 spin_lock(&cache
->space_info
->lock
);
7017 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
7018 spin_unlock(&cache
->space_info
->lock
);
7020 __link_block_group(space_info
, cache
);
7022 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
7025 set_avail_alloc_bits(root
->fs_info
, cache
->flags
);
7026 if (btrfs_chunk_readonly(root
, cache
->key
.objectid
))
7027 set_block_group_ro(cache
);
7030 list_for_each_entry_rcu(space_info
, &root
->fs_info
->space_info
, list
) {
7031 if (!(get_alloc_profile(root
, space_info
->flags
) &
7032 (BTRFS_BLOCK_GROUP_RAID10
|
7033 BTRFS_BLOCK_GROUP_RAID1
|
7034 BTRFS_BLOCK_GROUP_DUP
)))
7037 * avoid allocating from un-mirrored block group if there are
7038 * mirrored block groups.
7040 list_for_each_entry(cache
, &space_info
->block_groups
[3], list
)
7041 set_block_group_ro(cache
);
7042 list_for_each_entry(cache
, &space_info
->block_groups
[4], list
)
7043 set_block_group_ro(cache
);
7046 init_global_block_rsv(info
);
7049 btrfs_free_path(path
);
7053 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
7054 struct btrfs_root
*root
, u64 bytes_used
,
7055 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
7059 struct btrfs_root
*extent_root
;
7060 struct btrfs_block_group_cache
*cache
;
7062 extent_root
= root
->fs_info
->extent_root
;
7064 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
7066 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
7069 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
7071 if (!cache
->free_space_ctl
) {
7076 cache
->key
.objectid
= chunk_offset
;
7077 cache
->key
.offset
= size
;
7078 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
7079 cache
->sectorsize
= root
->sectorsize
;
7080 cache
->fs_info
= root
->fs_info
;
7082 atomic_set(&cache
->count
, 1);
7083 spin_lock_init(&cache
->lock
);
7084 INIT_LIST_HEAD(&cache
->list
);
7085 INIT_LIST_HEAD(&cache
->cluster_list
);
7087 btrfs_init_free_space_ctl(cache
);
7089 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
7090 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
7091 cache
->flags
= type
;
7092 btrfs_set_block_group_flags(&cache
->item
, type
);
7094 cache
->last_byte_to_unpin
= (u64
)-1;
7095 cache
->cached
= BTRFS_CACHE_FINISHED
;
7096 exclude_super_stripes(root
, cache
);
7098 add_new_free_space(cache
, root
->fs_info
, chunk_offset
,
7099 chunk_offset
+ size
);
7101 free_excluded_extents(root
, cache
);
7103 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
7104 &cache
->space_info
);
7107 spin_lock(&cache
->space_info
->lock
);
7108 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
7109 spin_unlock(&cache
->space_info
->lock
);
7111 __link_block_group(cache
->space_info
, cache
);
7113 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
7116 ret
= btrfs_insert_item(trans
, extent_root
, &cache
->key
, &cache
->item
,
7117 sizeof(cache
->item
));
7120 set_avail_alloc_bits(extent_root
->fs_info
, type
);
7125 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
7126 struct btrfs_root
*root
, u64 group_start
)
7128 struct btrfs_path
*path
;
7129 struct btrfs_block_group_cache
*block_group
;
7130 struct btrfs_free_cluster
*cluster
;
7131 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
7132 struct btrfs_key key
;
7133 struct inode
*inode
;
7137 root
= root
->fs_info
->extent_root
;
7139 block_group
= btrfs_lookup_block_group(root
->fs_info
, group_start
);
7140 BUG_ON(!block_group
);
7141 BUG_ON(!block_group
->ro
);
7144 * Free the reserved super bytes from this block group before
7147 free_excluded_extents(root
, block_group
);
7149 memcpy(&key
, &block_group
->key
, sizeof(key
));
7150 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
7151 BTRFS_BLOCK_GROUP_RAID1
|
7152 BTRFS_BLOCK_GROUP_RAID10
))
7157 /* make sure this block group isn't part of an allocation cluster */
7158 cluster
= &root
->fs_info
->data_alloc_cluster
;
7159 spin_lock(&cluster
->refill_lock
);
7160 btrfs_return_cluster_to_free_space(block_group
, cluster
);
7161 spin_unlock(&cluster
->refill_lock
);
7164 * make sure this block group isn't part of a metadata
7165 * allocation cluster
7167 cluster
= &root
->fs_info
->meta_alloc_cluster
;
7168 spin_lock(&cluster
->refill_lock
);
7169 btrfs_return_cluster_to_free_space(block_group
, cluster
);
7170 spin_unlock(&cluster
->refill_lock
);
7172 path
= btrfs_alloc_path();
7175 inode
= lookup_free_space_inode(root
, block_group
, path
);
7176 if (!IS_ERR(inode
)) {
7177 btrfs_orphan_add(trans
, inode
);
7179 /* One for the block groups ref */
7180 spin_lock(&block_group
->lock
);
7181 if (block_group
->iref
) {
7182 block_group
->iref
= 0;
7183 block_group
->inode
= NULL
;
7184 spin_unlock(&block_group
->lock
);
7187 spin_unlock(&block_group
->lock
);
7189 /* One for our lookup ref */
7193 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
7194 key
.offset
= block_group
->key
.objectid
;
7197 ret
= btrfs_search_slot(trans
, tree_root
, &key
, path
, -1, 1);
7201 btrfs_release_path(path
);
7203 ret
= btrfs_del_item(trans
, tree_root
, path
);
7206 btrfs_release_path(path
);
7209 spin_lock(&root
->fs_info
->block_group_cache_lock
);
7210 rb_erase(&block_group
->cache_node
,
7211 &root
->fs_info
->block_group_cache_tree
);
7212 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
7214 down_write(&block_group
->space_info
->groups_sem
);
7216 * we must use list_del_init so people can check to see if they
7217 * are still on the list after taking the semaphore
7219 list_del_init(&block_group
->list
);
7220 up_write(&block_group
->space_info
->groups_sem
);
7222 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
7223 wait_block_group_cache_done(block_group
);
7225 btrfs_remove_free_space_cache(block_group
);
7227 spin_lock(&block_group
->space_info
->lock
);
7228 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
7229 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
7230 block_group
->space_info
->disk_total
-= block_group
->key
.offset
* factor
;
7231 spin_unlock(&block_group
->space_info
->lock
);
7233 memcpy(&key
, &block_group
->key
, sizeof(key
));
7235 btrfs_clear_space_info_full(root
->fs_info
);
7237 btrfs_put_block_group(block_group
);
7238 btrfs_put_block_group(block_group
);
7240 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
7246 ret
= btrfs_del_item(trans
, root
, path
);
7248 btrfs_free_path(path
);
7252 int btrfs_init_space_info(struct btrfs_fs_info
*fs_info
)
7254 struct btrfs_space_info
*space_info
;
7255 struct btrfs_super_block
*disk_super
;
7261 disk_super
= &fs_info
->super_copy
;
7262 if (!btrfs_super_root(disk_super
))
7265 features
= btrfs_super_incompat_flags(disk_super
);
7266 if (features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
7269 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
7270 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
7275 flags
= BTRFS_BLOCK_GROUP_METADATA
| BTRFS_BLOCK_GROUP_DATA
;
7276 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
7278 flags
= BTRFS_BLOCK_GROUP_METADATA
;
7279 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
7283 flags
= BTRFS_BLOCK_GROUP_DATA
;
7284 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
7290 int btrfs_error_unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
7292 return unpin_extent_range(root
, start
, end
);
7295 int btrfs_error_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
7296 u64 num_bytes
, u64
*actual_bytes
)
7298 return btrfs_discard_extent(root
, bytenr
, num_bytes
, actual_bytes
);
7301 int btrfs_trim_fs(struct btrfs_root
*root
, struct fstrim_range
*range
)
7303 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7304 struct btrfs_block_group_cache
*cache
= NULL
;
7311 cache
= btrfs_lookup_block_group(fs_info
, range
->start
);
7314 if (cache
->key
.objectid
>= (range
->start
+ range
->len
)) {
7315 btrfs_put_block_group(cache
);
7319 start
= max(range
->start
, cache
->key
.objectid
);
7320 end
= min(range
->start
+ range
->len
,
7321 cache
->key
.objectid
+ cache
->key
.offset
);
7323 if (end
- start
>= range
->minlen
) {
7324 if (!block_group_cache_done(cache
)) {
7325 ret
= cache_block_group(cache
, NULL
, root
, 0);
7327 wait_block_group_cache_done(cache
);
7329 ret
= btrfs_trim_block_group(cache
,
7335 trimmed
+= group_trimmed
;
7337 btrfs_put_block_group(cache
);
7342 cache
= next_block_group(fs_info
->tree_root
, cache
);
7345 range
->len
= trimmed
;